Temperature compensation method for piezoresistive pressure sensor based on IGWO-SVR

Abstract To mitigate the nonlinearity in the response characteristics of silicon piezoresistive pressure sensors caused by temperature drift, this paper proposes an improved grey wolf optimization optimized support vector regression (IGWO-SVR) temperature compensation method. First, an IGWO algorithm is proposed that combines a nonlinear convergence factor with an adaptive Levy flight strategy to globally optimize the penalty coefficient, insensitive loss function coefficient, and SVR kernel function. Then, based on the characteristics of a wide temperature compensation range, high nonlinear complexity, and strong interference of pressure sensors under high temperature and high-pressure working conditions, a temperature compensation model of piezoresistive pressure sensors based on IGWO-SVR is developed. The results of an algorithm transplant experiment using MATLAB 2024b and the STM32F407VET6 microcontroller show that the proposed algorithm’s nonlinear convergence factor can dynamically adjust the optimal pressure search behavior based on the problem’s complexity, reducing ineffective iterations during grey wolf optimization (GWO) training. The adaptive Levy flight strategy allows the GWO algorithm to quickly exit the current search area when local development slows, which not only improves global adaptive optimization but also enhances the robustness of the SVR temperature compensation model. The actual pressure compensation results in the high pressure range of 0–50 MPa and the wide temperature range of −20 °C–70 °C demonstrate that the maximum value of the quoted error, the mean absolute error, the root mean square error (RMSE), and the accuracy are only 9.1857 × 10 −3 , 2.8699 × 10 −3 , 1.5303 × 10 −4 , and 0.018%, respectively, and their compensation accuracies are better than those of the existing artificial bee colony-SVR, particle Swarm optimization-SVR, sparrow search algorithm-SVR, whale optimization algorithm-SVR, and GWO-SVR algorithms. Compared with the second-best performing algorithm among these five baseline methods, the proposed method reduces mean square error and RMSE by 73.7% and 48.7%, respectively, and improves accuracy by 57.1%. This model can be widely applied to the intelligent temperature compensation of piezoresistive pressure sensors under high-temperature and high-pressure working conditions, intelligent control in industrial fields, and parameter prediction.