Fast detection of business data reduces the impact of automated systems by anomaly scheduling. In this paper, we address the limitations of traditional unsupervised anomaly detection methods, such as leakage detection, and design an anomaly detection method based on meta-learning hybrid selection integration (Meta-HESAD). Methods such as compact time series and isolated Senri are introduced to accomplish the base detector required for downscaling and anomaly detection of multimodal data. The adaptive filtering algorithm is optimized with improved Sigmoid function to improve the stability and convergence speed of the data anomaly detection process. The results show that the average value of the adaptive filtering algorithm is 0.9851, which is higher than that of the comparison algorithms, 0.8007, 0.8286, 0.8456. When the detection probability reaches 1.0, the false alarm rate of this paper’s algorithm is smaller than that of the three comparison algorithms. In practice, the detection accuracy probability of this paper’s algorithm exceeds 90% in all 10 iterations.
