Prediction Of Strength For Recycled Aggregate Thermal Insulation Concrete Based On GA-BP Neural Network

Compressive strength is an important index to evaluate the performance of concrete.It has great practical significance to propose an accurate and reasonable concrete strength prediction method that can save resources,time and cost.Recycled aggregate thermal insulation concrete?RATIC?is composed of many materials and its internal effects are complex.The prediction of its strength is a typical multivariable and nonlinear problem,and it is difficult to predict its strength with a certain mathematical formula.Therefore,there is an urgent need to seek a new way of thinking or a new method to solve this problem.As a branch of artificial intelligence,in recent years,artificial neural network and genetic algorithm and other new technologies have been used widely.Many satisfactory results have been achieved in the field of concrete performance prediction.Therefore,this paper uses these two techniques to predict the compressive strength of RATIC.The main research work is as follows:?1?Establishing prediction model of compressive strength of RATIC based on BP neural network and GA-BP neural networkAfter collecting the relevant experimental data of the research group as a training sample,and analyzing the affecting factors of compressive strength of RATIC,eleven factors are selected as the input parameters of the neural network which including the water-binder ratio,the total amount of cementitious materials,the replacement rate of silica fume,the replacement rate of nano-Si O₂,the amount of fine aggregate,and sand rate,admixture dosage,recycled coarse aggregate substitution rate,mixed coarse aggregate crushing index,water absorption rate,and apparent density.After trial calculation and theoretical analysis,all the parameters that needed for the network establishment can be confirmed.Finally,after the network is trained,strength prediction model of RATIC based on the BP neural network and GA-BP neural network can be obtained with the network structure of 11-20-1.?2?Evaluating the two strength prediction models of RATIC based on the test dataThe two strength prediction models were used to predict the strength for twelve groups of RATIC tests which were designed as the network's predictive samples.The prediction accuracy of two models were evaluated by comparing the difference between the network predicted values and experimental values.The prediction results show that,the prediction accuracy of GA-BP neural network strength prediction model is higher than the BP neural network strength prediction model of RATIC,and the relative error of prediction is within 10%,which can meet the needs of the actual project.