Research On Collaborative Knowledge Mining Technology And Intelligence Load Forecasting Method

Power industry is one of the most important basic industries in the energy field of our nation, it is the lifeline of national economy, and the economic development follows the electricity development. The power plays a very important role in China's economic construction, national security and social stability. Accurate load forecasting has an important significance of power generation plan, the reasonable development plan of the distribution system, reducing rotation storage capacity, avoid major accidents and prevent risks. However, load forecasting is a very complex problem because of the influenced factors includes such as economic factors, irregular event, date, season, weather and other non-numeric description factors. If we don't consider these factors, the load forecasting accuracy will not be improved further. In this paper, a collaborative model based on knowledge mining and intelligent load forecasting model is present. Through using the knowledge mining technology to deal with the factors, the relevancy of the history load data and the forecasting goal is discovered. The highly similar load pattern will be extracted in historical data, the intelligent load forecasting methods will be used to forecast the load, combined with adding the appropriate text mining interventions, large errors can be further overcomed. The model can make a breakthrough to improve prediction accuracy. The major work carried out is as follows:(1) The criterion of data storage and the corresponding pre-data specification method is present. Based on the variables classification, the different data storage criterions are constructed and the corresponding data views are also constructed.(2) A combined text mining classification techniques and BP neural network load forecasting model is established. When the data only contains load data, the similarity of load curve is calculated at first, and then the BP neural network error correction is used to get higher prediction accuracy. If the data contains weather data availables for analysis, the cluster technology is used to classify the load curve, and the desion trees technology is used to find the corresponding knowledge rules, in the classification process, the rough set can be used to reduce the attributes. Based on the classification data, the different BP neural network models are trained for load forecast. According to the rules, the appropriate data can choosed for prediction. When the BP neural network model is trained, a simple adaptive method is used, and the adaptive network can automatically determine the number of nodes in the hidden layer without human experience.(3) An adaptative SVM long-term load forecasting model is proposed with the differential evolution algorithm. For long-term load forecasting, the number of sample data is far less than the short-term load forecasting, SVM can effectively forecast in this situation. The experiment proves the differential evolution algorithm can select corresponding parameters, which can effectively improve the accuracy of long-term load forecasting.(4) A novle model combined time series forecasting techniques, support vector machine methods and knowledge mining correcting technical of daily maximum load forecasting method is proposed. Daily maximum load forecasting not only need consider the impact of meteorological factors, but also need consider different types of dates and the effects of irregular events, the proposed method not only can treat the trend of time series, but also non-linear factors and irregular effects. The experimental results show that the method can effectively improve the accuracy of load forecasting.(5) The early warning indicators for load monitoring and meteorological disaster are proposed. Based on above load forecasting methods, the short-term daily load curve deviation degree and long-term supply and demand ratio are present, then the three typical climate monitoring conditions including ice disaster, dust storms and typhoons are also given. In the ice disaster conditions presented, the decision tree classification technology is used to pick out the important monitoring industries of national economics.(6) The collaborative knowledge mining technology and intelligence load forecasting method system is studied. Based on the above models, it is unlike the traditional methods and other improvements based on algorithms developed for load forecasting system.