Nonlinear Correction of Methane Sensor Based on Functional Link Neural Network

Abstract

The nonlinear relation between methane concentration and the output voltage of the sensor is indicated by analysis of detection principle of catalytic methane sensor. This paper proposes a nonlinear correction model based on functional link neural network (FLNN) with the output voltage of methane sensor as input and the methane concentration as output to eliminate the nonlinear errors in methane detection. By adding some high-order terms, the model applies the single-layer network to realize the network supervised learning. The approach has advantages of nonlinear approach ability and independent on accurate mathematical model, it can improve network learning speed and simplify the network structure. The experimental result shows that the maximum relative error of simulation curves is reduced to 0.86%, which is much smaller than that of piecewise linear fitting curve with 3.09%. The detection accuracy of methane sensor is improved.