Mass concrete hydration heat internal and external temperature difference and temperature strain prediction method

Abstract

The invention relates to a mass concrete hydration heat internal and external temperature difference and temperature strain prediction method, which is based on a hydration reaction of a concrete cementing material, through observation records of parameters such as concrete mold entering temperature, environment temperature and center temperature, a function relationship between the parameters and a pouring area temperature is established, and through a function model, an overall temperature field is predicted. Forming an internal and external temperature difference prediction model, and outputting an internal and external temperature difference; and in combination with concrete strain records, establishing a relation function of temperature and strain, forming a concrete hydration heat temperature-strain prediction model, and outputting a strain prediction result. The problem of hydration heat of mass concrete pouring construction puzzling major engineering can be solved. The problems of internal damage and surface cracking of the concrete due to overlarge internal and external temperature difference and temperature strain exceeding the initial tensile strain of the concrete caused by low concrete strength, poor bonding force of gravel aggregate, centralized hydration of a large amount of cement, and easy generation of adiabatic temperature rise in the initial stage of pouring are solved.

Description

Technical Field

[0001] The invention belongs to the technical field of model establishment prediction, and in particular relates to a method for predicting the internal and external temperature difference and temperature strain of hydration heat of large-volume concrete. Background Art

[0002] The characteristics of large-volume concrete are large cement dosage, concentrated hydration reaction, and large heat release of cement hydration. Aggregates such as gravel and sand in concrete are poorly heat-dissipated under the wrapping of cement and water, and are prone to adiabatic temperature rise, resulting in excessive temperature difference between the inside and outside of the concrete, thus causing temperature cracks. Large-volume concrete usually refers to concrete structures with a minimum geometric cross-section size of not less than 1m. The internal and external temperature difference usually refers to the difference between the geometric center temperature of the concrete and the external surface temperature. The center temperature refers to the temperature at the centroid position according to the concrete casting shape. Relevant studies at home and abroad have clearly shown that the internal hydration heat of concrete and the external temperature restrict the material's mechanical properties and durability, and the temperature effect of large-volume concrete is more obvious. In particular, in the construction of large-volume concrete, controlling the temperature difference between the inside and outside of the concrete, the ambient temperature, etc., plays a key role in the casting quality and structural performance of the concrete. In order to prevent the temperature crack problem of large-volume concrete hydration heat, my country's standard "Technical Specification for Temperature Measurement and Control of Large-Volume Concrete GB / T51028-2015" gives specific internal and external temperature difference limits, which stipulates: when the concrete pouring thickness is less than 1.5m, the internal and external temperature difference shall not exceed 20℃; when the pouring thickness is in the range of 1.5m-2.5m, the internal and external temperature difference shall not exceed 25℃; when the pouring thickness is greater than 2.5m, the internal and external temperature difference shall not exceed 28℃. In order to effectively control the hydration heat temperature of large-volume concrete, it is very important to obtain information on parameters such as the internal and external temperature difference and strain of concrete in a timely and accurate manner. It is necessary to monitor the temperature and strain of large-volume concrete, and understand the changes in hydration heat temperature and strain by collecting and analyzing temperature data.

[0003] However, due to the uncertainty of the boundary conditions of large-volume concrete structures, especially parameters such as ambient temperature and mold entry temperature, the calculation results of hydration heat analysis will be significantly affected, which may cause the prediction results to deviate seriously from the actual observed data. Although there are some technical means to monitor the hydration heat process of large-volume concrete and obtain its temperature field data, due to the deviation of the model itself, these data are often difficult to directly use to predict the hydration heat temperature field, damage and future cracks of concrete structures.

[0004] In order to solve this problem, it is particularly important to correct the model parameters used in the hydration heat analysis of large-volume concrete structures. A feasible method is to combine the measured temperature field data, perform inversion analysis on the measured data, and build a temperature gradient convolutional neural network by continuously adjusting the model parameters to deduce the internal and external temperature difference at the next moment. In this way, not only can the accuracy of model prediction be improved, but also a more reliable basis can be provided for the construction control and crack prevention of concrete structures. Summary of the invention

[0005] Purpose of the Invention

[0006] In order to solve the problem that hydration heat damage and cracks are difficult to predict during the existing mass concrete pouring process, the present invention provides a method for predicting the internal and external temperature difference and temperature strain of hydration heat of mass concrete.

[0007] Technical Solution

[0008] A method for predicting the internal and external temperature difference and temperature strain of hydration heat of large-volume concrete comprises the following steps:

[0009] S1: Based on the records of the hydration heat temperature measurement points of large-volume concrete, the pouring start time is taken as the starting point to record the hydration heat temperature value of the concrete, and a data training sample set of hydration heat temperature is established according to the time series; nonlinear dynamic system modeling is performed to obtain the prediction results of the internal and external temperatures at the i-th moment;

[0010] Construct a posterior probability density function based on the samples of the internal and external temperature difference of the hydration heat; extract N actual sampling data from the sampling density function according to the weighted data; calculate the weight of the data and normalize the weight, perform resampling steps based on the filtering algorithm, fit the functional relationship between the mold temperature, ambient temperature and center temperature and the hydration heat temperature, and then predict the hydration heat temperature in real time through data filtering, so as to make a real-time prediction of the overall temperature field of the hydration heat of large-volume concrete;

[0011] S2: The support vector regression algorithm is used to fit the relationship between the hydration heat of concrete and its corresponding hydration time-concrete temperature field. The relationship between the hydration heat of concrete and its corresponding internal and external temperature difference is fitted by combining the mold entry temperature, ambient temperature and center temperature.

[0012] For the temperature prediction area (for large volumes, it is divided into different prediction areas), the extended Kalman filter is used to estimate the temperature difference between the inside and outside of large concrete volumes in real time;

[0013] S3: A real-time prediction model for concrete strain under the action of hydration heat of large-volume concrete is established based on parameters that affect hydration heat, such as cement type and retarder dosage. The convolutional neural network algorithm is used to predict concrete hydration heat strain. The neural network model is constructed using observation data to fit the functional relationship between concrete temperature strain and cement type and retarder dosage parameters. The weights of the convolutional neural network are fitted using the historical observation data of the temperature field of large-volume concrete hydration heat and concrete strain. After training the convolutional neural network model, the concrete temperature strain is trained, and the concrete hydration heat temperature strain can be predicted in real time at future moments.

[0014] S4: Based on the monitoring records of concrete hydration heat temperature strain and the prediction results of the overall hydration heat temperature field and the internal and external temperature difference, a real-time prediction model of concrete hydration heat strain based on the long short-term memory network (LSTM) under the action of hydration heat is established, and the prediction results of hydration heat temperature strain in the entire prediction area are updated in real time; a neural network model is constructed using monitoring data to fit the complex functional relationship between the temperature field and the concrete temperature strain, and the accuracy is compared according to different regional ranges of concrete. The neural network model trained according to the center of the hydration heat temperature field is used to make real-time predictions of the concrete temperature strain under the action of the hydration heat temperature field;

[0015] S5: According to the distribution range of the hydration heat temperature field, the concrete temperature strain prediction area is divided into three-dimensional grids, and each strain variable is counted in real time. Thiessen polygons are constructed in the statistical area for assignment, and the array of strain monitoring point position coordinates is triangulated to generate the triangulation network corresponding to the variable distribution prediction results, and the regional distribution is statistically analyzed; the Gaussian smoothing function is used to smooth the prediction results to increase the continuity of strain distribution between prediction areas.

[0016] Furthermore, the S1 specifically includes the following steps:

[0017] S1-1: Based on the measurement point records of the hydration heat temperature of large-volume concrete, take the pouring start time as the starting point, record the hydration heat temperature value of concrete, and establish a data training sample set of hydration heat temperature according to the time series Z(k) represents the temperature record from the kth measuring point 1 to time t; Represents the internal and external temperature observation results of the kth measuring point at the i-th moment; performs nonlinear dynamic system modeling where f k represents the state transition function, represents the predicted results of internal and external temperatures at the kth measuring point at the i-th moment, represents the predicted result of internal and external temperature at the k measuring point i-1, let where h k represents the measurement function, Represents the internal and external temperature observation results at the k measuring point i-1 time;

[0018] S1-2: Construct a posterior probability density function based on the internal and external temperature difference samples of hydration heat in It represents the prediction result sequence of hydration heat parameters generated from k measuring point 1 to time t. represents the sequence of hydration heat parameter observation results generated from k measuring point 1 to time t; extract N sample data, i = 1, 2, ..., N, the hydration heat parameter prediction result sequence generated by the i-th data from k measuring point 1 to time t is expressed as Indicates that the weight of the i-th data is expressed as Represents, and the probability weight According to the weighted data, from the sampling density function N actual sampling data are extracted from the , m means the m-order data filtering algorithm is used, Where q(.) represents the sampling density function; x represents the predicted result of the internal and external temperature difference of hydration heat, z represents the observed result of the internal and external temperature difference of hydration heat, t represents the t time, m represents the use of the m-order filtering algorithm, k represents the k-th measuring point, and i represents the i-th sample data; according to Recursion, p(·) represents the posterior probability density function of the hydration heat temperature parameter sample, k represents the kth measuring point, i represents the i-th sample data, the data weight is calculated and the weight is normalized, based on the filtering algorithm, the resampling step is performed, the functional relationship between the mold temperature, ambient temperature and center temperature and the hydration heat temperature is fitted, and then the hydration heat temperature is predicted in real time through data filtering, and then the overall temperature field of the hydration heat of mass concrete is predicted in real time.

[0019] Furthermore, the S2 specifically includes the following steps:

[0020] S2-1: Train the hydration heat-internal and external temperature difference model function in the hydration heat temperature and internal and external temperature difference sample space, the decision surface width b and the decision surface normal vector Δ, determine the position and direction of the decision surface, t = 0, 1, 2, ..., t n , introduce the slack variable ξ 1：t and ⅴ 1:t , represents the slack variable up to the t-th time node, under the following constraints μ 1:t ∈{-1,1}, slack variable ξ 1:t ≥0,υ 1:t ≥0,t=0,1,2,…,t n , t represents the tth time point, t n Represents the total number of observation time nodes, loss function: L 1:t represents the loss function from 1 to t time points, where c is a constant > 0, ξ 1:t and ⅴ 1:t is a relaxation variable that satisfies the optimization objective loss function and the internal and external temperature difference prediction model function based on hydration heat Where 1:t represents the time series from the 1st to the tth time point, x represents the prediction result of the internal and external temperature difference of hydration heat, K represents the temperature observation point number, and b is the width of the decision surface, which can predict the internal and external temperature difference at future moments;

[0021] S2-2: The prediction result of the temperature difference between the inside and outside under the action of hydration heat in step S2-1 is simulated within the prediction area. Combined with the actual observation record of the temperature difference between the inside and outside, the prediction area of ​​the large volume concrete is divided into n units. Predict the temperature distribution of the area before the next time point Same as the actual temperature observation record u at time t t Gradually predict the distribution of temperature prediction area under the action of hydration heat temperature field at the current moment Q represents the distribution of temperature in the prediction area, t-1 represents the time node before time t, n represents that the domain is divided into n units, the length of the actual temperature observation data is used as a vector, which represents the number of prediction results of the inside and outside temperature difference, and E represents the calculation process of the temperature distribution of the area at time t under the action of hydration heat:

[0022] in, Indicates the temperature difference distribution inside and outside at the current moment, It represents the temperature difference distribution of the inside and outside at the time node before time t at the hydration heat temperature, Δq t The observed value variable representing the temperature difference between inside and outside, It represents the temperature difference distribution between the inside and outside of the current moment and the next time point, and κ represents the correction value predicted at a time point by the extended Kalman filter.

[0023] Furthermore, S3 specifically includes the following steps:

[0024] S3-1: According to the coordinates and distribution of the strain observation points, the prediction area of ​​mass concrete is delineated, and a corresponding strain observation point position coordinate is constructed, and a two-dimensional array V of mass concrete hydration heat temperature strain data is constructed. M ={(x 1 ,y 1 ),(x 2 ,y 2 ),…,(x M-1 ,y M-1 ),(x M ,yM )},(x M ,y M ) indicates V M The real-time monitoring data of the strain variable corresponding to the Mth strain observation point in the prediction area is obtained, and the original data is normalized to eliminate the dimension for the subsequent neural network training data and optimize the loss function of the training model;

[0025] S3-2: Divide the data set normalized in step S3-1 into a training set and a test set, design a neural network model using the strain monitoring data of each observation point before the current time t, and introduce a convolutional neural network, in which the hidden layer unit can be designed to include a convolutional layer, an excitation layer, and a neural network. 6 The excitation layer, slicing layer and fusion layer operations are used to fit the weights of the convolutional neural network using the historical observation data of the hydration heat temperature field and strain. After training the convolutional neural network model, the hydration heat temperature strain variable is trained;

[0026] S3-3: Test in the actual hydration heat temperature strain prediction case, and construct the temperature strain prediction model function f(t,x',y',d,z), where t represents the time, x' and y' represent the coordinates of the hydration heat temperature strain observation point, and d represents the strain. According to the temperature strain prediction function, the strain in the prediction area at future moments can be predicted in real time.

[0027] Furthermore, S4 specifically includes the following steps:

[0028] S4-1: Multiple time windows are used to introduce the long short-term memory network (LSTM) for time series prediction. The hydration heat temperature strain and hydration heat temperature data information are used through the forget gate Ft, input gate It and output gate Ot. The forget gate Ft controls the state of the memory unit Ct-1 at the previous moment and determines the strain information that needs to be forgotten at time t. The input gate It controls the candidate state ct at the current moment and determines the strain information that needs to be stored. The output gate Ot controls the memory unit Ct at the current moment and determines the state Ht that the hydration heat temperature and strain information need to output to the outside. The formula is as follows:

[0029]

[0030] in,

[0031] C t =F t *C t-1 +I t *c t

[0032] H t =O t *tanh(ct ),

[0033] σ represents the sigmoid function, tanh represents the hyperbolic tangent function, W represents the weights corresponding to the forget gate, input gate, and output gate, β represents the bias values ​​corresponding to the forget gate, input gate, and output gate, and H t represents the hidden state at time t, X t Represents the hydration heat temperature strain and hydration heat temperature data input at time t;

[0034] S4-2: Establish a neural network for the temporal dependency between strain monitoring and hydration heat temperature, use the long short-term memory network to predict the strain data test set in the prediction area, and compare the accuracy of the prediction results with the strain in the actual prediction area in step S3 according to different regional ranges. Define the application scope of the neural network model trained in step S4-1 according to the distance between the temperature field core and the measuring point.

[0035] Furthermore, S5 specifically includes the following steps:

[0036] S5-1: Divide the prediction area into grids according to the hydration temperature strain of the prediction area, and create a two-dimensional array of grid point area centers with an N×N area V centered on the horizontal and vertical coordinates N ={(x 1 ,y 1 ),(x 2 ,y 2 ),…,(x N-1 ,y N-1 ),(x N ,y N )}, the array V is processed by the identity matrix function N Initialize the data information contained in the grid, where L and H are the spacing between adjacent grid points in the horizontal and vertical directions, respectively, and x N =L×N,y N =H×N; construct Thiessen polygons with adjacent observation points as perpendicular medians for each hydration heat temperature strain variable monitoring point in the prediction area, and divide it into R Thiessen polygons with the strain variable monitoring point coordinates as the center, where R is a positive integer, and the array V is calculated according to the strain prediction result in step S4. R Assign values ​​and make preliminary assignments to the distribution of the predicted results of the hydration heat temperature dependence variable of the entire area;

[0037] S5-2: Triangulate the array containing the position coordinates of the concrete temperature strain monitoring points to generate a triangulated network, simulate the temperature strain distribution prediction result of step S5-1, and perform point-by-point correction on the strain prediction result corresponding to the point set generated by the triangulated network to simulate the numerical prediction result of the temperature strain distribution in the prediction area;

[0038] S5-3: Use a Gaussian smoothing function to smooth the prediction result of step S5-2, and construct a Gaussian kernel function for the two-dimensional array of the center of the grid point area with horizontal and vertical coordinates σ0 is the standard deviation of the Gaussian distribution in the Gaussian kernel generator, which characterizes the discreteness of the prediction result data; thereby optimizing and updating the distribution of the strain prediction results generated in S5-2 in the prediction area, and increasing the continuity between the prediction results of each temperature strain variable distributed in the prediction area.

[0039] Advantages and effects

[0040] The present invention utilizes the characteristics of hydration heat of large-volume concrete, combined with machine learning methods, to perform real-time prediction of the internal and external temperature difference and temperature strain of hydration heat of large-volume concrete. This method only uses the ambient temperature and mold temperature hydration heat parameters as initial sample assignments, and does not consider the concrete grade, water-cement ratio and other construction parameters. It utilizes the characteristics of hydration heat and machine learning methods to iterate and update data without considering too many complicated parameters, which brings more convenience to construction technicians. Utilizing hydration heat observation data, by constructing a deep neural network algorithm, multiple models of hydration heat temperature field, internal and external temperature difference change distribution, and hydration heat temperature strain distribution are simultaneously predicted and updated in real time in the prediction domain. In terms of the training strategy of the deep neural network, unlike the use of a single neural network model, multiple neural network training strategies are integrated. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present invention is further described below in conjunction with the accompanying drawings and specific implementation methods. The protection scope of the present invention is not limited to the following description.

[0042] Figure 1 Schematic diagram of the layout of temperature and strain sensors (taking the circular and rectangular shapes commonly used in large-volume concrete construction as examples). DETAILED DESCRIPTION

[0043] A method for predicting the internal and external temperature difference and temperature strain of hydration heat of large-volume concrete comprises the following steps:

[0044] S1: Based on the records of the hydration heat temperature measurement points of large-volume concrete, the pouring start time is taken as the starting point to record the hydration heat temperature value of the concrete, and a data training sample set of hydration heat temperature is established according to the time series; nonlinear dynamic system modeling is performed to obtain the prediction results of the internal and external temperatures at the i-th moment;

[0045] Construct a posterior probability density function based on the samples of the internal and external temperature difference of the hydration heat; extract N actual sampling data from the sampling density function according to the weighted data; calculate the weight of the data and normalize the weight, perform resampling steps based on the filtering algorithm, fit the functional relationship between the mold temperature, ambient temperature and center temperature and the hydration heat temperature, and then predict the hydration heat temperature in real time through data filtering, so as to make a real-time prediction of the overall temperature field of the hydration heat of large-volume concrete;

[0046] S2: The support vector regression algorithm is used to fit the relationship between the hydration heat of concrete and its corresponding hydration time-concrete temperature field. The relationship between the hydration heat of concrete and its corresponding internal and external temperature difference is fitted by combining the mold entry temperature, ambient temperature and center temperature.

[0047] For the temperature prediction area (for large volumes, it is divided into different prediction areas), the Extended Kalman Filter (EKF) is used to estimate the temperature difference between the inside and outside of large concrete volumes in real time.

[0048] S3: A real-time prediction model for concrete strain under the action of hydration heat of large-volume concrete is established based on parameters that affect hydration heat, such as cement type and retarder dosage. The convolutional neural network algorithm is used to predict concrete hydration heat strain. The neural network model is constructed using observation data to fit the functional relationship between concrete temperature strain and cement type and retarder dosage parameters. The weights of the convolutional neural network are fitted using the historical observation data of the temperature field of large-volume concrete hydration heat and concrete strain. After training the convolutional neural network model, the concrete temperature strain is trained, and the concrete hydration heat temperature strain can be predicted in real time at future moments.

[0049] S4: Based on the monitoring records of concrete hydration heat temperature strain and the prediction results of the overall hydration heat temperature field and the internal and external temperature difference, a real-time prediction model of concrete hydration heat strain based on the long short-term memory network (LSTM) under the action of hydration heat is established, and the prediction results of hydration heat temperature strain in the entire prediction area are updated in real time; a neural network model is constructed using monitoring data to fit the complex functional relationship between the temperature field and the concrete temperature strain, and the accuracy is compared according to different regional ranges of concrete. The neural network model trained according to the center of the hydration heat temperature field is used to make real-time predictions of the concrete temperature strain under the action of the hydration heat temperature field;

[0050] S5: According to the distribution range of the hydration heat temperature field, the concrete temperature strain prediction area is divided into three-dimensional grids, and each strain variable is counted in real time. Thiessen polygons are constructed in the statistical area for assignment, and the array of strain monitoring point position coordinates is triangulated to generate the triangulation network corresponding to the variable distribution prediction results, and the regional distribution is statistically analyzed; the Gaussian smoothing function is used to smooth the prediction results to increase the continuity of strain distribution between prediction areas.

[0051] S1 specifically includes the following steps:

[0052] S1-1: Based on the measurement point records of the hydration heat temperature of large-volume concrete, take the pouring start time as the starting point, record the hydration heat temperature value of concrete, and establish a data training sample set of hydration heat temperature according to the time series Z(k) represents the temperature record from the kth measuring point 1 to time t; Represents the internal and external temperature observation results of the kth measuring point at the i-th moment; performs nonlinear dynamic system modeling where f k represents the state transition function, represents the predicted results of internal and external temperatures at the kth measuring point at the i-th moment, represents the predicted result of internal and external temperature at the k measuring point i-1, let where h k represents the measurement function, Represents the internal and external temperature observation results at the k measuring point i-1 time;

[0053] S1-2: Construct a posterior probability density function based on the internal and external temperature difference samples of hydration heat in represents the prediction result sequence of hydration heat parameters generated from k measurement point 1 to time t, represents the sequence of hydration heat parameter observation results generated from k measuring point 1 to time t; extract N sample data, i = 1, 2, ..., N, the sequence of hydration heat parameter prediction results generated by the i-th data from k measuring point 1 to time t is expressed as Indicates that the weight of the i-th data is expressed as Represents, and the probability weight According to the weighted data, from the sampling density function N actual sampling data are extracted from the , m means the m-order data filtering algorithm is used, Where q(.) represents the sampling density function; x represents the predicted result of the internal and external temperature difference of hydration heat, z represents the observed result of the internal and external temperature difference of hydration heat, t represents the t time, m represents the use of the m-order filtering algorithm, k represents the k-th measuring point, and i represents the i-th sample data; according to Recursion, p(·) represents the posterior probability density function of the hydration heat temperature parameter sample, k represents the kth measuring point, i represents the i-th sample data, the data weight is calculated and the weight is normalized, based on the filtering algorithm, the resampling step is performed, the functional relationship between the mold temperature, ambient temperature and center temperature and the hydration heat temperature is fitted, and then the hydration heat temperature is predicted in real time through data filtering, and then the overall temperature field of the hydration heat of mass concrete is predicted in real time.

[0054] S2 specifically includes the following steps:

[0055] S2-1: Train the hydration heat-internal and external temperature difference model function in the hydration heat temperature and internal and external temperature difference sample space, the decision surface width b and the decision surface normal vector Δ, determine the position and direction of the decision surface, t = 0, 1, 2, ..., t n , introduce the slack variable ξ 1：t and ⅴ 1:t , represents the slack variable up to the t-th time node, under the following constraints μ 1:t ∈{-1,1}, slack variable ξ 1:t ≥0,υ 1:t ≥0,t=0,1,2,…,t n , t represents the tth time point, t n Represents the total number of observation time nodes, loss function: L 1:t represents the loss function from 1 to t time points, where c is a constant > 0, ξ 1:t and ⅴ 1:t is a relaxation variable that satisfies the optimization objective loss function and the internal and external temperature difference prediction model function based on hydration heat Where 1:t represents the time series from the 1st to the tth time point, x represents the prediction result of the internal and external temperature difference of hydration heat, K represents the temperature observation point number, and b is the width of the decision surface, which can predict the internal and external temperature difference at future moments;

[0056] S2-2: The prediction result of the temperature difference between the inside and outside under the action of hydration heat in step S2-1 is simulated within the prediction area. Combined with the actual observation record of the temperature difference between the inside and outside, the prediction area of ​​the large volume concrete is divided into n units. Predict the temperature distribution of the area before the next time point Same as the actual temperature observation record u at time t t Gradually predict the distribution of temperature prediction area under the action of hydration heat temperature field at the current moment Q represents the distribution of temperature in the prediction area, t-1 represents the time node before time t, n represents that the domain is divided into n units, the length of the actual temperature observation data is used as a vector, which represents the number of prediction results of the inside and outside temperature difference, and E represents the calculation process of the temperature distribution of the area at time t under the action of hydration heat:

[0057] in, Indicates the temperature difference distribution inside and outside at the current moment, It represents the temperature difference distribution of the inside and outside at the time node before time t at the hydration heat temperature, Δq t The observed value variable representing the temperature difference between inside and outside, It represents the temperature difference distribution between the inner and outer surfaces before the next time point at the current moment, and κ represents the correction value predicted at a time point by the Extended Kalman Filter (EKF).

[0058] S3 specifically includes the following steps:

[0059] S3-1: According to the coordinates and distribution of the strain observation points, the prediction area of ​​mass concrete is delineated, and a corresponding strain observation point position coordinate is constructed, and a two-dimensional array V of mass concrete hydration heat temperature strain data is constructed. M ={(x 1 ,y 1 ),(x 2 ,y 2 ),…,(x M-1 ,y M-1 ),(x M ,y M )},(x M ,y M ) indicates V M The real-time monitoring data of the strain variable corresponding to the Mth strain observation point in the prediction area is obtained, and the original data is normalized to eliminate the dimension for the subsequent neural network training data and optimize the loss function of the training model;

[0060] S3-2: Divide the data set normalized in step S3-1 into a training set and a test set, design a neural network model using the strain monitoring data of each observation point before the current time t, introduce a convolutional neural network, wherein the hidden layer unit can be designed to include convolutional layer, excitation layer, slicing layer and fusion layer operations, and use the historical observation data of hydration heat temperature field and strain to fit the weight of the convolutional neural network. After training the convolutional neural network model, train the hydration heat temperature strain variable;

[0061] S3-3: Test in the actual hydration heat temperature strain prediction case, and construct the temperature strain prediction model function f(t,x',y',d,z), where t represents the time, x' and y' represent the coordinates of the hydration heat temperature strain observation point, and d represents the strain. According to the temperature strain prediction function, the strain in the prediction area at future moments can be predicted in real time.

[0062] S4 specifically includes the following steps:

[0063] S4-1: Multiple time windows are used to introduce the long short-term memory network (LSTM) for time series prediction. The hydration heat temperature strain and hydration heat temperature data information are used through the forget gate Ft, input gate It and output gate Ot. The forget gate Ft controls the state of the memory unit Ct-1 at the previous moment and determines the strain information that needs to be forgotten at time t. The input gate It controls the candidate state ct at the current moment and determines the strain information that needs to be stored. The output gate Ot controls the memory unit Ct at the current moment and determines the state Ht that the hydration heat temperature and strain information need to output to the outside. The formula is as follows:

[0064]

[0065] in,

[0066] C t =F t *C t-1 +I t *c t

[0067] H t =O t *tanh(c t ),

[0068] σ represents the sigmoid function, tanh represents the hyperbolic tangent function, W represents the weights corresponding to the forget gate, input gate, and output gate, β represents the bias values ​​corresponding to the forget gate, input gate, and output gate, and H t represents the hidden state at time t, X t Represents the hydration heat temperature strain and hydration heat temperature data input at time t;

[0069] S4-2: Establish a neural network for the temporal dependency between strain monitoring and hydration heat temperature, use the long short-term memory network to predict the strain data test set in the prediction area, and compare the accuracy of the prediction results with the strain in the actual prediction area in step S3 according to different regional ranges. Define the application scope of the neural network model trained in step S4-1 according to the distance between the temperature field core and the measuring point.

[0070] S5 specifically includes the following steps:

[0071] S5-1: Divide the prediction area into grids according to the hydration temperature strain of the prediction area, and create a two-dimensional array of grid point area centers with an N×N area V centered on the horizontal and vertical coordinates N ={(x 1 ,y 1 ),(x 2 ,y 2 ),…,(x N-1 ,y N-1 ),(x N ,y N )}, the array V is processed by the identity matrix function N Initialize the data information contained in the grid, where L and H are the spacing between adjacent grid points in the horizontal and vertical directions, respectively, and x N =L×N,y N =H×N; construct Thiessen polygons with adjacent observation points as perpendicular medians for each hydration heat temperature strain variable monitoring point in the prediction area, and divide it into R Thiessen polygons with the strain variable monitoring point coordinates as the center, where R is a positive integer, and the array V is calculated according to the strain prediction result in step S4. R Assign values ​​and make preliminary assignments to the distribution of the predicted results of the hydration heat temperature dependence variable of the entire area;

[0072] S5-2: Triangulate the array containing the coordinates of the concrete temperature strain monitoring points to generate a triangulation network. 15 The distribution prediction results are simulated, and the corresponding strain variable prediction results of the point set generated by the triangulation network are corrected point by point to simulate the numerical prediction results of the temperature strain variable distribution in the prediction area;

[0073] S5-3: Use a Gaussian smoothing function to smooth the prediction result of step S5-2, and construct a Gaussian kernel function for the two-dimensional array of the center of the grid point area with horizontal and vertical coordinates σ0 is the standard deviation of the Gaussian distribution in the Gaussian kernel generator, which characterizes the discreteness of the prediction result data; thereby optimizing and updating the distribution of the strain prediction results generated in S5-2 in the prediction area, and increasing the continuity between the prediction results of each temperature strain variable distributed in the prediction area.

[0074] Obviously, the above-mentioned embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not limitations on the embodiments of the present invention. For ordinary technicians in the relevant field, other different forms of changes or modifications can be made on the basis of the above description. It is impossible to list all the embodiments here. All obvious changes or modifications derived from the technical solution of the present invention are still within the protection scope of the present invention.

Claims

1. A method for predicting the internal and external temperature difference and temperature strain of hydration heat of large-volume concrete, characterized by: The steps include: S1: Based on the measurement point records of the hydration heat temperature of large-volume concrete, the pouring start time is taken as the starting point to record the hydration heat temperature value of concrete, and a data training sample set of hydration heat temperature is established according to the time series; Conduct nonlinear dynamic system modeling to obtain the prediction results of internal and external temperatures at the i-th moment; Construct a posterior probability density function based on the samples of the internal and external temperature difference of the hydration heat; extract N actual sampling data from the sampling density function according to the weighted data; calculate the weight of the data and normalize the weight, perform resampling steps based on the filtering algorithm, fit the functional relationship between the mold temperature, ambient temperature and center temperature and the hydration heat temperature, and then predict the hydration heat temperature in real time through data filtering, so as to make a real-time prediction of the overall temperature field of the hydration heat of large-volume concrete; S2: The support vector regression algorithm is used to fit the relationship between the hydration heat of concrete and its corresponding hydration time-concrete temperature field. The relationship between the hydration heat of concrete and its corresponding internal and external temperature difference is fitted by combining the mold entry temperature, ambient temperature and center temperature. For the temperature prediction area, the extended Kalman filter is used to estimate the temperature difference between the inside and outside of the large volume of concrete in real time; S3: A real-time prediction model for concrete strain under the action of hydration heat of large-volume concrete is established based on parameters that affect hydration heat, such as cement type and retarder dosage. The convolutional neural network algorithm is used to predict concrete hydration heat strain. The neural network model is constructed using observation data to fit the functional relationship between concrete temperature strain and cement type and retarder dosage parameters. The weights of the convolutional neural network are fitted using the historical observation data of the temperature field of large-volume concrete hydration heat and concrete strain. After training the convolutional neural network model, the concrete temperature strain is trained, and the concrete hydration heat temperature strain can be predicted in real time at future moments. S4: Based on the monitoring records of concrete hydration heat temperature strain and the prediction results of the overall hydration heat temperature field and the internal and external temperature difference, a real-time prediction model of concrete hydration heat strain under the action of hydration heat based on the long short-term memory network is established, and the prediction results of hydration heat temperature strain in the entire prediction area are updated in real time; a neural network model is constructed using monitoring data to fit the complex functional relationship between the temperature field and the concrete temperature strain, and the accuracy is compared according to different regional ranges of concrete. The neural network model trained according to the center of the hydration heat temperature field is used to make real-time predictions of concrete temperature strain under the action of the hydration heat temperature field; S5: According to the distribution range of the hydration heat temperature field, the concrete temperature strain prediction area is divided into three-dimensional grids, and each strain variable is counted in real time. Thiessen polygons are constructed in the statistical area for assignment, and the array of strain monitoring point position coordinates is triangulated to generate the triangulation network corresponding to the variable distribution prediction results, and the regional distribution is statistically analyzed; the Gaussian smoothing function is used to smooth the prediction results to increase the continuity of strain distribution between prediction areas.

2. The method for predicting the internal and external temperature difference and temperature strain of hydration heat of large volume concrete according to claim 1 is characterized in that: The S1 specifically includes the following steps: S1-1: Based on the measurement point records of the hydration heat temperature of large-volume concrete, take the pouring start time as the starting point, record the hydration heat temperature value of concrete, and establish a data training sample set of hydration heat temperature according to the time series Z(k) represents the temperature record from the kth measuring point 1 to time t; Represents the internal and external temperature observation results of the kth measuring point at the i-th moment; performs nonlinear dynamic system modeling where f k represents the state transition function, represents the predicted results of internal and external temperatures at the kth measuring point at the i-th moment, represents the predicted result of internal and external temperature at the k measuring point i-1, let where h k represents the measurement function, Represents the internal and external temperature observation results at the k measuring point i-1 time; S1-2: Construct a posterior probability density function based on the internal and external temperature difference samples of hydration heat in represents the prediction result sequence of hydration heat parameters generated from k measurement point 1 to time t, represents the sequence of hydration heat parameter observation results generated from k measuring point 1 to time t; extract N sample data, i = 1, 2, ..., N, the hydration heat parameter prediction result sequence generated by the i-th data from k measuring point 1 to time t is expressed as Indicates that the weight of the i-th data is expressed as Represents, and the probability weight According to the weighted data, from the sampling density function N actual sampling data are extracted from the , m means the m-order data filtering algorithm is used, Where q(.) represents the sampling density function; x represents the predicted result of the internal and external temperature difference of hydration heat, z represents the observed result of the internal and external temperature difference of hydration heat, t represents the t time, m represents the use of the m-order filtering algorithm, k represents the k-th measuring point, and i represents the i-th sample data; according to Recursion, p(·) represents the posterior probability density function of the hydration heat temperature parameter sample, k represents the kth measuring point, i represents the i-th sample data, the data weight is calculated and the weight is normalized, based on the filtering algorithm, the resampling step is performed, the functional relationship between the mold temperature, ambient temperature and center temperature and the hydration heat temperature is fitted, and then the hydration heat temperature is predicted in real time through data filtering, and then the overall temperature field of the hydration heat of mass concrete is predicted in real time.

3. The method for predicting the internal and external temperature difference and temperature strain of hydration heat of large volume concrete according to claim 1 is characterized in that: The S2 specifically includes the following steps: S2-1: Train the hydration heat-internal and external temperature difference model function in the hydration heat temperature and internal and external temperature difference sample space, the decision surface width b and the decision surface normal vector Δ, determine the position and direction of the decision surface, t = 0, 1, 2, ..., t n , introduce the slack variable ξ 1：t and ⅴ 1:t , represents the slack variable to the t-th time node, under the following constraints Slack variable ξ 1:t ≥0,υ 1:t ≥0,t=0,1,2,…,t n , t represents the tth time point, t n Represents the total number of observation time nodes, loss function: L 1:t represents the loss function from 1 to t time points, where c is a constant > 0, ξ 1:t and ⅴ 1:t is a relaxation variable that satisfies the optimization objective loss function and the internal and external temperature difference prediction model function based on hydration heat Where 1:t represents the time series from the 1st to the tth time point, x represents the prediction result of the internal and external temperature difference of hydration heat, K represents the temperature observation point number, and b is the width of the decision surface, which can predict the internal and external temperature difference at future moments; S2-2: The prediction result of the temperature difference between the inside and outside under the action of hydration heat in step S2-1 is simulated within the prediction area. Combined with the actual observation record of the temperature difference between the inside and outside, the prediction area of ​​the large volume concrete is divided into n units. Predict the temperature distribution of the area before the next time point Same as the actual temperature observation record u at time t t Gradually predict the distribution of temperature prediction area under the action of hydration heat temperature field at the current moment Q represents the distribution of temperature in the prediction area, t-1 represents the time node before time t, n represents that the domain is divided into n units, the length of the actual temperature observation data is used as a vector, which represents the number of prediction results of the inside and outside temperature difference, and E represents the calculation process of the temperature distribution of the area at time t under the action of hydration heat: in, Indicates the temperature difference distribution inside and outside at the current moment, It represents the temperature difference distribution of the inside and outside at the time node before time t at the hydration heat temperature, Δq t The observed value variable representing the temperature difference between inside and outside, It represents the temperature difference distribution between the inside and outside of the current moment and the next time point, and κ represents the correction value predicted at a time point by the extended Kalman filter.

4. The method for predicting the internal and external temperature difference and temperature strain of hydration heat of large volume concrete according to claim 1 is characterized in that: The S3 specifically includes the following steps: S3-1: According to the coordinates and distribution of the strain observation points, the prediction area of ​​mass concrete is delineated, and a corresponding strain observation point position coordinate is constructed, and a two-dimensional array V of mass concrete hydration heat temperature strain data is constructed. M ={(x1,y1),(x2,y2),…,(x M-1 ,y M-1 ),(x M ,y M )},(x M ,y M ) indicates V M The real-time monitoring data of the strain variable corresponding to the Mth strain observation point in the prediction area is obtained, and the original data is normalized to eliminate the dimension for the subsequent neural network training data and optimize the loss function of the training model; S3-2: Divide the data set normalized in step S3-1 into a training set and a test set, design a neural network model using the strain monitoring data of each observation point before the current time t, introduce a convolutional neural network, wherein the hidden layer unit can be designed to include convolutional layer, excitation layer, slicing layer and fusion layer operations, and use the historical observation data of hydration heat temperature field and strain to fit the weight of the convolutional neural network. After training the convolutional neural network model, train the hydration heat temperature strain variable; S3-3: Test in the actual hydration heat temperature strain prediction case, and construct the temperature strain prediction model function f(t,x',y',d,z), where t represents the time, x' and y' represent the coordinates of the hydration heat temperature strain observation point, and d represents the strain. According to the temperature strain prediction function, the strain in the prediction area at future moments can be predicted in real time.

5. The method for predicting the internal and external temperature difference and temperature strain of hydration heat of large volume concrete according to claim 1 is characterized in that: The S4 specifically includes the following steps: S4-1: Multiple time windows are used to introduce long short-term memory networks for time series prediction. The hydration heat temperature strain and hydration heat temperature data information are used through the forget gate Ft, input gate It and output gate Ot. The forget gate Ft controls the state of the memory unit Ct-1 at the previous moment and determines the strain information that needs to be forgotten at time t. The input gate It controls the candidate state ct at the current moment and determines the strain information that needs to be stored. The output gate Ot controls the memory unit Ct at the current moment and determines the state Ht that the hydration heat temperature and strain information need to be output to the outside. The formula is as follows: in, C t =F t *C t-1 +I t *c t H t =O t *tanh(c t ), σ represents the sigmoid function, tanh represents the hyperbolic tangent function, W represents the weights corresponding to the forget gate, input gate, and output gate, β represents the bias values ​​corresponding to the forget gate, input gate, and output gate, and H t represents the hidden state at time t, X t Represents the hydration heat temperature strain and hydration heat temperature data input at time t; S4-2: Establish a neural network for the temporal dependency between strain monitoring and hydration heat temperature, use the long short-term memory network to predict the strain data test set in the prediction area, and compare the accuracy of the prediction results with the strain in the actual prediction area in step S3 according to different regional ranges. Define the application scope of the neural network model trained in step S4-1 according to the distance between the temperature field core and the measuring point.

6. The method for predicting the internal and external temperature difference and temperature strain of hydration heat of large volume concrete according to claim 1 is characterized by: The S5 specifically includes the following steps: S5-1: Divide the prediction area into grids according to the hydration temperature strain of the prediction area, and create a two-dimensional array of grid point area centers with an N×N area V centered on the horizontal and vertical coordinates N ={(x1,y1),(x2,y2),…,(x N-1 ,y N-1 ),(x N ,y N )}, the array V is processed by the identity matrix function N Initialize the data information contained in the grid, where L and H are the spacing between adjacent grid points in the horizontal and vertical directions, respectively, and x N =L×N,y N =H×N; construct Thiessen polygons with adjacent observation points as perpendicular medians for each hydration heat temperature strain variable monitoring point in the prediction area, and divide it into R Thiessen polygons with the strain variable monitoring point coordinates as the center, where R is a positive integer, and the array V is calculated according to the strain prediction result in step S4. R Assign values ​​and make preliminary assignments to the distribution of the predicted results of the hydration heat temperature dependence variable of the entire area; S5-2: Triangulate the array containing the position coordinates of the concrete temperature strain monitoring points to generate a triangulated network, simulate the temperature strain distribution prediction result of step S5-1, and perform point-by-point correction on the strain prediction result corresponding to the point set generated by the triangulated network to simulate the numerical prediction result of the temperature strain distribution in the prediction area; S5-3: Use a Gaussian smoothing function to smooth the prediction result of step S5-2, and construct a Gaussian kernel function for the two-dimensional array of the center of the grid point area with horizontal and vertical coordinates σ0 is the standard deviation of the Gaussian distribution in the Gaussian kernel generator, which characterizes the discreteness of the prediction result data; thereby optimizing and updating the distribution of the strain prediction results generated in S5-2 in the prediction area, and increasing the continuity between the prediction results of each temperature strain variable distributed in the prediction area.

Images