基于遗传算法的城市雨水排水管网状态预测研究

摘要：结合遗传算法（Genetic Algorithm， GA）与长短期记忆网络（Long Short-Term Memory，LSTM）构建预测模型，并基于管龄、管长、埋深等多维度特征参数，采用GA动态调整LSTM超参数组合，以准确评估城市雨水排水管网状态。模型验证结果表明，优化后的GA-LSTM模型在结构性缺陷F和功能性缺陷G预测中表现突出，决定系数（R²）、平均绝对误差（Mean Absolute Error，MAE）、平均绝对误差（Mean Absolute Error，MAE）3项指标分别达到0.916、0.309和0.135，相较于传统LSTM模型，R²提高了4.9%，MAE和RMSE值分别减少了12.2%和17.7%

关键词：遗传算法 长短期记忆网络 雨水管网 状态预测

Research on State Prediction of Urban Rainwater Drainage Network Based on Genetic Algorithm

WU Rongping

Yuanhong Liangguo Shuangyuan （Fujian） Holdings Group Co.， Ltd.， Fuzhou， Fujian Province， 350300 China

Abstract： Combining genetic algorithm （GA） with long-term and short-term memory network （LSTM， a prediction model is constructed， and based on multi-dimensional characteristic parameters such as pipe age， pipe length and buried depth， the GA is used to dynamically adjust the hyperparameter combination of LSTM to accurately evaluate the state of urban rainwater drainage pipe network. The model verification results show that the optimized GA-LSTM model is outstanding in the prediction of structural defect F and functional defect G， and the three indexes of determination coefficient （R ²）， mean absolute error （MAE） and MAE reach 0.916， 0.309 and 0.135 respectively. Compared with the traditional LSTM model， R2 is increased by 4.9%， and the values of MAE and RMSE are decreased by 12.2% and 17.7% respectively.

Key Words： Genetic algorithm; Long-term and short-term memory network; Rainwater pipe network; State prediction

随着城市化进程的加快，城市雨水排水管网作为城市的重要基础设施，其健康状态直接关系到城市的防汛排涝能力和居民的生活安全[1]。（剩余3743字）