| With increasing demand for energy, the development and improvement of powersystem becomes more important. Power system load forecasting is very important for thesystem scheduler automation. And it has great significance for the safe, stable andeconomic operation of power system. The accuracy of load forecasting directly affects thesecurity and stability of the grid, and its predicted results provide help to the running ofgenerators, and provide the basis for power plant fuel supply plan, while the control of thesystem can be improved. The predictions’ inaccurate or errors over the general will affectthe fuel rational allocation of power generation sector, and reduce its income. Studies withhigh precision and high practical load forecasting method for development of theelectricity market and the smart grid are very essential.Through access to relevant literature, this paper describes the status of the powersystem load forecasting, analyses and compares the characteristics of different existingprediction methods. Introduce specifically the theory and learning algorithms of artificialneural network method. It is through abstracting and simulating the basic features of thehuman brain to form an adaptive parallel information processing method with the featureof self-learning and nonlinear mapping, and it has important application value, for thepower system load forecasting. This paper detailed introduce the structure and its learningalgorithm of error back propagation model (Back Propagation, BP) and radial basisfunction neural network model (Radial Basis Function, RBF), and established the modelsbased on BP neural network and RBF neural network for power load forecastingrespectively. Through the process of establish model, preprocess the input of the raw datafirstly, then remove bad data and add missing data;the input sample need normalizationprocessing to avoid the saturation of neurons; analyze the selection of initial weights andlearning parameters in the model last. Comparing the two established models, theestablished BP neural network model’s required learning and training time is longer, andits convergence is poor, besides it is easy to fall into local minimum conditions;the RBFneural network model’s train speed is faster, and has good convergence, so it has a greater advantage for the power system load forecasting.Then，this paper introduced the fuzzycontrol theory. Using the fuzzy control theory methods, we can achieve control of complexsystems without establishing a precise mathematical model. Concretely introduced thestructure and design process of fuzzy controller, including the selected of input variablesand fuzzy reasoning and judgment. Using fuzzy control theory to adjust and enhance RBFneural network model, and improve its convergence rate and decrease its training time.Establish the power system load forecasting model combining RBF neural network andfuzzy control.Using the established BP neural network, RBF neural network model and the modelcombining RBF neural network and fuzzy control, this paper predicted the actual load onan are, then analyzed and compared the errors of results. The resulting accuracy of the twomethods are able to meet the actual requirements of the power sector, so illustrate theeffectiveness and practicality of these methods. But the obtained error using the model ofcombining RBF neural network and fuzzy control is smallest, and predictive effect is better,so it indicated that this method for power system load forecasting has practicalsignificance. |
