Output temperature estimation of the gas flow temperature control system based on artificial neural network

Abstract

Due to the strong nonlinearity, and the difficulty to establish the analytical models for plasma heaters and mixing chambers, the performance of the gas flow temperature control system is severely restricted. To address this issue, three artificial neural network methods were used to estimate the output of the system. Simulation and experimental results indicate that, compared to the simulation models based on mathematical models and the system output estimation models based on feedforward neural networks, both time-delay neural network and nonlinear autoregressive with exogenous input neural network can improve the accuracy of output estimation. The nonlinear autoregressive with exogenous input neural network reduced the average system output estimation error to 0.099 K, significantly enhancing the accuracy of the estimated output for the real system. For output estimation in time varying systems, incorporating time delays to account for historical data can improve the estimation accuracy of neural networks. Furthermore, in systems where the output can be dynamically collected, adding output feedback can further enhance the estimation accuracy. This is crucial for improving the control accuracy of the gas flow temperature control system and is also insightful for output estimation in similar time-varying control systems