Abstract: Accurately predicting total carbon emissions of various industries is crucial for informing decision-making in carbon emission control, significantly contributing to China's dual carbon goal. However, the accounting of carbon emission data in China is relatively late, characterized by small samples, multiple driving factors, high volatility, and non-monotonic patterns, which makes precise predictions rather challenging. To enhance the feature selection and generalization capabilities of prediction models under the conditions of small samples and multiple driving factors, this paper proposes a novel approach for predicting industrial carbon emissions. The approach utilizes support vector regression as the primary predictive model, integrating deep Q-network based feature engineering to identify key factors from the multi driving factors, and optimizing the predictive model parameters by the sparrow search algorithm, which effectively improve the model's predictive ability and generalization performance, particularly in scenarios characterized by small samples. Numerical experiments are conducted utilizing carbon emission data from two representative industries: construction and transportation in China. The results reveal that the proposed predictive methodology surpasses traditional models, demonstrating more significant robustness and predictive ability, and yielding more precise and effective forecasts of carbon emission trends in industries marked by small samples and multiple driving factors.

Key words: small samples, multiple driving factors, industry carbon emissions forecast, feature selection, parameter optimization