CNN-Attention-LSTM Model for Short-Term Electricity Price Forecasting Optimized by PID-Based Search Algorithm

Accurate short-term electricity price forecasting is crucial for optimizing electricity market transactions, ensuring the reliability and safety of power systems, and accelerating the adoption of renewable energy sources. To enhance forecasting accuracy, this paper proposes a short-term electricity price forecasting method based on a CNN-Attention-LSTM model optimized by a PID-based search algorithm. Considering the fact that electricity prices are influenced by multiple factors and exhibit strong nonlinearity, high volatility, and non-stationarity, the proposed approach first employs Pearson correlation analysis to identify and select features with high relevance to electricity prices. Then, Convolutional Neural Networks (CNN) are used for feature extraction and dimensionality reduction, while the attention mechanism highlights key features. Long Short-Term Memory (LSTM) networks are applied to model the temporal dependencies in the price data. Furthermore, the PID-based Search Algorithm (PSA) is utilized to optimize the hyperparameters of the model. The proposed method is validated using historical electricity price data from the Spanish power market and is compared with various baseline models and optimization algorithms. Experimental results demonstrate the effectiveness of the proposed approach.