Research And Implementation Of Short-term Load Forecasting In Micro Grid

With the development of society,the shortage of power resources is becoming more and more prominent.How to make the limited power resources play the greatest value is the problem we need to study.At the same time,the application of Microgrid technology is more and more extensive,especially in the optimal dispatch and use of power resources,which has great advantages.The main power load forecasting methods in Microgrid are long-term load forecasting,medium-term and long-term load forecasting,short-term load forecasting and ultra-short-term load forecasting.Short-term load forecasting refers to the load data forecasting for the next 24 hours,which is the most important in Microgrid.Therefore,this thesis focuses on the short-term load forecasting algorithm in Microgrid.Short-term power load forecasting has the problem of few characteristics and low correlation between the forecasted load and some data in the training set.Based on the above problems,the main research work of this thesis is as follows:Firstly,the feature engineering of load forecasting for Microgrid is established,and three kinds of features are extracted according to the user's load characteristics: structural features,transformation features and frequency features,and the extracted features are filled with missing values and standardized processing,thus expanding the number of features and obtaining more effective features.Secondly,a multi-labeling algorithm based on K-means and K-nearest neighbors is proposed.The main idea is to divide the data into K clusters by K-means algorithm,and then find N nearest neighbor data of the data to be predicted in the cluster by K-nearest neighbor algorithm.According to the distribution of the number of the N data in the cluster,a set of weights can be obtained.Through these weights,the data to be predicted can be obtained.Similarity matrix with different data clusters makes a weight relationship between the samples to be predicted and the training samples.Finally,a short-term load forecasting algorithm based on multi-label learning is proposed.The main idea is as follows: after obtaining the correlation matrix between the sample to be predicted and the cluster by using multi-label algorithm based on K-means and K-nearest neighbors,when the correlation is equal to 0,the training samples in the corresponding cluster will not be used to train the BPNN model,and this part of the data will not be used for the prediction sample.When the correlation degree is not equal to 0,the training samples of the corresponding clustering cluster will be used to train the BPNN model.Then,the model trained with this part of the data is used to predict the power load of the forecasting sample.The load data obtained is multiplied by the correlation degree,and the multiplied result is taken as the load forecasting result of the cluster pair of forecasting samples.Finally,the forecasting results of K clustering clusters are added up.The value is taken as the total load forecasting result.