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.