In order to solve the problem of data fusion between network nodes of users with different reputation in spectrum sensing, a distributed cooperative spectrum sensing method based on reinforcement learning and consensus fusion is proposed. This method regards each perceived user as an agent, and the agent uses the reinforcement learning algorithm to select a cooperative user from adjacent nodes for consensus fusion. The reputation value is used as reward to ensure that agent tends to merge with high reputation nodes. At the same time, the reputation value of malicious users is reduced, causing that they gradually drop out of the cognitive wireless network. Finally, the conformance fusion method is adopted to make the whole network reach consensus and compare with the decision threshold to complete the cooperative spectrum sensing. The simulation results show that this method can identify malicious users effectively, and enhance the perception performance of the whole cognitive wireless network through reinforcement learning. As a result the cooperative spectrum sensing network is more intelligent and stable.