Construction of Spatiotemporal Correlation Mechanism and Linear Control Method for Long-Span Steel Truss Arch Bridges

Abstract

The construction of the spatio-temporal correlation mechanism and the linear control method for long-span steel truss arch bridges involve the field of bridge construction control technology. Collect the environmental temperature, component quality, and actual elevation after beam dropping in the first n hoisting stages; calculate the theoretical elevation before beam dropping, the theoretical elevation after beam dropping, and the error adjustment variable in the first n hoisting stages; define intervals, and normalize the data to form a data set; divide the data set into a training set and a test set; use the training set to construct a spatio-temporal correlation mechanism model; use the test set to test the effectiveness of the model; use the model to predict the error adjustment variables of l control points in the (n + 1)-th hoisting stage; correct the theoretical elevation before beam dropping in the (n + 1)-th hoisting stage to achieve linear control. It can calculate the theoretical elevation considering the different influence weights of various spatio-temporal factors, making up for the disadvantage of traditional construction prediction methods that cannot consider the adverse effects of environmental temperature, component quality, and construction cumulative error on the weights.

Description

Technical Field

[0001] The present invention relates to the technical field of bridge construction control, and particularly to a construction spatio-temporal correlation mechanism construction and alignment control method for long-span steel truss arch bridges. Background Technique

[0002] The steel truss arch bridge is composed of two structural systems, namely, the truss and the arch, and combines the force-bearing characteristics of the truss and the arch. The truss part bears axial force, and the horizontal thrust of the arch reduces the mid-span bending moment, so that the solid web section in the mid-span mainly bears axial pressure under the action of the dead load. It combines the favorable factors of the truss and the arch, mainly bears axial force, is conducive to saving materials, and compared with the same-span beam bridge of other structures, the steel truss arch bridge saves more steel and has good adaptability.

[0003] The existing construction control prediction methods for steel truss arch bridges mainly include the grey system theory method and the Kalman filtering method. The grey system theory method can grey-process various factors affecting the bridge construction state, collect these influencing factors and centrally feedback them into the target vector, and finally complete the entire prediction process through the target vector; the Kalman filtering method is to extract the true signal from the signal that has been contaminated by noise and estimate the true state of the system. These two methods can be well applied in the construction of cable-stayed bridges. However, for steel truss arch bridges, due to the influence of environmental temperature and component quality on construction errors, which need to be considered emphatically, due to the limitations of their own methods, these two methods cannot consider the influence weights of various factors.

[0004] In view of the above reasons, on the basis of the traditional long short-term memory network, the present invention establishes a spatio-temporal correlation mechanism for the hoisting construction of steel truss arch bridges, which can consider the influence of environmental temperature, component quality and construction cumulative error while considering the influence of historical hoisting errors, avoiding the gradient explosion and gradient disappearance of other neural network models, combining the Dropout method and the self-attention mechanism to prevent model overfitting, increasing the influence weight of the hoisting error in the adjacent time, and improving the accuracy of model prediction. Summary of the Invention

[0005] To solve the deficiencies in the background technique and aiming at the theoretical elevation calculation problem in the construction control of long-span steel truss arch bridges, the present invention provides a construction spatio-temporal correlation mechanism construction and alignment control method for long-span steel truss arch bridges, which can calculate the theoretical elevation considering the different influence weights of various spatio-temporal factors, making up for the adverse effects of the traditional construction prediction method that cannot consider the environmental temperature, component quality and construction cumulative error on the weight.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A construction spatio-temporal correlation mechanism construction and alignment control method for long-span steel truss arch bridges, including the following steps:

[0007] Step 1: Collect measured data

[0008] Define the hoisting stages \(i = 1, 2, \cdots, n - 1, n\). For each hoisting stage, there are control points \(j = 1, 2, \cdots, l - 1, l\). Collect the ambient temperature \(T\), the component mass \(M\) for the first \(n\) hoisting stages, and the actual elevations \(f\) of the \(n\times l\) control points after the girder is lowered 3 , and the forms of each data are as follows:

[0009] \(T=(t_1,t_2,\cdots,t\) n-1 ,t n ) T , \(M=(m_1,m_2,\cdots,m\) n-1 ,m n ) Τ ,

[0010] In the formula, \(t\) i and \(m\) i respectively represent the ambient temperature and the component mass of the \(i\)-th hoisting stage, represents the actual elevation of the \(j\)-th control point of the \(i\)-th hoisting stage after the girder is lowered;

[0011] Step 2: Calculate theoretical data

[0012] Calculate the theoretical elevation \(f\) 1 before the girder is lowered and the theoretical elevation \(f\) 2 after the girder is lowered, as well as the error adjustment variable \(\delta\) for the \(n\times l\) control points in the first \(n\) hoisting stages. The forms of each data are as follows:

[0013]

[0014] In the formula, and \(\delta\) i,j respectively represent the theoretical elevation before the girder is lowered, the theoretical elevation after the girder is lowered, and the error adjustment variable of the \(j\)-th control point of the \(i\)-th hoisting stage;

[0015] Step 3: Data processing to form a data set

[0016] Define the interval \([x\) min ,x max , where:

[0017] x max =Q3 + 2×IQR, \(x\) min =Q1 - 2×IQR

[0018] In the formula, IQR represents the interquartile range, \(IQR = Q3 - Q1\), \(Q3\) is the third quartile, and \(Q1\) is the first quartile;

[0019] Mark the data outside the interval as outliers and delete them. Process them using the mean filling method, and then normalize the data. The normalized dataset is as follows:

[0020]

[0021] In the formula, and are the normalized theoretical elevation before beam drop, environmental temperature, component mass, and error adjustment variable respectively;

[0022] Step 4: Divide the dataset

[0023] Perform time series division on the normalized dataset X sd to obtain the training set X T and the test set X E . Define the hoisting stage in the training set as i1 = 1, 2,..., k - 1, k, then:

[0024]

[0025] In the formula, and are the normalized theoretical elevation before beam drop, environmental temperature, component mass, and error adjustment variable in the training set respectively;

[0026] Define the hoisting stage in the test set as i2 = k + 1, k + 2,..., n - 1, n, then:

[0027]

[0028] In the formula, and are the normalized theoretical elevation before beam drop, environmental temperature, component mass, and error adjustment variable in the test set respectively;

[0029] Step 5: Use the training set X T to construct a spatio-temporal correlation mechanism model

[0030] Specify the T in the training set X as the input data, and the T in the training set X as the output data. Establish a Dropout layer to perform regularization processing on the input data:

[0031]

[0032] In the formula, ο is the Hadamard product, and D is a random vector drawn from the Bernoulli distribution;

[0033] Build a long short-term memory layer, and define the input weight matrices of the forget gate, input gate, cell candidate, and output gate as W f , W i , W c , W o , and the hidden weight matrices as U f , U i , U c , U o , and the bias terms as b f , b i , b c , b o . Taking X i' as the input of the long short-term memory layer at time step i', we have:

[0034] Forget gate:

[0035] F i' = σ(W f X i' + U f H i'-1 + b f )

[0036] Input gate:

[0037] I i' = σ(W i X i' + U i H i'-1 + b i )

[0038] Cell candidate:

[0039]

[0040] Cell state:

[0041]

[0042] Output gate:

[0043] O i' = σ(W o X i' + U o H i'-1 + b o )

[0044] Hidden state:

[0045] H i' = O i' · tanh(C i' )

[0046] where, σ(·) is the sigmoid function, tanh(·) is the hyperbolic tangent function, H i'-1 is the hidden state of the long short-term memory layer at the time step i'-1, C i'-1 is the cell state of the long short-term memory layer at the time step i'-1;

[0047] Build a self-attention mechanism layer to calculate the attention weights:

[0048]

[0049] Apply it to the hidden state of the long short-term memory layer, then there is the hidden state processed by the self-attention mechanism:

[0050]

[0051] where, softmax(·) is the activation function, is the weight vector, w a is the weight matrix, b a is the bias term. Use and to calculate the gradient of the loss function, and update W f 、W i 、W c 、W o and U f 、U i 、U c 、U o as well as b f 、b i 、b c 、b o , and perform iterations until the gradient is 0 to complete the construction and training of the spatio-temporal correlation mechanism model;

[0052] Step Six: Use the test set X E to verify the effectiveness of the model

[0053] Input the test set X E into the spatio-temporal correlation mechanism model to output the predicted value of the normalized error adjustment variable Calculate its root mean square error with . If the root mean square error meets the required accuracy requirements, go to Step Seven, otherwise return to Step Five to update the parameters of the spatio-temporal correlation mechanism model until the root mean square error meets the required accuracy requirements;

[0054] Step Seven: Use the model to predict the error adjustment variables of l control points in the (n + 1)-th hoisting stage

[0055] ​

[0056]

[0057] In the formula, is the theoretical elevation before beam lowering normalized for the l control points in the (n + 1)-th hoisting stage, and are the normalized ambient temperature and component mass in the (n + 1)-th hoisting stage, respectively;

[0058] Input into the spatio-temporal correlation mechanism model, output the predicted value of the normalized error adjustment variable for the l control points in the (n + 1)-th hoisting stage, and perform inverse normalization on it to obtain the predicted error adjustment variable δ n+1,j ;

[0059] Step Eight: Modify the theoretical elevation before beam lowering in the (n + 1)-th hoisting stage to achieve linear control

[0060] Modify and calculate the theoretical elevation before beam lowering for the l control points in the (n + 1)-th hoisting stage according to the following formula:

[0061]

[0062] In the formula, is the theoretical elevation before beam lowering for the j-th control point in the (n + 1)-th hoisting stage, and use the corrected value of the theoretical elevation before beam lowering to replace the theoretical elevation before beam lowering

[0063] Furthermore, and δ i,j in Step Two are calculated as follows:

[0064]

[0065]

[0066]

[0067] In the formula, and are the designed elevation, designed pre-camber, pre-cast height, and theoretical elastic deformation of the j-th control point in the i-th hoisting stage, respectively.

[0068] Furthermore, in Step Four, the normalized dataset X sd is divided into time series in a ratio of 4:1, and the first 4 / 5 of the divided data is used as the training set X T , and the last 1 / 5 of the data is used as the test set X E .

[0069] Further, the inverse normalization process in Step 7 is expressed as follows:

[0070]

[0071] In the formula, δ n+1,j is the error adjustment variable predicted for the j-th control point in the (n + 1)-th hoisting stage, is the predicted value of the normalized error adjustment variable for the j-th control point in the (n + 1)-th hoisting stage, δ max is the maximum value of the error adjustment variable in the dataset, δ min is the minimum value of the error adjustment variable in the dataset.

[0072] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention is applicable to the alignment control of long-span steel truss arch bridges in hoisting construction, and can calculate the theoretical elevation considering the influence weights of various spatio-temporal factors. It makes up for the disadvantages of traditional construction prediction methods that cannot consider the adverse effects of environmental temperature, component quality, and construction cumulative error on the weights, and can effectively improve the hoisting construction accuracy, having the following advantages:

[0073] 1. The constructed spatio-temporal correlation mechanism model only needs to collect basic parameters such as theoretical elevation, environmental temperature, and component quality, without the need for complex finite element modeling and nonlinear analysis;

[0074] 2. The constructed spatio-temporal correlation mechanism model can consider the influence weights of different factors and the influence of historical construction errors at the same time, and can effectively improve the hoisting construction accuracy, making the alignment in the completed bridge stage closer to the ideal state;

[0075] 3. The constructed spatio-temporal correlation mechanism model can focus on considering the influence of environmental temperature, thereby accurately predicting the error adjustment variable at a specified environmental temperature and proposing a more reasonable theoretical elevation. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 is the flow chart of the method of the present invention;

[0077] Figure 2 is the comparison chart of the predicted value and the true value of the test set data in the embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0078] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0079] As Figure 1As shown in the figure, the construction of the spatio-temporal correlation mechanism and the linear control method for long-span steel truss arch bridges include the following steps:

[0080] Step 1: Collect measured data

[0081] Define the hoisting stage \(i = 1, 2, \cdots, n - 1, n\), and there are control points \(j = 1, 2, \cdots, l - 1, l\) in each hoisting stage. Collect the ambient temperature \(T\), component mass \(M\) in the first \(n\) hoisting stages, and the actual elevation \(f\) of \(n\times l\) control points after the beam drops 3 , and the forms of each data are as follows:

[0082] \(T=(t_1,t_2,\cdots,t\) n-1 ,t\) n ) T , \(M=(m_1,m_2,\cdots,m\) n-1 ,m\) n ) Τ ,

[0083] In the formula, \(t\) i represents the ambient temperature of the \(i\)-th hoisting stage, \(m\) i represents the component mass of the \(i\)-th hoisting stage, represents the actual elevation of the \(j\)-th control point after the beam drops in the \(i\)-th hoisting stage;

[0084] Step 2: Calculate theoretical data

[0085] Calculate the theoretical elevation \(f\) of \(n\times l\) control points before the beam drops in the first \(n\) hoisting stages, 1 the theoretical elevation \(f\) after the beam drops, 2 and the error adjustment variable \(\delta\). The forms of each data are as follows:

[0086]

[0087] In the formula, represents the theoretical elevation of the \(j\)-th control point before the beam drops in the \(i\)-th hoisting stage, represents the theoretical elevation of the \(j\)-th control point after the beam drops in the \(i\)-th hoisting stage, \(\delta\) i,j represents the error adjustment variable of the \(j\)-th control point in the \(i\)-th hoisting stage, and the calculation is as follows:

[0088]

[0089]

[0090]

[0091] In the formula, is the designed elevation of the \(j\)-th control point in the \(i\)-th hoisting stage, is the designed pre-camber of the j-th control point in the i-th hoisting stage, is the pre-camber of the j-th control point in the i-th hoisting stage, is the theoretical elastic deformation of the j-th control point in the i-th hoisting stage;

[0092] Step 3: Data processing to form a data set

[0093] Define the interval [x min , x max , where:

[0094] x max = Q3 + 2 × IQR, x min = Q1 - 2 × IQR

[0095] In the formula, IQR represents the interquartile range, IQR = Q3 - Q1, Q3 is the third quartile, and Q1 is the first quartile.

[0096] Mark the data outside the interval as outliers and delete them, process them using the mean filling method, and then normalize the data as follows:

[0097]

[0098] In the formula, x sd represents the normalized data, x represents the original data, including the theoretical elevation before beam settlement ambient temperature t i , component mass m i and error adjustment variable δ i,j .

[0099] Then the data set after normalization is as follows:

[0100]

[0101] In the formula, is the normalized theoretical elevation before beam settlement, is the normalized ambient temperature, is the normalized component mass, is the normalized error adjustment variable, all of which are solved through the x sd calculation formula;

[0102] Step 4: Divide the data set

[0103] Divide the data set X sd after normalization into time series in a ratio of 4:1. The first 4 / 5 of the divided data is used as the training set X T , and the last 1 / 5 of the data is used as the test set X E, of course, it is also feasible to divide with other ratios, and the division ratio can be selected according to the actual situation. Define the hoisting stage \(i_1 = 1, 2, \cdots, k - 1, k\) in the training set, then:

[0104]

[0105] In the formula, is the normalized theoretical elevation before beam lowering in the training set, is the normalized environmental temperature in the training set, is the normalized component mass in the training set, is the normalized error adjustment variable in the training set.

[0106] Define the hoisting stage \(i_2 = k + 1, k + 2, \cdots, n - 1, n\) in the test set, then:

[0107]

[0108] In the formula, is the normalized theoretical elevation before beam lowering in the test set, is the normalized environmental temperature in the test set, is the normalized component mass in the test set, is the normalized error adjustment variable in the test set;

[0109] Step Five: Use the training set \(X\) T to construct a spatio-temporal correlation mechanism model

[0110] Specify T in the training set \(X\) as the input data, and T in the training set \(X\) as the output data, and establish a Dropout layer to perform regularization processing on the input data:

[0111]

[0112] In the formula, \(\circ\) is the Hadamard product, and \(D\) is a random vector drawn from the Bernoulli distribution, whose value is 0 or 1.

[0113] Establish a long short-term memory layer, and define \(W\) f , \(W\) i , \(W\) c , \(W\) o as the input weight matrices of the forget gate, input gate, cell candidate, and output gate respectively, and \(U\) f , \(U\) i , \(U\) c , \(U\) o as the hidden weight matrices of the forget gate, input gate, cell candidate, and output gate respectively, and \(b\) f , \(b\)i , b c , b o are the bias terms of the forget gate, input gate, cell candidate, and output gate respectively. Taking X i' as the input at the i'-th time step of the long short-term memory layer, we have:

[0114] Forget gate:

[0115] F i' = σ(W f X i' + U f H i'-1 + b f )

[0116] Input gate:

[0117] I i' = σ(W i X i' + U i H i'-1 + b i )

[0118] Cell candidate:

[0119]

[0120] Cell state:

[0121]

[0122] Output gate:

[0123] O i' = σ(W o X i' + U o H i'-1 + b o )

[0124] Hidden state:

[0125] H i' = O i' · tanh(C i' )

[0126] In the formula, σ(·) is the sigmoid function, tanh(·) is the hyperbolic tangent function, H i'-1 is the hidden state at the (i'-1)-th time step of the long short-term memory layer, and its dimension is determined by the number of cells in the long short-term memory layer. C i'-1 is the cell state at the (i'-1)-th time step of the long short-term memory layer.

[0127] Build a self-attention mechanism layer to calculate the attention weights:

[0128]

[0129] Applying it to the hidden state of the long short-term memory layer, there is a hidden state processed by the self-attention mechanism:

[0130]

[0131] In the formula, softmax(·) is the activation function, is the weight vector, w a is the weight matrix, b a is the bias term.

[0132] Using and to calculate the gradient of the loss function, and update W f , W i , W c , W o and U f , U i , U c , U o as well as b f , b i , b c , b o , and perform iteration until the gradient is 0 to complete the construction and training of the spatio-temporal correlation mechanism model;

[0133] Step Six: Use the test set X E to verify the effectiveness of the model

[0134] Input the test set X E into the spatio-temporal correlation mechanism model, and output the predicted value of the normalized error adjustment variable Use the following formula to calculate its root mean square error with :

[0135]

[0136] If the root mean square error MSE meets the required accuracy requirements, enter Step Seven; otherwise, return to Step Five to update the parameters of the spatio-temporal correlation mechanism model until the root mean square error MSE meets the required accuracy requirements;

[0137] Step Seven: Use the model to predict the error adjustment variables of l control points in the (n + 1)-th hoisting stage

[0138] Normalize the theoretical elevation, ambient temperature, and component mass of l control points before beam lowering in the (n + 1)-th hoisting stage as follows:

[0139]

[0140] In the formula, is the theoretical elevation before girder lowering normalized for l control points in the (n + 1)-th hoisting stage, is the normalized environmental temperature in the (n + 1)-th hoisting stage, is the normalized component mass in the (n + 1)-th hoisting stage, all obtained through the x sd calculation formula.

[0141] Input into the spatio-temporal correlation mechanism model, and output the predicted value of the normalized error adjustment variable for l control points in the (n + 1)-th hoisting stage and perform inverse normalization on it to obtain the predicted error adjustment variable, expressed as follows: [[ID=1"15]]

[0142]

[0143] In the formula, δ n+1,j is the predicted error adjustment variable for the j-th control point in the (n + 1)-th hoisting stage, is the predicted value of the normalized error adjustment variable for the j-th control point in the (n + 1)-th hoisting stage, δ max is the maximum value of the error adjustment variable in the dataset, δ min is the minimum value of the error adjustment variable in the dataset;

[0144] Step Eight: Modify the theoretical elevation before girder lowering in the (n + 1)-th hoisting stage to achieve linear control

[0145] Modify and calculate the theoretical elevation before girder lowering for l control points in the (n + 1)-th hoisting stage according to the following formula:

[0146]

[0147] In the formula, is the theoretical elevation before girder lowering for the j-th control point in the (n + 1)-th hoisting stage, and use the corrected value of the theoretical elevation before girder lowering to replace the theoretical elevation before girder lowering to achieve linear control.

[0148] Example

[0149] Taking the large-span steel truss arch bridge over the Chaobai River on the Changtong Road in Beijing as an example, predict the error adjustment variable and use the corrected theoretical elevation before girder lowering for hoisting construction.

[0150] ① Collect the environmental temperature T, component mass M in the first 50 hoisting stages and the actual elevation f after girder lowering for 4 control points in each stage 3 , and the collected measured data is shown in Table 1:

[0151] Table 1 Collected Measured Data

[0152]

[0153]

[0154] ② Calculate the theoretical elevation f before the girder drop in the first 50 hoisting stages 1 、the theoretical elevation f after the girder drop 2 and the error adjustment variable δ. Some calculated theoretical data are shown in Table 2 as follows:

[0155] Table 2 Calculated Theoretical Data (Partial)

[0156] Hoisting stage 1 2 3 4 5 <![CDATA[f 1 n,1 > 32.403 32.369 32.262 32.343 32.387 <![CDATA[f 1 n,2 > 32.286 32.504 32.376 32.484 32.267 <![CDATA[f 1 n,3 > 32.382 32.351 32.271 32.359 32.389 <![CDATA[f 1 n,4 > 32.271 32.484 32.383 32.455 32.274 <![CDATA[f 2 n,1 > 32.383 32.355 32.252 32.335 32.384 <![CDATA[f 2 n,2 > 32.273 32.488 32.371 32.482 32.259 <![CDATA[f 2 n,3 > 32.376 32.334 32.262 32.354 32.388 <![CDATA[f 2 n,4 > 32.265 32.484 32.370 32.439 32.258 <![CDATA[δ n,1 > -0.085 -0.003 -0.224 32.335 32.384 <![CDATA[δ n,2 > -0.001 0.114 0.090 32.482 32.259 <![CDATA[δ n,3 > -0.110 -0.032 -0.221 32.354 32.388 <![CDATA[δ n,4 > -0.126 0.218 0.002 32.439 32.258

[0157] ③ Normalize the data, and some of the normalized data sets are shown in Table 3 as follows:

[0158] Table 3 Normalized Data Set (Partial)

[0159] phase 1 2 3 4 <![CDATA[f 1 n,1 > 0.115196 0.087418 0 0.066176 <![CDATA[f 1 n,2 > 0.015807 0.197171 0.090682 0.180532 <![CDATA[f 1 n,3 > 0.102024 0.076923 0.012146 0.083401 <![CDATA[f 1 n,4 > 0.204502 0.341479 0.276527 0.32283 T 0.055556 0 0 0 M 1 0.454553 0.984466 0.503854 <![CDATA[f 2 n,1 > 0.11156 0.088925 0.005659 0.072757 <![CDATA[f 2 n,2 > 0.011647 0.190516 0.093178 0.185524 <![CDATA[f 2 n,3 > 0.100324 0.066343 0.008091 0.082524 <![CDATA[f 2 n,4 > 0.203215 0.344051 0.27074 0.315113 <![CDATA[δ n,1 > 0.522727 0.363636 0.386364 0.204545 <![CDATA[δ n,2 > 0.378378 0.702703 0.486486 1 <![CDATA[δ n,3 > 0.651163 0.069767 0.906977 0.906977 <![CDATA[δ n,4 > 0.447368 0.921053 0.578947 0

[0160] ④ Divide the normalized data set X sd into time series. The first 4 / 5 of the data is used as the training set X T , and the last 1 / 5 of the data is used as the test set X E . Then, the hoisting stage in the training set is i1 = 1, 2,..., 40, and the hoisting stage in the test set is i2 = 41, 42,..., 50. There are control points j = 1, 2, 3, 4 in each hoisting stage. The training set X T and the test set X E are respectively expressed as follows:

[0161]

[0162]

[0163] ⑤ Call the tensorflow open-source library to build a spatio-temporal correlation mechanism model, and select an optimizer and a loss function:

[0164] Create an input layer, and specify the input shape of the model as [X.shape[1], X.shape[2]], where X.shape[1] is the number of time steps and X.shape[2] is the number of features at each time step;

[0165] Create a Dropout layer and specify the loss parameter of this layer;

[0166] Create a long short-term memory layer, specify the number of neurons, and use the output of the Dropout layer as the input of this layer;

[0167] Add a self-attention mechanism layer, take the input of the long short-term memory layer as the output of this layer, and convert the output into two-dimensional data;

[0168] Create an output layer, which will take the input two-dimensional data and output the data of 4 prediction targets;

[0169] Specify the time step as 1, the number of loops as 40, and the number of training samples as 2, and train this model.

[0170] ⑥ Call the trained spatio-temporal correlation mechanism model to predict the test set data, and the results are combined Figure 2 as shown.

[0171] ⑦ Use the spatio-temporal correlation mechanism model to predict the error adjustment variables of 4 control points in the 51st hoisting stage. The results are shown in Table 4:

[0172] Table 4 Predicted error adjustment variables

[0173] phase <![CDATA[δ 51,1 > <![CDATA[δ 51,2 > <![CDATA[δ 51,3 > <![CDATA[δ 51,4 > 51 0.008 -0.002 -0.020 -0.008

[0174] ⑧ Calculate the theoretical elevation correction value before beam lowering, use the correction value to replace the original theoretical elevation before beam lowering to achieve alignment control. The correction value results are shown in Table 5:

[0175] Table 5 Theoretical elevation correction value before beam lowering

[0176]

[0177] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent conditions of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claims involved.

[0178] In addition, it should be understood that although this specification is described according to embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. Construction spatio-temporal correlation mechanism and linear control method for long-span steel truss arch bridges, characterized in that: The following steps are involved: Step 1: Collect measured data Define the hoisting stages \(i = 1, 2,\cdots, n - 1, n\). For each hoisting stage, there are control points \(j = 1, 2,\cdots, l - 1, l\). Collect the ambient temperature \(T\), the component mass \(M\) in the first \(n\) hoisting stages, and the actual elevations \(f\) of the \(n\times l\) control points after the beam is lowered 3 , and the forms of the data are as follows: T=(t1,t2…,t n-1 ,t n ) T ,M=(m1,m2…,m n-1 ,m n ) Τ , Where t i and m i respectively represent the ambient temperature and component mass at the i-th hoisting stage, represents the actual elevation after the girder drops at the j-th control point in the i-th hoisting stage; Step 2: Calculate theoretical data Calculate the theoretical elevations \(f\) of \(n\times l\) control points before the girder lowering for the first \(n\) hoisting stages 1 and the theoretical elevations \(f\) after the girder lowering 2 as well as the error adjustment variable \(\delta\). The forms of each data are as follows: In the formula, and δ i,j respectively represent the theoretical elevation before beam lowering, the theoretical elevation after beam lowering, and the error adjustment variable of the j-th control point in the i-th hoisting stage; Step 3: Data processing to form a data set Define the interval [x min , x max , where: x max = Q3 + 2 × IQR, x min = Q1 - 2 × IQR Where, IQR represents the interquartile range, IQR = Q3-Q1, Q3 is the third quartile, Q1 is the first quartile; The data outside the range are marked as outliers and deleted, and the mean filling method is used to process the data. Then the data is normalized. The normalized data set is as follows: In the formula, and are respectively the theoretical elevation before the girder drop, the ambient temperature, the component mass, and the error adjustment variable after normalization; Step 4: Divide the dataset The normalized dataset X sd is divided into a training set X T and a test set X E . Define the lifting stages in the training set as i1 = 1, 2,..., k - 1, k, then: In the formula, and are respectively the normalized theoretical elevation before beam drop, ambient temperature, component mass and error adjustment variable in the training set; Define the hoisting stage i2=k+1,k+2,...,n-1,n in the test set, then: In the formula, and are respectively the theoretical elevation before the girder fall, the environmental temperature, the component mass, and the error adjustment variable normalized in the test set; Step 5: Use the training set X T Construct a spatio-temporal correlation mechanism model Specify the training set X T in is the input data, and the training set X T in is the output data. A Dropout layer is established to regularize the input data as follows: Where ο is the Hadamard product, D is a random vector drawn from the Bernoulli distribution; Build a long short-term memory layer, and define the input weight matrices of the forget gate, input gate, cell candidate, and output gate as W f 、W i 、W c 、W o respectively, and the hidden weight matrices as U f 、U i 、U c 、U o respectively, and the bias terms as b f 、b i 、b c 、b o respectively. Taking X i' as the input at the i'-th time step of the long short-term memory layer, we have: Forget Gate: F i' = σ(W f X i' + U f H i'-1 + b f ) Input Gate: I i' = σ(W i X i' + U i H i'-1 + b i ) Unit Candidates: Unit Status: Output Gate: O i' = σ(W o X i' + U o H i'-1 + b o ) Hidden state: H i' =O i' ·tanh(C i' ) wherein, σ(·) is the sigmoid function, tanh(·) is the hyperbolic tangent function, and H i'-1 is the hidden state of the long short-term memory layer at the time step i'-1, and C i'-1 is the cell state of the long short-term memory layer at the time step i'-1; Establish a self-attention mechanism layer and calculate the attention weight: Applying this to the hidden state of the long short-term memory layer, we get the hidden state after the self-attention mechanism: Where softmax(·) is the activation function, is the weight vector, w a is the weight matrix, b a As the bias term, use and Calculate the gradient of the loss function and update W through the backpropagation algorithm f , W i , W c , W o and U f , U i , U c , U o and b f , b i , b c , b o , iterate until the gradient is 0, completing the construction and training of the spatiotemporal correlation mechanism model; Step 6: Use the test set X E Test model validity The test set X E Input the spatiotemporal correlation mechanism model and output the predicted value of the normalized error adjustment variable Calculate its If the root mean square error meets the required accuracy requirements, proceed to step seven, otherwise return to step five to update the spatiotemporal correlation mechanism model parameters until the root mean square error meets the required accuracy requirements; Step 7: Use the model to predict the error adjustment variable of the l control point in the n+1th lifting stage The theoretical elevation, ambient temperature and component mass before the beam is dropped at the l control point in the n+1th hoisting stage are normalized as follows: In the formula, is the normalized theoretical elevation of the l control point before beam drop in the n+1th hoisting stage, and are the normalized ambient temperature and component mass at the n+1th lifting stage, respectively; Input into the spatio-temporal correlation mechanism model, output the predicted values of the normalized error adjustment variables of the l control points in the (n + 1)-th hoisting stage, and perform inverse normalization on them to obtain the predicted error adjustment variable δ n+1,j ; Step 8: Correct the theoretical elevation before the beam is dropped in the n+1th hoisting stage to achieve linear control The theoretical elevation of the l control point before the beam is dropped in the n+1th hoisting stage is corrected and calculated according to the following formula: In the formula, is the theoretical elevation before the girder settlement of the j-th control point in the (n + 1)-th hoisting stage, and the theoretical elevation correction value before the girder settlement is used to replace the theoretical elevation before the girder settlement 2. The construction spatio-temporal correlation mechanism construction and alignment control method of the long-span steel truss arch bridge according to claim 1, characterized in that: In the second step, and δ i,j are calculated as follows: Wherein, and are respectively the design elevation, design camber, pre-camber and theoretical elastic deformation of the j-th control point in the i-th hoisting stage.

3. The construction time-space correlation mechanism construction and alignment control method of the long-span steel truss arch bridge according to claim 1, characterized in that: In the fourth step, the normalized dataset X sd is divided into time series in a ratio of 4:

1. The first 4 / 5 of the divided data is used as the training set X T , and the last 1 / 5 of the data is used as the test set X E .

4. The construction spatio-temporal correlation mechanism construction and alignment control method for long-span steel truss arch bridges according to claim 1, characterized in that: The inverse normalization process in step seven is expressed as follows: where δ n+1,j is the error adjustment variable predicted for the j-th control point in the (n + 1)-th hoisting stage, is the predicted value of the normalized error adjustment variable for the j-th control point in the (n + 1)-th hoisting stage, δ max is the maximum value of the error adjustment variable in the dataset, and δ min is the minimum value of the error adjustment variable in the dataset.

Images