Green Light Optimized Speed Advisory (GLOSA) is a cutting-edge technology for intelligent transportation and is being increasingly applied on the ground worldwide. However, evaluating the environmental benefits of vehicles using GLOSA has become a challenge owing to the complexity and unpredictability of real-traffic. Here, a standardized environmental benefit evaluation framework for GLOSA is first proposed, which combined standard driving cycle (DC) database and high-precision emission models. The DC database was constructed using Markov chain approach. The database has certain randomness and diversity while considering the real driving characteristics. The plug-in hybrid electric vehicle (PHEV) CO2 and NOx emission models were constructed based on machine learning approach. These models have good generalization ability and prediction accuracy, with 0.87 and 0.80 R2 on the test set, respectively. In the constructed model, vehicle specific power (VSP) and state of charge (SOC) were identified as the most significant features affecting CO2 and NOx emissions. According to the evaluation results, the use of GLOSA reduces the emission factors of CO2 and NOx by 16.3% and 18.4%, respectively, when the initial SOC of PHEV is 20%. When the initial SOC is 50%, the emission factors of CO2 and NOx of the PHEV with GLOSA are reduced by 22.1% and 26.6%, respectively. The generality of this framework solves the different evaluation scales problem due to the randomness traffic. The framework serves the construction of intelligent and clean transportation while also assisting in the promotion, optimization, and recognition of GLOSA's limits.