Research On Load Profiling And Short-term Load Forecast In Smart Grid

With the rapid development of society, the demand for electricity is growing rapidly, and the available energy is decreasing gradually. So there is a huge contradiction between load demand and supply. Demand Side Management (DSM) is an important tool which can be used to balance the load demand and supply. With the maturity of smart grid, the dynamic Demand Response (DR) technology provides a new idea for the effective DSM, which is used to motivate consumers to cooperate for regulation of demand and supply by shifting the load from peak hours to off peak hours to achieve the balance of demand and supply.With the massive load data collected in smart grid, using data mining technique to analysis these data can better promote the effective dynamic demand response. On the ground of large scale of load data, the application of clustering algorithm in load characteristics analysis can cluster the load profiles of smart grid customers more precisely and effectively to catch the load characteristic of different clusters of customers, which can provide theoretical basis for making more effective strategy of demand response.Accurate forecast the changes of the load by using more novel prediction algorithm can support to put forward the demand response strategies and improve the sensitivity of the demand response.The research of this thesis is supported by the national power grid technology project, "Research on key technology of the network traffic analysis based on the traffic visualization". Based on the massive load data of smart grid, the related theory of data mining is applied to the load profiling and the short-term load forecasting.In this thesis, first of all, the related research directions of the demand response of smart grid and the common methods of data mining are summarized. And then, based on the large scale characteristic of smart grid load, an Adaptive Weighted Fuzzy C Means clustering algorithm based on Principle Component Analysis (PCA-WFCM) is proposed to analyze the load characteristics. The data sets based on the real smart meter readings are applied to evaluate the proposed PCA-WFCM algorithm. The simulation results show that the proposed algorithm could achieve considerable improvement both in time complexity and clustering accuracy compared with the conventional fuzzy C means clustering algorithm through comparing four clustering validity indexes. Finally, in order to improve the forecasting accuracy, an improved Radial Basis Function neural network model (PCA-WFCM- RBF) for short-term load forecast is proposed, which use the PCA-WFCM algorithm to determine the basis function centers, and use the gradient descent algorithm to train the output layer weights. The simulation results show that better forecasting accuracy could be achieved by using the proposed model comparing with the conventional RBF neural network model based on K-Means. Based on the different load characteristics of different cluster of customers, a new short-term load forecasting method based on clustering analysis is proposed to further improve the forecasting accuracy, which cluster the load profiles first, and then forecast each clusters of load, and finally sum all the forecast load and obtain the final total load forecast results. The simulation results show that the load forecasting method based on clustering is superior to direct load forecasting method with higher prediction precision.