To solve the problem of missing fault data and dividing fault degree in roadheader rotary table anomaly detection, a novel anomaly detection method based on variation self-adapting particle swarm optimization (VSAPSO) BP neural network under the single category learning is proposed. Using support vector data description (SVDD) to train the healthy data, constructing a non-healthy sample dataset based on the experience in the field, the non-healthy sample data is divided into fault critical data and fault data based on the recognition rate of SVDD for non-healthy sample datasets, and the variation self-adapting particle swarm optimization algorithm is proposed, the VSAPSO-BP neural network is constructed to detect healthy data, fault critical data and fault data, the detection accuracy is 91.7%, compared with the traditional PSO-BP neural network, VSAPSO-BP neural network has a higher accuracy and stability. The results show that the abnormal of roadheader rotary table can be detected accurately and effectively by using anomaly detection method based on VSAPSO-BP neural network under single learning, this method has a high application value.