Research On Data-driven State Estimation Method Of Electric-gas Coupled Integrated Energy System

With the development of the national energy revolution strategy and the power system reform,the integrated energy system(IES)has gradually emerged as a new energy development form that integrates various energy attributes to maximize the energy utilization efficiency.Among them,state estimation is an important part of the energy management system in IES,and its accuracy has the direct impact on the energy dispatch,monitoring and forecasting of IES.However,the problems of low redundancy of measurement data,large measurement errors of measurement devices,and inconsistency in the unit time scale of data collection between power system and gas network measurement devices bring serious challenges to the state estimation of gas and electric integrated energy system(GEIES).Therefore,this paper addresses the state estimation problem of GEIES,and the main work is as follows:(1)In this paper,we research the GEIES measurement model.The different measurement equipment systems of gas network and power system is analyzed in GEIES.The current problem of low observability is explored in GEIES state estimation.The GEIES measurement equations are constructed by using Bayesian learning.The two computational models of Bayesian learning: Gaussian inferential model and Gaussian mixed distribution model are established;the two computational models of Bayesian learning are applied to solve the probability distributions of gas network and grid measurement in GEEIS respectively.(2)It is difficult to calculate the state estimation of GEIES network by the traditional method due to the low redundancy of measurement data.Considering that the data-driven method has high portability and the ability to refine and summarize different information,the GEIES Bayesian state estimation model based on long-short term memory(LSTM)is established.Based on the probability distribution of GEIES measurement data obtained by Bayesian learning,enough GEIES measurement data is generated by Monte Carlo sampling.The rationality of the generated data is verified by the GEIES energy flow to obtain the training sample set of the LSTM network.The root-mean-square error criterion is used to train the LSTM network,which effectively improves the error accuracy of GEIES state estimation.In comparison with the classical model-driven state estimation method,the simulation verifies the effectiveness and robustness of the proposed method.(3)Monte Carlo sampling method has the problem that the generated data are concentrated around the mean value.The Latin hypercube sampling method is established to realize the data generation of quantitative measurements and cover all probability distribution intervals.LSTM network has a high fit for data with temporal characteristics,but the quantitative measurements of GEIES still have certain spatial characteristics.Therefore,a hybrid deep learning network based on CNN-LSTM is constructed to achieve the state estimation of GEIES,which improves the estimation accuracy.