基于改进CS优化BP神经网络的降水量预测模型

Precipitation Prediction Model Based on Improved CS Optimization BP Neural Network

针对传统BP神经网络易陷入局部极值和连接权值难以确定的问题,提出了一种基于融合PSO(Particle Swarm Optimization)和CS(Cuckoo Search)的混合算法优化设计BP神经网络(PCS-BP)的预测模型.该优化方法主要利用混合算法优秀的全局搜索能力和收敛速度设计优化BP神经网络的连接权值和网络结构,解决了BP神经网络由于参数随机取值引起的网络震荡和过拟合的问题,提高了预测模型的准确性.结合具体实例,分别采用BP神经网络、CS-BP模型和PCS-BP模型对汉中地区的月降水量进行预测,实验结果表明,PCS-BP的平均绝对误差(MAE)为0.3966,均方根误差(RMSE)为2.3793,平均绝对百分比误差(MAPE)为0.46%,均优于其他模型,具有较好的预测能力.

In view of the problem that traditional BP neural networks are prone to local extreme values and connection weights,a predictive model of BP neural network(PCS-BP) is proposed based on a hybrid algorithm optimizing the fusion PSO(Particle Swarm Optimization) and Cuckoo Search.This optimization method mainly optimizes the connection weight and network structure of BP neural network by using the excellent global search ability and convergence speed design of mixed algorithm,solves the problem of network oscillation and overfitting caused by random parameter value of BP neural network,and improves the accuracy of the prediction model.Based on concrete examples,BP neural network,CS-BP model and PCS-BP model were used to predict the monthly precipitation in Han Zhong region,and the experimental results showed that the average absolute error(MAE) of PCS-BP was 0.3966,the mean square root error(RMSE) was 2.3793,and the average absolute percentage error(MAPE) was 0.46%,which was better than other models and had better prediction ability.