Abstract: 【Objective】 Aiming at the uncertain parameters of tomato water demand law and the non-dynamic analysis of system, the dynamic prediction model of tomato stem diameter change in greenhouse was studied in the hope providing some decision support for tomato water demand. 【Method】This research mainly uses the method of integrating the multi-layer perceptron algorithm with plant physiological and ecological information, established a stem diameter variations prediction model, which contains many types of data ,like air temperature, humidity, soil humidity, leaf temperature, stem diameter variations and photosynthetic effective radiation. In this paper, the three-layer hidden layer neural network is adopted to conduct fully connected training on the data vector of the 6-dimensional training set after regularization and normalization, and then the 1-dimensional output vector is obtained after input the validation set data . Finally, the output vector is inverse normalized to obtain the predicted value of stem diameter variations.【Result】The regression coefficient of the predicated value and the measured value (R²) and the root mean square error (RMSE) were 0.901 and 0.175, respectively, based on the dynamic prediction model of the short-term tomato stem diameter variations on which multilayer perceptron was established. 【Conclusion】The model is applicable to the dynamic prediction of short-term stem diameter variations of greenhouse tomato, and has a good application scenario, which can provide a strong basis for the decision to meet the crop water demand.