Environment-enabled fault diagnosis system for photovoltaic power stations

Area of Science:

• Renewable Energy Systems
• Electrical Engineering
• Artificial Intelligence in Energy

Background:

• Photovoltaic (PV) power plants face operational faults like short circuits and degradation, reducing efficiency and posing safety risks.
• Current fault detection methods using only electrical data lack adaptability and interpretability in dynamic environments.
• Existing static models struggle with complex fault patterns and environmental influences.

Purpose of the Study:

• To develop an online fault detection and classification method for PV systems that incorporates environmental factors.
• To enhance the accuracy, robustness, and interpretability of PV fault diagnosis.
• To provide an intelligent monitoring solution for PV systems in diverse field conditions.

Main Methods:

• Constructed a multidimensional feature space integrating electrical and environmental parameters (irradiance, temperature).
• Analyzed fault feature distribution using kernel density estimation to understand environmental influences.
• Designed a lightweight fully connected neural network with ReLU activation and Adam optimization for complex fault patterns.

Main Results:

• Achieved a 99.67% classification accuracy on the test set.
• Demonstrated the effectiveness of integrating environmental information for improved fault detection.
• Validated the method's ability to handle complex nonlinear fault patterns.

Conclusions:

• The proposed method offers a significant improvement in PV fault detection performance.
• Integrating environmental data enhances the robustness and accuracy of PV system monitoring.
• The developed technique provides an efficient, interpretable solution for intelligent PV operation.