基于机器学习的城市积水风险预测研究

Research on Urban Waterlogging Risk Prediction Based on Machine Learning

摘要:

随着城市化进程的加剧，城市道路硬化后，地表的雨水渗透能力显著降低，雨水吸纳能力逐渐减弱，导致地表径流增加，城市内涝风险加剧，积水预测成为城市防洪减灾的重要技术手段。结合机器学习方法，提出了一种综合预测模型，用于模拟降雨过程中的径流量、排水能力以及积水深度的变化。使用某地降雨径流和排水能力参数，使用粒子群优化的BP神经网络对积水深度进行预测。通过2024年8月24日的降雨与积水深度的数据实验验证，结果表明粒子群优化的BP神经网络在预测准确度上优于传统BP网络，为城市积水预测和防洪提供了新的技术思路。

Abstract:

With the acceleration of urbanization, after the hardening of urban roads, the rainwater infiltration capacity of the surface has significantly decreased, and the rainwater absorption capacity has gradually weakened, leading to an increase in surface runoff and exacerbating the risk of urban waterlogging. Waterlogging prediction has become an important technical means for urban flood control and disaster reduction. Combined with the machine learning methods, a comprehensive prediction model is proposed for simulating changes in runoff, drainage capacity and waterlogging depth during rainfall. Using rainfall-runoff and drainage capacity parameters from a certain area, a particle swarm optimization (PSO)-based BP neural network is used to predict the waterlogging depth. The experimental verification with rainfall and waterlogging depth data on August 24, 2024 shows that the PSO-based BP neural network is superior to the traditional BP network in prediction accuracy, providing a new technical approach for urban waterlogging prediction and flood control.