Tbm surrounding rock quality real-time sensing method based on online transfer learning guided by physics

Abstract

The application discloses a TBM surrounding rock quality real-time sensing method based on physical guidance online migration learning, which comprises the following steps: 1, source domain and target domain tunnel engineering object division; 2, feature parameter extraction based on source domain field monitoring data sharing parameter space; 3, construction of a thrust prediction model based on LSTM; 4, construction of a surrounding rock state prediction model based on a similarity index; 5, target domain surrounding rock quality online prediction. The application extracts the characteristic value of the data stream of the working section from the shared space parameters in the different tunnel engineering field monitoring data in the source domain, and constructs a surrounding rock quality online prediction module based on shared features for real-time identification of the surrounding rock state across projects. The application widens the migration application feasibility of related intelligent models, and the online prediction module can realize rapid diagnosis of the surrounding rock state by relying on real-time monitoring data flow in the TBM construction process of the newly-built tunnel, thereby providing an important basis for guaranteeing efficient and safe TBM construction.

Description

Technical Field

[0001] This invention relates to the field of intelligent application of tunnel boring machine (TBM) construction monitoring, and in particular to a real-time sensing method for TBM surrounding rock quality based on physical-guided online transfer learning. Background Technology

[0002] Tunnel boring machines (TBMs) are widely used in underground space development and construction as a highly efficient, safe, and environmentally friendly mechanized excavation method. However, TBM performance is affected by changes in surrounding rock geological conditions. If the machine cannot promptly identify adverse rock conditions during excavation, it can cause equipment jamming, rockfalls, and other engineering disasters, leading to equipment damage and even seriously threatening the lives of personnel on site. Furthermore, due to the obstruction caused by the machine's shield, methods such as advanced exploration and geological mapping are inconvenient to implement during TBM construction, and the analysis of related results is time-consuming, failing to meet the needs of rapid, real-time diagnosis on-site. Therefore, tunnel engineering requires the development of real-time sensing methods for TBM surrounding rock quality.

[0003] In recent years, advancements in sensor technology have laid the foundation for the acquisition of data based on on-site monitoring. TBM excavation data is the result of the interaction between the machine and the rock mass; therefore, on-site monitoring of excavation data offers the possibility of real-time intelligent evaluation of surrounding rock quality during TBM construction. However, related research mainly focuses on training and testing intelligent diagnostic methods for surrounding rock quality using monitoring data from individual tunnel projects, lacking verification of cross-project transfer learning applications of relevant models in newly constructed tunnels. Furthermore, the initial construction of new tunnels faces challenges such as insufficient monitoring and learning sample data, and the on-site monitoring data is collected in the form of real-time data streams. Therefore, how to achieve online transfer learning-based intelligent diagnostic model prediction of surrounding rock quality under the conditions of TBM construction site monitoring data streams is crucial to ensuring the cross-project application of related methods.

[0004] Developing online transfer learning methods suitable for on-site monitoring data streams in new tunnel engineering still requires addressing the following key issues. First, how to effectively process the real-time data streams acquired in new tunnel engineering to extract effective working segments of the TBM and features that enable transfer learning for different tunnels; second, how to develop online learning models for rock mass quality diagnosis based on real-time monitoring data streams in new engineering projects without supervised learning labels; and finally, how to implement the transfer learning application of the online diagnostic model during the construction process of new tunnels. The establishment of these methods will enable the application of rock mass quality evaluation models trained on completed engineering monitoring data during TBM construction in new tunnel engineering, achieving real-time evaluation of rock mass stability, reducing geological exploration costs for new tunnel engineering, and ensuring the safe and efficient progress of TBM construction. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to address the shortcomings of the prior art by providing a real-time sensing method for TBM surrounding rock quality based on physical-guided online transfer learning. This real-time sensing method for TBM surrounding rock quality based on physical-guided online transfer learning uses the physical process of TBM rock breaking to extract characteristic parameters that are widely present in different tunnel projects as the basis for transfer learning. By constructing a standardized online data stream processing and physical-guided feature extraction module, an online prediction module for surrounding rock categories, and a transfer learning application module, real-time sensing of surrounding rock quality across tunnel projects is achieved.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0007] A real-time sensing method for TBM surrounding rock quality based on physical-guided online transfer learning includes the following steps.

[0008] Step 1: Delineation of Tunnel Engineering Objects (Source and Target Domains): The source domain refers to the tunnel engineering project that has been excavated and completed, and its on-site monitoring data space is represented by X. S The source domain data space contains on-site monitoring data of M tunnel projects, where M≥1; the surrounding rock quality status of the M tunnel projects in the source domain is divided into S types, where S≥2.

[0009] The target domain refers to the N tunnel projects that have not yet been excavated, and the space for their on-site monitoring data is denoted by X. T It means that N≥1;

[0010] Step 2, Feature parameter extraction based on the shared parameter space of source domain field monitoring data, includes the following steps:

[0011] Step 2-1: Segment the data stream of the effective rock-breaking working stage of the TBM in different tunnel projects within the source domain to obtain n effective rock-breaking working segment data streams.

[0012] Step 2-2: Determine the shared parameter space X within the source domain. Shard A shared parameter space X is established for the data stream of each effective rock-breaking section. Shard ;X Shard The number M of on-site monitoring data for tunnel engineering in the source domain is determined, specifically as follows:

[0013] A. When M = 1, X Shard These are the rock-breaking parameters.

[0014] B. When M > 1, X Shard It is the union of rock-breaking parameters and non-rock-breaking parameters.

[0015] Steps 2-3: Extract shared parameter feature values: Shared parameter feature values ​​include rock breaking parameter feature values ​​and non-rock breaking parameter feature values; among them, rock breaking parameter feature values ​​include rock breaking parameter statistical feature values ​​and excavability indicators; among them, excavability indicators are physical indicators that are related to the rock breaking process; non-rock breaking parameter feature values ​​are the mean values ​​of each non-rock breaking parameter.

[0016] Step 3: Construct an LSTM-based thrust prediction model: Construct a thrust prediction model based on the LSTM neural network algorithm; Let the current working segment to be predicted be the t-th working segment, then the input layer of the thrust prediction model is the shared parameter feature values ​​of the L historical working segments before the t-th working segment; the output layer is the thrust prediction value of the t-th working segment; Based on the source domain shared parameter feature values ​​extracted in Step 2, train the constructed thrust prediction model to obtain the trained thrust prediction model.

[0017] Step 4: Construct a surrounding rock condition prediction model based on similarity index: Calculate the similarity error between the predicted thrust value of the t-th working segment and the actual thrust value of the t-th working segment; Based on the source domain shared parameter feature values ​​extracted in Step 2, obtain the similarity error sequence SI corresponding to the n effective rock breaking working segments; Combine this with the S surrounding rock quality states divided in Step 1, and thus construct a surrounding rock condition prediction model based on the similarity index sequence SI and the corresponding surrounding rock classification sequence S.

[0018] Step 5, Online prediction of surrounding rock quality in the target area, includes the following steps:

[0019] Step 5-1, Initial Excavation of the Target Area: The TBM excavates the target area to the h-th working section and collects on-site monitoring data for each working section; at the same time, the surrounding rock type of the first to h-th excavated working sections is surveyed in the field.

[0020] Step 5-2: Continue tunneling in the target area: The TBM tunnels from the (h+1)th working segment to the (t)th working segment in the target area, and collects on-site monitoring data for each working segment; where t ≥ h+1.

[0021] Step 5-3: Extract shared parameter feature values ​​of the target domain: Extract shared parameter feature values ​​for each working segment in Step 5-1 and Step 5-2.

[0022] Step 5-4, Thrust Prediction for the t-th Working Segment: Input the shared parameter feature values ​​of the L historical working segments preceding the t-th working segment extracted in Step 3 into the LSTM-based thrust prediction model constructed in Step 3 to obtain the thrust prediction value for the t-th working segment in the target domain.

[0023] Step 5-5: Calculate similarity: Calculate the true thrust value of the t-th working segment of the target domain obtained in Step 5-2. Compared with the thrust prediction value obtained in step 5-4 The similarity error SI of the t-th working segment in the target domain was calculated. t .

[0024] Steps 5-6: Online prediction of surrounding rock quality in the (t+1)th working segment: The surrounding rock type of the (t-1)th working segment is taken as the surrounding rock type s before migration. t-1 , will s t-1 and SI t Substituting into the surrounding rock state prediction model constructed in step 4, we find the value in s t-1 and SI t The probability values ​​of different surrounding rock states under the conditions; among them, the surrounding rock state corresponding to the maximum probability value is the predicted surrounding rock quality state of the (t+1)th working section; among them, the surrounding rock type of the (t-1)th working section is obtained directly from the survey in step 5-1 or based on the survey data in step 5-1 and obtained through online quality prediction.

[0025] In step 1, the surrounding rock quality status of the M tunnel projects in the source domain adopts the national standard "Code for Geological Investigation of Water Conservancy and Hydropower Projects GB50487-2008", which classifies the surrounding rock of the tunnels into stable I, basically stable II, locally stable III, unstable IV and extremely unstable V, for a total of S=5 types.

[0026] In step 2-2, the rock breaking parameters include cutterhead thrust, cutterhead torque, feed rate, cutterhead rotation speed, and penetration depth; where penetration depth is the ratio of feed rate to cutterhead rotation speed.

[0027] In steps 2-3, the excavability indices include the linear fitting slope 'a' and thrust intercept 'b' of the single-blade thrust and penetration data streams, as well as the linear fitting slope TPI of the single-blade torque and penetration.

[0028] In steps 2-3, the statistical feature values ​​of rock breaking parameters include the statistical feature values ​​of rock breaking operation parameters and the statistical feature values ​​of rock breaking non-operation parameters; in step 3, the LSTM neural network includes an LSTM layer and a fully connected layer; the input of the LSTM layer is the statistical feature values ​​of rock breaking non-operation parameters, the excavability index, and the non-rock breaking parameter feature values ​​of the L historical working segments before the t-th working segment; the input of the fully connected layer is the output of the LSTM layer and the statistical feature values ​​of rock breaking operation parameters; the output of the fully connected layer is the thrust prediction value of the t-th working segment.

[0029] In step 3, the thrust prediction model is trained based on the source domain shared parameter feature values ​​extracted in step 2. During the training process, the hyperparameters of the thrust prediction model are optimized. The specific optimization method includes the following steps:

[0030] Step 3B-1, Sample Library Division: Divide the n source domain working segment data streams loaded in Step 3A-2 into training set and validation set according to the set ratio.

[0031] Step 3B-2: Design hyperparameters: Design several combinations of hyperparameters involved in the thrust prediction model.

[0032] Step 3B-3, Model Training: Use the training set data to train each hyperparameter combination in the thrust prediction model to obtain several trained thrust prediction models.

[0033] Step 3B-4, Hyperparameter Optimization: Validate each thrust prediction model trained in Step 3B-3 using the validation set, and select the hyperparameter combination with the best validation performance as the optimal hyperparameter. Freeze and fix the optimal hyperparameter. The thrust prediction model with the optimal hyperparameter is also called the optimal thrust prediction model.

[0034] In step 4, let the similarity error sequence of the source domain be SI={si1,...,si i ,...,si n Simultaneously, based on the results of the source domain's on-site surrounding rock quality status classification, the surrounding rock classification sequence S = {s0, s1, ..., s} corresponding to the working section data stream is obtained. i ,...,s n A rock mass classification state transition probability model is constructed based on the similarity index sequence SI and the corresponding surrounding rock classification sequence, as shown in the following equation:

[0035] M(s,si)={p0(s0|s,si),...,p i (s i |s,si),...,p n (s n |s,si)}

[0036] In the formula, p i (s i |s,si) is the conditional probability density distribution function when transitioning from the excavated state s to any other arbitrary state with a similarity index of si. This distribution function establishes a non-parametric probability density statistical model by using the similarity index of different rock mass categories during state transitions in the surrounding rock classification sequence.

[0037] In step 5-1, h = 20.

[0038] The present invention has the following beneficial effects:

[0039] (1) This invention can segment the monitoring data stream during TBM construction, extract feature parameters, and calculate similarity indices to achieve real-time prediction of the surrounding rock condition probability. This method can complete online intelligent prediction of surrounding rock quality. Compared with traditional geological exploration and advanced geological drilling, it does not affect the normal excavation process of TBM, has high time efficiency, and reduces related exploration costs.

[0040] (2) The feature parameter extraction in the source and target domain transfer learning method considers parameters related to the physical guidance of rock breaking process and other shared feature parameters. Among them, the rock breaking parameter features based on the interaction between TBM and rock mass, as the main parameters of the shared feature space, have the convenience and feasibility of transfer application. This is because although the dimensions of the on-site monitoring parameters of different tunnel projects are not entirely the same, the main rock breaking parameters for cutting rock mass during TBM excavation, such as thrust, torque, advance speed, and cutterhead rotation speed, are simple and easy to monitor in different tunnels. This provides a data feature foundation for the establishment of transfer learning related models and cross-engineering applications.

[0041] (3) This invention realizes an online intelligent perception method for surrounding rock condition based on source domain shared features to construct an online learning module and apply it across engineering projects in the target domain. The established online learning module includes LSTM similarity index prediction and surrounding rock condition prediction models. It does not require the prior acquisition of monitoring data and supervised learning labels in the target domain. Compared with traditional offline learning methods, it does not require the accumulation of monitoring data for newly constructed tunnel projects. It enables real-time prediction of the surrounding rock condition in the early and middle stages of newly constructed tunnel projects, significantly expanding the cross-engineering application scope of the transfer learning module. This makes it possible to predict the surrounding rock condition of more mechanized construction tunnel projects, thereby ensuring the safety of the TBM construction process. Attached Figure Description

[0042] Figure 1 This is a flowchart of the TBM surrounding rock quality real-time sensing method based on physical-guided online transfer learning according to the present invention.

[0043] Figure 2 This is a schematic diagram of the rock-breaking process and feature extraction related to TBM cutting of rock mass based on the working segment data stream. Sub-figure (a) shows the extracted representative working segment data stream, sub-figure (b) shows the effective rock-breaking process of the cutter intruding into the rock mass, sub-figure (c) shows the linear fitting of single-cutter thrust and penetration, and sub-figure (d) shows the single-cutter torque and penetration.

[0044] Figure 3 This is a schematic diagram of the prediction process for the online prediction module of surrounding rock quality.

[0045] Figure 4 This is a schematic diagram illustrating the application of parameter freezing and target domain transfer in an online learning module based on a shared feature parameter space.

[0046] Figure 5 This is a schematic diagram of the thrust prediction model based on LSTM.

[0047] Figure 6 This is an example diagram of the probability density distribution of surrounding rock state transition based on the similarity error index.

[0048] Figure 7 The following are examples of the application results of the online transfer learning TBM surrounding rock condition real-time perception method in the target domain. Sub-figure (a) shows the single-blade thrust prediction and measured values ​​of the working section data stream obtained by using the LSTM prediction model between chainages 65334 and 65325 of the Yinchuo-Jiliao No. 6 Tunnel in the target domain; sub-figure (b) shows the change process of the average percentage error of the similarity index within the corresponding chainage range; and sub-figure (c) shows an example of the change curve of the probability prediction results of different surrounding rock condition categories within the corresponding chainage range.

[0049] Figure 8 The results of surrounding rock state transition and classification prediction of the online working section data stream of the Jiliao Tunnel in the target domain are presented. Subfigure (a) shows the example results of the accuracy of surrounding rock state transition, accuracy of surrounding rock category prediction and F1 index of the method in the example tunnel sections of the source domain and the target domain. Subfigure (b) shows the confusion matrix results of the data stream of all working sections in the target domain under different surrounding rock states. Detailed Implementation

[0050] The present invention will now be described in further detail with reference to the accompanying drawings and specific preferred embodiments.

[0051] This invention uses the on-site monitoring data from the construction process of the open-type TBM in Section 4 of the Jilin Province Yinsong Project water conveyance tunnel between Chaluhe and Yinmahe as the source domain data for transfer learning, and the real-time data stream of the on-site monitoring of the open-type TBM construction process in Section 6 of the Inner Mongolia Yinchuojiliao Project as the target domain for transfer learning. This invention will be described in detail.

[0052] like Figure 1 As shown, a real-time sensing method for TBM surrounding rock quality based on physical-guided online transfer learning includes the following steps.

[0053] Step 1: Delineation of Tunnel Engineering Objects in Source and Target Domains

[0054] The source domain refers to the tunnel project that has been excavated and completed, and its on-site monitoring data is represented spatially.

[0055] The source domain data space can contain M tunnel engineering field monitoring data, where M≥1; in this embodiment, M=1 is preferred. In this invention, the source domain relies on the field monitoring data of the 19.77km section of the Jilin Province Yinsong Water Conveyance Tunnel No. 4, which has been excavated and completed.

[0056] Furthermore, the method for assessing the surrounding rock quality of the M tunnel projects in the source region must adopt the same standard. The relevant standards classify the surrounding rock quality state into S types, where S≥2. In this example, the national standard "Code for Geological Investigation of Water Conservancy and Hydropower Engineering GB50487-2008" is adopted, classifying the tunnel surrounding rock into S=5 types: stable (I), basically stable (II), locally stable (III), unstable (IV), and extremely unstable (V).

[0057] The target domain refers to the N tunnel projects that have not yet been excavated, and the space of their on-site monitoring data is represented by N, where N≥1; in this example, N=1 is preferred, and the target domain is based on the Inner Mongolia Yinchuojiliao No. 6 section tunnel project that has not yet been excavated.

[0058] The same type of TBM is used in the excavation of the source and target tunnels. The TBM types include open-face, single-shield, and double-shield.

[0059] Step 2: Feature parameter extraction based on the shared parameter space of source domain field monitoring data. This step mainly extracts shared features from different tunnel engineering data obtained from monitoring within the source domain to establish an online transfer learning model. The steps include:

[0060] Step 2-1: Segmentation of data streams for effective rock-breaking stages of TBMs in different tunnel projects within the source domain. Data on the TBM's shutdown status within the source domain is zero-valued data, which is invalid and therefore needs to be segmented to extract valid data for the working stages. The segmentation of valid working segments is determined based on the cutterhead rotational speed (RPM) data stream, using a data stream of length n [D1, D2, ..., D...]. n-1 D n For example, the formula for extracting the working segment is as follows:

[0061]

[0062] Based on the above formula, the on-site monitoring data of the Yinsong Water Conveyance Tunnel was divided into working segment data streams, resulting in a total of n = 13200 TBM working segment data.

[0063] The cutter head rotation speed is used as the online segmentation index for the real-time data stream of the working section because the TBM operator sets the cutter head rotation speed value when the TBM starts working. Additionally, the real-time segmentation criterion for the working section is to identify that the product of the cutter head rotation speed (RPM) values ​​in the online data stream is greater than zero.

[0064] Step 2-2: Determine the shared parameter space within the source domain

[0065] By comparing the dimensional characteristics of field monitoring data from different tunnel projects within the source domain, a shared parameter space encompassing all tunnel projects within the source domain is determined. in

[0066] The shared parameter space consists of two parts. One part consists of rock-breaking parameters that are closely related to the TBM cutting of rock mass during tunnel construction. These parameters can be monitored and acquired in tunnels constructed using the TBM method. The other part consists of parameters that are not related to the rock-breaking process but are collected in various tunnel projects within the source domain.

[0067] A shared parameter space is established for the data stream of each effective rock-breaking section. The number M of on-site monitoring data for tunnel engineering in the source domain is determined.

[0068] When the number of TBM-constructed tunnel projects in the source domain is M=1, the rock breaking parameters guided by physical processes are used as the shared parameter space. When the number of tunnel projects in the source domain is M>1, the union of the rock breaking parameters and other parameters is used as the shared parameter space.

[0069] In this example, M=1 field monitoring data points in the Yinsong Water Conveyance Tunnel are used as the source domain for transfer learning. Therefore, rock breaking parameters guided by physical processes are used as the shared parameter space.

[0070] Shared parameters closely related to the rock-breaking process in the source domain include: single-blade thrust, single-blade torque, feed rate, and cutterhead rotation speed. These parameters are the performance and control parameters for TBM cutting rock. The penetration depth can be obtained by dividing the feed rate by the cutterhead rotation speed.

[0071] Therefore, the shared parameter space X of the Yinsong Water Conveyance Tunnel Shard It includes parameters such as cutterhead thrust, cutterhead torque, feed speed, cutterhead rotation speed, and penetration depth, where penetration depth can be obtained by dividing the feed speed by the cutterhead rotation speed.

[0072] Steps 2-3: Extract the shared parameter space X Shard The characteristic values ​​of each parameter in the TBM working segment data stream.

[0073] Shared parameter characteristic values ​​include rock breaking parameter characteristic values ​​and non-rock breaking parameter characteristic values; among them, rock breaking parameter characteristic values ​​include rock breaking parameter statistical characteristic values ​​and excavability index.

[0074] The statistical characteristic values ​​of rock breaking parameters include the average values ​​of single-blade thrust, cutterhead torque, feed speed, and cutterhead rotation speed for each working segment's data stream. Furthermore, the statistical characteristic values ​​of rock breaking parameters include both operational and non-operational rock breaking parameters; among these, the average feed speed, average cutterhead rotation speed, and penetration depth belong to the operational rock breaking parameter statistical characteristic values; while the average single-blade thrust and average cutterhead torque belong to the non-operational rock breaking parameter statistical characteristic values.

[0075] The tunnelability index is a physically-guided indicator related to the rock-breaking process; its calculation uses data from the effective rock-breaking stages in the work segment data stream. For example... Figure 2 As shown, the rock-breaking stage of the TBM corresponds to stage ② in sub-figure (a), which is the process shown in figure (b) where the cutter penetrates the rock mass and the cracks expand to form slag flakes; among them, the excavability index includes the linear fitting slope a and thrust intercept b of the single cutter thrust and penetration data stream, as shown in sub-figure (c), and the linear fitting slope TPI of the single cutter torque and penetration, as shown in sub-figure (d).

[0076] The characteristic value of non-rock breaking parameters is the mean value of each non-rock breaking parameter, such as cutterhead shield pressure, cutterhead shield displacement, belt conveyor speed, and slag particle size.

[0077] Step 3: Construct a thrust prediction model based on LSTM

[0078] like Figure 3 and Figure 5 As shown, a thrust prediction model is constructed based on the LSTM neural network algorithm. Let the current working segment to be predicted be the t-th working segment. Then, the input layer of the thrust prediction model is the shared parameter feature values ​​of the L historical working segments before the t-th working segment; the output layer is the thrust prediction value of the t-th working segment.

[0079] Furthermore, the aforementioned LSTM neural network preferably includes an LSTM layer and a fully connected layer; the input of the LSTM layer is the statistical feature value of the rock-breaking non-operational parameters, the excavability index, and the feature value of the non-rock-breaking parameters of the L historical working segments before the t-th working segment; the input of the fully connected layer is the output of the LSTM layer and the statistical feature value of the rock-breaking operation parameters; the output of the fully connected layer is the thrust prediction value of the t-th working segment.

[0080] Based on the source domain shared parameter feature values ​​extracted in step 2, the constructed thrust prediction model is trained to obtain the trained thrust prediction model. The preferred training method for the thrust prediction model includes the following steps.

[0081] Step 3A-1: Preparation of source domain working segment data streams. Arrange the shared feature working segment data streams of each tunnel project within the source domain in chronological order of excavation time, and then obtain n tunnel working segment data streams.

[0082] Step 3A-2: Loading the source domain working segment data stream. The data stream and features of each tunnel engineering working segment are loaded sequentially to construct a quantitative model of surrounding rock similarity based on LSTM. This is achieved by sequentially reading the working segment data stream and shared features. When the TBM excavates to the t-th working segment, the current data stream and the shared parameter data streams and features of the previous L working segments are loaded.

[0083] Step 3A-3: Use the loaded source domain working segment data to train the constructed thrust prediction model.

[0084] Furthermore, based on the source domain shared parameter feature values ​​extracted in step 2, hyperparameter optimization of the thrust prediction model can be performed during the training process. The specific optimization method includes the following steps.

[0085] Step 3B-1, Sample Library Division: The n source domain working segment data streams loaded in Step 3A-2 are divided into training and validation sets according to a set ratio. In this embodiment, the preferred ratio is 90%:10%.

[0086] Step 3B-2: Design hyperparameters: Design several combinations of hyperparameters involved in the thrust prediction model. The specific design methods are shown in Table 1.

[0087] Step 3B-3, Model Training: Use the training set data to train each hyperparameter combination in the thrust prediction model to obtain several trained thrust prediction models.

[0088] Step 3B-4, Hyperparameter Optimization: Validate each thrust prediction model trained in Step 3B-3 using the validation set, and select the hyperparameter combination with the best validation performance as the optimal hyperparameter. Freeze and fix the optimal hyperparameter. The thrust prediction model with the optimal hyperparameter is also called the optimal thrust prediction model.

[0089] In this embodiment, the optimal hyperparameters after freezing are shown in Table 2.

[0090] Table 1. LSTM Model Structure Parameters and Online Prediction Process Hyperparameter Range

[0091]

[0092] Table 2 Optimal Hyperparameter Combinations for Freezing

[0093]

[0094] Step 4: Construct a surrounding rock condition prediction model based on similarity index.

[0095] Calculate the similarity error between the predicted thrust value and the actual thrust value of the t-th working segment.

[0096] In this example, the rock mass similarity index uses the mean percentage error (MPE) between the predicted thrust and the actual monitoring data stream t, and its calculation formula is as follows:

[0097]

[0098] In the formula, A iP represents the monitoring sequence of the true thrust values ​​of the data stream in the t-th working segment. i Let n represent the thrust prediction sequence of the LSTM model for the data stream of the t-th working segment, where n is the number of samples in the data stream sequence of the t-th working segment.

[0099] Based on the source domain shared parameter feature values ​​extracted in step 2, the similarity error sequence SI corresponding to the n effective rock breaking working sections is obtained; then, combined with the S surrounding rock quality states divided in step 1, a surrounding rock state prediction model based on the similarity index sequence SI and the corresponding surrounding rock classification sequence S is constructed.

[0100] Let the similarity error sequence of the source domain be SI={si1,...,si i ,...,si n Simultaneously, based on the results of the source domain's on-site surrounding rock quality status classification, the surrounding rock classification sequence S = {s0, s1, ..., s} corresponding to the working section data stream is obtained. i ,...,s n A rock mass classification state transition probability model is constructed based on the similarity index sequence SI and the corresponding surrounding rock classification sequence, as shown in the following equation:

[0101] M(s,si)={p0(s0|s,si),...,p i (s i |s,si),...,p n (s n |s,si)}

[0102] In the formula, p i (s i |s,si) is the conditional probability density distribution function when transitioning from state s to any other arbitrary state with similarity index si. This distribution function establishes a non-parametric probability density statistical model by using the similarity index of different rock mass categories during state transitions in the surrounding rock classification sequence.

[0103] In this embodiment, the average percentage MPE error index corresponding to the surrounding rock state category transfer conditions in the data streams of 13,200 working sections of the source-field water diversion tunnel is statistically analyzed, and the probability density distribution under different surrounding rock category transfer conditions is obtained by using the density estimation method, as shown below. Figure 6 As shown, the row direction represents the surrounding rock condition before the transfer, and the column direction represents the surrounding rock condition after the transfer. An example of the prediction process is as follows: when the TBM excavates to the t-th working section, the MPE is obtained based on the LSTM similarity prediction model. t = -70%, and s is known. t-1 =III, enter it into Figure 6 From the probability density distribution function of different columns under Category III, the probability p of the surrounding rock state of the current data stream is obtained.35 >p 34 >p 33 >p 32 Therefore, it is predicted that the surrounding rock of the current data stream is Class V rock mass. In the formula, p 35 p 34 p 33 and p 32 These respectively represent the tags as Figure 6 The probability density values ​​for Class III surrounding rock in the second row, and the subsequent surrounding rock categories after transfer as Class V, Class IV, Class III, and Class II.

[0104] Step 5, online prediction of the surrounding rock quality in the target area, includes the following steps.

[0105] Step 5-1, Initial Excavation of the Target Area: The TBM excavates the target area to the h-th working section, and collects on-site monitoring data for each working section; simultaneously, it conducts on-site surveys of the surrounding rock types of the excavated working sections from the 1st to the h-th section. The preferred value is h = 20.

[0106] Step 5-2: Continue tunneling in the target area: The TBM tunnels from the (h+1)th working segment to the (t)th working segment in the target area, and collects on-site monitoring data for each working segment; where t ≥ h+1.

[0107] Step 5-3: Extract shared parameter feature values ​​of the target domain: Extract shared parameter feature values ​​for each working segment in Step 5-1 and Step 5-2.

[0108] The method for extracting the feature values ​​of shared parameters in the target domain is specifically referred to in step 2-1, which involves segmenting the data stream during the working stage; step 2-2, which involves determining the shared parameter space within the target domain; and step 2-3, which involves extracting the feature values ​​of shared parameters within the target domain.

[0109] In this embodiment, the dimensionality of the on-site monitoring data for the Yinchuo-Jiliao No. 6 section water conveyance tunnel in the target domain is not entirely consistent with the dimensionality of the on-site monitoring data for the Yinsong No. 4 section water conveyance tunnel in the source domain. The target domain parameter space of the Yinchuo-Jiliao No. 6 section is adopted. It shares the characteristic space with the source region of Section 4 of the Yinsong Water Conveyance Tunnel. Intersection: As a parameter space for migration applications. It includes two parameter types: characteristic parameters related to the rock-breaking process and those unrelated. In this example, This includes parameters related to rock breaking, such as cutterhead thrust, cutterhead torque, feed rate, cutterhead rotation speed, and penetration depth. These parameters, directly related to cutting the rock mass, can be easily obtained on-site, thus ensuring the smooth operation of the process. It is a non-empty set.

[0110] Then, the target domain is extracted based on steps 2-3. The feature values ​​related to the rock breaking process in the real-time working segment data stream. The feature values ​​include physical guides for excavability indices related to the rock breaking process, where the excavability indices include the linear fit slope 'a' and thrust intercept 'b' of the single-blade thrust and penetration data stream, and the linear fit slope TPI of the single-blade torque and penetration.

[0111] Statistical features of each real-time working segment data stream in the target domain, such as the average single-tool thrust, average feed rate, and average cutterhead rotation speed, were also extracted. The target domain parameter space was then used. Shared feature space with source domain Intersection: As a parameter space for migration applications.

[0112] Step 5-4, Thrust Prediction for the t-th Working Segment: Input the shared parameter feature values ​​of the L historical working segments preceding the t-th working segment extracted in Step 3 into the LSTM-based thrust prediction model constructed in Step 3 to obtain the thrust prediction value for the t-th working segment in the target domain.

[0113] Step 5-5: Calculate similarity: Calculate the true thrust value of the t-th working segment of the target domain obtained in Step 5-2. Compared with the thrust prediction value obtained in step 5-4 The similarity error SI of the t-th working segment in the target domain was calculated. t .

[0114] Steps 5-6: Online prediction of surrounding rock quality in the (t+1)th working segment: The surrounding rock type of the (t-1)th working segment is taken as the surrounding rock type s before migration. t-1 , will s t-1 and SI t Substituting into the surrounding rock state prediction model P(s) constructed in step 4 t ) = p ij (s t-1 |s t MPE i In ), the search results for s were found. t-1 and SI t The probability values ​​of different surrounding rock states under the conditions; among them, the surrounding rock state corresponding to the maximum probability value is the predicted surrounding rock quality state of the (t+1)th working section; among them, the surrounding rock type of the (t-1)th working section is obtained directly from the survey in step 5-1 or based on the survey data in step 5-1 and obtained through online quality prediction.

[0115] When the TBM excavates to the next working section, repeat steps 5-3 to 5-6 above to obtain the data for the t+1th working section and predict the surrounding rock type and state.

[0116] Based on the above steps, online prediction was performed on 861 real-time data streams during the TBM construction of the No. 6 section of the Yinchuo-Jiliao Railway in the target domain. The predicted sequence of single-blade thrust, the similarity error (MPE) sequence, and the probability prediction sequence of surrounding rock condition were obtained for each online data stream. This example demonstrates the predictive performance of the online transfer learning method using the example of a decrease in the surrounding rock condition category from chainage 65334 to 65325 in the target domain.

[0117] The actual category of the surrounding rock condition in the vicinity of chainages 65334 to 65325 in the target area was determined by the geological engineer according to relevant national standards, and it was changed from Category II to Category IV. Figure 7 As shown in the figure. Subfigure (a) shows the variation process of predicted thrust and actual thrust in the online data streams. The actual thrust values ​​in data streams 1 and 2 are consistent with the predicted values, while data stream 3 shows a larger relative error, with the actual thrust being 19.12% lower than the predicted value. The subsequent three data streams also show similar phenomena, indicating that the thrust required for the excavated surrounding rock is less than that of the excavated section, which means that the strength of the surrounding rock is reduced. Subfigure (b) shows the variation trend of the rock mass similarity index MPE calculated by the online data streams. The MPE value of data stream 3 in the transition zone is -19.12%, while when the surrounding rock state changes from II to IV, the MPE values ​​of data streams 4, 5 and 6 drop to -33.53%, -31.43% and -25.59%, respectively. After inputting the above MPE into the surrounding rock state prediction model, the probability changes under different states are obtained.

[0118] As shown in subfigure (c), Class II rock mass exhibits a trend of first decreasing and then increasing, while MPE first increases and then decreases. Classes III, IV, and V show a trend of first increasing and then stabilizing. However, Class IV rock mass has the highest probability value. Therefore, the online prediction model predicts all rock mass after station 65331 as Class IV surrounding rock. The results show that the real-time prediction method for surrounding rock condition based on online transfer learning can accurately identify the change process of surrounding rock quality.

[0119] In addition, a real-time prediction method for surrounding rock condition based on online transfer learning was used to statistically analyze the prediction results of surrounding rock condition for all working segments of the target domain data stream, such as... Figure 8 As shown in Figure (b), the accuracy rate for identifying the surrounding rock state transition in the source domain of the Yinsong Water Conveyance Tunnel was 86.23%, and the accuracy rate for classifying the surrounding rock in the working section was 86.70%. In the target domain, the accuracy rate for the surrounding rock state transition in the Yinchuo-Jiliao No. 6 section was 70.83%, and the accuracy rate for classifying the surrounding rock in the real-time working section data stream was 83.39%. Furthermore, the confusion matrix for the surrounding rock classification in the target domain tunnel is shown in sub-figure (b), with micro and macro F1 values ​​of 62.09% and 84.62%, respectively. This indicates that the online transfer learning method can achieve ideal prediction results in the target domain.

[0120] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of 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 protection scope of the present invention.

Claims

1. A real-time sensing method for TBM surrounding rock quality based on physics-guided online transfer learning, characterized by: Includes the following steps: Step 1: Delineation of Tunnel Engineering Objects (Source and Target Domains): The source domain refers to the tunnel engineering project that has been excavated and completed, and its on-site monitoring data space is represented by X. S The source domain data space contains on-site monitoring data for M tunnel projects, where M≥1; the surrounding rock quality status of the M tunnel projects in the source domain is divided into S types, where S≥2; The target domain refers to the N tunnel projects that have not yet been excavated, and the space for their on-site monitoring data is denoted by X. T It means that N≥1; Step 2, Feature parameter extraction based on the shared parameter space of source domain field monitoring data, includes the following steps: Step 2-1: Segment the data stream of the effective rock-breaking working stage of the TBM in different tunnel projects within the source domain to obtain n effective rock-breaking working segment data streams; Step 2-2: Determine the shared parameter space X within the source domain. Shard A shared parameter space X is established for the data stream of each effective rock-breaking section. Shard ;X Shard The number M of on-site monitoring data for tunnel engineering in the source domain is determined, specifically as follows: A. When M = 1, X Shard These are the rock-breaking parameters; B. When M > 1, X Shard It is the union of rock-breaking parameters and non-rock-breaking parameters; Steps 2-3: Extract shared parameter feature values: Shared parameter feature values ​​include rock breaking parameter feature values ​​and non-rock breaking parameter feature values; among them, rock breaking parameter feature values ​​include rock breaking parameter statistical feature values ​​and excavability indicators; the excavability indicators are physically guided indicators related to the rock breaking process; the non-rock breaking parameter feature values ​​are the mean values ​​of each non-rock breaking parameter. Step 3: Construct an LSTM-based thrust prediction model: Construct a thrust prediction model based on the LSTM neural network algorithm; Let the current working segment to be predicted be the t-th working segment, then the input layer of the thrust prediction model is the shared parameter feature values ​​of the L historical working segments before the t-th working segment; the output layer is the thrust prediction value of the t-th working segment; Based on the source domain shared parameter feature values ​​extracted in Step 2, train the constructed thrust prediction model to obtain the trained thrust prediction model; Step 4: Construct a surrounding rock condition prediction model based on similarity index: Calculate the similarity error between the predicted thrust value of the t-th working segment and the actual thrust value of the t-th working segment; Based on the source domain shared parameter feature values ​​extracted in Step 2, obtain the similarity error sequence SI corresponding to the n effective rock breaking working segments; Combine this with the S surrounding rock quality states divided in Step 1, and thus construct a surrounding rock condition prediction model based on the similarity index sequence SI and the corresponding surrounding rock classification sequence S. Step 5, Online prediction of surrounding rock quality in the target area, includes the following steps: Step 5-1, Initial Excavation of the Target Area: The TBM excavates the target area to the h-th working section and collects on-site monitoring data for each working section; at the same time, the surrounding rock type of the excavated working sections from the 1st to the h-th section is surveyed in the field. Step 5-2: Continue tunneling in the target area: The TBM tunnels from the (h+1)th working segment to the (t)th working segment in the target area, and collects on-site monitoring data for each working segment; where t ≥ h+1; Step 5-3: Extract shared parameter feature values ​​of the target domain: Extract shared parameter feature values ​​for each working segment in Step 5-1 and Step 5-2; Step 5-4, Thrust Prediction for the t-th Working Segment: Input the shared parameter feature values ​​of the L historical working segments preceding the t-th working segment extracted in Step 3 into the LSTM-based thrust prediction model constructed in Step 3 to obtain the thrust prediction value for the t-th working segment in the target domain. Step 5-5: Calculate similarity: Calculate the true thrust value of the t-th working segment of the target domain obtained in Step 5-2. Compared with the thrust prediction value obtained in step 5-4 The similarity error SI of the t-th working segment in the target domain was calculated. t ; Steps 5-6: Online prediction of surrounding rock quality in the (t+1)th working segment: The surrounding rock type of the (t-1)th working segment is taken as the surrounding rock type s before migration. t-1 , will s t-1 and SI t Substituting into the surrounding rock state prediction model constructed in step 4, we find the value in s t-1 and SI t The probability values ​​of different surrounding rock states under the conditions; among them, the surrounding rock state corresponding to the maximum probability value is the predicted surrounding rock quality state of the (t+1)th working section; among them, the surrounding rock type of the (t-1)th working section is obtained directly from the survey in step 5-1 or based on the survey data in step 5-1 and obtained through online quality prediction.

2. The real-time sensing method for TBM surrounding rock quality based on physical-guided online transfer learning according to claim 1, characterized in that: In step 1, the surrounding rock quality status of the M tunnel projects in the source domain adopts the national standard "Code for Geological Investigation of Water Conservancy and Hydropower Projects GB50487-2008", which classifies the surrounding rock of the tunnels into stable I, basically stable II, locally stable III, unstable IV and extremely unstable V, for a total of S=5 types.

3. The real-time sensing method for TBM surrounding rock quality based on physical-guided online transfer learning according to claim 1, characterized in that: In step 2-2, the rock breaking parameters include cutterhead thrust, cutterhead torque, feed rate, cutterhead rotation speed, and penetration depth; where penetration depth is the ratio of feed rate to cutterhead rotation speed.

4. The real-time sensing method for TBM surrounding rock quality based on physical-guided online transfer learning according to claim 3, characterized in that: In steps 2-3, the excavability indices include the linear fitting slope 'a' and thrust intercept 'b' of the single-blade thrust and penetration data streams, as well as the linear fitting slope TPI of the single-blade torque and penetration.

5. The real-time sensing method for TBM surrounding rock quality based on physical-guided online transfer learning according to claim 1, characterized in that: In steps 2-3, the statistical feature values ​​of rock breaking parameters include the statistical feature values ​​of rock breaking operation parameters and the statistical feature values ​​of rock breaking non-operation parameters; in step 3, the LSTM neural network includes LSTM layers and fully connected layers; the input of the LSTM layer is the statistical feature values ​​of rock breaking non-operation parameters, the excavability index, and the non-rock breaking parameter feature values ​​of the L historical working segments before the t-th working segment; the input of the fully connected layer is the output of the LSTM layer and the statistical feature values ​​of rock breaking operation parameters. The output of the fully connected layer is the thrust prediction value for the t-th working segment.

6. The real-time sensing method for TBM surrounding rock quality based on physical-guided online transfer learning according to claim 1, characterized in that: In step 3, the thrust prediction model is trained based on the source domain shared parameter feature values ​​extracted in step 2. During the training process, the hyperparameters of the thrust prediction model are optimized. The specific optimization method includes the following steps: Step 3B-1, Sample library division: The n source domain working segment data streams loaded in step 3A-2 are divided into training set and validation set according to a set ratio. Step 3B-2: Design hyperparameters: Design several combinations of hyperparameters involved in the thrust prediction model; Step 3B-3, Model Training: Use the training set data to train each hyperparameter combination in the thrust prediction model to obtain several trained thrust prediction models. Step 3B-4, Hyperparameter Optimization: Validate each thrust prediction model trained in Step 3B-3 using the validation set, and select the hyperparameter combination with the best validation performance as the optimal hyperparameter. Freeze and fix the optimal hyperparameter. The thrust prediction model with the optimal hyperparameter is also called the optimal thrust prediction model.

7. The real-time sensing method for TBM surrounding rock quality based on physical-guided online transfer learning according to claim 1, characterized in that: In step 4, let the similarity error sequence of the source domain be SI={si1,...,si i ,...,si n Simultaneously, based on the results of the source domain's on-site surrounding rock quality status classification, the surrounding rock classification sequence S = {s0, s1, ..., s} corresponding to the working section data stream is obtained. i ,...,s n A rock mass classification state transition probability model is constructed based on the similarity index sequence SI and the corresponding surrounding rock classification sequence, as shown in the following equation: M(s,si)={p0(s0|s,si),...,p i (with i |s,si),...,p n (with n |s,si)} In the formula, p i (s i |s,si) is the conditional probability density distribution function when transitioning from the excavated state s to any other arbitrary state with a similarity index of si. This distribution function establishes a non-parametric probability density statistical model by using the similarity index of different rock mass categories during state transitions in the surrounding rock classification sequence.

8. The real-time sensing method for TBM surrounding rock quality based on physical-guided online transfer learning according to claim 1, characterized in that: In step 5-1, h = 20.

Images