Concrete performance prediction method based on principal component and PSO-BP network

Abstract

The invention discloses a concrete performance prediction method based on a principal component and a PSO-BP network, and belongs to the field of concrete performance prediction.The method comprises the following steps that a concrete data set is established, and data preprocessing is carried out; the method comprises the following steps: processing concrete proportion data by using a principal component analysis method, and dividing the processed data into a training set and a prediction set; establishing a neural network prediction model of concrete strength, slump, yield stress and plastic viscosity; optimizing the neural network by using a particle swarm algorithm; and importing the training set into the optimized neural network for training, inputting the concrete proportion in the prediction set into the trained network to obtain a prediction result, and comparing the prediction result with a corresponding result of the prediction set. The method provided by the invention overcomes the problems of large deviation and low efficiency of concrete performance prediction by a traditional empirical formula and an existing prediction model, and can improve the accuracy and rapidity of concrete strength, slump, yield stress and plastic viscosity prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of concrete performance prediction, and in particular to a concrete performance prediction method based on principal components and a PSO-BP network. Background Art

[0002] With underground coal mining, tunnel surrounding rock deformation and tunnel collapse are very likely to occur, especially during high-intensity mining, which seriously threatens the lives of underground coal miners. Higher requirements are also put forward for the quality of support. Concrete spraying is an important link in tunnel support. The most critical thing in the spraying link is the quality of concrete. Concrete refers to the general term for engineering composite materials that combine water, cement, sand, gravel, water reducer, etc. into a whole. It is the main engineering material that affects the safety of support. In the process of concrete spraying, the fluidity and strength of concrete are crucial to the quality of support. The slump, yield stress and plastic viscosity of concrete slurry are the main indicators for measuring the fluidity of concrete slurry; compressive strength is an important indicator for measuring the strength of concrete. The above concrete properties are related to concrete components such as water consumption, water-cement ratio, sand ratio, and water reducer dosage. Concrete slurries with different proportions have different slumps, compressive strength, yield stress and plastic viscosity, which in turn affects the economic benefits and safety of tunnel wet spraying support. Designers often need to design the mix ratio according to the actual construction conditions and the specified specifications, and use the design strength and slump as the basis to preliminarily evaluate whether the strength and slump of the designed concrete meet the target requirements. During the multiple designs and experiments, experts and scholars used relevant data for regression analysis, and then derived prediction formulas related to strength and slump. However, the traditional empirical formula has large deviations and low efficiency in predicting concrete performance. In addition, the existing statistical learning methods and their prediction models have large prediction errors and poor prediction model stability. Therefore, it is very necessary to improve the accuracy and robustness of the concrete performance prediction model. Efficient and accurate prediction of concrete strength, slump, yield stress, and plastic viscosity is an urgent problem to be solved. Summary of the invention

[0003] The purpose of the present invention is to provide a concrete performance prediction method based on principal components and PSO-BP network, which overcomes the problems of large deviation and low efficiency in concrete performance prediction of traditional empirical formulas and existing prediction models, and can improve the accuracy and rapidity of concrete strength, slump, yield stress and plastic viscosity prediction.

[0004] To achieve the above object, the present invention provides a method for predicting concrete properties based on principal components and PSO-BP network, comprising the following steps:

[0005] Step S1: Establishing a concrete data set: Collecting data, including mix ratio parameters, yield stress, plastic viscosity, slump, and strength parameters of concrete raw materials, pre-processing the collected data through data cleaning and normalization, and removing abnormal data;

[0006] Step S2: using principal component analysis to process the concrete mix data, eliminating the correlation of input factors, reducing the number of input factors, and dividing the data set processed by the principal component into a training set and a prediction set;

[0007] Step S3: Establishing a BP neural network prediction model for concrete strength CCS, slump S, yield stress YS and plastic viscosity PV;

[0008] Step S4: Optimizing the network structure, weights and thresholds of the neural network using the particle swarm algorithm PSO;

[0009] Step S5: Import the training set into the optimized PSO-BP neural network for training, and input the concrete mix ratio in the prediction set into the trained network to obtain the prediction results of concrete strength, slump, yield stress, and plastic viscosity, and compare them with the corresponding results of the prediction set.

[0010] Preferably, in step S2,

[0011] The original mix sample information of the concrete dataset is recorded as:

[0012]

[0013] In the formula, y 1 is the amount of water used in the concrete mix, y i1 is the water consumption in the concrete mix of group i; 2 is the water-cement ratio in the concrete mix, y i2 is the water-cement ratio in the i-th group of concrete mix; y 3 is the sand ratio in the concrete mix, y i3 is the sand ratio in the i-th group of concrete mix; y 4 is the amount of water reducing agent in the concrete mix, y i4 is the water reducing agent dosage in the i-th group of concrete mix, 1 <i<n;

[0014] The sample information of the concrete mix after standardization is recorded as:

[0015]

[0016] In the formula, x 1 is the amount of water in the standardized concrete mix, x i1 is the water content in the concrete mix of group i after standardized treatment; x2 is the water-cement ratio in the standardized concrete mix, x i2 is the water-cement ratio of the concrete mix after standardization in the i-th group; x 3 is the sand ratio in the standardized concrete mix, x i3 is the sand ratio in the concrete mix of the i-th group after standardized treatment; x 4 is the water reducing agent dosage in the standardized concrete mix, x i4 is the water reducing agent dosage in the concrete mix of the i-th group after standardized treatment;

[0017] The sample information of concrete mix after principal component analysis is recorded as:

[0018]

[0019] In the formula, a ij are the coefficients of the matrix F, k=1,2,…,p, so that F m , F n Mutually unrelated, m≠n, n=1,2,…,p, F 1 Variance > F 2 Variance > F 3 The variance of , and so on.

[0020] Preferably, step S3 comprises the following steps:

[0021] Step S31, establishing a BP neural network that determines the network structure, transfer function and learning rules;

[0022] Step S32, substituting the data after principal component analysis as network input value.

[0023] Preferably, in step S31,

[0024] The BP neural network structure is three layers, including input layer, hidden layer, and output layer. The input layer includes F 1 、F 2 , the output layer includes concrete strength, slump, yield stress, and plastic viscosity, and the number of neurons in the hidden layer is determined according to the following formula:

[0025]

[0026] Where n h is the number of hidden layer nodes, n in is the number of nodes in the input layer, n ois the number of output layer nodes, m is a regulation constant between 1 and 10, the hidden layer neuron transfer function uses log-sigmod function, the output layer neuron transfer function uses tan-sigmod function, and the Levenberg-Marquardt algorithm is used to optimize the BP neural network.

[0027] Preferably, step S4 comprises the following steps:

[0028] Step S41, select the parameters of the PSO algorithm and initialize the population;

[0029] Step S42, determining the fitness function Fitness to guide population search and calculating the particle fitness value;

[0030] Step S43, compare the current fitness of the individual particle with that before iteration. If the current fitness of the particle is better than the fitness before iteration, perform individual extreme value p b Update of the global extreme value p gb is the individual extreme value with the best fitness, and the global extreme value is the current optimal solution of the particle swarm;

[0031] Step S44, updating the speed and position of the particle;

[0032] Step S45, judging whether the stop condition is met, if the current number of iterations has reached the set upper limit, or the error has reached the set minimum error, then the iteration is stopped and the next step is performed, otherwise, the process goes to step S42;

[0033] Step S46: output the p corresponding to the current fitness value b As the global optimal solution, the obtained weights and thresholds are the optimal initial weights;

[0034] Step S47, substitute the optimal initial weights and thresholds searched by PSO into the BP neural network, and use the training samples to train the PSO-optimized neural network until the error requirements are met, thereby completing the construction of the concrete strength, slump, yield stress, and plastic viscosity prediction model based on the PSO-BP neural network.

[0035] Preferably, in step S41, the speed and position of each particle in the population are initialized, and the population P, fitness function Fitness, population size N, number of iterations I and maximum allowed speed v are set. max ;

[0036] Introducing inertia weight ω and learning factor c 1 and c 2 , the inertia weight adopts the linear decreasing method, setting c 1 =c 2 =2.

[0037] Preferably, in step S44, the velocity and position of the particle are updated by the following two formulas:

[0038] v i+1 (t+1)=ωv i (t)+c 1 r 1 (p bi (t)-x i (t))+c 2 r 2 (p gb (t)-x i (t));

[0039] x i+1 (t+1)=x i (t)+v i+1 (t+1);

[0040] In the formula, v i is the velocity vector, x i is the position vector, ω is the inertia weight, c 1 and c 2 is the learning factor, r 1 and r 2 are two random constants between 0 and 1, p bi is the historical optimal position of the particle, p gb is the contemporary optimal position of the population, and t is the current iteration number.

[0041] Therefore, the present invention adopts the above-mentioned concrete performance prediction method based on principal component and PSO-BP network, which has the following beneficial effects:

[0042] (1) The method provided by the present invention overcomes the problems of large deviation and low efficiency in predicting concrete properties with traditional empirical formulas, and also overcomes the problems of large prediction errors and poor stability of the existing statistical learning methods and their prediction models;

[0043] (2) The method provided by the present invention eliminates the correlation of input factors through principal component analysis, while reducing the number of input factors and improving the prediction accuracy of concrete performance; introduces a particle swarm optimization algorithm to optimize the initial weight threshold of the neural network, thereby improving the convergence speed and prediction accuracy of the network;

[0044] (3) The method provided by the present invention can simultaneously meet the prediction requirements of concrete strength, slump, yield stress and plastic viscosity performance, overcoming the difficulty of traditional methods in meeting the requirements of multiple performance predictions, which requires a large number of tests, thereby improving design efficiency and reducing manpower and resource costs;

[0045] (4) The method provided by the present invention solves the problem of multiple variables in the process of concrete performance prediction, can quickly and accurately realize the prediction of multiple concrete properties, and provide technical support for the mix design of green and low-carbon concrete.

[0046] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A schematic diagram of a prediction method flow chart of an embodiment of the present invention;

[0048] Figure 2 A schematic diagram of a concrete data set according to an embodiment of the present invention;

[0049] Figure 3 This is a comparison result diagram of the predicted values ​​and measured values ​​of the yield stress, plastic viscosity, slump and compressive strength of the concrete prediction set according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical scheme and advantages disclosed in the embodiments of the present invention clearer, the embodiments of the present invention are further described in detail in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not intended to limit the embodiments of the present invention. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions.

[0051] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or inherent to these processes, methods, products or devices.

[0052] Like reference numerals and letters denote similar items in the following drawings, and thus, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0053] In the description of the present invention, it should be noted that the terms "upper", "lower", "inside", "outside", etc. indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings, or are directions or positional relationships in which the product of the invention is usually placed when in use. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as a limitation on the present invention.

[0054] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "setting", "installation" and "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0055] Example

[0056] like Figure 1 As shown, the concrete performance prediction method based on principal component and PSO-BP network of the present invention comprises the following steps:

[0057] Step S1: Figure 2 As shown, a concrete data set is established: data is collected, including the mix ratio parameters, yield stress, plastic viscosity, slump, and strength parameters of concrete raw materials. The collected data is preprocessed through data cleaning and normalization to remove abnormal data and eliminate the dimensional influence between the collected data. The normalization method uses the MIN-MAX method:

[0058]

[0059] Among them, x i is the input and output data, x ij is the output data, x min The minimum value of the data change range is 0, x max The maximum value of the data variation range is 1;

[0060] Step S2: Use the principal component analysis method to process the concrete mix data, eliminate the correlation of input factors, reduce the number of input factors, and divide the data set processed by the principal component into a training set and a prediction set.

[0061] The original mix sample information of the concrete dataset is recorded as:

[0062]

[0063] In the formula, y1 is the water consumption (kg) in the concrete mix ratio, y i1 is the water consumption (kg) in the i-th (1 < i < n) concrete mix ratio; y 2 is the water-cement ratio in the concrete mix ratio, y i2 is the water-cement ratio in the i-th concrete mix ratio; y 3 is the sand ratio in the concrete mix ratio, y i3 is the sand ratio in the i-th concrete mix ratio; y 4 is the admixture dosage of water reducer in the concrete mix ratio, y i4 is the admixture dosage of water reducer in the i-th concrete mix ratio.

[0064] The sample information of the concrete mix ratio after standardization is denoted as:

[0065]

[0066] In the formula, x 1 is the water consumption (kg) in the concrete mix ratio after standardization, x i1 is the water consumption (kg) in the i-th (1 < i < n) concrete mix ratio after standardization; x 2 is the water-cement ratio in the concrete mix ratio after standardization, x i2 is the water-cement ratio in the i-th concrete mix ratio after standardization; x 3 is the sand ratio in the concrete mix ratio after standardization, x i3 is the sand ratio in the i-th concrete mix ratio after standardization; x 4 is the admixture dosage of water reducer in the concrete mix ratio after standardization, x i4 is the admixture dosage of water reducer in the i-th concrete mix ratio after standardization.

[0067] The sample information of the concrete mix ratio after principal component analysis is denoted as:

[0068]

[0069] In the formula, a ij is the coefficient of matrix F, (k = 1, 2, …, p), such that F m , F n are uncorrelated (m ≠ n, n = 1, 2, …, p), F 1 has a variance > F 2 has a variance > F 3 has a variance, and so on.

[0070] Step S3: Establish a BP neural network prediction model for concrete strength CCS, slump S, yield stress YS and plastic viscosity PV. The specific steps are as follows:

[0071] Step S31, establishing a BP neural network with a determined network structure, transfer function and learning rules.

[0072] The BP neural network structure is three layers, including input layer, hidden layer, and output layer. The input layer includes F 1 、F 2 , the output layer includes concrete strength, slump, yield stress, and plastic viscosity, and the number of neurons in the hidden layer is determined according to the following formula:

[0073]

[0074] Where n h is the number of hidden layer nodes, n in is the number of nodes in the input layer, n o is the number of output layer nodes, m is a regulation constant between 1 and 10, and the hidden layer neuron transfer function is Function, the output layer neuron transfer function adopts Function, using Levenberg-Marquardt (trainlm) algorithm to optimize BP neural network, training times 1000, learning rate 0.1, training error 0.0001.

[0075] Step S32, the data after principal component analysis is substituted as the network input value, such as F 1 、F 2 .

[0076] Step S4: Use the particle swarm algorithm PSO to optimize the network structure, weights and thresholds of the neural network. The specific steps are as follows:

[0077] Step S41, select the parameters of the PSO algorithm and initialize the population.

[0078] Initialize the speed and position of each particle in the population, and set the population P, fitness function Fitness, population size N = 20, number of iterations I = 100, maximum allowed speed v max =0.0151;

[0079] Introducing inertia weight ω and learning factor c 1 and c 2 , the inertia weight adopts the linear decreasing method, setting c 1 =c 2 =2, the algorithm target error is 0.001.

[0080] Step S42, determining the fitness function Fitness to guide the population search and calculating the particle fitness value.

[0081] set up Where M is the total number of learning samples, N is the number of output nodes, and a ij is the actual output value of the parameter, b ij is the expected output value of the corresponding parameter.

[0082] Step S43, compare the current fitness of the individual particle with that before iteration. If the current fitness of the particle is better than the fitness before iteration, perform individual extreme value p b Update of the global extreme value p gb It is the individual extreme value with the best fitness among all individual extreme values, and the global extreme value is the current optimal solution of the particle swarm.

[0083] Step S44, update the velocity and position of the particle using the following two formulas:

[0084] vi +1 (t+1)=ωv i (t)+c 1 r 1 (p bi (t)-x i (t))+c 2 r 2 (p gb (t)-x i (t));

[0085] x i+1 (t+1)=x i (t)+v i+1 (t+1);

[0086] In the formula, v i is the velocity vector, x i is the position vector, ω is the inertia weight, c 1 and c 2 is the learning factor, r 1 and r 2 are two random constants (usually between 0 and 1), p bi is the historical optimal position of the particle, p gb is the contemporary optimal position of the population, and t is the current iteration number.

[0087] Step S45, determine whether the stop condition is met. If the current number of iterations has reached the set upper limit, or the error has reached the set minimum error, stop the iteration and proceed to the next step. Otherwise, go to step S42.

[0088] Step S46: output the p corresponding to the current fitness value bAs the global optimal solution, the weights and thresholds obtained at this time are the optimal initial weights.

[0089] Step S47, substitute the optimal initial weights and thresholds searched by PSO into the BP neural network, and use the training samples to train the PSO-optimized neural network until the error requirements are met, thereby completing the construction of the concrete strength, slump, yield stress, and plastic viscosity prediction model based on the PSO-BP neural network.

[0090] Step S5: The training set is imported into the optimized PSO-BP neural network for training, and the concrete mix ratio in the prediction set is input into the trained network to obtain the prediction results of concrete strength, slump, yield stress, and plastic viscosity. The results are compared with the corresponding results of the prediction set. The RMSE are 6.10, 11.68, 4.71, and 1.84, respectively, which are in line with the expected results. The comparison results are shown in the figure below. Figure 3 shown.

[0091] Therefore, the present invention adopts the above-mentioned concrete performance prediction method based on principal components and PSO-BP network, which overcomes the problems of large deviation and low efficiency in concrete performance prediction by traditional empirical formulas, and also overcomes the problems of large prediction error and poor stability of the existing statistical learning method and its prediction model.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A concrete performance prediction method based on principal component and PSO-BP network, characterized by: The following steps are involved: Step S1: Establishing a concrete data set: Collecting data, including mix ratio parameters, yield stress, plastic viscosity, slump, and strength parameters of concrete raw materials, pre-processing the collected data through data cleaning and normalization, and removing abnormal data; Step S2: using principal component analysis to process the concrete mix data, eliminating the correlation of input factors, reducing the number of input factors, and dividing the data set processed by the principal component into a training set and a prediction set; Step S3: Establishing a BP neural network prediction model for concrete strength CCS, slump S, yield stress YS and plastic viscosity PV; Step S4: Optimizing the network structure, weights and thresholds of the neural network using the particle swarm algorithm PSO; Step S5: Import the training set into the optimized PSO-BP neural network for training, and input the concrete mix ratio in the prediction set into the trained network to obtain the prediction results of concrete strength, slump, yield stress, and plastic viscosity, and compare them with the corresponding results of the prediction set.

2. The method for predicting concrete properties based on principal component and PSO-BP network according to claim 1 is characterized in that: In step S2, The original mix sample information of the concrete dataset is recorded as: In the formula, y1 is the amount of water used in the concrete mix, y i1 is the water content in the concrete mix of group i; y2 is the water-cement ratio in the concrete mix, y i2 is the water-cement ratio in the concrete mix of group i; y3 is the sand ratio in the concrete mix, y i3 is the sand ratio in the concrete mix of group i; y4 is the water reducing agent dosage in the concrete mix, y i4 is the water reducing agent dosage in the i-th group of concrete mix, 1 <i<n; The sample information of the concrete mix after standardization is recorded as: In the formula, x1 is the water content in the concrete mix after standardization, x i1 is the water content in the standardized concrete mix of the i-th group; x2 is the water-cement ratio in the standardized concrete mix, x i2 is the water-cement ratio of the concrete mix after standardization in the i-th group; x3 is the sand ratio of the concrete mix after standardization, x i3 is the sand ratio in the concrete mix after standardization of the i-th group; x4 is the water reducing agent dosage in the concrete mix after standardization, x i4 is the water reducing agent dosage in the concrete mix of the i-th group after standardized treatment; The sample information of concrete mix after principal component analysis is recorded as: In the formula, a ij are the coefficients of the matrix F, Make F m , F n Uncorrelated, m≠n, n=1,2,…,p, the variance of F1>the variance of F2>the variance of F3, and so on.

3. The method for predicting concrete properties based on principal component and PSO-BP network according to claim 2 is characterized in that: Step S3 includes the following steps: Step S31, establishing a BP neural network that determines the network structure, transfer function and learning rules; Step S32, substituting the data after principal component analysis as network input value.

4. The method for predicting concrete properties based on principal component and PSO-BP network according to claim 3 is characterized in that: In step S31, The BP neural network structure consists of three layers, including input layer, hidden layer and output layer. The input layer includes F1 and F2, and the output layer includes concrete strength, slump, yield stress and plastic viscosity. The number of neurons in the hidden layer is determined according to the following formula: Where n h is the number of hidden layer nodes, n in is the number of nodes in the input layer, n o is the number of output layer nodes, m is a regulation constant between 1 and 10, the hidden layer neuron transfer function uses log-sigmod function, the output layer neuron transfer function uses tan-sigmod function, and the Levenberg-Marquardt algorithm is used to optimize the BP neural network.

5. The method for predicting concrete properties based on principal component and PSO-BP network according to claim 4 is characterized in that: Step S4 includes the following steps: Step S41, select the parameters of the PSO algorithm and initialize the population; Step S42, determining the fitness function Fitness to guide population search and calculating the particle fitness value; Step S43, compare the current fitness of the individual particle with that before iteration. If the current fitness of the particle is better than the fitness before iteration, update the individual extreme value pb. The global extreme value pgb is the individual extreme value with the best fitness, and the global extreme value is the current optimal solution of the particle swarm. Step S44, updating the speed and position of the particle; Step S45, judging whether the stop condition is met, if the current number of iterations has reached the set upper limit, or the error has reached the set minimum error, then the iteration is stopped and the next step is performed, otherwise, the process goes to step S42; Step S46, outputting pb corresponding to the current fitness value as the global optimal solution, and the obtained weights and thresholds are the optimal initial weights; Step S47, substitute the optimal initial weights and thresholds searched by PSO into the BP neural network, and use the training samples to train the PSO-optimized neural network until the error requirements are met, thereby completing the construction of the concrete strength, slump, yield stress, and plastic viscosity prediction model based on the PSO-BP neural network.

6. The method for predicting concrete properties based on principal component and PSO-BP network according to claim 5 is characterized in that: In step S41, the speed and position of each particle in the population are initialized, and the population P, fitness function Fitness, population size N, number of iterations I and maximum allowed speed v are set. max ; Inertia weight ω and learning factors c1 and c2 are introduced. The inertia weight adopts a linear decreasing method and is set to c1=c2=2.

7. The method for predicting concrete properties based on principal component and PSO-BP network according to claim 6 is characterized in that: In step S44, the velocity and position of the particle are updated by the following two formulas: vi +1 (t+1)=ωvi(t)+c1r1(pbi(t)-xi(t))+c2r2(pgb(t)-xi(t)); x i+1 (t+1)=x i (t)+v i+1 (t+1); In the formula, v i is the velocity vector, x i is the position vector, ω is the inertia weight, c1 and c2 are learning factors, r1 and r2 are two random constants between 0 and 1, and p bi is the historical optimal position of the particle, p gb is the contemporary optimal position of the population, and t is the current iteration number.

Images