EMLP: short-term gas load forecasting based on ensemble multilayer perceptron with adaptive weight correction

Area of Science:

• Smart City Technologies
• Energy Systems Analysis
• Computational Intelligence

Background:

• Accurate short-term natural gas load forecasting is crucial for smart city energy management.
• Existing combined forecasting models often suffer from redundancy, limiting prediction accuracy.
• The volatility of load data presents a significant challenge for traditional forecasting methods.

Purpose of the Study:

• To develop an advanced natural gas load forecasting scheme that overcomes the limitations of existing models.
• To enhance the accuracy and stability of short-term natural gas load predictions.
• To introduce a novel ensemble multilayer perceptron (EMLP) with adaptive weight correction.

Main Methods:

• Data normalization and interpolation were applied to multi-source data.
• A window model segmented the normalized data, and abnormal data points were handled.
• An ensemble forecasting model was constructed using multiple multilayer perceptron (MLP) networks.
• An adaptive weight correction function was implemented to adjust prediction weights dynamically.

Main Results:

• The proposed EMLP method demonstrated significantly improved prediction accuracy.
• The adaptive weight correction effectively handled the volatility characteristics of load data.
• Experimental results confirmed the superiority of the EMLP method over existing state-of-the-art schemes.
• Enhanced stability in forecasting was observed.

Conclusions:

• The EMLP with adaptive weight correction is a highly effective approach for short-term natural gas load forecasting.
• This method offers a substantial improvement in accuracy and stability for smart city applications.
• The adaptive weight correction mechanism is key to successfully modeling load data volatility.