基于粒子群优化算法的RBF神经网络在泾惠渠灌区地下水位埋深预测中的应用

Groundwater Depth Prediction Model in Jinghui Irrigation District of RBF Neutral Network Based on Particle Swarm Optimization

针对不恰当地选取RBF神经网络的网络结构和参数会使网络收敛慢的问题,采用粒子群优化算法对RBF神经网络参数进行优化,建立了基于粒子群优化算法的RBF神经网络模型（PSO-RBF模型）,对泾惠渠灌区地下水位埋深进行了模拟和预测。结果表明,与单一的RBF神经网络相比,PSO-RBF模型具有较高的预测精度。再根据时间序列预测法预测的降水量、径流量、蒸发量、渠灌引水量、地下水开采量、气温等模型的输入变量,用训练好的PSO-RBF模型预测了泾惠渠灌区2009-2020年地下水位埋深,发现该灌区地下水位埋深呈下降趋势。

As inappropriate parameters of RBF neural network will slow down the network convergence rate, particle swarm optimization （PSO） is used to optimize the parameters of RBF neural network. And then the PSO-RBF model is established to simulate and predict the groundwater depth of Jinghui irrigation district. Compared with single RBF neural network model, the results show that PSO-RBF model has higher prediction accuracy. Finally, based on the data of precipitation, runoff, volume of groundwater exploitation and other model input variables which were predicted by time series forecasting method, PSO-RBF model is applied to forecast groundwater depth of Jinghui irrigation district from 2009 to 2020, it finds out that the groundwater depth of this irrigation district appears downward trend.