The nonlinearity compensation of piezoresistive sensor based on Levenberg-Marquardt algorithm

Abstract

The nonlinear compensation based on artificial neural network trained by improved Levenberg-Marquardt (LM) algorithm is proposed in the paper. In order to reduce the computation complexity of LM algorithm, the input variable space is divided into several subsections and the structure of neural network is simplified in subsection LM algorithm. During the training process, a new method of matrixes multiplying called column-row multiplication is used to reduce the temporary elements for storage. As the simulation shown, the required memory space of training is reduced 77% compared with LM algorithm, furthermore, the accuracy and convergence speed is in same lever. The improvement makes the artificial neural network training process with LM algorithm could run in microprocessor of piezoresistive sensor.