With the increasing deployment of photovoltaic (PV) systems in extreme environments such as deserts and Gobi areas, the fault signatures of PV arrays tend to become more compound and intricate. These adverse conditions significantly increase the difficulty of accurate health evaluation, thereby posing a critical threat to the power yield and operational reliability of PV power stations. To address the increasing complexity of PV array faults, this study proposes a quantitative health evaluation method for PV arrays that integrates multi-dimensional features, including I-V curve shape features, model parameters, and electrical parameters. First, the I-V output characteristics under similar and compound fault scenarios are analyzed in detail. A novel multidimensional health status evaluation framework is established based on the fusion of shape features, model parameters, and electrical parameters. Subsequently, an Improved Global Harmony Search algorithm is developed for model parameter identification. By dynamically adjusting the pitch adjusting rate and integrating elite harmony guidance from the harmony memory, the global search capability of the algorithm is effectively enhanced, leading to a significant improvement in parameter estimation accuracy. Finally, a health index is constructed based on efficiency loss analysis, as well as electrical parameters, geometric features, and model parameters that reflect the characteristics of various PV fault types. The proposed method is validated through both simulation and experimental studies, covering recoverable faults, unrecoverable faults, and compound fault conditions.