A Multi-Strategy Enhanced Whale Optimization Algorithm for Long Short-Term Memory-Application to Short-Term Power...

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Time Series Analysis

Background:

• Accurate short-term electric load forecasting is crucial for power system security and energy efficiency.
• Electric load data exhibits inherent randomness, non-stationarity, and nonlinearity.
• Existing forecasting methods may struggle with the complexity of microgrid load patterns.

Purpose of the Study:

• To develop a novel hybrid model for precise and efficient short-term electric load forecasting in microgrids.
• To improve the optimization of Long Short-Term Memory (LSTM) neural networks for load prediction.
• To enhance the robustness and reliability of microgrid load forecasting.

Main Methods:

• Complementary Ensemble Empirical Mode Decomposition (CEEMD) was used to decompose the electric load time series.
• A multi-strategy enhanced Whale Optimization Algorithm (WOA) with chaotic initialization and reverse learning optimized LSTM model parameters.
• Individual LSTM models were trained for each decomposed component, and predictions were aggregated.

Main Results:

• The proposed CEEMD-WOA-LSTM hybrid model demonstrated high forecasting accuracy in case studies.
• The enhanced WOA effectively avoided local optima and improved parameter search capabilities.
• The method proved reliable for microgrid load forecasting using UC San Diego data.

Conclusions:

• The hybrid CEEMD-WOA-LSTM model offers a reliable solution for high-accuracy short-term electric load forecasting in microgrids.
• This approach provides a strong foundation for microgrid system planning and stable operational management.
• The integration of advanced decomposition and optimization techniques significantly improves forecasting performance.