Abstract: Partial shading on photovoltaic（PV） strings can significantly affect the power generation efficiency and safety of PV systems, so special attention is required. Currently, it is challenging to quantitatively calculate and characterize the operating state of PV strings using the voltage-current implicit equivalent circuit model. In order to perceive the operation state of photovoltaic strings in the local shadow state, a multimodal data-driven photovoltaic string state evaluation method is proposed. Firstly, the electrical characteristics of PV strings, the product of open circuit voltage and short-circuit current（referred to as ideal power）, and maximum power, are determined by analyzing the current-voltage curve. Subsequently, the pixel features within the curve contour are calculated by using the Canny edge detection algorithm and area formula. Multi-modal data sets are constructed through the above feature sets. Then, considering the causal relationship between electrical features and pixel features, an improved kernel Gaussian process regression model is used to calculate the contribution of electrical features to pixel size, and the optimization of weight factors for each feature is achieved based on the principle of pixel feature normalization. Finally, the multi-criteria decision-making method with optimized weights（VIKOR） is used to quantitatively evaluate the operating state of PV strings. Experimental comparison shows that the proposed method has a better state evaluation effect.