Study On Visualization And Forecasting Method Of Massive Power Load Data

With the improvement of the intelligent level of the power grid,the measurement data collected and transmitted by the power system will increase explosively.The storage,transmission,mining and analysis methods of a large number of user load data have become a research hotspot in power system industry.According to the results of power system load forecasting,the corresponding generation plans and dispatching schemes can be formulated in advance,which is beneficial to improve the stability and economy of power grid.Therefore,it is necessary to conduct in-depth research on load forecasting methods.In recent years,the rapid development of deep learning provides advanced research ideas for visualization,analysis and prediction of massive power load data.The main research findings of this paper are as follows.(1)The data has been cleaned,including missing value filling,outlier processing,repeated value deduplication,normalization and other operations,providing a more reliable data set for the following research;The visualization method of load data is proposed,and the mutual conversion between data and image is realized.The visualization of long-term load pattern and the preliminary compression of data are realized by power data image;Based on the embedded zero-tree wavelet image coding,the zero-tree structure and successive approximation quantization are combined to realize the compression and reconstruction of power data image;Power data images of typical categories are generated,and the embedded zero-tree wavelet algorithm and the original wavelet transform algorithm are used for simulation verification.Then the size of compressed power data image and original data are compared.(2)A comprehensive model based on the improved VGG16 and K-means is established.The power data image is used as the input of VGG16,the fully connected layer of VGG16 is improved,and the feature extraction of power data image is completed;The extracted image features are used as the input of K-means,and the gap statistics are used to select optimal K value of clusters automatically.K-means is improved to realize the classification of the load image;Simulation of the classification of users in the dataset is carried out,and the classified categories are analyzed.(3)A load forecasting model considering load categories is established.Taking time series data as input,short-term load forecasting is carried out by using gate recurrent unit;Taking the image as the input,the idea of image completion in computer vision field is adopted.The power data image is outpainted by using the training network of generator,local discriminator and global discriminator,which realized medium-term load forecasting;The results of two methods based on time series analysis and image analysis are analyzed.Three different training sets are used to train Gate Recurrent Unit,and the prediction accuracy with different training sets is analyzed.Therefore,the advantages of load forecasting based on classification are verified.The outpainted image is transformed into time series data using the improved Generative Adversarial Networks trained by power data images of users in categories,and the validity of this load forecasting method is verified.