Research On City Microgrids Short-term Load Forecasting

Microgrids short-term load forecasting is the important precondition for the microgrids to achieve its security, energy saving and efficient operation and is the prerequisite and fundamental basis for the optional microgrids energy management. Its prediction accuracy directly affects supply plans of power generation, power quality and power market transactions of microgrids and large power systerms. The small base load, high fluctuation and big randomness of the microgrids make its short-term load forecasting different from the large power system’s and lead to difficult load forecasting. However, the foreign and domestic research of the microgrids short-term load forecasting is mainly for the field of micro power sources forecasting and is few or far between on the user side. Therefore, it is of importance to study on urban microgrids short-term load forecasting.This paper introduces the least square support vector machine(LS-SVM) algorithm theory, and then points out the advantages of LS-SVM used on short-term load forecasting. In view of various shortcomings found in LS-SVM short-term load forecasting model, it also puts forward one improved LS-SVM microgrids short-term load forecasting.In order to determine the input vector for the improved forecasting model, this paper accomplish a sequence of operations on the original load data, including three areas: make up for the missing data,horizontal smoothing and vertical smoothing. And, it also completes the quantitative analyses of the major impact on the microgrids short-term load forecasting, including the day type, temperature, weather patterns and wind. While, a new and more objective reasonable detailed quantitative method of day type is provided. At last, normalization processing correlative is caarrid out.In order to choose the trainin samples, a bidirectional weighted similar days trainin samples selecting model is proposed, which is based on load point scale. With consideration of cumulative effects of the weather factors, the characteristic of continuity and periodicity of short-term load and saturation effect of time distance, day character similarity as the transverse weights is combined with the local shape similarity of a couple of days before in the function of similar days, in which the time factor as the longitudinal weights is also considered.In order to ensure the learning capability and the generalization ability in the same time, one mixed kernel function is used in the proposed forecasting model. Meanwhile, particle swarm optimization(PSO) algorithm is used in optimizing the model parameters.The examples show that microgrids short-term load forecasting is different from the large power systems. Mealwhile, the results demonstrate that the selection of the trainin samples, the choice of kernel function and parameter optimization are reasonable. This paper also proves thatthe short-term forecasting effect can be improved by using the method proposed in this paper, and that the proposed method is more appropriate for the microgrids short-term load forecasting and is of great value in practical application.