基于小波神经网络替代模型的抽水井涌水量计算方法

影响抽水井涌水量的水文地质因素多且存在不确定性,采用确定性模型模拟会导致较大的误差。以某研究区抽水井涌水量为依据,建立该地区地下水不确定性数值模型。采用参数敏感性分析确定对抽水井涌水量影响较大的水文地质参数,随后用改进的随机进化算法在参数取值范围内抽样,最后将抽样参数输入地下水数值模型中计算抽水井涌水量。根据输入、输出数据集建立小波神经网络模型代替地下水数值模型。研究结果表明,不同工况下基于小波神经网络替代模型计算结果与数值模拟结果相差不大,两者误差在10%以内,表明用小波神经网络替代模型满足精度要求,且避免了传统数值模型的反复试算,计算简便、效率高,且因其可公式化,扩大了在推求地下水抽水井涌水量方面的应用。

Calculating Pumping Well Inflow Based on Surrogate Model of Wavelet Neural Network

There are many hydrogeological factors affecting the water quantity of pumping well as well as uncertainties. The simulation with deterministic model will lead to large error. Based on the water inflow of a pumping well in a research area, an uncertain numerical model of groundwater was established. The hydrogeological parameters that have a great influence on the pumping well water inflow were determined by parameter sensitivity analysis. Then the improved stochastic evolutionary algorithm was used to sample within the range of parameter value. Finally, the sampling parameters were input into the groundwater numerical model to calculate the pumping well water inflow. Based on the input and output data sets, a wavelet neural network model was established to replace the groundwater numerical model. The results show that there is little difference between the surrogate model based on wavelet neural network and that of numerical simulation under different working conditions, and the error between the two models is less than 10%, which indicates that the surrogate model based on wavelet neural network meets the precision requirement, and avoids the repeated trial calculation of traditional numerical model, so the calculation is simple and efficient. Due to its formulation, its application in the calculation of groundwater pumping well inflow is expanded.