Photovoltaic (Pv) System Performance Forecasting and Modelling Using Real -time Observation and Weather Data

Abstract

Photovoltaic (PV) systems are an indispensable source of renewable energy supply for both commercial and domestic use in many developing countries including Kenya. However, it remains difficult to fully integrate solar energy into the power grid. This is because solar energy is intermittent and highly dependent on weather conditions. Therefore, proper modelling and assessment of the influence of environmental parameters on the PV system performance is essential. In this study, a detailed performance analysis of a 1.5 kW PV system was done to study the effect of selected weather parameters on the power output. A weather station was setup on the site to provide real-time measurements of the ambient temperature and relative humidity. Solar irradiance was measured using a HT304N reference cell and the PV module temperature measured using a HT instrument PT300N temperature sensor. A current - voltage values of the solar PV system were obtained using a currentvoltage solar (I-V) analyzer. Data collection was done daily between 10:00 a.m. to 3:00 p.m. EAT at 30 minutes’ interval for a period of 21 days. Data analysis and visualization was performed using the R –software statistical package and Origin 9.1 software. An interactive application based on the single diode model was also developed and the results compared to measured data. The results obtained show that the ambient temperature increases with increasing solar irradiance with correlation coefficient (P) of 0.53 and Adj R2 of 0.27 showing a weak relationship. It was also noted that relative humidity varies inversely with solar irradiance with a correlation coefficient P of -0.50 and Adj R2 of 0.27. Relative humidity and ambient temperature exhibited a strong negative relationship yielding a correlation of - 0.94 and Adj R2 of 0.90. It was also noted that the ambient temperature had a relatively strong linear relationship with module temperature having a correlation P of 0.84 and Adj R2 of 0.71. The study further showed that maximum PV power output varies linearly with solar irradiance with strong positive relationship evidenced by a correlation P of 0.99 and Adj R2 of 0.98. However, the PV system efficiency was noted to decrease with increasing solar irradiance with a negative correlation P of-0.85 and Adj R2 of 0.72. Series resistance was found to have a strong negative non-linear relationship with solar irradiance with Adj R2 of 1 while shunt resistance decreased non-linearly with solar irradiance of Adj R2 of 0.64. The open circuit voltage was found to vary inversely with the module temperature with correlation P of -0.50 and low Adj R2 of 0.25 indicating a weak relationship. The maximum power and module temperature exhibited a positive linear relationship with P of 0.70 and Adj R2 of 0.49. It was established that the module temperature decreased with the efficiency of the PV system with P of -0.87 and Adj R2 of 0.76. Due to the high correlation between ambient temperature, solar irradiance,

relative humidity, module temperature principal component analysis (PCA) was done to remove redundant information. Support Vector regression (SVR) and random forest regression (RFR) models were therefore trained, tested and validated using data obtained from PCA to forecast real-time PV power output. SVR model employing leave one out cross validation technique (LOOCV) yielded the best model compared to 𝑘-fold and CV (Random resampling) cross validation techniques with root mean square (RMSE) of 40.4, Adj R2 of 0.98 and mean absolute error (MAE) of 29.01 on training dataset and RMSE of 45.10, Adj R2. of 0.97 and MAE of 29.27 on testing dataset. RFR model employing LOOCV yielded best model 𝑘-fold and CV (Random resampling) cross validation techniques with RMSE of 65, Adj R2 of 0.95 and MAE of 51.8 on training dataset whereas for testing set RMSE of 94, Adj R2 of 0.87, MAE of 68 were obtained. The trained models were further evaluated using validation dataset, SVR model outperformed RFR with RMSE of 43.16, Adj R2 of 0.97 and MAE of 32.57 compared to RMSE of 86, Adj R 2 of 0.90 and MAE of 69 obtained from RFR model. Furthermore, an app for carrying out real-time 1.5 kW PV power output prediction based on the SVR model was developed in this study. This research work therefore demonstrates that variability of solar irradiance, ambient temperature and relative humidity have significant effect on the performance of solar PV systems and must be considered when predicting PV power output. This is a significant step towards realizing a site-specific and dynamic solar PV performance analysis and forecasting technique.