The intelligent penetration attack based on reinforcement learning aims to model the penetration process as a Markov decision process, and train the attacker to optimize the penetration path in a trial-and-error manner, so as to achieve strong attack performance.In order to prevent intelligent penetration attacks from being maliciously exploited, a deception defense method for intelligent penetration attack based on reinforcement learning was proposed.Firstly, obtaining the necessary information for the attacker to construct the penetration model, which included state, action and reward.Secondly, conducting deception defense against the attacker through inverting the state dimension, disrupting the action generation, and flipping the reward value sign, respectively, which corresponded to the early, middle and final stages of the penetration attack.At last, the three-stage defense comparison experiments were carried out in the same network environment.The results show that the proposed method can effectively reduce the success rate of intelligent penetration attacks based on reinforcement learning.Besides, the deception method that disrupts the action generation of the attacker can reduce the penetration attack success rate to 0 when the interference ratio is 20%.