Empowering data-driven load forecasting by leveraging long short-term memory recurrent neural networks.

Area of Science:

• Electrical Engineering
• Computer Science
• Data Science

Background:

• The increasing integration of renewable energy sources complicates power system management.
• Accurate load forecasting is vital for grid stability, considering dynamic climatic and socioeconomic factors.
• Traditional methods struggle with the complex temporal dynamics of load data.

Purpose of the Study:

• To develop and evaluate a Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) model for precise power load forecasting.
• To assess the model's performance in capturing intricate temporal relationships in load data.
• To demonstrate the model's utility in integrating demand response with renewable energy sources for smart grids.

Main Methods:

• Implementation of a Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) architecture.
• Training and validation of the LSTM-RNN model using historical power load data.
• Evaluation of forecasting accuracy using Mean Absolute Percentage Error (MAPE) and Root Mean Squared Error (RMSE).

Main Results:

• The LSTM-RNN model achieved a 1.5% MAPE and 26.5 RMSE for hourly load forecasts.
• Yearly load estimations showed a 1.77% MAPE and 30 RMSE.
• The hourly forecasting model demonstrated superior accuracy and robustness against noisy or inadequate input data.

Conclusions:

• LSTM-RNN presents a practical and efficient solution for accurate power load forecasting.
• The model enhances operational efficiency and resilience in power systems.
• This technology is crucial for effective demand response and smart grid stability with distributed renewables.