Quantum-Inspired Reinforcement Learning for High-Dimensional Optimization Problems

High-dimensional optimization problems are real challenges for classical reinforcement learning (RL) based on its limited exploration capacity and poor scalability. In response to these limitations, this research proposes hybrid Quantum-Inspired Evolutionary Deep Reinforcement Learning (QIEDRL) framework. The method integrates the global search ability of quantum-inspired evolutionary algorithms with the adaptive learning advantages of deep reinforcement learning models such as DQN and PPO. Working on high-dimensional data with 537 features algorithm involves quantum rotation-based initialization and a policy improvement loop to refine convergence and exploration-exploitation balance. The feature selection scheme efficiently compacts dimensionality with a 22% feature selection ratio without degrading predictive performance. Experimental assessments show that QIEDRL results in a cumulative reward of 35.2 and converges after 50 training epochs, improving over several state-of-the-art baselines. Implemented in Python, the system provides a hardware-independent, scalable solution for complex, real-world settings, thus broadening the usability of deep RL in high-dimensional optimization problems.