An Incremental Radial Basis Function Network Based on Information Granules and Its Application

Area of Science:

• Building energy modeling
• Artificial intelligence in civil engineering
• Computational intelligence for energy systems

Background:

• Accurate prediction of heating and cooling loads is crucial for residential building energy efficiency.
• Existing models like Linear Regression (LR) and standard Radial Basis Function Networks (RBFN) have limitations in capturing complex nonlinearities.
• Integrating global and local models can enhance prediction accuracy.

Purpose of the Study:

• To design an Incremental Radial Basis Function Network (IRBFN) for improved heating and cooling load prediction in residential buildings.
• To combine Linear Regression (LR) for global modeling with local RBFN for capturing nonlinearities.
• To utilize Context-based Fuzzy C-Means (CFCM) clustering for creating information granules guided by LR model errors.

Main Methods:

• Development of an Incremental Radial Basis Function Network (IRBFN).
• Integration of Linear Regression (LR) as a global model, refined by a local RBFN.
• Application of Context-based Fuzzy C-Means (CFCM) clustering for information granule generation.
• Experimental validation on energy performance data of 768 diverse residential buildings.

Main Results:

• The proposed IRBFN demonstrated superior performance in predicting heating and cooling loads.
• IRBFN outperformed Linear Regression (LR), standard RBFN, RBFN with information granules, and Linguistic Model (LM).
• The model effectively captured localized nonlinearities missed by the global LR model.

Conclusions:

• The IRBFN offers a robust and accurate approach for estimating residential building energy performance.
• Combining global linear models with local fuzzy-based RBFN significantly enhances prediction accuracy.
• The proposed method provides a valuable tool for optimizing energy management in buildings.