Short-term Electricity Demand Forecasting Based On Artificial Neural Network

Short-term load forecasting is an essential work in power system operation,and control.Consequently,much research has been devoted to it.Since the historic data of electricity load is a random and non-stationary time series in nature,the prediction with no error is impossible at present.Thereby,the increasing interests are dedicated to the improvement of prediction accuracy.Owing to its high self-learning and generalization abilities,artificial neural network(ANN)has been widely used in the load forecasting and could bring about good prediction results.In recent years,the new achievements have been made in ANN and deep learning has risen as a new research topic.With the latest achievements of ANN,the short-term load forecasting is studied based on the actual historical load data of customers.The main research contents include:(1)The point forecasting model of customer load using extreme learning machine(EML)combined with an improved particle swarm optimization(IPSO)algorithm is presented.Due to the fact that the input weighting matrix and hidden layer offset of EML are initialized by random assignment,EML usually leads to the poor generalization ability and forecasting performance.So,the powerful global searching ability of IPSO algorithm is applied to find out the optimal input weighting matrix and hidden layer offset of EML.With the actual historical load data of customers,the experiment simulations are conducted to verify the validity of the model.(2)Based on the architecture of deep learning,the point forecasting model of long short-term memory(LSTM)network with the embedding layer and that with both the embedding and convolution layers are proposed.With the actual historical load data of customers,the two models are respectively demonstrated and compared with the model using EML combined with IPSO.(3)Aiming at describing the uncertainty of load data,the interval prediction model of customer load by using EML combined with IPSO is proposed.Also,an improved scalar method based on the point prediction results is introduced to generate the prediction interval,and the corresponding case studies on the load forecasting of actual customers are given.