A deep LSTM network for the Spanish electricity consumption forecasting.

Area of Science:

• Electrical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Electricity is essential for modern societies, and increasing consumption necessitates efficient smart grid management.
• Accurate short-term electricity forecasting is vital for grid stability, sustainability, and safety.
• Traditional forecasting methods often struggle with the complex, sequential nature of electricity consumption data.

Purpose of the Study:

• To propose and evaluate a deep neural network for short-term electricity consumption forecasting.
• To optimize a Long Short-Term Memory (LSTM) network using a novel metaheuristic algorithm.
• To compare the LSTM model's performance against various deep learning and traditional machine learning techniques.

Main Methods:

• A Long Short-Term Memory (LSTM) network, suitable for time-series data, was employed for electricity demand prediction.
• Hyper-parameter optimization for the LSTM was performed using both random search and the Coronavirus Optimization Algorithm (CVOA).
• The optimized LSTM model was applied to predict electricity demand with a 4-hour forecast horizon using extensive Spanish electricity consumption data (9.5 years at 10-min frequency).

Main Results:

• The LSTM network optimized with the Coronavirus Optimization Algorithm (CVOA) demonstrated superior performance in electricity demand forecasting.
• The proposed LSTM model achieved the lowest prediction error, falling below 1.5%.
• Performance was benchmarked against deep neural networks (Deep Feed-Forward Neural Network, Temporal Fusion Transformers) and traditional methods (Linear Regression, Decision Trees, Gradient-Boosted Trees, Random Forest).

Conclusions:

• The Coronavirus Optimization Algorithm (CVOA) is an effective metaheuristic for optimizing LSTM hyper-parameters in electricity forecasting.
• The proposed LSTM model offers a highly accurate and reliable solution for short-term electricity demand prediction.
• This research contributes to the development of more efficient and sustainable smart grid management systems through advanced forecasting techniques.