AI-based Failure Detection System for Photovoltaic Power Plants

January 26, 2019 by Scott McMahan

Sumitomo Electric Industries, Ltd. developed a data storage and analysis system that uses artificial intelligence to quickly detect a reduction of photovoltaic (PV) power output. The (AI) works in conjunction with a string monitoring system for solar power generation plants. The system will be available for sales beginning in April 2019.

Most solar power generating plants currently under construction in Japan are anticipated to run for more than 20 years after installation. To ensure steady output power over a long time period, quick detection and restoration of failures is crucial. For this reason, a monitoring system to detect such failures is installed in each plant.

Conventional monitoring systems use threshold value, or critical value, to determine the reduction in power output. For instance, once the output is less than the prescribed threshold, the system determines it to be a failure and alerts the operator of the solar power generation plant.

However, output power is known to vary based on various factors including time of day, season, installation area or surrounding environment. Therefore using the threshold value alone lowers reliability, and adds the need for additional human analysis and subsequent decision making, both of which demand experience and more significant effort.

The string monitoring system employs AI to detect failures from the measured string power value and reports the cause according to predefined emergency levels. In detecting a reduction in power, the system identifies its cause (such as failure of fuses or breakers, shadows from trees and grass or deterioration of solar panels), and reports the cause according to a predefined emergency level.

Failures classified into seven categories

Failure detection is monitored on each string and the minimum unit of solar power generation system and the failed area is displayed on the plant's map in different colors classified according to the emergency level, thereby enabling quick identification of the failure area and the provision of precise instructions to workers.

The system uses power line communication (PLC) technology, which employs existing power lines for data communication, thus eliminating the need for additional installation of a communication line, enabling simple installation to existing plants already operating. According to Sumitomo, using the recently-developed failure detection and reporting system promptly and precisely detects and reports failures at an early stage, reducing operating costs of the solar power plant and loss of power generation.


  • Advanced string current data analysis using AI classifies the cause of the failure into a maximum of seven categories and reports it according to a predefined emergency level.
  • Daily reports can be sent to designated e-mail addresses. In addition to any failures measured on the day, other information such as power generation output is visually displayed.
  • String with failure occurring is overlaid on the power plant map. (Optional functions)
  • Abnormal string map option