In order to resolve the security monitoring problem of critical infrastructure based on Wireless Sensor Network(WSN), an adaptive intrusion detection scheme in data fusion stage is presented. The presented scheme takes the cluster network structure based on weight as the foundation, respectively using the anomaly detection subsystem and the misuse detection subsystem to detect the known attacks and the unknown attacks, and then adaptively adjusts the proportion of fusion data transmitted to the two sub-systems by tracking the Receiver Operating Characteristics(ROC) and the reward and punishment mechanism of the two sub-systems, so the intrusion detection of critical infrastructure in the data fusion stage can be realized. The simulation analysis indicates that the accuracy rate and detection rate of the presented algorithm is as high as 99.6% and 94.9% respectively, and can be improved at least 0.5% and 10.2% by comparing with other classic intrusion detection systems based on cluster structure.