TBM tunneling section feature extraction and cutterhead thrust force prediction method based on kernel density estimation

Abstract

The application discloses a TBM tunneling section feature extraction and cutter disc thrust prediction method based on kernel density estimation, which comprises the following steps: step 1, surrounding rock classification; step 2, construction of a cutter disc thrust prediction model; step 3, construction of a model training sample library, specifically comprising: collection of effective tunneling section data, tunneling section internal stage segmentation, extraction and loading of section operation parameters and calculation of stable section cutter disc average thrust; step 4, training of the cutter disc thrust prediction model; step 5, selection of the cutter disc thrust prediction model; step 6, tunneling; step 7, extraction of loading section operation parameters; and step 8, prediction of the stable section cutter disc average thrust. The application establishes a cutter disc thrust prediction model and engineering application by extracting rock mass quality indexes and stable excavation section operation parameters from loading section data, realizes timely prediction of the cutter disc thrust in the TBM construction process, and the prediction result can provide guidance for setting of operation parameters of on-site machine drivers, and further improves the excavation efficiency of the tunnel boring machine.

Description

Technical Field

[0001] The present invention relates to the field of automatic processing of monitoring data during tunnel engineering tunnel boring machine excavation, and in particular to a method for extracting TBM excavation segment features and predicting cutterhead thrust based on kernel density estimation. Background Art

[0002] Tunnel boring machines (TBMs), mechanized equipment that integrates multiple functions including rock breaking, slag removal, and support, are widely used in the construction of major deep tunnel projects. During excavation, TBM operators must promptly adjust machine operating parameters based on changing geological conditions to achieve optimal excavation efficiency under varying rock mass qualities. Therefore, predicting TBM response parameters can effectively guide the adjustment of operating parameters during construction. Conversely, failure to promptly adjust machine operating parameters, such as applying excessive thrust in weak rock formations, can easily lead to accidents such as landslides and machine jams, posing a serious threat to the lives of construction workers and causing significant economic losses.

[0003] An effective TBM excavation section occurs within the maximum travel range, where the TBM operator adjusts operating parameters to gradually increase the excavation speed from zero, then maintains steady excavation until the machine is unloaded and shut down. The steady excavation phase is the primary stage in which the TBM cuts through the rock mass to form a tunnel. Currently, operating parameter adjustments during TBM construction are primarily based on the operator's experience with machine parameters during the rock-breaking and loading phase. However, there is no reference for selecting and setting operating parameters during the steady excavation phase. Therefore, predicting the machine's response cutterhead thrust during the steady excavation phase based on rock-breaking data from the loading phase of the TBM construction process is of great engineering significance for guiding operating parameter adjustments, improving parameter compatibility with the rock mass, and ultimately enhancing machine excavation efficiency. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to address the deficiencies of the above-mentioned prior art and provide a method for extracting the features of the TBM excavation section and predicting the cutterhead thrust based on kernel density estimation. To predict the cutterhead thrust of the stable excavation section during the TBM construction process, it is necessary to collect the excavation section data under different rock mass conditions on site, accurately segment the loading section and stable excavation section data, and establish a thrust prediction model that reflects the interaction between rock and machine based on the loading section data. Based on the characteristics of the TBM on-site excavation process, the present invention establishes a feature extraction method for the excavation section and its internal loading section and stable excavation section, providing an accurate data basis for the training of the prediction model. The prediction model established on this basis is applied to the prediction of cutterhead thrust during the TBM excavation process, which can provide parameter guidance for on-site machine operators and improve the efficiency of machine excavation.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A method for extracting features of a TBM excavation section and predicting cutterhead thrust based on kernel density estimation includes the following steps.

[0007] Step 1. Surrounding rock classification: Based on the relevant national or industry surrounding rock classification standards, the surrounding rock is divided into N types according to rock strength, integrity, structural surface state, occurrence and groundwater conditions.

[0008] Step 2: Construct a cutterhead thrust prediction model: A neural network-based cutterhead thrust prediction model is constructed for each surrounding rock type. The input layer of the cutterhead thrust prediction model is the loading section operation and rock mass index parameter vector [n, v, a, b, TPI], and the output layer is the average cutterhead thrust prediction value in the stable section. Where n is the cutterhead speed during the loading section; v is the TBM propulsion speed during the loading section; a is the slope of the linear fitting curve between the single cutter thrust and penetration during the loading section; b is the initial penetration thrust of the roller cutter during the loading section; and TPI is the slope of the linear fitting curve between the single cutter torque and penetration during the loading section.

[0009] Step 3: Build a model training sample library, which includes the following steps:

[0010] Step 3-1, collecting effective excavation section data: Collect effective excavation section data of all types of surrounding rock in step 1; wherein, the number of effective excavation sections of each type of surrounding rock is not less than M; M ≥ 100.

[0011] Step 3-2, internal stage segmentation of the excavation section: Based on the two-way sliding window method of kernel density estimation, each effective excavation section of each type of surrounding rock is divided into loading section, stable section and unloading section.

[0012] Step 3-3, extracting operating parameters of the loading section and rock mass: extract operating parameters for each loading section of each type of surrounding rock.

[0013] Step 3-4, calculate the average thrust of the cutterhead in the stable section: for each stable section of each type of surrounding rock, calculate the average thrust of the cutterhead in the stable section.

[0014] Step 4: Training the cutterhead thrust prediction model: Using the model training sample library constructed in step 3, the cutterhead thrust prediction model corresponding to the surrounding rock type constructed in step 2 is trained to obtain N trained cutterhead thrust prediction models.

[0015] Step 5: Select a cutterhead thrust prediction model: Determine the surrounding rock type at the tunneling site and select a trained cutterhead thrust prediction model with the same surrounding rock type as determined in step 4.

[0016] Step 6: Tunnel excavation: The TBM performs the construction of the loading section 1 of the tunnel to be excavated.

[0017] Step 7, extracting the operating parameters of the loading section 1: extract the operating parameters of the loading section 1 in step 6 to obtain the operating parameters of the loading section 1 and the rock mass index vector [n′, v′, a′, b′, TPI′].

[0018] Step 8. Predict the average thrust of the cutterhead in the stable section: Substitute the parameter vector [n′, v′, a′, b′, TPI′] of the loading section obtained in step 7 into the trained cutterhead thrust prediction model selected in step 5 to predict the average thrust of the cutterhead in the stable section of the tunnel to be excavated.

[0019] In step 1, based on relevant national or industry standards for surrounding rock classification, for example, according to the national standard "Specifications for Geological Investigation of Water Conservancy and Hydropower Engineering GB50487-2008", the surrounding rock is divided into N=5 types, namely Class I, II, III, IV and V surrounding rock.

[0020] In step 3-1, the effective excavation section refers to the process in which the TBM driver turns on the cutterhead motor current switch, causes the cutterhead to operate at the specified rotational speed n, and adjusts the control speed parameters so that the cutterhead thrust and torque overcome the cutterhead's own friction and idling resistance to cut the rock mass, forming the designed excavation face. Therefore, the method for determining the effective excavation section includes the following steps:

[0021] Step 3-1A, TBM excavation: The TBM excavates the tunnel and monitors the excavation operation data according to the set data sampling interval, thereby obtaining several excavation sections. The excavation operation data of each excavation section includes the cutterhead speed n, cutterhead propulsion speed v, cutterhead thrust F, and cutterhead torque T.

[0022] Step 3-1B, calculate the mean value of the excavation section operation parameters: calculate the mean value of each excavation operation data of each excavation section; the mean values ​​of the excavation operation data of any excavation section are: the mean cutterhead speed Average cutterhead advancement speed Average cutterhead thrust and the mean cutter head torque

[0023] Step 3-1C, determine the effective excavation section: When the mean excavation operation data of a certain excavation section meets the following four basic conditions at the same time, the excavation section is considered to be an effective excavation section; otherwise, it is judged to be an ineffective excavation section and no effective rock cutting and breaking is achieved. The four basic conditions are:

[0024]

[0025] Where, F f is the friction thrust of the cutter disc.

[0026] T fis the friction torque value of the cutter head.

[0027] In step 3-1C, the cutter head friction thrust F f , measured through an on-site TBM friction propulsion test. The test method is to measure the maximum static friction force when the machine hydraulic system pushes the cutter head from a standstill to a speed greater than zero.

[0028] In step 3-1C, the cutter head friction torque value T f , measured through on-site TBM torque idling test. The test method is: when the cutter head speed is set to the maximum rated speed, the maximum torque value during the cutter head rotation process.

[0029] In step 3-2, the method for segmenting the internal stages of the effective excavation section based on the bidirectional sliding window of kernel density estimation includes the following steps:

[0030] Step 3-2A, Identify the loading section starting index value Index1: Traverse the excavation operation data of the valid excavation section A in excavation time order; record the first time index value of the excavation operation data that meets the following Index1 calculation formula as the loading section starting index value Index1; the Index1 calculation formula is:

[0031]

[0032] Step 3-2B, Probability density estimation of cutterhead advancement speed: Perform kernel density estimation on the cutterhead advancement speed of the effective excavation section A to obtain the probability density curve of the cutterhead advancement speed.

[0033] Step 3-2C, obtain the maximum probability density of the cutterhead propulsion speed: find the maximum probability density v of the cutterhead propulsion speed in the stable section of the effective excavation section A from the probability density curve of the cutterhead propulsion speed. max .

[0034] Step 3-2D, identify the index value of the starting point of the stable segment Index2: Use the forward sliding window method to traverse from the starting point to the end point of the effective excavation segment A, and determine the cutterhead advancement speed and v in turn. max When the cutter head advance speed equals or exceeds v for the first time max The time index value corresponding to the time is the index value Index2 of the starting point of the stable segment.

[0035] Step 3-2E, identify the index value of the unloading section starting point Index3: Use the reverse sliding window method to traverse from the end point of the effective excavation section A to the starting point, and determine the cutterhead advancement speed and v in turn. max When the cutter head advance speed equals or exceeds v for the first time max The time index value corresponding to the time is the starting index value Index3 of the unloading segment.

[0036] Step 3-2F, stage division: the effective excavation section A between the starting index value Index1 of the loading section and the starting index value Index2 of the stable section is recorded as the loading section; the effective excavation section A between the starting index value Index2 of the stable section and the starting index value Index3 of the unloading section is recorded as the stable section; the effective excavation section A between the starting index value Index3 of the unloading section and the end point is recorded as the unloading section.

[0037] In step 3-2B, assume that the effective excavation section A has t cutterhead advancement speeds, which are: x1, x2, ..., x t , the probability density of the cutterhead advancement speed is estimated to be The calculation formula is:

[0038]

[0039] Where K(x) represents the Gaussian kernel function; h is the bandwidth. The cutterhead propulsion speed is used as the indicator of density estimation, and its bandwidth size is h = 2 mm / min.

[0040] In step 2, the loss function of the cutterhead thrust prediction model is the root mean square error.

[0041] In step 2, the cutterhead thrust prediction model adopts a 5-fully-connected hidden layer structure; 80% of the model training sample library constructed in step 3 is used as the training set, and 20% is used as the test set.

[0042] The present invention has the following beneficial effects:

[0043] (1) The present invention establishes a criterion for identifying the effective rock breaking stage based on the construction characteristics during TBM excavation. According to the physical characteristics of the excavation parameters during the cutterhead cutting of the rock mass, more accurate excavation section data can be obtained.

[0044] (2) The method of the present invention can quickly and accurately realize the internal segmentation of the excavation section data, and the identification of the starting index value of different stages can achieve high accuracy; the kernel density estimation method can be applied to various data change processes of the excavation section data, and the maximum probability density value is used to accurately identify the stable section data in the excavation section; the bidirectional traversal method based on the excavation section data can minimize the number of data queries and improve the efficiency of index value identification. At the same time, the proposed method is also applicable to shield monitoring data with obvious loading, stabilization and unloading trends or other field monitoring big data with similar laws, and has a wide range of applications.

[0045] (3) A cutterhead thrust prediction model is trained based on rock mass and operating parameter indicators extracted from TBM loading section data reflecting the rock-machine interaction pattern. This model enables thrust prediction during the stable excavation phase. This method can effectively reduce the limitations of on-site construction personnel who rely solely on experience when setting operating parameters. Using thrust prediction results to guide the setting of machine control parameters during the stable excavation phase can effectively improve excavation efficiency and ensure a safe and efficient construction process. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 The overall flow chart of the present invention including segmentation of tunneling sections, extraction of operating parameters, establishment of thrust model and application is shown.

[0047] Figure 2 The curves of cutterhead thrust, cutterhead torque, propulsion speed and cutterhead speed changing with time in the effective excavation section that meets the effective excavation section judgment conditions are shown.

[0048] Figure 3 The curves of cutterhead thrust, cutterhead torque and propulsion speed changing with time in the effective excavation section that does not meet the effective excavation section judgment conditions are shown.

[0049] Figure 4 The schematic diagram shows the principle of data segmentation process of typical effective rock breaking tunneling section data using forward and reverse sliding methods based on kernel density estimation.

[0050] Figure 5 A schematic diagram shows the results of internal segmentation of cutterhead thrust, cutterhead torque, and propulsion speed parameters in a representative effective excavation section using the kernel density forward and reverse sliding segmentation method.

[0051] Figure 6 The curve showing the root mean square error between the predicted and actual cutterhead thrust at location 51735.28 in the fourth section of the Yinsong Project is shown.

[0052] Figure 7 The figure shows the fitting curve of the predicted and actual thrust of the cutterhead at location 51735.28 in the fourth section of the Yinsong Project. DETAILED DESCRIPTION

[0053] The present invention will be further described in detail below with reference to the accompanying drawings and specific preferred embodiments.

[0054] The invention is elaborated in detail based on the collected on-site monitoring data of the tunnel boring machine construction process of the fourth section of the Yinsong Project in Jilin Province.

[0055] A method for extracting features of a TBM excavation section and predicting cutterhead thrust based on kernel density estimation includes the following steps.

[0056] Step 1: Classification of surrounding rock

[0057] During tunnel excavation, rock-breaking mechanisms vary depending on rock mass quality. When the rock mass is intact and has few joints and fissures, the primary mechanism for rock fragmentation is the intrusion and cutting of the machine cutter. However, when the rock mass is well-developed with joints and fissures, weak fractures dominate the formation of rock blocks. Therefore, it is necessary to collect rock-breaking data under different surrounding rock mass quality conditions to ensure the generalizability and robustness of the established prediction model. Rock mass quality conditions are categorized according to the national or industry technical classification standards for the specific tunneling project. Furthermore, the data collection frequency during on-site tunnel boring machine construction must meet the minimum requirements of the corresponding standards.

[0058] The surrounding rock is divided into N types based on rock strength, integrity, structural surface state, occurrence, and groundwater conditions. In this embodiment, according to the national standard "Code for Geological Investigation of Water Conservancy and Hydropower Engineering GB50487-2008," the surrounding rock is divided into N = 5 types, namely Class I, II, III, IV, and V surrounding rock. The higher the surrounding rock type, the worse the rock mass quality. For the fourth section of the Yinsong Project, it includes four surrounding rock types: II, III, IV, and V.

[0059] Step 2: Build a cutterhead thrust prediction model

[0060] A neural network-based cutterhead thrust prediction model is constructed for each surrounding rock type. The input layer of the cutterhead thrust prediction model is the operating parameter vector [n, v, a, b, TPI] in the loading section, and the output layer is the predicted value of the cutterhead thrust in the stable section. Where n is the cutterhead speed during the loading section; v is the TBM propulsion speed during the loading section; a is the slope of the linear fitting curve between the single cutter thrust and penetration during the loading section; b is the initial thrust of the roller cutter during the loading section; and TPI is the slope of the linear fitting curve between the single cutter torque and penetration during the loading section.

[0061] The above-mentioned cutterhead thrust prediction model preferably adopts a 5-fully connected hidden layer structure, where the number of neurons in each layer is [5, 16, 32, 64, 32, 16, 1]. The root mean square error is selected as the loss function. When the learning rate is 1e-1 through the hyperparameter selection method, the optimal prediction performance in the test set is achieved.

[0062] Step 3: Build a model training sample library, which includes the following steps:

[0063] Step 3-1: Collect effective excavation section data

[0064] Collect the valid excavation section data for all surrounding rock types in step 1; the number of valid excavation sections for each type of surrounding rock is no less than M; M ≥ 100.

[0065] During TBM construction, factors such as idle machine shifts and test excavations can cause the cutterhead to idle, resulting in ineffective rock breaking. Therefore, it is necessary to determine whether a section of excavation is effectively rock breaking based on field monitoring data collected under varying surrounding rock quality conditions. This determination of effective rock breaking and cutting sections is based on the characteristic parameters of the tunnel boring machine's effective excavation process.

[0066] An effective rock-breaking excavation phase, also known as the effective excavation phase, occurs when the TBM driver switches on the cutterhead motor current, sets the cutterhead at the specified rotational speed, and then adjusts control parameters to ensure that the cutterhead thrust and torque overcome friction and idling resistance, effectively cutting the rock mass to form the designed excavation face. However, the key difference between the effective rock-breaking phase and processes such as machine step change or idling is that, in addition to the cutterhead rotational speed and thrust speed being greater than zero, the thrust required for the disc cutter to effectively cut the rock must exceed the machine's own frictional resistance, and the cutterhead torque must exceed the machine's own idling resistance.

[0067] Therefore, the above-mentioned method for determining the effective excavation section includes the following steps.

[0068] Step 3-1A, TBM excavation: The TBM excavates the tunnel and monitors the excavation operation data at set intervals, thereby obtaining several excavation sections. The excavation operation data for each excavation section includes the cutterhead speed n, cutterhead propulsion speed v, cutterhead thrust F, and cutterhead torque T.

[0069] Step 3-1B, calculate the mean value of the excavation section operation parameters: calculate the mean value of each excavation operation data of each excavation section; the mean values ​​of the excavation operation data of any excavation section are: the mean cutterhead speed Average cutterhead advancement speed Average cutterhead thrust and the mean cutter head torque

[0070] Step 3-1C: Determine the effective excavation section

[0071] For a field monitoring data of a tunneling section with a length of t, the tunneling section contains a sequence of cutterhead speed n {n1,n2,n3,...,n t}, cutterhead thrust F sequence {F1, F2, F3, ..., F t}, cutter head torque T sequence {T1,T2,T3,...,T t}, the cutter head propulsion speed v sequence {v1,v2,v3,...,v tWhen the mean value of the excavation operation data of an excavation section with a length of t satisfies the following four basic conditions at the same time, the excavation section is considered to be an effective excavation section; otherwise, it is judged to be an ineffective excavation section and no effective rock cutting and breaking is achieved. The four basic conditions are:

[0072]

[0073] Where, F f The friction thrust of the cutterhead is measured through an on-site TBM friction thrust test. The test method is to measure the maximum static friction force when the machine hydraulic system pushes the cutterhead from a standstill to a speed greater than zero.

[0074] T f The friction torque of the cutterhead is measured by an on-site TBM torque idling test. The test method is: when the cutterhead speed is set to the maximum rated speed, the maximum torque value during the cutterhead rotation process is obtained.

[0075] Based on the above formula, the data of each excavation section during the TBM construction process are judged to determine whether it is an effective rock breaking section. The excavation section sequence of the cutterhead speed, cutterhead thrust, cutterhead torque and propulsion speed of the excavation section that does not meet the above conditions is deleted. The TBM construction machinery of the Yinsong Project is an open tunnel boring machine produced by China Railway Equipment Engineering Co., Ltd. Based on the field test data, the maximum friction thrust of the tunnel boring machine is 4000kN, that is, F f =4000kN; the maximum idling torque of the cutter head at the rated maximum speed of 7.6r / min is 200kN·m, that is, T f =200kN·m.

[0076] Preparation of effective on-site excavation section data under different rock mass qualities. Prepare the excavation section data of the on-site tunnel boring machine according to the rock mass quality classification standard to form a database of excavation sections under different rock mass qualities. The effective excavation section data is judged as a criterion, and the effective excavation section sequence that meets the conditions constitutes a database. Based on the above criteria, the effective excavation section of Section 4 of the Yinsong Project is judged, and the change curves of the cutterhead thrust, cutterhead torque, propulsion speed and cutterhead speed in a typical effective excavation section are obtained as follows Figure 2 As shown in the figure; the relevant parameter change curves in the typical invalid excavation section are as follows Figure 3 shown.

[0077] Step 3-2: Internal segmentation of the tunneling section

[0078] The complete excavation process of the tunneling section is the process in which the thrust, torque, and penetration of the TBM cutterhead, driven by the propulsion system, gradually increase from zero to stable rock-breaking cutting, and then decrease to zero after reaching maximum travel. The operating parameters within the tunneling section are divided into four parts: idling period, loading period, stable period, and unloading period.

[0079] Based on the bidirectional sliding window method of kernel density estimation, each effective excavation section of each type of surrounding rock is divided into loading section, stable section and unloading section.

[0080] like Figure 3 As shown in the figure, a method for segmenting the internal stages of the effective excavation section based on a bidirectional sliding window of kernel density estimation is shown in the figure. The principle of bidirectional sliding segmentation is shown in the figure. Figure 4 As shown, it preferably includes the following steps.

[0081] Step 3-2A: Identify the loading segment starting index value Index1

[0082] The loading start point is the point where the cutterhead disc initially penetrates the rock mass and subsequently begins to effectively cut the rock. When the cutterhead reaches the loading start point, the propulsion system must overcome its own gravity and rock resistance, the cutterhead torque must exceed the idling torque, and the penetration rate must be greater than zero. Therefore, the first data index that meets all three of these conditions corresponds to the loading start point.

[0083] The excavation operation data of the effective excavation section A is traversed in the order of excavation time. The first time index value of the excavation operation data that satisfies the following Index1 calculation formula is recorded as the loading section starting index value Index1; the Index1 calculation formula is:

[0084] Index1=argfirst(And(F>F f ,T>F f ,v>0))

[0085] An investigation was conducted on the cutterhead propulsion resistance in the Yinsong and Yinchojiliao projects, and the value was 4000kN. Therefore, the value corresponding to the first data in the excavation section that meets the following conditions is the starting point of the loading section.

[0086] Index1=argfirst(And(F>4000,T>200,v>0))

[0087] The above formula is used to identify the starting point of the loading phase in a typical effective excavation section of the fourth section of the Yinsong Project. Figure 3 As shown, the excavation section is judged according to the above conditions and the starting point of the excavation section is at the position of Index1=180s.

[0088] Step 3-2B, Probability density estimation of cutterhead advancement speed: Perform kernel density estimation on the cutterhead advancement speed of the effective excavation section A to obtain the probability density curve of the cutterhead advancement speed.

[0089] The curve is obtained by performing kernel density estimation on the advancement speed of the representative tunneling section of the Yinsong Project, as shown in the figure below: Figure 3 As shown in the figure on the right.

[0090] Assume that the effective excavation section A has t cutterhead advancement speeds, which are: x1, x2, ..., x t , the probability density of the cutterhead advancement speed is estimated to be The calculation formula is:

[0091]

[0092] Where K(x) represents the Gaussian kernel function; h is the bandwidth, a parameter that determines the smoothness of the overall probability density function of f(x). A smaller bandwidth results in a less smooth probability density curve. In this paper, a Gaussian kernel function is used, and the cutterhead propulsion speed is used as the density estimation indicator to estimate the probability density. The bandwidth is h = 2 mm / min.

[0093] Step 3-2C, obtain the maximum probability density of the cutterhead propulsion speed: find the maximum probability density v of the cutterhead propulsion speed in the stable section of the effective excavation section A from the probability density curve of the cutterhead propulsion speed. max .

[0094] For the Yinsong Project, the propulsion speed value corresponding to the peak position of the probability density is 70mm / min.

[0095] Step 3-2D: Identify the starting index value of the stable segment, Index2

[0096] Using the forward sliding window method, traverse from the starting point to the end point of the effective excavation section A, and judge the cutter head advancement speed and v in turn. max For comparison, that is:

[0097] Index2=argfirst(v≥v max )

[0098] When the cutterhead advance speed equals or exceeds v for the first time max The time index value corresponding to the time is the index value Index2 of the starting point of the stable segment.

[0099] For the Songyuan Project, the first time index value that meets the following conditions is the starting point of the stable excavation section:

[0100] Index2=argfirst(v≥70)

[0101] The results show that when the current excavation section is excavated to the 580s position, the advancement speed of the excavation section exceeds 70mm / min for the first time, so the time index value position of the starting point of the stable section of the excavation section is Index2=580s.

[0102] Step 3-2E, identify the index value of the unloading section starting point Index3: The starting point of the unloading section is also the end point of the stable section. Use the reverse sliding window method to traverse from the end point of the effective excavation section A to the starting point, and determine the cutterhead advancement speed and v in turn. max When the cutter head advance speed equals or exceeds v for the first time max The time index value corresponding to the time is the starting index value Index3 of the unloading segment.

[0103] For the pine project, the time when the propulsion speed value first exceeds 70mm / min is 1950s. Figure 3 Therefore, the time index value of the unloading section starting point of the tunneling section is obtained as Index3=1950s.

[0104] Step 3-2F, stage division: the effective excavation section A between the starting index value Index1 of the loading section and the starting index value Index2 of the stable section is recorded as the loading section; the effective excavation section A between the starting index value Index2 of the stable section and the starting index value Index3 of the unloading section is recorded as the stable section; the effective excavation section A between the starting index value Index3 of the unloading section and the end point is recorded as the unloading section.

[0105] The tunneling segment data is segmented and stored independently based on the identified index values ​​of the loading, stable excavation, and unloading segments. Repeating these steps, the collected valid tunneling segment excavation data under different surrounding rock conditions is segmented into loading, stable excavation, and unloading segment datasets.

[0106] For the Yinsong Project, the time index values ​​of the starting point of the loading section, the starting point of the stable section, and the starting point of the unloading section in this tunneling section are 180s, 580s, and 1950s respectively. Based on the above index values, the data of different tunneling parameters in this tunneling section between 181s and 580s, 581s and 1950s, and 1951 and 2200s are retrieved respectively. The data after internal segmentation is as follows Figure 5 The segmented data is stored in three separate CSV files. Repeat the above steps to segment the effective rock-breaking excavation section data under different surrounding rock conditions of the Yinsong Project into loading section, stable excavation section, and unloading section data.

[0107] Step 3-3, extracting the operating parameters of the loading section: extracting the operating parameters of each loading section of each surrounding rock type.

[0108] Step 3-4, calculate the average thrust of the cutterhead in the stable section: for each stable section under each surrounding rock type, calculate the average thrust of the cutterhead in the stable section.

[0109] Rock mass indices are empirically based on the difficulty of rock excavation. These include the field penetration index and the torque penetration index. The thrust penetration index is the slope a and the thrust intercept b of the linear fit between the single-pole thrust and penetration during the loading phase. The torque penetration index is the slope TPI of the linear fit between the single-pole torque and penetration during the loading phase. Rock mass indices under different working conditions are obtained by linearly fitting the single-pole thrust and penetration, and the single-pole torque and penetration, during the loading phase under different working conditions.

[0110] The operating parameters extract the cutterhead rotational speed and thrust velocity during the loading section of the tunneling segment. Based on this, the rock mass indicators, operating parameters, and thrust values ​​for each effective tunneling segment's loading section are combined to form a thrust prediction model database, whose matrix form is [n, v, a, b, TPI, F]. Similarly, the extracted and predicted parameters for the loading section under different operating conditions are combined to form the following initial model training database for a total of m tunneling segments.

[0111]

[0112] Step 4: Training the cutterhead thrust prediction model: Using the model training sample library constructed in step 3, the cutterhead thrust prediction model corresponding to the surrounding rock type constructed in step 2 is trained to obtain N cutterhead thrust prediction models trained for different surrounding rock types.

[0113] 80% of the data under each working condition is randomly selected as the training set, and the remaining 20% ​​is used as the test set. Then, the training set data is input into the neural network for training.

[0114] Step 5: Select a cutterhead thrust prediction model: Determine the surrounding rock type at the tunneling site and select a trained cutterhead thrust prediction model with the same surrounding rock type as determined in step 4.

[0115] Step 6: Tunnel excavation: The thrust prediction model is deployed to the intelligent excavation platform in the on-site tunnel boring machine control room, and the TBM constructs the loading section 1 of the tunnel to be excavated.

[0116] Step 7, extracting the operating parameters of the loading section 1: perform operations on the loading section 1 in step 6 and extract rock mass index parameters to obtain the operating parameter vector of the loading section 1 [n′, v′, a′, b′, TPI′].

[0117] Taking the on-site cutterhead thrust prediction at pile number 51735.28 in the fourth section of the Yinsong Project as an example, the rock mass parameters were extracted from the loading section data. The single cutterhead thrust and penetration slope a'=5.44, the intercept initial penetration thrust b'=88.89, and the torque penetration coefficient TPI'=1.32.

[0118] Step 8: Predict the average thrust of the cutterhead in the stable section: Substitute the loading section operation parameter vector [n′, v′, a′, b′, TPI′] obtained in step 7 into the trained cutterhead thrust prediction model selected in step 5 to predict the average thrust of the cutterhead in the stable section of the tunnel to be excavated.

[0119] For the pile number 51735.28 of the fourth section of the Yinsong Project, the operating parameter vector of the loading section 1 is input into the thrust prediction model to obtain the predicted thrust and actual thrust values ​​as shown in the following figure: Figure 6 and Figure 7 As shown in the figure, the average percentage error between the actual value and the predicted value of the thrust of the tunneling section is 5.34%, and the root mean square error is 8.94; the prediction results show that the established prediction model can accurately predict the cutterhead thrust during the construction process.

[0120] The proposed bidirectional sliding traversal method based on kernel density estimation significantly improves data segmentation efficiency. Based on the data characteristics of different stages, a feature extraction method for corresponding segments is proposed. This method can also achieve high accuracy for data with similar trends, and has a wide range of applications. By extracting rock mass quality indicators from loading section data and operating parameters from the stable excavation section, a cutterhead thrust prediction model is established and applied in engineering projects, achieving real-time, high-precision prediction of cutterhead thrust during TBM construction. The prediction results can provide guidance for on-site machine operators in setting operating parameters, thereby improving the excavation efficiency of tunnel boring machines.

[0121] The preferred embodiments of the present invention are described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the scope of protection of the present invention.

Claims

1. A method for extracting features from TBM excavation sections and predicting cutterhead thrust based on kernel density estimation, characterized by: The steps include: Step 1: Classification of surrounding rocks: Classify surrounding rocks into N types based on rock strength, integrity, structural surface state, occurrence, and groundwater conditions; Step 2: Construct a cutterhead thrust prediction model: A neural network-based cutterhead thrust prediction model is constructed for each surrounding rock type. The input layer of the cutterhead thrust prediction model is the loading section operation and rock mass index parameter vector [n, v, a, b, TPI] of the TBM construction process, and the output layer is the average cutterhead thrust prediction value in the stable section. Where n is the cutterhead speed during the loading section; v is the TBM propulsion speed during the loading section; a is the slope of the fitted curve of the single cutter thrust and penetration during the loading section; b is the initial penetration thrust of the roller cutter during the loading section; TPI is the linear fitting slope of the single cutter torque and penetration during the loading section; Step 3: Build a model training sample library, which includes the following steps: Step 3-1, collecting effective excavation section data: Collect effective excavation section data for all types of surrounding rock in step 1; wherein the number of effective excavation sections for each type of surrounding rock is not less than M; M ≥ 100; The effective excavation section refers to the process in which the TBM driver turns on the cutterhead motor current switch, causes the cutterhead to operate at the specified rotational speed n, and adjusts the control parameters so that the cutterhead thrust and torque overcome the cutterhead's own friction and idling resistance to cut the rock mass, forming the designed excavation face. Therefore, the method for determining the effective excavation section includes the following steps: Step 3-1A, TBM excavation: The TBM excavates the tunnel and monitors the excavation operation data at the set data sampling interval, thereby obtaining several excavation sections. The excavation operation data for each excavation section includes the cutterhead speed n, cutterhead propulsion speed v, cutterhead thrust F, and cutterhead torque T; Step 3-1B, calculate the mean value of the excavation section operation parameters: calculate the mean value of each excavation operation data of each excavation section; the mean values ​​of the excavation operation data of any excavation section are: the mean cutterhead speed Average cutterhead advancement speed Average cutterhead thrust and the mean cutter head torque Step 3-1C, determine the effective excavation section: When the mean excavation operation data of a certain excavation section meets the following four basic conditions at the same time, the excavation section is considered to be an effective excavation section; otherwise, it is judged to be an ineffective excavation section and no effective rock cutting and breaking is achieved. The four basic conditions are: Where, F f is the friction thrust of the cutter head; T f is the friction torque value of the cutter head; Step 3-2: Segmentation of internal stages of the tunneling section: Based on the two-way sliding window method of kernel density estimation, each effective tunneling section of each type of surrounding rock is divided into loading section, stable section, and unloading section; Step 3-3, extracting loading section operation and rock mass index parameters: extracting operation parameters for each loading section of each type of surrounding rock; Step 3-4, calculate the average thrust of the cutterhead in the stable section: for each stable section of each type of surrounding rock, calculate the average thrust of the cutterhead in the stable section; Step 4: Training the cutterhead thrust prediction model: Using the model training sample library constructed in step 3, the cutterhead thrust prediction model corresponding to the surrounding rock type constructed in step 2 is trained to obtain N trained cutterhead thrust prediction models; Step 5: Select a cutterhead thrust prediction model: Determine the surrounding rock type at the tunneling site and select a trained cutterhead thrust prediction model that matches the surrounding rock type determined in step 4. Step 6: Tunnel excavation: The TBM performs loading section 1 on the tunnel to be excavated. Step 7, extracting the operating parameters of the loading section 1: extracting the operating parameters of the loading section 1 in step 6, and obtaining the loading section 1 operation and rock mass index parameter vector [n′, v′, a′, b′, TPI′]; Step 8: Predict the average thrust of the cutterhead in the stable section: Substitute the loading section operation parameter vector [n′, v′, a′, b′, TPI′] obtained in step 7 into the trained cutterhead thrust prediction model selected in step 5 to predict the average thrust of the cutterhead in the stable section of the tunnel to be excavated.

2. The method for extracting features of TBM excavation sections and predicting cutterhead thrust based on kernel density estimation according to claim 1 is characterized by: In step 1, according to the national standard "Specifications for Geological Investigation of Water Conservancy and Hydropower Engineering GB50487-2008", the surrounding rock is divided into N=5 types, namely Class I, II, III, IV and V surrounding rock.

3. The method for extracting features of TBM excavation sections and predicting cutterhead thrust based on kernel density estimation according to claim 1 is characterized by: In step 3-1C, the cutter head friction thrust F f , measured through an on-site TBM friction propulsion test. The test method is to measure the maximum static friction force when the machine hydraulic system pushes the cutter head from a standstill to a speed greater than zero.

4. The method for extracting features of TBM excavation sections and predicting cutterhead thrust based on kernel density estimation according to claim 1 is characterized in that: In step 3-1C, the cutter head friction torque value T f , measured through on-site TBM torque idling test. The test method is: when the cutter head speed is set to the maximum rated speed, the maximum torque value during the cutter head rotation process.

5. The method for extracting features of TBM excavation sections and predicting cutterhead thrust based on kernel density estimation according to claim 1 is characterized in that: In step 3-2, the method for segmenting the internal stages of the effective excavation section based on the bidirectional sliding window of kernel density estimation includes the following steps: Step 3-2A, Identify the loading section starting index value Index1: Traverse the excavation operation data of the valid excavation section A in excavation time order; record the first time index value of the excavation operation data that meets the following Index1 calculation formula as the loading section starting index value Index1; the Index1 calculation formula is: Step 3-2B, Probability density estimation of cutterhead advancement speed: Perform kernel density estimation on the cutterhead advancement speed in the effective excavation section A to obtain the probability density curve of the cutterhead advancement speed; Step 3-2C, obtain the maximum probability density of the cutterhead propulsion speed: find the maximum probability density v of the cutterhead propulsion speed in the stable section of the effective excavation section A from the probability density curve of the cutterhead propulsion speed. max ; Step 3-2D, identify the index value of the starting point of the stable segment Index2: Use the forward sliding window method to traverse from the starting point to the end point of the effective excavation segment A, and determine the cutterhead advancement speed and v in turn. max When the cutter head advance speed equals or exceeds v for the first time max The time index value corresponding to the time is the index value Index2 of the starting point of the stable segment; Step 3-2E, identify the index value of the unloading section starting point Index3: Use the reverse sliding window method to traverse from the end point of the effective excavation section A to the starting point, and determine the cutterhead advancement speed and v in turn. max When the cutter head advance speed equals or exceeds v for the first time max The time index value corresponding to the time is the starting index value of the unloading segment Index3; Step 3-2F, stage division: the effective excavation section A between the starting index value of the loading section Index 1 and the starting index value of the stable section Index 2 is recorded as the loading section; The effective excavation section A between the starting index value Index2 of the stable section and the starting index value Index3 of the unloading section is recorded as the stable section; the effective excavation section A between the starting index value Index3 of the unloading section and the end point is recorded as the unloading section.

6. The method for extracting features of TBM excavation sections and predicting cutterhead thrust based on kernel density estimation according to claim 5 is characterized by: In step 3-2B, assume that the effective excavation section A has t cutterhead advancement speeds, which are: x1, x2, ..., x t , the probability density of the cutterhead advancement speed is estimated to be The calculation formula is: Where K(x) represents the Gaussian kernel function and h is the bandwidth.

7. The method for extracting features of TBM excavation sections and predicting cutterhead thrust based on kernel density estimation according to claim 6 is characterized in that: The cutter head propulsion speed is used as the index of density estimation, and its bandwidth size is h = 2 mm / min.

8. The method for extracting features of TBM excavation sections and predicting cutterhead thrust based on kernel density estimation according to claim 1 is characterized by: In step 2, the loss function of the cutterhead thrust prediction model is the root mean square error.

9. The method for extracting features of TBM excavation sections and predicting cutterhead thrust based on kernel density estimation according to claim 1 is characterized by: In step 2, the cutterhead thrust prediction model adopts a 5-fully-connected hidden layer structure; 80% of the model training sample library constructed in step 3 is used as the training set, and 20% is used as the test set.

Images