MAFCD: Multi-level and adaptive conditional diffusion model for anomaly detection

Zhichao Wu, Li Zhu, Zitao Yin, Xirong Xu, Jianmin Zhu, Xiaopeng Wei, Xin Yang

Abstract: Highlights•An adaptive conditional fusion framework is proposed to effectively guide the diffusion generation.•A novel perspective to comprehensively and accurately capture the information of time series.•The exponential and sign ensemble functions are designed to weight the multi-step sampling results.•Generate time series with high fidelity and diversity to enhance the generalization of the model.•Verify the performance and utility of the algorithm in five typical industrial domains.