Electric Power System Short-term Load Forecasting Based On Particle Filter Algorithm

The short-term load forecasting of power systems has become one of the significant approach for power sector to economize the electric energy and ensure the quality of electricity. The accuracy of load forecasting could reduce the cost of electricity effectively and increase the economic efficiency. Although the topic of load forecasting of power system had become the study focus of researchers for a long time, this work has never stopped. Through regression analysis and time series methods to intelligent algorithm which appeared later, people witnessed that more and more methods were applied into the area of load forecasting to pursuit the better result. Based on this viewpoint, the method of short-term load forecasting based on particle filter was proposed in this thesis and some favorable results were obtained finally.On the basis of the current international research situation of short-term load forecasting, the particle filter was applied into the area of short-term load forecasting problem. And this algorithm was improved by applying a better resampling algorithm to deal with actual problem like particle degeneration. Therefore the advantage of resolving the problems in dynamic systems that are typically nonlinear and non-Gaussian could be brought into effect more properly. The model of short-term load forecasting that is based on the particle filter is constructed in this thesis. After analyzing the load data, we found that the load that would be predicted is highly related with the load in previous time, the same type load in a week before, the load at the same moment in a day before and the average temperature of the day. Therefore those four relation parameters were chose to be the state vector for the model to update form time to time. To test the property of this model and algorithm, simulations were conducted based on the load data given by a worldwide net competition organized by EUNITE Network on1st of August,2001. The research works have been done to illustrate that the accuracy of the prediction is improved when the number of the state particles is increased within a certain range while, at the same time, the time wasted in computation is also increased highly. The reasonable number of the particles were given under the promise that the prediction accuracy is up to the mustard. Also this thesis presents the appropriate parameters region for this system to obtain the optimized result.Finally compared with kalman filter, the particle filter algorithm used in the model of the short-term load forecasting problem showed a better performance and was therefore testified to be an effective method.