Research on the methods of improving the training speed of LMBP algorithm

The paper talks about several numerical methods of improving the training speed of LMBP algorithm and the corresponding amount of computation in LMBP algorithms. According to the characteristic of formula LMBP algorithm, we propose a suitable method to reduce the calculation quantity and improve the training speed of LMBP algorithm and apply it into the basic LMBP algorithm. This method has less computation compared to several other numerical methods and improves network training speed greatly when calculating the increments of weights and biases. Finally we do several training simulations with several typical network training swatches. The simulation results indicate that total training speed of single hidden layer BP neural network based on improved LMBP algorithm converges very rapidly and has good precision compared with the basic LMBP algorithm.