数据驱动的上海市日需水量预报建模研究

Data-driven Modeling for Daily Water Demand Forecast of Shanghai City

为了提高城市供水调度的品质和效率,需要高精度的日需水量预报信息作为参考。分析影响城市需水量变化的主要因素,以近期需水量、降水及气温实测值为输入,辅以星期、节假日信息校正,采用RBF神经网络与支持向量机相结合的数据驱动建模技术,进行超前一天需水量预报研究。为了提高黑箱模型的训练效果,对数据进行一系列预处理,包括分离出历史需水量中的变化量;提取降水量的连续等级信息;非线性处理温度对需水量的影响。通过模型验证,结果表明预报误差在1%以内的占总预报天数的62.0%。

For better quality and efficiency of water supply scheduling,high-precision water demand forecast is required as a important reference.Main factors influencing the water demand of a city are presented and analyzed.A model based on RBF neural network and support vector machine is used to forecast the water demand 24 hours in advance.Air temperature,precipitation and recent water demand data are taken as the model input.Water demand variations associated with weekend and holidays are also taken into account. While related data are pretreated to improve train effect of black-box method,including detaching variant information of water demand, calculating indiscrete scale value of precipitation,and processing nonlinear treatment to the air temperature information.As a result, the forecast data whose relative error are less than 1%are 62.0%of the whole data.