Outlier Detection of Photovoltaic Power Generation System Data Based on GMM-IF Algorithm

The Gaussian Mixture Model (GMM) is widely used in anomaly detection due to its flexibility in handling complex data distributions and its soft classification mechanism. However, its modeling process relies on the assumption that data follows a Gaussian distribution. In contrast, time-series data from photovoltaic (PV) systems often exhibit high nonlinearity and non-stationarity, making it difficult to represent with a single Gaussian distribution, which affects detection performance. To address this issue and improve detection accuracy, this paper proposes a segmented GMM-IF outlier detection algorithm. In this method, the decision tree segmentation modeling strategy is used to structure the data to construct a local Gaussian-like distribution dataset to overcome the prior assumptions. Additionally, it integrates an Isolation Forest (IF) as the decision-making mechanism for detailed detection, further enhancing detection accuracy. Under different operating conditions and actual photovoltaic array datasets, comparative experiments were conducted using different evaluation metrics and algorithms to verify the effectiveness and superiority of the GMM-IF algorithm. Experimental results show that compared with the single model of GMM and IF, the GMM-IF algorithm has an average improvement of 3.8% in accuracy and 9.1% in F-value. Compared with other comparison algorithms, GMM-IF has obvious advantages, especially in terms of recall.