计算机科学
集成学习
人工智能
服务(商务)
机器学习
集合预报
钥匙(锁)
数据挖掘
领域(数学)
特征(语言学)
作者
Zhongyao Feng,Aizhu Zhang,linyilin xu,Genyun Sun
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-08-01
卷期号:: 257-260
标识
DOI:10.1109/eicars71536.2026.11694741
摘要
Accurate quantification of urban cooling service supply and cooling demand is critical for refined thermal-environment governance. However, existing studies often suffer from (i) subjective indicator weighting, (ii) limited stability of single-model simulations, and (iii) insufficient interpretability. This study proposes an interpretable framework that integrates an improved three-layer stacking ensemble with SHAP-based explanations to address these limitations. Taking Hohhot-a mid-temperate arid and semi-arid continental city of China-as a case study, we build regression models to quantify cooling supply and heat-risk demand using remotely sensed land surface temperature (LST) and the modified temperature-humidity index (MTHI) as dependent variables. An out-of-fold correlation screening strategy is used to optimize the ensemble set, and weighted SHAP values are computed to estimate indicator contributions. Based on these contributions, we construct an Urban Cooling Service Supply Index (UCSI) and an Urban Cooling Service Demand Index (UCDI), and identify supply-demand matching zones via a four-quadrant model. The optimized stacking model outperforms individual machine learning algorithms. Results reveal a pronounced mismatch: blue-green spaces in the urban periphery exhibit surplus cooling supply, whereas central built-up areas face cooling shortages. The proposed framework improves the objectivity of weighting and enhances interpretability, providing scientific support for differentiated heat-mitigation planning in arid cities.
科研通智能强力驱动
Strongly Powered by AbleSci AI