黄河流域是“能源流域”，兼具生态环境治理和经济社会发展的重任，涉煤产业场地类型、数量及特征的精准智能识别是流域能源资源−低碳发展−生态保护的关键基础问题。研究融合多源数据与深度学习算法，从流域−基地−场地尺度对黄河流域13个大型煤电基地的煤基场地特征精准解析，获得煤电基地高精度、高质量的本底信息，提出一种实时实景智能识别涉煤产业空间特征的新方法。① 筛选Google image、GF-6影像、Sentinel-2影像等多源数据，采集13个大型煤电基地煤基场地样本，构建煤炭场地(露天)、煤炭场地(井工)、煤电场地、煤化工场地4类数据集，涵盖21种样本类型。按照每种样本六面体设定6×10个样本，共计1 260个场地样本，分析得出最适样本数量−最高识别效率−最优识别模型的置信区间为80%～86%。② 建立了煤基场地类型量化模型(Coal-based Site Classification Quantitative Model, CSCQM)和煤基场地范围特征模型(Coal-based Site Range Characteristic Model, CSRCM)，模型平均精准度为0.837。明析了黄河流域涉煤产业场地本底信息，提出Google image底图叠加场地智能识别模型解算结果的高精度场地智能识别方法。③ 解析了流域神东煤炭−煤电产业集聚区精准本底数据，依据遥感生态指数(Remote Sensing Based Ecological Index，I_RSE)分析，煤基场地分布2 km核心区地表生态质量受煤炭、煤电产业影响明显，5 km缓冲区则影响不明显，而8 km控制区基本不受煤炭、煤电产业影响，从而给出了“动态修复”与分区域、分阶段重点治理等低碳路径。④ 解析了流域宁东煤炭−煤电−煤化工产业集聚区精准本底数据，2022年煤炭场地17.81 km²、占比34.1%，煤化工场地22.3 km²、占比42.6%，煤电场地12.2 km²、占比23.3%，煤化工场地 > 煤炭场地 > 煤电场地。进而采用PSR(Pressure-State-Response)模型得到风险管控综合得分53.93分，较2003年提高了27.2%。划分生态维护区、生产监测预警区、损毁修复重建区、其他调控区的分区管控模式。研究为涉煤产业煤基场地潜在污染控制、场地治理及区域生态修复提供技术方法与实践支撑。

The Yellow River Basin is an energy basin that has the dual responsibility of ecological environment governance and economic and social development. The precise and intelligent recognition of the categories, numbers and characteristics of coal-related industrial sites is a key basic issue for energy resources-low carbon development-ecological protection in the basin. This study integrated the multi-source data and deep learning algorithms to precisely analyze the characteristics of coal-based sites in 13 large-scale coal-fired power bases in the Yellow River Basin from the basin-base-site scale, obtained the high-precision and high-quality background information of coal-power bases, and proposed a new method of real-time real-scene intelligent recognition of spatial characteristics of coal-related industries. In this study, ① Multi-source data such as Google image, GF-6 image, Sentinel-2 image, etc. were collected as coal-based site samples from 13 large-scale coal-fired power bases to build four datasets of coal mine sites (open-pit), coal mine sites (underground), coal-power sites, and coal chemical sites, covering 21 categories of samples. According to each type of sample, 6×10 samples were set for each hexagonal cell, totaling 1260 site samples. The confidence interval of the optimal sample number-highest recognition efficiency-optimal recognition model was 80%−86%. ② A coal-based site classification quantitative model (CSCQM) and a coal-based site range characteristic model (CSRCM) were established. The average accuracy of the models was 0.837. The background information of coal-related industrial sites in the Yellow River Basin were clarified, and a high-precision site intelligent recognition method based on Google image base map overlaying site intelligent recognition model calculation results was proposed. ③ The precise background data of the Shendong coal-power industrial agglomeration area in the basin were analyzed. Analyzed by remote sensing based ecological index (RSEI), the surface ecological quality of the 2 km core area of coal-based sites was significantly affected by coal mine and coal-power industries, while the 5 km buffer zone was not significantly affected, and the 8 km control zone was basically not affected by coal mine and coal power industries. Thus, the low-carbon pathways such as dynamic remediation and key management by region and stage were proposed. ④ The precise background data of the Ningdong coal-power-chemical industrial agglomeration area in the basin were analyzed. In 2022, the area of coal mine sites covered an area of 17.81 km², accounting for 34.1% of the total area, the area of coal chemical sites covered an area of 22.3 km², accounting for 42.6% of the total area, and the area of coal-power sites covered an area of 12.2 km², accounting for 23.3% of the total area. The area ratio was coal chemical sites > coal mine sites > coal-power sites. Then, using the PSR (Pressure-State-Response) model, the comprehensive score of risk management was obtained as 53.93 points, which was 27.2% higher than that in 2003. A zoning management mode of ecological maintenance zone, production monitoring and early warning zone, damage repair and reconstruction zone, and other regulation zone were implemented. The study provided some technical methods and practical support for the potential pollution control, site management and regional ecological restoration of coal-related industrial sites.