Continuous stirred tank reactor (CSTR) is a common reactor in the chemical industry. The accurate observation of the concentration conversion rate of the mixture and the internal temperature of the reaction vessel is a prerequisite for obtaining the desired mixture. This paper proposes a novel observer based on residual neural networks for CSTR systems. Firstly, the mathematical model of the CSTR reaction is given, as well as a detailed description of the structure and equations of the residual neural networks and the designed observer. Then the matrix method is used for the nonlinear isolation of the residual neural networks and the theory of quadratic constraints for nonlinear activation functions of the neural networks is applied. Thus, the convergence of the proposed observer is analyzed theoretically in detail. Finally, the numerical simulations are implemented to demonstrate that the proposed residual neural network-based observer can quickly and accurately observe the state changes during the CSTR reaction.

Continuous stirred tank reactor (CSTR) is one of the most common industrial equipment in petroleum and chemical industry, and is widely used in regrouping, fermentation engineering and additive preparation. In general, CSTR is used to prepare a fixed concentration of the output product. Accurate and fast monitoring of the changes in the state quantities of the CSTR chemical reaction process becomes the most important aspect before implementing excellent control. This paper presents a neural network observer with a residual network as the core component. In addition, the operations of the neural network are also matrixed to isolate the nonlinearities as much as possible. Finally we conduct numerical experiments in MATLAB R2018b based on SIMULINK framework to verify the feasibility of our strategy.

The traditional BP network is the most typical and widely used artificial neural network, which easily falls into local minimum with the slow convergence speed, and other shortcomings, seriously affecting its performance and application. In this paper, considering the error signal features in PID(Proportional Integral Derivative) parameters tuning, the variable DNN (Dynamic Neural Network) structure is introduced to optimize the traditional BP network as a new strategy, especially for some time-delay systems. The simulation experiments are realized by simulation module in MATLAB. The result shows that the performance based on the designed BP-DNN neural network is better than that of traditional BPNN (Backpropagation Neural Network), with the faster convergence speed and fewer oscillations. Further experiments suggest that the parameter tuning algorithm for PID parameters based on the variable network structure is also better, presenting characteristics-faster response, higher steady-state accuracy, stronger robustness, etc.