The invention discloses a method for predicting the concrete strength based on a hybrid model. The method comprises the following steps: conducting strength experiments on concrete at different mixing ratios on site according to a standard concrete strength detection method, thereby obtaining a plurality of learning samples about the function relationships between cement x1, blast-furnace slag powder x2, fly ash x3, water x4, a water reducing agent x5, a coarse aggregate x6, a fine aggregate x7, a curing period x8 and other concrete component mixing information as well as the concrete strength y; training extreme learning machines, artificial neural networks and support vector machines in the hybrid model, and confirming the optimal extreme learning machine, artificial neural network and support vector machine according to the optimization goal of minimized relative errors; on the basis of the optimal extreme learning machine, artificial neural network and support vector machine, confirming the optimally predicted concrete strength y with a decision function based on adaptive weight according to the predicted values of three modeling methods, wherein the optimally predicted concrete strength y can be utilized for judging whether concrete component input information meets the engineering design requirements or not. Through adoption of the method, the advantages of the three modeling methods are enhanced while the disadvantages of the three modeling methods are avoided, and the comprehensive predicting effect is better, so that the adaptability to different actual work conditions, namely the robustness, can be improved, and important significance is provided for rapid mixing ratio design and quality control for concrete.