| 摘要: |
| 电力政策通过构建制度框架和激励机制引导电力系统优化升级,是服务于国家能源战略的重要工具。电力政策主要以文本形式呈现,基于文本信息采取的量化分析和效力研究对完善和优化电力政策有重要意义。通过构建基于提示工程与微调深度学习模型的政策量化框架,结合主题分析与可解释机器学习对中国电力政策效力展开研究。研究发现:当前国家电力政策大致归为五类主题,其中可再生电力开发与电力市场化改革这两个主题近年来具有较高热度；不同省份电力政策强度存在显著差异,其中平均强度最高的省份为广东、浙江和福建；除政策主题1外,剩余四类电力政策对于电力系统运行效率分别存在显著的“U”型、“倒U”型、“倒U”型与负向影响。在重视电力政策组合效应与协同效果的前提下,建议重点布局可再生能源发展的配套政策体系,同时推行渐进式的电力市场化改革。 |
| 关键词: 电力政策 提示工程 微调深度学习模型 量化框架 政策效力 |
| DOI：10.13216/j.cnki.upcjess.2026.01.0002 |
| 分类号:TP 391.1;F426.61 |
| 基金项目:国家自然科学基金资助项目(72004218；72140006；72074212)；中国博士后科学基金面上项目(2022M710049) |
|
| A Quantitative Framework and Efficacy Assessment of Electricity Policies in China 's Mainland Based on Large Language Models and Explainable Machine Learning |
|
MA Tieju1, CAI Yunhang2, LIU Feng2,3
|
|
(1.Antai College of Economics and Management, Shanghai Jiao Tong University, Shanghai 200030, China;2.School of Economics and Management, China University of Mining and Technology, Xuzhou 221116, Jiangsu, China;3.School of Business, East China University of Science and Technology, Shanghai 200237, China)
|
| Abstract: |
| Electricity policies serve as pivotal instruments for advancing national energy strategies by shaping institutional frameworks and incentive mechanisms that guide the system 's optimization and upgrading. Given the predominant textual form of the policy texts, quantitative analysis and efficacy assessment are of great significance for policy refinement and optimization. This study develops a policy quantification framework based on prompt engineering and fine-tuned deep learning models and combines topic modelling with explainable machine learning to examine the efficacy of China 's electricity policies. Key findings reveal that current national electricity policies can be broadly classified into five thematic categories, among which renewable energy development and market-oriented reforms have attracted relatively high attention in recent years. Substantial disparities in policy intensity have been observed across provinces, with Guangdong, Zhejiang, and Fujian exhibiting the highest average policy intensity. With an exception of the first policy theme, the remaining four categories exert heterogeneous effects on electricity system operational efficiency, manifesting U-shaped, inverted U-shaped, and negative relationships respectively. This study suggests prioritizing a supporting policy system for renewable energy while simultaneously advancing gradual market-oriented reforms on the premise of fully considering the policy portfolio effects and policy coordination. |
| Key words: electricity policies prompt engineering fine-tuned deep learning models quantitative framework policy efficacy |
