With the increase of real-time sensor data in the actual environment, it becomes more and more difficult to locate an abnormal situation. At the same time, in the field of image-based anomaly detection, generative adversarial networks (GAN) have been developed because of their ability to model complex high-dimensional image distribution. In order to accurately and quickly locate an abnormality of a photovoltaic inverter in a photovoltaic power generation system, a new abnormal detection and location framework based on GAN is proposed. The multi-variable time series is transformed into a series of two-dimensional image using the angle field, and a structure of encoder and decoder is used. Particularly for a series of images, a convolutional long-term and short-term memory network encoder is used to ensure the extraction of the time information of each time series data and the relevant information between variables. Finally, the anomaly can be detected and located by executing the anomaly score function, and the relevant experiments prove the effectiveness of this method in the actual photovoltaic inverter data anomaly detection task. This work is supported by the Ningxia Natural Science Foundation (No. 2020AAC03034).