Research On Urban Water Supply Forecast Based On Multi-granularity Deep Echo State Network

Under the strategic goal of"carbon dioxide emissions peak and carbon neutrality",accurately forecast the urban water supply,help the water companies to reasonably plan and effectively manage the urban water resources,achieve the purpose of energy conservation and emission reduction,and help the urban water supply system achieve the development mode of low energy consumption,low emissions,high efficiency,low carbon and green,which is of great significance for the sustainable development of the economy and the realization of the"double carbon"goal,and under the promotion of smart city construction,It is particularly important to improve the urban smart water system,which is also the trend of urban water supply system in the future.In order to improve the forecast accuracy of urban daily water supply,this paper proposes a model framework of"sequence decomposition-secondary granularity reconstruction-integrated forecast",under which a multi-granularity deep echo state network is constructed and applied to urban water supply forecast.The main work of this paper is as follows:1.Model selection.In this paper,four machine learning models are compared and studied.The models of Extreme Learning Machine(ELM),Back Propagation Neural Network(BPNN),Echo State Network(ESN)and Leaky Integrator Echo State Network(Li ESN)are used to forecast the urban daily water supply.Finally,Li ESN is determined as the forecast model for the follow-up study of this paper.2.Multi-granularity feature analysis.According to the framework of"sequence decomposition-secondary granularity reconstruction-integrated forecast",the original sequence of urban daily water supply is first decomposed into multiple intrinsic mode functions(IMFs)using the time varying filtering based empirical mode decomposition(TVFEMD)technology,and then the secondary granularity reconstruction is carried out using a variety of technologies.Specifically,fast Fourier transform(FFT)realizes the first granularity reconstruction in time-frequency domain,and sample entropy(SE)realizes the secondary granularity reconstruction in entropy domain.The comparison results show that the secondary granularity reconstruction can further identify the granularity feature of urban daily water supply,making it easier for the forecast model to learn the implicit relationship between the data.3.Integrated depth forecast.On the basis of the previous two studies,the determined multi-granularity technology and the Deep Leaky Integrator Echo State Network(DeepLiESN)based on the deep learning framework is used to make integrated forecast of urban daily water supply,and the results of integrated forecast after the granularity reconstruction are compared and verified,and then the final combined forecast model of this paper is determined as TVFEMD-FFT-SE-DeepLiESN.Its forecast performance is as follows:mean absolute error(MAE)is 1053m~3/d,root mean square error(RMSE)is1397 m~3/d,mean absolute percentage error(MAPE)is 0.5765%,and coefficient of determination(R~2)is 0.9898.The model shows the strongest forecast performance and stronger applicability,and provides a feasible scheme for accurate forecast of urban daily water supply.