Журнал «Современная Наука»

Russian (CIS)English (United Kingdom)
 MOSCOW +7(495)-142-86-81

STATISTICAL AND MACHINE LEARNING APPROACHES TO DYNAMIC PROCESS CONTROL BASED ON HYBRID NEURAL NETWORK ARCHITECTURES

Novikov Kirill Sergeevich  (Postgraduate student Penza State University )

Yurkov Nikolay Kondratievich  (Doctor of Technical Sciences Penza State University )

Koshelev Nikita Dmitrievich  (Postgraduate student Penza State University )

Background and Objectives. Controlling complex dynamic systems under uncertainty requires models that combine precision, adaptability, and interpretability. Classical methods rely on predefined equations and constraints but often lose efficiency when observations are incomplete or operating regimes shift. On the other hand, machine learning methods offer flexibility but may fail to provide reliable control. This study aims to develop a hybrid neural network architecture that integrates principles of optimal control and statistical modeling. Materials and methods. The proposed model comprises two primary components: a dynamics prediction module and a trainable control unit. The training process optimizes a joint loss function that includes prediction accuracy and control effort minimization. Evaluation metrics include MSE, MAE, determination coefficient R^2, and deviation from target states. Results. The hybrid model reduced MSE by 51% compared to an LSTM and by 41% relative to the LQR approach. The average deviation from the target state decreased by 32%, while control energy was reduced. Under regime shifts, maximum trajectory deviation was cut by half. Conclusions. The proposed hybrid approach successfully combines the structural reliability of classical control with the flexibility of neural networks. It is applicable to autonomous regulation, predictive control, and system adaptation tasks where both precision and interpretability are critical.

Keywords:hybrid neural networks, dynamic system control, optimal control, statistical modeling, trainable regulators, nonlinear dynamics, adaptive architectures.
 

Read the full article …

Citation link:
Novikov K. S., Yurkov N. K., Koshelev N. D. STATISTICAL AND MACHINE LEARNING APPROACHES TO DYNAMIC PROCESS CONTROL BASED ON HYBRID NEURAL NETWORK ARCHITECTURES // Современная наука: актуальные проблемы теории и практики. Серия: Естественные и Технические Науки. -2025. -№06/2. -С. 165-169 DOI 10.37882/2223-2966.2025.06-2.27
LEGAL INFORMATION:
Reproduction of materials is permitted only for non-commercial purposes with reference to the original publication. Protected by the laws of the Russian Federation. Any violations of the law are prosecuted.
© ООО "Научные технологии"

Terms Publications

Issues of the journals (Archive)

ECONOMICS AND LAW
HUMANITIES
NATURAL AND TECHNICAL SCIENCES
COGNITION

Серия - Natural and Technical Sciences