This paper proposes a novel online streaming data anomaly detection method. By using the new method, the improved $$L_{1}$$ L 1 detection neighbor region optimizes the initial hyper-grid-based anomaly detection method by decreasing the quantity of neighbor detection region, and online ensemble learning adapts to the distribution evolving characteristic of streaming data and overcomes the difficulty of obtaining the optimal hyper-grid structure. To validate the proposed method, the paper uses a real-world dataset and two simulated datasets and finds out that the experimental results are near to the optimal results.