改进的粒子群BP神经网络算法在天气预测中的应用

Application of improved particle swarm optimization based BP neural network in weather prediction

介绍了一种基于改进的粒子群算法BP神经网络(IBPPSO)来预测天气情况。粒子群算法BP神经网络(BPPSO)弥补了BP神经网络迭代次数多,运算速度慢的缺点。粒子群算法中学习因子(c1、c2)、惯性权重(w)设定为常数,实际情况下,c1、c2、w随迭代次数而变化。文章对粒子群算法进行适当改进,对最佳位置的搜索从全局的大范围渐渐向局部的小范围转变,既满足了较大的全局搜索能力,又增加了小范围局部搜索的精度。该方法应用于上海虹桥机场近三年来12000组天气情况数据。结果表明,IBPPSO预测准确率和成功预测雨天概率两方面都优于BP神经网络与BPPSO。

This paper introduces an improved particle swarm optimization based BP neural network (IBPPSO) to predict weather conditions. The particle swarm optimization BP neural network (BPPSO) makes up for the shortcomings of the traditional BP neural network, which has many iterations and slow operation speed, and enhances the global search ability of the BP neural network algorithm. In particle swarm algorithm, the learning factor (c1、c2) and inertia weight (w) are set as constants. In practice, c1、c2、w vary with the number of iterations. In this paper, the particle swarm algorithm is improved appropriately, and the search for the best location is gradually changed from the global large-scale to the local small-scale, which not only satisfies the larger global search ability, but also increases the precision of the small-scale local search. The method is applied to test 12,000 sets of weather data of Shanghai Hongqiao Airport in the past three years. The results show that IBPPSO is superior to BP neural network and BPPSO in both accuracy and probability of successful prediction of rainy days.