Intelligent construction method for global time-varying temperature field of large-span concrete bridge structure

Abstract

The invention discloses an intelligent construction method for a global time-varying temperature field of a large-span concrete bridge structure, and relates to the technical field of bridge structure temperature field prediction. Discretizing the bridge and calculating position coordinates; dividing according to position coordinates by considering solar radiation received by the whole region of the bridge; collecting actually measured environment data; constructing a data set, and carrying out normalization processing; constructing a physical constraint loss function and a data constraint loss function; a neural network model is constructed, and an output target is a final prediction value; in a model training process, configuring a weight coefficient for a constraint item, and updating until loss convergence to obtain a temperature prediction result; evaluating the neural network model and completing training; and performing temperature prediction on an actual bridge by using the neural network model. A correlation mechanism of bridge temperature prediction is established, the bridge temperature at a certain moment in the future is predicted while historical temperature data is considered, a physical mechanism is introduced, constraint loss of physical data double constraints is formed, and prediction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of temperature field prediction of bridge structures, and in particular to a method for intelligently constructing a global time-varying temperature field of a long-span concrete bridge structure. Background Art

[0002] With the development of bridge construction in my country, more and more bridges have been put into use and have become the main hubs connecting various places. With the long-term use of most bridges, the main task of bridges has gradually shifted from traditional bridge construction to bridge monitoring.

[0003] It is necessary to consider the long-term operation and maintenance of bridges. However, bridges are affected by various loads during actual use, which reduces the life of bridges. The impact of temperature loads on bridges is also an issue that cannot be ignored. The traditional method of monitoring bridge status based on the deployment of sensors requires a lot of initial costs and has limited controllability. Therefore, it is necessary to explore a simple method for obtaining the temperature of bridge structures.

[0004] With the rapid development of artificial intelligence, in data-driven modeling technology, artificial neural networks can handle complex nonlinear relationships, are applicable to various problem areas, and adapt to high-dimensional data, complex inputs, and diversified forms. They have a certain tolerance to noise and missing data, and they also have significant advantages in image processing. Therefore, proposing a convenient and fast bridge temperature prediction scheme based on neural networks is of great significance for the safe operation monitoring of bridges. Summary of the invention

[0005] In view of the shortcomings of traditional sensor deployment methods for monitoring bridge temperature, such as high construction and maintenance costs, long equipment deployment and data acquisition time, and low efficiency in massive data analysis, the present invention provides a method for intelligently constructing the global time-varying temperature field of a large-span concrete bridge structure. The method establishes an association mechanism for bridge temperature prediction based on traditional long short-term memory networks, and can predict the bridge temperature at a certain moment in the future while considering historical temperature data. Physical mechanisms are introduced into the model to form a constraint loss of dual constraints of physical data, effectively improving the accuracy of model prediction.

[0006] To achieve the above object, the present invention adopts the following technical solution: a method for intelligently constructing the global time-varying temperature field of a long-span concrete bridge structure, comprising the following steps:

[0007] Step 1: Calculate the bridge location coordinates

[0008] The bridge is discretized into N equal intervals along the length direction L, width direction W and vertical direction Z, and the position coordinates of a certain point are expressed as:

[0009]

[0010] Then the position coordinates of all points are expressed as:

[0011] X=(X0 X1…X N-1 X N ) T

[0012] Among them, for a certain element X i i=0,1,...,N-1,N, and its form is:

[0013]

[0014] Step 2: Calculate solar radiation parameters

[0015] Calculate the solar radiation intensity I at time t solar , considering the solar radiation received by the entire bridge area, the solar radiation intensity I is obtained solar According to the division of position coordinates, it is expressed as:

[0016] I solar =(I0 I1…I N-1 I N ) T

[0017] Among them, for a certain element I i , which is expressed in the form of:

[0018]

[0019] Step 3: Collect measured environment data

[0020] Collect environmental data at different bridge locations at time t, including the ambient temperature T env , wind speed V mind , relative humidity RH and historical temperature data T history ,in:

[0021] Ambient temperature T env It is expressed as:

[0022] For an element Its form is:

[0023]

[0024] Wind speed V mind Expressed as: V mind =(V0 V1…V N-1 V N ) T

[0025] For an element V i , which is expressed in the form of:

[0026]

[0027] Relative humidity RH is expressed as: RH = (R0 R1…R N-1 R N ) T

[0028] For an element R i , which is expressed in the form of:

[0029]

[0030] Historical temperature data T history It is expressed as:

[0031] For an element Its form is:

[0032]

[0033] Step 4: Build the dataset

[0034] The position coordinate X and the corresponding characteristic data, including the solar radiation intensity I solar 、Ambient temperature T env , wind speed V mind , relative humidity RH and historical temperature data T history The data set is constructed and normalized to limit its range to the bridge size. Data outside this range are marked as outliers and deleted. The normalized data set is expressed as follows:

[0035] G=[XI solar T env V mind RH T history ];

[0036] Step 5: Constructing a physical constraint loss function

[0037] The heat conduction equation is defined to describe the time and space variation of the internal temperature of the bridge when only concrete is considered, and it is converted into heat conduction constraint loss, which is expressed as:

[0038]

[0039] Where N' is the number of data samples, ρ is the material density, c is the specific heat capacity, is the predicted temperature value, is the time derivative of temperature, κ is the thermal conductivity, is the spatial derivative of temperature, x, y, and z correspond to the length direction L, width direction W, and vertical direction Z of the bridge discretization, respectively;

[0040] Convective heat transfer is defined to describe the heat exchange boundary conditions between the bridge surface and the air. The physical process is described by Newton's cooling law, and then the convection heat transfer constraint loss is constructed, which is expressed as:

[0041]

[0042] Where h is the convective heat transfer coefficient, q i is the heat flow at the corresponding point, and T surface is the temperature distribution on the bridge surface, i.e. the temperature prediction value n is the normal direction of the bridge surface;

[0043] Then, the physical constraint loss function is expressed as:

[0044] L physics =Loss conduct +Loss conv ;

[0045] Step 6: Constructing a data-constrained loss function

[0046] Construct a data constraint loss to measure the gap between the predicted temperature and the actual temperature, and use the mean square error to represent the data constraint loss function, which is expressed as:

[0047]

[0048] Step 7: Build a neural network model

[0049] Divide the normalized data set G into training set G T and the test set G E ;

[0050] Define the input target as the training set G T Excluding historical temperature data T history Other data outside

[0051] Define the output target as the final predicted value The historical temperature data T history As validation data;

[0052] Use the long short-term memory network to build a neural network model and output the final prediction value It is expressed as:

[0053]

[0054] Among them, the temperature prediction value for a certain element Its form is:

[0055]

[0056] Step 8: Constructing physical data dual constraints

[0057] In the process of neural network model training, weight coefficients are configured for physical constraints and data constraints respectively, and the sum of the weight coefficients is set to 1. The loss function expression of physical data dual constraints is as follows:

[0058] L total =λ1Loss data +λ2L physics

[0059] In the formula, λ1 is the weight coefficient of data constraint loss, and λ2 is the weight coefficient of physical constraint loss;

[0060] Use the gradient optimization algorithm to update to determine the best weights, and repeat the update on the entire normalized data set G until the loss converges to obtain the temperature prediction result. It is expressed as:

[0061]

[0062] Among them, for an element Its form is:

[0063]

[0064] Step 9: Neural Network Model Evaluation

[0065] Using the test set G E To measure the capability of the neural network model, the mean square error is used to evaluate the neural network model. If the neural network model meets the mean square error condition during training, the training is completed. If not, back propagation is continued until the condition is met to obtain a trained neural network model.

[0066] Step 10: Use the neural network model to make predictions

[0067] The temperature of the entire bridge at time point t′ is predicted based on the actual bridge working condition information, and the position coordinate X′, solar radiation intensity I s ' olar 、Ambient temperature T e ' nv , wind speed V m ' ind , relative humidity RH′ and historical temperature data T h ' istoryAfter normalization, the data set G′ is obtained. The data set G′ is input into the trained neural network model to obtain the final temperature prediction output.

[0068] Furthermore, in step 2, the solar radiation intensity I solar For direct radiation I dir , Scattered Radiation I diff and surface reflected radiation I ref The sum of , where:

[0069]

[0070] Where, t u is the Link turbidity coefficient, k a is the relative atmospheric pressure, m is the light mass, I0 is the solar constant, φ is the angle between the illuminated surface and the incident light, P is the atmospheric transparency coefficient, β s is the solar altitude angle, β n is the angle between the external normal of the illuminated surface and the ground, r e is the ground reflectivity.

[0071] Furthermore, in step 7, a long short-term memory network is used to model time series data, and W is defined f , W i , W c and W o are the input weight matrices of the forget gate, input gate, memory unit update, and output gate, respectively, and b f 、b i 、b c and b o are the bias terms of the forget gate, input gate, memory unit update and output gate respectively, then:

[0072] Input Gate:

[0073] i(t)=σ(W i [h(t-1),X(t)]+b i )

[0074] Forget Gate:

[0075] f(t)=σ(W f [h(t-1),X(t)]+b f )

[0076] Output Gate:

[0077] o(t)=σ(W o [h(t-1),X(t)]+b o )

[0078] Memory unit update:

[0079]

[0080] Hide status updates:

[0081] h(t)=o(t)*tanh(C(t))

[0082] Output prediction:

[0083] The output h(t) of LSTM can be used to obtain the final prediction value through the fully connected layer.

[0084] Where σ(·) is the sigmoid function, tanh(·) is the hyperbolic tangent function, and W out is the weight of the output layer, b out is the bias term of the output layer.

[0085] Furthermore, in step nine, during the training of the neural network model, the mean square error (MSE) is defined to evaluate the performance of the neural network model every 10 epochs, as shown below:

[0086]

[0087] Where N valid is the number of samples in the test set, is the loss of the ith sample.

[0088] Furthermore, in step 10, the final temperature prediction output is The data is displayed through a three-dimensional cloud map.

[0089] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention trains the neural network model input features and the model training algorithm according to the significant time-varying characteristics, time history accumulation characteristics and spatial characteristics of the bridge temperature, and applies them to the prediction of the bridge structure temperature. Compared with the traditional method of deploying sensors to monitor the bridge temperature, it can help save a lot of time and cost, and can infer the bridge temperature at a certain moment in the future based on the existing data. In addition, physical mechanism constraints are introduced into the model to connect the theoretical and actual bridges, help to reasonably set the boundary conditions, ensure the accuracy of the model prediction, and achieve a time-saving, labor-saving and high-precision temperature monitoring effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 is a flow chart of the method of the present invention;

[0091] Figure 2 It is a three-dimensional cloud map of the final temperature prediction output obtained in the embodiment. DETAILED DESCRIPTION

[0092] The technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0093] like Figure 1 As shown, a method for intelligently constructing the global time-varying temperature field of a long-span concrete bridge structure includes the following steps:

[0094] Step 1: Calculate the bridge location coordinates

[0095] The bridge is discretized into N equal intervals along the length direction L, width direction W and vertical direction Z. The position coordinates of a certain point can be expressed as:

[0096]

[0097] Then the position coordinates of all points can be expressed as:

[0098] X=(X0 X1…X N-1 X N ) T

[0099] Among them, for a certain element X i i=0,1,...,N-1,N, and its form is:

[0100]

[0101] Step 2: Calculate solar radiation parameters

[0102] Calculate the solar radiation intensity I at time t solar , considering the solar radiation intensity I solar For direct radiation I dir , Scattered Radiation I diff and surface reflected radiation I ref The sum of, that is:

[0103] I solar =I dir +I diff +I ref

[0104] in,

[0105] Direct Radiation I dir Considering the effects of impurities such as air molecules, aerosols, and dust in the atmosphere, the formula is defined as:

[0106]

[0107] Scattered Radiation I diff Considering the influence of the top and bottom plates of the box girder, the formula is defined as:

[0108]

[0109] Surface reflected radiation I ref Considering the different surface morphologies, the formula is defined as:

[0110]

[0111] In the formula, t u is the Link turbidity coefficient, k a is the relative atmospheric pressure, m is the light mass, I0 is the solar constant, φ is the angle between the illuminated surface and the incident light, P is the atmospheric transparency coefficient, β s is the solar altitude angle, β n is the angle between the external normal of the illuminated surface and the ground, r e is the ground reflectivity.

[0112] Considering the solar radiation received by the entire bridge area, the solar radiation intensity I can be obtained solar According to the division of position coordinates, it is expressed as:

[0113] I solar =(I0 I1…I N-1 I N ) T

[0114] Among them, for a certain element I i i=0,1,...,N-1,N, and its form is:

[0115]

[0116] Step 3: Collect measured environment data

[0117] Collect environmental data at different bridge locations at time t, including the ambient temperature T env , wind speed V mind , relative humidity RH and historical temperature data T history ,in:

[0118] Ambient temperature T env It is expressed as:

[0119]

[0120] Among them, for an element Its form is:

[0121]

[0122] Wind speed V mind It is expressed as:

[0123] V mind =(V0 V1…V N-1 V N ) T

[0124] Among them, for a certain element V i i=0,1,...,N-1,N, and its form is:

[0125]

[0126] Relative humidity RH is expressed as:

[0127] RH=(R0 R1…R N-1 R N ) T

[0128] Among them, for a certain element R i i=0,1,...,N-1,N, and its form is:

[0129]

[0130] Historical temperature data T history It is expressed as:

[0131]

[0132] Among them, for an element Its form is:

[0133]

[0134] Step 4: Build the dataset

[0135] The position coordinate X and the corresponding characteristic data, including the solar radiation intensity I solar 、Ambient temperature T env , wind speed V mind , relative humidity RH and historical temperature data T history A data set is constructed, and the data and location coordinates in the data set are normalized so that their range is limited to the bridge size. Data outside this range are marked as outliers and deleted.

[0136] The normalization process is expressed as follows:

[0137]

[0138] In the formula, α normis the normalized data, α is the original data of each feature data, α min and α max are the minimum and maximum values ​​of the original data of each feature data respectively.

[0139] The normalized data set is expressed as follows:

[0140] G=[XI solar T env V mind RH T history ]

[0141] Step 5: Constructing a physical constraint loss function

[0142] The heat conduction equation is defined to describe the change of the internal temperature of the bridge with time and space when only concrete is considered, which is expressed as:

[0143]

[0144] In the formula, ρ is the material density, c is the specific heat capacity, is the rate of change of temperature with time, κ is the thermal conductivity, is the Laplace operator of the temperature field, x, y and z correspond to the length direction L, width direction W and vertical direction Z of the discretization of the bridge respectively.

[0145] Converting it into a loss function, the temperature prediction value is The heat conduction constraint loss is calculated by the following formula:

[0146]

[0147] Where N' is the number of data samples, is the time derivative of temperature, is the spatial derivative of temperature,

[0148] Convective heat transfer is defined to describe the boundary conditions of heat exchange between the bridge surface and the air. The physical process is described by Newton's law of cooling, which is expressed as:

[0149]

[0150] Where h is the convective heat transfer coefficient, T surface is the temperature distribution on the bridge surface, i.e. the temperature prediction value q i is the heat flow at the corresponding point, and n is the normal direction of the bridge surface.

[0151] Then, the convection heat transfer constraint loss is constructed so that the temperature prediction value satisfies the convection heat transfer law, which is expressed as:

[0152]

[0153] Then, the physical constraint loss function is expressed as:

[0154] L physics =Loss conduct +Loss conv

[0155] Step 6: Constructing a data-constrained loss function

[0156] Construct a data constraint loss to measure the gap between the predicted temperature and the actual temperature, and use the mean square error to represent the data constraint loss function, which is expressed as:

[0157]

[0158] Step 7: Build a neural network model

[0159] The normalized data set G is divided into two subsets in a time series with a ratio of 4:1, and 80% of the data is used as the training set G T , 20% of the data is used as the test set G E ;

[0160] Define the input target as the training set G T Excluding historical temperature data T history Other data, namely, solar radiation intensity I solar 、Ambient temperature T env , wind speed V mind , relative humidity RH and position coordinate X;

[0161] Define the output target as the final predicted value The historical temperature data T history As validation data;

[0162] Use long short-term memory networks to model time series data and define W f , W i , W c and W o are the input weight matrices of the forget gate, input gate, memory unit update, and output gate, respectively, and b f , b i , b c and b o are the bias terms of the forget gate, input gate, memory unit update and output gate respectively, then:

[0163] Input Gate:

[0164] i(t)=σ(W i[h(t-1),X(t)]+b i )

[0165] Forget Gate:

[0166] f(t)=σ(W f [h(t-1),X(t)]+b f )

[0167] Output Gate:

[0168] o(t)=σ(W o [h(t-1),X(t)]+b o )

[0169] Memory unit update:

[0170]

[0171] Hide status updates:

[0172] h(t)=o(t)*tanh(C(t))

[0173] Output prediction:

[0174] The output h(t) of LSTM can be used to obtain the final prediction value through the fully connected layer.

[0175] Where σ(·) is the sigmoid function, tanh(·) is the hyperbolic tangent function, and W out is the weight of the output layer, b out is the bias term of the output layer, It is expressed as:

[0176]

[0177] Among them, the temperature prediction value for a certain element Its form is:

[0178]

[0179] Step 8: Constructing physical data dual constraints

[0180] Preliminary temperature prediction Finally, the physical constraint loss function in step 5 and the data constraint loss function in step 6 are called for further fitting.

[0181] In the process of neural network model training, weight coefficients are configured for physical constraints and data constraints respectively, and the sum of the weight coefficients is set to 1. The loss function expression of physical data dual constraints is as follows:

[0182] L total=λ1Loss data +λ2L physics

[0183] Where λ1 is the weight coefficient of data constraint loss, and λ2 is the weight coefficient of physical constraint loss.

[0184] Use the gradient optimization algorithm to update to determine the best weights, and repeat the update on the entire normalized data set G until the loss converges to obtain the temperature prediction result. It is expressed as:

[0185]

[0186] Among them, for an element Its form is:

[0187]

[0188] Step 9: Neural Network Model Evaluation

[0189] Using the test set G E To measure the ability of the neural network model, every 10 epochs during the training process, the mean square error MSE is defined to evaluate the performance of the neural network model, which is expressed as follows:

[0190]

[0191] Where N valid is the number of samples in the test set, is the loss of the ith sample.

[0192] If the neural network model meets the mean square error condition during training, the training is completed. If not, back propagation continues until the condition is met to obtain a trained neural network model.

[0193] Step 10: Use the neural network model to make predictions

[0194] Based on the trained neural network model, the full-bridge temperature at a certain time point t′ is predicted according to the actual bridge working condition information.

[0195] The position coordinate X′ and solar radiation intensity I at the time point t′ are s ' olar 、Ambient temperature T e ' nv , wind speed V m ' ind , relative humidity RH′ and historical temperature data T h ' istory After normalization, the data set G′ is expressed as follows:

[0196] G′=[X′I s ' olar T e ' nv V m ' ind RH′T h ' istory ]

[0197] Input the data set G′ into the trained neural network model to obtain the final temperature prediction output It is expressed as:

[0198]

[0199] Finally, a three-dimensional cloud map of the time point is obtained through data visualization for display.

[0200] Example

[0201] Take a concrete bridge in Hebei Province as an example. It is summer there and the time is 12 noon. The whole temperature of the bridge is predicted at 12 noon the next day.

[0202] (1) Calculate the bridge location coordinates

[0203] The latitude of the bridge is 39°N, the area is temperate monsoon climate, the direct angle of the sun is 23.44°, the bridge is northwest-southeast oriented, the angle between the longitudinal axis and the north-south direction is 30°, the length L of the bridge is 100m, the width W is 10m, and the vertical Z is 10m. The position coordinates of all points can be expressed as follows according to N=10 equal intervals:

[0204] X=(X0 X1…X9 X 10 ) T

[0205] For an element X i , which is expressed in the form of:

[0206]

[0207] Taking X1 as an example, it can be expressed as

[0208] By analogy, the position coordinates of the entire bridge can be obtained.

[0209] (2) Calculation of solar radiation parameters

[0210] By inputting the atmospheric parameters and the coordinates of the entire bridge position, the solar radiation intensity I can be obtained. solar According to the division of position coordinates, it is expressed as:

[0211]

[0212] (3) Collect measured environmental data

[0213] When the time is 12 noon and the ambient temperature is 33.4 degrees Celsius, the ambient temperature T env Element T e1 It is expressed as:

[0214]

[0215] Wind speed V mind The element V1 is represented by:

[0216]

[0217] The element R1 in relative humidity RH is expressed as:

[0218]

[0219] Collect the temperature data of the bridge in the past week and construct the historical temperature data T history , where the element T h1 It is expressed as:

[0220]

[0221] (4) Constructing a dataset

[0222] The data was normalized and some results were obtained as shown in Table 1:

[0223] Table 1 Normalized data set (part)

[0224] data 1 2 3 4 <![CDATA[I solar ]]> 0.19925 0.06082 0.05385 0.10750 <![CDATA[V mind ]]> 0.38322 0.19717 0.00228 0.15428 RH 0.18340 0.076923 0.22215 0.08157 <![CDATA[T history ]]> 0.15266 0.063122 0.129991 0.02112

[0225] Then we can get the normalized data set G: G = [XI solar T env V mind RH T history ]

[0226] (5) Constructing physical constraint loss function

[0227] The bridge is made of C40 concrete with a density of 2400 (kg / m 3 ), the specific heat capacity is 900 (J / kg·k), the thermal conductivity is 1.8 (W / m·k), and the convection heat transfer coefficient under this condition is 34 (W / m 2 K), and introduce it into the physical formula.

[0228] Then the heat conduction equation: Substituting the parameters into the loss function, we can get:

[0229]

[0230] Convective heat transfer equation, substituting the parameters into the loss function:

[0231]

[0232] (6) Constructing a data-constrained loss function

[0233]

[0234] (7) Constructing a neural network model

[0235] The normalized data set G is divided into two subsets in a time series with a ratio of 4:1, and 80% of the data is used as the training set G T , 20% of the data is used as the test set G E .

[0236] Define the input target as the training set G T Excluding historical temperature data T history Other data outside.

[0237] Define the output target as the final predicted value The historical temperature data T history As verification data.

[0238] (8) Constructing dual constraints on physical data

[0239] Call the physical constraint loss function and data constraint loss function to get the temperature prediction value preliminarily Weight allocation is performed. During the training of the neural network model, weight coefficients are configured for the physical constraint items and the data constraint items, and the initial weight coefficients are 0.5 and 0.5 respectively. The total is limited to 1, the time step is specified to 1, and the number of cycles is 100. This model is trained and the training results are shown in Table 2 below:

[0240] Table 2 Weight distribution results (partial)

[0241]

[0242]

[0243] When the weight distribution is λ1=0.502392344 and λ2=0.497607656, the loss is minimized, the weight distribution is completed, and the final data training is performed. The final prediction results are shown in Table 3 below:

[0244] Table 3 Final predicted temperature results (partial)

[0245]

[0246] (9) Neural network model evaluation

[0247] Using the test set G E To measure the capability of the neural network model, the mean square error (MSE) is defined every 10 epochs during the training process to evaluate the performance of the neural network model. The training results are shown in Table 4 below:

[0248] Table 4 Error verification results (partial)

[0249]

[0250] When training reaches 91 times, the loss and mean square error remain unchanged, that is, the mean square error condition is met, the optimal solution of the model is obtained, and the neural network model training is completed.

[0251] (10) Using neural network models for prediction

[0252] Based on the trained neural network model, the full-bridge temperature prediction was performed at 12:00 on the second day, given that the local ambient temperature was 34.5°C.

[0253] The position coordinates X′ and solar radiation intensity I at this time point s ' olar 、Ambient temperature T e ' nv , wind speed V m ' ind , relative humidity RH′ and historical temperature data T h ' istory After normalization, the data are shown in Table 5 below:

[0254] Table 5 Data parameters (partial)

[0255]

[0256]

[0257] The processed data set is input into the trained neural network model to obtain the final temperature prediction output, and finally a three-dimensional cloud map is used to visualize the combined Figure 2 shown.

[0258] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other forms of assembly without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered in all respects as exemplary and non-restrictive, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations within the meaning and range of equivalents of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0259] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A method for intelligently constructing the global time-varying temperature field of a long-span concrete bridge structure, characterized by: The following steps are involved: Step 1: Calculate the bridge location coordinates The bridge is discretized into N equal intervals along the length direction L, width direction W and vertical direction Z, and the position coordinates of a certain point are expressed as: Then the position coordinates of all points are expressed as: X=(X0 X1…X N-1 X N ) T Among them, for a certain element X i i=0,1,...,N-1,N, and its form is: Step 2: Calculate solar radiation parameters Calculate the solar radiation intensity I at time t solar , considering the solar radiation received by the entire bridge area, the solar radiation intensity I is obtained solar According to the division of position coordinates, it is expressed as: I solar =(I0 I1…I N-1 I N ) T Among them, for a certain element I i , which is expressed in the form of: Step 3: Collect measured environment data Collect environmental data at different bridge locations at time t, including the ambient temperature T env , wind speed V mind , relative humidity RH and historical temperature data T history ,in: Ambient temperature T env It is expressed as: For an element Its form is: Wind speed V mind Expressed as: V mind =(V0 V1…V N-1 V N ) T For an element V i , which is expressed in the form of: Relative humidity RH is expressed as: RH = (R0 R1…R N-1 R N ) T For an element R i , which is expressed in the form of: Historical temperature data T history It is expressed as: For an element Its form is: Step 4: Build the dataset The position coordinate X and the corresponding characteristic data, including the solar radiation intensity I solar 、Ambient temperature T env , wind speed V mind , relative humidity RH and historical temperature data T history The data set is constructed and normalized to limit its range to the bridge size. Data outside this range are marked as outliers and deleted. The normalized data set is expressed as follows: G=[X I solar T env V mind RH T history ]; Step 5: Constructing a physical constraint loss function The heat conduction equation is defined to describe the time and space variation of the internal temperature of the bridge when only concrete is considered, and it is converted into heat conduction constraint loss, which is expressed as: Where N' is the number of data samples, ρ is the material density, c is the specific heat capacity, is the predicted temperature value, is the time derivative of temperature, κ is the thermal conductivity, is the spatial derivative of temperature, x, y, and z correspond to the length direction L, width direction W, and vertical direction Z of the bridge discretization, respectively; Convective heat transfer is defined to describe the heat exchange boundary conditions between the bridge surface and the air. The physical process is described by Newton's cooling law, and then the convection heat transfer constraint loss is constructed, which is expressed as: Where h is the convective heat transfer coefficient, q i is the heat flow at the corresponding point, and T surface is the temperature distribution on the bridge surface, i.e. the temperature prediction value n is the normal direction of the bridge surface; Then, the physical constraint loss function is expressed as: L physics =Loss conduct +Loss conv ; Step 6: Constructing a data-constrained loss function Construct a data constraint loss to measure the gap between the predicted temperature and the actual temperature, and use the mean square error to represent the data constraint loss function, which is expressed as: Step 7: Build a neural network model Divide the normalized data set G into training set G T and the test set G E ; Define the input target as the training set G T Excluding historical temperature data T history Other data outside Define the output target as the final predicted value The historical temperature data T history As validation data; Use the long short-term memory network to build a neural network model and output the final prediction value It is expressed as: Among them, the temperature prediction value for a certain element Its form is: Step 8: Constructing physical data dual constraints In the process of neural network model training, weight coefficients are configured for physical constraints and data constraints respectively, and the sum of the weight coefficients is set to 1. The loss function expression of physical data dual constraints is as follows: L total =λ1Loss data +λ2L physics In the formula, λ1 is the weight coefficient of data constraint loss, and λ2 is the weight coefficient of physical constraint loss; Use the gradient optimization algorithm to update to determine the best weights, and repeat the update on the entire normalized data set G until the loss converges to obtain the temperature prediction result. It is expressed as: Among them, for an element Its form is: Step 9: Neural Network Model Evaluation Using the test set G E To measure the capability of the neural network model, the mean square error is used to evaluate the neural network model. If the neural network model meets the mean square error condition during training, the training is completed. If not, back propagation is continued until the condition is met to obtain a trained neural network model. Step 10: Use the neural network model to make predictions The temperature of the entire bridge at time point t′ is predicted based on the actual bridge working condition information, and the position coordinate X′, solar radiation intensity I s ' olar 、Ambient temperature T e ' nv , wind speed V m ' ind , relative humidity RH′ and historical temperature data T h ' istory After normalization, the data set G′ is obtained. The data set G′ is input into the trained neural network model to obtain the final temperature prediction output.

2. According to claim 1, a method for intelligently constructing a global time-varying temperature field of a long-span concrete bridge structure is characterized by: In the step 2, the solar radiation intensity I solar For direct radiation I dir , Scattered Radiation I diff and surface reflected radiation I ref The sum of , where: In the formula, t u is the Link turbidity coefficient, k a is the relative atmospheric pressure, m is the light mass, I0 is the solar constant, φ is the angle between the illuminated surface and the incident light, P is the atmospheric transparency coefficient, β s is the solar altitude angle, β n is the angle between the external normal of the illuminated surface and the ground, r e is the ground reflectivity.

3. The method for intelligently constructing the global time-varying temperature field of a long-span concrete bridge structure according to claim 1 is characterized by: In step 7, a long short-term memory network is used to model time series data, and W is defined f , W i , W c and W o are the input weight matrices of the forget gate, input gate, memory unit update, and output gate, respectively, and b f 、b i 、b c and b o are the bias terms of the forget gate, input gate, memory unit update and output gate respectively, then: Input Gate: i(t)=σ(W i [h(t-1),X(t)]+b i ) Forget Gate: f(t)=σ(W f [h(t-1),X(t)]+b f ) Output Gate: o(t)=σ(W o [h(t-1),X(t)]+b o ) Memory unit update: Hide status updates: h(t)=o(t)*tanh(C(t)) Output prediction: The output h(t) of LSTM can be used to obtain the final prediction value through the fully connected layer. Where σ(·) is the sigmoid function, tanh(·) is the hyperbolic tangent function, and W out is the weight of the output layer, b out is the bias term of the output layer.

4. The method for intelligently constructing the global time-varying temperature field of a long-span concrete bridge structure according to claim 1 is characterized by: In step nine, during the training of the neural network model, the mean square error (MSE) is defined to evaluate the performance of the neural network model every 10 epochs, as shown below: Where N valid is the number of samples in the test set, is the loss of the ith sample.

5. The method for intelligently constructing the global time-varying temperature field of a long-span concrete bridge structure according to claim 1 is characterized by: In step 10, the final temperature prediction output is The data is displayed through a three-dimensional cloud map.

Images