抽水蓄能电站地下水位预测的优化神经网络模型

Optimized neural network model for groundwater level prediction in pumped-storage power stations

摘要:

抽水蓄能电站上、下水库落差大，水头高，针对输水系统沿线山体地下水位变化的监测和预测对电站安全运行过程中的监测分析具有重要意义。为实现施工期山体水位预测，通过环境监测站获取多项环境监测数据，结合 PCA(主成分分析)和 GA(遗传算法)优化 BP 神经网络方法，建立 PCA-GA-BP 优化模型对地下水位进行预测。选取广东某抽水蓄能电站环境量及输水系统沿线山体水位孔数据，在分析测点、测站布置及地下水位影响因素基础上，对优化算法模型进行验证、比较。实验结果表明：优化模型具有较高预测精度，在高、 中、低水位预测中综合相对误差较低，决定系数更高，均优于单 BP 预测模型，并通过 PCA 法使得网络拓扑结构更简单，提高综合预测精度，具有较好的预测效果，在实际运用中可以为安全分析、工程预警等领域提供一定参考。

Abstract:

The drop between upper and lower reservoirs of a pumped-storage power station is large, and the water head is high. The monitoring and prediction of mountain groundwater level change along the water conveyance system is of great significance to safe operation monitoring of the power station. In order to predict mountain groundwater level during construction period, environmental monitoring data were obtained through the environmental monitoring station. By combining PCA (Principal Component Analysis) and GA (Genetic Algorithm) to optimize the BP neural network method, a PCA-GA-BP optimization model was established for groundwater level prediction. One pumpedstorage power station in Guangdong is selected, and its environmental factors and mountain groundwater well data along the water delivery system are used. The optimized algorithm model is verified on the basis of analyzing measuring points, layout of the measuring station and impact factors of the groundwater level. The results show that the optimized model has high prediction accuracy, low comprehensive relative error and high determination coefficient in high, medium and low water level prediction, which is better than the single BP prediction model. And the network topology is simpler than the PCA method, which improves comprehensive prediction accuracy and thus has a better prediction performance. In practice, the optimized model can provide reference for safety analysis, engineering early warning and other fields.