Prediction Of High Performance Concrete Strength Based On Artificial Neural Network

Compressive strength has been generally considered to be one of the most essential qualities of concrete. Developing accurate and reliable compressive strength prediction methods is an important issue in concrete construction, and could lead to saving costs and time.Because of the wide application of mineral admixtures and chemical additives in high performance concrete(HPC), the relations between the composition and the performance of concrete are complicated increasingly. Strength prediction of high performance concrete is a typical non-linear system with muli-variables. Current most present models could not be widely used to their inaccurate prediction. So it is urgent to make use of the new idea, new methodology and new technology in probing into the regularity and predicting the performance of modern concrete. Prediction of strength for the high performance concrete are extremely significant in industry.In order to compensate the drawbacks of traditional methods, more attention have been paid to the artificial intelligence techniques like artificial neural network in these years. The artificial neural network method has been proved in achieving ideal results in the prediction of compressive strength of concrete.So in this study, through established the direct mapping relationship between the constituent of HPC and the compressive strength, several different artificial neural network forecasting models for predicting the strength of HPC have been developed. The results showed that it was practicable and feasible to predict the strength of HPC by neural networks forecasting models, and each of those models we presented achieve an accurate prediction result.Among three single neural network forecasting models, BP model and RBF model were superior to GRNN model in the prediction precision, and the prediction results of RBF were the nearest approximation of the actual.Otherwise, the RBF-BP composite neural network forecasting model promoted in this paper performed well in predicting strength of HPC. Compared with three single neural network forecasting models, composite neural network forecasting model had better precision.Finally, a model of non-linear combination forecasting based on PSO-RBF neural network were set up in this paper. Compared to other single models and linear combination forecasting model, the model of non-linear combination forecasting based on PSO-RBF neural network is the minimal. Because all kinds of single models could be combined completely and the available information of single models could be used completely on the model of non-linear combination forecasting based on PSO-RBF neural network. Moreover, this model overcame the weakness of linear combination forecasting model.