CO2 was one of the main raw materials for plant photosynthetic rate, CO2 optimal regulation model to meet the crops’ requirements was pivotal to afford a fine growth environment in crops’ whole life cycle. CO2 optimal regulation model fusing the support vector machine-improved fish swarm algorithm was proposed to provide a quantitative basis for precise regulation of CO2 in greenhouse. Taking the cucumber plant as research object, considering the mechanism of its photosynthesis, a photosynthesis rate nest test with threefactor combinations consisted of temperature, photon flux density and CO2 concentration was constructed. In the test, temperatures, photon flux densities and CO2 concentrations were set at 9, 7, 10 gradients, respectively. Totally 630 groups of CO2 response data were obtained by LI-6400XT portable photosynthesis rate instrument, in which 81% of the data was employed to construct the support vector machine (SVM) photosynthetic rate prediction model, while the remaining data was used for model validation. Furthermore, through improved fish swarm algorithm with SVM photosynthetic rate prediction model network as input, optimized photosynthetic rate values were acquired with variety of variables. Accordingly, CO2 saturation points were generated at different temperatures and photon flux density conditions for CO2 optimal regulation model. Compared the proposed SVM photosynthetic rate prediction model with conventional non-linear regression (NLR) prediction model and error back propagation (BP) prediction model, results showed that SVM prediction model was obviously superior to NLR prediction model and BP prediction model with correlation coefficient of 0.994 and mean absolute error of 0.879μmol/(m2·s). Then, XOR checkout was adopted to validate the CO2 optimal regulation model, results showed that the correlation coefficient between the simulated values and measured values was 0.965 and the maximum relative error was 3.056%, which indicated that the proposed CO2 optimization model could be applied to predict CO2 saturation points dynamically and provide a feasible way for CO2 concentration precise controlling for plants in greenhouse.