A Hybrid Algorithm of GA Wavelet-BP Neural Networks to Predict Near Space Solar Radiation

Solar radiation is affected by many factors, solar radiation prediction is a highly nonlinear problem. It is hard to establish any analytical mathematical model. Considering solar radiation ray is composed of a series of different frequency bands with different characteristics, wavelet is introduced to decompose the radiation signal into high and low frequency hefts. By respectively inputting the hefts into BP neural networks, which have strong, fault-tolerant ability and nonlinear mapping ability, better prediction precision can be obtained. But BP neural networks are apt to converge at local optimal point. so genetic algorithm is embedded to optimize BP neural networks' weights and threshold values, hybrid algorithm's prediction precision is receivable through these improvements.