Improved BP Neural Network Based On CPA And Its Application Research In Meteorological Data

BP (Back propagation) neural network is one of the most widely used artificial neural networks. It is known that the performance of the BP neural network mainly depends on neural network’s structure. In the process of iteration, some nodes in hidden layer become redundancy, which may affect the neural network training efficiency. What’s more, the BP neural network takes long time to achieve convergence and the results may fill in local optimum.Based on above-mentioned problems, the Correlation Pruning Algorithm (CPA) is introduced into the BP neural network optimized by variable learning rate and adding momentum in this paper. This method improves the convergence speed and prediction accuracy. Besides, we put it into the application of meteorological data restoration algorithm. The main works of this paper include two following aspects:(1) We have proposed an improved BP neural network pruning algorithm based on CPA algorithm, named as LMCPA neural network. The improved BP neural network pruning algorithm combines CPA algorithm with variable learning rate and additional momentum. In this paper, we introduce the CPA correlation pruning algorithm into Zhang’s method to optimize BP neural network. In order to verify the performance of the proposed method, we have compared LMCPA algorithm with the original BP neural network, Zhang’s proposed BP neural network and CPA pruning algorithm respectively. The experimental results show that the performance of the neural network has been improved by LMCPA algorithm.(2) We have proposed a meteorological data repairing algorithm. Basing on the LMCPA algorithm proposed in this work, we have taken Rough Set to reduce the redundant data. Firstly, we have taken Rough Set to calculate the importance of meteorological attributes (such as wind speed, humidity). After that, we have reduced the insignificant attributes and remove the related meteorological data. Basing the reduced data, we have applied LMCPA algorithm to repair the error weather data. We have compared the two methods:Rough Set algorithm combines with LMCPA and applied LMCPA directly. The results show that by using Rough Set to reduce the weather data, the convergence rate and accuracy of LMCPA algorithm has been improved.