Figure 1.
The variable-wise unfolding method of batch process dataset.
Citation: Valeriia Mikhailova, Valeriia Gulaia, Vladlena Tiasto, Stanislav Rybtsov, Margarita Yatsunskaya, Alexander Kagansky. Towards an advanced cell-based in vitro glioma model system[J]. AIMS Genetics, 2018, 5(2): 91-112. doi: 10.3934/genet.2018.2.91
| [1] | Helena Hanusová, Karolína Juřenová, Erika Hurajová, Magdalena Daria Vaverková, Jan Winkler . Vegetation structure of bio-belts as agro-environmentally-climatic measures to support biodiversity on arable land: A case study. AIMS Agriculture and Food, 2022, 7(4): 883-896. doi: 10.3934/agrfood.2022054 |
| [2] | Jan Willem Erisman, Nick van Eekeren, Jan de Wit, Chris Koopmans, Willemijn Cuijpers, Natasja Oerlemans, Ben J. Koks . Agriculture and biodiversity: a better balance benefits both. AIMS Agriculture and Food, 2016, 1(2): 157-174. doi: 10.3934/agrfood.2016.2.157 |
| [3] | Guizhen Wang, Limin Hua, Victor R. Squires, Guozhen Du . What road should the grazing industry take on pastoral land in China?. AIMS Agriculture and Food, 2017, 2(4): 354-369. doi: 10.3934/agrfood.2017.4.354 |
| [4] | Babatope Samuel Ajayo, Baffour Badu-Apraku, Morakinyo A. B. Fakorede, Richard O. Akinwale . Plant density and nitrogen responses of maize hybrids in diverse agroecologies of west and central Africa. AIMS Agriculture and Food, 2021, 6(1): 381-400. doi: 10.3934/agrfood.2021023 |
| [5] | Eric Tzyy Jiann Chong, Lucky Poh Wah Goh, Mariam Abd. Latip, Zaleha Abdul Aziz, Noumie Surugau, Ping-Chin Lee . Genetic diversity of upland traditional rice varieties in Malaysian Borneo based on mitochondrial cytochrome c oxidase 3 gene analysis. AIMS Agriculture and Food, 2021, 6(1): 235-246. doi: 10.3934/agrfood.2021015 |
| [6] | Ezekiel Mugendi Njeru . Exploiting diversity to promote arbuscular mycorrhizal symbiosis and crop productivity in organic farming systems. AIMS Agriculture and Food, 2018, 3(3): 280-294. doi: 10.3934/agrfood.2018.3.280 |
| [7] | Janice Liang, Travis Reynolds, Alemayehu Wassie, Cathy Collins, Atalel Wubalem . Effects of exotic Eucalyptus spp. plantations on soil properties in and around sacred natural sites in the northern Ethiopian Highlands. AIMS Agriculture and Food, 2016, 1(2): 175-193. doi: 10.3934/agrfood.2016.2.175 |
| [8] | Boris Boincean, Amir Kassam, Gottlieb Basch, Don Reicosky, Emilio Gonzalez, Tony Reynolds, Marina Ilusca, Marin Cebotari, Grigore Rusnac, Vadim Cuzeac, Lidia Bulat, Dorian Pasat, Stanislav Stadnic, Sergiu Gavrilas, Ion Boaghii . Towards Conservation Agriculture systems in Moldova. AIMS Agriculture and Food, 2016, 1(4): 369-386. doi: 10.3934/agrfood.2016.4.369 |
| [9] | Aliou Badara Kouyate, Vincent Logah, Robert Clement Abaidoo, Francis Marthy Tetteh, Mensah Bonsu, Sidiki Gabriel Dembélé . Phosphorus sorption characteristics in the Sahel: Estimates from soils in Mali. AIMS Agriculture and Food, 2023, 8(4): 995-1009. doi: 10.3934/agrfood.2023053 |
| [10] | Simon Wambui Mburu, Gilbert Koskey, Ezekiel Mugendi Njeru, John M. Maingi . Revitalization of bacterial endophytes and rhizobacteria for nutrients bioavailability in degraded soils to promote crop production. AIMS Agriculture and Food, 2021, 6(2): 496-524. doi: 10.3934/agrfood.2021029 |
In chemical processes, accurate real-time predictions of product quality are highly desirable, which is critical to realize successful process control, monitoring and optimization [1,2]. However, due to the costs of online analyzer and offline laboratory analysis, the process often encounters the great challenge of lacking reliable quality estimation. Soft sensing technique, which aims to construct theoretical or statistical models that can describe the functional relationship between process variables (easy-to-measure variables) and quality variables (difficult-to-measure variables), is proposed to address this issue and attracts much attention in both academia and industry. Generally, soft sensors can be classified into three groups: Model-driven, data-driven and mixed models [3,4,5,6]. Compared with model-driven method, data-driven one does not require in-depth mechanical knowledge of processes and only relies on recorded process datasets, which shows great flexibility and low complexity. Many dynamical models, such as nonlinear autoregressive with exogenous inputs (NARX) [7], and data-driven models, such as partial least squares (PLS), artificial neural networks (ANN), support vector machine (SVM), and Gaussian process regression (GPR) [8,9,10,11,12], have been successfully applied to online quality prediction.
Batch processes play an important role in the production of food, drugs, special chemicals and biological industrial products, which have high requirements for product quality and safe operation. In addition to the nonlinear and time-varying characteristics, other distinct characteristics, such as instability, finite duration, and batch-to-batch variations, are quite different from those of continuous processes [13,14]. It is difficult to construct accurate predictive models as the operating conditions vary. Furthermore, datasets obtained from batch processes are high-dimensional, including different batches, variables, and sampling time. Thus, they cannot be directly used for modeling and need to be preprocessed. Generally, multidimensional datasets contain abundant process information that can contribute to informative models, but it may also lead to information redundancy and complex model structure. Thus, dimension reduction and significant feature information extraction are crucial in satisfactory soft sensor development. Multiway principal component analysis (MPCA) [15,16,17] and multiway PLS (MPLS) [18,19] have been successfully applied in the fault diagnosis and soft sensing for batch processes. MPCA can be used to realize data analysis and preprocessing. Variable-wise unfolding method, which tends to keep the track of variables and retain the overall change information of process variables in batch and time, is introduced to obtain the two-dimensional datasets. Then, ordinary PCA is applied to dimensionality reduction and extract maximum amount of process information, making it more effective to soft sensor modeling.
Traditional nonlinear soft sensors can achieve a universal generalization performance in quality prediction of chemical processes. However, many of them rely on a single global model under the assumption that the operating phases and conditions are constant in the whole process. With operating conditions or product demands changing, processes exhibit apparent multiphase behaviors while different phases present various process characteristics, thereby resulting in the poor regression accuracy of global models.
Ensemble learning has been investigated and developed to be an effective tool to improve the generalization performance of soft sensors, especially for multiphase or multimode batch processes [20]. Under ensemble learning framework, the process dataset is partitioned into several local domains, then a series of local high-performance models are constructed and integrated to make a final quality prediction. Instead of global model construction, ensemble model based soft sensors can greatly enhance estimation accuracy and maintain satisfactory performance for a long time even though process characteristics change. The first step of ensemble learning method is to generate subsets from process data samples. Several popular data partition approaches include bagging [21], boosting [22], clustering [23] and the subspace method [24]. Clustering based methods, such as K-means, fuzzy C-means (FCM) [25], and Gaussian mixture model (GMM) [26], have been widely used and have shown their effectiveness in data clustering for multiphase processes. For example, Wang et al. used GMM to create local partitions and verified the feasibility and reliability of the proposed soft sensor [26]. However, this method only considers one batch of process data and does not take multiphase characteristics into account. In addition, the dataset length of each batch may not be equal because of the complex operating conditions in actual processes. Prediction combination is another important step of ensemble learning method. Traditional approaches for this purpose are averaging, voting, Bayesian inference, and learning method [2,20,26]. Bayesian fusion method has been proven to be a natural fit for ensemble model combination due to its strong statistical learning and analytical abilities from datasets [27,28]. It can remarkably and effectively utilize the limited process information.
Motivated to address the aforementioned issues, a novel ensemble learning based soft sensor, namely ensemble least squares support vector regression (LSSVR) [29,30] based on GMM method (GMM-LSSVR), is developed in this paper for the quality prediction of multiphase/multimode nonlinear batch processes. Firstly, MPCA is applied to data unfolding and information extraction for original 3-dimensional process datasets. In this method, the feature vectors corresponding to the large feature values are selected to form a subspace, where original datasets are mapped, then the preprocessed low-dimensional data matrix can be obtained for soft sensor modeling. Secondly, the Bayesian information criterion (BIC) [31] technique is introduced to determine the optimal number of Gaussian components for phase partition. And the newly obtained datasets are divided into several different subsets by GMM method to produce ensemble components. Thirdly, the grid search (GS) [32] method is used to generate all possible parameter pairs (σ, γ) due to its significant search effect and easy implementation. Meanwhile, ten-fold cross-validation [33] technique is employed to calculate the average relative error and evaluate the optimality of each pair. In such cases, an optimal parameter pair can be determined for each local LSSVR model, which greatly contributes to reliability enhancement. Finally, the Bayesian inference strategy is used to estimate the posterior probability of each test sample with respect to different operation dynamics and multiple models are combined with posterior probability based weightings for the final prediction.
The remainder of this paper is organized as follows. Section 2 briefly reviews LSSVR model, MPCA and GMM methods. Section 3 presents some details of the proposed novel soft sensor, including its modeling method, parameters determination, and combination strategy. Section 4 evaluates the effectiveness of the modeling method via simulation results in a batch process. Finally, Section 5 draws the conclusions of this paper.
The LSSVR model is modified from support vector regression (SVR) [29]. Instead of inequality constraints applied, LSSVR uses equality constraints in the optimization problem in order to turn the convex quadratic optimization procedure into the solution of linear equations, which has shown its great ability in dealing with significant nonlinearity in batch processes. Thus, LSSVR is applied to construct local models upon the several partitioned regions in this paper.
Given
|
$
\left\{ minJ(ωω,ζ)=12ωωTωω+γ12ζζTζζs.t.y=ZTωω+bI+ζζ \right.
$
|
(1) |
where
By introducing Lagrange multipliers
| $ L\left( {{\bf{ \pmb{\mathsf{ ω}} }}, b, \zeta , {\bf{ \pmb{\mathsf{ α}} }}} \right) = J\left( {{\bf{ \pmb{\mathsf{ ω}} }}, \zeta } \right) - {{\bf{ \pmb{\mathsf{ α}} }}^{\rm{T}}}\left( {{{\bf{Z}}^{\rm{T}}}{\bf{ \pmb{\mathsf{ ω}} }} + b{\bf{I}} + {\bf{ \pmb{\mathsf{ ζ}} }} - {\bf{y}}} \right) $ | (2) |
The following linear equations can be obtained by referring to the Karush-Kuhn-Tucker (KKT) condition for optimality.
|
$
\left\{ ∂L∂ωω=0⇒ωω=Zαα∂L∂b=0⇒ααTI=0∂L∂ζζ=0⇒αα=γζζ∂L∂αα=0⇒ZTωω+bI+ζζ−y=0 \right.
$
|
(3) |
Then, a linear system can be described by simplifying equations and eliminating
|
$
\left[ {bαα } \right] = {\left[ {01T1H } \right]^{ - 1}}\left[ {0y } \right]
$
|
(4) |
where
| $ {K_{i, j}} = \varphi {\left( {{{\bf{x}}_i}} \right)^{\rm{T}}}\varphi \left( {{{\bf{x}}_j}} \right) = k\left( {{{\bf{x}}_i}, {{\bf{x}}_j}} \right),\ \ \ \forall \left( {i, j} \right) \in {{\rm{N}}_N} \times {{\rm{N}}_N} $ | (5) |
In this work, the Gaussian kernel function is adopted to be the kernel function of LSSVR:
| $f\left( {\bf{x}} \right) = \varphi {\left( {\bf{x}} \right)^{\rm{T}}}{{\bf{ \pmb{\mathsf{ ω}} }}^*} + {b^*} = \varphi {\left( {\bf{x}} \right)^{\rm{T}}}{\bf{Z}}{{\bf{ \pmb{\mathsf{ α}} }}^*} + {b^*} = \sum\limits_{i = 1}^N {{\bf{ \pmb{\mathsf{ α}} }}_i^*\varphi {{\left( {\bf{x}} \right)}^{\rm{T}}}\varphi \left( {{{\bf{x}}_i}} \right)} + {b^*} = \sum\limits_{i = 1}^N {{\bf{ \pmb{\mathsf{ α}} }}_i^*k\left( {{\bf{x}}, {{\bf{x}}_i}} \right)} + {b^*} $ | (6) |
In batch processes, the collected datasets are related to batches, variables and sampling time. Excessive data information may lead to information redundancy and deteriorate the estimation performance of soft sensor models. MPCA method has been proven to be effective in dimensionality reduction and widely used in data preprocessing of batch processes.
The dataset of a batch process can be given as a three-dimensional matrix
In this way, the original dataset can be rewritten into a new KI-dimensional variable space, then the new data matrix is preprocessed by
|
$
\left\{ ¯xi,j,k=xi,j,k−¯xjsj,k¯xj=1KIK∑k=1I∑i=1xi,j,ksj,k=√1KIK∑k=1I∑i=1(xi,j,k−¯xj)2 \right.
$
|
(7) |
where
For the standard dataset
| ${\bf{X}}\left( {{\bf{KI}}{\rm{ \times }}{\bf{J}}} \right) = {\bf{T}}{{\bf{P}}^{\rm{T}}} + {\bf{E}} $ | (8) |
where
As an effective probabilistic approach for data clustering, GMM is widely used for process monitoring and soft sensor application. The main purpose of GMM method is to identify and localize phase of data samples in batch processes.
Consider a training dataset consisting of
| $ p({\bf{x}}\mathit{\boldsymbol{| \boldsymbol{\varTheta} }}) = \sum\limits_{g = 1}^G {{\pi _g}p({\bf{x}}\mathit{\boldsymbol{|}}{\mathit{\boldsymbol{ \boldsymbol{\varTheta} }}_g})} $ | (9) |
where
| $\sum\limits_{g = 1}^G {{\pi _g} = 1} , \ \ 0 \leqslant {\pi _g} \leqslant 1 $ | (10) |
And
| $p({\bf{x}}\mathit{\boldsymbol{|}}{\mathit{\boldsymbol{ \boldsymbol{\varTheta} }}_g}) = \frac{1}{{\sqrt {{{(2\pi )}^m}\left| {{\sum _g}} \right|} }} \times \exp \left[ { - \tfrac{1}{2}{{\left( {{\bf{x}}\mathit{\boldsymbol{ - }}{\mu _g}} \right)}^{\rm{T}}}{\sum _g}^{ - 1}\left( {{\bf{x}}\mathit{\boldsymbol{ - }}{\mu _g}} \right)} \right] $ | (11) |
Assume that data samples follow a mixture of a finite number of Gaussian distributions, it can be seen that each Gaussian component has three parameters (
| $ L({\bf{x}}\mathit{\boldsymbol{| \boldsymbol{\varTheta} }}) = \log \mathop \prod \limits_{i = 1}^N \sum\limits_{g = 1}^G {{\pi _g}p({\bf{x}}\mathit{\boldsymbol{|}}{\mathit{\boldsymbol{ \boldsymbol{\varTheta} }}_g})} = \sum\limits_{i = 1}^N {\log } \sum\limits_{g = 1}^G {{\pi _g}p({\bf{x}}\mathit{\boldsymbol{|}}{\mathit{\boldsymbol{ \boldsymbol{\varTheta} }}_g})} $ | (12) |
Then, expectation maximization (EM) algorithm, is introduced to estimate the optimal parameters by iterative calculation, which consist of E step and M step:
E step: Evaluate the posterior probability that ith training data samples, which belongs to the gth Gaussian component by using current parameter values.
| $p({C_g}|{{\bf{x}}_i}) = \frac{{{\pi _g}p({{\bf{x}}_i}|{\mathit{\Theta} _g})}}{{\sum\limits_{g = 1}^G {{\pi _g}p({{\bf{x}}_i}|{\mathit{\Theta} _g})} }}, \ \ i = 1, 2, \cdots N $ | (13) |
M step: Obtain the corresponding likelihood function via the posterior probability calculated by E step. Re-estimate the parameters using the current value.
| $ {\mu _g} = \frac{{\sum\limits_{i = 1}^N {p\left( {{C_g}\left| {{{\bf{x}}_i}} \right.} \right)} {{\bf{x}}_i}}}{{\sum\limits_{i = 1}^N {p\left( {{C_g}\left| {{{\bf{x}}_i}} \right.} \right)} }} $ | (14) |
| $ {\pi _g} = \frac{{\sum\limits_{i = 1}^N {p\left( {{C_g}\left| {{{\bf{x}}_i}} \right.} \right)} }}{N} $ | (15) |
| $ {\sum _g} = \frac{{\sum\limits_{i = 1}^N {p\left( {{C_g}\left| {{{\bf{x}}_i}} \right.} \right)} \left( {{{\bf{x}}_i} - {{\bf{ \pmb{\mathsf{ μ}} }}_g}} \right){{\left( {{{\bf{x}}_i} - {{\bf{ \pmb{\mathsf{ μ}} }}_g}} \right)}^{\rm{T}}}}}{{\sum\limits_{i = 1}^N {p\left( {{C_g}\left| {{{\bf{x}}_i}} \right.} \right)} }} $ | (16) |
The parameter estimation process is not completed until the convergence is satisfied. For batch processes, the number of Gaussian components of GMM model corresponds to the number of stages of the process. Moreover, the mixing coefficient of each Gaussian component for a data sample is determined by the average posterior probability of the sample with respect to the corresponding component.
Parameter determination is an important step of model construction, and it can directly affect the generalization behavior of regression models. The multi-model parameter optimization method shows its strong superiority in tackling parametric uncertainty problems when industrial processes are complex and time-varying [34].
LSSVR models need to determine regularization coefficients and kernel parameters. The commonly used methods for parameter determination include GS and swarm intelligence optimization [36,37,38,39]. In this study, the parameters of the LSSVR model are determined by ten-fold cross-validation and GS methods. First, for the parameter pair (σ, γ) that needs to be determined, GS method is used to form the grid in the given parameter selection interval. Second, the average relative error (Eq. (17)) of the corresponding model is calculated by ten-fold cross-validation method at the grid point. Finally, the parameter pair with the minimum error value is selected as an optimal pair.
| $ \delta = \frac{1}{N}\sum\limits_{i = 1}^N {\frac{{\left| {{y_i} - {{\widehat y}_i}} \right|}}{{{y_i}}}} $ | (17) |
where
The steps of LSSVR parameter determination are presented as follows:
Step 1: Assign an initial value to σ and γ.
Step 2: Determine the search range of σ and γ.
Step 3: Determine the grid point position of the first cross-validation calculation according to the initial value.
Step 4: Select ten-fold cross-validation as the objective function of grid point calculation. Then calculate the errors of all grid points.
Step 5: Compare the error results and determine an optimal parameter pair.
Some traditional soft sensors construct a global regression model for quality prediction; it ignores the multiphase and multistage characteristics of batch processes. Fortunately, ensemble learning based local modeling methods, which can better meet the requirements of prediction accuracy by combining multiple local models, have drawn increasing attention to improving the performance of soft sensors. Therefore, a novel soft sensor, referred to as ensemble LSSVR based on GMM (GMM-LSSVR), is proposed for quality prediction in multiphase batch processes. First, MPCA is employed to data preprocessing, including three-dimensional data unfolding and dimensionality reduction. And GMM method is applied to divide the preprocessed dataset into multiple local domains. Then, several local LSSVR models are constructed for all identified subsets. Meanwhile, optimal hyperparameters are determined by combining ten-fold cross-validation with GS method. Finally, according to the posterior probability of the new sample to each operation period (Eq (18)), the high-performance predictions of local LSSVR models are integrated to produce the overall prediction results by using the Bayesian inference and finite mixture mechanism, as shown in Eq (19).
| $ p({S_g}|{{\bf{x}}_q}) = \frac{{{\pi _g}p({{\bf{x}}_q}|{\mathit{\Theta} _g})}}{{\sum\limits_{g = 1}^G {{\pi _g}p({{\bf{x}}_q}|{\mathit{\Theta} _g})} }} $ | (18) |
| $ {y_p} = \sum\limits_{g = 1}^G {y_q^gp\left( {{S_g}\left| {{{\bf{x}}_q}} \right.} \right)} $ | (19) |
where
When GMM method is applied, the BIC technique is introduced to determine the number of Gaussian components in an intuitive and persuasive way, which can be formulated as
| ${\rm{BIC}} = - 2\log L({\bf{x}}\left| \mathit{\boldsymbol{ \boldsymbol{\varTheta} }} \right.) + dlog(N) $ | (20) |
where
Figure 2 illustrates the online prediction steps of test samples based on GMM-LSSVR method. The proposed soft sensor modeling strategy is shown in Figure 3.
Penicillin fermentation process is a typical microbial fermentation reaction and is often used to be a benchmark process for monitoring, controlling, and quality prediction. This process is a complex multivariable coupled biochemical procedure and often contains significant nonlinearity and time-varying behavior, which can be generally divided into three stages: growth, penicillin synthesis and autolysis stages [20]. Figure 4 shows the flow diagram of penicillin fermentation process. During the whole cultivation process, bacterial growth and antibiotic synthesis process are completed under suitable fermentation conditions such as temperature, pH, and oxygen concentration and so on. Considering the costs of offline chemical analysis and hardware sensors, designing a high-performance soft sensor plays an important role in real-time estimation of penicillin concentration.
A simulation platform named PenSim has been widely used to simulate fed-batch penicillin fermentation process under different operating conditions [20]. In this study, all process data samples for experiments are collected via running the PenSim platform. There are total 16 process variables in the simulation plant, and 11 highly related variables are selected as input variables, which are listed in Table 1. The typical trend plots of input and quality variables are depicted in Figure 5. The entire duration of each batch is set as 400 hours, while the sampling interval is set as 1 hour. Under the normal operating condition, a total of 4 training batches (named as Batches 1 to 4) are obtained for soft sensor model construction, while the additional 2 test batches (named as Batches 5 and 6) are collected for model performance evaluation.
| NO. | Variable description (Unit) | NO. | Variable description (Unit) |
| 1 | Aeration rate (L/h) | 7 | Carbon dioxide concentration (g/L) |
| 2 | Agitator power (W) | 8 | PH (-) |
| 3 | Substrate feed rate (L/h) | 9 | Fermenter temperature (K) |
| 456 | Substrate feed temperature (K)Dissolved oxygen concentration (g/L)Culture volume (L) | 1011 | Generated heat (kcal)Cooling water flow rate (L/h) |
DownLoad:
CSV
For model construction, 100 data samples are collected evenly from Batches 1 to 4, respectively. As a result, training dataset is composed of 400 samples, while additional 200 samples that collected evenly from Batch 5 compose the test dataset 1, and other 200 samples from Batch 6 compose the test dataset 2. Here, two test datasets are used for model evaluation: test dataset 1 in Batch 5 and test dataset 2 in Batch 6 with noisy condition. Suppose that the measure noise is the zero-mean Gaussian noise with variance of 0.01, the dataset 2 is used to study the behavior of the proposed soft sensor model under noisy measure environment. In order to show the sampling strategy more intuitively, for examples, we collect the aeration rate (one of the input variables) values every 4 hours in the training Batch 1. The sampling time plots of aeration rate are illustrated in Figure 6, where the red points represent the data samples selected for modeling. Figure 6a–d and e–f gives the sampling time of aeration rate in the training batches and test batches, respectively. The sampling time plots of other input variables are like that of aeration rate.
Then, MPCA is applied to data preprocessing. Firstly, two-dimensional modeling datasets can be obtained from original multidimensional datasets by variable-wise data unfolding method. Then, PCA, as a well-known technique in statistics and machine learning, is used to compress the input variables, and extract the most important information of the process. The relationship between principal components number and cumulative contribution rate for input dataset is illustrated in Figure 7. In this study, the principal component number can be set as 7 because the corresponding cumulative contribution rate achieves 0.98. As a result, the dimensionally reduced data is obtained and imported into soft sensor models for training.
The BIC value is calculated according to the obtained data matrix to determine the optimal number of Gaussian components. The relationship between the number of Gaussian components and BIC values is shown in Figure 8. When the number of Gaussian components is small, BIC values decrease dramatically. As the number increases, which changes from 3 to 6, BIC values change smoothly. In order to simplify model structure as much as possible and prevent the model from overfitting, the optimal number of Gaussian components is set as 3.
Four soft sensor models have been constructed in the following study:
(ⅰ) GPR: A global GPR model constructed from the preprocessed dataset.
(ⅱ) LSSVR: A global LSSVR model constructed from the preprocessed dataset.
(ⅲ) GMM-GPR: An ensemble model based on several local GPR models constructed from local preprocessed datasets that are obtained by using GMM method.
(ⅳ) GMM-LSSVR: An ensemble model based on several local LSSVR models constructed from local preprocessed datasets that are obtained by using GMM method.
To verify the prediction capabilities of the soft sensors with penicillin concentration, three performance indices including root-mean-square error (RMSE), tracking precision (TP) and coefficient of determination (R2) are used, which are defined as follows:
| $ {\rm{RMSE}} = \sqrt {\frac{1}{N}\sum\limits_{i = 1}^N {{{\left( {{y_i} - {{\widehat y}_i}} \right)}^2}} } $ | (21) |
| $ {\rm{TP}} = 1 - \frac{{\sigma _{error}^2}}{{\sigma _{true}^2}} $ | (22) |
| $ {{\rm{R}}^{\rm{2}}} = 1 - \frac{{\sum\limits_{i = 1}^N {{{\left( {{{\widehat y}_i} - {y_i}} \right)}^2}} }}{{\sum\limits_{i = 1}^N {{{\left( {{y_i} - {{\overline y }_i}} \right)}^2}} }} $ | (23) |
where
Table 2 shows the quantitative comparison of the performance indicators for different four soft sensors. The comparison of global modeling and local learning methods shows that ensemble GPR model and ensemble LSSVR model perform better than global GPR and global LSSVR, respectively, because the RMSE value of the former is smaller than that of the latter. Clearly, GMM based multiple models can accurately and effectively describe the multiphase characteristics of batch process and enhance the ability of model interpretation. Therefore, for penicillin fermentation process, multi-model modeling has higher estimation accuracy and smaller prediction error. Similarly, by comparing GMM-GPR model with GMM-LSSVR model, it can be found that the ensemble LSSVR model based soft sensor has higher prediction accuracy and better tracking effect for penicillin concentration, whereas the ensemble GPR model based soft sensor has bigger RMSE values and smaller TP values. This result shows that, although the prediction performance of GMM- ensemble GPR model is improved compared with the global GPR model, poor predictions for test samples are still observed. As presented, the prediction performance of GMM-GPR model is far inferior to that of GMM-LSSVR model. Despite the presence of noise, as studied for dataset 2 in Batch 6 with noise, GMM-LSSVR based soft sensor still outperforms other different soft sensors. Three performance indicators can demonstrate the feasibility and superiority of the proposed method.
| Method | Batch 5 with no noise | Batch 6 with noise | ||||
| RMSE | TP | R2 | RMSE | TP | R2 | |
| GPR | 0.0101 | 0.9996 | 0.9995 | 0.0206 | 0.9982 | 0.9980 |
| LSSVR | 0.0119 | 0.9994 | 0.9993 | 0.0224 | 0.9981 | 0.9977 |
| GMM-GPR | 0.0094 | 0.9996 | 0.9996 | 0.0177 | 0.9986 | 0.9985 |
| GMM-LSSVR | 0.0039 | 0.9999 | 0.9999 | 0.0125 | 0.9993 | 0.9993 |
DownLoad:
CSV
To present the regression performance of different soft sensors, the prediction results of penicillin concentration for global modeling and local learning methods is depicted in detail in Figures 9 and 10. As shown in Figure 9, the prediction curve of penicillin concentration by GMM-LSSVR model is more in line with the true value curve, thereby showing that the predicted value of penicillin concentration in this method is closer to the true value, and the prediction accuracy is also significantly higher than that of global LSSVR model. Furthermore, the prediction error of GMM-LSSVR model for test samples is reduced, and its generalization performance is better compared with that of GMM-GPR model. Similar analysis conclusions can be made according to the quality prediction results of Batch 6, which is given in Figure 10. This soft sensor modeling method can effectively improve the prediction capability and regression accuracy of global LSSVR model and can better complete the prediction of penicillin concentration.
To further illustrate the effectiveness of the proposed method, Figure 11 shows the comparison of the prediction errors of penicillin concentration for four soft sensors. It can be clearly seen that whether there is noise or not, the prediction error of the GMM-LSSVR model fluctuates less near 0, showing that the prediction results of the model are more consistent with the real results, and the tracking ability is stronger. Compared with different modeling methods, the GMM-LSSVR based soft sensor provides an accurate prediction of the true value of penicillin concentration and has good regression performance. In addition, the scatter plots of prediction results for penicillin concentration is presented in Figure 12. Compared with other scatters, the red asterisk scatters that correspond to GMM-LSSVR are more compactly distributed in the diagonal line, which shows that the proposed method can further improve the tracking performance and regression accuracy of the soft sensor. It can deliver reliable and accurate estimation of quality variable despite the presence of noise.
A smart soft sensor based on ensemble LSSVR models is proposed to deal with nonlinear, time-varying, and multiphase characteristics in batch processes. First, MPCA method is applied to be an effective tool for data unfolding and dimensionality reduction. Then, the new obtained dataset can be partitioned into several local regions, where local LSSVR models are constructed. Second, local LSSVR models are constructed for each operation period, respectively. Meanwhile, GS method and ten-fold cross-validation procedure are introduced to local model parameter determination. In this way, each local LSSVR model with a pair of optimal parameters can provide superior regression accuracy. Finally, an ensemble regression model is established by combining different local models by Bayesian fusion strategy and we can obtain the final prediction for test samples from ensemble LSSVR model online. Detailed analyses and comparative studies for penicillin fermentation process show that the proposed soft sensor is feasible and can deliver reliable and accurate quality prediction. In addition, we may be able to improve our future work for soft sensor development by applying cellular neural network approach.
This work was supported by the National Natural Science Foundation of China (Grant No.61773182), and the Subtopics of National Key Research and Development Program of China (Grant No.2018YFC1603705-03).
The authors declare no conflict of interest in this paper.
| [1] |
Lindsey AT, Freddie B, Rebecca LS, et al. (2015) Global cancer statistics, 2012. CA Cancer J Clin 65: 87–108. doi: 10.3322/caac.21262
|
| [2] | Kaprina AD, Starinskiy VV, Petrova GB. (2018) Malignant neoplasms in Russia in 2016, Moscow : P.Herzen Moscow Oncology Research Institute publishing house. |
| [3] | Ostrom QT, Gittleman H, Xu J, et al. (2017) CBTRUS Statistical Report: Primary Brain and Other Central Nervous System Tumors Diagnosed in the United States in 2010–2014. Neuro Onco 18:iv1–iv89. |
| [4] |
Ferlay J, Soerjomataram I, Ervik M, et al. (2015) Cancer incidence and mortality worldwide: sources, methods and major patterns in GLOBOCAN 2012. Int J Cancer 136: E359–E386. doi: 10.1002/ijc.29210
|
| [5] |
Ostrom QT, Gittleman H, Xu J, et al. (2016) CBTRUS Statistical Report: Primary Brain and Other Central Nervous System Tumors Diagnosed in the United States in 2009–2013. Neuro Oncol 18: v1–v75. doi: 10.1093/neuonc/now207
|
| [6] |
Claes A, Idema AJ, Wesseling P. (2007) Diffuse glioma growth: a guerilla war. Acta Neuropathol 114: 443–458. doi: 10.1007/s00401-007-0293-7
|
| [7] |
Louis DN, Ohgaki H, Wiestler OD, et al. (2007) The 2007 WHO Classification of Tumours of the Central Nervous System. Acta Neuropathol 114: 97–109. doi: 10.1007/s00401-007-0243-4
|
| [8] |
Ostrom QT, Gittleman H, Fulop J, et al. (2015) CBTRUS statistical report: primary brain and central nervous system tumors diagnosed in the United States in 2008–2012. Neuro- Oncology 17: iv1–iv62. doi: 10.1093/neuonc/nov189
|
| [9] |
Stupp R, Mason WP, van den Bent MJ, et al. (2005) Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma. N Engl J Med 352: 987–996. doi: 10.1056/NEJMoa043330
|
| [10] |
Roy S, Lahiri D, Maji T, et al. (2015) Recurrent Glioblastoma: Where we stand. South Asian J Cancer 4: 163–173. doi: 10.4103/2278-330X.175953
|
| [11] |
Suzuki H, Aoki K, Chiba K, et al. (2015) Mutational landscape and clonal architecture in grade II and III gliomas. Nat Genet. 47: 458–468. doi: 10.1038/ng.3273
|
| [12] |
Schomas DA, Laack NN, Rao RD, et al. (2009) Intracranial low-grade gliomas in adults: 30-year experience with long-term follow-up at Mayo Clinic. Neuro-Oncology 11: 437–445. doi: 10.1215/15228517-2008-102
|
| [13] |
Darlix A, Zouaoui S, Virion JM, et al. (2014) Significant heterogeneity in the geographical distribution of diffuse grade II/III gliomas in France. J Neuro-Oncology 120: 547–555. doi: 10.1007/s11060-014-1585-0
|
| [14] | Tseng MY, Tseng JH, Merchant E. (2006) Comparison of effects of socioeconomic and geographic variations on survival for adults and children with glioma. J Neurosurg 105: 297–305. |
| [15] |
Frosina G. (2011) Frontiers in targeting glioma stem cells. Eur J Cancer 47: 496–507. doi: 10.1016/j.ejca.2010.11.017
|
| [16] | Gasch С, Ffrench B, O'Leary JJ, et al. Catching moving targets: cancer stem cell hierarchies, therapy-resistance & considerations for clinical intervention. Mol Cancer 16: 1–15. |
| [17] | Clarke J, Butowski N, Chang S. (2010) Recent advances in therapy for glioblastoma. Arch Neurol 67: 279–283. |
| [18] |
Lenting K, Verhaak R, Ter Laan M, et al. (2017) Glioma: experimental models and reality. Acta Neuropathol 133: 263–282. doi: 10.1007/s00401-017-1671-4
|
| [19] |
Meric-Bernstam F, Mills GB. (2012) Overcoming implementation challenges of personalized cancer therapy. Nat Rev Clin Oncol 9: 542–548. doi: 10.1038/nrclinonc.2012.127
|
| [20] |
Fialkow PJ. (1979) Clonal origin of human tumors. Annu Rev Med 30: 135–143. doi: 10.1146/annurev.me.30.020179.001031
|
| [21] |
Rabkin CS, Janz S, Lash A, et al. (1997) Monoclonal origin of multicentric Kaposi's sarcoma lesions. N Engl J Med 336: 988–993. doi: 10.1056/NEJM199704033361403
|
| [22] |
Yachida S, Jones S, Bozic I, et al. (2010) Distant metastasis occurs late during the genetic evolution of pancreatic cancer. Nature 467: 1114–1117. doi: 10.1038/nature09515
|
| [23] |
Al-Hajj M, Wicha MS, Benito-Hernandez A, et al. (2003) Prospective identification of tumorigenic breast cancer cells. Proc Natl Acad Sci U.S.A 100: 3983–3988. doi: 10.1073/pnas.0530291100
|
| [24] |
Wicha MS, Liu S, Dontu G. (2006) Cancer stem cells: an old idea--a paradigm shift. Cancer Res 66: 1883–1890. doi: 10.1158/0008-5472.CAN-05-3153
|
| [25] | Chen W, Dong J, Haiech J, et al. (2016) Cancer Stem Cell Quiescence and Plasticity as Major Challenges in Cancer Therapy. Stem Cells Int 2016:1740936. |
| [26] |
Valent P, Bonnet D, De Maria R, et al. (2012) Cancer stem cell definitions and terminology: the devil is in the details. Nat Rev Cancer. 12: 767–775. doi: 10.1038/nrc3368
|
| [27] |
Chen J, Li Y, Yu TS, et al. (2012) A restricted cell population propagates glioblastoma growth after chemotherapy. Nature 488: 522–526. doi: 10.1038/nature11287
|
| [28] | Liebelt BD, Shingu T, Zhou X, et al.(2016) Glioma Stem Cells: Signaling, Microenvironment, and Therapy. Stem Cells Int 2016: 7849890. |
| [29] | Fidoamore A, Cristiano L, Antonosante A, et al. (2016) Glioblastoma Stem Cells Microenvironment: The Paracrine Roles of the Niche in Drug and Radioresistance. Stem Cells Int 2016: 6809105. |
| [30] |
Michor F, Polyak K. (2010) The origins and implications of intratumor heterogeneity. Cancer Prev Res (Phila) 3: 1361–1364. doi: 10.1158/1940-6207.CAPR-10-0234
|
| [31] | Gerdes MJ, Sood A, Sevinsky C, et al. (2014) Emerging understanding of multiscale tumor heterogeneity. Front Oncol 4: 366. |
| [32] |
Lathia JD, Mack SC, Mulkearns-Hubert EE. (2015) Cancer stem cells in glioblastoma. Genes Dev 29: 1203–1217. doi: 10.1101/gad.261982.115
|
| [33] |
Anido J, Saez-Borderıas A, Gonzalez-Junca A, et al. (2010) TGF-β receptor inhibitors target theCD44high/Id1 high glioma-initiating cell population in human glioblastoma. Cancer Cell 18: 655–668. doi: 10.1016/j.ccr.2010.10.023
|
| [34] |
Lathia JD, Gallagher J, Heddleston JM, et al. (2010) Integrin alpha 6 regulates glioblastoma stem cells. Cell Stem Cell 6: 421–432. doi: 10.1016/j.stem.2010.02.018
|
| [35] |
Thon N, Damianoff K, Hegermann J, et al. (2010) Presence of pluripotent CD133+ cells correlates with malignancy of gliomas. Mol Cell Neurosci 43: 51–59. doi: 10.1016/j.mcn.2008.07.022
|
| [36] |
Bexell D, Gunnarsson S, Siesjo P, et al. (2009) CD133+ and nestin+ tumor-initiating cells dominate in N29 andN32 experimental gliomas. Int J Cancer 125: 15–22. doi: 10.1002/ijc.24306
|
| [37] |
Mathieu J, Zhang Z, Zhou W, et al. (2011) HIF induces human embryonic stem cell markers in cancer cells. Cancer Res 71: 4640–4652. doi: 10.1158/0008-5472.CAN-10-3320
|
| [38] |
Ikushima H, Todo T, Ino Y, et al. (2011) Glioma-initiating cells retain their tumorigenicity through integration of the Sox axis and Oct4 protein. J Biol Chem 286: 41434–41441. doi: 10.1074/jbc.M111.300863
|
| [39] |
Hagerstrand D, He X, Bradic Lindh M, et al. (2011) Identification of a SOX2-dependent subset of tumor- and sphere-forming glioblastoma cells with a distinct tyrosine kinase inhibitor sensitivity profile. Neuro-Oncology 13: 1178–1191. doi: 10.1093/neuonc/nor113
|
| [40] |
Ahlenius H, Kokaia Z. (2010) Isolation and generation of neurosphere cultures from embryonic and adult mouse brain. Methods Mol Biol 633: 241–252. doi: 10.1007/978-1-59745-019-5_18
|
| [41] |
Galli R. (2013) The neurosphere assay applied to neural stem cells and cancer stem cells. Methods Mol Biol. 986: 267–277. doi: 10.1007/978-1-62703-311-4_17
|
| [42] |
Rahman M, Reyner K, Deleyrolle L, et al. (2015) Neurosphere and adherent culture conditions are equivalent for malignant glioma stem cell lines. Anatomy Cell Biol 48: 25–35. doi: 10.5115/acb.2015.48.1.25
|
| [43] |
Pastrana E, Silva-Vargas V, Doetsch F. (2011) Eyes wide open: a critical review of sphere-formation as an assay for stem cells. Cell Stem Cell 8: 486–498. doi: 10.1016/j.stem.2011.04.007
|
| [44] |
Patel AP, Tirosh I, Trombetta JJ, et al. (2014) Single-cell RNA-seq highlights intratumoral heterogeneity in primary glioblastoma. Science 344: 1396–1401. doi: 10.1126/science.1254257
|
| [45] | Venteicher AS, Tirosh I, Hebert C, et al. (2017) Decoupling genetics, lineages, and microenvironment in IDH-mutant gliomas by single-cell RNA-seq. Science 355: 1391–1403. |
| [46] |
Barker N, Bartfeld S, Clevers H. (2010) Tissue-resident adult stem cell populations of rapidly self-renewing organs. Cell Stem Cell 7: 656–670. doi: 10.1016/j.stem.2010.11.016
|
| [47] |
Bao S, Wu Q, McLendon RE, et al. (2006) Glioma stem cells promote radioresistance by preferential activation of the DNA damage response. Nature 444: 756–760. doi: 10.1038/nature05236
|
| [48] |
Perazzoli G, Prados J, Ortiz R, et al. (2015) Temozolomide Resistance in Glioblastoma Cell Lines: Implication of MGMT, MMR, P-Glycoprotein and CD133 Expression. PLoS One. 10: e0140131. doi: 10.1371/journal.pone.0140131
|
| [49] |
Paik JH, Ding Z, Narurkar R, et al. (2009) FoxOs cooperatively regulate diverse pathways governing neural stem cell homeostasis. Cell Stem Cell 5: 540–553. doi: 10.1016/j.stem.2009.09.013
|
| [50] |
Martynoga B, Mateo JL, Zhou B, et al. (2013) Epigenomic enhancer annotation reveals a key role for NFIX in neural stem cell quiescence. Genes Dev 27: 1769–1786. doi: 10.1101/gad.216804.113
|
| [51] |
Louis DN, Perry A, Reifenberger G, et al. (2016) The 2016 World Health Organization classification of tumors of the central nervous system: a summary. Acta Neuropathol 131: 803–820. doi: 10.1007/s00401-016-1545-1
|
| [52] |
Brat DJ, Verhaak RG, Aldape KD, et al. (2015) Comprehensive, integrative genomic analysis of diffuse lower-grade gliomas. N Engl J Med 372: 2481–2498. doi: 10.1056/NEJMoa1402121
|
| [53] |
Chi AS, Batchelor TT, Yang D, et al. (2013) BRAF V600E mutation identifies a subset of low-grade diffusely infiltrating gliomas in adults. J Clin Oncol 31: e233–236. doi: 10.1200/JCO.2012.46.0220
|
| [54] |
Suzuki Y, Takahashi-Fujigasaki J, Akasaki Y, et al. (2016) BRAF V600E-mutated diffuse glioma in an adult patient: a case report and review. Brain Tumor Pathol 33: 40–49. doi: 10.1007/s10014-015-0234-4
|
| [55] | Leeper HE, Caron AA, Decker PA, et al. (2015) IDH mutation, 1p19q codeletion and ATRX loss in WHO grade II gliomas. Oncotarget 6: 30295–30305. |
| [56] |
Weller M, Weber RG, Willscher E, et al. (2015) Molecular classification of diffuse cerebral WHO grade II/III gliomas using genome- and transcriptome-wide profiling improves stratification of prognostically distinct patient groups. Acta Neuropathol 129: 679–693. doi: 10.1007/s00401-015-1409-0
|
| [57] |
Eckel-Passow JE, Lachance DH, Molinaro AM, et al. (2015) Glioma groups based on 1p/19q, IDH, and TERT promoter mutations in tumors. N Engl J Med 372: 2499–2508. doi: 10.1056/NEJMoa1407279
|
| [58] |
Cloughesy TF, Cavenee WK, Mischel PS. (2014) Glioblastoma: from molecular pathology to targeted treatment. Annu Rev Pathol 9: 1–25. doi: 10.1146/annurev-pathol-011110-130324
|
| [59] |
Liu XY, Gerges N, Korshunov A, et al. (2012) Frequent ATRX mutations and loss of expression in adult diffuse astrocytic tumors carrying IDH1/IDH2 and TP53 mutations. Acta Neuropathol 124: 615–625. doi: 10.1007/s00401-012-1031-3
|
| [60] |
Ohgaki H, Kleihues P. (2013) The definition of primary and secondary glioblastoma. Clin Cancer Res 19: 764–772. doi: 10.1158/1078-0432.CCR-12-3002
|
| [61] |
Egan KM, Thompson RC, Nabors L, et al. (2011) Cancer susceptibility variants and the risk of adult glioma in a US case–control study. J Neuro-Oncol 104: 535–542. doi: 10.1007/s11060-010-0506-0
|
| [62] | Jenkins RB, Xiao Y, Sicotte H, et al. (2012) A low-frequency variant at 8q24. 21 is strongly associated with risk of oligodendroglial tumors and astrocytomas with IDH1 or IDH2 mutation. Nat Genet 44: 1122–1125. |
| [63] |
Kinnersley B, Labussiere M, Holroyd A, et al. (2015) Genome-wide association study identifies multiple susceptibility loci for glioma. Nat Commun 6:8559. doi: 10.1038/ncomms9559
|
| [64] |
Rice T, Zheng S, Decker PA, et al. (2013) Inherited variant on chromosome 11q23 increases susceptibility to IDH-mutated but not IDH-normal gliomas regardless of grade or histology. Neuro-Oncology 15: 535–541. doi: 10.1093/neuonc/nos324
|
| [65] |
Wrensch M, Jenkins RB, Chang JS, et al. (2009) Variants in the CDKN2B and RTEL1 regions are associated with high-grade glioma susceptibility. Nat Genet 41: 905–908. doi: 10.1038/ng.408
|
| [66] |
Shete S, Hosking FJ, Robertson LB, et al. (2009) Genome-wide association study identifies five susceptibility loci for glioma. Nat Genet 41: 899–904. doi: 10.1038/ng.407
|
| [67] |
Walsh KM, Codd V, Smirnov IV, et al. (2014) Variants near TERT and TERC influencing telomere length are associated with high-grade glioma risk. Nat Genet 46: 731–735. doi: 10.1038/ng.3004
|
| [68] |
Lu C, Ward PS, Kapoor GS, et al. (2012) IDH mutation impairs histone demethylation and results in a block to cell differentiation. Nature 483: 474–478. doi: 10.1038/nature10860
|
| [69] |
Sundarraj N, Schachner M, Pfeiffer SE. (1975) Biochemically differentiated mouse glial lines carrying a nervous system specific cell surface antigen (NS-1). Proc Natl Acad Sci U.S.A 72: 1927–1931. doi: 10.1073/pnas.72.5.1927
|
| [70] | Dawson G, Sundarraj N, Pfeiffer SE. (1977) Synthesis of myelin glycosphingolipids (galactosylceramide and galactosyl (3-O-sulfate) ceramide (sulfatide) by cloned cell lines derived from mouse neurotumors. J Biol Chem 252: 2777–2779. |
| [71] |
Fields KL, Gosling C, Megson M, et al. (1975) New cell surface antigens in rat defined by tumors of the nervous system. Proc Natl Acad Sci U.S.A 72: 1296–1300. doi: 10.1073/pnas.72.4.1296
|
| [72] | Allen M, Bjerke M, Edlund H, et al. (2016) Origin of the U87MG glioma cell line: Good news and bad news. Sci Transl Med 8: 354–353. |
| [73] |
Torsvik A, Stieber D, Enger PO, et al. (2014) U-251 revisited: genetic drift and phenotypic consequences of long-term cultures of glioblastoma cells. Cancer Med 3: 812–824. doi: 10.1002/cam4.219
|
| [74] |
De Vries GH, Boullerne AI. (2010) Glial cell lines: an overview. Neurochem Res 35: 1978–2000. doi: 10.1007/s11064-010-0318-9
|
| [75] |
Louis JC, Magal E, Muir D, et al. (1992) CG-4, a new bipotential glial cell line from rat brain, is capable of differentiating in vitro into either mature oligodendrocytes or type-2 astrocytes. J Neurosci Res 31: 193–204. doi: 10.1002/jnr.490310125
|
| [76] |
Richter-Landsberg C, Heinrich M. (1996) OLN-93: a new permanent oligodendroglia cell line derived from primary rat brain glial cultures. J Neurosci Res 45: 161–173. doi: 10.1002/(SICI)1097-4547(19960715)45:2<161::AID-JNR8>3.0.CO;2-8
|
| [77] |
Jung M, Kramer E, Grzenkowski M, et al. (1995) Lines of murine oligodendroglial precursor cells immortalized by an activated neu tyrosine kinase show distinct degrees of interaction with axons in vitro and in vivo. Eur J Neurosci 7: 1245–1265. doi: 10.1111/j.1460-9568.1995.tb01115.x
|
| [78] |
Foster LM, Phan T, Verity AN, et al. (1992) Generation and analysis of normal and shiverer temperature-sensitive immortalized cell lines exhibiting phenotypic characteristics of oligodendrocytes at several stages of differentiation. J Neurosci Res 31: 193–204. doi: 10.1002/jnr.490310125
|
| [79] |
Post GR, Dawson G. (1992) Characterization of a cell line derived from a human oligodendroglioma. Mol Chem Neuropathol 16: 303–317. doi: 10.1007/BF03159976
|
| [80] |
McLaurin J, Trudel GC, Shaw IT, et al. (1995) A human glial hybrid cell line differentially expressing genes subserving oligodendrocyte and astrocyte phenotype. J Neurobiol 26: 283–293. doi: 10.1002/neu.480260212
|
| [81] |
Benda P, Lightbody J, Sato G, et al. (1968) Differentiated rat glial cell strain in tissue culture. Science 161: 370–371. doi: 10.1126/science.161.3839.370
|
| [82] |
Radany EH, Brenner M, Besnard F, et al. (1992) Directed establishment of rat brain cell lines with the phenotypic characteristics of type 1 astrocytes. Proc Natl Acad Sci U.S.A. 89: 6467–6471. doi: 10.1073/pnas.89.14.6467
|
| [83] |
Loo DT, Fuquay JI, Rawson CL, et al. (1987) Extended culture of mouse embryo cells without senescence: inhibition by serum. Science 236: 200–202. doi: 10.1126/science.3494308
|
| [84] |
Giard DJ, Aaronson SA, Todaro GJ, et al. (1973) In vitro cultivation of human tumors: establishment of cell lines derived from a series of solid tumors. J Natl Cancer Inst 51: 1417–1423. doi: 10.1093/jnci/51.5.1417
|
| [85] | Pontén J, Macintyre EH. (1968) Long term culture of normal and neoplastic human glia. Acta Pathol Microbiol Scand 74: 465–486. |
| [86] | Westermark B, Ponten J, Hugosson R. (1973) Determinants for the establishment of permanent tissue culture lines from human gliomas. Acta Pathol Microbiol Scand A 81: 791–805. |
| [87] |
Lee J, Kotliarova S, Kotliarov Y, et al. (2006) Tumor stem cells derived from glioblastomas cultured in bFGF and EGF more closely mirror the phenotype and genotype of primary tumors than do serum-cultured cell lines. Cancer Cell 9: 391–403. doi: 10.1016/j.ccr.2006.03.030
|
| [88] |
Luca AC, Mersch S, Deenen R, et al. (2013) Impact of the 3D Microenvironment on Phenotype, Gene Expression, and EGFR Inhibition of Colorectal Cancer Cell Lines. PloS One 8: e59689. doi: 10.1371/journal.pone.0059689
|
| [89] |
Storch K, Eke I, Borgmann K, et al. (2010) Three-dimensional cell growth confers radioresistance by chromatin density modification. Cancer Res 70: 3925–3934. doi: 10.1158/0008-5472.CAN-09-3848
|
| [90] |
Hehlgans S, Lange I, Eke I, et al. (2009) 3D cell cultures of human head and neck squamous cell carcinoma cells are radiosensitized by the focal adhesion kinase inhibitor TAE226. Radiotherapy oncology: j Eu Soc Therapeutic Radiology Oncology 92:371–378. doi: 10.1016/j.radonc.2009.08.001
|
| [91] | Bristow RG, Hill RP. (2008) Hypoxia and metabolism. Hypoxia, DNA repair and genetic instability. Nat Rev. Cancer 8: 180–192. |
| [92] | Gomez-Roman N, Stevenson K, Gilmour L, et al. (2017) A novel 3D human glioblastoma cell culture system for modeling drug and radiation responses. Neuro Oncol 19:229–241. |
| [93] |
Mullins CS, Schneider B, Stockhammer F, et al. (2013) Establishment and characterization of primary glioblastoma cell lines from fresh and frozen material: a detailed comparison. PLoS One 8: e71070. doi: 10.1371/journal.pone.0071070
|
| [94] |
Cheng L, Huang Z, Zhou W, et al. (2013) Glioblastoma stem cells generate vascular pericytes to support vessel function and tumor growth. Cell 153: 139–152. doi: 10.1016/j.cell.2013.02.021
|
| [95] |
Soda Y, Marumoto T, Friedmann-Morvinski D, et al. (2011) Transdifferentiation of glioblastoma cells into vascular endothelial cells. Proc Natl Acad Sci U.S.A 108: 4274–4280. doi: 10.1073/pnas.1016030108
|
| [96] |
Wang R, Chadalavada K, Wilshire J, et al. (2010) Glioblastoma stem-like cells give rise to tumour endothelium. Nature 468: 829–833. doi: 10.1038/nature09624
|
| [97] |
Singec I, Knoth R, Meyer RP, et al. (2006) Defining the actual sensitivity and specificity of the neurosphere assay in stem cell biology. Nat Methods 3: 801–806. doi: 10.1038/nmeth926
|
| [98] |
Gritti A, Galli R, Vescovi AL. (2008) Clonal analyses and cryopreservation of neural stem cell cultures. Methods Mol Biol 438: 173–184. doi: 10.1007/978-1-59745-133-8_14
|
| [99] |
Brewer GJ, Torricelli JR, Evege EK, et al. (1993) Optimized survival of hippocampal neurons in B27-supplemented Neurobasal, a new serum-free medium combination. J Neurosci Res 35: 567–576. doi: 10.1002/jnr.490350513
|
| [100] |
Luchman HA, Stechishin OD, Dang NH, et al. (2012) An in vivo patient-derived model of endogenous IDH1-mutant glioma. Neuro Oncol 14: 184–191. doi: 10.1093/neuonc/nor207
|
| [101] |
Rohle D, Popovici-Muller J, Palaskas N, et al. (2013) An inhibitor of mutant IDH1 delays growth and promotes differentiation of glioma cells. Science 340: 626–630. doi: 10.1126/science.1236062
|
| [102] |
Alcantara LSR, Wang Z, Sun D, et al. (2015) Adult Lineage-Restricted CNS Progenitors Specify Distinct Glioblastoma Subtypes. Cancer Cell 28: 429–440. doi: 10.1016/j.ccell.2015.09.007
|
| [103] | Lindberg N, Jiang Y, Xie Y, et al. (2014) Oncogenic signaling is dominant to cell of origin and dictates astrocytic or oligodendroglial tumor development from oligodendrocyte precursor cells. J Neuro Sci 34: 14644–14651. |
| [104] | Wang J, Wakeman TP, Lathia JD, et al. (2010) Notch promotes radioresistance of glioma stem cells. Stem Cells 28: 17–28. |
| [105] |
Adorno-Cruz V, Kibria G, Liu X, et al. (2015) Cancer stem cells: targeting the roots of cancer, seeds of metastasis, and sources of therapy resistance. Cancer Res 75: 924–929. doi: 10.1158/0008-5472.CAN-14-3225
|
| [106] |
Shlush LI, Mitchell A, Heisler L, et al. (2017) Tracing the origins of relapse in acute myeloid leukaemia to stem cells. Nature 547: 104–108. doi: 10.1038/nature22993
|
| [107] |
Stopschinski BE, Beier CP, Beier D. (2013) Glioblastoma cancer stem cells--from concept to clinical application. Cancer Lett 338: 32–40. doi: 10.1016/j.canlet.2012.05.033
|
| [108] |
Holmberg Olausson K, Maire CL, Haidar S, et al. (2014) Prominin-1 (CD133) defines both stem and non-stem cell populations in CNS development and gliomas. PLoS One 9: e106694. doi: 10.1371/journal.pone.0106694
|
| [109] |
Wee B, Charles N, Holland EC. (2011) Animal models to study cancer-initiating cells from glioblastoma. Front Bio Sci (Landmark Ed) 16: 2243–2258. doi: 10.2741/3851
|
| [110] |
Mack SC, Hubert CG, Miller TE, et al. (2016) An epigenetic gateway to brain tumor cell identity. Nat Neuro Sci 19: 10–19. doi: 10.1038/nn.4190
|
| [111] |
Irvin DM, McNeill RS, Bash RE, et al. (2017) Intrinsic Astrocyte Heterogeneity Influences Tumor Growth in Glioma Mouse Models. Brain Pathol 27: 36–50. doi: 10.1111/bpa.12348
|
| [112] |
Chen W, Wang D, Du X, et al. (2015) Glioma cells escaped from cytotoxicity of temozolomide and vincristine by communicating with human astrocytes. Med Oncol 32:43 . doi: 10.1007/s12032-015-0487-0
|
| [113] | Chen W, Xia T, Wang D, et al. (2016) Human astrocytes secrete IL-6 to promote glioma migration and invasion through upregulation of cytomembrane MMP14. Oncotarget 7: 62425–62438. |
| [114] |
Graeber MB, Scheithauer BW, Kreutzberg GW. (2002) Microglia in brain tumors. Glia 40: 252–259. doi: 10.1002/glia.10147
|
| [115] |
Watters JJ, Schartner JM, Badie B. (2005) Microglia function in brain tumors. J Neurosci Res 81: 447–455. doi: 10.1002/jnr.20485
|
| [116] |
Hambardzumyan D, Gutmann DH, Kettenmann H. (2016) The role of microglia and macrophages in glioma maintenance and progression. Nat Neurosci 19: 20–27. doi: 10.1038/nn.4185
|
| [117] |
Brooks WH, Markesbery WR, Gupta GD, et al. (1978) Relationship of lymphocyte invasion and survival of brain tumor patients. Ann Neurol 4: 219–24. doi: 10.1002/ana.410040305
|
| [118] |
Hao C, Parney IF, Roa WH, et al. (2002) Cytokine and cytokine receptor mRNA expression in human glioblastomas: evidence of Th1, Th2 and Th3 cytokine dysregulation. Acta Neuropathol 103: 171–178.. doi: 10.1007/s004010100448
|
| [119] |
Rodrigues JC, Gonzalez GC, Zhang L, et al. (2010) Normal human monocytes exposed to glioma cells acquire myeloid-derived suppressor cell-like properties. Neuro Oncol 12: 351–365. doi: 10.1093/neuonc/nop023
|
| [120] | Jackson C, Ruzevick J, Phallen J, et al. (2011) Challenges in immunotherapy presented by the glioblastoma multiforme microenvironment. Clin Dev Immunol 2011: 732413 . |
| [121] | Wischhusen J, Jung G, Radovanovic I, et al. (2002) Identification of CD70-mediated apoptosis of immune effector cells as a novel immune escape pathway of human glioblastoma. Cancer Res 62: 2592–2599. |
| [122] |
Chahlavi A, Rayman P, Richmond AL, et al. (2005) Glioblastomas induce T-lymphocyte death by two distinct pathways involving gangliosides and CD70. Cancer Res 65: 5428–5438. doi: 10.1158/0008-5472.CAN-04-4395
|
| [123] |
Charles NA, Holland EC, Gilbertson R, et al. (2012) The brain tumor microenvironment. Glia 60: 502–514. doi: 10.1002/glia.21264
|
| [124] |
Wen PY, Kesari S. (2008) Malignant gliomas in adults. N Engl J Med 359: 492–507. doi: 10.1056/NEJMra0708126
|
| [125] |
Calabrese C, Poppleton H, Kocak M, et al. (2007) A perivascular niche for brain tumor stem cells. Cancer Cell 11: 69–82. doi: 10.1016/j.ccr.2006.11.020
|
| [126] |
Charles N, Ozawa T, Squatrito M, et al. (2010) Perivascular nitric oxide activates notch signaling and promotes stem-like character in PDGF-induced glioma cells. Cell Stem Cell 6: 141–152. doi: 10.1016/j.stem.2010.01.001
|
| [127] |
Hambardzumyan D, Becher OJ, Rosenblum MK, et al. (2008) PI3K pathway regulates survival of cancer stem cells residing in the perivascular niche following radiation in medulloblastoma in vivo. Genes Dev 22: 436–448. doi: 10.1101/gad.1627008
|
| [128] |
Shaw KM, Wrobel C, Brugge J. (2004) Use of Three-Dimensional Basement Membrane Cultures to Model Oncogene-Induced Changes in Mammary Epithelial Morphogenesis. J. Mammary Gland Biol 9: 297–310. doi: 10.1007/s10911-004-1402-z
|
| [129] |
Trédan O, Galmarini CM, Patel K, et al. (2007) Drug Resistance and the Solid Tumor Microenvironment. J Natl. Cancer Inst 99: 1441–1454. doi: 10.1093/jnci/djm135
|
| [130] |
Caliari SR, Burdick JA. (2016) A practical guide to hydrogels for cell culture. Nat Methods 13: 405–414. doi: 10.1038/nmeth.3839
|
| [131] |
Ahmed EM. (2015) Hydrogel: Preparation, characterization, and applications: A review. J Adv Res 6: 105–121. doi: 10.1016/j.jare.2013.07.006
|
| [132] |
Dawson E, Mapili G, Erickson K, et al. (2008) Biomaterials for stem cell differentiation. Adv Drug Deliv Rev 60: 215–228. doi: 10.1016/j.addr.2007.08.037
|
| [133] |
Kiefer JA, Farach-Carson MC. (2001) Type I collagen-mediated proliferation of PC3 prostate carcinoma cell line: Implications for enhanced growth in the bone microenvironment. Matrix Bio 20: 429–437. doi: 10.1016/S0945-053X(01)00159-7
|
| [134] | Menke A, Philippi C, Vogelmann R, et al. (2001) Down-Regulation of E-Cadherin Gene Expression by Collagen Type I and Type III in Pancreatic Cancer Cell Lines. Cancer Res 61: 3508–3517. |
| [135] |
Kim YJ, Bae HI, Kwon OK, et al. (2009) Three-dimensional gastric cancer cell culture using nanofiber scaffold for chemosensitivity test. Int J Biol Macromol 45: 65–71. doi: 10.1016/j.ijbiomac.2009.04.003
|
| [136] |
Sun W, Incitti T, Migliaresi C, et al. (2017) Viability and neuronal differentiation of neural stem cells encapsulated in silk fibroin hydrogel functionalized with an IKVAV peptide. J Tissue Eng Regen Med 11: 1532–1541. doi: 10.1002/term.2053
|
| [137] |
Musah S, Morin SA, Wrighton PJ, et al. (2012) Glycosaminoglycan-binding hydrogels enable mechanical control of human pluripotent stem cell self-renewal. ACS Nano 6: 10168–10177. doi: 10.1021/nn3039148
|
| [138] |
Souza GR, Molina JR, Raphael RM, et al. (2010) Three-dimensional tissue culture based on magnetic cell levitation. Nat. Nanotechnol 5: 291–296. doi: 10.1038/nnano.2010.23
|
| [139] |
Carpenter PM, Dao AV, Arain ZS, et al. (2009) Motility induction in breast carcinoma by mammary epithelial laminin 332 (laminin 5). Mol. Cancer Res 7: 462–475. doi: 10.1158/1541-7786.MCR-08-0148
|
| [140] |
Zhou Z, Wang J, Cao R, et al. (2004) Impaired Angiogenesis, Delayed Wound Healing and Retarded Tumor Growth in Perlecan Heparan Sulfate-Deficient Mice. Cancer Res 64: 4699–4702. doi: 10.1158/0008-5472.CAN-04-0810
|
| [141] |
Miyamoto H, Murakami T, Tsuchida K, et al. (2004) Tumor‐stroma interaction of human pancreatic cancer: acquired resistance to anticancer drugs and proliferation regulation is dependent on extracellular matrix proteins. Pancreas 28: 38–44. doi: 10.1097/00006676-200401000-00006
|
| [142] |
Nguyen TV, Sleiman M, Moriarty T, et al. (2014) Sorafenib resistance and JNK signaling in carcinoma during extracellular matrix stiffening. Biomaterials 35: 5749–5759. doi: 10.1016/j.biomaterials.2014.03.058
|
| [143] |
Vinci M, Gowan S, Boxall F, et al. (2012) Advances in establishment and analysis of three-dimensional tumor spheroid-based functional assays for target validation and drug evaluation. BMC Biol 10: 29–49. doi: 10.1186/1741-7007-10-29
|
| [144] |
Rybtsov S, Batsivari A, Bilotkach K, et al. (2014) Tracing the Origin of the HSC Hierarchy Reveals an SCF-Dependent, IL-3-Independent CD43- Embryonic Precursor. Stem Cell Rep 3: 489–501. doi: 10.1016/j.stemcr.2014.07.009
|
| [145] |
Herter S, Morra L, Schlenker R, et al. (2017) A novel three-dimensional heterotypic spheroid model for the assessment of the activity of cancer immunotherapy agents. Cancer Immunol Immunother 66: 129–140. doi: 10.1007/s00262-016-1927-1
|
| [146] |
Hirt C, Papadimitropoulos A, Mele V, et al. (2014) "In vitro" 3D models of tumor-immune system interaction. Adv Drug Deliv Rev 79-80:145–54. doi: 10.1016/j.addr.2014.05.003
|
| [147] | Ma FX, Chen F, Chi Y, et al. (2013) Culture of pancreatic progenitor cells in hanging drop and on floating filter. Zhongguo Yi Xue Ke Xue Yuan Xue Bao 35: 270–274. |
| [148] |
Ingthorsson S., Sigurdsson V., Fridriksdottir JR A. (2010) Endothelial cells stimulate growth of normal and cancerous breast epithelial cells in 3D culture. BMC Res 3: 184–195. doi: 10.1186/1756-0500-3-184
|
| [149] |
Li L, Lu YJ. (2011) Optimizing a 3D Culture System to Study the Interaction between Epithelial Breast Cancer and Its Surrounding Fibroblasts. Cancer 2: 458–466. doi: 10.7150/jca.2.458
|
| [150] |
Touboul C, Raphael L, Farsi H AI, et al. (2013) Mesenchymal stem cells enhance ovarian cancer cell infiltration through IL6 secretion in an amniochorionic membrane based 3D model. J Transl. Med 11: 28–39. doi: 10.1186/1479-5876-11-28
|
| [151] | Li L, Fukunaga-Kalabis M, Herlyn M. (2011) The three-dimensional human skin reconstruct model: a tool to study normal skin and melanoma progression. J Vis Exp 54: e2937. |
| [152] |
Vörsmann H, Groeber F, Walles H, et al. (2013) Development of a human three-dimensional organotypic skin-melanoma spheroid model for in vitro drug testing. Cell Death Dis 4: e719. doi: 10.1038/cddis.2013.249
|
| [153] |
Carmeliet P, Jain RK. (2000) Angiogenesis in cancer and other diseases. Nature. 407: 249–257. doi: 10.1038/35025220
|
| [154] |
Upreti M, Jamshidi-Parsian A, Koonce NA, et al. (2011) Tumor-Endothelial Cell Three-dimensional Spheroids: New Aspects to Enhance Radiation and Drug Therapeutics. Transl Oncol 4: 365–376. doi: 10.1593/tlo.11187
|
| [155] |
Ramgolam K, Lauriol J, Lalou C, et al. (2011) Melanoma spheroids grown under neural crest cell conditions are highly plastic migratory/invasive tumor cells endowed with immunomodulator function. PLoS One 6: e18784. doi: 10.1371/journal.pone.0018784
|
| [156] |
Giannattasio A, Weil S, Kloess S, et al. (2015) Cytotoxicity and infiltration of human NK cells in in vivo-like tumor spheroids. BMC Cancer 15: 351–363. doi: 10.1186/s12885-015-1321-y
|
| [157] | Marchwicka A, Cebrat M, Sampath P, et al. (2014) Perspectives of Differentiation Therapies of Acute Myeloid Leukemia: The Search for the Molecular Basis of Patients' Variable Responses to 1,25-Dihydroxyvitamin D and Vitamin D Analogs. Frontiers in Oncology 4:125–136. |
| [158] |
Caren H, Beck S, Pollard SM. (2016) Differentiation therapy for glioblastoma–too many obstacles? Mol Cellr Oncol 3: e1124174. doi: 10.1080/23723556.2015.1124174
|
| 1. | Mariano M. Perdomo, Carlos I. Sanseverinatti, Luis A. Clementi, Jorge R. Vega, 2022, Sensor Inferencial Multi-modelo Aplicado a un Proceso Simulado para la Producción Continua de Látex para Caucho, 978-1-6654-8014-7, 1, 10.1109/ARGENCON55245.2022.9939849 | |
| 2. | Yan-Ning Sun, Wei Qin, Hong-Wei Xu, Run-Zhi Tan, Zhan-Luo Zhang, Wen-Tian Shi, A multiphase information fusion strategy for data-driven quality prediction of industrial batch processes, 2022, 608, 00200255, 81, 10.1016/j.ins.2022.06.057 | |
| 3. | Nobuhito Yamada, Hiromasa Kaneko, Adaptive soft sensor ensemble for selecting both process variables and dynamics for multiple process states, 2021, 219, 01697439, 104443, 10.1016/j.chemolab.2021.104443 | |
| 4. | Wangwang Zhu, Zhengjiang Zhang, Yi Liu, Dynamic Data Reconciliation for Improving the Prediction Performance of the Data-Driven Model on Distributed Product Outputs, 2022, 61, 0888-5885, 18780, 10.1021/acs.iecr.2c02536 | |
| 5. | Hongyu Tang, Zhenli Yang, Feng Xu, Qi Wang, Bo Wang, Soft Sensor Modeling Method Based on Improved KH-RBF Neural Network Bacteria Concentration in Marine Alkaline Protease Fermentation Process, 2022, 194, 0273-2289, 4530, 10.1007/s12010-022-03934-4 | |
| 6. | Joyce Chen Yen Ngu, Wan Sieng Yeo, 2022, Prediction Of Dissolved Oxygen Using Least Square Support Vector Regression Model, 978-1-6654-8663-7, 70, 10.1109/GECOST55694.2022.10010638 | |
| 7. | Feng Xu, Kaihao Hu, Ali Mohsin, Jie Wu, Lihuan Su, Yuan Wang, Rong Ben, Hao Gao, Xiwei Tian, Ju Chu, Recent advances in the biosynthesis and production optimization of gentamicin: A critical review, 2024, 2405805X, 10.1016/j.synbio.2024.11.003 | |
| 8. | Ling Zhao, Jinlin Zhu, Zheng Zhang, Zhenping Xie, Furong Gao, 2023, A Novel Semi-supervised Two-dimensional Dynamic Soft Sensor for Quality Prediction in Batch Processes, 979-8-3503-2529-4, 1, 10.1109/IAI59504.2023.10327600 | |
| 9. | Jameson Malang, Wan Sieng Yeo, Zhen Yang Chua, Jobrun Nandong, Agus Saptoro, A. Saptoro, R. Nagarajan, J.S.Y. Lau, Y.Y. Tiong, V. Rowtho, C.P. Selvan, A. Tan, C. Koh, A comparison study between different kernel functions in the least square support vector regression model for penicillin fermentation process, 2023, 377, 2261-236X, 01025, 10.1051/matecconf/202337701025 | |
| 10. | Wenlong Li, Xi Wang, Houliu Chen, Xu Yan, Haibin Qu, In-Line Vis-NIR Spectral Analysis for the Column Chromatographic Processes of the Ginkgo biloba L. Leaves. Part II: Batch-to-Batch Consistency Evaluation of the Elution Process, 2022, 9, 2297-8739, 378, 10.3390/separations9110378 | |
| 11. | Wei Zou, Yanxia Shen, Lei Wang, Design of robust fuzzy iterative learning control for nonlinear batch processes, 2023, 20, 1551-0018, 20274, 10.3934/mbe.2023897 | |
| 12. | Na Lu, Bo Wang, Xianglin Zhu, Soft Sensor Modeling Method for the Marine Lysozyme Fermentation Process Based on ISOA-GPR Weighted Ensemble Learning, 2023, 23, 1424-8220, 9119, 10.3390/s23229119 | |
| 13. | Xinmin Zhang, Bocun He, Hongyu Zhu, Zhihuan Song, Information Complementary Fusion Stacked Autoencoders for Soft Sensor Applications in Multimode Industrial Processes, 2024, 20, 1551-3203, 106, 10.1109/TII.2023.3257307 | |
| 14. | Yi Shan Lee, Junghui Chen, A novel reinforced incomplete cyber-physics ensemble with error compensation learning for within-batch quality prediction, 2025, 65, 14740346, 103172, 10.1016/j.aei.2025.103172 |
Valeriia Mikhailova, Valeriia Gulaia, Vladlena Tiasto, Stanislav Rybtsov, Margarita Yatsunskaya, Alexander Kagansky. Towards an advanced cell-based in vitro glioma model system[J]. AIMS Genetics, 2018, 5(2): 91-112. doi: 10.3934/genet.2018.2.91
| NO. | Variable description (Unit) | NO. | Variable description (Unit) |
| 1 | Aeration rate (L/h) | 7 | Carbon dioxide concentration (g/L) |
| 2 | Agitator power (W) | 8 | PH (-) |
| 3 | Substrate feed rate (L/h) | 9 | Fermenter temperature (K) |
| 456 | Substrate feed temperature (K)Dissolved oxygen concentration (g/L)Culture volume (L) | 1011 | Generated heat (kcal)Cooling water flow rate (L/h) |
DownLoad:
CSV
| Method | Batch 5 with no noise | Batch 6 with noise | ||||
| RMSE | TP | R2 | RMSE | TP | R2 | |
| GPR | 0.0101 | 0.9996 | 0.9995 | 0.0206 | 0.9982 | 0.9980 |
| LSSVR | 0.0119 | 0.9994 | 0.9993 | 0.0224 | 0.9981 | 0.9977 |
| GMM-GPR | 0.0094 | 0.9996 | 0.9996 | 0.0177 | 0.9986 | 0.9985 |
| GMM-LSSVR | 0.0039 | 0.9999 | 0.9999 | 0.0125 | 0.9993 | 0.9993 |
DownLoad:
CSV
| NO. | Variable description (Unit) | NO. | Variable description (Unit) |
| 1 | Aeration rate (L/h) | 7 | Carbon dioxide concentration (g/L) |
| 2 | Agitator power (W) | 8 | PH (-) |
| 3 | Substrate feed rate (L/h) | 9 | Fermenter temperature (K) |
| 456 | Substrate feed temperature (K)Dissolved oxygen concentration (g/L)Culture volume (L) | 1011 | Generated heat (kcal)Cooling water flow rate (L/h) |
| Method | Batch 5 with no noise | Batch 6 with noise | ||||
| RMSE | TP | R2 | RMSE | TP | R2 | |
| GPR | 0.0101 | 0.9996 | 0.9995 | 0.0206 | 0.9982 | 0.9980 |
| LSSVR | 0.0119 | 0.9994 | 0.9993 | 0.0224 | 0.9981 | 0.9977 |
| GMM-GPR | 0.0094 | 0.9996 | 0.9996 | 0.0177 | 0.9986 | 0.9985 |
| GMM-LSSVR | 0.0039 | 0.9999 | 0.9999 | 0.0125 | 0.9993 | 0.9993 |
