An improved BP neural network algorithm for researching on stability of reverse unloading diaphragm pressure reducing regulator

To investigate the stability of PRR(pressure reducing regulator) with adjusting multiple structure parameters simultaneously, BFGS (Broyden-Fletcher-Goldfarb-Shanno) quasi-Newton method and line search with Wolfe conditions were applied to optimize the BP (back propagation) algorithm. Results showed that the improved BP algorithm reduced number of iterations by 1-2 orders of magnitude, making it easy to reach a global minimal point. When the improved BP algorithm was used for reverse unloading diaphragm PRR, it could adapt to coupling of 2-3 structure parameters and predict the data set with more than 10⁶ data points. It was easy to find a combination of structure parameters to stabilize PRR when multiple structure parameters change simultaneously. More importantly, when these parameters changed together, the stability of PRR was much better than the case with change of just one of the parameters.