基于粒子群和投影寻踪的样本重构神经网络集成模型在降水预报中的应用

首先利用粒子群算法和投影寻踪技术构造神经网络的学习矩阵,基于负相关学习的样本重构方法生成神经网络集成个体,进一步用粒子群算法和投影寻踪回归方法对集成个体集成,生成神经网络集成的输出结论,建立基于粒子群算法-投影寻踪的样本重构神经网络集成模型。该方法应用于广西全区的月降水量预报,结果表明该方法在降水预报中能有效从众多天气因子中构造神经网络的学习矩阵,而且集成学习预测精度高、稳定性好,具有一定的推广能力。

Neural Network Ensemble Model with Specimen Reconstruct Based on Projection Pursuit and Particle Swarm Optimization for Rainfall Forecasting

First of all,learning matrix of neural network is got by projection pursuit and particle swarm optimization algorithm,which particle swarm optimization algorithm optimize projection index from high dimensionality to a lower dimensional subspace,and then many individual neural networks are generated by samples reconstruction based on negative correlation learning method.Secondly,the result of ensemble is generated by projection pursuit regression based on particle swarm optimization algorithm.Finally,the forecasting model is established by neural network ensemble with specimen reconstruct based on projection pursuit and particle swarm optimization.The method is used as an alternative forecasting tool for a meteorological application in the monthly precipitation forecasting of the Guangxi region.The results show that the method can effectively improve the generalization ability of the system in achieving greater forecasting accuracy and improving prediction quality further.