Short-term Power Load Forecasting Based On Fuzzy Clustering Similar Days

Power load forecasting is an important aspect of grid management.Through accurate forecasting, the power plant can arrange the on-offoperation of machine set economically and reduce the spare capacity, soto lower the cost of power generating and improve the profits. Thus,seeking an efficient way to forecast the power load more efficientlymeans a lot to power industry.In this thesis, first introduce the background and meaning of powerindustry’s development and demonstrate the current situation of researchat home and abroad in the power load forecasting domain, especiallyemphasizing on the means of neural network. Next aiming to thedrawbacks of over-fitting and easily getting stuck into local extremes ofBack Propagation Neural Network, a combination of wavelet transformand PSO-SVM (Particle Swarm Optimization-Support Vector Machine) isput forward. PSO, which is a heuristic bionic optimization algorithm, canfigure out the parameters’ selection of SVM preferably. Finally byemploying wavelet transform, decompose the time sequences of power load into high-frequency and low-frequency parts. As to thelow-frequency parts, forecast with this model. However, to thehigh-frequency parts, put weighted average method to use. Forecastingresults of these two parts consist of the ultimate value. In order toenhance the accuracy more, here a novel method using Similar Daysbased on fuzzy clustering analysis is proposed. In order to describe theweather factors’ influence on power load forecasting more clearly, firstlyit categorized the weather factors as temperature, air pressure, wind speed,overcast day, rainy day, etc., then together with week type and day typethese factors formed the influence items. According to the items above,fuzzy rules were applied to establish the mapping table to get the factorsquantized. Next, the cluster technology was utilized to classify thecontent in the mapping table, and the similar days were chosen based onthe clustering level, which was to reduce the numbers of samples andaccelerate the speed of selection. This model takes weather and otherfactors into consideration to weaken the randomness of power loadfurther.Finally, a software of short-term power load forecasting isestablished. It mainly includes the functions of data importing, powerforecasting, similar days’ selection, error evaluation and data saving.What’s more, it can forecast the power load of future24hours, which ishelpful for power generation plan and improve the enterprise management level of technology and economic benefits.