Research On Combination Of Data Preprocessing And Machine Learning In Power Forecasting

Forecasting is an important part in electric power system and it is the basis of economic operation of electric power system. Forecasting is a kind of nonlinear optimization problem with complex and large scale. In recent years, many countries have devoted to the wind power generation. Although there are many mature methods in the field of wind power generation, but because of the randomness and the characteristics of the wind power series, it is difficult to establish the statistical model of historical data. Therefore, more accurate and efficient forecasting model and data preprocessing technology are important to short-term wind power.At present, there are many electric power forecasting techniques and apply to different kinds of electric power forecasting. In this field, forecasting method is divided into two categories:Based on the statistical mode and machine learning model, and the models are combine with data preprocessing.In this paper, we make an overview of different techniques, and then discuss the relationship between the planning model and the algorithm, and the advantages and disadvantages of the models. Set up a deep model SAE_BP which has three hidden layer. SAE_BP model compare with the traditional BP model and the SVM model in step 1-9 in advance on the experimental analysis, The MAPE of SAE_BP model was reduced by 10%-30% than the other two model. And then based on particle swarm optimization algorithm, the model parameter are optimized. PSO solved difficulties of artificial experience setting up the complex parameters. Finally, we put forward a kind of sample classification method based on similarity and the trend of similarity, using similar day of forecasting day organize the reference vector of model, solve the disadvantage of the multi-step prediction precision is low. The results show that the proposed data preprocessing method combine with the artificial experience processing method is effective. The results of the study have important guidance for the prediction of short-term wind power.