Short-term And Ultra-short-term Electrical Load Forecasting Based On IWOA-Bi-LSTM-Attention Model

With the development and progress of society,there is a high demand for electricity throughout social life.Electricity load forecasting is currently an important area of research,as load forecasting exists in all aspects of the power system.Excellent power load forecasting models are not only effective in improving the resource utilization,but also enhance the flexibility of loads in the power market,which is vital to the power system.In order to improve the prediction accuracy,a bi-directional long and short-term memory network(Bi-LSTM)based on improved whale optimization algorithm(IWOA)and attention mechanism is proposed for power load prediction.The method receives input variables,builds a model to learn the dynamic change pattern within the data variable features,introduces the attention mechanism on the basis of Bi-LSTM,gives differential weights to the states of the hidden layer of Bi-LSTM,and independently mines the correlation between the power load output and each feature to reduce the loss of historical information and enhance the this fixes the shortcomings of traditional methods that rely on expert experience to select thresholds.For neural networks,the selection of hyperparameters has a great impact on the performance of the model,so the setting of hyperparameters is important.In the thesis,the hyperparameters of this model are optimized by improving the whale optimization algorithm.For the model proposed in this thesis,the simulation experiment is carried out,and the short-term power load forecasting are respectively carried out by using the corresponding real data.BP neural network,signal LSTM neural network,LSTM network based on WOA,Bi-LSTM network based on attention mechanism and Bi-LSTM network based on WOA and attention mechanism are used as the benchmark model of this thesis.In this thesis,based on the load data of the 2016 Electrician Mathematical Modeling Competition and the data of related influence factors,the benchmark model and IWOA-BiLSTM-Attention model are used for short-term electrical load forecasting and ultra-short-term power load forecasting respectively.The experimental results are that the MAPE value of the IWOA-Bi-LSTM-Attention model on short-term power load forecasting is 0.076851,and the MAPE values on ultra-short-term electrical load forecasting with four steps are 0.008838(step3),0.007264(step 6),0.006480(step 9)and 0.005218(step 15),compared with the MAPE values of each benchmark model,all have different degrees of reduction.This result proves that using the Bi-LSTM-Attention model which based on the improved whale optimization algorithm can effectively improve the accuracy of short-term forecasting and ultra-short-term forecasting,which verifies the effectiveness of the method.