Short-Term Load Forecasting Method Based on Copula Correlation Measurement Combined With Attention Mechanism

Accurate short-term load forecasting is of great significance to ensure the security and economic operation of power grid. Long and short term memory (LSTM) based predictive model in machine learning is a typical method for short-term load forecasting, but it is difficult to extract the weight of entering feature. This paper proposes a combined forecast model based on bidirectional LSTM (Bi-LSTM) based on attention mechanism. Firstly, the K-SVD algorithm is used to train the historical load data samples, and an over complete dictionary with the characteristics of data classification and sample sparsity is obtained. Then, the similarity matrix of sparse representation is used to cluster the load data. In the next step, the correlation analysis of Copula function is used to quantitatively calculate the correlation between weather, electricity price and load. Finally, Bi-LSTM-Attention is used to establish the prediction model. Taking the data set of a certain region in Australia as practical example, the prediction accuracy of the method reaches 95.84%. Comparing with support vector regression (SVR), back propagation (BP) and LSTM, the proposed method has higher prediction accuracy for short-term power load.