Study On The Algorithm For Intelligent Urban Water Supply Forecasting Based On Incremental Learning

Incremental learning is a type of algorithm that simulates the learning logic of human brain.It does not require all historical samples,it only uses important features or individual representative samples to effectively implement knowledge learning of new data.It is suitable for the scenario such as repeated learning of data in online systems and model parameters which are dynamically updated.Urban water supply is the guarantee of the normal operation of the city,accurate and efficient water supply prediction is crucial to the production scheduling decisions of water utilities.Based on the self-developed smart water system,the author designed an incremental water supply prediction algorithm based on machine learning for the purpose of achieving high-precision water supply prediction,and completed data acquisition,modeling analysis and visual design of prediction results.The author’s work and contributions are mainly reflected in the following three aspects:(1)Filling in the missing abnormal data in the water supply prediction model.Water supply forecast is a key part of water supply enterprises to provide efficient water supply services.The author clarified the short-term forecast task of water supply by analyzing the forecast demand of water supply from water company.By analyzing the daily water supply data of the relevant water plants,it was found that there are about20% of outlier samples are missing.by introducing external environmental data including auxiliary information such as temperature,weather,and holidays,the detection and filling of outliers are realized,and the accuracy of network prediction is improved.(2)Introducing incremental learning method to solve the problem of insufficient sample size in water supply prediction.Water supply prediction methods can be divided into two categories,namely traditional batch offline models and online learning methods based on incremental training.The author establishes a prediction regression model of water supply by analyzing the short-term distribution trend of daily water supply and combining the characteristics of newly added data in the online system.Considering that the water supply is affected by external environmental factors,the new data distribution mode is complex,and the prediction model needs to learn the new data knowledge in time.The traditional batch learning method will no longer be applicable,so it is proposed to use incremental learning as a model parameter update algorithm for data training,constantly using the new data to improve the relevant models,effectively solving the problem of small data samples in the water system.(3)An incremental SVR algorithm based on exponential smoothing correction is proposed to improve prediction accuracy.Daily water supply forecasting is a regression forecasting model,the author first applied the traditional incremental support vector regression method in the field of water supply prediction.Based on the analysis of traditional incremental SVR prediction methods and their advantages and disadvantages,an incremental SVR algorithm based on exponential smoothing correction is proposed to learn the local trend information of the newly added data in the incremental interval.This model not only has the advantages of traditional incremental SVR learning data general distribution law,but also has the ability to synthesize the local trend information of the data obtained by the exponential smoothing method to improve the prediction accuracy.Finally,the method is applied to the prediction of daily water supply of actual water plants,and a comparative analysis is performed using incremental Mondrian forest and BP incremental regression algorithm.The experimental results show that the exponential smoothing modified incremental SVR method fully learns the local trend information of the data and obtains the best prediction accuracy.