Convolutional neural network with feature reconstruction for monitoring mismatched photovoltaic systems

Abstract The cutting-edge monitoring technique of photovoltaic power systems tends to employ cloud servers for big data analysis in the wake of smart sensors and internet of things (IoT). Various mismatch phenomena of photovoltaic arrays are generated due to the external and internal interactions, which would be identified by an appropriate monitoring approach. However, high dimensional sequential data provided by multiple sensors challenges the existing monitoring technologies. This paper proposes a dimension reduction technology mapping multiple sequence signals to a sequence of images which are processed further by a convolutional neural network (CNN), resulting in a novel condition monitoring system for photovoltaic array systems. Firstly, multiple sources of 1-dimensional time-series data are rearranged to construct 2-dimensional time-series images. Then, the CNN algorithm automatically extracts the underlying graphical features from data of 2-dimensional images for condition monitoring. Experiments were carried out upon self-made solar power stations to verify the effectiveness of the proposed method for real-world solar power systems. It shows that the data driven approach could identify effectively key operation conditions from the historical data with a negligible loss of features at the presence of mismatched phenomena.