Abstract <p>Traditional control methods struggle to achieve fine-grained and adaptive optimization for the lead–zinc flotation process due to its inherent challenges such as strong coupling, nonlinearity, significant time delay, and reliance on empirical knowledge. To address this, this paper proposes an intelligent optimization deep reinforcement learning control algorithm (IFPA-PPO-PID) that integrates an improved flower pollination algorithm (IFPA) with proximal policy optimization (PPO) for the adaptive optimization of proportional–integral–derivative (PID) parameters in the flotation process. Firstly, chaotic mapping is utilized to enhance the initialization process of the flower pollination algorithm (FPA), improving its global optimization capability and convergence speed. The superior performance of the improved algorithm (IFPA) is validated through standard test functions. Secondly, the PPO algorithm is applied within the flotation process control framework. An actor–critic network structure outputs continuous actions (PID parameters) to enable interactive learning with the environment (the flotation process). Simulation experiments demonstrate that, compared to the unoptimized PPO-PID approach, the IFPA-PPO-PID algorithm increases the reward value by approximately 2.9%, significantly improves control performance, achieves faster convergence, and enables real-time adaptive adjustment of PID parameters. This effectively enhances the control precision and automation level of the flotation process, offering a novel approach for the intelligent optimization control of complex industrial processes.</p> Graphical Abstract <p></p>