Abstract: In order to explore the correlation between the environmental factors and photosynthetic rate in facility grape, and to achieve accurate prediction of photosynthetic rate of facility grape under the coupling of multiple environmental features, this study proposed a facility grape photosynthetic rate prediction model based on grey wolf algorithm optimizing BP (GWO-BP). Firstly, the photosynthetic data of nina queen grape from June to September 2024 were collected, under controlled environmental conditions, covering the key growth stages of the crop. And then the data were normalized to unify the data scale. Secondly, the four key factors affecting photosynthetic rate, including carbon dioxide concentration, air temperature, air humidity and light intensity, were screened out by using random forest gini importance and maximum information coefficient. A BP neural network model with a structure of 4-12-1 was constructed. Thirdly, the xavier uniform initialization and zero initialization algorithms were used to initialize the weights and thresholds of the BP model. And then the adaptive moment estimation (ADAM) algorithm was introduced to dynamically adjust the learning rate. The algorithm combined the advantages of momentum gradient descent and adaptive learning rate, and set different learning rates for different parameters by calculating the first-order moment estimation and second-order moment estimation of the gradient, which can speed up the convergence speed of the model and improve the stability of convergence. Finally, the wolf algorithm (GWO) was used to optimize its initial weights and thresholds to solve the problem of the BP model being easy to fall into local optimum. The weights and thresholds of the BP neural network were initialized by xavier uniform initialization and zeros initialization algorithms, and then the adaptive moment estimation algorithm was introduced to dynamically adjust the learning rate. The algorithm combined the advantages of momentum gradient descent and adaptive learning rate, and set different learning rates for different parameters by calculating the first-order moment estimation and second-order moment estimation of the gradient, which can speed up the convergence speed of the model and improve the stability of convergence. At last, the initial weights and thresholds were optimized by the gray wolf algorithm GWO to solve the problem that the BP model was prone to local optimum. The experimental results showed that compared with the four models of BP, support vector regression (SVR), random forest (RF) and extreme learning machine (ELM), were compared and analyzed to determine the basic model with the best effect. The results show that the BP model had the determination coefficients (R²) of 0.832 and 0.834 on the verification set and test set, which was better than the traditional model, but the prediction results had a large deviation in the extreme value area of the photosynthetic rate. After the optimization of GWO, the verification set R² of the GWO-BP model was increased to 0.920, the root mean square error decreased by 31.2%, and the mean absolute error decreased by 25.1%; the test set R² reached 0.918, the root mean square error decreased by 29.8%, and the mean absolute error decreased by 22.5%; the Huber loss function was reduced from 0.01 to less than 0.006, indicating a significant reduction in prediction errors. The experimental results showed that the GWO effectively avoided the problem of the BP model being easy to fall into local optimum, and improved the global search ability and convergence stability of the model, so that its prediction performance in the extreme value area was better. The model provided a reliable technical means for accurately capturing the coupling relationship between the photosynthetic rate of facility grape and multiple environmental factors, and realized the accurate prediction of the photosynthetic rate of facility grape.