Method for reconstructing three-dimensional temperature field of concrete structure

Abstract

The present invention provides a method for reconstructing the three-dimensional temperature field of a concrete structure, comprising: establishing a grid model based on the actual concrete structure, performing finite element calculations on the three-dimensional temperature field of the concrete structure under arbitrary initial and boundary conditions in actual working conditions; establishing a neural network for reconstructing the three-dimensional temperature field of the concrete structure; determining the optimal spacing of concrete temperature measurement points based on the numerical calculation results of the three-dimensional temperature field of the concrete structure and the neural network for reconstructing the three-dimensional temperature field of the concrete structure, and actually arranging the temperature measurement points; obtaining spatial position information of each temperature measurement point in the actual arrangement and actual temperature history monitoring data, and obtaining a reconstruction result of the actual three-dimensional temperature field of the concrete structure through the neural network for reconstructing the three-dimensional temperature field of the concrete structure. The three-dimensional temperature field of the concrete structure reconstructed by the present invention has higher accuracy and is more consistent with actual conditions. The number of temperature measurement points required is minimized under the same temperature field reconstruction accuracy conditions.

Description

Technical Field

[0001] The present invention relates to a method for reconstructing a three-dimensional temperature field of a concrete structure and belongs to the technical field of reconstructing a three-dimensional temperature field of a concrete structure. Background Art

[0002] The reconstruction of the three-dimensional temperature field of concrete structures refers to solving the temperature at any spatial and temporal position of the concrete structure based on the monitoring data of a limited number of known concrete temperature measurement points, thereby realizing the reconstruction of the three-dimensional temperature field of the concrete structure. At present, there are two problems in the reconstruction of the three-dimensional temperature field of concrete structures: (1) The reconstruction of the three-dimensional temperature field of concrete structures is mainly achieved through mathematical methods such as inverse distance weighted interpolation, cubic spline function interpolation, and Kriging interpolation. The disadvantage of these methods is that they ignore the physical laws followed by the temporal and spatial changes of concrete temperature. There is a large error between the reconstructed three-dimensional temperature field of concrete structures and the actual temperature field. In order to reduce the error, a large number of temperature measurement points need to be arranged to achieve a high-precision reconstruction of the three-dimensional temperature field of concrete structures. The embedding of a large number of temperature monitoring devices in the concrete structure will greatly increase the investment in manpower, material resources, and financial resources. (2) When reconstructing the three-dimensional temperature field of concrete structures, there is no clear optimization scheme for how to arrange temperature measurement points in the concrete structure. It is highly blind. Blindly arranging concrete temperature measurement points will lead to a waste of resources if too many temperature measurement points are arranged; and too few temperature measurement points will lead to the failure to ensure the accuracy of the reconstructed temperature field. Summary of the Invention

[0003] In order to solve the problems of low accuracy, large number of temperature measurement points and strong blindness in the arrangement of temperature measurement points in the current reconstruction of the three-dimensional temperature field of concrete structures, the purpose of the present invention is to propose a method for reconstructing the three-dimensional temperature field of concrete structures.

[0004] To achieve the above object, the present invention adopts the following technical solution: a method for reconstructing the three-dimensional temperature field of a concrete structure, comprising:

[0005] S1. Establish a grid model based on the actual concrete structure, and use finite element method to calculate the three-dimensional temperature field of the concrete structure under any initial and boundary conditions in actual working conditions;

[0006] S2. Establish a neural network for reconstructing the three-dimensional temperature field of concrete structure;

[0007] Specific methods:

[0008] S2.1. Determine the temperature control equation at each node of the grid model according to the actual working conditions;

[0009] When there is no water pipe cooling, the temperature control equation at each grid node is:

[0010]

[0011] When there is water pipe cooling, the temperature control equation at each grid node is:

[0012]

[0013]

[0014]

[0015]

[0016] Where: T = T(x, y, z, τ) represents temperature; τ represents time variable; x, y, z represent spatial variables; a is the thermal conductivity of concrete; θ0 and m are the adiabatic temperature rise parameters of concrete, where θ0 is the final value of the adiabatic temperature rise of concrete and m is a constant; T0 is the concrete pouring temperature; T w is the water temperature at the cooling water inlet; λ is the thermal conductivity of concrete; λ1 is the thermal conductivity of the cooling water pipe; c is the outer diameter of the water pipe; r0 is the inner diameter of the water pipe; S1 is the horizontal spacing of the water pipes; S2 is the vertical spacing of the water pipes; L is the length of the cooling water pipe; c w is the specific heat of water; ρ w is the density of water; q w is the cooling water flow rate;

[0017] S2.2. Build a fully connected neural network for reconstructing the three-dimensional temperature field of concrete structures.

[0018] Assume that the number of neurons in the input layer of the established neural network is 4, the number of neurons in the output layer is 1, and the neural network relationship is T net =net(x,y,z,τ,w,b), where the neural network activation function is preferably ELU, and the neural network parameters w and b are initialized using the He method;

[0019] S2.3. Construct a loss function based on the control equation determined in step S2.1;

[0020] The loss function of the temperature control equation is:

[0021] When cooling without water pipe:

[0022]

[0023] When there is water pipe cooling:

[0024]

[0025] Temperature measurement point data loss function:

[0026] loss monitoring =mse(T net (xm ,y m ,z m ,τ m )-(T m ))

[0027] Loss function:

[0028]

[0029] Where mse is the mean square error function; T net (x p ,y p ,z p ,τ m ) is the neural network output value corresponding to the training sample data of the temperature control equation; T net (x m ,y m ,z m ,τ m ) is the neural network output value corresponding to the temperature measurement point training sample data; (T m ) Temperature measurement point temperature data corresponding to the temperature measurement point training sample data; X i ∈(x p ,y p ,z p ) represents any spatial variable in the training sample data of the temperature control equation; τ m Represents the time variable in the training sample data of the temperature control equation;

[0030] S3. Determine the optimal spacing between temperature measurement points and optimize the number of temperature measurement points to be arranged;

[0031] Specific methods:

[0032] S3.1. Determine the maximum characteristic dimensions of the concrete structure in three dimensions: length L, width W, and height H;

[0033] S3.2. Calculate the spacing between temperature measurement points using the following formula: Determine the spatial position of the temperature measuring point when N=1, and obtain the numerical calculation results of the temperature history at each temperature measuring point when N=1 from the numerical calculation results of the three-dimensional temperature field of the concrete structure calculated in step S1;

[0034] S3.3, obtain the spatial position information (x p ,y p ,z p ), the spatial position information of the temperature measurement point at time N in step S3.2 (x m ,y m ,z m ) and temperature history (τ m )、(Tm ), construct the temperature control equation training sample data (x p ,y p ,z p ,τ m ) and temperature measurement point training sample data (x m ,y m ,z m ,τ m );

[0035] S3.4, input the temperature control equation training sample data and temperature measurement point training sample data obtained in step S3.3 into the neural network obtained in step S2.2 to train the neural network; optimize the neural network parameters through back propagation of the loss function, and when the loss function tends to be stable Or the number of iterations is greater than the maximum number of iterations iter≥iter max When , the neural network training is completed, and the reconstruction result of the three-dimensional temperature field of the concrete structure is obtained;

[0036] S3.5. Calculate the mean absolute error between the reconstructed three-dimensional temperature field of the concrete structure obtained in step S3.4 and the numerical calculation result of the three-dimensional temperature field of the concrete structure obtained in step S1. and the maximum absolute error ε max =max{|(T i ) 重构 -(T i ) 数值计算 |}(i=1,...,M),

[0037] Where: M is the number of grid model nodes; (T i ) 重构 is the temperature reconstruction value at the i-th grid model node; (T i ) 数值计算 is the numerical value of the temperature at the i-th grid model node;

[0038] S3.6, repeat steps S3.2-S3.5, iteratively calculate the error between the temperature field reconstruction result and the numerical calculation result at time N+1, until the error And ε max ≤2℃, the optimal concrete temperature measurement point arrangement spacing d is obtained;

[0039] S4, actually arranging temperature measuring points according to the optimal concrete temperature measuring point arrangement spacing determined in step S3, obtaining spatial position information of the actually arranged temperature measuring points and actual temperature history monitoring data, and constructing temperature control equation training sample data and temperature measuring point training sample data;

[0040] Input into the neural network obtained in step S2.2 to train the neural network; optimize the neural network parameters through back propagation of the loss function. When the loss function tends to be stable Or the number of iterations is greater than the maximum number of iterations iter≥iter max When , the neural network training is completed and the actual three-dimensional temperature field of the reconstructed concrete structure is obtained.

[0041] The present invention utilizes the physical laws followed by the concrete temperature field, and on the basis of establishing a nonlinear mapping relationship between the spatiotemporal position coordinates of the concrete structure and the corresponding temperature values using a conventional fully connected neural network, adds the constraints of the physical equation loss function, and establishes a neural network for reconstructing the three-dimensional temperature field of the concrete structure. The present invention also determines the optimal temperature measurement point arrangement spacing that meets the accuracy requirements through iterative calculation. The temperature measurement points are actually arranged according to this spacing, and the temperature history monitoring data of the measurement points are collected. The neural network for reconstructing the three-dimensional temperature field of the concrete structure is used to obtain the reconstruction result of the actual three-dimensional temperature field of the concrete structure.

[0042] The three-dimensional temperature field of the concrete structure is reconstructed by the method disclosed in the present invention. On the one hand, the reconstructed three-dimensional temperature field of the concrete structure has higher accuracy and is more consistent with the actual situation. Under the same temperature field reconstruction accuracy condition, the number of temperature measuring points required is small. On the other hand, the blindness of the arrangement of temperature measuring points can be avoided, and the minimum number of temperature measuring points can be arranged on the basis of ensuring the accuracy of the temperature field reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a flow chart of the method for reconstructing the three-dimensional temperature field of a concrete structure according to the present invention;

[0044] Figure 2 This is a framework diagram for reconstructing the three-dimensional temperature field of a concrete structure according to the present invention;

[0045] Figure 3 A schematic diagram of constructing training sample data for a specific embodiment of the present invention;

[0046] Figure 4 A schematic diagram of the process of determining the optimal arrangement spacing of temperature measurement points in a concrete structure according to a specific embodiment of the present invention;

[0047] Figure 5A This is an iterative calculation diagram of the maximum absolute error of the reconstructed three-dimensional temperature field and the spacing between the temperature measurement points of the concrete structure according to a specific embodiment of the present invention;

[0048] Figure 5B This is an iterative calculation diagram of the distance between temperature measurement points of a concrete structure and the average absolute error of the reconstructed three-dimensional temperature field according to a specific embodiment of the present invention;

[0049] Figure 6 This is a diagram of the reconstruction result of the three-dimensional temperature field of the concrete structure according to a specific embodiment of the present invention. DETAILED DESCRIPTION

[0050] The structure and features of the present invention are described in detail below with reference to the accompanying drawings and embodiments. It should be noted that various modifications can be made to the embodiments disclosed herein. Therefore, the embodiments disclosed in the specification should not be regarded as limiting the present invention, but are merely examples of embodiments, the purpose of which is to make the features of the present invention apparent.

[0051] like Figure 1 、 Figure 2 As shown, the method of reconstructing the three-dimensional temperature field of the concrete structure of the present invention is:

[0052] S1. A grid model is established based on the actual concrete structure, and the finite element method is used to calculate the three-dimensional temperature field of the concrete structure under arbitrary initial and boundary conditions in actual working conditions.

[0053] The actual working conditions of concrete structures include concrete thermal parameters, material partition information, water pipe and water flow information, and pouring progress information.

[0054] S2. Establish a neural network for reconstructing the three-dimensional temperature field of concrete structure;

[0055] Specific methods:

[0056] S2.1. Determine the temperature control equation at each node of the grid model according to the actual working conditions;

[0057] When there is no water pipe cooling, the temperature control equation at each grid node is:

[0058]

[0059] When there is water pipe cooling, the temperature control equation at each grid node is:

[0060]

[0061]

[0062]

[0063]

[0064] Where: T = T(x, y, z, τ) represents temperature; τ represents time variable; x, y, z represent spatial variables; a is the thermal conductivity of concrete; θ0 and m are the adiabatic temperature rise parameters of concrete, where θ0 is the final value of the adiabatic temperature rise of concrete and m is a constant; T0 is the concrete pouring temperature; T wis the water temperature at the cooling water inlet; P, k, and g are intermediate process quantities and have no practical significance; λ is the thermal conductivity of concrete; λ1 is the thermal conductivity of the cooling water pipe; c is the outer diameter of the water pipe; r0 is the inner diameter of the water pipe; S1 is the horizontal spacing of the water pipes; S2 is the vertical spacing of the water pipes; L is the length of the cooling water pipe; c w is the specific heat of water; ρ w is the density of water; q w is the cooling water flow rate;

[0065] S2.2. Build a fully connected neural network for reconstructing the three-dimensional temperature field of concrete structures.

[0066] Assume that the number of neurons in the input layer of the established neural network is 4, the number of neurons in the output layer is 1, and the neural network relationship is T net =net(x,y,z,τ,w,b), where the preferred neural network activation function is ELU, and the neural network parameters w and b are initialized using the He method.

[0067] S2.3. Construct a loss function based on the control equation determined in step S2.1;

[0068] The loss function of the temperature control equation is:

[0069] When cooling without water pipe:

[0070]

[0071] When there is water pipe cooling:

[0072]

[0073] Temperature measurement point data loss function:

[0074] loss monitoring =mse(T net (x m ,y m ,z m ,τ m )-(T m ))

[0075] Loss function:

[0076]

[0077] Where mse is the mean square error function; T net (x p ,y p ,z p ,τ m ) is the neural network output value corresponding to the training sample data of the temperature control equation; T net (xm ,y m ,z m ,τ m ) is the neural network output value corresponding to the temperature measurement point training sample data; (Tm) is the temperature data of the temperature measurement point corresponding to the temperature measurement point training sample data; X i ∈(x p ,y p ,z p ) represents any spatial variable in the training sample data of the temperature control equation; τ m Represents the time variable in the training sample data of the temperature control equation;

[0078] S3. Determine the optimal spacing between temperature measurement points and optimize the number of temperature measurement points to be arranged;

[0079] Specific methods:

[0080] S3.1. Determine the maximum characteristic dimensions of the concrete structure in three dimensions: length L, width W, and height H;

[0081] S3.2. Calculate the spacing between temperature measurement points using the following formula: Determine the spatial position of the temperature measuring point when N=1, and obtain the numerical calculation results of the temperature history at each temperature measuring point when N=1 from the numerical calculation results of the three-dimensional temperature field of the concrete structure calculated in step S1;

[0082] S3.3, obtain the spatial position information (x p ,y p ,z p ), the spatial position information of the temperature measurement point at time N in step S3.2 (x m ,y m ,z m ) and temperature history (τ m )、(T m ), construct the temperature control equation training sample data (x p ,y p ,z p ,τ m ) and temperature measurement point training sample data (x m ,y m ,z m ,τ m );

[0083] S3.4, input the temperature control equation training sample data and temperature measurement point training sample data obtained in step S3.3 into the neural network obtained in step S2.2 to train the neural network; optimize the neural network parameters through back propagation of the loss function, and when the loss function tends to be stable Or the number of iterations is greater than the maximum number of iterations iter≥itermax When , the neural network training is completed, and the reconstruction result of the three-dimensional temperature field of the concrete structure is obtained;

[0084] S3.5. Calculate the mean absolute error between the reconstructed three-dimensional temperature field of the concrete structure obtained in step S3.4 and the numerical calculation result of the three-dimensional temperature field of the concrete structure obtained in step S1. and the maximum absolute error ε max =max{|(T i ) 重构 -(T i ) 数值计算 |}(i=1,...,M),

[0085] Where: M is the number of grid model nodes; (T i ) 重构 is the temperature reconstruction value at the i-th grid model node; (T i ) 数值计算 is the numerically calculated value of the temperature at the i-th grid model node.

[0086] S3.6, repeat steps S3.2-S3.5, iteratively calculate the error between the temperature field reconstruction result and the numerical calculation result at time N+1, until the error And ε max ≤2℃, the optimal concrete temperature measurement point arrangement spacing d is obtained;

[0087] S4, actually arranging temperature measuring points according to the optimal concrete temperature measuring point arrangement spacing determined in step S3, obtaining spatial position information of the actually arranged temperature measuring points and actual temperature history monitoring data, and constructing temperature control equation training sample data and temperature measuring point training sample data;

[0088] Input into the neural network obtained in step S2.2 to train the neural network; optimize the neural network parameters through back propagation of the loss function. When the loss function tends to be stable Or the number of iterations is greater than the maximum number of iterations iter≥iter max When , the neural network training is completed and the actual three-dimensional temperature field of the reconstructed concrete structure is obtained.

[0089] The technical solution of the present invention is further described below through specific embodiments.

[0090] Actual working conditions: A concrete structure is a 1.5m x 1.5m x 1.5m cubic block, poured in a single pour, with no internal cooling water pipes. The thermal conductivity of the concrete is 216kJ / (m·d·°C), the thermal diffusivity is 0.0864㎡ / d, and the adiabatic temperature rise parameters θ0 = 25 and m = 0.4. By embedding several thermometers within the concrete structure to monitor the temperature history at each temperature measurement point, the overall temperature field of the concrete structure can be reconstructed.

[0091] The method for reconstructing the three-dimensional temperature field of a concrete structure proposed in the present invention reconstructs the three-dimensional temperature field of the concrete structure. First, a neural network is constructed for reconstructing the three-dimensional temperature field of the concrete structure. Then, the optimal spacing of concrete temperature measurement points that meets the temperature field reconstruction accuracy is determined, and the temperature measurement points are arranged according to the optimal spacing. The spatial position information and actual temperature monitoring data of each optimally arranged temperature measurement point are obtained, and the three-dimensional temperature field of the concrete structure is reconstructed based on the constructed neural network. The specific implementation steps are as follows:

[0092] 1. A grid model was established based on an actual 1.5m×1.5m×1.5m cubic concrete structure. Based on the actual working conditions: single pouring, no material partitioning, no cooling water pipes, and known concrete thermal parameters, the temperature field of this concrete structure was calculated by finite element method under arbitrary pouring temperature, ambient temperature, and heat dissipation coefficient.

[0093] 2. Establish a neural network for reconstructing the three-dimensional temperature field of concrete structure.

[0094] Steps to establish a neural network for reconstructing the 3D temperature field of concrete structure:

[0095] 2.1 According to the actual working conditions (one-time pouring, no material partitioning, no cooling water pipes, and known concrete thermal parameters), the temperature control equation at each node of the grid model is the same, which is:

[0096]

[0097] 2.2 Establish a fully connected neural network with the structural parameters of [4, 20, 20, 20, 20, 1], the activation function of the neural network is ELU, and the neural network parameters w and b are initialized using the He method.

[0098] 2.3. Construct the loss function based on the control equation determined in step 2.1.

[0099] Temperature control equation loss function:

[0100]

[0101] Temperature measurement point data loss function:

[0102] loss monitoring =mse(T net (x m ,y m ,z m ,τ m )-(T m ))

[0103] Loss function:

[0104] loss = 100 × loss monitoring +loss f1

[0105] 3. Based on the numerical calculation results of the three-dimensional temperature field of the concrete structure and the neural network reconstructed based on the three-dimensional temperature field of the concrete structure, the optimal spacing of the concrete temperature measurement points is determined, and the temperature measurement points of the concrete structure are actually arranged according to the optimal spacing.

[0106] 3.1 The maximum characteristic dimension of the concrete structure in three dimensions is L = W = H = 1.5 m. The calculation formula for the optimal arrangement spacing of the temperature measuring points is: like Figure 4 As shown in the figure, the spatial position information of the measuring points when N=1 is (0, 0, 0), (0, 0, 1.5), (0, 1.5, 0), (0, 1.5, 1.5), (1.5, 0, 0), (1.5, 0, 1.5), (1.5, 1.5, 0), (1.5, 1.5, 1.5). From the numerical calculation results of the three-dimensional temperature field of the concrete structure, the temperature history at each temperature measuring point when N=1 is obtained.

[0107] 3.2 Obtaining the spatial position information of each node in the grid model (x p ,y p ,z p ), spatial position information of temperature measurement point at time N (x m ,y m ,z m ) and the numerical calculation results of temperature history (τ m )、(T m ), construct the temperature control equation training sample data (x p ,y p ,z p ,τ m ) and temperature measurement point training sample data (x m ,y m ,z m ,τ m ),like Figure 3 shown.

[0108] 3.3 Input the temperature control equation training sample data and the temperature measurement point training sample data into the neural network to train the neural network; the loss function is back-propagated to optimize the neural network parameters. When the loss function tends to be stable Δloss≤0.0001 or the number of iterations is greater than the set maximum number of iterations iter≥50000, the neural network training is completed, and the reconstruction result of the three-dimensional temperature field of the concrete structure is obtained.

[0109] 3.4 Calculation of the maximum absolute error ε between the reconstruction results of the three-dimensional temperature field of the concrete structure and the numerical calculation results of the three-dimensional temperature field of the concrete structure max =max{|(T i ) 重构 -(T i ) 数值计算 |}(i=1,...,M) and mean absolute error

[0110] 3.5 Repeat steps 3.1-3.4, iteratively calculate the error between the temperature field reconstruction result and the numerical calculation result at time N+1, until the error And ε max ≤2℃, the optimal concrete temperature measurement point arrangement spacing is obtained; from the iterative calculation results in Figure 5, it can be seen that when N=4, the optimal arrangement spacing d=0.375m is obtained.

[0111] 4. Arrange the temperature measurement points according to the optimal spacing, obtain the spatial position information of each temperature measurement point and the actual temperature history monitoring data, construct the temperature control equation training sample data and the temperature measurement point training sample data; input them into the neural network to train the neural network, and optimize the neural network parameters through the back propagation of the loss function. When the loss function tends to be stable Δloss ≤ 0.0001 or the number of iterations is greater than the set maximum number of iterations iter ≥ 50000, the neural network training ends and the reconstruction result of the actual three-dimensional temperature field of the concrete structure is obtained. The results are as follows Figure 6 shown.

[0112] Table 1 shows the error analysis of the results of reconstructing the 3D temperature field of concrete structures using different methods. As can be seen from Table 1, the method proposed in this invention for reconstructing the 3D temperature field of concrete structures has a higher reconstruction accuracy than the Kriging interpolation and cubic spline function interpolation methods.

[0113] Table 1 Error analysis of reconstruction results of different methods (N=4)

[0114]

[0115] Finally, it should be noted that the embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for reconstructing the three-dimensional temperature field of a concrete structure, characterized by: It includes: S1. Establish a grid model based on the actual concrete structure, and use finite element method to calculate the three-dimensional temperature field of the concrete structure under any initial and boundary conditions in actual working conditions; S2. Establish a neural network for reconstructing the three-dimensional temperature field of concrete structure; Specific methods: S2.

1. Determine the temperature control equation at each node of the grid model according to the actual working conditions; When there is no water pipe cooling, the temperature control equation at each grid node is: When there is water pipe cooling, the temperature control equation at each grid node is: Where: T = T(x, y, z, τ) represents temperature; τ represents time variable; x, y, z represent spatial variables; a is the thermal conductivity of concrete; θ0 and m are the adiabatic temperature rise parameters of concrete, where θ0 is the final value of the adiabatic temperature rise of concrete and m is a constant; T0 is the concrete pouring temperature; T w is the water temperature at the cooling water inlet; λ is the thermal conductivity of concrete; λ1 is the thermal conductivity of the cooling water pipe; c is the outer diameter of the water pipe; r0 is the inner diameter of the water pipe; S1 is the horizontal spacing of the water pipes; S2 is the vertical spacing of the water pipes; L is the length of the cooling water pipe; c w is the specific heat of water; ρ w is the density of water; q w is the cooling water flow rate; S2.

2. Build a fully connected neural network for reconstructing the three-dimensional temperature field of concrete structures. Assume that the number of neurons in the input layer of the established neural network is 4, the number of neurons in the output layer is 1, and the neural network relationship is T net =net(x,y,z,τ,w,b), where the neural network activation function is preferably ELU, and the neural network parameters w and b are initialized using the He method; S2.

3. Construct a loss function based on the control equation determined in step S2.1; The loss function of the temperature control equation is: When cooling without water pipe: When there is water pipe cooling: Temperature measurement point data loss function: loss monitoring =mse(T net (x m ,y m ,z m ,τ m )-(T m )) Loss function: Where mse is the mean square error function; T net (x p ,y p ,z p ,τ m ) is the neural network output value corresponding to the training sample data of the temperature control equation; T net (x m ,y m ,z m ,τ m ) is the neural network output value corresponding to the temperature measurement point training sample data; (Tm) is the temperature data of the temperature measurement point corresponding to the temperature measurement point training sample data; X i ∈(x p ,y p ,z p ) represents any spatial variable in the training sample data of the temperature control equation; τ m Represents the time variable in the training sample data of the temperature control equation; S3. Determine the optimal spacing between temperature measurement points and optimize the number of temperature measurement points to be arranged; Specific methods: S3.

1. Determine the maximum characteristic dimensions of the concrete structure in three dimensions: length L, width W, and height H; S3.

2. Calculate the spacing between temperature measurement points using the following formula: Determine the spatial position of the temperature measuring point when N=1, and obtain the numerical calculation results of the temperature history at each temperature measuring point when N=1 from the numerical calculation results of the three-dimensional temperature field of the concrete structure calculated in step S1; S3.3, obtain the spatial position information (x p ,y p ,z p ), the spatial position information of the temperature measurement point at time N in step S3.2 (x m ,y m ,z m ) and temperature history (τ m )、(T m ), construct the temperature control equation training sample data (x p ,y p ,z p ,τ m ) and temperature measurement point training sample data (x m ,y m ,z m ,τ m ); S3.4, input the temperature control equation training sample data and temperature measurement point training sample data obtained in step S3.3 into the neural network obtained in step S2.2 to train the neural network; optimize the neural network parameters through back propagation of the loss function, and when the loss function tends to be stable Or the number of iterations is greater than the maximum number of iterations iter≥iter max When , the neural network training is completed, and the reconstruction result of the three-dimensional temperature field of the concrete structure is obtained; S3.

5. Calculate the mean absolute error between the reconstructed three-dimensional temperature field of the concrete structure obtained in step S3.4 and the numerical calculation result of the three-dimensional temperature field of the concrete structure obtained in step S1. and the maximum absolute error ε max =max{|(T i ) 重构 -(T i ) 数值计算 |}(i=1,...,M), Where: M is the number of grid model nodes; (T i ) 重构 is the temperature reconstruction value at the i-th grid model node; (T i ) 数值计算 is the numerical value of the temperature at the i-th grid model node; S3.6, repeat steps S3.2-S3.5, iteratively calculate the error between the temperature field reconstruction result and the numerical calculation result at time N+1, until the error And ε max ≤2℃, the optimal concrete temperature measurement point arrangement spacing d is obtained; S4, actually arranging temperature measuring points according to the optimal concrete temperature measuring point arrangement spacing determined in step S3, obtaining spatial position information of the actually arranged temperature measuring points and actual temperature history monitoring data, and constructing temperature control equation training sample data and temperature measuring point training sample data; Input into the neural network obtained in step S2.2 to train the neural network; optimize the neural network parameters through back propagation of the loss function. When the loss function tends to be stable Or the number of iterations is greater than the maximum number of iterations iter≥iter max When , the neural network training is completed and the actual three-dimensional temperature field of the reconstructed concrete structure is obtained.

Images