Fast Discovery Of Cement Hydration Kinetic Equation In GPU Environment

Cement is one of the most commonly used building materials,its physical properties will affect the safety of the building.And a variety of performance indicators of cement,such as compressive strength,flexural strength,durability will be directly affected by cement hydration reaction.Through the modeling and simulation of the cement hydration process,we can understand the internal mechanism of the cement hydration process.So as to achieve the purpose of assisting cement production and design.Among the various methods for modeling the hydration process of cement,dynamic modeling is an important method.The hydration process is described by constructing the dynamic equation of hydration degree and time,which makes the whole process easy to understand.At the same time,during the hydration process of cement,the hydration reaction in the early stage is the most intense and gets stable gradually in the middle and late stages.Therefore,it has important research significance and application prospects to construct the early dynamic equation of cement hydration.Materials scientists who study cement hydration kinetics has made a lot of progress.At present,the cement hydration dynamics research also exists some problems which need to solve urgently,As the cement hydration process of physical change and chemical reaction having high complexity,it makes finding cement hydration kinetics equation by hand particularly difficult.Besides that,in order to reduce the difficulty of the research and simplify the derivation process,materials scientists had to ignore some characteristics of the cement,for example,assume that cement particle is spherical,ignore the hydration reaction of the interaction between the particle and so on.Relevant studies have shown that evolutionary computation can extract natural laws or formulas that reflect phenomena from data.Schmidt and Lipson pointed out that without any prior knowledge,evolutionary algorithm can be used to extract Lagrange’s theorem,Hamiltonian formula and other laws from the data in the physical system.Therefore,it is feasible and effective to apply the technique of the evolution calculation to extract the dynamic equation of cement hydration.This study mainly aims at the early stage of cement hydration to establish a correct and effective kinetic model,the main contents of this paper are as follows.(1)Application of single gene similarity weight tournament operator in the discovery of kinetic equation in the early stage of cement hydration.The similarity weight tournament operator is firstly applied to cement hydration kinetic modeling.The experimental results show that the algorithm can effectively solve the problem of premature convergence and local extremum in the traditional gene expression programming algorithm.The effectiveness of gene expression programming algorithm in dynamic discovery is verified.(2)Early stage kinetic equation discovery of cement hydration based on similarity weight tournament gene expression programming algorithm.As it is difficult to fit the cement hydration process with single gene chromosome,the multi-genes similarity weight tournament operator is proposed.The experimental results show that it can effectively improve the fitting accuracy of dynamic equations and further verify the validity of evolutionary computation in dynamic modeling.(3)Discovery the early dynamic equation of cement hydration rapidly.In the case of large scales models or a huge amount of data,the high time complexity is an inevitable problem in evolutionary computation.It gets particularly prominent in the modeling of cement dynamics.In this paper,the genetic expression programming algorithm and particle swarm optimization algorithm are combined to quickly discover the dynamic equation of cement hydration under the GPGPU environment.To reduce the dependence on a single initial point in the process of numerical solution calculation,this work proposed a numerical solution calculation strategy based on multiple initial points.At the same time,the selection mechanism of the initial point is designed to solve the problem of the high time complexity of the numerical solution.Besides,aiming at the problem of premature convergence of small populations in evolutionary computation,the similarity weight selection strategy of multiple genes was proposed to improve the diversity of the population,to avoid the problem of premature convergence of small populations.