A multi-objective control optimization method for concrete mix ratio
Through the combination of genetic algorithm and BP neural network, a multi-objective optimization design of concrete mix ratio is achieved, which solves the problem that performance and price are difficult to optimize simultaneously in the existing technology, improves the design efficiency and accuracy, and is suitable for different regions. Raw materials and performance indicators.
Patent Information
- Application Number
- CN202111140305.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-09-28
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2041-09-28
AI Technical Summary
The existing concrete mix design is difficult to achieve optimal performance and price at the same time, and due to the geographical differences of raw materials, the design process is time-consuming and labor-intensive, and there is a lack of unified standards.
Genetic algorithm optimization neural network is used to establish a concrete mix performance prediction model, combine BP neural network and particle swarm optimization algorithm, and find the mix ratio that meets strength, ease and price indicators through the iteration of the objective function to achieve multi-objective optimization design.
It can design according to raw materials in different regions, improve the design accuracy and efficiency, reduce the number of adaptations, save manpower and material resources, and achieve good control over concrete performance and price.
Smart Images

Figure CN113836811B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of concrete mix design, and in particular relates to a concrete mix multi-objective control optimization method. Background Art
[0002] Due to the complex composition of concrete and the large regional differences in raw materials, it is difficult to unify standards in mix design. In the process of mix adaptation, the test cycle is long and requires a lot of time, manpower and material resources. Conventional design methods generally calculate according to specifications and design mixes with the goal of meeting performance requirements. The resulting mixes can generally meet performance requirements, but because cost factors are ignored, it is usually difficult to achieve the best performance and price at the same time. Summary of the invention
[0003] The main purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art and to propose a multi-objective control optimization method for concrete mix ratio, which optimizes the concrete mix ratio by taking strength, workability and price as the three targets.
[0004] In order to achieve the above object, the present invention adopts the following technical solutions:
[0005] A multi-objective control optimization method for concrete mix ratio comprises the following steps:
[0006] S1. Establish a prediction model for concrete mix ratio performance, use genetic algorithm to optimize neural network to establish a prediction model for concrete mix ratio performance, and train the prediction model with a large amount of mix ratio sample data;
[0007] S2. Establishing the objective function of mix ratio optimization;
[0008] S3, iterative optimization, find the mix ratio that meets the target, use particle swarm optimization as a strategy for random search, calculate the objective function value for each particle in the population and compare it, update the particle swarm, find the group extreme value, and the particle corresponding to the group extreme value is the target mix ratio.
[0009] Furthermore, a BP neural network is used as the modeling carrier of the prediction model, and the network includes an input layer, a hidden layer and an output layer;
[0010] The number of neurons in the input layer is the number of raw material types used in the mix;
[0011] The number of neurons in the output layer is the number of performance indicators of concrete;
[0012] The number of neurons in the hidden layer is N h Determined according to the empirical formula, specifically:
[0013]
[0014] Among them, N i is the number of neurons in the input layer, N o is the number of neurons in the output layer, N s is the number of samples in the training set, and α is a coefficient ranging from 2 to 10.
[0015] Furthermore, the BP neural network was trained, and the input units were the amounts of cement, fly ash, mineral powder, fine aggregate, coarse aggregate, water, and admixtures per cubic meter of concrete;
[0016] The output units are the performance indicators of the 28d compressive strength of concrete, slump and mix ratio, which have a nonlinear relationship.
[0017] Furthermore, the specific training steps of the BP neural network are as follows:
[0018] Define the neural network input layer to have n neurons, the hidden layer to have p neurons, and the output layer to have q neurons;
[0019] The input variables are X=(X1,X2,...,X n ); the hidden layer input variable is hi=(hi1,hi2,...,hi p ); the hidden layer output variable is ho=(ho1,ho2,...,ho p ); the input variable of the output layer is yi=(yi1,yi2,...,yi q ); the output variable of the output layer is yo=(yo1,yo2,...yo q ); the expected output variable is Y=(Y1,Y2,...,Y q );
[0020] Randomly select the kth sample and the corresponding expected output:
[0021] X(k)=(X1(k),X2(k),...,X n (k))
[0022] Y(k)=(Y1(k),Y2(k),...,Y q (k))
[0023] Among them, the number of data samples k = 1, 2, ..., m;
[0024] Calculate the input and output of each neuron in the hidden layer:
[0025]
[0026] ho h (k) = f(hi h(k)), h=1,2,...,p
[0027]
[0028] yo o (k) = f(yi o (k)), o = 1, 2, ..., q
[0029] Among them, w ih is the connection weight between the input layer and the hidden layer, w ho is the connection weight between the hidden layer and the output layer, b h is the threshold of neurons in the hidden layer, b o is the threshold of each neuron in the output layer, and f(·) is the activation function;
[0030] Using the δ of each neuron in the output layer o (k) and the output value of each neuron in the hidden layer, and correct the connection weight w between the output layer and the hidden layer ho (k):
[0031]
[0032]
[0033] Where e is the error function,
[0034] Using the hidden layer neurons δ h (k) and the input values of each neuron in the input layer, and modify the connection weight w between the hidden layer and the input layer ih (k):
[0035]
[0036]
[0037] Calculate the global error E:
[0038]
[0039] Determine whether the error meets the requirements. If so, end the training. If not, use the updated weight threshold to return to the step to calculate the input and output of each neuron in the hidden layer, and repeat the subsequent steps until the global error E meets the requirements.
[0040] After training, a prediction model is obtained that can predict concrete properties based on mix ratio data.
[0041] Furthermore, the objective function is specifically:
[0042] Fun(CE,F,M,SA,ST,W,A)=a×P+b×C+PF (2)
[0043] Where a and b are weights, which are determined according to actual needs, and P is the performance index calculated by formula (3):
[0044] P=(PP-AP) 2 (3)
[0045] Where PP is the predicted performance value obtained by inputting the mix ratio into the trained artificial neural network, AP is the optimized target performance, and C is the price index, which is calculated by formula (4):
[0046] C=|PC-AC| (4)
[0047] Among them, PC is the predicted unit cost of the mix, AC is the optimized target unit cost of the mix, PF is the penalty function, and the provisions of the concrete mix design code for the amount of each component are used as the boundary constraint construction.
[0048] Further, iterative optimization is performed to find the mix ratio that meets the target:
[0049] The random search is performed using particle swarm optimization as a strategy. The specific search process is as follows:
[0050] S31. Randomly generate an initial population S n , where each particle S i Represents a set of mix ratios S i =(CE i F i M i SA i ST i W i A i ), i = 1, 2, 3, ..., n, with speed V id , position P id represents the eigenvalue of the point;
[0051] S32, Substitute each particle in the population into formula (2) to calculate its corresponding objective function value Fun i =Fun(S i ), by comparing the objective function value Fun corresponding to each particle i The size of , we get a group extreme value and an individual extreme value;
[0052] S33, updating the speed and position of each particle according to formula (5) and formula (6), formula (5) and formula (6) are as follows:
[0053]
[0054]
[0055] Among them, c1 and c2 are learning coefficients, r1 and r2 are random numbers between 0 and 1, and ω is the inertia weight. is the individual velocity of the kth generation particle, The individual positions of the kth generation particles, is the individual extreme value of the kth generation particle, is the population extreme value of the kth generation particles;
[0056] S34, after the update, the objective function value of the new generation of particles is calculated again, and new individual extreme values and group extreme values are obtained;
[0057] S35. Update the particle swarm according to the new individual extreme value and group extreme value using formula (7), calculate the objective function value, and obtain the individual extreme value and group extreme value. Formula (7) is as follows:
[0058]
[0059] After looping steps S32 to S35 for a specified number of times, the final group extreme value is obtained, and the corresponding particles are the mix ratio closest to the target performance and target price.
[0060] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0061] 1. The present invention can be studied according to the raw materials in different regions, and the mix ratio can be designed according to the prices of different raw materials in different regions. In addition, the present invention can be applied to different concrete carriers, and the required concrete mix ratio can be designed according to different performance indicators.
[0062] 2. The prediction model of the present invention can memorize the data input in the past. With the increase of accumulated data, the optimization effect and accuracy will continue to improve. In addition, in the process of searching for the mix ratio, the prediction model can memorize the optimal mix ratio that appears, and can continuously improve the convergence speed while ensuring the search range. The use of the present invention for concrete mix ratio design can have a better control over the various properties and prices of concrete, and can also reduce the number of adaptations, greatly saving manpower, material resources and resources. BRIEF DESCRIPTION OF THE DRAWINGS
[0063] Figure 1 is a flow chart of the method of the present invention;
[0064] Figure 2 It is the control principle diagram of the present invention;
[0065] Figure 3 It is a schematic diagram of BP neural network training of the present invention;
[0066] Figure 4 It is a flow chart of the present invention for finding a mix ratio that meets the target. DETAILED DESCRIPTION
[0067] like Figure 1 , Figure 2 As shown, the present invention provides a concrete mix ratio multi-objective control optimization method, comprising the following steps:
[0068] S1. Establish a prediction model for concrete mix ratio performance. Use genetic algorithm to optimize neural network to establish a prediction model for concrete mix ratio performance. Use a large amount of mix ratio sample data to train the prediction model. Specifically:
[0069] The BP neural network is used as the modeling carrier of the prediction model. The network includes an input layer, a hidden layer and an output layer.
[0070] The number of neurons in the input layer is the number of raw material types used in the mix;
[0071] The number of neurons in the output layer is the number of performance indicators of concrete;
[0072] The number of neurons in the hidden layer is N h Determined according to the empirical formula, specifically:
[0073]
[0074] Among them, N i is the number of neurons in the input layer, N o is the number of neurons in the output layer, N s is the number of samples in the training set, α is a coefficient ranging from 2 to 10.
[0075] The BP neural network is trained, and the input units are the amount of cement, fly ash, mineral powder, fine aggregate, coarse aggregate, water and admixture per cubic meter of concrete;
[0076] The output units are the performance indicators of the nonlinear relationship between the 28d compressive strength and slump of concrete and the mix ratio;
[0077] like Figure 3 As shown, the specific training steps of the BP neural network are as follows:
[0078] Define the neural network input layer to have n neurons, the hidden layer to have p neurons, and the output layer to have q neurons;
[0079] The input variables are X=(X1,X2,...,X n ); the hidden layer input variable is hi=(hi1,hi2,...,hi p); the hidden layer output variable is ho=(ho1,ho2,...,ho p ); the input variable of the output layer is yi=(yi1,yi2,...,yi q ); the output variable of the output layer is yo=(yo1,yo2,...yo q ); the expected output variable is Y=(Y1,Y2,...,Y q );
[0080] Randomly select the kth sample and the corresponding expected output:
[0081] X(k)=(X1(k),X2(k),...,X n (k))
[0082] Y(k)=(Y1(k),Y2(k),...,Y q (k))
[0083] Among them, the number of data samples k = 1, 2, ..., m;
[0084] Calculate the input and output of each neuron in the hidden layer:
[0085]
[0086] ho h (k) = f(hi h (k)), h=1,2,...,p
[0087]
[0088] yo o (k) = f(yi o (k)), o=1,2,...,q
[0089] Among them, w ih is the connection weight between the input layer and the hidden layer, w ho is the connection weight between the hidden layer and the output layer, b h is the threshold of neurons in the hidden layer, b o is the threshold of each neuron in the output layer, and f(·) is the activation function;
[0090] Using the δ of each neuron in the output layer o (k) and the output value of each neuron in the hidden layer, and correct the connection weight w between the output layer and the hidden layer ho (k):
[0091]
[0092]
[0093] Where e is the error function,
[0094] Using the hidden layer neurons δ h (k) and the input values of each neuron in the input layer, and modify the connection weight w between the hidden layer and the input layer ih (k):
[0095]
[0096]
[0097] Calculate the global error E:
[0098]
[0099] Determine whether the error meets the requirements. If so, end the training. If not, use the updated weight threshold to return to the step to calculate the input and output of each neuron in the hidden layer, and repeat the subsequent steps until the global error E meets the requirements.
[0100] After training, a prediction model is obtained that can predict concrete properties based on mix ratio data.
[0101] S2. Establish an objective function for mix ratio optimization; the specific objective function is:
[0102] Fun(CE,F,M,SA,ST,W,A)=a×P+b×C+PF (2)
[0103] Where a and b are weights, which are determined according to actual needs, and P is the performance index calculated by formula (3):
[0104] P=(PP-AP) 2 (3)
[0105] Where PP is the predicted performance value obtained by inputting the mix ratio into the trained artificial neural network, AP is the optimized target performance, and C is the price index, which is calculated by formula (4):
[0106] C=|PC-AC| (4)
[0107] Among them, PC is the predicted unit cost of the mix, AC is the optimized unit cost of the mix. PF is a penalty function, which is constructed as a boundary constraint according to the provisions of the concrete mix design code for the amount of each component.
[0108] S3, iterative optimization, find the mix ratio that meets the target, use particle swarm optimization as a strategy for random search, calculate the objective function value for each particle in the population and compare it, update the particle swarm, find the group extreme value, and the particle corresponding to the group extreme value is the target mix ratio.
[0109] like Figure 4 As shown, the iterative optimization and search for the mix ratio that meets the target are as follows:
[0110] S31. Randomly generate an initial population S n , where each particle S i Represents a set of mix ratios S i =(CE i F i M i SA i ST i W i A i ), i = 1, 2, 3, ..., n, with speed V id , position P id represents the eigenvalue of the point;
[0111] S32, Substitute each particle in the population into formula (2) to calculate its corresponding objective function value Fun i =Fun(S i ), by comparing the objective function value Fun corresponding to each particle i The size of , we get a group extreme value and an individual extreme value;
[0112] S33, updating the speed and position of each particle according to formula (5) and formula (6), formula (5) and formula (6) are as follows:
[0113]
[0114]
[0115] Among them, c1 and c2 are learning coefficients, r1 and r2 are random numbers between 0 and 1, and ω is the inertia weight. is the individual velocity of the kth generation particle, The individual positions of the kth generation particles, is the individual extreme value of the kth generation particle, is the population extreme value of the kth generation particles;
[0116] S34, after the update, the objective function value of the new generation of particles is calculated again, and new individual extreme values and group extreme values are obtained;
[0117] S35. Update the particle swarm according to the new individual extreme value and group extreme value using formula (7), calculate the objective function value, and obtain the individual extreme value and group extreme value. Formula (7) is as follows:
[0118]
[0119] After looping steps S32 to S35 for a specified number of times, the final group extreme value is obtained, and the corresponding particles are the mix ratio closest to the target performance and target price.
[0120] The present invention is further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.
[0121] Example 1
[0122] A mix design of sea sand concrete. The components of sea sand concrete are cement, fly ash, mineral powder, fine aggregate 1 (washed sea sand), fine aggregate 2, coarse aggregate 1, coarse aggregate 2, admixture 1, admixture 2, and water, a total of ten raw materials.
[0123] Taking 28-day compressive strength and slump as performance indicators of mix design, a 10-15-2-2 artificial neural network was established. 80 groups of mix ratios and corresponding performance data were used to train the network to obtain a model that can predict performance by mix ratio.
[0124] The weighted objectives of strength, slump and price are used as the objective function, and the particle swarm algorithm is used to search for the mix ratio that meets the requirements.
[0125] In this embodiment, the objective function is specifically:
[0126] Fun(CE,F,M,SA,ST,W,A)=a×P+b×C+PF
[0127] Among them, a and b are weights, which are determined according to actual needs, and P is the performance index calculated by the following formula:
[0128] P=(PP-AP) 2
[0129] Among them, PP is the predicted strength and slump value obtained by inputting the mix ratio into the trained artificial neural network, AP is the optimized target strength and slump value; C is the price index, which is calculated by the following formula:
[0130] C=|PC-AC|
[0131] Among them, PC is the predicted unit cost of the mix, AC is the optimized unit cost of the mix. PF is a penalty function, which is constructed as a boundary constraint according to the provisions of the concrete mix design code for the amount of each component.
[0132] As shown in Table 1 below, this is the mix ratio obtained in Example 1.
[0133]
[0134] Table 1
[0135] Example 2
[0136] A mix design for permeable concrete. The components of permeable concrete are cement, water, crushed stone, water reducer, silica fume, fly ash, slag and reinforcing agent, a total of 8 raw materials.
[0137] Taking 28-day compressive strength and effective porosity as performance indicators for mix design, an 8-12-2-2 artificial neural network was established. 120 groups of mix proportions and corresponding performance data were used to train the network to obtain a model that can predict performance by mix proportion.
[0138] The three weighted objectives of strength, effective porosity and price are taken as the objective function, and the particle swarm algorithm is used to search for the mix ratio that meets the requirements.
[0139] In this embodiment, the objective function is specifically:
[0140] Fun(CE,F,M,SA,ST,W,A)=a×P+b×C+PF
[0141] Among them, a and b are weights, which are determined according to actual needs, and P is the performance index calculated by the following formula:
[0142] P=(PP-AP) 2
[0143] Among them, PP is the predicted strength and effective porosity value obtained by inputting the mix ratio into the trained artificial neural network, AP is the optimized target strength and effective porosity value; C is the price index, which is calculated by the following formula:
[0144] C=|PC-AC|(4)
[0145] Among them, PC is the predicted unit cost of the mix, AC is the optimized unit cost of the mix. PF is a penalty function, which is constructed as a boundary constraint according to the provisions of the concrete mix design code for the amount of each component.
[0146] As shown in Table 2 below, this is the mix ratio obtained in Example 2.
[0147]
[0148] Table 2
[0149] It should also be noted that in this specification, terms such as "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of more restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.
[0150] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A multi-objective control optimization method for concrete mix ratio, characterized in that: The following steps are involved: S1. Establish a prediction model for concrete mix ratio performance, use genetic algorithm to optimize neural network to establish a prediction model for concrete mix ratio performance, and train the prediction model with a large amount of mix ratio sample data; S2. Establish an objective function for mix ratio optimization; the specific objective function is: Fun(CE,F,M,SA,ST,W,A)=a×P+b×C+PF (2) Where a and b are weights, which are determined according to actual needs, and P is the performance index calculated by formula (3): P=(PP-AP) 2 (3) Where PP is the predicted performance value obtained by inputting the mix ratio into the trained artificial neural network, AP is the optimized target performance, and C is the price index, which is calculated by formula (4): C=|PC-AC| (4) Among them, PC is the predicted unit cost of the mix, AC is the optimized target unit cost of the mix, PF is the penalty function, and the provisions of the concrete mix design code for the amount of each component are used as boundary constraints. S3, iterative optimization, find the mix ratio that meets the target, use particle swarm optimization as a strategy to perform random search, calculate the objective function value for each particle in the population and compare, update the particle swarm, find the group extreme value, and the particle corresponding to the group extreme value is the target mix ratio; Iterative optimization to find the mix ratio that meets the target is as follows: The random search is performed using particle swarm optimization as a strategy. The specific search process is as follows: S31. Randomly generate an initial population S n , where each particle S i Represents a set of mix ratios S i =(CE i F i M i S i ST i W i A i ), i = 1, 2, 3, ..., n, with speed V id , position P id represents the characteristic value of the particle; S32, Substitute each particle in the population into formula (2) to calculate its corresponding objective function value Fun i =Fun(S i ), by comparing the objective function value Fun corresponding to each particle i The size of , we get a group extreme value and an individual extreme value; S33, updating the speed and position of each particle according to formula (5) and formula (6), formula (5) and formula (6) are as follows: Among them, c1 and c2 are learning coefficients, r1 and r2 are random numbers between 0 and 1, and ω is the inertia weight. is the individual velocity of the kth generation particle, The individual positions of the kth generation particles, is the individual extreme value of the kth generation particle, is the population extreme value of the kth generation particles; S34, after the update, the objective function value of the new generation of particles is calculated again, and new individual extreme values and group extreme values are obtained; S35. Update the particle swarm according to the new individual extreme value and group extreme value using formula (7), calculate the objective function value, and obtain the individual extreme value and group extreme value. Formula (7) is as follows: After looping steps S32 to S35 for a specified number of times, the final group extreme value is obtained, and the corresponding particles are the mix ratio closest to the target performance and target price.
2. A concrete mix ratio multi-objective control optimization method according to claim 1, characterized in that: The BP neural network is used as the modeling carrier of the prediction model. The network includes an input layer, a hidden layer and an output layer. The number of neurons in the input layer is the number of raw material types used in the mix; The number of neurons in the output layer is the number of performance indicators of concrete; The number of neurons in the hidden layer is N h Determined according to the empirical formula, specifically: Among them, N i is the number of neurons in the input layer, N o is the number of neurons in the output layer, N s is the number of samples in the training set, and α is a coefficient ranging from 2 to 10.
3. A concrete mix ratio multi-objective control optimization method according to claim 2, characterized in that: The BP neural network is trained, and the input units are the amount of cement, fly ash, mineral powder, fine aggregate, coarse aggregate, water and admixture per cubic meter of concrete; The output units are the performance indicators of the 28d compressive strength of concrete, slump and mix ratio, which have a nonlinear relationship.
4. A concrete mix ratio multi-objective control optimization method according to claim 3, characterized in that: The specific training steps of BP neural network are as follows: Define the neural network input layer to have n neurons, the hidden layer to have p neurons, and the output layer to have q neurons; The input variables are X=(X1,X2,...,X n ); the hidden layer input variable is hi=(hi1,hi2,...,hi p ); the hidden layer output variable is ho=(ho1,ho2,...,ho p ); the input variable of the output layer is yi=(yi1,yi2,...,yi q ); the output variable of the output layer is yo=(yo1,yo2,...yo q ); the expected output variable is Y=(Y1,Y2,...,Y q ); Randomly select the kth sample and the corresponding expected output: X(k)=(X1(k),X2(k),...,X n (k)) Y(k)=(Y1(k),Y2(k),...,Y q (k)) Among them, the number of data samples k = 1, 2, ..., m; Calculate the input and output of each neuron in the hidden layer: ho h (k)=f(hi h (k)),h=1,2,...,p yo o (k) = f(yi o (k)), o=1,2,...,q Among them, w ih is the connection weight between the input layer and the hidden layer, w ho is the connection weight between the hidden layer and the output layer, b h is the threshold of neurons in the hidden layer, b o is the threshold of each neuron in the output layer, and f(·) is the activation function; Using the δ of each neuron in the output layer o (k) and the output value of each neuron in the hidden layer, and correct the connection weight w between the output layer and the hidden layer ho (k): Where e is the error function, Using the δ of each neuron in the hidden layer h (k) and the input values of each neuron in the input layer, and modify the connection weight w between the hidden layer and the input layer ih (k): Calculate the global error E: Determine whether the error meets the requirements. If so, end the training. If not, use the updated weight threshold to return to the step to calculate the input and output of each neuron in the hidden layer, and repeat the subsequent steps until the global error E meets the requirements. After training, a prediction model is obtained that can predict concrete properties based on mix ratio data.
Citation Information
Cited By
Concrete proportion multi-objective parallel design method based on sparse Gaussian process Bayesian optimization
CN122266563A
Concrete proportion optimization method and system based on gated attention and multi-objective differential evolution
CN122528589A
Concrete proportioning optimization method and system based on gated attention and multi-objective differential evolution
CN122528589B
