In some areas, there is a phenomenon that the landfill is full or even over-capacity with the extension of the service period. With the aging and damage of the protective facilities, this phenomenon may have a more serious impact on the surrounding environment. It is necessary to excavate and transport the waste beyond the part to control it. This process will inevitably produce many landfill gas emissions, which will pollute the air. Therefore, it is necessary to predict and control the landfill gas. This study utilizes the Grey Wolf Optimization (GWO) algorithm to optimize Support Vector Regression (SVR). It establishes prediction models for various LFG concentrations based on previous LFG concentration data and real-time environmental monitoring data. The models are compared with traditional Support Vector Regression and Random Forest (RF) algorithms, predicting the concentrations of odor, ammonia, hydrogen sulfide, methane, and nitrogen oxides. The results indicate that GWO-SVR demonstrates more stable and accurate predictions across various LFG, with the coefficient of determination R2 approximately 10% higher than that of SVR and RF, and most other error metrics significantly lower. In contrast, SVR and RF show substantial errors in predicting odor, hydrogen sulfide, and nitrogen oxides. Thus, the GWO-SVR algorithm substantially improves the performance in predicting LFG concentrations, meeting the needs of on-site management.