An Anomaly Detection Method for the Output Power of Photovoltaic Arrays Based on TKAN

One of the greatest challenges facing photovoltaic (PV) power generation systems today is maintaining their operation at the desired power generation efficiency. To achieve this goal, the anomaly detection of the output power of PV arrays is crucial for ensuring reliability and safety. This article proposes an anomaly detection for the output power of PV arrays based on temporal Kolmogorov–Arnold networks (TKANs). First, a dataset of PV array parameters is constructed by selecting the time series of output power, ambient temperature, component temperature, and irradiance of the PV array as input features. Second, the PV array parameter dataset undergoes feature normalization by obtaining the boundary values of environmental information and operating parameters, and scaling them to the range of 0–1. Then, the processed dataset is trained using the TKAN neural network to obtain the anomaly detection model of the output power of the PV array. Finally, the proposed method is compared and analyzed with three other methods, such as Isolation Forest, $K$-means, and long short-term memory, verifying its reliability and superiority. In addition, the effectiveness of the proposed method is validated in a self-built PV power plant.