Short term Load Forecasting

Short-term power load forecasting faces inherent challenges such as non-stationarity and strong randomness. Traditional modeling approaches are limited in their ability to capture spatiotemporal features and address complex uncertainties. To overcome these limitations, this study proposes an innovative predictive framework based on a Crowned Pig Optimization (CPO)-enhanced Convolutional Neural Network\u2013Long Short-Term Memory (CNN-LSTM) hybrid model. In this framework, the Convolutional Neural Network (CNN) autonomously extracts spatial features from power load data, while the Long Short-Term Memory (LSTM) network captures temporal dependencies and long-term trends. Subsequently, the CPO algorithm is employed to optimize critical hyperparameters of the CNN-LSTM model, including the learning rate, number of hidden nodes, and regularization coefficients. This iterative global optimization process reduces the likelihood of local optima and identifies optimal configurations. Comparative experiments with multiple forecasting models demonstrate that the proposed CPO-CNN-LSTM model achieves superior performance in short-term load prediction, exhibiting high accuracy and robust stability. Evaluation metrics, including a Mean Absolute Error (MAE) of 89.0866 and a Mean Squared Error (MSE) of 18,400.8438, confirm the model\u2019s effectiveness. Moreover, the framework demonstrates strong practical applicability, enhanced computational efficiency, and excellent capacity to handle complex power system data. Overall, the proposed model not only improves predictive accuracy and stability but also reduces training time, underscoring its significant potential for real-world smart grid applications.