Research on Economic Demand Forecasting Model of Power Industry Based on Deep Learning Algorithm

This study proposes a hybrid forecasting model based on deep learning, aiming at the challenges of nonstationary load series, multi factor coupling and real-time edge computing in power industry economic demand forecasting. Build an end-to-end framework of “data decomposition feature fusion time series modeling intelligent optimization”, which separates the load sequence into trend, period, and noise components through variational mode decomposition. Use onedimensional convolutional neural network to extract local correlation features from multi-source data, combine bidirectional long short-term memory network to capture bidirectional time series dependencies, and introduce attention mechanism to dynamically weight key factors such as meteorological parameters and economic indicators. Implementing global optimization of hyperparameters based on whale optimization algorithm, while achieving model lightweighting through parameter quantification and knowledge distillation techniques. The experimental results show that the model achieves a root mean square error of 1.8MW and an average absolute percentage error of 0.3% on the test set, significantly improving the accuracy compared to traditional ARIMA models, and enhancing robustness in extreme weather, holidays, and other scenarios. After lightweighting, the model volume is compressed by 78.9%, and the inference delay is reduced by 33.3%, balancing prediction accuracy and engineering deployment efficiency. The research provides a solution that combines theoretical innovation and practical value for demand forecasting in the power system. In the future, cross regional transfer learning, new energy feature fusion, and model interpretability technologies can be further explored to promote the digital transformation and intelligent upgrading of the power industry.