Search

Abstract Accurate parameter identification plays a crucial role in realizing precise modelling, design optimization, condition monitoring, and fault diagnosis of photovoltaic systems. However, due to the nonlinear, multivariate, and multistate characteristics of PV models, it is difficult to identify perfect model parameters using traditional analytical and numerical methods. Besides, some existing methods may stick in local optimum and have slow convergence speed. To address these challenges, this paper proposes an enhanced nature‐inspired OLARO algorithm for PV parameter identification under different conditions. OLARO is improved from ARO incorporating existing opposition‐based learning, Lévy flight and roulette fitness‐distance balance to improve global search capability and avoid local optima. Firstly, a novel data smoothing measure is taken to reduce the noises of I–V curves. Then, OLARO is compared with several common algorithms on different solar cells and PV modules using robustness analysis and statistical tests. The results indicate that OLARO has better ability than others to extract parameters from PV models such as single diode, double diode, and PV module models. Moreover, the convergence performance of OLARO is more excellent than the other algorithms. Additionally, the I–V curves of two PV modules under different irradiance and temperature conditions are applied to verify the robustness of the proposed parameter extraction algorithm. Besides, OLARO is successfully applied to two real operating PV modules, and it is compared with two recent well‐known methods improved by FDB. Finally, sensitivity analysis, stability analysis, and discussion of practical challenges are provided.

Abstract Irradiance prediction is a vital task in the renewable energy field. Its aim is to forecast the irradiance or power of a photovoltaic plant and thus provide a reference for stabilizing the power grid. In the real scenarios, missing data can significantly reduce the accuracy of the prediction. Meanwhile, due to the unawareness of the distribution of datasets, it is difficult to choose a suitable imputation method before modeling. Also, different imputation methods do not have the same good effects on different datasets. In this article, a recurrent neural network with an adaptive neural imputation module is proposed for forecasting direct solar irradiance using missing data. The model predicts future 4‐h irradiance based on the missing historical climate and irradiance data without imputing the data in pre‐processing stage. The proposed model is tested on the open access datasets, with missing values generated randomly in all input series. The model performance is compared under various missing rates and different input factors with other imputation methods. The results demonstrate that the proposed methods outperform other methods under different evaluation metrics. Furthermore, the authors integrate the model with the attention mechanism and find it has better performance at high irradiance.