Study On Soft Sensor Model Of Ready-mixed Concrete Compressive Strength

Concrete is the basic material of civil engineering construction,and its compressive strength is taken as an important index of design and acceptance.Using traditional methods to obtain concrete compressive strength requires 28 days of concrete curing,which seriously lags behind construction progress.If the concrete compressive strength after 28 days can be predicted by adopting soft-sensing method in the initial stage of concrete mixing,which will be of great significance to improve project management and ensure construction quality.The paper proposes a soft-sensor model of concrete compressive strength based on online sequential extreme learning machine with kernels(OS-ELMK)by analyzing the performance requirement of soft-sensor model of concrete compressive strength.OS-ELMK has the advantage of random setting of its partial model parameters,and has high training efficiency,which can meet the real-time requirements in construction.And OS-ELMK can update online based on new samples,so it is more accurate compared with the off-line model based on finite initial samples.On this basis,OS-ELMK with selection mechanism is proposed,which can ensure the accuracy of the model while reducing the update frequency of the model and improve easy use.Under the premise of guaranteeing the real-time of the model,the paper proposes using Bayesian Optimization Algorithm to optimize the hyper-parameters of OS-ELMK soft-sensor model.BOA fits the hyper-parameter combination with a Gaussian process,and uses the acquisition function to select the next optimal combination,which has a faster optimization speed.Based on the data of the mixing station and pressure from field collection,a data set of C50 concrete compressive strength is constructed,and then the comparison test and verification test are carried out on the data set.The results show that the soft sensor model of concrete compressive strength proposed by the paper takes less than 5 seconds to build model and the prediction error is less than 2 MPa,which has high real-time and accuracy.Finally,based on the proposed soft-sensor model,a soft-sensor simulation system for ready-mixed concrete compressive strength is designed and implemented,which provides support for the engineering application of soft-sensor model of concrete compressive strength.