摘要
为了提高水库和河流中长期径流预测精度,提出了弹性自适应人工鱼群算法(RAAFSA)。应用RAAFSA算法训练BP神经网络,实现BP神经网络参数优化,形成弹性自适应人工鱼群-BP神经网络混合算法(RAAFSA-BP),对石泉水库进行中长期径流预测。仿真计算表明,弹性自适应人工鱼群优化的BP神经网络算法收敛速度快于BP神经网络算法、人工鱼群-BP神经网络算法和RBF神经网络算法。该混合算法克服了BP神经网络和人工鱼群算法易陷于局部极值、搜索质量差和精度不高的缺点,改善了BP神经网络的泛化能力,输出稳定性好,预报精度显著提高,每次预测相对误差绝对值都小于6%,合格率达到100%。该算法成功地解决了石泉水库中长期径流预测精度不高的难题,可有效用于水库和河川中长期径流预测。
In order to improve the long-term runoff forecast accuracy of reservoirs and rivers, the paper analyzes the shortcomings of the traditional artificial fish school algorithm （AFSA） , and a resilient adaptive artificial fish school algorithm （RAAFSA） is brought up. The forecast model is set up by using a resilient adaptive artificial fish school algorithm and the back propagation （BP） neural network combined to form RAAFSA-BP hybrid algorithm, and then the neural network was trained by using the RAAFSA algorithm. It can automatically determine the parameters of the neural network from the sample data. The long-term runoff forecast model was formed based on the hybrid algorithm. Then the long-term runoff forecast of reservoirs was carried out by using the method and historic runoff data. The result shows the convergence of the method is faster and forecast accuracy is higher than that of the artificial fish school algorithm-BP neural network, RBF neural network and BP neural network. The method improves forecast accuracy and the BP neural network generalization capacity. It has a high computational precision with an average percentage error below 6% and the percentage eligibility of 100%. The model can successfully improve the long-term runoff forecast accuracy and speed of reservoirs in Shiquan. The hybrid algorithm can be used to perform the long-term runoff forecast of the reservoirs and rivers.
出处
《自然资源学报》
CSCD
北大核心
2009年第11期2005-2013,共9页
Journal of Natural Resources
基金
国家火炬计划基金(07C26213711606)
陕西省自然科学基础研究计划(SJ08E220)
山东省软科学基金(2009RKB190)
关键词
径流预测
RAAFSA-BP网络算法
弹性自适应人工鱼群算法
BP神经网络
long-term runoff forecast
RAAFSA-BP hybrid algorithm
resilient adaptive artificialfish school algorithm
the back propagation neural network
