Blockchain technology has emerged as a cornerstone of the Web3.0 ecosystem with its decentralized, traceable, and immutable properties. However, the exponentially growing demand for transaction processing in the new-generation Internet scenario is still restricted by the performance bottleneck of blockchain. Therefore, effective performance optimization strategies are urgently needed to improve the efficiency and scalability of blockchain systems. Although existing optimization methods improve the performance of blockchain by adjusting the blockchain configuration parameters, they still have the problem of poor performance in the environment of limited blockchain node resources and fluctuating network environment. To overcome these challenges, we propose RL-ABO, an adaptive blockchain control parameter optimization method based on reinforcement learning, designed to dynamically adjust the blockchain parameters according to the real-time network environment and transaction load demand under the premise of limited node resource consumption. Specifically, we design a new reward function to jointly optimize performance and resource utilization, and introduces the clip mechanism and experience replay mechanism to enhance the training efficiency and dynamic adaptability. Experimental results show that, compared with existing methods, RL-ABO shortens the convergence time by 32.7%, improves the throughput by 8.3%, and significantly decreases the utilization of the central processing unit (CPU) and memory. Furthermore, RL-ABO shows outstanding performance in scenarios with fluctuating network delays, effectively addressing the limitations of traditional blockchain performance optimization methods.