光伏阵列故障诊断的可解释性智能集成方法

针对现有光伏阵列故障检测和诊断智能方法存在的泛化性不强、可解释性差的问题，提出了一种可解释性智能集成方法。对采集的光伏阵列输出时序电压、电流波形进行特征挖掘，并将多个已成熟应用于光伏故障诊断的智能算法作为不同基学习器与元学习器，构建结合不同智能算法优势且更具泛化性的Stacking集成学习模型；以沙普利可加性特征解释方法为总框架，并结合局部近似可解释性方法，对模型训练过程与结果进行解释分析，通过获取各特征的贡献、分析该集成模型的决策机制，并了解其如何进行诊断，提高其可靠度和可信度。算例实验结果表明，所提可解释性智能集成方法在不同规模数据集的测试中均实现了高精度的故障诊断，模型的可解释性结果表明由该智能集成模型建立的故障特征和诊断结果的映射遵循物理见解，增强了智能方法的可信度和透明性。

Interpretable intelligent integration method for photovoltaic array fault diagnosis

Aiming at the problems of weak generalization and poor interpretability of existing intelligent methods in photovoltaic array fault detection and diagnosis, an interpretable intelligent integration method is proposed. The feature mining is performed on the collected output time-series of voltage and current waveforms of the photovoltaic array, and multiple mature intelligent algorithms that have been applied to photovoltaic fault diagnosis are used as different base learners and meta learners to construct a Stacking ensemble learning model that combines the advantages of different intelligent algorithms and is more generalized. Then, taking the Shapley additive explanation method as the overall framework, combined with the local approximate interpretable method, the model training process and results are explained and analyzed. By obtaining the contributions of each feature, analyzing the decision-making mechanism of the integrated model, and understanding how to diagnose it, the reliability and credibility of the model are improved. The experimental results of case study show that the proposed interpretable intelligent integration method achieves high-precision fault diagnosis in testing on datasets of different sizes. The interpretability results of the model indicate that the mapping of fault features and diagnostic results established by the intelligent integration model follows physical insights, enhancing the credibility and transparency of the intelligent method.