Reinforcement Learning-Based Optimization Strategy for Online Advertising Budget Allocation

This paper proposes a reinforcement learning-based optimization framework that defines a structured state space (real-time conversion rates, channel ROI, historical CTR), action space (budget-compliant allocations), and reward function (balancing revenue, cost, and placement effectiveness). To enhance adaptability, we introduce a multi-channel synergy mechanism using behavioral correlation matrices and a time-sequence update model for predictive, real-time budget adjustment. Trained with Proximal Policy Optimization (PPO) in a high-fidelity simulation, the model outperforms traditional rule-based and DQN baselines in CTR (+8.7%), ROI (+12.4%), and policy stability, while reducing latency and memory usage.