Research On Short-term Bus Load Forecasting Based On Deep Learning

Accurate bus load forecasting is the basis for achieving lean and intelligent power system dispatching,and it is also an important basis for the future electricity market to formulate timeof-use electricity prices.Compared with the system load,the power supply range of the bus load is smaller,so it is easily affected by the changes in the power consumption mode in the power supply area,and has the characteristics of large time-varying fluctuations and many sudden changes.Therefore,the accuracy of the bus load prediction is not as good as the system load.With the advent of the era of power big data,more and more power information can be used for bus load forecasting.The scientific use of relevant data will have a positive impact on improving the accuracy of bus load forecasting.The vigorous development of deep learning provides effective technical conditions for the application of big data in bus load forecasting.Focusing on the problem of bus load forecasting,this paper conducts research on related information processing technologies and forecasting methods based on deep learning;and takes the actual power network in a certain area of Guangdong Province as an example to verify the relevant research effectively.First,the research work on feature selection and bus load forecasting under multi-source data is introduced.According to the historical load data of the bus,the Pearson correlation coefficient is used to analyze the influence of different historical time data on the current value,and based on this,a reasonable set of strong correlation historical time is determined.In order to scientifically determine the multi-source variables introduced into the deep network,the Xgboost algorithm is used to determine the importance of different variables(features)to the prediction results;then feature selection is performed according to the importance of each feature,that is,the best feature set for network input is determined.The filtered features are input into the long short-term memory network and built into a predictive model through training.Then,the feature extraction technology of bus load information and the corresponding load prediction work are introduced.In order to extract the latent information contained in the bus load data,an autoencoder-based feature extraction model is designed.The extracted feature vector is spliced with other features as the input of the long short-term memory network,and a mixed prediction model(AE-LSTM)is constructed.In the study,the autoencoder was built using a long short-term memory network.Finally,a bus load forecasting study with renewable energy integration is discussed.In the study,the bus load of this type of bus is decomposed into two parts:electricity load and renewable energy generation.Therefore,the prediction of this kind of bus load can be transformed into:first predict the two separately,and then algebraically superimpose to obtain the predicted value of the bus load.Taking the injection of photovoltaic power plants as an example,a model based on k-means clustering and LightGBM algorithm is designed to predict photovoltaic power generation.The electricity load is predicted by the AE-LSTM model.