Estimation and Assessment of Partial Shading Patterns in Large PV Farms Using ANN Algorithm

The use of technology to optimize electricity from photovoltaic (PV) farms is growing. Partial shading often occurs in PV farms owing to cloud movement or dust on the PV modules. Therefore, estimating shading patterns is essential as preliminary information for subsequent power optimization. This paper proposed an artificial neural network (ANN) to estimate the pattern of partial shading conditions. Simulation using a 160kW peak PV farm consisting of 600 modules in a 20 series - 30 parallel configuration is conducted in this research. The proposed technique measures each string's current and the substring's voltages. The Matrix of the current and voltages represents the shaded area of the PV farm and is used as a dataset. When partial shading (PS) occurs on the PV farm, the ANN model receives input data from current and voltage. Then, the trained ANN processes the input to estimate shading patterns. Assessment of the estimation results was carried out to determine the shaded PV module ratio and the shading strength. Comparative analysis explains that the proposed method outperforms other techniques such as Multiple-output Support Vector Regression (M-SVR), ANN with Levenberg–Marquardt (LM) training, and conventional backpropagation networks. In a typical shading condition, the best estimation depicts the accuracy of the shaded PV module ratio as 97% and shading strength as 98%. While the low accuracy results in the shaded PV module ratio as 94% and shading strength of 96%. Thus, the proposed technique can satisfactorily estimate the pattern caused by clouds in large-scale PV farms.