On generalization error of neural network models and its application to predictive control of nonlinear processes

Alhajeri, Mohammed S.; Alnajdi, Aisha; Abdullah, Fahim; Christofides, Panagiotis D.

doi:10.1016/j.cherd.2022.12.001

Cited by 10 publications

(3 citation statements)

References 56 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Proses ini menunjukkan kemampuan kerja model dalam memprediksi dengan benar pada data baru yang belum pernah dilihat sebelumnya. Kemampuan generalisasi model klasifikasi yang dipelajari pada sampel data training terbatas pada data baru menjadi tantangan yang signifikan [2]. Semakin tinggi generalisasi dari sebuah model maka akan semakin meningkatkan akurasi prediksi.…”

Section: Data Training Dan Data Testingunclassified

Comparison of Naïve Bayes, CART, dan CART Adaboost Methods in Predicting Tire Product Sales

Aprihartha,

Astutik,

Sulistianingsih

2024

View full text Add to dashboard Cite

Data mining is a term to describe the process of moving through large databases in search of certain previously unknown patterns. In finding certain patterns, you need a supporting technique, called machine learning. Machine learning involves learning hidden patterns in data and further using patterns to classify or predict an event related to a problem. One of the problems can be solved with machine learning such as predicting the sales rate of tire products. This can help companies predict tire products that are selling well in the market. In producing an accurate prediction model, it will be compared with decision tree classification methods of CART, CART + Discrete Adaboost, and Naive Bayes applied to tire sales data by PT. Mitra Mekar Mandiri. The results of the study based on successive model performance evaluations are model Naive Bayes < model CART < model CART+Discrete Adaboost. The Discrete Adaboost model with a data proportion of 90:10 is the best model for predicting tire sales. The accuracy, sensitivity and specificity values for the model were 79.17%; 89.47%; and 68.84%. The AUC value is 0.8 which indicates the model is good

show abstract

Section: Data Training Dan Data Testingunclassified

Comparison of Naïve Bayes, CART, dan CART Adaboost Methods in Predicting Tire Product Sales

Aprihartha,

Astutik,

Sulistianingsih

2024

View full text Add to dashboard Cite

show abstract

“…As the prior art suggests, the LSTM can identify slow time-varying parameters while capturing faster system dynamics [6]. Similar architectures have been previously adopted for parameter identification problems [7,8].…”

Section: Introductionmentioning

confidence: 99%

Comparing the EKF with inverse black-box models in system identification of a chemical stirred tank

Fard

García

Kirchner

et al. 2023

2023 International Conference on Control, Automation and Diagnosis (ICCAD)

View full text Add to dashboard Cite

This paper compares the computational cost and accuracy of an extended Kalman filter with inverse black box models in multi-physical system identification of chemical stirred tanks. First, a physics-based coupled electro-thermo-mechanical lumped parameter model is presented with the heat transfer coefficient and stirring torque as unknowns. The developed model has two objectives: it is exploited by the extended Kalman filter to perform system identification and simulates the artificial data to train inverse black box models. The extended Kalman filter and the inverse black box models are compared in numerical and experimental cases in which the results show a good agreement between both approaches and reference values. The results show that introducing random variations to the heat transfer coefficient in simulating artificial data increases the accuracy of the inverse black box model.

show abstract

“…Moreover, the authors of [13] focused on deriving a generalization error bound for a feedforward neural network (FNN) that was used to construct a control Lyapunov-barrier function. The authors of [14] studied the generalization error bound for both a partially connected RNN (PCRNN) and an LSTM-RNN and subsequently carried out a comparison study between the generalization performance of an FCRNN and a PCRNN. Additionally, all the aforementioned works incorporated the designed machine learning models into Lyapunov-based MPC schemes to demonstrate the ability of the LMPCs to stabilize a chemical process.…”

Section: Introductionmentioning

confidence: 99%

Machine Learning-Based Model Predictive Control of Two-Time-Scale Systems

Alnajdi,

Abdullah,

Suryavanshi

et al. 2023

Mathematics

Self Cite

View full text Add to dashboard Cite

In this study, we present a general form of nonlinear two-time-scale systems, where singular perturbation analysis is used to separate the dynamics of the slow and fast subsystems. Machine learning techniques are utilized to approximate the dynamics of both subsystems. Specifically, a recurrent neural network (RNN) and a feedforward neural network (FNN) are used to predict the slow and fast state vectors, respectively. Moreover, we investigate the generalization error bounds for these machine learning models approximating the dynamics of two-time-scale systems. Next, under the assumption that the fast states are asymptotically stable, our focus shifts toward designing a Lyapunov-based model predictive control (LMPC) scheme that exclusively employs the RNN to predict the dynamics of the slow states. Additionally, we derive sufficient conditions to guarantee the closed-loop stability of the system under the sample-and-hold implementation of the controller. A nonlinear chemical process example is used to demonstrate the theory. In particular, two RNN models are constructed: one to model the full two-time-scale system and the other to predict solely the slow state vector. Both models are integrated within the LMPC scheme, and we compare their closed-loop performance while assessing the computational time required to execute the LMPC optimization problem.

show abstract

On generalization error of neural network models and its application to predictive control of nonlinear processes

Cited by 10 publications

References 56 publications

Comparison of Naïve Bayes, CART, dan CART Adaboost Methods in Predicting Tire Product Sales

Comparison of Naïve Bayes, CART, dan CART Adaboost Methods in Predicting Tire Product Sales

Comparing the EKF with inverse black-box models in system identification of a chemical stirred tank

Machine Learning-Based Model Predictive Control of Two-Time-Scale Systems

Contact Info

Product

Resources

About