Considering the spatial distribution and temporal evolution characteristics of extreme meteorological hazards, this paper constructs a combined model (CNN-LSTM) of convolutional neural network (CNN) and longshort-term memory network (LSTM), and designs the training process of the model. Certain salient features of the environmental change data are captured by the CNN spatial model and these features are used as inputs for constructing the LSTM time series dataset, which reveals the interactions between the hidden features in the data and the space, and thus improves the accuracy of the prediction results. The diffuse reflection coefficient of the solar panel is also calculated as well as the parameters of the model are determined to finalize the environmental change-sensitive nonlinear modeling of the current output characteristics of the solar cell, which is experimentally demonstrated and analyzed. The CNN-LSTM model in this paper outperforms the single LSTM model in the four evaluation indexes of RMSE, MAE, MAPE and R² in the training and test sets, and it is able to more accurately capture the small fluctuations of the solar cell current output power in response to the environmental changes, and shows stronger robustness and generalization ability. The reliability and sensitivity of the solar cell are better when the insulating film thickness is 0.2 mm and 0.8 mm, and it has better sensitivity to both temperature and relative humidity, which provides reference information for the sensitivity of solar cell current output to environmental changes using the time series data analysis method.
