Daily natural gas load prediction method based on APSO optimization and Attention-BiLSTM

Xinjing Qi ¹, Huan Wang ², Yubo Ji ²

⁰College of Metrology and Measurement Engineering, China Jiliang University, Hangzhou, Zhejiang, China.

Peerj. Computer Science +

|

March 4, 2024

View abstract on PubMed

Summary

Accurate short-term natural gas demand forecasting is crucial for energy management. A new model combining bidirectional long short-term memory (BiLSTM) networks, attention mechanisms, and adaptive particle swarm optimization significantly improves prediction accuracy.

Area Of Science

Energy Systems Engineering
Artificial Intelligence
Environmental Science

Background

Growing demand for clean energy drives natural gas adoption.
Accurate short-term natural gas demand prediction is vital for operational efficiency and safety.
Existing intelligent algorithms for prediction face limitations like local optimization and insufficient search capabilities.

Purpose Of The Study

To develop a novel methodology for accurately predicting daily natural gas loads.
To overcome the limitations of existing prediction models in terms of optimization and search capability.
To enhance the accuracy, generalization, and stability of natural gas demand forecasts.

Main Methods

Utilized bidirectional long short-term memory (BiLSTM) networks for bidirectional data learning.
Applied an attention mechanism to determine and focus on critical hidden layer weights within the BiLSTM.
Employed adaptive particle swarm optimization to optimize BiLSTM network structure, learning rate, and training epochs.

Main Results

The proposed integrated model achieved a mean absolute percentage error (MAPE) of 0.90%.
The model demonstrated a high coefficient of determination (R²) of 0.99.
The combined model outperformed comparative models in prediction accuracy, generalization, and stability.

Conclusions

The novel methodology effectively enhances the accuracy of short-term natural gas demand prediction.
The integration of BiLSTM, attention mechanism, and adaptive particle swarm optimization offers a robust solution.
This approach provides a stable and accurate forecasting tool for natural gas dispatch and pipeline management.

Keywords:

Adaptive particle swarm optimization (APSO) Attention mechanism Bidirectional long short-term memory network (BiLSTM) Natural gas daily load