Quantum-Inspired Deep Learning Model for Short-Term Load Forecasting in Smart Grid Energy Management

This study introduces a novel Quantum-Inspired Long Short-Term Memory (QI-LSTM) model for improved short-term electricity load forecasting. QI-LSTM combines quantum-inspired feature transformations with traditional LSTM structures to enhance prediction accuracy by producing richer entangled feature representations. Using hourly electricity demand and weather data from ISO New England from April 2019 to December 2024, the model performance was thoroughly compared with standard LSTM, stacked LSTM, and enhanced stacked LSTM models. This approach features a quantum-inspired mixing function based on the Hadamard transformation, which creates superpositions of the original inputs while maintaining their norms. The results indicate that QI-LSTM outperforms the other models, with a global Mean Absolute Percentage Error (MAPE) of 2.8044%, Mean Absolute Error (MAE) of 370.6929 MW, and R² of 0.9628 achieved during testing on the year 2024 data. The quantum-inspired transformation was highly beneficial in improving the feature representation by capturing complex dependence and nonlinear patterns in the power demand data. This study has shown the usefulness of quantum-inspired techniques in energy management and the development of more accurate and reliable forecasting models to improve the operation of power systems.