Abstract: Solar energy is characterized by enormous reserves, cleanliness, and low carbon emissions. Vigorously developing solar energy constitutes a key pathway to achieving the “dual carbon” goals. Photon-Enhanced Thermionic Emission （PETE） is an emerging solar power generation technology that couples the photovoltaic effect with the thermionic effect. It boasts advantages such as high efficiency, simple structure, and stable operation, thus demonstrating great development potential. Solar PETE systems are strongly coupled light-heat-electricity systems, and the establishment of their theoretical models is crucial for the development and research of PETE technology. However, the mathematical and physical models for light-heat-electricity coupling typically suffer from high computational costs and low efficiency. A back propagation （BP） neural network model is employed to predict the power generation performance of PETE systems. Three parameters, concentration ratio, electron affinity, and cathode thickness, are selected as input variables, while six key indicators, including cathode temperature and energy conversion efficiency, serve as output targets.A research sample is constructed using 990 sets of numerical simulation data. The dataset is divided based on the latin hypercube sampling （LHS） method, and logarithmic preprocessing is applied to the photon enhancement coefficient to optimize data distribution. The constructed BP neural network features a node configuration of 3-10-12-6. The Levenberg-Marquardt algorithm is adopted for model training, with two complementary metrics, the coefficient of determination （R²） and mean absolute percentage error （E_MAP）, used for evaluation. Results show that the model achieves extremely high prediction accuracy for all six parameters, with R² values exceeding 0.99 and E_MAP values approaching 0. Validation using the test set confirms that the model exhibits strong generalization ability. Compared with traditional numerical simulations, the prediction speed is significantly improved, enabling real-time optimization of the system. The effectiveness of the BP neural network model in predicting the performance of solar PETE power generation systems is verified, and an efficient approach for their development and application is provided.