An Error Evaluation Method of Temperature On-Site Measurement Based on Mgf-Sr-Ga-Bp Model

Abstract

Because currently available existing temperature sensors cannot be disassembled for measurement, we proposed an on-site testing method based on a model that integrates a mean generation function (MGF), stepwise regression (SR), a genetic algorithm (GA), and back propagation (BP) (i.e., the MGF-SR-GA-BP model). First, error data obtained through comparison with a standard device were periodically extended using an MGF, and SR was applied to establish an MGF-SR model. Second, a BP neural network was used to establish an MGF-BP model, and a GA was used to perform optimization and create an MGF-GA-BP model. Finally, a linear online evaluation model, namely the MGF-SR-GA-BP model, was established using the aforementioned models. This model uses the sum of squared prediction errors as the basis for its objective function. In experiments, compared with the MGF-SR and MGF-GA-BP models, the hybrid MGF-SR-GA-BP model achieved a lower average relative error (0.27 and 0.66 vs. 0.21, respectively).

Keywords: On-site measurement, mean-generating function, stepwise regression, back propagation