CN114388069A - Concrete mixing proportion optimization method with multiple performance controls - Google Patents

Concrete mixing proportion optimization method with multiple performance controls Download PDF

Info

Publication number
CN114388069A
CN114388069A CN202111582571.7A CN202111582571A CN114388069A CN 114388069 A CN114388069 A CN 114388069A CN 202111582571 A CN202111582571 A CN 202111582571A CN 114388069 A CN114388069 A CN 114388069A
Authority
CN
China
Prior art keywords
concrete
deep learning
strength
workability
durability
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111582571.7A
Other languages
Chinese (zh)
Inventor
周力
陈国辉
鄢烈祥
范阳春
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hangu Yunzhi Wuhan Technology Co ltd
Original Assignee
Hangu Yunzhi Wuhan Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hangu Yunzhi Wuhan Technology Co ltd filed Critical Hangu Yunzhi Wuhan Technology Co ltd
Priority to CN202111582571.7A priority Critical patent/CN114388069A/en
Publication of CN114388069A publication Critical patent/CN114388069A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
    • G16C20/10—Analysis or design of chemical reactions, syntheses or processes
    • C—CHEMISTRY; METALLURGY
    • C04—CEMENTS; CONCRETE; ARTIFICIAL STONE; CERAMICS; REFRACTORIES
    • C04B—LIME, MAGNESIA; SLAG; CEMENTS; COMPOSITIONS THEREOF, e.g. MORTARS, CONCRETE OR LIKE BUILDING MATERIALS; ARTIFICIAL STONE; CERAMICS; REFRACTORIES; TREATMENT OF NATURAL STONE
    • C04B28/00—Compositions of mortars, concrete or artificial stone, containing inorganic binders or the reaction product of an inorganic and an organic binder, e.g. polycarboxylate cements
    • C04B28/02—Compositions of mortars, concrete or artificial stone, containing inorganic binders or the reaction product of an inorganic and an organic binder, e.g. polycarboxylate cements containing hydraulic cements other than calcium sulfates
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
    • G16C20/70—Machine learning, data mining or chemometrics

Landscapes

  • Engineering & Computer Science (AREA)
  • Chemical & Material Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Computing Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Chemical Kinetics & Catalysis (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • General Engineering & Computer Science (AREA)
  • Ceramic Engineering (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Crystallography & Structural Chemistry (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Inorganic Chemistry (AREA)
  • Databases & Information Systems (AREA)
  • Materials Engineering (AREA)
  • Structural Engineering (AREA)
  • Organic Chemistry (AREA)
  • Analytical Chemistry (AREA)
  • Preparation Of Clay, And Manufacture Of Mixtures Containing Clay Or Cement (AREA)

Abstract

The invention belongs to the technical field of concrete mixing proportion optimization, and particularly discloses a multi-performance control concrete mixing proportion optimization method. The method comprises the following steps: collecting a plurality of groups of concrete mix proportion actual production data, respectively establishing deep learning prediction models of concrete strength, workability and durability according to the data, and establishing a concrete mix proportion mathematical optimization model which takes the minimum cost as a target and meets various performance requirements as constraints on the basis of the strength, workability and durability prediction models. The method combines the characteristics of the concrete mix proportion and the characteristics of the strength, the workability and the durability of the corresponding concrete, constructs a deep learning prediction model based on the consumption of the existing concrete raw materials, the performance of the raw materials and the corresponding strength, workability and durability, fully considers the influence of the properties of the raw materials on the performance of the concrete, has more accurate prediction result, considers the workability and the durability of the concrete as constraints besides the strength, and is more suitable for practical application.

Description

Concrete mixing proportion optimization method with multiple performance controls
Technical Field
The invention belongs to the technical field of concrete mixing proportion optimization, and particularly relates to a multi-performance control concrete mixing proportion optimization method.
Background
Concrete is the most widely applied building material in the world at present, and particularly, the rapid development of domestic capital construction and real estate projects in nearly 10 years brings great opportunities for the concrete industry. However, most domestic concrete production enterprises often have the problems of excessive quality or substandard quality of concrete products. On one hand, the quality is excessive, which causes the increase of production cost, and on the other hand, the quality is not up to the standard, which finally affects the engineering quality. The problems of excessive quality or substandard quality and the like can be well solved by optimizing the mixing ratio of the concrete. The overall idea of concrete mixing proportion optimization is to reduce the raw material cost as much as possible on the premise of meeting the performance indexes (strength, workability and durability) of concrete.
At present, the research on the optimization of the concrete mixing proportion only considers the dosage of each raw material in the aspect of influencing factors, but neglects the influence of the properties of the raw materials on the performance of the concrete, such as the mud content of aggregate, cement and the like; in the aspect of performance control, only strength is considered, and control of workability and durability is neglected. The optimization of the concrete mixing proportion mostly stays on theoretical research, and the actual application is rarely reported. .
Based on the above defects and shortcomings, there is a need in the art to provide a concrete mix proportion optimization method with multiple performance controls, which considers the performances of the raw materials, such as the strength of cement, the fineness of powder, the gradation of aggregate, and the like, and simultaneously considers the workability and durability of concrete, so that the constructed optimization model is more suitable for practical application.
Disclosure of Invention
Aiming at the defects or improvement requirements of the prior art, the invention provides a concrete mixing proportion optimization method with multiple performance controls, wherein the concrete mixing proportion optimization method with multiple performance controls is correspondingly designed by combining the characteristics of the concrete mixing proportion and the characteristics of the strength, the workability and the durability of the corresponding concrete, a deep learning prediction model is constructed on the basis of the consumption of the existing concrete raw materials, the performance of the raw materials and the corresponding strength, the workability and the durability of the raw materials, and the optimal concrete mixing proportion is obtained by taking the consumption of the concrete raw materials, the performance of the raw materials and the corresponding strength, the workability and the durability of the raw materials as constraint conditions with the minimum cost as an objective function. Therefore, the influence of the properties of the raw materials on the performance of the concrete is fully considered, the established optimization model is more accurate, the workability and durability of the concrete are also considered as constraints in addition to the strength of the concrete, and the model is more suitable for practical application.
In order to achieve the aim, the invention provides a multi-performance control concrete mixing proportion optimization method, which comprises the following steps:
s1, collecting multiple groups of concrete mix proportion actual production data and corresponding concrete strength, workability and durability, carrying out standardization processing on the mix proportion actual production data and the corresponding concrete strength, workability and durability, and dividing the standardized mix proportion actual production data and the corresponding concrete strength, workability and durability into a training set and a test set according to a specified proportion;
s2, establishing a deep learning prediction model of concrete strength, workability and durability, wherein a deep learning framework of the deep learning prediction model is TensorFlow, training the deep learning prediction model by using the consumption of raw materials in a training set, the mud content of aggregate, aggregate gradation, the strength of cement and the fineness data of powder as the input of the deep learning prediction model and using the concrete strength, the workability and the durability as the output of the deep learning prediction model, adjusting the weight and the bias of each layer of the deep learning prediction model, outputting the trained deep learning prediction model, and storing the trained deep learning prediction model on equipment;
s3, evaluating the built deep learning prediction model, bringing the data in the test set into the trained deep learning prediction model, and predicting the target values of the test set, namely the concrete strength, the workability and the durability;
s4, on the basis of the trained deep learning prediction model, establishing a concrete mix proportion mathematical optimization model taking the minimum cost as a target and taking various performance requirements of concrete as constraint conditions, substituting the mix proportion production data of the concrete to be tested and the corresponding concrete strength, workability and durability predicted by the deep learning prediction model into the concrete mix proportion mathematical optimization model, and outputting the mix proportion production data of the concrete under the target of the minimum concrete cost and the corresponding concrete strength, workability and durability.
More preferably, the actual production data of the mixing ratio is the dosage of raw materials, the mud content of aggregate, the grading of aggregate, the strength of cement and the fineness of powder;
wherein the raw materials comprise cement, mineral powder, fly ash, water, a water reducing agent, fine aggregate and coarse aggregate.
More preferably, in step S1, the calculation model of the normalization process is as follows:
Figure 459714DEST_PATH_IMAGE001
Figure 634343DEST_PATH_IMAGE002
Figure 766247DEST_PATH_IMAGE003
in the formula (I), the compound is shown in the specification,
Figure 26327DEST_PATH_IMAGE004
for values after data normalization i =1, … …, n being the number of samples, xiFor the values before the data is normalized,
Figure 839563DEST_PATH_IMAGE005
for the column of dataThe mean value, s, is the standard deviation of the data in the column.
More preferably, in step S2, the training of the deep learning prediction model specifically includes the following steps:
carrying out unsupervised layer-by-layer training, sequentially training each self-encoder in the deep learning prediction model, wherein the output of a hidden layer is as follows:
Figure 868699DEST_PATH_IMAGE006
wherein HiFor the output of the ith hidden layer,
Figure 664180DEST_PATH_IMAGE007
in order to perform the non-linear mapping,
Figure 411556DEST_PATH_IMAGE008
the weight from the input layer to the hidden layer of the ith self-encoder,
Figure 28482DEST_PATH_IMAGE009
for the biasing of the ith from the input layer to the hidden layer of the encoder,
Figure 177704DEST_PATH_IMAGE010
is the input of the ith self-encoder,
preserving weights between input and hidden layers
Figure 651410DEST_PATH_IMAGE011
And bias
Figure 823766DEST_PATH_IMAGE012
And outputs H of the hidden layeriObtaining coding processes of m self-encoders in a layer-by-layer training mode as the input of the (i + 1) th self-encoder;
secondly, supervised fine tuning is carried out, the weight W and the offset b of each layer of network are initialized, the weight and the offset of each layer of network are adjusted by using a loss function, and the expression of the loss function is as follows:
Figure 41120DEST_PATH_IMAGE013
wherein the content of the first and second substances,
Figure 44848DEST_PATH_IMAGE014
in the form of an actual value of the value,
Figure 423877DEST_PATH_IMAGE015
in order to predict the value of the target,
in the initial process of the stacked self-coding device, the result of the training process is fully utilized, the network weight and the bias obtained in the pre-training process are used as the initial values of the stacked self-coding neural network, and the loss function is solved by using a gradient descent method, namely the weight and the bias of each layer are adjusted.
Further preferably, in step S3, the mean square error MSE and the correlation coefficient R are calculated2Judging the prediction accuracy of the trained deep learning prediction model, wherein the calculation model of the mean square error MSE is as follows:
Figure 83529DEST_PATH_IMAGE016
the correlation coefficient R2The calculation model of (2) is as follows:
Figure 104574DEST_PATH_IMAGE017
in the formula (I), the compound is shown in the specification,
Figure 464274DEST_PATH_IMAGE018
in order to test the set of target values,
Figure 483045DEST_PATH_IMAGE019
for test set prediction values, n is the number of samples.
Further preferably, in step S2, the calculation model of the objective function is as follows:
Figure 692310DEST_PATH_IMAGE020
in the formula, C is the cost of the single raw material,
Figure 985888DEST_PATH_IMAGE021
is the unit price of the ith raw material, the unit is Yuan/kg,
Figure 901891DEST_PATH_IMAGE022
the raw materials are cement, mineral powder, fly ash, water, a water reducing agent, fine aggregate and coarse aggregate in sequence.
More preferably, in step S2, the constraint condition includes:
water-to-gel ratio constraint, i.e.
Figure 153881DEST_PATH_IMAGE023
Wherein, in the step (A),
Figure 584862DEST_PATH_IMAGE024
、
Figure 885394DEST_PATH_IMAGE025
respectively the lower limit and the upper limit of the water-to-gel ratio, x1Is the amount of cement, x2The amount of the mineral powder is x3Is the amount of fly ash, x4Is the amount of water;
② sand rate constraints, i.e.
Figure 718220DEST_PATH_IMAGE026
Wherein, in the step (A),
Figure 875532DEST_PATH_IMAGE027
respectively, the lower limit and the upper limit of the sand ratio, x6The amount of fine aggregate used, x7The dosage of the coarse aggregate;
③ powder quantity restraint, i.e.
Figure 997072DEST_PATH_IMAGE028
Wherein, in the step (A),
Figure 898032DEST_PATH_IMAGE029
respectively the lower limit and the upper limit of the powder dosage, x1Is the amount of cement, x2The amount of the mineral powder is x3The amount of the fly ash is used;
fourthly, the dosage of each raw material is restricted, namely
Figure 83900DEST_PATH_IMAGE030
Wherein, YLi1、YLi2The lower limit and the upper limit of the dosage of the raw material i, xiThe dosage of the ith raw material;
fifth restraint of concrete strength, i.e.
Figure 615376DEST_PATH_IMAGE031
Wherein, in the step (A),
Figure 20949DEST_PATH_IMAGE032
in order to achieve the lowest required strength of the concrete,
Figure 725600DEST_PATH_IMAGE033
predicting the strength of the model for the trained deep learning prediction;
sixth, the workability of the concrete is restricted
Figure 470702DEST_PATH_IMAGE034
Wherein, in the step (A),
Figure 235396DEST_PATH_IMAGE035
in order to minimize the required workability of the concrete,
Figure 862686DEST_PATH_IMAGE036
predicting the working performance of the trained deep learning prediction model;
strength constraint of concrete, i.e.
Figure 43132DEST_PATH_IMAGE037
Wherein, in the step (A),
Figure 705058DEST_PATH_IMAGE038
for the minimum required durability of the concrete,
Figure 640653DEST_PATH_IMAGE039
and predicting the durability of the model prediction for the trained deep learning.
Generally, compared with the prior art, the above technical solution conceived by the present invention mainly has the following technical advantages:
1. the method considers the influence of partial material properties on the performance of the concrete, the established optimization model is more accurate, the optimization model also considers the workability and durability of the concrete as constraints besides the strength, and the model is more suitable for practical application.
2. The invention takes the reverse error vector of the hidden layer neuron as the gradient term of the output layer, and the neural network takes the reverse error vector as the gradient descent strategy to carry out convergence calculation and timely adjust each weight and threshold value, thereby further improving the prediction precision of the prediction model after adjustment.
Drawings
Fig. 1 is a flow chart of a multi-performance controlled concrete mix proportion optimization method according to a preferred embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
As shown in fig. 1, a method for optimizing a concrete mix ratio with multiple performance controls provided by an embodiment of the present invention includes the following steps:
the method comprises the steps of firstly, acquiring multiple groups of actual production data of concrete mix proportion and corresponding concrete strength, workability and durability, carrying out standardization processing on the actual production data of mix proportion and corresponding concrete strength, workability and durability, and dividing the actual production data of mix proportion and corresponding concrete strength, workability and durability after standardization processing into a training set and a testing set according to a specified proportion.
In the step, the actual production data of the mixing proportion comprise the consumption of raw materials, the mud content of aggregate, the grading of aggregate, the strength of cement and the fineness of powder; wherein the raw materials comprise cement, mineral powder, fly ash, water, a water reducing agent, fine aggregate and coarse aggregate. The calculation model of the normalization process is as follows:
Figure 755239DEST_PATH_IMAGE040
Figure 739376DEST_PATH_IMAGE041
Figure 757272DEST_PATH_IMAGE042
in the formula (I), the compound is shown in the specification,
Figure 863769DEST_PATH_IMAGE043
for values after data normalization i =1, … …, n being the number of samples, xiFor the values before the data is normalized,
Figure 465651DEST_PATH_IMAGE044
is the average of the data in the column, and s is the standard deviation of the data in the column.
In the preferred embodiment of the present invention, the actual production data, the corresponding concrete strength, workability, and durability may be normalized to create a sample set, and the sample set may be divided into a training set and a testing set according to a predetermined ratio. In the invention, the preprocessing of the data is mainly to avoid the condition that the data is too large or too small, which causes the condition that the data is submerged or not converged, and the data is generally subjected to data normalization between-1 and 1.
In addition, all samples are divided into a training set and a testing set according to a certain proportion. The fraction of the fraction, test set, is typically 20% of the total number of samples. In the present invention, the ratio of the training set to the test set is not limited to the above ratio, and generally, the ratio of the total number of samples in the training set to the total number of samples in the test set is 2: 1 to 4: 1.
And step two, establishing a deep learning prediction model of concrete strength, workability and durability, wherein a deep learning framework of the deep learning prediction model is TensorFlow (symbolic mathematical system based on dataflow programming), the deep learning prediction model is trained by taking the consumption of raw materials in a training set, the mud content of aggregate, the aggregate gradation, the strength of cement and the fineness data of powder as the input of the deep learning prediction model and taking the concrete strength, the workability and the durability as the output of the deep learning prediction model, so that the weight and the bias of each layer of the deep learning prediction model are adjusted, the trained deep learning prediction model is output, and the trained deep learning prediction model is stored on equipment. And establishing a strength, working performance and durability prediction model. The method comprises the following specific steps:
first, the inputs and outputs of the model are determined. The input of the model is the dosage of various raw materials, the mud content of aggregate, the grading of aggregate, the strength of cement and the fineness of powder, and the output of the model is the concrete strength, the concrete workability and the concrete durability respectively. And establishing a deep learning prediction model according to the training set data after the standardization processing. The deep learning framework employs Tensorflow. The concrete description is as follows:
and (4) performing unsupervised layer-by-layer training. Training each self-encoder in turn, the output of the hidden layer is:
Figure 50216DEST_PATH_IMAGE006
wherein HiFor the output of the ith hidden layer,
Figure 624417DEST_PATH_IMAGE007
in order to perform the non-linear mapping,
Figure 636236DEST_PATH_IMAGE045
the weight from the input layer to the hidden layer of the ith self-encoder,
Figure 725414DEST_PATH_IMAGE046
for the biasing of the ith from the input layer to the hidden layer of the encoder,
Figure 316933DEST_PATH_IMAGE047
is the input of the ith self-encoder.
Preserving weights between input and hidden layers
Figure 542378DEST_PATH_IMAGE011
And bias
Figure 990677DEST_PATH_IMAGE048
And outputs H of the hidden layeriAnd as the input of the (i + 1) th self-encoder, obtaining the encoding process of the m self-encoders in a layer-by-layer training mode.
② there is supervised fine tuning. The method specifically comprises the following steps: initializing the weight W and the bias b of each layer of network; the weights and biases of the networks of each layer are adjusted using the loss function. The expression of the loss function is as follows:
Figure 770414DEST_PATH_IMAGE013
wherein the content of the first and second substances,
Figure 201176DEST_PATH_IMAGE049
in the form of an actual value of the value,
Figure 546707DEST_PATH_IMAGE050
is a predicted value.
In the initial process of the stacked self-coding device, the result of the training process is fully utilized, the network weight and the bias obtained in the pre-training process are used as the initial values of the stacked self-coding neural network, and the loss function is solved by using a gradient descent method, namely the weight and the bias of each layer are adjusted.
And thirdly, finishing the training of the prediction model and storing the model on the equipment.
In this step, a hidden layer of the deep learning prediction model is added, and when the neural network structure includes the hidden layer, an accumulated error of the back propagation error of the neural network is calculated, that is, the accumulated error of the back propagation error of the deep learning prediction model is calculated. Specifically, the method comprises the following steps: forward propagation input: and calculating the neural network net input vectors of the upper layer neurons i corresponding to all the neurons j of the neural network hidden layer or the output layer, and processing the neural network net input vectors by adopting a Sigmoid function. Calculating the net input vector of the neural network of the upper-layer neuron i corresponding to all the neurons j of the hidden layer or the output layer of the neural network, wherein the net input vector of the neural network is obtained by multiplying the neural network information of the upper-layer neuron i corresponding to all the neurons j by corresponding weight and adding bias. Meanwhile, in order to converge the result, neuron information is processed by adopting a Sigmoid function. And constructing an error square sum calculation model of the neural network, and calculating the error square sum of the neural network.
And calculating a back propagation error, namely, deriving the weight of the deep learning prediction model according to the chain rule of the error square sum, replacing the neural network information by adopting the actual production data of the concrete mix proportion in the training set and the target expected output vector corresponding to the sample in the corresponding concrete strength, workability and durability, and calculating the back error vector of the neuron j of the hidden layer of the neural network.
And (3) calculating an accumulated error: and accumulating and summing error vectors of each neuron j of the hidden layer to obtain accumulated errors of the back propagation errors of the deep learning prediction model.
And establishing a multi-classification support vector machine model of accumulated errors and sample data in a training set. The method comprises the steps of establishing a multi-classification support vector machine model with accumulated errors and input parameters (concrete mix proportion actual production data and corresponding concrete strength, workability and durability), taking a training set as input of the multi-classification support vector machine model, taking accumulated errors corresponding to a plurality of samples in the training set as output of the multi-classification support vector machine model, training the multi-classification support vector machine model, verifying the trained multi-classification support vector machine model by using a verification set to obtain the multi-classification support vector machine model with an error threshold meeting requirements, performing accumulated error training by using the trained multi-classification support vector machine model, and solving an optimal accumulated error. And feeding back the optimal error to the neural network, improving the training precision and adjusting the parameter model. And taking the optimal accumulated error as a descending strategy of the deep learning prediction model, performing convergence calculation, and adjusting the weight and the threshold value of the deep learning prediction model in time so that the prediction precision of the deep learning prediction model meets the requirement. The method transmits the accumulated error calculated by the deep learning prediction model to a multi-classification support vector machine for accumulated error training and solving the optimal accumulated error. The step is a very critical step, the traditional neural network adjusts parameters through self threshold values, and the solution result may have a divergent result due to the multidimensional and nonlinear data and the non-global representativeness of the sample. Therefore, the problem is solved to a limited extent by solving the error by virtue of the multi-classification support vector machine
And step three, evaluating the deep learning prediction model. And (5) standardizing the data of the test set in the same way, substituting the data into the established deep learning prediction model, and predicting the target value of the test set, namely the concrete strength. Calculating Mean Square Error (MSE) and correlation coefficient (R)2) The calculation formula is as follows:
Figure 369169DEST_PATH_IMAGE016
Figure 167361DEST_PATH_IMAGE017
in the formula (I), the compound is shown in the specification,
Figure 897419DEST_PATH_IMAGE051
in order to test the set of target values,
Figure 300719DEST_PATH_IMAGE052
for test set prediction values, n is the number of samples.
And fourthly, on the basis of the strength, workability and durability prediction model, establishing a concrete mix proportion mathematical optimization model which takes the minimum cost as a target and takes all performance requirements as constraints, specifically, establishing a concrete mix proportion mathematical optimization model which takes the minimum cost as a target and takes all performance requirements of the concrete as constraints on the basis of the trained deep learning prediction model, substituting mix proportion production data of the concrete to be tested and the corresponding concrete strength, workability and durability predicted by the deep learning prediction model into the concrete mix proportion mathematical optimization model, and outputting mix proportion production data of the concrete under the concrete cost minimum target and the corresponding concrete strength, workability and durability.
The concrete mixing proportion mathematical optimization model is described as follows:
(1) objective function
Optimizing the target: cost of unilateral raw materials is minimal, i.e.
Figure 90820DEST_PATH_IMAGE020
In the formula, C is the cost of the single raw material,
Figure 376308DEST_PATH_IMAGE021
is the unit price of the i-th raw material, yuan/kg,
Figure 910058DEST_PATH_IMAGE053
the raw materials are cement, mineral powder, fly ash, water, a water reducing agent, fine aggregate and coarse aggregate in sequence.
(2) Constraint conditions
Water-to-gel ratio constraint, i.e.
Figure 167864DEST_PATH_IMAGE023
Wherein, in the step (A),
Figure 128867DEST_PATH_IMAGE054
、
Figure 403115DEST_PATH_IMAGE055
respectively the lower limit and the upper limit of the water-to-gel ratio, x1Is the amount of cement, x2The amount of the mineral powder is x3Is the amount of fly ash, x4Is the amount of water;
② sand rate constraints, i.e.
Figure 678239DEST_PATH_IMAGE056
Wherein, in the step (A),
Figure 852868DEST_PATH_IMAGE057
respectively, the lower limit and the upper limit of the sand ratio, x6The amount of fine aggregate used, x7The dosage of the coarse aggregate;
③ powder quantity restraint, i.e.
Figure 984772DEST_PATH_IMAGE058
Wherein, in the step (A),
Figure 182535DEST_PATH_IMAGE029
respectively the lower limit and the upper limit of the powder dosage, x1Is the amount of cement, x2The amount of the mineral powder is x3The amount of the fly ash is used;
fourthly, the dosage of each raw material is restricted, namely
Figure 323667DEST_PATH_IMAGE030
Wherein, YLi1、YLi2The lower limit and the upper limit of the dosage of the raw material i, xiThe dosage of the ith raw material;
fifth restraint of concrete strength, i.e.
Figure 352803DEST_PATH_IMAGE031
Wherein, in the step (A),
Figure 593291DEST_PATH_IMAGE032
in order to achieve the lowest required strength of the concrete,
Figure 606247DEST_PATH_IMAGE033
predicting the strength of the model for the trained deep learning prediction;
sixth, the workability of the concrete is restricted
Figure 19910DEST_PATH_IMAGE034
Wherein, in the step (A),
Figure 106815DEST_PATH_IMAGE035
in order to minimize the required workability of the concrete,
Figure 580522DEST_PATH_IMAGE036
predicting the working performance of the trained deep learning prediction model;
strength constraint of concrete, i.e.
Figure 48150DEST_PATH_IMAGE037
Wherein, in the step (A),
Figure 468767DEST_PATH_IMAGE059
for the minimum required durability of the concrete,
Figure 206916DEST_PATH_IMAGE039
and predicting the durability of the model prediction for the trained deep learning.
And the model is adopted to optimize and solve the ratio of the concrete to be tested.
And fifthly, collecting actual production data of the same concrete production enterprise, updating the concrete strength, workability and durability depth prediction model, and further updating the concrete mix proportion optimization model.
The concrete formula data collected by the invention is from the No. 2 production line of a certain commercial concrete production enterprise in China. The concrete mix proportion mathematical optimization model is established by adopting the technical method, and the model is solved by adopting a queue competition algorithm. The data pairs before and after optimization are as in the table below.
TABLE 1 comparison of data before and after optimization
Figure 117103DEST_PATH_IMAGE060
In addition to the strength, the optimization model also considers the workability and durability of the concrete as constraints, the model is more suitable for practical application, and as can be seen from the table, the model is more accurate by adopting a deep learning technology to establish a concrete strength, workability and durability prediction model, so that the concrete cost is lower.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (7)

1. A concrete mixing proportion optimization method with multiple performance controls is characterized by comprising the following steps:
s1, collecting multiple groups of concrete mix proportion actual production data and corresponding concrete strength, workability and durability, carrying out standardization processing on the mix proportion actual production data and the corresponding concrete strength, workability and durability, and dividing the standardized mix proportion actual production data and the corresponding concrete strength, workability and durability into a training set and a test set according to a specified proportion;
s2, establishing a deep learning prediction model of concrete strength, workability and durability, wherein a deep learning framework of the deep learning prediction model is TensorFlow, training the deep learning prediction model by using the consumption of raw materials in a training set, the mud content of aggregate, aggregate gradation, the strength of cement and the fineness data of powder as the input of the deep learning prediction model and using the concrete strength, the workability and the durability as the output of the deep learning prediction model, adjusting the weight and the bias of each layer of the deep learning prediction model, outputting the trained deep learning prediction model, and storing the trained deep learning prediction model on equipment;
s3, evaluating the built deep learning prediction model, bringing the data in the test set into the trained deep learning prediction model, and predicting the target values of the test set, namely the concrete strength, the workability and the durability;
s4, on the basis of the trained deep learning prediction model, establishing a concrete mix proportion mathematical optimization model taking the minimum cost as a target and taking various performance requirements of concrete as constraint conditions, substituting the mix proportion production data of the concrete to be tested and the corresponding concrete strength, workability and durability predicted by the deep learning prediction model into the concrete mix proportion mathematical optimization model, and outputting the mix proportion production data of the concrete under the target of the minimum concrete cost and the corresponding concrete strength, workability and durability.
2. The method for optimizing the concrete mixing proportion under the multiple performance control according to claim 1, wherein the actual production data of the mixing proportion are the consumption of raw materials, the mud content of aggregates, the grading of the aggregates, the strength of cement and the fineness of powder;
wherein the raw materials comprise cement, mineral powder, fly ash, water, a water reducing agent, fine aggregate and coarse aggregate.
3. The method for optimizing the mix proportion of concrete with multiple performance controls according to claim 1, wherein in step S1, the calculation model of the standardization process is as follows:
Figure 908602DEST_PATH_IMAGE001
Figure 706793DEST_PATH_IMAGE002
Figure 436852DEST_PATH_IMAGE003
in the formula，
Figure 902468DEST_PATH_IMAGE004
For values after data normalization i =1, … …, n being the number of samples, xiFor the values before the data is normalized,
Figure 426991DEST_PATH_IMAGE005
is the average of the data in the column, and s is the standard deviation of the data in the column.
4. The method as claimed in claim 1, wherein the step S2 of training the deep learning prediction model specifically comprises the following steps:
carrying out unsupervised layer-by-layer training, sequentially training each self-encoder in the deep learning prediction model, wherein the output of a hidden layer is as follows:
Figure 915741DEST_PATH_IMAGE006
wherein HiFor the output of the ith hidden layer,
Figure 715070DEST_PATH_IMAGE007
in order to perform the non-linear mapping,
Figure 769613DEST_PATH_IMAGE008
the weight from the input layer to the hidden layer of the ith self-encoder,
Figure 668299DEST_PATH_IMAGE009
for the biasing of the ith from the input layer to the hidden layer of the encoder,
Figure 706662DEST_PATH_IMAGE010
is the input of the ith self-encoder,
preserving weights between input and hidden layers
Figure 282918DEST_PATH_IMAGE011
And bias
Figure 395230DEST_PATH_IMAGE012
And outputs H of the hidden layeriObtaining coding processes of m self-encoders in a layer-by-layer training mode as the input of the (i + 1) th self-encoder;
secondly, supervised fine tuning is carried out, the weight W and the offset b of each layer of network are initialized, the weight and the offset of each layer of network are adjusted by using a loss function, and the expression of the loss function is as follows:
Figure 792714DEST_PATH_IMAGE013
wherein the content of the first and second substances,
Figure 787215DEST_PATH_IMAGE014
in the form of an actual value of the value,
Figure 866029DEST_PATH_IMAGE015
in order to predict the value of the target,
in the initial process of the stacked self-coding device, the result of the training process is utilized, the network weight and the bias obtained in the pre-training process are used as the initial values of the stacked self-coding neural network, and the loss function is solved by utilizing a gradient descent method, namely the weight and the bias of each layer are adjusted.
5. The method as claimed in claim 1, wherein the step S3 is performed by calculating the mean square error MSE and the correlation coefficient R2Judging the prediction accuracy of the trained deep learning prediction model, wherein the calculation model of the mean square error MSE is as follows:
Figure 629586DEST_PATH_IMAGE016
the correlation coefficient R2The calculation model of (2) is as follows:
Figure 197970DEST_PATH_IMAGE017
in the formula (I), the compound is shown in the specification,
Figure 883030DEST_PATH_IMAGE018
in order to test the set of target values,
Figure 562273DEST_PATH_IMAGE019
for test set prediction values, n is the number of samples.
6. The method for optimizing the mix proportion of concrete with multiple performance controls according to claim 1, wherein in step S2, the calculation model of the objective function is as follows:
Figure 445915DEST_PATH_IMAGE020
in the formula, C is the cost of the single raw material,
Figure 122884DEST_PATH_IMAGE021
is the unit price of the i-th raw material,
Figure 91977DEST_PATH_IMAGE022
the raw materials are cement, mineral powder, fly ash, water, a water reducing agent, fine aggregate and coarse aggregate in sequence.
7. The method for optimizing the mix proportion of concrete with multiple performance controls according to claim 1, wherein in step S2, the constraint conditions include:
water-to-gel ratio constraint, i.e.
Figure 76376DEST_PATH_IMAGE023
Wherein, in the step (A),
Figure 17787DEST_PATH_IMAGE024
、
Figure 662395DEST_PATH_IMAGE025
respectively the lower limit and the upper limit of the water-to-gel ratio, x1Is the amount of cement, x2The amount of the mineral powder is x3Is the amount of fly ash, x4Is the amount of water;
② sand rate constraints, i.e.
Figure 384363DEST_PATH_IMAGE026
Wherein, in the step (A),
Figure 139830DEST_PATH_IMAGE027
respectively, the lower limit and the upper limit of the sand ratio, x6The amount of fine aggregate used, x7The dosage of the coarse aggregate;
③ powder quantity restraint, i.e.
Figure 201327DEST_PATH_IMAGE028
Wherein, in the step (A),
Figure 16836DEST_PATH_IMAGE029
respectively the lower limit and the upper limit of the powder dosage, x1Is the amount of cement, x2The amount of the mineral powder is x3The amount of the fly ash is used;
fourthly, the dosage of each raw material is restricted, namely
Figure 960521DEST_PATH_IMAGE030
Wherein, YLi1、YLi2The lower limit and the upper limit of the dosage of the raw material i, xiThe dosage of the ith raw material;
fifth restraint of concrete strength, i.e.
Figure 722941DEST_PATH_IMAGE031
Wherein, in the step (A),
Figure 701261DEST_PATH_IMAGE032
in order to achieve the lowest required strength of the concrete,
Figure 687672DEST_PATH_IMAGE033
predicting the strength of the model for the trained deep learning prediction;
sixth, the workability of the concrete is restricted
Figure 56336DEST_PATH_IMAGE034
Wherein, in the step (A),
Figure 917719DEST_PATH_IMAGE035
in order to minimize the required workability of the concrete,
Figure 750546DEST_PATH_IMAGE036
predicting the working performance of the trained deep learning prediction model;
strength constraint of concrete, i.e.
Figure 111120DEST_PATH_IMAGE037
Wherein, in the step (A),
Figure 29398DEST_PATH_IMAGE038
for the minimum required durability of the concrete,
Figure 930358DEST_PATH_IMAGE039
and predicting the durability of the model prediction for the trained deep learning.
CN202111582571.7A 2021-12-22 2021-12-22 Concrete mixing proportion optimization method with multiple performance controls Pending CN114388069A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111582571.7A CN114388069A (en) 2021-12-22 2021-12-22 Concrete mixing proportion optimization method with multiple performance controls

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111582571.7A CN114388069A (en) 2021-12-22 2021-12-22 Concrete mixing proportion optimization method with multiple performance controls

Publications (1)

Publication Number Publication Date
CN114388069A true CN114388069A (en) 2022-04-22

Family

ID=81196940

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111582571.7A Pending CN114388069A (en) 2021-12-22 2021-12-22 Concrete mixing proportion optimization method with multiple performance controls

Country Status (1)

Country Link
CN (1) CN114388069A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115010232A (en) * 2022-07-11 2022-09-06 湘潭中环水务有限公司 Drinking water plant coagulation dosing prediction method based on weak time-lag neural network model
WO2023223218A1 (en) * 2022-05-17 2023-11-23 Concrete Ai, Inc. Machine learning concrete optimization
CN117610435A (en) * 2024-01-22 2024-02-27 天津路联智通交通科技有限公司 Automatic proportioning method and system for civil and architectural construction mixed materials

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112016244A (en) * 2020-08-07 2020-12-01 中国交通建设股份有限公司吉林省分公司 Durable concrete multi-target mix proportion optimization method based on SVM and intelligent algorithm
CN112069656A (en) * 2020-08-07 2020-12-11 湖北交投十巫高速公路有限公司 Durable concrete mix proportion multi-objective optimization method based on LSSVM-NSGAII
CN112684156A (en) * 2021-01-06 2021-04-20 汉谷云智(武汉)科技有限公司 Concrete strength prediction method based on deep learning

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112016244A (en) * 2020-08-07 2020-12-01 中国交通建设股份有限公司吉林省分公司 Durable concrete multi-target mix proportion optimization method based on SVM and intelligent algorithm
CN112069656A (en) * 2020-08-07 2020-12-11 湖北交投十巫高速公路有限公司 Durable concrete mix proportion multi-objective optimization method based on LSSVM-NSGAII
CN112684156A (en) * 2021-01-06 2021-04-20 汉谷云智(武汉)科技有限公司 Concrete strength prediction method based on deep learning

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023223218A1 (en) * 2022-05-17 2023-11-23 Concrete Ai, Inc. Machine learning concrete optimization
US11897818B2 (en) 2022-05-17 2024-02-13 Concrete Ai, Inc. Machine learning concrete optimization
CN115010232A (en) * 2022-07-11 2022-09-06 湘潭中环水务有限公司 Drinking water plant coagulation dosing prediction method based on weak time-lag neural network model
CN117610435A (en) * 2024-01-22 2024-02-27 天津路联智通交通科技有限公司 Automatic proportioning method and system for civil and architectural construction mixed materials
CN117610435B (en) * 2024-01-22 2024-04-09 天津路联智通交通科技有限公司 Automatic proportioning method and system for civil and architectural construction mixed materials

Similar Documents

Publication Publication Date Title
CN114388069A (en) Concrete mixing proportion optimization method with multiple performance controls
CN101158674B (en) Method for predicting chlorophyll a concentration in water based on BP nerval net
CN111985796B (en) Method for predicting concrete structure durability based on random forest and intelligent algorithm
CN111079906B (en) Cement finished product specific surface area prediction method and system based on long-short-term memory network
CN110824915B (en) GA-DBN network-based intelligent monitoring method and system for wastewater treatment
CN112070356B (en) Method for predicting carbonization resistance of concrete based on RF-LSSVM model
CN109919356B (en) BP neural network-based interval water demand prediction method
CN112016244B (en) Durable concrete multi-target mix proportion optimization method based on SVM and intelligent algorithm
CN113761748B (en) Industrial process soft measurement method based on federal incremental random configuration network
CN108204944A (en) The Buried Pipeline rate prediction method of LSSVM based on APSO optimizations
CN112684156A (en) Concrete strength prediction method based on deep learning
CN106568647B (en) A kind of Strength Forecast of Concrete method neural network based
CN102033991A (en) Microwave drying prediction method through BP (back-propagation) neural network based on incremental improvement
CN110598929A (en) Wind power nonparametric probability interval ultrashort term prediction method
CN113191092A (en) Industrial process product quality soft measurement method based on orthogonal increment random configuration network
CN110364232A (en) It is a kind of based on memristor-gradient descent method neural network Strength of High Performance Concrete prediction technique
CN113393057A (en) Wheat yield integrated prediction method based on deep fusion machine learning model
CN110222825B (en) Cement product specific surface area prediction method and system
CN112819087A (en) Effluent BOD sensor abnormity detection method based on modular neural network
CN113033105A (en) Steel material fatigue performance prediction method based on mechanical theory guide transfer learning
CN110008566B (en) Filling slurry performance index prediction method based on big data
CN110648023A (en) Method for establishing data prediction model based on quadratic exponential smoothing improved GM (1,1)
CN113377075A (en) Method and device for optimizing rare earth extraction process in real time and computer readable storage medium
CN114037148B (en) Daily maximum power load prediction method based on genetic neural network and adaptive weight
Gai et al. Research on prediction model of National Railway Freight Volume based on GA-BP network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination