Short-term Load Forecasting Based On Data Mining Technology

Power system load forecasting is the foundation of power generation plan and power system development planning, and it is also an important basis for the safe operation in power system. With the development of power system reform, the introduction of price competition mechanism has put forward higher requirements for power system load forecasting. Load forecasting has a long history and mature forecasting theory and method, but it is necessary to improve the existing load forecasting method in order to improve the accuracy of load forecasting, thus to meet the power system requirements for the accuracy of load forecasting.With the extensive use of electric information acquisition system, we can collect a lot of historical load data from the user side, the data provides the basis for new prediction means and methods in load forecasting. Different from the traditional load forecasting based on the application of small scale data, the accumulation of a lot of new data allows us to forecast load based on the analysis of the load characteristics of different users using large scale of historical data.A load forecasting method based on data mining technology and support vector machine is proposed in this paper, this method presents a new idea for the selection of samples in support vector machines based on the existing large amount of historical data. First, the hierarchical clustering method is used to analyze the historical daily load, establish a decision tree using hierarchical clustering results and the historical weather, the week type, holiday type and other data. Then, according to the attribute of the forecast date, query the established decision tree to get a similar date, establish a SVM model based on history similar days, using the support vector machine forecasting model to forecast the load.In this paper, data for verifying new methods is the historical load data and historical weather data in a city of Zhejiang Province. In this paper, we predict the 96 points load in this region using the new method. And the results of the algorithm are compared with the traditional support vector machine algorithm. The method proposed in this paper solves the problem that the traditional model training date selection based on support vector machine can not accurately reflect the characteristics of the forecast day. Therefore, the algorithm proposed in this paper has higher prediction accuracy.Based on the principle of the prediction error can be reduced when adding up forecasting bus load, a new method is proposed in this paper. In this method we forecast the load of each bus bar in the area, accumulate the results to reduce the prediction error and improve the prediction accuracy. In this paper, the advantages of the combined load forecasting based on bus load and the principle of reducing error are introduced.