Accurately predicting the photosynthetic rate of greenhouse tomatoes is crucial for evaluating their growth and yield. However, due to the complexity and variability of the greenhouse environments, traditional photosynthetic rate prediction models often fail to meet the demand of precise prediction. To address this issue and enhance the accuracy and stability of prediction model, a multi-model fusion strategy for predicting the photosynthetic rate of greenhouse tomatoes was proposed. Initially, the photosynthetic rate of tomato was collected under various combinations of temperature, humidity, light intensity, and carbon dioxide concentration, and a sample set was constructed. The data was preprocessed by using five-fold cross-validation method. Based on preprocessed data, prediction models for tomato photosynthetic rate were established by using particle swarm optimization-support vector regression (PSO-SVR), cuckoo search optimization-extreme learning machine (CS-ELM), and northern goshawk optimization-Gaussian process regression (NGO-GPR) algorithms, and preliminary predictions were made. Next, the Stacking algorithm was used to combine the predictions of the basic models through training an ensemble tree meta-model (XGBoost), thereby achieving multi-model fusion. The results of simulation analysis demonstrated that compared with a single prediction model, the photosynthetic rate prediction model based on multi-model fusion effectively utilized the advantages of the basic models, enhancing the accuracy and stability of predicting photosynthetic rate. The MAE of the validation set for the model was 0.5697μmol/（m2·s）, and the RMSE was 0.7214μmol/（m2·s）. Therefore, the method proposed had significant advantages in predicting the photosynthetic rate of greenhouse crops, and can provide theoretical basis and technical support for the management and control of the light environment of greenhouse tomatoes and other crops.