Novel algorithm for ultra-short term forecast of PV system faults

A group of researchers from the National University of Sciences & Technology (NUST) in Pakistan has developed a PV system fault forecasting technique based on variations in solar cell current and voltage parameters.

“Existing fault detection techniques detect faults after their occurrence,” the research's lead author, Ihsan Ullah Khalil, told pv magazine. “Our proposed fault forecasting technique forecasts the fault so that predictive maintenance can be assured. It uses the rate of change of solar cell parameters to identify which fault is occurring.”

According to Khalil, solar cell parameters start changing even before a fault occurs. “The IV curve is divided into 172,000 data points, so we get 172,000 values of I and V,” he further explained. “Then, by using each value of I and V, and the values of Im and Vm, we extract the same number of values for each solar cell parameter. Finally, we model the rate of change of each variable for the first 100 data points. For the first hundred data points, I and V are almost the same up to the first decimal point.”

The proposed algorithm is claimed to be able to extract cell parameters at either no faulty conditions or faulty conditions and to sense fault at its inception level.

Machine learning-based regression techniques are used for the model, which the scientists said is able to detect variation trends of each parameter against each fault at the inceptive stage. The algorithm initially models the initial trend against a single voltage, current, and power value. It then splits the data set and models the variation of solar cell parameters using four variants of linear regression. “Linear regression has given excellent results,” Kahlil said. “One of the major contributions is introducing a lemma for the fault index formula that is not been discussed in the literature before.”

The scientists claim that the results demonstrate that the solar cell extraction method they used offer superior performance compared to existing forecasting techniques, as it analyzses the variation in cell current and voltage for the detection of faults at an incipient stage. “The significance of the proposed algorithm rests in its early fault detection capability, which contributes to the development of adaptive protection systems for photovoltaic installations,” stated the researchers.

Their findings were introduced in the study “A novel procedure for photovoltaic fault forecasting,” published in Electric Power Systems Research.