粒子群算法优化BP在降雨空间插值中的应用

为更好地表达降雨量的空间分布,将粒子群算法(PSO)优化后的反向传输(BP)神经网络分别运用于三峡区间流域日、月和年降雨量的空间插值中,并与单纯BP神经网络和克里金的插值效果作对比。研究结果表明:在日和年的时间尺度上,PSO-BP插值性能较BP有明显改善,且优于克里金的插值效果;在月时间尺度上,PSO-BP插值效果与BP接近且优于克里金。因此,PSO-BP能较好地揭示降雨量在空间的分布规律,也具备在不同时间尺度上对降雨量进行空间插值的能力,是一种较优的降雨空间插值方法。

Application of BP Neural Network Optimized by Particle Swarm Optimization to Rainfall Spatial Interpolation

To better describe the spatial distribution of rainfall,we applied BP neural network optimized by particle swarm optimization to the daily,monthly and yearly rainfall spatial interpolation of the Three Gorges reservoir area,and compared the performance with those of simple BP and Kriging interpolation.We found that in daily and yearly time-scale,PSO-BP neural network performs better than BP and Kriging;while in terms of monthly time-cale,PSO-BP result is close to BP and better than Kriging.We conclude that BP neural network optimized by particle swarm optimization could better reveal the law of spatial distribution of rainfall and has the ability of spatial interpolation in different timescales,and therefore is an excellent method for rainfall spatial interpolation.