Intelligent decision-making method and system for preventing and controlling geological disasters induced by TBM construction disturbance

Abstract

The present invention belongs to the technical field of geological disaster prevention and control in tunnel construction. An intelligent decision-making method and system for preventing and controlling geological disasters induced by TBM construction disturbance are provided, and a database of geological disasters and disaster prevention countermeasures for TBM construction tunnels is constructed. In the database, each feature node and the relationship between feature nodes are determined according to the correlation between the geological characteristics, rock mass characteristics, TBM excavation parameter characteristics and advance prediction characteristics of the TBM tunnel and the correlation with the disaster characteristics; a time series full feature sequence is constructed according to the geological characteristics, rock mass characteristics, advance prediction characteristics and TBM excavation parameter characteristics of the current construction stage, and a more comprehensive intelligent and accurate prediction of the disaster precursor feature prediction results and the corresponding disposal plan is achieved according to the graph neural network.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disaster prevention and control in tunnel construction, and in particular to an intelligent decision-making method and system for preventing and controlling geological disasters induced by TBM construction disturbance. Background Art

[0002] The statements in this section merely provide background art related to the present invention and do not necessarily constitute prior art.

[0003] Tunnel Boring Machine (TBM) has become the preferred construction method for deep and long-distance tunnel construction due to its advantages of efficient excavation, high excavation quality and strong safety. However, due to the special environment of complex and changeable rock geological conditions, it is difficult for TBM to always maintain a good working condition under various geological conditions. The core problem is how to obtain the geological conditions in front of the tunnel in advance, and then optimize the relevant parameters of the tunnel boring machine and establish a scientific and reliable disaster prevention decision-making plan.

[0004] The existing decision-making method for TBM construction parameters mainly relies on qualitative understanding of rock geological conditions and human experience judgment, and lacks quantitative standards. This makes it difficult to effectively match the optimization and adjustment of TBM operating parameters with the changes in rock geological conditions, and it is easy to cause adverse conditions such as slow excavation speed, severe wear of components such as cutterheads and cutters, and even serious consequences such as TBM jamming and whole machine damage. The main reason for these problems is that the current research on decision-making in TBM disaster prevention and control plans is not mature, TBM disaster prevention and control plans have not been systematically classified and summarized, and mainly rely on the experience of on-site technicians for adjustment. When encountering complex strata or changing geological conditions, TBM disaster prevention and control decisions lack scientific theoretical guidance and cannot be reasonably matched with the rock geological conditions in front of the TBM. Therefore, studying and proposing intelligent decision-making methods for the prevention and control of geological disasters induced by TBM construction disturbances is of vital significance for ensuring safe and efficient TBM excavation, solving the problem of low TBM excavation efficiency, and improving the level of tunnel safety construction.

[0005] At present, most TBM prevention and control decisions are based on engineering analogies and on-site expert discussions, but there are many problems: (1) There are a large number of disaster cases involving TBM, but these cases have not been systematically summarized, and no complete database has been established, which makes the engineering analogy relatively extensive. A large number of disaster cases lack effective sorting and integration, making it difficult to effectively apply them when making engineering analogies; (2) For a specific TBM construction tunnel, its geological conditions and the disaster characteristics it faces are unique and are not exactly the same as the disaster characteristics of existing projects; (3) There is no reliable information association between the different disaster types and characteristics encountered by TBM and the corresponding prevention and control decision-making plans.

[0006] Chinese patent CN116562656B discloses an intelligent decision-making method and auxiliary platform for early warning and prevention of geological disasters in tunnel construction. However, it has the following problems: 1) It is mainly used for tunnel construction by drilling and blasting. There are significant differences between drilling and blasting and TBM construction methods in terms of principles and processes, which makes the method not fully applicable to the prevention and control of TBM construction disasters. The drilling and blasting method mainly relies on blasting to break rocks, while the TBM advances by cutting rocks with a rotating cutter head. The difference in construction methods also determines the types of disasters, occurrence mechanisms and prevention and control strategies. 2) In addition to conventional geological analysis, advance forecasting, etc., the method is not fully applicable to the prevention and control of TBM construction disasters. In addition to the fact that conventional means are an effective way to obtain disaster information, for TBM, its excavation parameters, the characteristics of the rock mass in front, and the shield monitoring information are all key parameters for judging the evolution process of disasters and the timing of active prevention and control. The excavation parameters can reflect the working status of the TBM during excavation, the characteristics of the rock mass in front help to understand the upcoming geological conditions, and the shield monitoring information can provide timely feedback on the stress and deformation of the shield. These parameters are crucial for judging the development trend of disasters and choosing the appropriate prevention and control time. However, the above-mentioned patent does not involve these key parameters and cannot provide an effective prevention and control decision-making plan for TBM construction disasters. Summary of the invention

[0007] In order to address the deficiencies in the prior art, the present invention provides an intelligent decision-making method and system for the prevention and control of geological disasters induced by TBM construction disturbance, and constructs a database of geological disasters and disaster prevention countermeasures for TBM construction tunnels. In the database, according to the correlation between the geological characteristics, rock mass characteristics, TBM excavation parameter characteristics and advance prediction characteristics of the TBM tunnel and the correlation with the disaster characteristics, each feature node and the relationship between the feature nodes are determined, and the characteristics in the database are used as node inputs, the disaster characteristics are used as intermediate constraints, the relationship between the rock mass characteristics, the advance prediction characteristics and the disaster characteristics is used as physical constraints, and the treatment plan and countermeasure characteristics are used as prediction points to train a graph neural network. In combination with the geological characteristics of the current construction stage and the TBM excavation parameter characteristics, a time series full feature sequence is constructed, and according to the trained graph neural network, a more comprehensive, intelligent and accurate prediction of disaster precursor characteristics and corresponding treatment plans and countermeasures is achieved.

[0008] In order to achieve the above object, the present invention adopts the following technical solution:

[0009] In a first aspect, the present invention provides an intelligent decision-making method for preventing and controlling geological disasters induced by TBM construction disturbance.

[0010] An intelligent decision-making method for preventing and controlling geological disasters induced by TBM construction disturbance includes the following processes:

[0011] Constructing a database of geological hazards and disaster prevention measures for TBM-constructed tunnels, in which the correlations among geological characteristics, rock mass characteristics, TBM excavation parameter characteristics and advance prediction characteristics and their respective correlations with disaster characteristics are determined;

[0012] The graph neural network is trained with the geological characteristics and the TBM excavation parameter characteristics as input, the disaster characteristics as intermediate constraints, the relationship between rock mass characteristics, advanced prediction characteristics and disaster characteristics as physical constraints, and the disposal plan and countermeasures as prediction outputs. In the graph neural network, the weighted sum of the outputs of neighbor attention, disaster attention and decision attention is used as the information propagation weight;

[0013] According to the geological characteristics, rock mass characteristics, advance prediction characteristics and TBM excavation parameter characteristics of the current construction stage, a time series full feature sequence is constructed. Based on the trained graph neural network, the disaster precursor feature prediction results and corresponding disposal plans and countermeasures are obtained.

[0014] As a further limitation of the first aspect of the present invention, based on the correlation between geological characteristics, rock mass characteristics, TBM excavation parameter characteristics and advance prediction characteristics, and the correlation with disaster characteristics, the characteristics of disposal plans and countermeasures under different disaster precursors, disaster types and disaster intensities are obtained;

[0015] The characteristics of the treatment plan and countermeasures include TBM excavation parameter adjustment, support parameter adjustment and grouting parameter adjustment, and feature nodes are constructed based on geological characteristics, rock mass characteristics, TBM excavation parameter characteristics, advanced prediction characteristics, disaster characteristics and treatment plan and countermeasures characteristics, and the relationship between each feature node is constructed.

[0016] As a further limitation of the first aspect of the present invention, the graph neural network is adaptively optimized based on expert experience and field test verification results.

[0017] As a further limitation of the first aspect of the present invention, the disaster characteristics include: disaster type, occurrence location, scale and shape, precursor information, evolution process, disaster intensity, risk level, occurrence probability, and impact on TBM construction;

[0018] The geological characteristics include: lithology in front of the tunnel, surrounding rock classification, ground stress, rock density, degree of fragmentation, presence of faults, and presence of water bodies;

[0019] The rock mass characteristics include: uniaxial compressive strength, rock brittleness index, rock hardness and elastic strain energy;

[0020] The TBM excavation parameter characteristics include: shield confining pressure, excavation rate, thrust and downtime;

[0021] The advanced prediction features include: imaging results and rock fragmentation degree obtained by seismic wave method, resistivity distribution and water content obtained by induced polarization method, and structural conditions and water content obtained by geological radar method.

[0022] As a further limitation of the first aspect of the present invention, based on each feature node and the relationship between the feature nodes, the correlation between all features is obtained, and the correlation includes the causal relationship and the degree of correlation between the features, and the causal relationship includes: determination, being determined and mutual correlation, and the degree of correlation includes: strong correlation, correlation, weak correlation and irrelevance.

[0023] As a further limitation of the first aspect of the present invention, the graph neural network includes neighbor attention, disaster attention and decision attention;

[0024] The neighbor attention defines the first normalized weight according to the contribution of the neighbor nodes to the central node; the disaster attention considers the importance of different data to the disaster, quantifies the importance of the neighbor nodes according to the preference of the disaster for the neighbor node information, and then defines the second normalized weight; the decision attention considers the preference and method of the disposal schemes and countermeasures taken under different disasters and geological conditions, and then defines the weights of the quantified neighbor nodes and the central node, and then defines the third normalized weight;

[0025] The information propagation weight F is taken as the weighted sum of the first normalized weight, the second normalized weight and the third normalized weight;

[0026] Information dissemination P: ,in, , and To control the parameters of the attention mechanism, , and are the first normalized weight, the second normalized weight and the third normalized weight respectively, Represents the jth word in the i-th feature sentence;

[0027] Information Aggregation A: ;

[0028] Among them, S(·) is the Sigmoid activation function, , and b are trainable parameters, and H(·) is the Hadamard product.

[0029] As a further limitation of the first aspect of the present invention, the relationship between the disaster vector and the prevention and control countermeasures is predicted by calculating the inner product of the disaster vector and the prevention and control countermeasures vector, the disaster vector includes disaster characteristics, geological characteristics, rock mass characteristics, TBM excavation parameter characteristics and advance prediction characteristics, and the prevention and control countermeasures vector is a vectorized prevention and control countermeasure. When calculating each normalized weight, the weights of different features are used to obtain the potential correlation of each feature under different conditions, and the importance of the features is ranked and screened.

[0030] As a further limitation of the first aspect of the present invention, the updating of the graph neural network includes:

[0031] The first loss function is established with the relationship between rock mass characteristics, advance prediction characteristics and disaster characteristics as physical constraint attributes. The geological characteristics, rock mass characteristics, TBM excavation parameter characteristics and advance prediction characteristics are used as multi-node input graph neural networks. The disaster characteristics are predicted in the middle, and the second loss function is calculated with the difference from the actual disaster characteristics. The treatment plan and countermeasures are further predicted in combination with the predicted disaster characteristics, and the third loss function is calculated with the actual results.

[0032] The first loss function, the second loss function, and the third loss function are weighted, the gradient is calculated, and the graph neural network is updated.

[0033] As a further limitation of the first aspect of the present invention, the time series full feature sequence includes: geological characteristics of the current TBM construction face, real-time imaging results of seismic waves of the TBM rock-breaking source, advance prediction detection result information of the induced polarization of the current TBM construction face, real-time excavation parameters of the TBM and rock slag strength collected in situ.

[0034] As a further limitation of the first aspect of the present invention, TBM excavation parameters and monitoring parameters are dynamically acquired, and critical thresholds are established through the percentile method for active prevention and control based on changes in disaster characteristics, rock mass characteristics and forecast characteristics.

[0035] In a second aspect, the present invention provides an intelligent decision-making system for preventing and controlling geological disasters induced by TBM construction disturbances.

[0036] An intelligent decision-making system for preventing and controlling geological disasters induced by TBM construction disturbances, comprising:

[0037] The database generation unit is configured to: construct a database of geological hazards and disaster prevention countermeasures for tunnels constructed by TBM, in which the correlations among geological features, rock mass features, TBM excavation parameter features and advance prediction features and their respective correlations with disaster features are determined;

[0038] The graph neural network training unit is configured to: use the geological characteristics and the TBM excavation parameter characteristics as input, use the disaster characteristics as intermediate constraints, use the relationship between rock mass characteristics, advanced prediction characteristics and disaster characteristics as physical constraints, and use the disposal plan and countermeasures as prediction outputs to train the graph neural network, and in the graph neural network, use the weighted sum of the outputs of neighbor attention, disaster attention and decision attention as the information propagation weight;

[0039] The countermeasure generation unit is configured to: construct a time series full feature sequence according to the geological characteristics, rock mass characteristics, advance prediction characteristics and TBM excavation parameter characteristics of the current construction stage, and obtain the disaster precursor feature prediction results and corresponding disposal plans and countermeasures based on the trained graph neural network.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] The present invention innovatively proposes an intelligent decision-making strategy for the prevention and control of geological disasters induced by TBM construction disturbance, and constructs a database of geological disasters and disaster prevention countermeasures for TBM construction tunnels. In the database, according to the correlation between the geological characteristics, rock mass characteristics, TBM excavation parameter characteristics and advance prediction characteristics of the TBM tunnel and the correlation with the disaster characteristics, each feature node and the relationship between the feature nodes are determined, and the characteristics in the database are used as node inputs, the disaster characteristics are used as intermediate constraints, the relationship between the rock mass characteristics, the advance prediction characteristics and the disaster characteristics is used as physical constraints, and the disposal plan and countermeasure measures characteristics are used as prediction points to train the graph neural network. According to the geological characteristics, rock mass characteristics, advance prediction characteristics and TBM excavation parameter characteristics of the current construction stage, a time series full feature sequence is constructed, and according to the trained graph neural network, a more comprehensive intelligent and accurate prediction of disaster precursor characteristics and corresponding disposal plans is achieved, which effectively improves the accuracy of the selection of prevention and control countermeasures and effectively improves construction safety.

[0042] The present invention innovatively proposes a graph neural network with a three-attention mechanism, including neighbor attention, disaster attention and decision attention; the neighbor attention defines a first normalized weight according to the contribution of the neighbor node to the central node; the disaster attention considers the importance of different data to the disaster, and quantifies the importance of the neighbor node according to the preference of the disaster for the neighbor node information, and then defines the second normalized weight; the decision attention considers the preference and method of the disposal plan and countermeasure measures taken under different disasters and geological conditions, and then defines the weights of the quantized neighbor nodes and the central node, and then defines the third normalized weight; the weighted sum of the first normalized weight, the second normalized weight and the third normalized weight is used as the information propagation weight, and the relationship between the disaster vector and the prevention and control countermeasure vector is predicted by calculating the inner product of the disaster vector and the prevention and control countermeasure vector, the disaster vector includes disaster characteristics, geological characteristics, rock mass characteristics, TBM excavation parameter characteristics and advance prediction characteristics, and the prevention and control countermeasure vector is a vectorized prevention and control countermeasure. When calculating each normalized weight, the weights of different features are used to obtain the potential correlation of each feature under different conditions, and the importance of the features is sorted and screened, thereby improving the efficiency and accuracy of the graph neural network.

[0043] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0045] Figure 1A schematic diagram of a flow chart of an intelligent decision-making method for preventing and controlling geological disasters induced by TBM construction disturbances provided in Example 1 of the present invention;

[0046] Figure 2 A schematic diagram of a triple attention graph neural network model provided in Example 1 of the present invention;

[0047] Figure 3 A schematic diagram of the results of seismic detection provided by Example 1 of the present invention;

[0048] Figure 4 A schematic diagram of the results of induced polarization detection provided in Example 1 of the present invention;

[0049] Figure 5 A schematic diagram of an intelligent decision-making system for preventing and controlling geological disasters induced by TBM construction disturbance provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0050] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0051] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0052] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0053] Embodiment 1:

[0054] This implementation method proposes an intelligent decision-making method for preventing and controlling geological disasters induced by TBM construction disturbances. Figure 1 As shown, the following process is included:

[0055] According to the disaster characteristics, geological characteristics, rock mass characteristics, TBM excavation parameter characteristics and advance prediction characteristics in TBM tunnel construction, the corresponding treatment methods and countermeasures for various disasters are summarized and sorted out, and a database of geological disasters and disaster prevention countermeasures for TBM tunnel construction is constructed;

[0056] Analyze the correlation between the geological characteristics, rock mass characteristics, TBM excavation parameter characteristics and advance prediction characteristics of the TBM tunnel, as well as the correlation with the disaster characteristics, and then analyze the characteristics of the disposal plans and countermeasures under different disaster precursors, types and intensities, including TBM excavation parameter adjustment, support parameter adjustment, grouting parameter adjustment, etc., and construct each characteristic node and node relationship;

[0057] Based on the characteristic nodes and node relationships, a three-attention graph neural network structure is designed, with each feature as the node input, the disaster characteristics as the intermediate constraint point, and the disposal plan and countermeasure characteristics as the prediction point; the rock mass characteristics, the relationship between the detection results and the disaster characteristics are used as the physical constraint attributes, and the established database is used as the data-driven constraint to carry out the training of the three-attention graph neural network structure, and obtain the correlation between the characteristic points for the training of the three-attention graph neural network structure.

[0058] In this implementation, preferably, an active prevention and control decision-making system for geological disaster prevention and control and disaster evolution status judgment in tunnel construction is constructed. Intelligent decision-making is made by using a three-attention graph neural network structure, and the model algorithm is optimized through the prevention and control effect. At the same time, for tunnels where disasters are occurring, the disaster evolution status and the timing of prevention and control are judged through the change rules of excavation parameters and monitoring parameters, which greatly ensures construction safety.

[0059] More specifically, Figure 1 As shown in the figure, through the TBM rock-breaking source (to explore the type, location and morphology of the disaster-causing structure) as the morphological constraint of the induced polarization (to detect the location, size and structure of the disaster water body), the wave field information generated by the TBM rock-breaking source is input into the seismic-machine-rock relationship (to predict mechanical parameters such as compressive or tensile strength), and the rock mass information is generated through the seismic-machine-rock relationship and input into the induced polarization. The output of the seismic-machine-rock relationship and the output of the induced polarization is used to construct a comprehensive perception system of rock mass geological information. According to the comprehensive perception system of rock mass geological information, the geological characteristics, rock mass characteristics, TBM excavation parameter characteristics, advanced prediction characteristics and disaster characteristics of the TBM tunnel are obtained.

[0060] In this implementation, preferably, the TBM excavation parameters have a net thrust F p , Driving torque T , Cutter Radius R and penetration p , from which the engineering rock breaking specific energy SE can be calculated, that is, the energy consumed to break a unit volume of rock mass. The specific calculation formula is as follows:

[0061] (1);

[0062] When the surrounding rock quality deteriorates, the engineering rock breaking specific energy SE will decrease, and then the thrust of the disc cutter will be used. F , Cutter thrust T d 、Number of hobs n , Hob diameter d and penetration depth ρ , the rock mass strength at the TBM excavation face can be estimated σ c , the calculation formula is as follows:

[0063] σ c1 = F / ( c 1 ρ )(2);

[0064] σ c2 = T d / ( c 2 ρ 1.5 )(3);

[0065] c 1= n (4.37 d -1.43)(4);

[0066] c 2= c 1(2.19-0.1 R )(5);

[0067] σ c =( σ c1 + σ c2 ) / 2(6);

[0068] The estimated rock mass strength can be used as geological information for the model to improve the accuracy of model prediction when the geological conditions ahead are unclear.

[0069] In this implementation method, a database of geological hazards and prevention and control decisions for TBM construction tunnels is constructed based on geological characteristics, rock mass characteristics, TBM excavation parameter characteristics, advanced prediction characteristics, disaster characteristics, case libraries and experience libraries. The geological information and physical relationship are constructed based on the relationship analysis of the database of geological hazards and prevention and control decisions for TBM construction tunnels and expert experience. The physical constraints of the three-attention graph neural network are generated based on the geological information and physical relationship. The three-attention graph neural network is optimized and updated through expert experience and field verification. The three-attention graph neural network is used to predict disaster precursor characteristics and disposal plans in real time. The monitoring information is further combined to obtain the timing of active prevention and control decisions. The timing of active prevention and control decisions is input into the comprehensive perception system of rock mass geological information as dynamic information.

[0070] In this implementation, the construction of the database of geological hazards and disaster prevention measures for TBM tunnel construction includes the following processes:

[0071] Collect and process data: classify, summarize and conclude disaster characteristics, geological characteristics, rock mass characteristics, TBM excavation parameter characteristics, and advance forecast characteristics, and record prevention and control measures in a standardized manner, such as converting text materials into editable text, uniformly processing prevention and control measures with high similarity, and using standardized professional terms;

[0072] Data labeling and extraction: Label and extract the processed normalized data, and obtain the knowledge elements related to the prevention and control of geological hazards in tunnel construction, which can be combined with manual assistance to meet the expected effects of the two databases;

[0073] Fusion of target data: cleaning and removing irrelevant information in the processed data, unifying entities with similar expressions into a standard entity through entity linking operations, and realizing batch and automated data processing;

[0074] Construct a countermeasure database: classify disaster response methods by spatial location, size and type of disasters, and classify, summarize and integrate characteristic terms;

[0075] Establish a semantic network: represent the data features in the database with nodes and use edges to represent their relationships.

[0076] In this implementation, the update of the countermeasures database includes: with the continuous development of tunnel engineering examples, the content of the database is constantly updated, and better treatment countermeasures may be generated for the same type of disasters, making the countermeasures in the database more complete; at the same time, new data samples related to the prevention and control of geological disasters in tunnel construction are obtained, and knowledge framework sorting, knowledge annotation, knowledge extraction and knowledge fusion processing are carried out to optimize and iterate the knowledge system of prevention and control of geological disasters in tunnel construction.

[0077] In this implementation, the countermeasure database is updated by combining the experts' field experience and theoretical knowledge with the computer's data processing capabilities. For example, new strategies in the field of disaster prevention and control provided by experts are classified, summarized and stored through the computer system.

[0078] In this implementation, a multi-dimensional interactive attention knowledge graph neural network (MI-KGNN) is used. Its basic idea is to explore the user-item interaction matrix and semantic network to learn the characteristics of users and items to predict the possibility of future interactions. In this implementation, the user is the disaster model (including disaster characteristics, geological characteristics, rock mass characteristics, TBM excavation parameter characteristics, and advance prediction characteristics). Based on the four assumptions of collaborative filtering (similar disaster models, similar disasters related to the same project, similar projects with the same disaster characteristics, and similar disaster characteristics with the same prevention and control countermeasures), the graph neural network is used to explore structural information (interaction topology) and semantic information (interaction weight) to learn more reliable node representations;

[0079] Information propagation: This method allows nodes to propagate their embeddings through paths in the graph, that is, information propagation first generates three data streams that flow to the three attention mechanisms.

[0080] Information aggregation: The information transmitted by neighbors is aggregated with the node’s own information through the aggregator to obtain a personalized representation of the node. The design of the aggregator increases the interaction between the central node and the neighboring nodes, so that the parameter update direction of the central node is directly related to its neighbors. The information aggregation here is implemented using the triple attention mechanism.

[0081] In this implementation, the MI-KGNN model predicts the relationship between disasters and prevention and control measures by calculating the inner product of the disaster vector and the prevention and control countermeasure vector. Here, the disaster vector is a vector that integrates disaster characteristics, geological characteristics, rock mass characteristics, TBM excavation parameter characteristics, and advance prediction characteristics; the prevention and control countermeasure vector is a vectorized prevention and control countermeasure. For numerical data, one type of data can be regarded as a dimension for expression, and multiple types of data can be represented in multiple dimensions. For categorical data, such as rock mass hardness grade, 0 and 1 encoding can be used, and different categories can be represented by different combinations. For text data, such as prevention and control countermeasures, the text can be split and encoded. For example, different depths of drilling holes are encoded differently, and different methods such as drilling and grouting are also encoded. In this implementation, considering that there are already nonlinear activation functions in the triple attention graph neural network structure, and the calculation of the relative importance of nodes is based on the transmission path type and transmission information, the activation function can be pruned to optimize the model.

[0082] In this implementation, preferably, during the entire TBM construction process, based on geological characteristics, combined with the current construction plan and the dynamic characteristic parameter information obtained during the TBM excavation process, a time series full feature sequence is constructed, and the trained graph neural network is used to predict disaster precursor characteristics and disposal plans in real time.

[0083] In this implementation, preferably, the uncertainty provided by the prediction results of the disaster precursor characteristics and the disposal plan is combined to reflect the possible risks of landslides, machine jams, and sudden water inrush ahead, while dynamically analyzing TBM excavation parameters and rock mass parameters, shield surrounding rock deformation monitoring and other information to achieve early warning reminders throughout the entire process.

[0084] In this implementation, preferably, expert experience and field test verification results are used as optimization methods to adaptively optimize the graph neural network to make it more suitable for the current TBM construction tunnel and improve the prediction effect.

[0085] In the present implementation, preferably, the disaster characteristics include: type of disaster, location, scale and form, precursor information, evolution process, disaster intensity, risk level, probability of occurrence, impact on TBM construction, etc.; the geological characteristics include the lithology in front of the tunnel, surrounding rock classification, ground stress, rock density, degree of fragmentation, whether there are faults, whether there are water bodies, etc.; the rock characteristics include: uniaxial compressive strength, rock brittleness index, rock hardness and elastic strain energy; the excavation parameter characteristics include: shield confining pressure, excavation rate, thrust and downtime; the advanced prediction characteristics include: imaging results and rock fragmentation degree obtained by seismic wave method, resistivity distribution and water content obtained by induced polarization method, and structural conditions and water content obtained by geological radar method.

[0086] In this implementation, preferably, the disaster characteristics are common disaster characteristics for TBM construction tunnels, and their types include but are not limited to landslides, machine jams, and sudden water gushing.

[0087] In the present implementation, preferably, the relevant data in the TBM construction tunnel geological hazards and disaster prevention countermeasures database are structured, and the structured data are normalized through data mapping and entity matching, and the results of the normalization processing are used to preliminarily establish node and node relationships.

[0088] In the present implementation, preferably, the correlation and correlation degree between all features are analyzed based on the node-to-node relationships established in the database, such as the correlation between lithology in geological features, surrounding rock classification and strength in rock mass features, degree of fragmentation in advanced prediction features, and excavation efficiency in TBM excavation parameter features. The analyzed correlation includes the causal relationship and correlation degree between the features; the causal relationship includes: determination, being determined, and mutual correlation, and the correlation degree includes: strong correlation, correlation, weak correlation, and no correlation; the analysis of the above-mentioned node relationships comes from standards, theoretical research, and field experience.

[0089] In this implementation, preferably, the triple attention graph neural network structure includes:

[0090] Adopting neighbor attention, the first normalized weight is defined according to the contribution of neighbor nodes to the central node; the central node refers to the main object of current attention. Under different conditions, the relationship between the central node and the neighbor nodes may change dynamically. Moreover, the central node is the focus of information aggregation, the core of the relationship network and the basis for decision prediction. By analyzing the relationship between the central node and the neighbor nodes, we can better understand the structure and pattern in the data and improve the performance and accuracy of the model. The normalized weight refers to the processing of the attention weight values ​​so that they satisfy the weight value in the [0, 1] interval and the sum of all weight values ​​is 1.

[0091] Neighborhood attention, specifically, includes:

[0092] (7);

[0093] in, represents a layer of forward network, m is the number of feature sentences in the data set, and n is the feature sentence The number of words in represents the i-th feature sentence, represents the jth word in the i-th feature sentence, Indicates the label corresponding to this feature sentence;

[0094] (8);

[0095] in: Indicates the importance of each word in the feature of neighbor attention, For After normalization, the result will be Convert to probability distribution and optimize the processing of three types of attention by the triple-attention graph neural network model.

[0096] The disaster attention considers the importance of different data to the disaster, quantifies the importance of neighbor nodes according to the disaster's preference for neighbor node information, and further defines the second normalized weight;

[0097] Disaster attention, specifically, includes:

[0098] (9);

[0099] (10);

[0100] in, represents a layer of forward network; The importance of each word representing the characteristics of disaster attention, For The result after normalization.

[0101] The decision attention considers the preferences and methods of disposal plans and countermeasures taken under different disaster and geological conditions, defines the weights of quantified neighbor nodes and central nodes, and further defines the third normalized weight;

[0102] Decision-making attention, specifically, includes:

[0103] (11);

[0104] (12);

[0105] in, represents a layer of forward network; Indicates the importance of each word in the feature of decision attention, For The result after normalization.

[0106] The weighted sum of the first normalized weight, the second normalized weight and the third normalized weight is used as the information propagation weight, and the relationship between the disaster vector and the prevention and control countermeasures vector is predicted by calculating the inner product of the disaster vector and the prevention and control countermeasures vector. The disaster vector includes disaster characteristics, geological characteristics, rock mass characteristics, TBM excavation parameter characteristics and advance prediction characteristics, and the prevention and control countermeasures vector is a vectorized prevention and control countermeasure.

[0107] Information dissemination P:

[0108] (13);

[0109] (14);

[0110] in, and To control the parameters of the attention mechanism, by adjusting the parameters and , the impact of the three attention mechanisms can be fine-tuned to improve the model's adaptability to different data.

[0111] Information aggregation A (i.e. as the output of the triple attention graph neural network):

[0112] (15);

[0113] Among them, S(·) is the Sigmoid activation function, , and b are trainable parameters, and H(·) is the Hadamard product.

[0114] More specifically, Figure 2 As shown in the figure, according to the geological characteristics, rock mass characteristics, TBM excavation parameter characteristics and advance prediction characteristics input by the input layer, disaster attention calculation, neighbor attention calculation and decision attention calculation are performed respectively, and the three attention fusion modules are used to integrate and process the data combined with geological prior information and physical prior information, and finally the disaster prevention and control countermeasures are output through the output layer.

[0115] In this implementation, each normalized weight is obtained by using the weights of different features, and the potential associations of these features under different conditions are explored, and the importance of the features is sorted and screened in time to improve the efficiency and accuracy of the model.

[0116] In this implementation, the potential associations of these features under different conditions are explored, including: calculating the Pearson correlation coefficient. When building a prediction model, the Pearson correlation coefficient can help select appropriate predictor variables. If a variable has a high Pearson correlation coefficient with the target variable, it may be a potentially important predictor variable.

[0117] In this implementation, preferably, the updating of the triple attention graph neural network structure includes:

[0118] (1) Taking the relationship between rock mass characteristics, detection results and disaster characteristics as physical constraint attributes: that is, establishing a constraint function. Through the analysis and modeling of geological priors and physical relationships, the type and location of the disaster can be preliminarily obtained by analyzing the rock mass and detection results. For example, the seismic wave imaging results combined with the face geology and the geological conditions of the preliminary survey data can infer the possibility of collapse in front of the tunnel. This is used as loss function 1 (i.e., the first loss function) to perform effective constraints.

[0119] (2) Using the established database as a data-driven constraint: that is, using geological characteristics, rock mass characteristics, TBM excavation parameter characteristics and advance prediction characteristics as multi-node input graph neural networks, predicting disaster characteristics in the middle, and calculating loss function 2 (i.e., the second loss function) based on the difference between the disaster characteristics and the actual disaster characteristics, and continuing to predict the disaster vector and prevention and control countermeasure characteristics based on the predicted disaster characteristics, and calculating loss function 3 (i.e., the third loss function) based on the actual results; based on the loss function 1, loss function 2 and loss function 3 established above, analyzing the order of magnitude of the three loss functions, equalizing the three loss functions by weighting, and calculating the gradient, and finally updating the three-attention graph neural network structure.

[0120] In the present implementation, preferably, the time series full feature sequence includes: geological characteristics of the current TBM construction face, real-time imaging results of seismic waves of the TBM rock-breaking source, advance prediction detection result information of induced polarization of the current TBM construction face, real-time TBM excavation parameter characteristics, rock slag strength collected in situ, and other characteristics that all change with time series.

[0121] In this implementation, preferably, the real-time imaging result of the seismic wave of the TBM rock-breaking source is to use the strong vibration in the rock-breaking process of TBM construction as the source for propagation, and the seismic data is collected by the detector. Since it is detected during the TBM construction process, real-time imaging results can be obtained, which is one of the key information of the full feature sequence of the time series. In this implementation, preferably, the graph neural network predicts disaster precursor characteristics and disposal plans in real time. The advantage is that the graph neural network can establish almost real-time prediction results, combined with multi-parameter monitoring during the TBM construction process, disaster precursor characteristics and disposal plans can be obtained in real time.

[0122] In this implementation, active prevention and control opportunity prediction is also performed: a recurrent neural network (RNN) is used, neurons are repeatedly connected, and the input layer is the characteristic parameters of TBM excavation parameters, rock mass characteristic parameters, advance prediction characteristic parameters, geological characteristic parameters, and the parameters collected by monitoring at the current moment (i.e. Figure 1 The hidden layer is the aforementioned three neural network model, and the output layer is the active prevention and control opportunity. The neurons in the hidden layer not only receive the input data at the current moment, but also receive the output from the neurons in the hidden layer at the previous moment. RNN stores and transmits information through the cyclic connection of the hidden layer. This memory ability enables it to process sequence data with time order. When processing text data, it can remember the semantic information of previous words, so as to better understand the meaning of the entire sentence.

[0123] In this implementation, the optimization method based on expert experience and field test verification is preferred. Its connotation is that the early graph neural network obtains relationships by mining the disaster and disaster prevention database composed of all TBM construction tunnels. Its advantage is that it can quickly and reliably obtain preliminary prediction results. By combining the current TBM construction tunnel and TBM performance and analyzing new data, it can be made more suitable for the current TBM construction tunnel and improve the prediction effect.

[0124] Optionally, in some other implementations, through the above-mentioned time series full feature sequence, a comprehensive field of multiple parameters plus time dimension can be constructed, and the intrinsic relationship between multiple parameters and their evolution trend over time can be deeply analyzed, and the possible evolution trend of disasters and additional disasters can be analyzed, so as to provide real-time disaster prevention and control basis, reliably judge the prevention and control timing, and ensure that prevention and control measures can be implemented at the most appropriate time.

[0125] For example, if the confining pressure continues to increase over a period of time, while the excavation rate gradually decreases, and the degree of extrusion deformation also tends to increase, then we can infer that the disaster may be developing in a more serious direction, and additional disasters such as landslides may occur. Disaster prevention and control can be carried out in time by adjusting excavation parameters, strengthening support, etc.; if drilling and coring were carried out on site before the disaster occurred, the core parameters can be imported into the comprehensive field as known information to improve the accuracy of the analysis.

[0126] Intelligent decision-making is made based on the comprehensive field of multiple parameters and time dimensions collected and the disaster prevention and control decision-making database, and disaster prevention and control is carried out quickly and effectively. The multiple parameters collected cover key information related to disasters in all aspects. These parameters may come from different monitoring equipment and systems. Each parameter has its unique meaning and value. For example, it may include geological structure parameters, such as rock hardness, density, porosity, etc. These parameters can reflect the characteristics of the geological body itself and play an important role in judging the potential risk of disasters. The comprehensive field of the time dimension adds the key factor of time on the basis of considering these parameters.

[0127] As time goes by, various parameters will change. By integrating and analyzing the parameters collected at different time points, a comprehensive field containing time series information can be constructed. In this comprehensive field, the evolution of parameters over time can be observed, such as whether the hardness of rocks in a certain area gradually decreases over a period of time, or whether the excavation speed during construction fluctuates abnormally in a certain period of time. This kind of time dimension information can help us better predict the possibility of disasters and their development trends. Based on such a comprehensive field composed of multiple parameters and time dimensions and a disaster prevention and control decision library, intelligent decision-making can be made, which can quickly process and analyze the data in the comprehensive field, and at the same time, combined with the relevant knowledge in the decision library, quickly judge the current disaster risk situation and formulate corresponding prevention and control measures.

[0128] For example, when the intelligent algorithm detects that the hardness of rock in a certain area drops sharply in a short period of time, combined with the response strategies for such situations in the decision library, the system will make quick decisions and recommend construction personnel to take immediate reinforcement measures, such as adding support structures, adjusting excavation speeds, etc., thereby achieving rapid and effective disaster prevention and control, and minimizing the losses that may be caused by disasters.

[0129] This implementation method takes a TBM tunnel project as an example, specifically:

[0130] Geological characteristics, including: adverse geology that may be encountered in the tunnel crossing area, mainly high-stress rock bursts, dense joint zones, alteration zones, high ground temperatures, and sudden water inrush;

[0131] The rock mass characteristics include: the cave rock mass is gneiss with gneissic structure, mainly composed of quartz, feldspar and biotite, the rock mass is relatively broken, and joints and fissures are well developed;

[0132] Advanced prediction features: using seismic detection, such as Figure 3As shown (showing the X-axis and Z-axis, in meters), the seismic wave velocity varies greatly, and there are strong positive and negative reflection anomalies. It is inferred that there are developed joints and fissures in front of the tunnel face, the surrounding rock is relatively broken, and there is a risk of block falling in some areas. Using induced polarization detection, such as Figure 4 As shown (including x, y, and z axes, in meters), the resistivity value of the rock mass in front of the TBM is generally high, and no obvious low-resistance abnormal area is found. It is inferred that the groundwater in the surrounding rock of this area is not generated and the risk of water gushing is low.

[0133] Determine the cause of the machine jam: Based on the geological characteristics, rock characteristics and advance prediction characteristics combined with on-site construction, it can be concluded that the rock mass in front of the cutter disc has developed joints and fissures, and the rock mass is mostly cut into blocks. Under the action of the dense joint zone and high ground stress, the surrounding rock near the cutter disc is squeezed and crushed, forming a large number of loose stones. After excavation, the surrounding rock in front continues to collapse in the absence of timely support, and large and small stones are mixed. A large amount of collapsed bodies accumulate around the cutter disc, and large pieces of gravel block the cutter holes and scraping holes, causing the cutter disc to jam; the broken rock mass in the shield area under the action of high ground stress causes the surrounding rock to deform and converge, causing the machine to jam, resulting in the shield jamming.

[0134] The generated disposal plans and countermeasures may include: continuously monitoring the situation inside the tunnel, collecting information on the geological conditions of the surrounding rock inside the shield and in front of the face, analyzing the causes of the jam, and using advanced geological prediction methods to continuously explore the geological conditions ahead. During the excavation and expansion of the pilot tunnel, the surrounding rock conditions are monitored in real time. The pilot tunnel method is one of the methods for TBM jams to escape. A small pilot tunnel is excavated above the shield to the face. According to the space requirements for pipe roof construction and the needs for shield escape, the pilot tunnel is expanded in a circular manner to form a pipe roof studio and release the shield simultaneously. After the advanced pipe roof construction is completed, the cutterhead is cleaned and TBM excavation is resumed.

[0135] Embodiment 2:

[0136] like Figure 5 As shown, this implementation method provides an intelligent decision-making system for preventing and controlling geological disasters induced by TBM construction disturbance, including:

[0137] The database generation unit is configured to: construct a database of geological hazards and disaster prevention countermeasures for tunnels constructed by TBM, in which the correlations among geological features, rock mass features, TBM excavation parameter features and advance prediction features and their respective correlations with disaster features are determined;

[0138] The graph neural network training unit is configured to: use the geological characteristics and the TBM excavation parameter characteristics as input, use the disaster characteristics as intermediate constraints, use the relationship between rock mass characteristics, advanced prediction characteristics and disaster characteristics as physical constraints, and use the disposal plan and countermeasures as prediction outputs to train the graph neural network, and in the graph neural network, use the weighted sum of the outputs of neighbor attention, disaster attention and decision attention as the information propagation weight;

[0139] The countermeasure generation unit is configured to: construct a time series full feature sequence according to the geological characteristics, rock mass characteristics, advance prediction characteristics and TBM excavation parameter characteristics of the current construction stage, and obtain the disaster precursor feature prediction results and corresponding disposal plans and countermeasures based on the trained graph neural network.

[0140] It is understandable that the above-mentioned units can be separately or completely combined into one or several other units to constitute, or one (some) of the units can be further divided into multiple smaller units in function to constitute, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present application. The above-mentioned units are divided based on logical functions. In practical applications, the functions of one unit can also be implemented by multiple units, or the functions of multiple units can be implemented by one unit. In other embodiments of the present application, the system may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented by the collaboration of multiple units.

[0141] According to another embodiment of the present application, the system described in this embodiment can be constructed, and the method of Example 1 of the present application can be implemented by running a computer program (including program code) capable of executing the steps involved in the corresponding method described in Example 1 on a general-purpose computing device such as a computer, which includes processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM). The computer program can be recorded on, for example, a computer-readable recording medium, and loaded into the above-mentioned computing device through the computer-readable recording medium and run therein.

[0142] A person skilled in the art can appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0143] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from a website site, a computer, a server or a data center to another website site, a computer, a server or a data center by wired (e.g., coaxial cable, optical fiber, digital line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by the computer or a data processing device such as a server, a data center, etc. that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)), etc.

[0144] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent decision-making method for preventing and controlling geological disasters induced by TBM construction disturbance, characterized in that: The process includes: Constructing a database of geological hazards and disaster prevention measures for TBM-constructed tunnels, in which the correlations among geological characteristics, rock mass characteristics, TBM excavation parameter characteristics and advance prediction characteristics and their respective correlations with disaster characteristics are determined; The graph neural network is trained with the geological characteristics and the TBM excavation parameter characteristics as input, the disaster characteristics as intermediate constraints, the relationship between rock mass characteristics, advanced prediction characteristics and disaster characteristics as physical constraints, and the disposal plan and countermeasures as prediction outputs. In the graph neural network, the weighted sum of the outputs of neighbor attention, disaster attention and decision attention is used as the information propagation weight; According to the geological characteristics, rock mass characteristics, advance prediction characteristics and TBM excavation parameter characteristics of the current construction stage, a time series full feature sequence is constructed, and the disaster precursor feature prediction results and corresponding disposal plans and countermeasures are obtained based on the trained graph neural network; The graph neural network includes neighbor attention, disaster attention and decision attention; The neighbor attention defines a first normalized weight according to the contribution of the neighbor node to the central node; The disaster attention considers the importance of different data to the disaster, quantifies the importance of neighbor nodes according to the disaster's preference for neighbor node information, and further defines the second normalized weight; The decision attention considers the preferences and methods of disposal plans and countermeasures taken under different disaster and geological conditions, defines the weights of quantified neighbor nodes and central nodes, and further defines the third normalized weight; The information propagation weight F is taken as the weighted sum of the first normalized weight, the second normalized weight and the third normalized weight; Information dissemination P: ,in, , and To control the parameters of the attention mechanism, , and are the first normalized weight, the second normalized weight and the third normalized weight respectively, Represents the jth word in the i-th feature sentence; Information Aggregation A: ; Among them, S(·) is the Sigmoid activation function, , and b are trainable parameters, and H(·) is the Hadamard product.

2. The intelligent decision-making method for preventing and controlling geological disasters induced by TBM construction disturbance according to claim 1, characterized in that: Based on the correlation between geological characteristics, rock mass characteristics, TBM excavation parameter characteristics and advance prediction characteristics, as well as the correlation with disaster characteristics, the characteristics of disposal plans and countermeasures under different disaster precursors, disaster types and disaster intensities are obtained; The characteristics of the treatment plan and countermeasures include TBM excavation parameter adjustment, support parameter adjustment and grouting parameter adjustment, and feature nodes are constructed based on geological characteristics, rock mass characteristics, TBM excavation parameter characteristics, advanced prediction characteristics, disaster characteristics and treatment plan and countermeasures characteristics, and the relationship between each feature node is constructed.

3. The intelligent decision-making method for preventing and controlling geological disasters induced by TBM construction disturbance according to claim 1 or 2, characterized in that: The disaster characteristics include: disaster type, location, scale and form, precursor information, evolution process, disaster intensity, risk level, occurrence probability, and impact on TBM construction; The geological characteristics include: lithology in front of the tunnel, surrounding rock classification, ground stress, rock density, degree of fragmentation, presence of faults, and presence of water bodies; The rock mass characteristics include: uniaxial compressive strength, rock brittleness index, rock hardness and elastic strain energy; The TBM excavation parameter characteristics include: shield confining pressure, excavation rate, thrust and downtime; The advanced prediction features include: imaging results and rock fragmentation degree obtained by seismic wave method, resistivity distribution and water content obtained by induced polarization method, and structural conditions and water content obtained by geological radar method.

4. The intelligent decision-making method for preventing and controlling geological disasters induced by TBM construction disturbance according to claim 1 or 2, characterized in that: Based on the feature nodes in the database and the relationship between the feature nodes, the correlation between all features is obtained, and the correlation includes the causal relationship and the degree of correlation between the features. The causal relationship includes: determination, being determined and mutual correlation, and the degree of correlation includes: strong correlation, correlation, weak correlation and no correlation.

5. The intelligent decision-making method for preventing and controlling geological disasters induced by TBM construction disturbance according to claim 1, characterized in that: The relationship between the disaster vector and the prevention and control countermeasures is predicted by calculating the inner product of the disaster vector and the prevention and control countermeasures vector, wherein the disaster vector includes disaster characteristics, geological characteristics, rock mass characteristics, TBM excavation parameter characteristics and advance prediction characteristics; The prevention and control countermeasure vector is a vectorized prevention and control countermeasure. According to the weights of different features, the potential correlation of each feature under different conditions is obtained, and the importance of the features is ranked and screened.

6. The intelligent decision-making method for preventing and controlling geological disasters induced by TBM construction disturbance according to claim 1, characterized in that: The update of the graph neural network includes: The first loss function is established with the relationship between rock mass characteristics, advance prediction characteristics and disaster characteristics as physical constraint attributes. The geological characteristics, rock mass characteristics, TBM excavation parameter characteristics and advance prediction characteristics are used as multi-node input graph neural networks. The disaster characteristics are predicted in the middle, and the second loss function is calculated with the difference from the actual disaster characteristics. The treatment plan and countermeasures are further predicted in combination with the predicted disaster characteristics, and the third loss function is calculated with the actual results. The first loss function, the second loss function, and the third loss function are weighted, the gradient is calculated, and the graph neural network is updated.

7. The intelligent decision-making method for preventing and controlling geological disasters induced by TBM construction disturbance according to claim 1, characterized in that: The full time series includes: geological characteristics of the current TBM construction face, real-time imaging results of seismic waves from the TBM rock-breaking source, advance prediction detection results of induced polarization of the current TBM construction face, real-time excavation parameters of the TBM, and rock slag strength collected in situ.

8. The intelligent decision-making method for preventing and controlling geological disasters induced by TBM construction disturbance according to claim 1, characterized in that: Dynamically obtain TBM excavation parameters and monitoring parameters, and establish critical thresholds for active prevention and control through the percentile method based on changes in disaster characteristics, rock mass characteristics and forecast characteristics.

9. An intelligent decision-making system for preventing and controlling geological disasters induced by TBM construction disturbance, characterized in that: include: The database generation unit is configured to: construct a database of geological hazards and disaster prevention countermeasures for tunnels constructed by TBM, in which the correlations among geological features, rock mass features, TBM excavation parameter features and advance prediction features and their respective correlations with disaster features are determined; The graph neural network training unit is configured to: use the geological characteristics and the TBM excavation parameter characteristics as input, use the disaster characteristics as intermediate constraints, use the relationship between rock mass characteristics, advanced prediction characteristics and disaster characteristics as physical constraints, and use the disposal plan and countermeasures as prediction outputs to train the graph neural network, and in the graph neural network, use the weighted sum of the outputs of neighbor attention, disaster attention and decision attention as the information propagation weight; The countermeasure generation unit is configured to: construct a time series full feature sequence based on the geological characteristics, rock mass characteristics, advance prediction characteristics and TBM excavation parameter characteristics of the current construction stage, and obtain the disaster precursor feature prediction results and corresponding disposal plans and countermeasures based on the trained graph neural network; The graph neural network includes neighbor attention, disaster attention and decision attention; The neighbor attention defines a first normalized weight according to the contribution of the neighbor node to the central node; The disaster attention considers the importance of different data to the disaster, quantifies the importance of neighbor nodes according to the disaster's preference for neighbor node information, and further defines the second normalized weight; The decision attention considers the preferences and methods of disposal plans and countermeasures taken under different disaster and geological conditions, defines the weights of quantified neighbor nodes and central nodes, and further defines the third normalized weight; The information propagation weight F is taken as the weighted sum of the first normalized weight, the second normalized weight and the third normalized weight; Information dissemination P: ,in, , and To control the parameters of the attention mechanism, , and are the first normalized weight, the second normalized weight and the third normalized weight respectively, Represents the jth word in the i-th feature sentence; Information Aggregation A: ; Among them, S(·) is the Sigmoid activation function, , and b are trainable parameters, and H(·) is the Hadamard product.

Images