The Power Load Forecast Research And Application In Local Area

Power load forecasting is to establish a suitable mathematical model for the power consumption of a certain period of historical time in this area,in order to predict and analyze the power consumption of a certain time period in the future.Power load forecasting has a great connection with social and economic development,situation policy,seasonal climate and other factors.Power load forecasting affects the planning and operation,etc.of power systems.With the development of the power market,short-term power load forecasting provides a vital guarantee for the development of making reasonable electricity prices and the economic operation of the power grid.The power load forecast in Yingkou area started late,the original load forecasting technology used the trend extrapolation method and the exponential smoothing method for short-term load forecasting,while its low accuracy could hardly meet the requirements of the power sector.Through systematically analyzing,to list and statistic the historical data of electric load in Yingkou area.By comparing various load forecasting methods,summarize the variation law of load and explore the suitable method for power system load forecasting in Yingkou area.Both the neural network and the support vector machine can perform power load prediction,and the predicted result can guide the actual grid operation.This paper introduces the concept and basic principles of neural networks first,focusing on studying the model structure,learning rules and training process of neural networks.According to Yingkou historical power load data,analyze the operation of Yingkou power and conduct training.In order to compare the power load forecasting,the concept and principle of the support vector machine are introduced,and the support vector machine is used for regression prediction.The power load prediction results obtained by the two methods are compared to analyze the advantages and disadvantages of the two methods.Compared with the results of neural networks prediction,the results of support vector machine regression performs more generalization and show better effect when small amounts of data tests.While the model of neural network training requires more training time,and the obtained results sometimes are over-fitting,but it has a better performance for large amounts of data.In practical applications,it is necessary to select an appropriate regression prediction algorithm according to the amount of data.We use ensemble learning and combination forecasting methods to further reduce the bias and variance of forecasting results.