工业园区CO₂排放在全国CO₂排放总量中占比约31%，园区低碳发展对于缓解气候变化具有重要作用。以湖北省某工业区为例，对工业园区单、多CO₂排放源关键信息进行了反演研究。基于AERMOD系统建立了工业园区内CO₂扩散的正向模型，获取反演所需数据集。利用粒子群优化算法（particle swarm optimization，PSO）、鲸鱼优化算法（whale optimization algorithm，WOA）和鹈鹕优化算法（pelican optimization algorithm，POA）优化后的反向传播（backpropagation，BP）神经网络对工业园区中CO₂排放源位置及排放强度进行反演计算。结果表明：POA-BP反演模型对于单排放源坐标的反演结果R²为0.965，对排放强度的反演R²为0.938；对于多排放源坐标的反演结果R²为0.97，对排放强度的反演R²为0.988，相较于其他模型来说具有较高的反演精度和稳定性，可以对工业园区内CO₂排放源实现较为精确的定位，为工业园区应对气候变化和推进绿色转型提供决策支持。

CO₂ emissions from industrial parks account for about 31% of the total national CO₂ emissions, and the low-carbon development of parks plays an important role in mitigating climate change. In this paper, the key information of single and multiple CO₂ emission sources in industrial parks is inverted and studied by taking an industrial zone in Hubei Province as an example. A forward model of CO₂ diffusion in the industrial park is established based on the AERMOD system to obtain the data set required for inversion. The BP neural network optimized by particle swarm optimization （PSO）, whale optimization algorithm （WOA） and pelican optimization algorithm （POA） is used to calculate the inversion of CO₂ emission source location and emission intensity in the industrial park. The results show that the POA-BP inversion model has an R2 of 0.965 for single source coordinates and an R² of 0.938 for emission intensity, and an R² of 0.97 for multiple source coordinates and an R² of 0.988 for emission intensity, which has a higher inversion accuracy and stability than other models, and it can achieve a more accurate location of CO₂ sources in the industrial park and provide a better solution for the industrial park. It can realize more accurate positioning and provide decision support for industrial parks to cope with climate change and promote green transformation.