A Novel Algorithm for PV Fault Forecasting and Isolating Faulty Locations

Abstract

Efficient and more reliable operation of solar PV systems contributes effectively to addressing issues such as rising clean energy demand, soaring energy prices, and decarbonization of power generation. With the continuous increases in the adoption of solar energy, the probability of faults occurring in PV systems has also increased. Till 2023, global PV capacity is estimated around 1.6 TW, which was 1.2 TW in 2022. Underperformance of PV systems due to faults has led to annual revenue losses of $4.6 billion worldwide. With these challenge as a motivation, the goal of research presented in this thesis is to contribute to compensating reliability and enhancing efficiency of PV system. To achieve this goal, the research work is increasing its efficiency by mitigating power losses due to acute and chronic faults through PV array reconfiguration and isolation of faulty locations. With respect to the first perspective, this thesis proposes two fault forecasting algorithms, each utilizing variation in solar cell parameters under different fault conditions. The first algorithm employs Bayesian interactions in linear regression to model the rate of change of solar cell parameters for forecasting fault before their occurrence. While the second algorithm uses the concept of natural language processing by establishing a deep learning-based transformer model for robust fault prediction. Furthermore, forecasting algorithms also classify faults into different severity levels to determine the level of predictive maintenance required, which is another novel feature of the proposed algorithm. Performance of the proposed fault forecasting and classification algorithm is also compared with existing machine learning-based regression and classification techniques. With respect to the second perspective of power losses mitigation under faulty conditions, reconfiguring the connections of a PV array has been proven to be effective in mitigating the negative impact of chronic faults such as partial shading and Hotspot faults. In case of acute faults, such as short circuit and open circuit faults, isolating faulty location is considered better approach for mitigating power losses. This thesis has proposed a modified chess knight reconfiguration approach that improves power generation capability under different shadow footprints. The proposed method is simulated in MATLAB Simulink and validated experimentally under different shadowing footprints such as short or narrow wide, short or narrow long corners, and symmetrical and asymmetrical shadows. In addition, economic analysis of the investigated system is performed to show effectiveness of the proposed technique in terms of energy, cost savings, and profit. The proposed technique mitigates power losses by 71% and increases profit by 13.5%.