Numerical and experimental verification of the single neural adaptive PID real-time inverse method for solving inverse heat conduction problems

For the problems of difficult on-line tuning of PID parameters, weak anti-interference ability and poor self-adaptive ability in the classical proportional-integral-derivative (PID) inverse method, combining sin-gle artificial neuron with PID inverse algorithm, a single neural adaptive PID (SNA-PID) inverse method is established to estimate the thermal boundary condition. Compared with the classical PID inverse method, the main difference is that SNA-PID can adaptively adjust the weights of PID parameters. In the SNA-PID inverse method, firstly, the deviation signal between the output of the direct heat conduction model and the measured temperature and its first-order difference and second-order difference are taken as three inputs of a single neuron (dendritic input signal of neuron), and the connection strength of each input signal is determined according to the updated weight of the adaptive learning algorithm, The excitation value of the boundary heat flux increment of the heat conduction system is generated according to the linear truncation function. Finally, the real-time estimation of boundary heat flux of heat conduction sys-tem is realized. The unsteady boundary heat flux of one-dimensional heat conduction object is estimated through numerical and experimental tests, which verifies the effectiveness of the proposed SNA-PID in-verse method. (c) 2022 Elsevier Ltd. All rights reserved.

Automated summary

In this paper, a single neural adaptive PID (SNA-PID) inverse method is proposed to estimate the thermal boundary condition. Compared with the traditional PID inverse method, SNA-PID can adaptively adjust the weights of PID parameters, thus improving the tuning effect and anti-interference ability.