An intelligent decision-making method for TBM tunnel support based on multi-source fusion data

Abstract

The present invention discloses an intelligent decision-making method for TBM tunnel support based on multi-source fusion data, which relates to the technical field of tunnel construction. The key points of its technical solution are as follows: This method widely selects various influencing factors closely related to the support situation and surrounding rock situation, such as geological exploration data, rock mass information, TBM tunneling data, etc. Based on the support parameters actually implemented on-site, it deeply integrates various algorithms, thereby realizing the intelligent decision-making of TBM tunnel support parameters and improving the intelligent level of TBM tunnels.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel construction, and more specifically, it relates to an intelligent decision-making method for TBM tunnel support based on multi-source fusion data. Background Art

[0002] As a key equipment for tunnel construction, the Tunnel Boring Machine (TBM) is directly related to the quality and progress of the project in terms of its safe and efficient operation. However, the complex geological conditions and variable environments during tunnel construction pose great challenges to TBM tunnel support. In recent years, scholars and engineering technicians at home and abroad have carried out a large number of research works on TBM tunnel support and achieved remarkable progress. However, due to the diversity and complexity of geological conditions, the existing research still faces many challenges. For example, in soft strata and fractured zones, the TBM tunneling process is extremely likely to cause stratum deformation and support failure, seriously affecting the construction progress and safety. Therefore, how to make scientific and reasonable support decisions based on multi-source information fusion, and make real-time dynamic adjustments according to the construction situation, so as to achieve efficient, fast and safe support measures, has become a key technical problem to be solved urgently.

[0003] The selection of existing support parameters mainly relies on the engineering analogy method, which is greatly affected by human factors, the results are rough, it is difficult to quantify the support parameter decision-making, and there are problems of support waste and support failure in complex geological environments such as rock bursts and large deformations, and it is difficult to adapt to complex and changeable geological environments. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent decision-making method for TBM tunnel support based on multi-source fusion data. This method widely selects various influencing factors closely related to the support situation and surrounding rock situation, such as geological exploration data, rock mass information, TBM tunneling data, etc. Based on the support parameters actually implemented on site, it deeply integrates a variety of algorithms, so as to realize the intelligent decision-making of TBM tunnel support parameters and improve the intelligent level of TBM tunnels.

[0005] The above technical purpose of the present invention is achieved through the following technical solutions: An intelligent decision-making method for TBM tunnel support based on multi-source fusion data, including the following steps:

[0006] 1) Widely collect the tunneling parameters of existing or under-construction TBM tunnels to form an initial sample library of tunneling data;

[0007] 2) Divide the collected TBM tunnel tunneling parameters into tunneling states and non-tunneling states according to formula (1), and retain the data in the tunneling state;

[0008] f i =F i ·v i ·Ti ·n i (1)

[0009] Where: F i —— The total thrust at the i-th time step, unit: kN; v i —— The tunneling speed at the i-th time step, unit: mm / min; T i —— The cutterhead torque at the i-th time step, unit: MN·m; n i —— The cutterhead rotational speed at the i-th time step, unit: rpm; If f i = 0, then all data at the i-th time step are discarded;

[0010] 3) Process the tunneling state data set using the 3σ method, calculate the average value and standard deviation σ of each tunneling parameter, and remove outlier data points;

[0011] 4) Use the maximum inter-class difference adaptive method to extract tunneling cycles, divide them into cycle rising segments and stable segments, retain the average value of the stable segment of each tunneling cycle to represent the state of the tunneling parameters in this stage, and match them with the tunneling mileage. Finally, form a tunneling parameter database with the tunneling mileage as the coordinate;

[0012] 5) Collect the geological exploration data, designed support parameters, and changed design parameters of the built or under-construction TBM tunnel. Based on the designed support parameters, with the changed design parameters as the modified samples and the mileage as the coordinate, establish a field support parameter combination sample, and splice it with the tunneling parameter database in step 4) to form a TBM tunnel tunneling support sample database;

[0013] 6) Construct a support decision framework. First, arrange the data in a sequence form, and perform a convolution operation on the arranged input vector through Equation (4):

[0014]

[0015] Where, x ∈ R n×d is the input vector, W c ∈ R k×d is the convolution kernel, k is the size of the convolution kernel, b1 is the bias, and h1[t] is the result of the t-th time step of the convolution output.

[0016] Subsequently, perform a global average pooling operation, and perform a global average operation on the output h1 of the convolutional layer through Equation (5):

[0017]

[0018] Subsequently, perform a non-linear transformation through the fully connected layer using the ReLU activation function in Equation (6):

[0019] h3[i] = max(0, W2h2 + b2) (6)

[0020] Subsequently, a non - linear transformation is performed through a fully - connected layer using the sigmoid activation function in Equation (7):

[0021]

[0022] Subsequently, a transformation is performed using BiLSTM in Equation (8):

[0023]

[0024] In the formula, is the forward LSTM, and its calculation formula is shown in Equation (9), is the backward LSTM, and its calculation formula is shown in Equation (10);

[0025]

[0026] Subsequently, a non - linear transformation is performed through a fully - connected layer using the ReLU activation function in Equation (11):

[0027] h6[i] = max(0, W5h5 + b5) (11)

[0028] Subsequently, the output is transformed into a probability distribution using Softmax in Equation (12):

[0029]

[0030] (7) Based on the established support decision - making framework, train with the TBM tunnel boring support sample database established in step (5). Finally, obtain the TBM tunnel support decision - making algorithm based on multi - source fusion kernel data; this algorithm can be used for intelligent support decision - making in TBM tunnels to improve the intelligent level of TBM tunnels.

[0031] The present invention is further set as follows: The steps of the maximum between - class variance adaptive method are as follows:

[0032] (1) Initialize the threshold g0 between the rising section and the stable section in the total propulsion force of the cycle, and count the number of data with the total propulsion force less than g0 in the cycle as N0, and the number of data with the total propulsion force greater than g0 as N1;

[0033] (2) Denote the proportion of the data in the rising section of the total propulsion force of the cycle in the entire cycle data as w0, and calculate the average value μ0 of the total propulsion force in the rising section of the cycle. If the total propulsion force F < g0, it is regarded as the rising section; denote the proportion of the data in the stable section of the total propulsion force of the cycle in the entire cycle data as w1, and calculate the average value μ1 of the total propulsion force in the stable section of the cycle. If the total propulsion force F >= g0, it is regarded as the stable section;

[0034] (3) Based on the proportions w0, w1 and their average values μ0, μ1 of the rising section and the stable section data in the total cyclic propulsion force, the average value μ and the between-class variance S of the entire total cyclic propulsion force can be calculated:

[0035]

[0036] Through further simplification, an equivalent formula can be obtained:

[0037] S = w0w1(μ0 - μ1) 2 (3)

[0038] (4) Take the average value μ of the entire total cyclic propulsion force as the new threshold g1 between the rising section and the stable section of the total cyclic propulsion force, and then calculate the average value μˊ and the between-class variance S′ of the new total cyclic propulsion force according to steps (1) to (3). By comparing the between-class variances S and S′, if S′ > S, then take the average value μˊ of the total cyclic propulsion force as the final threshold between the rising section and the stable section of the total cyclic propulsion force. At this time, the data division of the cyclic rising section and the stable section ends; otherwise, continue to execute steps (1) to (3) until the cyclic division ends.

[0039] In summary, the present invention has the following beneficial effects:

[0040] 1. The selection of the current TBM tunnel support parameters is based on human experience for decision-making, which is slow and has a large amount of subjectivity. The TBM tunnel support decision-making method established based on this technical solution can quantify decision-making indicators, replace the influence of human factors, and achieve rapid decision-making at the same time;

[0041] 2. The input indicators such as the tunneling parameters adopted by the method provided by the present invention can reflect the current surrounding rock conditions during excavation in real time, with high accuracy, so as to realize real-time decision-making of TBM tunnel support parameters;

[0042] 3. In the past, the selected factors for support decision-making were relatively single, such as the parameters of surrounding rock strength and integrity. This technical method is based on machine learning algorithms and can widely adopt multiple tunneling parameters and geological exploration data that reflect the surrounding rock conditions in real time, and the results are more accurate. Description of the Drawings

[0043] Figure 1 is the support decision-making algorithm framework in Embodiment 1 of the present invention;

[0044] Figure 2 is part of the tunneling parameters in Embodiment 2 of the present invention;

[0045] Figure 3 is the tunneling cycle division in Embodiment 2 of the present invention;

[0046] Figure 4It is the result of the support decision-making model in Embodiment 2 of the present invention. Detailed implementation manners

[0047] The following further elaborates on the present invention in conjunction with the appended Figures 1-4 drawings.

[0048] Embodiment 1: A TBM tunnel support intelligent decision-making method based on multi-source fusion data, as Figure 1 shown, includes the following steps:

[0049] 1) Widely collect the tunneling parameters of built or under-construction TBM tunnels to form an initial sample library of tunneling data;

[0050] 2) Divide the collected TBM tunnel tunneling parameters into tunneling states and non-tunneling states according to Equation (1), and retain the data in the tunneling state;

[0051] f i = F i ·v i ·T i ·n i (1)

[0052] In the formula: F i ——The total thrust at the i-th time step, unit: kN; v i ——The tunneling speed at the i-th time step, unit: mm / min; T i ——The cutterhead torque at the i-th time step, unit: MN·m; n i ——The cutterhead rotation speed at the i-th time step, unit: rpm; if f i = 0, then all the data at the i-th time step are discarded;

[0053] 3) Use the 3σ method to process the tunneling state data set, calculate the average value and standard deviation σ of each tunneling parameter, and remove the abnormal data points outside;

[0054] 4) Use the maximum between-class variance adaptive method to extract the tunneling cycles, and divide them into a rising section and a stable section. Retain the average value of the stable section of each tunneling cycle to represent the state of the tunneling parameters at this stage, and match them with the tunneling mileage. Finally, form a tunneling parameter database with the tunneling mileage as the coordinate;

[0055] The steps of the maximum between-class variance adaptive method are as follows:

[0056] (1) Initialize the threshold g0 between the rising section and the stable section in the total cyclic thrust, and count the number of data with the total cyclic thrust less than g0 as N0, and the number of data with the total cyclic thrust greater than g0 as N1;

[0057] (2) Denote the proportion of the data in the rising section of the total cyclic thrust force to the entire cyclic data as \(w_0\), and calculate the average value \(\mu_0\) of the total cyclic thrust force in the rising section. If the total thrust force \(F < g_0\), it is regarded as the rising section; Denote the proportion of the data in the stable section of the total cyclic thrust force to the entire cyclic data as \(w_1\), and calculate the average value \(\mu_1\) of the total cyclic thrust force in the stable section. If the total thrust force \(F\geq g_0\), it is regarded as the stable section;

[0058] (3) According to the proportions \(w_0\), \(w_1\) of the data in the rising section and the stable section of the total cyclic thrust force and their average values \(\mu_0\), \(\mu_1\), the average value \(\mu\) of the total cyclic thrust force for the entire cycle and the between-class variance \(S\) can be calculated:

[0059]

[0060] Through further simplification, an equivalent formula can be obtained:

[0061] \(S = w_0w_1(\mu_0 - \mu_1)\) 2 (3)

[0062] (4) Take the average value \(\mu\) of the total cyclic thrust force for the entire cycle as the new threshold \(g_1\) between the rising section and the stable section of the cyclic thrust force, and then calculate the new average value \(\mu'\) of the cyclic thrust force and the between-class variance \(S'\) according to steps (1) to (3). By comparing the between-class variances \(S\) and \(S'\), if \(S' > S\), then take the average value \(\mu'\) of the cyclic thrust force as the final threshold between the rising section and the stable section of the cyclic thrust force. At this time, the data division of the rising section and the stable section of the cycle ends; otherwise, continue to execute steps (1) to (3) until the cycle division ends.

[0063] (5) Collect the geological exploration data, designed support parameters, and changed design parameters of the existing or under-construction TBM tunnels. Taking the designed support parameters as the benchmark, the changed design parameters as the modification samples, and the mileage as the coordinates, establish a field support parameter combination sample, and splice it with the tunneling parameter database in step 4) to form a TBM tunnel tunneling support sample database;

[0064] (6) Construct a support decision-making framework. First, arrange the data in a sequence form, and perform a convolution operation on the arranged input vector through formula (4):

[0065]

[0066] where \(x\in R\) n×d is the input vector, \(W\) c \(\in R\) k×d is the convolution kernel, \(k\) is the size of the convolution kernel, \(b_1\) is the bias, and \(h_1[t]\) is the result of the \(t\)-th time step of the convolution output.

[0067] Subsequently, perform a global average pooling operation, and perform a global average operation on the output \(h_1\) of the convolutional layer through formula (5):

[0068]

[0069] Subsequently, a non - linear transformation is performed through a fully - connected layer using the ReLU activation function in Equation (6):

[0070] h3[i]=max(0, W2h2 + b2) (6)

[0071] Subsequently, a non - linear transformation is performed through a fully - connected layer using the sigmoid activation function in Equation (7):

[0072]

[0073] Subsequently, a transformation is performed using BiLSTM in Equation (8):

[0074]

[0075] wherein, is the forward LSTM, and its calculation formula is shown in Equation (9), is the backward LSTM, and its calculation formula is shown in Equation (10);

[0076]

[0077] Subsequently, a non - linear transformation is performed through a fully - connected layer using the ReLU activation function in Equation (11):

[0078] h6[i]=max(0, W5h5 + b5) (11)

[0079] Subsequently, the output is transformed into a probability distribution using Softmax in Equation (12):

[0080]

[0081] 7) Based on the established support decision - making framework, train with the TBM tunnel boring support sample database established in step 5), and finally obtain the TBM tunnel support decision - making algorithm based on multi - source fusion kernel data; this algorithm can be used for intelligent support decision - making in TBM tunnels, improving the intelligent level of TBM tunnels.

[0082] Example 2: Taking a certain railway TBM tunnel as an example to demonstrate the implementation path of the present invention

[0083] (1) Collect TBM tunneling parameters including cutterhead rotation speed, cutterhead torque, cutterhead penetration, propulsion speed, total thrust, shield pressure, shield stroke displacement, etc., remove non - tunneling states and eliminate abnormal data points, Figure 2 The visualization display of some parameters is shown as follows;

[0084] (2) Use the maximum inter-class variance adaptive method to divide the tunneling cycle, as Figure 3 shown;

[0085] (3) Collect geological exploration data, design support parameters, change the design parameters. Taking the design support parameters as the benchmark, the changed design parameters as the modified samples, and the mileage as the coordinates, establish the on-site support parameter combination samples. Table 1 shows the classification of the actual on-site support combinations, with a total of 8 categories;

[0086] Table 1 Statistical form of support parameter combination

[0087]

[0088] (4) Based on the established TBM tunnel boring support sample database, conduct training on the support decision-making algorithm, Figure 4 which is the training result, with good effect and an accuracy rate as high as 99%.

[0089] This specific embodiment is only an explanation of the present invention and is not a limitation thereof. Those skilled in the art can make modifications to this embodiment without creative contributions according to needs after reading this specification, but as long as it is within the scope of the claims of the present invention, it is protected by the patent law.

Claims

1. An intelligent decision-making method for TBM tunnel support based on multi-source fusion data, characterized in that: It includes the following steps: 1) Widely collect the tunneling parameters of existing or under-construction TBM tunnels to form an initial sample library of tunneling data; 2) Divide the collected TBM tunnel tunneling parameters into tunneling states and non-tunneling states according to Equation (1), and retain the data in the tunneling state; f i = F i · v i · T i · n i (1) Where: F i —— The total thrust at the i-th time step, unit: kN; v i —— The tunneling speed at the i-th time step, unit: mm / min; T i —— The cutter head torque at the i-th time step, unit: MN·m; n i —— The cutter head rotation speed at the i-th time step, unit: rpm; If f i If f = 0, then all data at the i-th time step are discarded; 3) Process the tunneling state data set using the 3σ method to calculate the average value of each tunneling parameter and the standard deviation σ, and remove the outlier data points; 4) Use the maximum between-class variance adaptive method to extract the tunneling cycle, and divide it into a cycle ascending section and a stable section. Retain the average value of the stable section of each tunneling cycle to represent the state of the tunneling parameters in this stage, and match it with the tunneling mileage. Finally, form a tunneling parameter database with the tunneling mileage as the coordinate; 5) Collect the geological exploration data, design support parameters, and design change parameters of existing or under-construction TBM tunnels. Taking the design support parameters as the benchmark, the design change parameters as the modified samples, and the mileage as the coordinate, establish a combined sample of on-site support parameters, and splice it with the tunneling parameter database in step 4) to form a TBM tunnel tunneling support sample database; 6) Construct a support decision framework. First, arrange the data in a sequence form, and perform a convolution operation on the arranged input vector through Equation (4) Subsequently, perform a global average pooling operation, and perform a global average operation on the output h1 of the convolutional layer according to Equation (5): where \(x\in R\) n×d is the input vector, \(W\) c \(\in R\) k×d is the convolution kernel, \(k\) is the size of the convolution kernel, \(b_1\) is the bias, and \(h_1[t]\) is the result of the convolution output at the \(t\)-th time step; Subsequently, perform a non-linear transformation through a fully connected layer using the ReLU activation function according to Equation (6): h3[i]=max(0,W2h2 + b2) (6) Subsequently, perform a non-linear transformation through a fully connected layer using the sigmoid activation function according to Equation (7): Subsequently, perform a transformation using BiLSTM according to Equation (8): Subsequently, perform a non-linear transformation through a fully connected layer using the ReLU activation function according to Equation (11): In the formula, is the forward LSTM, and its calculation formula is shown in Equation (9). is the backward LSTM, and its calculation formula is shown in Equation (10). h6[i]=max(0,W5h5 + b5) (11) Subsequently, convert the output into a probability distribution using Softmax according to Equation (12): 7) Based on the established support decision framework, train it with the TBM tunnel tunneling support sample database established in step 5). Finally, obtain a TBM tunnel support decision algorithm based on multi-source fusion kernel data; using this algorithm can be applied to TBM tunnels for intelligent support decision-making to improve the intelligent level of TBM tunnels.

2. A TBM tunnel support intelligent decision-making method based on multi-source fusion data according to claim 1, characterized in that: The steps of the maximum between-class variance adaptive method are as follows: (1) Initialize the threshold g0 between the ascending section and the stable section in the total propulsion force of the cycle, and count the number of data with the total propulsion force less than g0 in the cycle as N0, and the number of data with the total propulsion force greater than g0 as N1; (2) Denote the proportion of the data in the ascending section of the total propulsion force of the cycle in the entire cycle data as w0, and calculate the average value μ0 of the total propulsion force in the ascending section of the cycle. If the total propulsion force F < g0, it is regarded as the ascending section; denote the proportion of the data in the stable section of the total propulsion force of the cycle in the entire cycle data as w1, and calculate the average value μ1 of the total propulsion force in the stable section of the cycle. If the total propulsion force F >= g0, it is regarded as the stable section; (3) According to the proportions w0, w1 of the data in the ascending section and the stable section in the total propulsion force of the cycle and their average values μ0, μ1, the average value μ of the total propulsion force of the entire cycle and the between-class variance S can be calculated: ​ Through further simplification, an equivalent formula can be obtained: S = w0w1(μ0 - μ1) 2 (3) (4) Take the mean value μ of the total propulsion force of the entire cycle as the new threshold g1 between the rising section and the stable section of the total propulsion force of the cycle. Then, calculate the mean value μˊ of the new total propulsion force of the cycle and the between-class variance S′ according to steps (1) to (3). By comparing the between-class variances S and S′, if S′>S, then take the mean value μˊ of the total propulsion force of the cycle as the final threshold between the rising section and the stable section of the total propulsion force of the cycle. At this time, the data division of the cycle rising section and the stable section ends; otherwise, continue to execute steps (1) to (3) until the cycle division ends.

Images