# An Enhanced Prediction Model for Energy Consumption in Residential Houses: A Case Study in China

## TL;DR

This paper introduces a new energy consumption prediction model for Chinese rural homes using an improved optimization algorithm and machine learning to support sustainable building design.

## Contribution

The novelty lies in the integration of an enhanced Bio-inspired Black-winged Kite Optimization Algorithm with Support Vector Regression for improved energy consumption prediction.

## Key findings

- The improved IBKA algorithm outperforms original BKA and other optimization methods in convergence accuracy and global search performance.
- The IBKA-SVR model reduces key error metrics by 37–40% and achieves an R2 of 0.9792, showing superior predictive accuracy.
- SHAP analysis identifies insulation thickness and window-wall ratios as dominant factors affecting energy consumption with non-linear relationships.

## Abstract

High energy consumption in Chinese rural residential buildings, caused by rudimentary construction methods and the poor thermal performance of building envelopes, poses a significant challenge to national sustainability and “dual carbon” goals. To address this, this study proposes a comprehensive modeling and analysis framework integrating an improved Bio-inspired Black-winged Kite Optimization Algorithm (IBKA) with Support Vector Regression (SVR). Firstly, to address the limitations of the original B-inspired BKA, such as premature convergence and low efficiency, the proposed IBKA incorporates diversification strategies, global information exchange, stochastic behavior selection, and an NGO-based random operator to enhance exploration and convergence. The improved algorithm is benchmarked against BKA and six other optimization methods. An orthogonal experimental design was employed to generate a dataset by systematically sampling combinations of influencing factors. Subsequently, the IBKA-SVR model was developed for energy consumption prediction and analysis. The model’s predictive accuracy and stability were validated by benchmarking it against six competing models, including GA-SVR, PSO-SVR, and the baseline SVR and so forth. Finally, to elucidate the model’s internal decision-making mechanism, the SHAP (SHapley Additive exPlanations) interpretability framework was employed to quantify the independent and interactive effects of each influencing factor on energy consumption. The results indicate that: (1) The IBKA demonstrates superior convergence accuracy and global search performance compared with BKA and other algorithms. (2) The proposed IBKA-SVR model exhibits exceptional predictive accuracy. Relative to the baseline SVR, the model reduces key error metrics by 37–40% and improves the R2 to 0.9792. Furthermore, in a comparative analysis against models tuned by other metaheuristic algorithms such as GA and PSO, the IBKA-SVR consistently maintained optimal performance. (3) The SHAP analysis reveals a clear hierarchy in the impact of the design features. The Insulation Thickness in Outer Wall and Insulation Thickness in Roof Covering are the dominant factors, followed by the Window-wall Ratios of various orientations and the Sun space Depth. Key features predominantly exhibit a negative impact, and a significant non-linear relationship exists between the dominant factors (e.g., insulation layers) and the predicted values. (4) Interaction analysis reveals a distinct hierarchy of interaction strengths among the building design variables. Strong synergistic effects are observed among the Sun space Depth, Insulation Thickness in Roof Covering, and the Window-wall Ratios in the East, West, and North. In contrast, the interaction effects between the Window-wall Ratio in the South and other variables are generally weak, indicating that its influence is approximately independent and linear. Therefore, the proposed bio-inspired framework, integrating the improved IBKA with SVR, effectively predicts and analyzes residential building energy consumption, thereby providing a robust decision-support tool for the data-driven optimization of building design and retrofitting strategies to advance energy efficiency and sustainability in rural housing.