Photovoltaic Fault Diagnosis Model Incorporating Improved KAN-Based Feature Extraction and HPO-Optimized LightGBM

Engineering notes

Engineering notes will be added by the Power for Solar editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为光伏效率、钙钛矿太阳能电池、储能技术、太阳能热利用、BIPV、并网技术等高价值论文补充中文说明。

Original abstract

Abstract To enhance the fault diagnosis accuracy of photovoltaic (PV) arrays in harsh environments and the adaptive performance of diagnostic models, this paper proposes a PV array fault diagnosis model that integrates feature extraction via Kolmogorov-Arnold Network (KAN) with hyperparameter optimization of Light Gradient Boosting Machine (LightGBM) using the Hunter-Prey Optimization (HPO) algorithm. First, a PV array simulation model is constructed on the MATLAB/Simulink platform. By simulating typical fault scenarios under diverse operating conditions, the screening and construction of characteristic parameters are completed. Second, to address the inherent limitations of traditional KAN in nonlinear fitting capability and feature discrimination efficiency, a multilayer deep basis function network is proposed to replace the conventional univariate basis functions. Meanwhile, a feature-level attention mechanism is introduced, and a residual connection structure is integrated to achieve effective extraction of deep-seated features from time-series data, thereby strengthening the feature representation capability of the model. Finally, considering that the performance of LightGBM is highly susceptible to the configuration of key hyperparameters, the HPO algorithm is adopted to optimize the core hyperparameters of LightGBM, so as to further improve the fault diagnosis performance of the model.Experimental results based on measured data demonstrate that the proposed improved algorithm achieves a fault diagnosis classification accuracy of 99.31%, which is significantly superior to that of existing comparative models. This verifies its superiority and high precision in the task of PV array fault diagnosis.

Links and sources

• PDF from original source

Comments