In order to improve the prediction accuracy of the Reference Crop Evapotranspiration (ET0) in the greenhouse environment, Fireworks Algorithm (FWA) is proposed to optimize the crop reference evapotranspiration model established by Extreme Learning Machine (ELM), which effectively solves the data fluctuation problem caused by the random input weight matrix and the bias matrix of the Extreme Learning Machine in the process of data prediction, improves the prediction accuracy of Extreme Learning Machine. By taking the greenhouse environment data as the input of the model, the FWAELM model is established with reference evapotranspiration ET0 as the output, and the results are compared with the ELM model prediction results. The outcome shows that the root mean square error, mean absolute error and model deterministic coefficients of the FWAELM model are: 0.115 6, 0.143 6, 0.943 8, better than ELM's 0.403 5, 0.346 7 and 0.819 0, FWAELM model has higher prediction accuracy. In addition, the prediction accuracy of the model under the absence of meteorological parameters is studied. The results show that the prediction accuracy of the model is still high when the parameters are three or more, which is suitable for the prediction research on greenhouse ET0.