摘要
本文将小波分析与BP人工神经网络相结合,并使用遗传算法优化神经网络,建立起遗传算法优化的小波神经网络———遗传小波神经网络。从合肥城市需水密切相关的14个社会经济指标中,筛选出主要的影响因子,根据长时间序列数据,构建了城市水资源需求量预测模型。通过遗传小波神经网络和传统BP的网络训练输出效果比较,表明该预测模型收敛速度较快,对神经网络的性能优化有明显效果,拟合精度较高,泛化能力较好,对城市需水预测能取得较好的效果。
In this paper, an iterative method which combines the strength of back-propagation （BP） in weight training and genetic algorithms＇ （GA） capability of searching the satisfying solution was proposed for optimizing wavelet neural networks （WNN）. Taking provincial capital city of Hefei as an example, the proposed GA optimized WNN that required a few representative properties as possible for input data was applied to predict the urban water demand of the near future. It selected 14 socio-economic influence factors of urban water demand, and partial least squares （PLS） method was used to select the important factors influencing urban water demand. The study used 11 important socio-economic factors impacting urban water demand as input data and total water consumption during the study period as output data to construct the WNN model. Finally, the prediction performance of the GA optimized WNN was compared with traditional BP neural networks, and simulation results demonstrated the accuracy and the reliability of the prediction methodology based on the proposed model.
出处
《测绘科学》
CSCD
北大核心
2013年第5期28-31,共4页
Science of Surveying and Mapping
基金
上海重点学科建设项目(人文地理学:B410)
关键词
小波神经网络
遗传算法
预测
城市需水
偏最小二乘回归
wavelet neural network
genetic algorithms
prediction
urban water demand
partial least squares （PLS） method
