Day-Ahead Electricity Price Forecasting Based on GCM Filtering and Higher-Order Pooling Feature Enhancement

Abstract

Redundant complexities and inadequate representation of spatiotemporal features are included in the electricity price data. To address complex data redundancy and inadequate representation of spatiotemporal features, a gated channel mechanism (GCM) combined with a high-order pooling feature enhanced convolutional LSTM network (GCHCon-LSTM) electricity price prediction model is proposed. On the standardized processed electricity price dataset, the data vertical correlation information is expanded using gated dual convolutional neural network (GDCNN) integrated adjustment features. Redundant features are filtered using GCM. Temporal and spatial features are extracted by LSTM and ASPConv. Key information on spatial features is extracted using higher-order pooling. Combined with temporal features, the spatiotemporal feature representation is enhanced in a time-dominant and spatial manner. The prediction result is obtained by error correction. On the ERCOT Houston area electricity price dataset, compared to LSTM, CNN-LSTM, GHTnet and ILRCN, the experimental results showed that MAE, MSE and RMSE are reduced by the highest 21.50%, 29.56% and 40.18%, respectively, and the lowest 7.95%, 6.39% and 13.60%.

Keywords

• Electricity price forecasting
• GDCNN
• gated channel mechanism
• higher-order pooling enhancement
• LSTM