Research On Data-Driven State Estimation For Integrated Energy System

The Integrated Energy System(IES)has received more and more attention in recent years for its potential of improving energy utilization efficiency and hosting capacity of renewable energy.To enhance the monitoring and control of IES,several Model-Driven State Estimation(MDSE)methods for IES have been proposed for filtering raw measurement data,improving system visibility,and helping validate the operation state of system.However,these MDSE methods are highly dependent on the accurate system topological information,which is difficult to achieve in the context of increasing penetration rate of renewable energy sources and frequently reconstructed network topology of IES.Therefore,when the topological information of current snapshot cannot be accurately obtained,the estimation accuracy of MDSE cannot be guaranteed.To this end,a Data-Driven State Estimation(DDSE)method for IES that based merely on historical operational data and current measurement is needed,in order to get rid of the dependence of the traditional MDSE method on the topological information of the system.Thus,accurate prediction,decision and control if IES can be achieved.In this paper,static state estimation and dynamic state estimation based on the idea of data-driven for IES are studied respectively.The main work is as follows:(1)In order to relize an Integrated Energy System State Estimation(IES-SE)method independent of current topological information,a data-driven state estimation method for Integrated Electricity-Heat System(IEHS)is proposed,which can be divided into off-line learning stage and online application stage.Firstly,the accurate linearization of IEHS nonlinear measurement equation is realized by introducing auxiliary measurement and state variables,and the exact linear regression equation is obtained.In the off-line learning stage,the estimated value of the linear mapping matrix in the exact linear regression equation was obtained using Partial Least Absolute Deviation(PLAD)regression algorithm.In the online application stage,multiply the linear mapping matrix with the auxiliary measurement of current snapshot to get the estimated value of the auxiliary state variables of current snapshot,and then carry out linear and nonlinear transformation to get the estimated value of the state variables of current snapshot.Through the simulation example analysis,compared with MDSE method,the DDSE method has higher estimation accuracy,higher computational efficiency and better numerical stability.(2)In order to extend the single topology-oriented DDSE method to time-varying topology-oriented DDSE method,a data-driven robust state estimation method for Integrated Electricity-Heat System considering topological changes is proposed.On the basis of the proposed DDSE method,a preprocessing method for the bad data in the current snapshot is added in the online application stage.Based on the linear relationship between the node injection measurement and the branch measurement,the bad data in the measurement is initially identified and replaced.Then,a topology identification method based on Support Vector Classification(SVC)is proposed to quickly match the current measurement with the historical data set with the same system topology.The simulation examples show that the proposed identification and replacement method of bad data improves the robustness of DDSE in online application stage.The proposed SVC-based topology identification method extends the DDSE method which is only suitable for single system topology to time-varying system topology as well.(3)Considering the difference of time scale and sampling period between power system and natural gas system,as well as the increasing nonlinearity of the Integrated Electricity-Gas System(IEGS)due to different types of load and access to new energy,Traditional model-driven dynamic state estimation methods for IEGS are difficult to accurately track the nonlinear dynamic characteristics of IEGS.Therefore,a Koopman-Kalman filter based data-driven dynamic state estimation method for IEGS is proposed.In the off-line learning stage,based on the Koopman operator theory,the nonlinear IEGS dynamic system is raised to the high-dimensional Hilbert space through the ascending function,and its global linearization is realized.And the Koopman tuples are obtained by the Extended Dynamic Mode Decomposition(EDMD)algorithm.In the online application stage,firstly,the interpolation method is used to generate the pseudo-measurement of the natural gas system to make the measurement data of each subsystem "align".Then,the dynamic state estimation of the high-dimensional linear dynamic system based on the Koopman tuples is carried out by Kalman filter method.The estimated state variables of high dimensional linear dynamic system is restored to the original estimated state variables by using Koopman mode.Simulation results show that this method can track the actual operational state of IEGS in real time and accurately.