CN110163430A - Concrete material Prediction of compressive strength method based on AdaBoost algorithm - Google Patents

Abstract

The invention discloses a method for predicting the compressive strength of concrete materials based on the AdaBoost algorithm. First, a large amount of existing concrete compressive strength test data is collected as a training set, and the proportion of each component of the concrete material is regarded as an input variable. The compressive strength is used as the output variable, and the test data is trained through the weak classifier in the AdaBoost algorithm, and the weights of different weak classifiers are determined according to the accuracy of the training results, and the weight of the weak classifier with a small prediction error rate is increased to reduce The weight of the weak classifier with a large prediction error rate, so that the weak classifiers are combined into a strong classifier with high prediction accuracy, and the compressive strength of concrete can be directly given according to the relevant input parameters. The invention only needs simple data collection and application of machine learning methods to quickly and accurately predict the compressive strength of concrete materials, which is convenient for the popularization and application of professionals such as structural design, identification and reinforcement.

Description

Prediction method of compressive strength of concrete material based on AdaBoost algorithm

technical field

The invention relates to a method for predicting the compressive strength of materials, in particular to a method for predicting the compressive strength of concrete materials based on the AdaBoost algorithm.

Background technique

Concrete material has the advantages of easy material acquisition, low cost, superior performance and high durability, so it is widely used in practical engineering. At present, concrete structure has become the structure type with the largest proportion in civil engineering. The compressive strength of concrete materials is one of the key parameters in the design of concrete structures, which is seriously related to the safety performance of the overall structure. The determination of the compressive strength of concrete is generally through the test method. The concrete test block sample is poured according to the designed mix ratio, and it is cured for a certain period of time. After forming, the strength test is carried out to obtain the compressive strength of the concrete. However, this method is cumbersome, time-consuming, consumes a lot of resources, and is very inefficient, which is not conducive to the application of actual engineering.

Contents of the invention

Purpose of the invention: the purpose of this invention is to provide a kind of concrete material compressive strength prediction method based on AdaBoost algorithm, solve the time-consuming, inefficient, resource waste problem of measuring concrete compressive strength with the mode of experiment.

Technical scheme: the concrete material compressive strength prediction method based on AdaBoost algorithm of the present invention is characterized in that, comprises the following steps:

(1) Collect N groups of concrete compressive strength test samples, obtain the concrete material component information and final compressive strength information of each group of tests, and combine the concrete material mix ratio information and The maintenance information is used as the input variable X, and the compressive strength is used as the output variable y;

(2) Import N groups of test samples into the AdaBoost algorithm as a training set, and initialize the weight distribution of the training set;

(3) Perform multiple rounds of iterations to determine the error rate e _t and weight α _t of different weak learners;

(4) Combine each weak learner according to the weight of the weak learner to obtain the final strong learner;

(5) Input the mixture ratio information and curing information of the concrete material to be obtained into the strong learner after training to obtain the predicted value of the compressive strength of the concrete material.

Wherein, the weight distribution of initializing training set in described step (2) is:

D ₁ ={ω ₁ ,ω ₂ ,…,ω _N },ω _i =1/N,i=1,2,…,N

In the formula, D ₁ is the weight distribution of the samples in the training set, and ω _i in the training set is the weight of the i-th sample, that is, the weight of each training sample is 1/N.

Described step (3) is specifically:

Assume T rounds of overall iterations, for the tth round of iterations, select the basic weak learner H _t (X), and use the training set with weight distribution D _t to train it, and calculate the weak learner on the sample distribution D _t Error rate:

In the formula, e _t is the error rate of the weak learner in the current round, y _i is the output value of the i-th group of samples, H _t (X _i )=y _i means that the training of the i-th group of samples is correct, H _t (X _i ) ≠y _i means training error on the i-th group of samples;

Then calculate the weight α _t of the weak learner in the final learner:

Update the weights of the training samples so that the weights of the samples that made mistakes in the previous round of training increase, and you can focus on learning them in the next study:

In the formula, Z _t = sum(D _t ) is the normalization factor;

Repeat the above steps to train multiple weak learners H _t (X) to obtain the corresponding weight α _t .

In the step (4), each weak learner is combined by the weight of the weak learner to obtain the final strong learner F(X):

The step (5) inputs the mixing ratio of the concrete material to be obtained and the curing parameter X to the strong learner after the training, and the learner outputs the predicted value of compressive strength F(X).

Beneficial effects: the present invention adopts machine learning technology, which can avoid the process of physical experiment, can complete the analysis and prediction process in a very short time, and greatly improves the efficiency and accuracy of calculation; since the machine learning algorithms are all built on the black box model On the basis of the algorithm, the prediction and analysis process of the algorithm does not need to rely on any definite formula or function, which can avoid the systematic error caused by the inaccuracy of the parameters; the learning process of the machine learning algorithm of the present invention is based on actual engineering or experiment On the basis of the actual data generated in the algorithm, the algorithm will also take the random factors of the scene into consideration during the analysis during the learning process, so that the predicted effect is closer to reality; the present invention predicts in the last round of iteration Wrong samples will be tested emphatically, and the sample weight distribution after multiple iterations has high credibility, and the resulting prediction results also have high accuracy.

Description of drawings

Fig. 1 is the adaptive evolution process of the present invention;

Fig. 2 is a specific flow chart of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings.

As shown in Figure 1-2, the method for predicting the compressive strength of concrete materials based on the AdaBoost algorithm includes the following steps:

Step 1: Construction of machine learning training set database:

Collect N groups of concrete compressive strength test sample data through literature search, network search, etc., and construct a database Θ=[θ ₁ ,θ ₂ ,...,θ _N ], where θ _i ,i=1,2,. ..N is the i-th group of data.

Step 2: Algorithm input and output parameter settings:

Each set of data in the database Θ contains two types of information: (1) the proportion and age of the mix of concrete materials such as cement, mortar, water, coarse aggregate, fine aggregate, and admixtures; (2) the age of concrete materials For compressive strength, set the information of type (1) as an input variable, denoted as X; the information of type (2) as an output variable, denoted as y, then each set of data can be written as θ _i = (X _i , y _i ).

Specifically, each set of data has 9 components, of which the input parameters have 8 components, including each component in the common mix ratio and the curing age, and the output parameters have 1 component, then X=(X ₁ ,X ₂ ,...,X ₈ ) is a 1×8 vector, and y is a scalar, as shown in Table 1 below:

Table 1 Relationship between input variables and output variables

In summary, the completed database can be expressed as:

Θ=[θ ₁ ,θ ₂ ,...,θ _N ]=[(X ₁ ,y ₁ ),(X ₂ ,y ₂ ),...,(X _N ,y _N )] (1)

Step 3: Build a high-precision learner based on the AdaBoost algorithm:

The AdaBoost algorithm in machine learning is used to train or learn the constructed database Θ. The AdaBoost algorithm includes two levels of learners, namely the weak learner H(X) and the strong learner F(X), where the strong learner is Multiple weak learners are combined according to a certain weight. The training process is shown in Figure 1. The specific process is as follows:

Initialize the sample weights in the training set Θ, and define the weight ω _i of each sample θ _i to be equal, then the initial sample weight vector is:

D ₁ ={ω _1,1 ,ω _1,2 ,…,ω _1,N },ω _1,i =1/N,i=1,2,…,N (2)

Carry out multiple rounds of iterations to determine the error rate and weight of the weak learner in each round. Suppose the overall iteration is T rounds. For the tth round of iterations, the weak learner is denoted as H _t (·), then for each set of samples input parameters X _i , the output result of the weak learner is H _t (X _i ), H _t (X _i )=y _i means that the prediction of the i-th group of samples is correct, and H _t (X _i )≠y _i means that the prediction of the i-th group of samples is wrong;

For N sets of training samples, calculate the error rate e _t of the t-th round weak learner

The meaning of this formula is that the error rate e of the weak learner in the _t -th round is the sum of the weights ω _i of the wrongly predicted samples;

Calculate the weight α _t of the t-th round weak learner according to the error rate e _t

From the above formula, it can be known that when e _t ≤ 1/2, α _t ≥ 0, and α _t increases with the decrease of e _t , which means that the weaker learner with smaller prediction error rate has a better performance in the final learner. greater effect;

Update the weight ω _t+1,i of each group of samples in the training set, so that the weight of samples with correct prediction (H _t (X _i )=y _i ) is reduced, and the weight of samples with wrong prediction (H _t (X _i )≠y _i ) Increase to focus on learning wrongly predicted samples during the next round of training

D _t+1 ＝{ω _t+1,1 ,ω _t+1,2 ,…,ω _t+1,N } (5)

In the formula, Z _t = sum(D _t ) is the normalization factor;

For t<T, repeat 2)-5), obtain the weak learner H _t ( ) and its weight α _t for each round of training, and combine them into a strong learner

Step 4: Application of the trained strong learner:

Input the mixture ratio of the concrete material to be obtained and the curing parameter X to the strong learner F(·) after training, and the learner outputs the predicted value of compressive strength F(X).

The method disclosed in the present invention learns all data and undergoes multiple rounds of self-correction and evolution. Since the operation of the algorithm is completely based on the black box mode, there is no link of fitting functions, and it only relies on known experiments The data is analyzed and learned, so the results obtained by this kind of prediction are also completely based on reality, and it is no longer necessary to repeatedly analyze the situation in various non-ideal states, and truly avoid tedious and idealized simulation calculations.

In order to more clearly illustrate the execution steps of the method disclosed in the present invention, a technical flow chart of the method is drawn, as shown in FIG. 2 . Now take the concrete compressive strength test data collected from the Internet from Prof. I-cheng Yeh of Chung Hwa University in Taiwan as an example. There are 1030 sets of test data. The data set is divided into 10 parts, and 9 of them are used as training data in turn. After the strong learner is obtained by the method disclosed in the invention, the remaining data is tested, and the error rate of 10 result comparisons is always no more than 5%, which shows that the method of the invention is effective.

Claims (5)

1. A method for predicting the compressive strength of concrete based on AdaBoost algorithm, is characterized in that, comprises the following steps: (1) Collect N groups of concrete compressive strength test samples, obtain the concrete material component information and final compressive strength information of each group of tests, and combine the concrete material mix ratio information and The maintenance information is used as the input variable X, and the compressive strength is used as the output variable y; (2) Import N groups of test samples into the AdaBoost algorithm as a training set, and initialize the weight distribution of the training set; (3) Perform multiple rounds of iterations to determine the error rate e _t and weight α _t of different weak learners; (4) Combine each weak learner according to the weight of the weak learner to obtain the final strong learner; (5) Input the mixture ratio information and curing information of the concrete material to be obtained into the strong learner after training to obtain the predicted value of the compressive strength of the concrete material. 2. the concrete material compressive strength prediction method based on AdaBoost algorithm according to claim 1, is characterized in that, in the described step (2), the weight distribution of initialization training set is: D ₁ ={w ₁ ,w ₂ ,...,w _N },w _i =1/N,i=1,2,...,N In the formula, D ₁ is the weight distribution of the samples in the training set, and w _i in the training set is the weight of the i-th sample, that is, the weight of each training sample is 1/N. 3. the concrete material compressive strength prediction method based on AdaBoost algorithm according to claim 1, is characterized in that, described step (3) is specially: Assume T rounds of overall iterations, for the tth round of iterations, select the basic weak learner H _t (X), and use the training set with weight distribution D _t to train it, and calculate the weak learner on the sample distribution D _t Error rate: In the formula, e _t is the error rate of the weak learner in the current round, y _i is the output value of the i-th group of samples, H _t (X _i )=y _i means that the training of the i-th group of samples is correct, H _t (X _i ) ≠y _i means training error on the i-th group of samples; Then calculate the weight α _t of the weak learner in the final learner: Update the weights of the training samples so that the weights of the samples that made mistakes in the previous round of training increase, and you can focus on learning them in the next study: In the formula, Z _t = sum(D _t ) is the normalization factor; Repeat the above steps to train multiple weak learners H _t (X) to obtain the corresponding weight α _t . 4. the concrete material compressive strength prediction method based on AdaBoost algorithm according to claim 3, is characterized in that, in described step (4), combine each weak learner by the weight of weak learner, obtain final strong study Device F(X): 5. the method for predicting the concrete material compressive strength based on AdaBoost algorithm according to claim 4, is characterized in that, described step (5) imports the mixing ratio and the maintenance parameter of concrete material to be sought to the strong learner after training completes X, the learner outputs the predicted value of compressive strength F(X).