科研管理

科研管理 ›› 2025, Vol. 46 ›› Issue (8): 178-189.DOI: 10.19571/j.cnki.1000-2995.2025.08.017

• 论文 • 上一篇    下一篇

基于机器学习的重污染企业绿色创新绩效影响因素研究

王晓岭1,2,庞梦茵1,金家华1,陶星伊1,董恒敏3   

  1. 1.北京科技大学经济管理学院，北京100083；
    2.北京企业低碳运营战略研究基地，北京100083；
    3.中国航天系统科学与工程研究院，北京100086

  • 收稿日期:2024-05-28 修回日期:2025-05-29 出版日期:2025-08-20 发布日期:2025-08-14
  • 通讯作者: 金家华
  • 基金资助:
    国家社会科学基金一般项目：“‘双碳’背景下高耗能制造业绿色转型绩效测度与环境政策研究”（22BJY138）；中央高校基本科研业务费（FRF-IDRY-23-013）；北京科技大学科技与文明中外人文交流研究课题（2025KFYB020）。

Research on the influencing factors of green innovation performance in heavy polluting enterprises based on machine learning

Wang Xiaoling1,2,  Pang Mengyin1, Jin Jiahua1, Tao Xingyi1, Dong Hengmin3   

  1. 1. School of Economics and Management, University of Science and Technology Beijing, Beijing 100083, China; 
    2. Institute of Low Carbon Operations Strategy for Beijing Enterprises, Beijing 100083, China;
    3. China Aerospace Academy of Systems Science and Engineering, Beijing100086, China
  • Received:2024-05-28 Revised:2025-05-29 Online:2025-08-20 Published:2025-08-14

摘要：    作为我国环境治理的重点领域，重污染企业绿色创新绩效的提高对于加快促进工业绿色转型发展具有重要促进作用。本文首先基于制度理论与资源基础观构建重污染企业绿色创新绩效影响因素分析框架，并采用机器学习XGBoost（eXtreme Gradient Boosting）模型预测企业绿色创新水平。在此基础上，进一步利用基于博弈论的可加性解释模型SHAP（Shapley Additive exPlanations）明确各个特征变量对企业绿色创新绩效的影响程度和作用方向。基于2011—2021年重污染行业上市公司的实证分析表明：外部环境、企业特征和公司治理维度对重污染企业绿色创新绩效具有关键影响。其中，市场型环境规制和企业规模是驱动绩效提升的最重要因素，体现出了外部制度环境和规模效应在创新中的关键作用。同时，机构投资者持股比例、前十大股东持股比例、企业年龄、市场价值、公众关注度、命令控制型环境规制以及政府补贴等次要因素的作用效果进一步强调了股权结构、市场表现以及外部激励对绿色创新的重要影响。值得注意的是，命令控制型环境规制表现出对绿色创新的抑制效应，说明强制性政策工具在激励创新的效率和灵活性方面有待提高。基于上述发现，本文进一步提出加快促进我国重污染企业绿色创新能力提升的治理对策与优化建议。 

关键词: 绿色创新绩效, 重污染企业, XGBoost模型, 可解释机器学习

Abstract:    As a key area of China′s environmental governance, the improvement of green innovation performance of heavy polluting enterprises plays an important role in accelerating industrial green transformation and development. In this paper, an analytical framework of the factors influencing green innovation performance of heavy polluting enterprises was first constructed based on the institutional theory and the natural resourcebased view. The machine learning XGBoost (eXtreme Gradient Boosting) model was also applied to predict corporate green innovation performance. On this basis, the additive explanatory model Shapley Additive exPlanations (SHAP) based on the game theory was further used to identify the degree of influence and the direction of the role of each characteristic variable on corporate green innovation performance. The empirical analysis based on the listed firms in the heavily polluting industries from 2011 to 2021 demonstrated that external environment, corporate characteristics, and corporate governance significantly influence the green innovation performance of heavily polluting enterprises. Among them, market-oriented environmental policies and firm size are the most critical factors driving performance improvement, reflecting the key role of external institutional environment and scale effect in innovation. Meanwhile, the effects of secondary factors such as institutional shareholdings, shareholding ratio of the top 10 shareholders, firm age, market value, public concern, command-and-control environmental regulation, and government subsidy further emphasized the significant impact of shareholding structure, market performance, and external incentives on green innovation. It is worth noting that command-and-control environmental regulation exhibits an inhibitory effect on green innovation, suggesting that the efficiency and flexibility of mandatory policy tools in stimulating innovation needs to be improved. Based on the above findings, this paper further proposed governance countermeasures and recommendations to accelerate the promotion of green innovation performance of heavy polluting enterprises in China.

Key words: green innovation performance, heavy polluting enterprise, XGBoost model, interpretable machine learning