Research On Load Data Processing Method In Distribution Systems

Highly accurate load data are the base of implementing kind of function modules in distribution management system (DMS). At present, plenty of sets that possess functions of measuring and noting have been fixed to 10kV distribution transformers. It is a new problem to research personnel how to use collected load data build a whole and accurate database. In this paper, new data processing ideas have been proposed based on several aspects, such as non-measurement load estimation, meter placement, bad data identification and revision. Satisfied results are obtained. The major contributions of this thesis are detailed as follows:Firstly, a new idea that adopts state estimation algorithm to revise non-measurement load data in distribution systems is presents. Test results show the method can efficiently make the accuracy of pseudo measurements reach or approach that of real measurements.By improving the measurement transform technique, a new method for dealing with voltage measurements is proposed on the basis of the branch-current-based distribution system state estimation algorithm so that it can increase measurement redundancy and enhance estimation precision of measurements. Aiming at the problem that the WLS algorithm is prone to failure as pseudo-measurement load data include gross errors, a new robust estimation algorithm is puts forward for resisting bad influence of gross errors. In engineering application, the effect that leverage-measurements come into bring on the load estimation issue should not be ignored. A comprehensive robust approach is presented to solve the problem of leverage-measurement gross errors in distribution systems.Secondly, the relationship between measurement precise (or measurement variance) and total variance of estimation error is analyzed in details. Results show that the total variance of measurement estimation errors will be dropped and measurement estimation precision will be advanced, if a non-measured point is replaced as a measured point. On the base of the theory, a meter placement method based on sensitivity genes is developed. Test results show this approach is simple and practical.Thirdly, an identifying bad data technique based on wavelet singularity detection theory for distribution systems is proposed for solving the problem that doubtful bad data of measuring nodes make non-measurement load estimation invalid. This method checks up the modulus max values of the wavelet coefficients on the first scale by hypothesis testing, and then identifies doubtful bad data points according to the definition of modulus max value line. Test results show that the new method has higher veracity and better practicability.Fourthly, a short load forecasting method is adopted to correct doubtful power data after they are eliminated in real-measurement data. This paper proves compactly the conclusion that the first datum of original data lists doesn't work in the course of building GM(1,1) model.Further, A zero addend GM(1,1) combining forecasting approach that takes into account the first datum is developed. This method selects two original data lists to forecast two values, and combines them to obtain final result by relative-degree analysis approach. Test results show that this method can improve the forecasting precise and make it satisfy the request of practice application.Fifthly, apply the above theories in practice and develop the software of load data processing aiming to the running analysis and automation manage system of 10kV overhead line network.Lastly, conclusions are made on the base of research outcomes, and directions for future research are pointed out.