Power Load Forecasting Method Based On Neural Network Research And Implementation

Over the past decade, with the rapid development of the national economy, China’selectric power system also will be facing a growing challenge. Given the special natureof electricity, load forecasting has become the important work of the schedulingoperations of power systems. However, traditional load forecasting method is toestablish the linear hypothesis on the basis of its low prediction accuracy. It is difficultto meet the requirements of the electricity sector. How to build effective predictivemodels to improve short-term load forecasting accuracy to improve power system hasbecome a top priority in the forecasting. Artificial neural networks have been applied inpower load forecasting because of their own adaptive self-learning, high fault tolerance,and a series of advantages. It has achieved the ideal results. Actually, the discrete signalacquisition to the historical load data are influenced by various factors, like signalinterference and noise shortcomings. People wonder that how to maximize the reductionof noise and how to greatest degree of approximation of the original signal. They have adirect impact on the accuracy of power load forecasting. In addition, according to thesome of the existing problems, to develop a framework for flexible, scalable,platform-independent integration of cities power load forecasting software system alsohas an important value.In this thesis, the main work and innovation are as follows:(1) Comparing of traditional forecasting methods, artificial neural networks have theadvantage in the analysis of nonlinear problems; we propose a load forecasting modelbased on advanced BP neural network. We design the input layer, hidden layer andoutput and implement a load forecasting model based on neural networks.(2) Actual power load data signal have the problems of strong interference, widespectral range.We propose wavelet denoising algorithm to make the problems for theactual power load signal.(3) We design and develop the load forecasting system---GVMS Load Forecasting System. We achieve the interface of data access and algorithm interface. We alsoachieve the database platform independence through the use of Hibernate technologyand GIS thematic map.(4) Finally, according to some of the problems encountered in the actualdevelopment process, as well as some ideas of my own, we made some directions forimprovement. They need to be further researching and verification.