Abstract: Carbon dioxide（CO₂） is a principal component of greenhouse gases, significantly impacting global climate change and environmental quality. Coal-fired power plants, being the largest source of CO₂emissions in China, face severe challenges. Therefore, to accurately, rapidly, and cost-effectively monitor the CO₂concentration in coal-fired power plants and promote their low-carbon development, this paper utilizes a distributed feedback semiconductor laser to construct a high-sensitivity CO₂gas detection system. It also validates and employs the HITRAN database as the dataset for a deep learning model, establishes a one-dimensional convolutional neural network（1D-CNN） model, and a backpropagation neural network for CO₂concentration detection. These models are compared with direct absorption spectroscopy technique, and the performance of the 1D-CNN model is enhanced through K-fold cross-validation and parameter adjustment. The results show that the determination coefficient（R2） of the 1D-CNN model can reach 0.9997, with a relative error of 1.07% and an absolute error of 7.88 mg/m3, indicating the model’s suitability. By utilizing the optimal parameters of the 1D-CNN model, a comparison between predicted and actual data reveals an average relative error of6.06%, an average absolute error of 17.97 mg/m3, and an R2of 0.99941, demonstrating high accuracy in the model’s predictions. This gas concentration detection model based on direct absorption spectroscopy and a deep learning neural network exhibits high accuracy and reliability in measuring CO₂concentrations, offering robust technical support for environmental monitoring and energy conservation and emission reduction within the power industry.