This paper proposes a Deep Reinforcement Learning (DRL)-assisted online operator selection framework for constrained multi-objective optimization (CMOP). The framework uses population dynamics (convergence, diversity, feasibility) as the state, candidate operators as actions, and population state improvement as the reward. A Deep Q-Network (DQN) is trained to learn a policy that estimates Q-values for actions, enabling adaptive operator selection. The framework is embedded into four popular CMOEAs and tested on 42 benchmark problems. Experimental results show that the DRL-assisted operator selection significantly improves the performance of these CMOEAs, outperforming nine state-of-the-art CMOEAs in terms of versatility and solution quality. The proposed method addresses the challenge of selecting suitable operators for CMOPs, which is critical for algorithm performance. The DQL model considers constraint satisfaction and feasibility in the state and reward design, making it suitable for CMOPs. The framework is flexible, allowing any number of operators and easy integration into existing CMOEAs. The method demonstrates effectiveness in handling complex CMOPs with varying constraints and objectives. The DQL-assisted operator selection improves the algorithm's ability to adapt to population states, leading to better convergence and diversity. The study highlights the potential of DRL in adaptive operator selection for CMOPs, offering a promising approach for future research in constrained multi-objective optimization.This paper proposes a Deep Reinforcement Learning (DRL)-assisted online operator selection framework for constrained multi-objective optimization (CMOP). The framework uses population dynamics (convergence, diversity, feasibility) as the state, candidate operators as actions, and population state improvement as the reward. A Deep Q-Network (DQN) is trained to learn a policy that estimates Q-values for actions, enabling adaptive operator selection. The framework is embedded into four popular CMOEAs and tested on 42 benchmark problems. Experimental results show that the DRL-assisted operator selection significantly improves the performance of these CMOEAs, outperforming nine state-of-the-art CMOEAs in terms of versatility and solution quality. The proposed method addresses the challenge of selecting suitable operators for CMOPs, which is critical for algorithm performance. The DQL model considers constraint satisfaction and feasibility in the state and reward design, making it suitable for CMOPs. The framework is flexible, allowing any number of operators and easy integration into existing CMOEAs. The method demonstrates effectiveness in handling complex CMOPs with varying constraints and objectives. The DQL-assisted operator selection improves the algorithm's ability to adapt to population states, leading to better convergence and diversity. The study highlights the potential of DRL in adaptive operator selection for CMOPs, offering a promising approach for future research in constrained multi-objective optimization.