A Hybrid BP Network and Its Application in Virtual Sensor of Dynamic Flow

摘要

An accurate on-line measurement of dynamic flow signals is essential for the successful monitoring of hydraulic systems. However, flow signals usually were difficult to measure on-line due to the limitations such as the time delay, high cost and reliability, et al. The virtual sensor, an inferential model, has been used as an alternative of flowmeter for predicting flow signals. In order to overcome the disadvantages such as slow convergent speed and low generalization as back propagation (BP) neural network (NN) was used to build virtual sensor model, we present a novel model combined with working principium of the process and hybrid modeling of BPNN and partial least square(PLS) algorithm. The proposed method was illustrated by comparisons with other methods. Simulation results have shown that the proposed method gives a better or equal performance over the conventional PLS and BPNN method.