PSO-ESN在城市内涝点降雨积水预测中的应用

Application of Echo State Network in the Prediction of Water Level at Urban Waterlogging Point

针对城市内涝监控系统内涝点监测数据利用有限问题,提出一种基于PSO-ESN的内涝点降雨积水预测方法。分别选取历史降雨和液位为输入向量、当前液位为输出向量,通过回声状态网络动态逼近输入和输出向量之间的映射关系建立预测模型,以迭代多步预测方法得到未来液位,并采用粒子群算法解决回声状态网络模型关键参数和时间序列嵌入维数选择主观性问题。应用实例表明所述模型在内涝点降雨积水预测中的适用性,与传统Elman神经网络和BP神经网络相比,所述模型预测精度分别提高52.9%和82.4%。该方法能够有效利用监测数据,为内涝预警以及排水系统优化调度提供科学依据。

Regarding the low utilization rate of monitoring data from urban waterlogging monitoring system,a method based on PSO-ESN for prediction of water level at waterlogging sites is proposed. The historical rainfall and water level data are selected as input vector and the current water level is selected as output vector,the prediction model is established by dynamically approximating the mapping relationship between input and output vectors by echo state network,the future water level is predicted by iterative multi-step prediction method. Particle swarm optimization algorithm is used to relieve the subjective choice of key parameters of the model and the time series embedded dimension.The applicability of the model in the prediction of water level at waterlogging sites is shown by the example. Compared with traditional Elman neural network and BP neural network,the prediction accuracy of the model is respectively increased by 52.9 percent and 82.4 percent. The method can effectively use the monitoring data and provide a scientific foundation for the waterlogging warning and optimized scheduling of drainage system.