Abstract: In order to achieve energy and emission reduction for scheduling plans in manufacturing without reducing processing efficiency, a mathematical model of the multi-objective green flexible job shop scheduling problem is established with the objectives of minimizing makespan, total energy consumption of machines and total carbon emissions. To address the problem that the traditional dual-population genetic algorithm tends to generate highly repetitive scheduling schemes due to the low diversity of the initial population when solving the model, an improved dual population genetic algorithm is proposed. Firstly, a two-stage coding strategy is adopted to simplify algorithm processes, and a multi-objective improved global-local-random search initialization method is proposed to increase the population diversity. Then, a fitness-based population segmentation method is developed to divide the population. Subsequently, evolutionary operations are carried out on the two populations separately, and population selection is conducted after merging to improve the quality of the next generation. Finally, a normalization method is adopted for comprehensive evaluation to select the optimal schedule. Comparative experiments on an improved Brandimarte dataset and a practical case show that the proposed algorithm has great advantages in solving the multi-objective green flexible job shop scheduling problem.