Research On Prediction Of Concrete Strength Based On Extreme Learning Machine

Compressive strength of concrete is an important indicator of quality assessment in engineering practice, to ensure the safety and economy of construction, it is necessary that early age can grasp the development trend of the strength of concrete.28-day strength of concrete prediction of great significance in the actual production of the project, is a typical multi-variable nonlinear systems. The accuracy of the traditional forecasting methods is poor, difficult to be popularized in the actual application. On the other hand, some scholars artificial intelligence algorithms, such as artificial neural network to predict the strength of concrete, has made more satisfactory results, but the traditional neural network training algorithm is based on a multi-gradient, easily lead to slow convergence, easy local optimum, over-fitting problems, the optimal number of nodes in the network is difficult to determine the hidden layer and other defects.Extreme Learning Machine (ELM) is a novel single hidden layer feedforward neural network proposed recently, due to its simple structure, fast learning and good generalization performance advantages are increasingly subject to numerous research scholars favor. Firstly, the concrete compressive strength of several factors:the components, such as water-cement ratio of concrete, fly ash, slag, aggregate, etc., curing conditions such as the relationship between temperature and humidity, as well as age and other concrete compressive strength were analysis, and gives concrete compressive strength test method analysis.On this basis, we propose a concrete compressive strength prediction model based on extreme learning machine, cement, fly ash, blast furnace slag, water, superplasticizer, dosage and age of concrete coarse aggregate, fine aggregate as a model input to the compressive strength as a model output. To verify the effectiveness of the algorithm, using actual data simulation of concrete compressive strength, and with the traditional BP neural network were compared. Simulation results show that:the proposed model prediction errors are within1OMPa, meet the practical applications.The number of nodes in the hidden layer is the main parameter to predict the effects of extreme learning machine performance. When the number of nodes in the hidden layer is too large, the extreme learning machine prone to morbid solution, resulting in the model to predict performance. To effectively solve this problem, we propose an extreme learning machine principal component analysis, in the hidden layer space extreme learning machine principal component analysis transform to eliminate pathological features hidden layer output matrix to improve the forecasting performance of extreme learning machine. Simulation results based on actual data show that based on principal component analysis of extreme learning machine learning machine get better than traditional extreme predictions.