Abnormal data detection method of power system based on mass dissimilarity

With the increasing complexity of power systems, the complexity and large-scale nature of power data have brought about problems such as data redundancy, anomalous data mixing, and degradation of data quality. To address these problems, anomaly data detection has become the key to system security. Traditional methods are usually limited to a single feature analysis or simple statistics, which is difficult to deal with high-dimensional, high-noise and nonlinear power data. In this paper, a novel anomaly data detection method based on quality dissimilarity is proposed. First, particle swarm optimization is applied to reduce the data dimensionality and identify key features associated with anomalies. Then, a detection model is built by quantifying the inter-sample differences through quality dissimilarity assessment to effectively distinguish normal and abnormal patterns while reducing system risk. Experiments demonstrate that the approach outperforms traditional statistical and machine learning methods, especially in dealing with noisy non-linear datasets. By focusing on intrinsic data quality differences, the framework improves the accuracy and reliability of anomaly identification without relying on predefined thresholds. This approach provides novel solutions for power system security, improves detection in complex operational environments, and supports real-world grid monitoring.