Research On Online Verification For Clustered Smart Meters

Smart meter is one of the most important measurement instruments in smart grid.By the end of 2020,the total number of smart meters installed in China has reached 540 million and the number is still increasing.The huge deployed smart meters put forward the demand for large-scale,high real-time and high accuracy detection technology.The existing energy meter metering technology cannot meet the above needs.In recent years,with the establishment of smart grid information acquisition system,online verification technology based on big data analysis provides a solution to the above problems.The main content of energy meter online verification is the meter error online estimation.However,the existing smart meter online error estimation algorithms have poor real-time and low accuracy performance.Also,the abnormal measurement data and line energy loss in low-voltage system in real working conditions will seriously interfere the online error estimation accuracy.The existing methods cannot meet the verification requirements of batch smart meters.This paper aims to establish a smart meter online measurement model based on the cluster meter distribution structure and estimation accuracy interference causes such as abnormal measurement data and system energy line loss.Different algorithms have been proposed to estimate meter error,detect abnormal meter and analysis the system line energy loss.This paper makes a depth theoretical and experimental research on the above methods.The main research contents are as follows:Aiming at the problems of poor real-time performance and low accuracy of meter error online estimation algorithms,an error online estimation algorithm based on variable attenuation factor weighted recursion is proposed to solve the above problems.By using the energy measurement data in different periods,the error matrix equation is constructed based on the characteristics of cluster topology distribution structure of smart meters and the system energy conservation principle to estimate meter error.The error estimation value can be obtained by solving the matrix equation.By assigning weights to the measurement data in each period and constructing a variable attenuation factor based on the condition number of the coefficient matrix,the weights are updated in real time to improve the accuracy of equation solving and error estimation.The iterative recursive idea is also introduced to update the error estimated value.The error of electric energy meter can be estimated immediately after inputting new metering data to the system,which improves the real-time performance of error online estimation.The test results show that when the abnormal data and line energy loss are not considered,for the system with 100 submeters,the root mean square error and average absolute percentage errors of the error estimated values obtained by this method are 0.029% and 2.33%respectively.Compared with the traditional least square method and iterative method,the root mean square error is reduced by 0.013% and 0.032% respectively,the average absolute percentage errors values are reduced by 1.33% and 0.68%.The accuracy of error estimation reachs to 97.79% in ideal conditions.Aiming at the problem that various types of abnormal measurement data in low-voltage power grid affect the accuracy of online error estimation under actual working conditions,an abnormal meter data detection method based on heuristic deep Q-learning is proposed.In this method,the abnormal data detection problem is modeled based on Markov decision process.A designed heuristic action learning module is introduced into the deep Q-network to guide the output of the Q-learning algorithm based on a prior knowledge to improve the learning ability of the abnormal data characteristics.At the same time,the depth Q network is improved to accelerate the convergence speed and improve the learning efficiency of the algorithm.The experimental results show that the precision of detecting mixture abnormal data is 95.36%.When the anomaly coefficient is less than 0.04 and the number of submeters is no larger than 100,the accuracy of the error online estimation algorithm is improved from 84% to 94% compared with without this algorithm.The accuracy is improved by 11.9%,which reduces the impact of abnormal data on the error online estimation.Aiming at the problem that the line energy loss in low-voltage grid affects the accuracy of online error estimation under actual working conditions,a line energy loss analysis method based on parallel hybrid depth neural network is proposed.In this method,only the electric energy measurement data of the electric energy meter is used as the input,and a parallel deep neural network model based on deep belief network and long-term and short-term neural network is constructed.The nonlinear fitting characteristics of deep belief neural network are used to analyze the complex relationship between meter measurement data and line energy loss,The long-term and short-term neural network is deployed to analyze the relationship between energy loss sequences in different periods from the time dimension,so as to reduce the impact of single data type on line loss estimation;Finally,in order to improve the accuracy of line energy analysis,the final line energy loss estimation is obtained by integrating the output results of the above two networks.The test results show that when the line loss rate is less than 4% and the number of submeters is 100,the accuracy of the online error estimation algorithm is improved from 76% to 93%after using this method.The accuracy is improved by 22.4%,which reduces the impact of line energy loss on the online error estimation.This paper analyzes the overall experimental results of the algorithm after using the proposed error online estimation,abnormal data detection and line energy loss analysis algorithms.The experimental results show that when the number of submeter is 100 and the comprehensive loss rate is less than 10%,the error estimation deviation interval is [-0.09%,0.66%].When the comprehensive loss rate is less than 8%,the estimation accuracy of the overall algorithm is above 90%.The performance of the algorithm is analyzed based on the measurement data.The algorithm can track the trend of error change and accurately estimate the error value of abnormal meter.The accuracy of the overall algorithm is 91.80%.The experimental results show that the algorithm in this paper has strong feasibility and practicability,which lays a foundation for the follow up application of online verification technology of smart meters.