Research On Power Load Short-term Forecasting Of Tianjin Regions

The accuracy of short-term power load forecasting is the basis for the planning of power system operation.Excessive forecast errors result in waste of power and result insocioeconomic loss and damage to electrical equipment due to insufficient power supply.How to improve the accuracy of short-term power load forecast and accurately grasp the power supply of district has become an important work of the Tianjin Power Grid Corporation.Due to the characteristics of the strong non-linearity and high frequency,it is greatly troubled by short-term forecasting of power load.Therefore,the paper summarizes the methods of power load forecasting and analyzing the law of daily load characteristics and weekly load characteristics,the holiday date is the main influencing factor of short-term load changes,put the original load sample divide into non-holiday Sample set and holiday sample set to predicte.In the aspect of model establishment,this paper conpare with the results of non-holiday sample set that use the generalized regression neural network,wavelet neural network and Elman neural network.Final,the paper choice Elman neural network and use the twoparameter joint estimation method construct Elman neural network prediction model.Due to the gray system theory has a good ability to deal with poor information.This paper chooses the GM(1,1)model to predict the holiday sample set.In the aspect of data preprocessing,this paper proposes a threshold reduction method suitable for non-holiday sample set,it was improved the data level processing method.In this paper,the method is applied to the noise reduction of non-holiday sample data.Compare with the threshold reduction method result and data level processing method result,it prove by comparison,the threshold reduction method is effectively.Finally,the simulation results show that the threshold noise reduction method can reduce the prediction error.At the same time,the prediction after the classification of the sample can effectively improve the short-term prediction of the power load.