基于回溯搜索优化算法的BP神经网络年径流预测

Annual Runoff Prediction,Adopting BP Neural Network based on Backtracking Search Optimization Algorithm

针对BP神经网络易陷入局部极值及初始权阈值参数难以确定的不足,利用一种新型进化算法——回溯搜索优化算法(BSA)优化BP神经网络初始权值和阈值,提出BSA-BP年径流预测模型,并构建PSO-BP、GA-BP及BP模型作对比模型,以云南省清水江站年径流预测为例进行实例研究。结果表明:BSA-BP模型预测精度优于PSO-BP、GA-BP及BP模型。利用BSA算法优化BP神经网络的初始权值和阈值,可有效提高BP神经网络的预测精度和泛化能力。

As forBP Neural Network, the local extrema and initial weight and threshold parameters are difficult to be determined. Con- sidering the above deficiencies, Backtracking Search Optimization Algorithm （BSA） is utilized for optimizing the initial weights and thresholds of the BP Neural Network. Thus, a BSA - BP model for predicting annual runoffs is established, together with the construc- tion of the PSO - BP model, the GA - BP model and the BP model for comparison. In this research, the Qingshuijiang Station of Yun- nan is used as an example for case study. The results show that： the BSA - BP model has better precision and adapability than the PSO - BP model, the GA - BP model and the BP model.