Research And Design On Recognition System Of Electricity Customer Load Pattern

In the electricity market, the integration of power and electricity should be achieved to improve the energy efficiency and make full use of existing assets. Thus the load demand characteristics must be taken into account in the planning, designing and operating of power system. Load pattern extraction and users’ consumption pattern recognition based on customers’ consumption data are extremely the most important work. Based on this, load control, electric anomaly detection (measurement, stealing etc.), power marketing (segmentation of customers and markets) can be developed and implemented. Therefore, electricity customers’ load pattern extraction and classification are of theoretical significance and practical value for the grids’ safety, reliability and economy.This thesis focused on the design of load pattern extraction and recognition system. It has given the steps of extracting and identifying load profiles with clustering algorithms, and also given the functions of each module. Simultaneously, the paper has studied the problems corresponding to all aspects of each module. Including:(1) for the data preprocessing module, the effects of the preprocessing methods used in the clustering procedure with clustering algorithms were studied. To find a better preprocessing method for clustering, we compared the clustering results based on three sides, they are the number of samples incorrectly clustered, running times and average accuracy rates and the IRIS dataset and the actual load profiles are used.(2) for the load patterns extraction module, in order to solve the problem that the number of load patterns is difficult to determine, we has derived a formula to determine the optimal number of load patterns based on a common cluster validity index MIA. Also, the formula’s validity is verified through numerical examples.(3) for the user classification module, due to randomness and time-varying properties of the load data, completely accurate load models are difficult to be determined. We established a load pattern recognition model based on data pre-processing, clustering, clustering validity assessment and decision tree. First the load data are transformed into load characteristic indicators and the characteristic indicators are clustered, and then discreted. Last the discrete characteristic indicators together with class indexes were taken as the inputs of ID3algorithm to build a decision tree. The load patterns were interpreted by the decision tree. Then the rules of interpretations were saved into the knowledge database which provided basis for the patterns recognition. The case shows that the method can effectively determine the categories electricity customers belonged to and achieve fast classification.Finally, based on the design and the research about the above issues, the algoritlims and the functions of each module were programmed and implemented with MATLAB software. Functions of the system were demonstrated by the application of load pattern extraction and recognition in non-technical loss detection of the distribution network.