A method and system for constructing a digital scene of a immersed tube tunnel construction process

Abstract

The present application relates to the field of geotechnical engineering intelligent technology, especially to a method and system for constructing a digital scene of a immersed tunnel construction process, the method comprising the following steps: establishing a construction event set, and configuring an event state code and an event topology mask; identifying the event state code of the current construction, and determining the event topology mask; converting a construction object into a node, converting an object relationship into an edge, and constructing an event constraint dynamic scene graph; establishing a six-degree-of-freedom mechanism baseline prediction model, and obtaining a mechanism baseline prediction value; obtaining a residual prediction value to obtain a final prediction value; calculating a construction risk value, when the construction risk value exceeds a set threshold, taking a construction control parameter as a decision variable, and solving to obtain a construction adjustment scheme; and constructing a digital construction scene of an immersed tunnel. The present application improves the state prediction reliability, risk identification advancement, and construction parameter feedback capability in the immersed tunnel construction process.

Description

Technical Field

[0001] This invention relates to the field of intelligent geotechnical engineering technology, and in particular to a method and system for constructing a digital scene of immersed tunnel construction process. Background Technology

[0002] Immersed tunnels are a type of tunnel structure widely used in the construction of cross-river and cross-sea passages. Their construction process typically includes key stages such as segment prefabrication, undocking, floating, sinking, docking, joint treatment, and backfilling. Among these, the floating, sinking, and docking of segments are the stages with the highest concentration of risks and the most stringent control requirements. During the floating stage, the segments are affected by a combination of factors, including tugboat traction, water flow velocity and direction, tide level, wind speed, channel boundaries, and the segment's own inertia, making them prone to yaw, lateral drift, and path deviation. During the sinking stage, the segment's attitude is affected by ballast water volume, ballast change rate, buoyancy difference, hydrodynamic disturbances, stress on the lifting points, and near-bottom effect, making it prone to longitudinal and transverse inclination deviations, abnormal sinking speeds, and near-bottom attitude instability. During the docking stage, there is a strong coupling relationship between the segment end face gap, axis deviation, waterstop pressure, and docking speed. Improper control may lead to abnormal docking gaps, uneven stress on the waterstop, or waterproofing risks at the joint. Existing methods for managing immersed tunnel construction have begun to incorporate Building Information Modeling (BIM), Geographic Information System (GIS), drone inspections, edge computing, sensor monitoring, digital twin platforms, and construction information management systems to achieve integrated construction process information and 3D visualization. These methods improve the efficiency of construction data management and, to a certain extent, enable real-time display of on-site data.

[0003] However, existing methods still have the following shortcomings: First, most existing digital construction scenarios are based on static model displays or conventional data associations, typically simply binding pipe sections, equipment, monitoring points, and environmental data, lacking the expression of changes in object relationships under different construction events. The immersed tunnel construction process has obvious stages, and the dominant relationships between different construction objects are not the same in the floating, sinking, near-bottom adjustment, and docking stages. For example, the key relationship in the floating stage is the relationship between tugboat traction, water flow direction, and pipe section yaw; the key relationship in the sinking stage is the relationship between ballast changes, vertical sinking speed, and pipe section attitude; and the key relationship in the docking stage is the relationship between end face gap, waterstop pressure, and attitude fine-tuning. If a fixed topological relationship is still used to express the construction scenario, it is difficult to accurately reflect the dynamic transmission path of construction risks. Second, existing construction status prediction methods mostly use ordinary neural networks, genetic algorithms to optimize neural networks, recurrent neural networks, graph neural networks, or multi-algorithm fusion models to directly fit the monitoring data. These methods typically use historical monitoring data as input and attitude deviation, displacement deviation, or risk level as output. While they can learn certain statistical patterns, they fail to adequately consider the physical relationships between tunnel segment movement, ballast adjustment, tugboat traction, and hydrodynamic disturbances. When the construction sample is limited, the hydrological environment changes significantly, or construction events shift, purely data-driven models are prone to prediction drift, leading to inconsistencies between predictions and engineering mechanisms. Third, existing construction early warning methods often employ threshold judgment, comparing monitored values ​​with preset control limits and triggering an early warning when the value exceeds the threshold. This method can detect anomalies that have already occurred, but it struggles to identify abnormal trends in advance and determine the causes of anomalies. For example, an increase in the tunnel segment's lateral tilt angle may be caused by uneven local ballast, lateral water flow disturbance, tugboat traction direction deviation, or the proximity effect of the subgrade; abnormal end-face gaps may be caused by attitude deviation, axis deviation, excessive pulling speed, or changes in the stress on the waterstop. Traditional threshold-based early warning systems struggle to distinguish the coupling relationships between these factors, making it difficult to directly translate warning results into construction adjustment measures. Fourth, existing digital twin platforms typically emphasize "data acquisition - model display - status early warning," but lack a calculable inverse mapping relationship between prediction results and construction control parameters. Even if the system can indicate a risk at a certain location, construction personnel still need to rely on experience to determine adjustment schemes for ballast water volume, ballast change rate, tugboat traction force, tugboat course, sinking speed, and measurement verification frequency, making it difficult to form a true closed-loop construction control.

[0004] Therefore, there is an urgent need to propose a new method for constructing digital scenarios in the construction process of immersed tunnels. This method should not only enable the fusion and digital representation of multi-source data, but also dynamically change the topological relationship of the scenario graph based on construction events. Furthermore, it should introduce mechanistic constraints such as pipe segment movement, ballast control, tugboat traction, and hydrodynamic effects into the machine learning prediction process, and further transform the prediction results into executable construction adjustment schemes. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for constructing a digital scene for immersed tunnel construction, which solves problems such as object stacking, insufficient constraints on construction events, lack of mechanistic constraints in prediction models, and difficulty in feeding prediction results back to construction control parameters in existing digital construction scenes for immersed tunnels. It is particularly suitable for the floating, sinking, and underwater docking construction processes of large-scale cross-sea and cross-river immersed tunnels in complex hydrological environments.

[0006] To achieve the above objectives, in a first aspect, the present invention provides a method for constructing a digital scene of immersed tunnel construction, the method comprising the following steps: establishing a set of construction events based on the immersed tunnel construction event chain, and configuring event status codes and event topology masks; acquiring construction status data, construction control instructions, and construction log information to identify the event status code of the current construction and determine the event topology mask; converting construction objects into nodes, object relationships into edges, and constructing an event-constrained dynamic scene graph by combining the event status code of the current construction and the event topology mask; establishing a six-degree-of-freedom mechanism baseline prediction model, and obtaining a mechanism baseline prediction value based on the six-degree-of-freedom mechanism baseline prediction model; obtaining a residual prediction value based on the event status code of the current construction, the event-constrained dynamic scene graph, the mechanism baseline prediction value, and the construction status history sequence to obtain a final prediction value; calculating a construction risk value based on the final prediction value, and when the construction risk value exceeds a set threshold, using construction control parameters as decision variables and solving for a construction adjustment scheme; and dynamically updating based on the final prediction value and the construction adjustment scheme to construct a digital construction scene of the immersed tunnel.

[0007] Optionally, the step of establishing a set of construction events based on the immersed tunnel construction event chain and configuring event status codes and event topology masks includes: the construction events in the set of construction events include segment launching, floating, arrival at the immersion area, start of immersion, attitude adjustment, near-bottom immersion, connection and docking, and retesting and locking; determining the event status code based on the event's implicit state, classification weight, and bias; and establishing an event topology mask library to determine the event topology mask for the construction event based on the event status code.

[0008] Optionally, the step of acquiring construction status data, construction control instructions, and construction log information to identify the event status code of the current construction and determine the event topology mask includes: collecting the construction status data of the current construction, including pipe section movement parameters and docking control parameters; acquiring the construction control instructions of the current construction, including ballast control instructions, tugboat traction instructions, sinking speed control instructions, and measurement verification instructions; obtaining the construction log features of the current construction through a log encoding function, and using the construction log features as the construction log information; establishing a construction event recognition machine learning model, constructing an event recognition input feature vector based on the construction status data, the construction control instructions, and the construction log information as input, outputting the event status code, and calling the corresponding event topology mask.

[0009] Optionally, the step of converting construction objects into nodes, converting object relationships into edges, and constructing an event-constrained dynamic scene graph by combining the event status code and the event topology mask of the current construction includes: the construction objects include pipe section control points, ballast tanks, tugboats, sinking equipment, measurement targets, docking faces, local areas of the foundation bed, and hydrodynamic influence areas to obtain the nodes; the object relationships include traction control relationships, ballast control relationships, hydrodynamic disturbance relationships, docking constraint relationships, measurement binding relationships, spatial proximity relationships, and risk transmission relationships to obtain the edges; the nodes and edges are effectively filtered through the event topology mask to determine the node set and edge set; the node attribute matrix of the node set is obtained, the edge attribute matrix and adjacency weight matrix of the edge set are obtained, and the event-constrained dynamic scene graph is constructed by combining the event status code.

[0010] Optionally, establishing the six-degree-of-freedom mechanism baseline prediction model includes: using the generalized configuration vector and the generalized velocity vector of the pipe section as state variables; based on the state variables, introducing the tugboat traction generalized force, ballast adjustment generalized force, hydrodynamic disturbance generalized force, near-bottom effect generalized force, and auxiliary control generalized force to establish a six-degree-of-freedom mechanism motion equation; and using the six-degree-of-freedom mechanism motion equation as the six-degree-of-freedom mechanism baseline prediction model.

[0011] Optionally, obtaining the mechanistic baseline prediction value based on the six-degree-of-freedom mechanistic baseline prediction model includes: using the six-degree-of-freedom mechanistic baseline prediction model to obtain the mechanistic baseline prediction value at each time in the future prediction time domain under a discrete prediction step size; the mechanistic baseline prediction value includes the pipe section configuration mechanistic baseline prediction value and the pipe section velocity mechanistic baseline prediction value.

[0012] Optionally, obtaining the residual prediction value based on the event status code of the current construction, the event constraint dynamic scene graph, the mechanism baseline prediction value, and the construction state history sequence to obtain the final prediction value includes: establishing an event-gated mechanism residual spatiotemporal graph network, including an event embedding layer, a multi-relationship graph attention layer, an event-gated time recursion layer, a mechanism residual gating layer, and an uncertainty output layer; inputting the event status code, the event constraint dynamic scene graph, the mechanism baseline prediction value, and the construction state history sequence into the event-gated mechanism residual spatiotemporal graph network to output the residual prediction value; and summing the residual prediction value and the mechanism baseline prediction value to obtain the final prediction value.

[0013] Optionally, the step of inputting the event status code, the event-constrained dynamic scene graph, the mechanism baseline prediction value, and the construction state history sequence into the event-gated mechanism residual spatiotemporal graph network to output the residual prediction value includes: obtaining a construction event embedding vector through the event embedding layer based on the event status code, and constructing initial latent features of the node by combining node features, spatial location encoding, the mechanism baseline prediction value, and the construction state history sequence; obtaining attention weights in the event-constrained dynamic scene graph by using the multi-relationship graph attention layer and combining the initial latent features of the node; obtaining the node time memory state by performing time recursion on the attention weights according to the event-gated time recursion layer; obtaining the residual gating coefficient of the mechanism residual gating layer based on the node time memory state; and determining the residual prediction value through the uncertainty output layer based on the residual gating coefficient.

[0014] Optionally, the step of calculating the construction risk value based on the final predicted value, and when the construction risk value exceeds a set threshold, using construction control parameters as decision variables and solving for the construction adjustment scheme, includes: calculating the construction risk value based on the final predicted value, using the risk weights, standardized excess quantities, and excess probabilities of the prediction indicators, and combining them with the current construction hazard coefficient; when the construction risk value exceeds the set threshold, solving for the decision variables through a construction parameter rolling optimization model to obtain the construction adjustment scheme.

[0015] Secondly, this invention provides a digital scene construction system for immersed tunnel construction processes. The system executes the digital scene construction method for immersed tunnel construction processes provided by this invention. The system includes input devices, output devices, a processor, and a memory, which are interconnected. The memory stores a computer program, which includes program instructions, and the processor is configured to call the program instructions. This invention, through high-performance hardware collaboration, ensures stable execution of the method and facilitates business deployment and expansion.

[0016] Compared with the prior art, the present invention has at least the following beneficial effects: First, this invention transforms the digital construction scenario from "full object display" to "event-driven control relationship expression," avoiding the need to merely stack BIM, GIS, point cloud, video, and sensor data to represent the technical solution.

[0017] Secondly, this invention uses construction event status codes and event topology masks to enable different node and edge relationships in the floating, sinking, near-bottom adjustment, docking and retesting locking stages, which can reflect the unique stage control logic in the construction of immersed tunnels.

[0018] Third, this invention combines six-degree-of-freedom mechanism baseline prediction with event-gated mechanism residual spatiotemporal graph network, enabling the prediction model to utilize the engineering laws of pipe section stress, ballast and hydrodynamic disturbance, and to learn on-site measured deviations, thereby improving the prediction stability under small sample or cross-operating condition conditions.

[0019] Fourth, this invention reduces the impact of latency, anomalies, missing data, and low-precision data on prediction results by fusing multi-source data credibility.

[0020] Fifth, this invention converts the prediction results into executable construction parameters such as ballast, traction, sinking speed and measurement verification frequency through rolling optimization of construction parameters, enabling digital construction scenarios to have prediction, correction and feedback capabilities. Attached Figure Description

[0021] Figure 1 This is a flowchart of a method for constructing a digital scene during the construction process of an immersed tunnel, according to an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the switching of the event topology mask during the construction phase in an embodiment of the present invention; Figure 3 This is a flowchart of the reliability fusion processing of construction status data according to an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the construction of an event-constrained dynamic scene graph according to an embodiment of the present invention; Figure 5 This is a flowchart illustrating the joint prediction process of the six-degree-of-freedom mechanism baseline prediction model and the event-gated mechanism residual spatiotemporal graph network in an embodiment of the present invention. Figure 6 This is a closed-loop feedback diagram illustrating the prediction results driving the rolling optimization of construction parameters in an embodiment of the present invention. Figure 7 This is a framework diagram of a digital scene construction system for immersed tunnel construction according to an embodiment of the present invention; Figure 8 This is a schematic diagram of a module of a digital scene construction system for immersed tunnel construction process according to an embodiment of the present invention; Figure labeling: 1. Construction event identification module; 2. Data credibility fusion module; 3. Event constraint dynamic scene graph construction module; 4. Six-degree-of-freedom mechanism baseline prediction module; 5. Event gating mechanism residual spatiotemporal graph network prediction module; 6. Risk calculation module; 7. Construction parameter rolling optimization module; 8. Digital scene update module. Detailed Implementation

[0022] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0023] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0024] Please see Figure 1 An embodiment of the present invention provides a method for constructing a digital scene during the construction process of an immersed tunnel, the method comprising the following steps: S1. Establish a set of construction events based on the immersed tunnel construction event chain, and configure event status codes and event topology masks.

[0025] In this embodiment, a set of construction events is established according to the immersed tunnel construction event chain, satisfying the following relationship:

[0026] in, A collection of construction events. For construction incidents, This serves as an index for construction events.

[0027] Specifically, construction events include at least the following: tunnel segment launch, floating, arrival at the immersion area, commencement of immersion, attitude adjustment, near-bottom immersion, docking and connection, and re-measurement and locking. By dividing the immersed tunnel construction process into construction events with clear engineering significance, the continuous construction process can be transformed into an event-driven state expression, giving clear construction stage constraints to the subsequent construction of dynamic scene diagrams and the invocation of prediction models.

[0028] In this embodiment, for each construction event ( Configure event status codes ( ) and event topology mask ( ).

[0029] Specifically, the event status code is determined based on the implicit state, classification weight, and bias of the construction event, satisfying the following relationship:

[0030] in, For event status codes, For construction incidents, For the first Classification weights for construction-related events. This is the implicit state of the event. For the first Bias towards construction-like events, The total number of construction event categories. An index for construction event categories.

[0031] It should be noted that the implicit state of the event is obtained through a construction event recognition machine learning model.

[0032] In this embodiment, an event topology mask library is established, and the event topology mask for construction events is determined based on the event status code.

[0033] Specifically, the event topology mask library satisfies the following relationship:

[0034] in, For event topology masking library, For the first Event topology template matrix for construction-type events. The spatial action threshold, As a time window, This is the event priority parameter. An index for construction event categories. This represents the total number of construction event categories.

[0035] Furthermore, the current construction event is determined based on the event status code, and an event topology mask corresponding to the current construction event is generated, satisfying the following relationship:

[0036] in, For event topology mask, Based on the candidate topology matrix, For event template fusion matrix, This is the construction time-space constraint matrix.

[0037] It should be noted that the event topology mask is used to constrain the information propagation relationship between nodes in the dynamic scene graph.

[0038] Please see Figure 2 The diagram illustrates the switching of event topology masks as construction stages occur. The event topology mask library, through switching operations, corresponds to three different floating construction stage topologies, from left to right representing the floating stages. Floating steps Floating phase In the diagram, solid lines represent active relationships between nodes, and dashed lines represent inactive relationships. All three types of stage topologies use the pipe section control point node as the top-level core node, branching downwards to connect different child nodes. Within the topology, the pipe section control point nodes are connected to the tugboat node and the hydrodynamic influence area node by solid lines, and the two are then connected to the measurement target node by solid lines. The measurement target node is connected to the ballast tank node and the submerged equipment node by dashed lines. The ballast tank node and the submerged equipment node are then connected to the end face node and the local area node of the foundation bed by dashed lines. Within the topology, the control point nodes of the pipe section are separated into ballast tank nodes and sinking equipment nodes by solid lines. At the same time, solid lines connect the nodes of the hydrodynamic influence area and the nodes of the local area of ​​the foundation bed. The nodes of the hydrodynamic influence area are connected to the tugboat nodes by solid lines. The nodes of the local area of ​​the foundation bed are connected to the measurement target nodes by dashed lines. The tugboat nodes and the measurement target nodes are then connected to the docking end face nodes by dashed lines. Within the topology, the control point nodes of the pipe section are connected by solid lines to the measurement target node and the docking end face node. The measurement target node is connected by dashed lines to the tugboat node. The docking end face node is connected by dashed lines to the ballast tank node. The tugboat node and the ballast tank node are respectively connected by dashed lines to the sinking equipment node and the hydrodynamic influence area node. The sinking equipment node and the hydrodynamic influence area node finally converge together by dashed lines to the local area node of the foundation bed.

[0039] S2. Obtain construction status data, construction control instructions, and construction log information to identify the event status code of the current construction and determine the event topology mask.

[0040] In this embodiment, construction status data of the current construction is collected, including pipe segment motion parameters and docking control parameters. The pipe segment motion parameters include the three-dimensional position of the pipe segment, heel angle, pitch angle, bow and stern heading angle, velocity and angular velocity. The docking control parameters include ballast tank water volume, ballast change rate, tugboat traction force, tugboat traction direction, water level, flow velocity, flow direction, wind speed, GNSS measurement values ​​of the Global Navigation Satellite System (GNSS), IMU attitude measurement values ​​of the Inertial Measurement Unit (IMU), docking end face gap, waterstop pressure, local elevation of the foundation bed, and measurement verification data.

[0041] In this embodiment, the construction control commands include ballast control commands, tugboat traction commands, sinking speed control commands, and measurement verification commands.

[0042] In this embodiment, the log embedding vector of the current construction is obtained through the log encoding function as the construction log feature to obtain construction log information.

[0043] In this embodiment, the process entities, equipment entities, spatial location entities, environmental disturbance entities, monitoring status entities, and abnormal status entities in the construction log are encoded into construction events using a log encoding function. The encoding results are then used for event status code generation, event topology mask selection, and information propagation constraints of the mechanism residual spatiotemporal graph network.

[0044] Specifically, when the construction log is unstructured text, it is represented as a sequence of terms that satisfies the following relationship:

[0045] in, For construction logs, The first one in the construction log Each word element, This represents the number of lexical units.

[0046] For each lexical, construct a lexical input vector that satisfies the following relationship:

[0047] in, For word input vectors, For word embedding vectors, For position embedding vectors, Embed vectors for construction entity types.

[0048] It should be noted that the construction entity type includes at least one of the following: process type, equipment type, monitoring type, environment type, spatial location type, and anomaly type; therefore, the log encoding function not only preserves the textual semantic information of the construction log, but also explicitly introduces the entity type information related to the construction event.

[0049] The sequence of word input vectors is fed into the log encoding network to obtain the hidden state sequence, which satisfies the following relationship:

[0050]

[0051] in, Given a sequence of word input vectors, For word input vectors, For the hidden state sequence, For log encoding networks, This is the hidden vector.

[0052] It should be noted that the log coding network can be implemented using existing Transformer encoders, bidirectional recurrent neural network encoders, convolutional text encoders, or pre-trained language models as the basic coding structure.

[0053] Furthermore, to highlight the contributions of construction event trigger words, equipment status words, environmental disturbance words, and abnormal risk words to the text embedding vector, construction event attention weights are introduced, satisfying the following relationship:

[0054] in, For the first The attention weight of construction events for each word element. For trainable parameters, For the first The hidden vector corresponding to each word element The total number of lexical units, For word index, For the first The hidden vectors corresponding to each word.

[0055] Furthermore, the text embedding vector is represented to satisfy the following relationship:

[0056] in, For text embedding vectors, For log encoding functions, For construction logs, The total number of lexical units, For word index, For the first The hidden vector corresponding to each word element For the first The attention weight of each word in the construction event.

[0057] When the construction log is a structured process record, the structured fields are represented to satisfy the following relationship:

[0058] in, For structured fields, This is a field for the construction phase. For device status fields, For spatial location fields, For environmental disturbance fields, This is an abnormal status field.

[0059] Furthermore, after performing field embedding on the structured fields, we obtain field embedding vectors that satisfy the following relationship:

[0060] in, Embed vectors for fields. For log encoding functions, For structured fields.

[0061] When both unstructured construction logs and structured process records exist simultaneously, the text embedding vector and the field embedding vector are fused to obtain the log embedding vector, satisfying the following relationship:

[0062] in, For log embedding vectors, For text embedding vectors, Embed vectors for fields. This represents vector concatenation. These are the parameters for the fusion layer.

[0063] In this embodiment, a construction event recognition machine learning model is established. Based on construction status data, construction control instructions and construction log information, an event recognition input feature vector is constructed as input to output an event status code. The corresponding event topology mask is called according to the current construction event status code.

[0064] In this embodiment, the construction event recognition machine learning model includes an input feature construction layer, a feature fusion layer, a temporal coding layer, an event probability output layer, and an event status code generation layer; it is used to identify the current construction event type based on the current construction status, construction control instructions, and construction log information, and output the corresponding event activation variables for subsequent event topology mask selection.

[0065] Specifically, numerical construction status features, construction control command features, and construction log features are jointly modeled, and the event activation variables output by the model are further converted into event status codes, thereby driving the invocation of the event topology mask library and the information propagation constraints of the spatiotemporal graph network.

[0066] In the At each construction moment, numerical construction status characteristics, construction control command characteristics, and construction log characteristics are constructed. The numerical construction status characteristics include at least one of the following: pipe section position, attitude angle, velocity, angular velocity, ballast state, water level, water flow velocity, equipment operating status, and environmental disturbance parameters. The construction control command characteristics include at least one of the following: traction command, anchor cable adjustment command, ballast adjustment command, attitude correction command, sinking control command, and docking control command. The construction log characteristics are obtained from the construction log text or structured process records through a log encoding function.

[0067] Specifically, the inputs to the construction event recognition machine learning model include numerical construction status features, construction control instruction features, and construction log features. For each construction moment, the event recognition input feature vector satisfies the following relationship:

[0068] in, Input feature vectors for event recognition. This is a numerical construction state feature vector. This is the feature vector of construction control instructions. This is the log embedding vector.

[0069] To eliminate the dimensional differences between different types of features, the numerical construction state feature vector is normalized to satisfy the following relationship:

[0070] in, This is a normalized numerical feature vector of construction state. This is a numerical construction state feature vector. This represents the mean of the numerical construction state feature vector. is the standard deviation of the numerical construction state feature vector.

[0071] The feature vector of construction control instructions is encoded to obtain the control instruction embedding vector, which satisfies the following relationship:

[0072] in, To embed control instructions into vectors, For control instruction encoding functions, This is the feature vector of construction control instructions.

[0073] It should be noted that when the construction control command is a discrete command, the control command encoding function adopts the command category embedding method; when the construction control command contains continuous control quantities, the control command encoding function adopts the linear mapping or multilayer perceptron mapping method.

[0074] Furthermore, the processed numerical construction status feature vector, control command embedding vector, and log embedding vector are fused to obtain a fused event feature vector, satisfying the following relationship:

[0075] in, To fuse event feature vectors, It is a non-linear activation function. For trainable parameters, This is a normalized numerical feature vector of construction state. To embed control instructions into vectors, This is the log embedding vector.

[0076] In this embodiment, considering the temporal continuity and phase switching characteristics of immersed tunnel construction events, the construction length is... The time window feature sequences satisfy the following relationship:

[0077] in, For time window feature sequences, This represents the fused event feature vector for each time window.

[0078] Furthermore, the time window feature sequence is input into the temporal coding layer to obtain the hidden state vector of the construction event at the current moment, which satisfies the following relationship:

[0079] in, Construction event hidden state vector, It is a time-series coding network. For network parameters, This is a time window feature sequence.

[0080] It should be noted that the temporal coding network can be implemented using any of the existing temporal modeling structures, such as gated recurrent unit networks, long short-term memory networks, temporal convolutional networks, or Transformer encoders.

[0081] The event probability output layer outputs the activation probabilities of various construction events based on the hidden state vector of construction events, satisfying the following relationship:

[0082] in, For activation probability, Indicates the first Construction-related events at any time activation probability, For trainable parameters, Construction event hidden state vector.

[0083] Based on the activation probability and preset threshold of various construction events, the event activation variables are determined, satisfying the following relationship:

[0084] in, Activate the variable for the event. For indicator functions, Indicates the first Construction-related events at any time activation probability, For the first The activation threshold corresponding to construction events.

[0085] It should be noted that when hour, , indicating the first Construction-type events are activated; otherwise, .

[0086] Therefore, the construction event recognition machine learning model converts multi-source construction status information into event activation variables to determine the event status code, so that the construction event recognition results can be directly used for calling the event topology mask library and information propagation constraints of the spatiotemporal graph network.

[0087] During the model training phase, a training sample set is constructed based on historical construction data, satisfying the following relationship:

[0088] in, For the training sample set, Input feature vectors for event recognition. For a moment Construction event tags, This represents the number of training samples.

[0089] A construction event recognition machine learning model is trained using a weighted binary cross-entropy loss function, satisfying the following relationship:

[0090] in, This is the total loss value. For the first Category weights for different construction events (used to reduce the impact of imbalanced sample sizes for different construction events on model training). For real labels, Indicates the first Construction-related events at any time The activation probability.

[0091] Furthermore, the event latent states of the construction event recognition machine learning model satisfy the following relationship:

[0092] in, This is the implicit state of the event. It is a non-linear activation function. The weight matrix to be trained. Input feature vectors for event recognition. This is the construction control command vector. For log embedding vectors, This refers to the implicit state of the event at the previous construction moment. This is a bias term.

[0093] At the current construction moment, the system first identifies the event status code of the current construction through the construction event identification module 1, and then calls the event topology mask corresponding to the current construction event from the event topology mask library.

[0094] Specifically, during the floating phase, the event topology mask primarily retains the traction control edges between the tugboat node and the tunnel section control point node, the hydrodynamic disturbance edges between the hydrodynamic influence area node and the tunnel section control point node, the spatial constraint edges between the channel boundary and the tunnel section control point node, and the measurement binding edges between the GNSS measuring points and the tunnel section control point node. At this stage, end face gaps, waterstop pressure, and near-bottom effects of the subgrade are not the primary control targets, and the relevant edge relationships are suppressed by the topology mask. During the sinking phase, the event topology mask primarily retains the ballast control edges between the ballast tank node and the tunnel section control point node, the sinking control edges between the sinking equipment node and the tunnel section control point node, the hydrodynamic disturbance edges between the hydrodynamic influence area node and the tunnel section control point node, the near-bottom risk edges between the subgrade local area node and the tunnel section control point node, and the attitude measurement binding edges between the IMU measuring points and the tunnel section control point node. At this stage, the model focuses on the relationship between ballast changes, sinking speed, hydrodynamic disturbances, and attitude deviations. During the connection and docking phase, the event topology mask mainly preserves the docking constraint edges between the pipe section to be installed and the installed pipe section, the measurement binding edges between the docking end face node and the measurement target node, the risk transmission edges between the waterstop pressure measurement point and the docking end face node, and the geometric constraint edges between the pipe section control point node and the docking end face node. At this time, the model focuses on the relationship between the end face normal clearance, misalignment, waterstop pressure, and attitude fine-tuning.

[0095] S3. Convert the construction object into a node, convert the object relationship into an edge, and construct an event-constrained dynamic scene graph by combining the event status code and the event topology mask of the current construction.

[0096] In this embodiment, the construction objects include pipe section control points, ballast tanks, tugboats, sinking equipment, measurement targets, docking faces, local areas of the foundation bed, and hydrodynamic influence areas, which are converted into nodes; the object relationships include traction control relationships, ballast control relationships, hydrodynamic disturbance relationships, docking constraint relationships, measurement binding relationships, spatial proximity relationships, and risk transmission relationships, which are converted into edges.

[0097] In this embodiment, the node set and edge set are determined by effectively filtering nodes and edges using an event topology mask.

[0098] Specifically, the event topology mask is used to control whether nodes and edges in the event-constrained dynamic scene graph under the current construction event participate in the calculation.

[0099] In the At each construction moment, the event constraint dynamic scene graph corresponding to the current event status code is represented, satisfying the following relationship:

[0100] in, This is a dynamic scene diagram of the event constraints corresponding to the current event status code. This refers to the set of nodes participating in the calculation under the current construction event. Let this be the set of edges that participate in the calculation under the current construction event. For event-constrained adjacency matrix, This is the node feature matrix after event constraints.

[0101] For the There are nodes, and the nodes participating in the mask are defined to satisfy the following relationship:

[0102] in, For nodes to participate in masking, For the first The data validity flag of each node (when the sensor, equipment or construction object corresponding to the node is at time...) The value is 1 when it is observable and the data is not missing. For the first Data reliability of each node This represents the node reliability threshold. For the first The spatial distance between each node and the area affected by the current construction event. The threshold value for the spatial effect of the node corresponding to the current construction event. For the first The most recent valid observation time for each node The time window that applies to the node corresponding to the current construction event.

[0103] when At that time, the first Each node participates in the state calculation under the current construction event; when At that time, the first Each node does not participate in the state calculation under the current construction event.

[0104] The node control standard satisfies the following relationship:

[0105] in, This refers to the set of nodes participating in the calculation under the current construction event. For the nodes participating in the calculation under the current construction event, For nodes to participate in masking, For node indexing, This represents the total number of nodes.

[0106] Furthermore, construct a node mask matrix that satisfies the following relationship:

[0107] in, Let be the node mask matrix. The first node participates in the masking process. The second node participates in the masking process. For the first Each node participates in the masking process.

[0108] In this embodiment, the node attribute matrix is ​​gated using a node mask matrix to obtain an event-constrained node feature matrix, which satisfies the following relationship:

[0109] in, This is the node feature matrix after event constraints. Let be the node mask matrix. This is a matrix of node attributes.

[0110] Therefore, for nodes that do not meet the current construction event control standards, their node characteristics are set to zero or removed in subsequent spatiotemporal graph network calculations, and they do not participate in node state updates and risk propagation calculations.

[0111] In this embodiment, for the first The node and the first Candidate edges between nodes are defined to participate in the mask, satisfying the following relationship:

[0112] in, For edge participation masking, Basic adjacency marker, For event-edge type adaptation coefficients, For nodes The event involved masking, For nodes The event involved masking, For indicator functions, For spatial distance, The edge space action threshold corresponding to the current construction event. For nodes The observation time For nodes The observation time The edge time window corresponding to the current construction event. The strength of the relationship between the edges. This is the intensity threshold.

[0113] For example, during the floating phase, the event topology mask prioritizes preserving the edges between the tugboat and the tunnel section, the hydrodynamic area and the tunnel section, the channel boundary and the tunnel section, and the GNSS measuring point and the tunnel section; during the sinking phase, the event topology mask prioritizes preserving the edges between the ballast tank and the tunnel section, the tunnel section and the foundation bed, the tunnel section and the IMU measuring point, and the tunnel section and the hydrodynamic area; during the docking phase, the event topology mask prioritizes preserving the edges between the tunnel section to be installed and the installed tunnel section, the docking face and the measuring target, and the watertight area and the face gap. This design avoids indiscriminately inputting all object relationships into the prediction model, thereby improving the model's ability to identify key relationships at the current construction stage.

[0114] In this embodiment, the construction status data undergoes credibility fusion before entering the event-constrained dynamic scene graph; for the first... The construction object is from the first Raw features provided by class data source The credibility weights and the fused node features satisfy the following relationships:

[0115]

[0116] in, For the first Credibility weight of data sources This is the credibility attenuation coefficient. For the first Real-time measurement errors or verification deviations of similar data sources For sampling delay, As the reference time interval, Mark missing or abnormal data. For the degree of historical residual fluctuation, The total number of data sources, For the index of the data source, For node features, For data source mapping matrix, These are the original features.

[0117] It should be noted that for different data sources such as GNSS, IMU, total station, survey vessel, hydrological monitoring system, ballast control system and video recognition system, the confidence weight is calculated based on real-time measurement error, sampling delay, abnormal missing markers and historical residual fluctuation degree, and the multi-source data of the same construction object is fused, which can reduce the impact of abnormal data, delayed data and low-precision data on subsequent prediction results.

[0118] Please see Figure 3The diagram shows the flowchart of the credibility fusion processing of construction status data. Six types of multi-source construction monitoring data, including GNSS data, IMU attitude data, total station / survey target data, hydrological monitoring data, ballast control system data, and video recognition data, are simultaneously input into the data credibility fusion module 2. Three types of verification results, namely error assessment, time delay assessment, and anomaly / missing data assessment, are fed back to this module through dotted lines as the constraint basis for fusion calculation. After completing the comprehensive credibility calculation of multi-source data, the data credibility fusion module 2 outputs credibility weights. The credibility weights are then used to generate the final fused node features.

[0119] In this embodiment, the node attribute matrix of the node set, the edge attribute matrix and the adjacency weight matrix of the edge set are obtained, and an event-constrained dynamic scene graph is constructed by combining the event status codes, satisfying the following relationship:

[0120] in, For event-constrained dynamic scene graphs, For a set of nodes, Let be the set of edges. This is a matrix of node attributes. This is the edge attribute matrix. This is the adjacency weight matrix. This is the event status code.

[0121] Please see Figure 4 The diagram illustrates the construction of a dynamic scene diagram with event constraints. The left dashed box marks the event topology mask library, and the dashed arrows marked as event constraints point to the dynamic scene topology structure enclosed by the large dashed box on the right. This structure takes the pipe section control point node as the core node and is connected to seven functional nodes through various functional control edges: the pipe section control point node connects upward to the sinking equipment node via the sinking control edge, connects to the tugboat node to the upper left via the traction control edge, connects to the measurement target node to the upper right via the measurement binding edge, connects to the docking end face node to the right via the docking constraint edge, connects to the hydrodynamic influence area node to the lower right via the hydrodynamic disturbance edge, connects to the local area node of the foundation bed via the near-bottom risk edge, and connects to the ballast tank node to the lower left via the ballast control edge. Each node and the pipe section control point node form a complete immersed tunnel construction event constraint topology network through corresponding business logic edges.

[0122] It should be noted that the event-constrained dynamic scenario diagram is not a static display object, but a computable structure used to express the control relationships, disturbance relationships and risk transmission relationships under different construction events.

[0123] Specifically, the node attribute matrix is ​​determined by the type attributes, real-time monitoring status, control status, environmental status, and data validity of each construction node; assuming there are a total of [number missing] nodes in the current construction scenario diagram. If there are n nodes, then the node attribute matrix satisfies the following relationship:

[0124] in, This is a matrix of node attributes. For the first Each node at time... The node attribute vector, For node attribute dimensions.

[0125] No. The original attribute vectors of the nodes satisfy the following relationship:

[0126] in, For the first Each node at time... The original attribute vector, For the spatial location of the node, For node attitude, For node speed, For node monitoring data, For the construction control quantities related to this node, For environmental or disturbance characteristics, For nodes Data validity indicators.

[0127] It should be noted that when nodes have missing or abnormal data, they can be filled using the observation value from the previous valid time step, the mean, or preset invalid values. Indicate whether the node participates in subsequent calculations.

[0128] After standardizing the original attribute vectors of the nodes, we obtain continuous attribute vectors for the nodes. By introducing node type embedding, we obtain the final node attribute vectors, which satisfy the following relationship:

[0129] in, For the first Each node at time... The node attribute vector, For mapping parameters, This is the standardized continuous attribute vector of the nodes. Embed vectors for node types. Number the node type. This is a vector for indicating missing attributes.

[0130] It should be noted that the node type includes at least one of the following: pipe joint node, traction equipment node, anchor cable node, ballast system node, measurement node, environmental node, joint node, and subgrade node.

[0131] Specifically, the edge attribute matrix is ​​used to describe the relative spatial relationships, time synchronization relationships, process collaboration relationships, and event constraint relationships between nodes; the edge attribute matrix is ​​organized according to node pairs and satisfies the following relationship:

[0132] in, This is the edge attribute matrix. For the first The node and the first Edge attribute vector between nodes This represents the total number of nodes.

[0133] For each edge in the candidate edge set, the edge attribute vector satisfies the following relationship:

[0134] in, For the first The node and the first Edge attribute vector between nodes Relative position For relative attitude difference, For spatial distance, For the observation time difference, Embed vectors for edge types. Number the edge type. The construction constraint strength coefficient, The value is the corresponding edge value in the event topology mask.

[0135] The relative state quantities in the above equation satisfy the following relationship:

[0136]

[0137]

[0138]

[0139] in, Relative position For nodes Location, For nodes Location, For relative attitude difference, For nodes posture, For nodes posture, For spatial distance, For the observation time difference, For nodes The observation time For nodes The observation time.

[0140] Furthermore, whether an edge participates in the calculation is determined according to the following criteria: First, the edge belongs to the basic candidate edge set; second, the current event topological mask allows the edge to exist; third, the distance between nodes satisfies the spatial scope of the current event; fourth, the observation time difference between nodes satisfies the time window of the current event; fifth, both ends of the edge are in a valid observation state. The corresponding edge participation criteria satisfy the following relationship:

[0141] in, The edge is used as a flag for calculation. Based on candidate topological matrix elements, This represents the value of the corresponding edge in the event topology mask. For spatial distance, The spatial impact threshold corresponding to the current event. For the observation time difference, The time window corresponding to the current event. For nodes Data validity indicators For nodes Data validity indicators.

[0142] when When, this edge participates in subsequent graph calculations; when When this occurs, the edge is not included in subsequent graph calculations, and its edge attribute vector is set to zero; the corrected edge attribute vector satisfies the following relationship:

[0143] in, This is the corrected edge attribute vector. The edge is used as a flag for calculation. For the first The node and the first Edge attribute vectors between nodes.

[0144] Furthermore, based on the corrected edge attribute vector, the corrected edge attribute matrix is ​​obtained, satisfying the following relationship:

[0145] in, This is the corrected edge attribute matrix. This is the corrected edge attribute vector. This represents the total number of nodes.

[0146] Specifically, for nodes With nodes The adjacency weights of the two satisfy the following relationship:

[0147]

[0148] in, For nodes With nodes Adjacency weights between them This represents the value of the corresponding edge in the event topology mask. These are the weighting coefficients. For spatial distance, For distance attenuation scale parameters, For the strength of the tugboat traction control relationship, For the strength of the ballast control relationship, The intensity of the hydrodynamic disturbance relationship. To determine the strength of the connection constraint relationship, To measure the strength of the binding relationship, This represents the strength of the risk transmission relationship.

[0149] It should be noted that the adjacency weights mentioned above are not ordinary spatial distance weights, but rather a comprehensive relationship strength formed under the constraints of the event topology mask. Through the control of the event topology mask, invalid object relationships in the current construction event are not included in the subsequent prediction model calculations, thus avoiding the stacking of all object relationships in digital construction scenarios. If the event topology mask does not allow nodes in the current construction event... With nodes Establish a connection, then Corresponding adjacency weight If the edge is allowed to exist, the adjacency weight is determined by the spatial distance attenuation term and the strength of various construction relationships.

[0150] The above-mentioned tugboat traction control relationship strength is used to characterize the traction control influence of the tugboat node on the immersed tunnel segment node, and satisfies the following relationship:

[0151] in, For the strength of the tugboat traction control relationship, This is a marker indicating the tugboat's traction relationship. For nodes event status codes, For nodes event status codes, For normalized traction force, The traction direction consistency coefficient. For spatial distance, This is the parameter for the attenuation of traction distance.

[0152] The above-mentioned ballast control relationship strength is used to characterize the influence of ballast system nodes on the buoyancy adjustment, sinking depth, and attitude leveling of the immersed tunnel sections, and satisfies the following relationship:

[0153]

[0154] in, For the strength of the ballast control relationship, This serves as a marker for ballast control relationships. For nodes event status codes, For nodes event status codes, These are the weighting coefficients. To normalize the ballast water flow rate, To normalize the change in ballast water volume, This refers to normalized buoyancy.

[0155] The above hydrodynamic disturbance intensity satisfies the following relationship:

[0156] in, The intensity of the hydrodynamic disturbance relationship. For spatial distance, The distance attenuation parameter is affected by hydrodynamics. The angle of water flow direction. For nodes Pointing to node The direction angle, For water flow velocity, For the normalized parameter of water flow velocity, This is the amplification factor for water level changes. This is the current water level. This is the baseline water level.

[0157] The above-mentioned docking constraint strength is used to characterize the degree of geometric constraint between the pipe sections to be docked, the installed pipe sections, the joint nodes, or the docking measurement nodes, and satisfies the following relationship:

[0158] in, To determine the strength of the connection constraint relationship, To mark the connection constraint relationship, For nodes event status codes, For nodes event status codes, To account for the spatial deviation between the mating end faces, This is the spatial deviation scale parameter. For relative attitude difference, This is the attitude deviation scale parameter.

[0159] The aforementioned measurement binding strength is used to characterize the reliability of data binding between the measurement base station, sensor node, and the node of the construction object being measured, and satisfies the following relationship:

[0160] in, To measure the strength of the binding relationship, To measure the binding relationship flag, For nodes event status codes, For nodes event status codes, To measure the quality coefficient, For the observation time difference, This is the time synchronization decay parameter.

[0161] The measured quality coefficient satisfies the following relationship:

[0162] in, To measure the quality coefficient, As a preset reference error, This is the current measurement error. To prevent constants with a denominator of zero.

[0163] The strength of the risk transmission relationship described above is used to characterize the degree to which an abnormal or risky state of one construction node may be transmitted to another node, and satisfies the following relationship:

[0164] in, The strength of the risk transmission relationship, As a marker of risk transmission relationship, For nodes event status codes, For nodes event status codes, For nodes To node The risk transmission coefficient, For nodes The intensity of the risk state, For spatial distance, This is the parameter for risk propagation distance attenuation.

[0165] node The intensity of the risk status is determined by the degree to which monitoring indicators exceed limits, satisfying the following relationship:

[0166] in, For nodes The intensity of the risk state, For nodes Current monitoring indicators For safety reference value, This is the risk limit.

[0167] It should be noted that when the node With nodes When no corresponding construction relationship exists, the corresponding relationship flag is taken as... The strength of this relationship is When there is a corresponding construction relationship between the two, the strength of the relationship is determined based on the construction control quantity, spatial distance, constraint deviation, measurement quality, or risk status.

[0168] Through the above methods, event-driven dynamic mapping of the immersed tunnel construction process is realized, so that the digital construction scene is no longer a simple three-dimensional display model, but a computable graph structure that can express construction events, control relationships, disturbance relationships, measurement relationships and risk transmission relationships.

[0169] S4. Establish a six-degree-of-freedom mechanism baseline prediction model, and obtain the mechanism baseline prediction value based on the six-degree-of-freedom mechanism baseline prediction model.

[0170] In this embodiment, a six-degree-of-freedom mechanistic baseline prediction model is established based on the stress on the pipe section, ballast changes, tugboat traction, hydrodynamic disturbance, and near-bottom effect to obtain the mechanistic baseline prediction value of the pipe section construction state in the future prediction time domain.

[0171] Specifically, the generalized configuration vector and the generalized velocity vector of the pipe segment are used as state variables.

[0172] The generalized configuration vectors of the aforementioned pipe segments satisfy the following relationship:

[0173] in, For the first The generalized configuration vector of the tube segment at time t. Let the three-dimensional position vector of the centroid of the immersed tunnel segment be denoted as . Let be the attitude angle vector of the pipe segment. Position along the tunnel axis. This is the horizontal offset position. This is the vertical elevation. The tilt angle is... The pitch angle is the angle of inclination. This refers to the bow-stern heading angle or course angle.

[0174] The generalized velocity vector of the above pipe segment satisfies the following relationship:

[0175] in, For the first The generalized velocity vector of the tube segment at time t. Let be the translational velocity vector of the tube segment. The angular velocity vector of the tube section. The translational velocity is along the tunnel axis. The translational velocity is in the lateral offset direction. The vertical translational velocity, The rate of change of the pitch angle, The rate of change of the pitch angle. This represents the rate of change of the bow-stern heading angle.

[0176] Furthermore, based on state variables, a six-degree-of-freedom mechanism motion equation is established by introducing the tractor traction generalized force, ballast adjustment generalized force, hydrodynamic disturbance generalized force, near-bottom effect generalized force, and auxiliary control generalized force, serving as a six-degree-of-freedom mechanism baseline prediction model.

[0177] The above six-degree-of-freedom mechanism equations of motion satisfy the following relationship:

[0178] in, The generalized equivalent mass matrix is ​​related to construction events, pipe section configuration, and ballast state. The generalized acceleration vector of the pipe segment. For inertial coupling terms, For the first The generalized velocity vector of the tube segment at time t. For hydrodynamic damping and construction control damping, The term is the generalized restoring force formed by the combined forces of gravity, buoyancy, and the weight of ballast water. For the generalized traction force of tugboats, To adjust the generalized force for ballast, For generalized forces of hydrodynamic disturbance, For the proximity effect generalized force of the subgrade during the near-bottom settling stage, To assist in controlling generalized forces.

[0179] The generalized traction force of the tugboat described above satisfies the following relationship:

[0180]

[0181] in, For the generalized traction force of tugboats, The total number of tugboats, For tugboat index, For the first The vector of the traction force exerted by the tugboat on the tunnel section. Let be the relative position vector from the traction point to the centroid of the tube segment. For the first The traction of the tugboat For the first The traction direction angle of the tugboat.

[0182] The above-mentioned generalized ballast adjustment forces satisfy the following relationship:

[0183]

[0184] in, To adjust the generalized force for ballast, This represents the total number of ballast tanks. For ballast tank indexing, For the first The equivalent vertical force generated by the ballast changes in each ballast tank For the first The position vector of each ballast tank relative to the center of mass of the tube segment. This is the mapping matrix between the ballast change rate and the pipe section motion. This is the ballast change rate vector. For the density of water, It is the acceleration due to gravity. For the first Changes in water volume in each ballast tank It is a vertical unit vector.

[0185] The generalized forces of the above hydrodynamic disturbances satisfy the following relationship:

[0186]

[0187] in, For generalized forces of hydrodynamic disturbance, The resultant hydrodynamic force generated by the water flow on the pipe section, Let be the position vector of the equivalent hydrodynamic point of action relative to the centroid of the pipe section. For the density of water, The drag coefficient, The equivalent water-facing area of ​​the pipe section, Let be the relative velocity vector of the water flow with respect to the pipe section. For the additional quality coefficient, The equivalent drainage volume of the pipe section. This is the relative acceleration vector of the water flow with respect to the pipe section.

[0188] When the bottom of the pipe section is far from the foundation bed When the value is less than a set threshold, the generalized force of the proximity effect satisfies the following relationship:

[0189] in, For the proximity effect generalized force of the subgrade during the near-bottom settling stage, This is the near-bottom effect intensity coefficient. This refers to the distance from the bottom of the pipe section to the subgrade. The near-bottom effect attenuation scale, Near-bottom effect Influence coefficient on yaw angle The tilt angle is... This is the coefficient representing the influence of the near-bottom effect on the pitch angle. It is the pitch angle.

[0190] The aforementioned auxiliary control generalized force is generated by the lifting point, guiding device, or temporary auxiliary control, and is used to characterize the equivalent force and equivalent moment of temporary external control actions other than tugboat traction, ballast adjustment, hydrodynamic disturbance, and bed proximity effect in the six-degree-of-freedom motion equation of the pipe section; it is obtained by converting the force, position, and direction of action of each auxiliary control point.

[0191] Assume there are a total of [number] construction moments at the current time. The auxiliary control point, the first The auxiliary force generated by each auxiliary control point on the pipe section is The auxiliary torque is Its position vector relative to the pipe section reference point is: Then the auxiliary control generalized force satisfies the following relationship:

[0192] in, To assist in controlling generalized forces, This represents the total number of auxiliary control points. For the index of auxiliary control points, For the first The activation flag for each auxiliary control point The reliability coefficient for auxiliary control action. For the first The auxiliary force generated by each auxiliary control point on the pipe section The position vector of the point of application relative to the reference point of the pipe section. This is an auxiliary torque.

[0193] When the auxiliary control point is not activated, the control signal is invalid, or the corresponding device is not involved in the current construction event. This auxiliary control point does not participate in the generalized force calculation; when the auxiliary control point is enabled and the control signal is valid, .

[0194] The auxiliary force can be determined according to the control command, guide deviation, or lifting point correction, and satisfies the following relationship:

[0195] in, For the first The auxiliary force generated by each auxiliary control point on the pipe section To assist in controlling the intensity of the action, For the first The direction of action of each auxiliary control point is a unit vector.

[0196] The strength of the auxiliary control effect satisfies the following relationship:

[0197] in, To assist in controlling the intensity of the action, To assist in the control command quantity, To assist in controlling the stiffness coefficient, This refers to the deviation of the auxiliary control point relative to the target position, target attitude, or guidance reference. The damping coefficient is... The rate of change of deviation This is the maximum permissible intensity of action of the auxiliary control device.

[0198] Whether auxiliary control points participate in the calculation is determined by their activation status and data validity, satisfying the following relationship:

[0199] in, For the first The activation flag for each auxiliary control point For the first The activation status of each auxiliary control point This serves as a validity indicator for the corresponding control signal or measurement data. This is an indicator function.

[0200] Considering the time synchronization of the auxiliary control signals, the reliability coefficient satisfies the following relationship:

[0201] in, The reliability coefficient for auxiliary control action. To assist in controlling the time difference between the control signal and the current construction time, This is the time decay coefficient.

[0202] It should be noted that, through the above methods, the generalized force of auxiliary control is determined based on the activation status, direction of action, control command, deviation correction amount, and signal reliability of the auxiliary control device; if there are no lifting points, guiding devices, or temporary auxiliary control functions under the current construction event, then... or ,at this time .

[0203] In this embodiment, under a discrete prediction step size, the mechanistic baseline prediction values ​​at each time point in the future prediction time domain are obtained using a six-degree-of-freedom mechanistic baseline prediction model. These include the pipe section configuration mechanism baseline prediction values ​​and the pipe section velocity mechanism baseline prediction values, which respectively satisfy the following relationships:

[0204]

[0205] in, This is the baseline prediction value for the pipe segment configuration mechanism at the next moment. For discrete prediction step size, For the first The generalized configuration vector of the tube segment at time t. The kinematic transformation matrix is ​​determined by the attitude angles of the tube segments. For the first The generalized velocity vector of the tube segment at time t. The inverse of the generalized equivalent mass matrix. For the generalized traction force of tugboats, To adjust the generalized force for ballast, For generalized forces of hydrodynamic disturbance, For the proximity effect generalized force of the subgrade during the near-bottom settling stage, The generalized auxiliary control force generated by the lifting point, guiding device, or temporary auxiliary control. For inertial coupling terms, For hydrodynamic damping and construction control damping, The term is the generalized restoring force formed by the combined forces of gravity, buoyancy, and the weight of ballast water. This is the baseline prediction value for the pipe segment velocity mechanism at the next moment.

[0206] It should be noted that the six-degree-of-freedom mechanism baseline prediction model is not required to completely replace the fine hydrodynamic simulation. Instead, it provides engineering priors with correct orientation and clear physical meaning, avoiding the direct use of pure data-driven models to predict the pipe segment state. This ensures that the subsequent event-gated mechanism residual spatiotemporal graph network does not need to learn the pipe segment motion law from scratch.

[0207] S5. Based on the event status code of the current construction, the event constraint dynamic scene diagram, the mechanism baseline prediction value, and the construction status history sequence, obtain the residual prediction value to obtain the final prediction value.

[0208] In this embodiment, an event-gated mechanism residual spatiotemporal graph network is established, including an event embedding layer, a multi-relation graph attention layer, an event-gated time recursion layer, a mechanism residual gating layer, and an uncertainty output layer.

[0209] Furthermore, the event status code, event constraint dynamic scene graph, mechanism baseline prediction value, and construction status history sequence are input into the event-gated mechanism residual spatiotemporal graph network to output residual prediction values; specifically, the following steps are included: First, based on the event status code, the construction event embedding vector is obtained through the event embedding layer, and the initial implicit features of the node are constructed by combining node features, spatial location coding, mechanism baseline prediction value and construction status historical sequence.

[0210] The above construction event embedding vectors satisfy the following relationship:

[0211] in, Embed vectors for construction events. Embedded functions for events, This is the event status code.

[0212] The initial latent features of the above nodes satisfy the following relationship:

[0213]

[0214] in, For nodes The initial implicit features, This is the weight matrix for node features. For node features, The weight matrix for event embedding. Embed vectors for construction events. The weight matrix for position encoding. Encoding the spatial location of nodes, The weight matrix for the mechanism prediction values. For nodes The corresponding mechanistic baseline prediction value, The weight matrix is ​​the historical residual. The historical residuals between the current baseline prediction and the measured value of the mechanism. This is the measured value at the current moment. This is the baseline prediction value for the mechanism at the current moment.

[0215] Secondly, in the event-constrained dynamic scene graph, attention weights are obtained by utilizing the attention layer of the multi-relationship graph and combining the initial implicit features of the nodes.

[0216] Specifically, for edge relationship types ,node For nodes The attention weights satisfy the following relationship:

[0217] in, For nodes For nodes In the Attention weights on edge relationships, The weight matrix of the query vector. For the first Nodes in the attention layer of the layer graph The implicit features, The weight matrix is ​​the weight matrix of the key vectors. For the first Nodes in the attention layer of the layer graph The implicit features, The dimension scaling factor for the attention head. Let be the attention weight vector for the edge attributes. For the first The edge attribute vector of the class edge, The attention weight vector embedded in the event. Embed vectors for construction events. This is the attention weight vector for spatial location relationships. For nodes With nodes The relative position vector, This is a penalty term for the event topology mask.

[0218] The above event topology mask penalty term satisfies the following relationship:

[0219] in, As a penalty item, The penalty coefficient is... For the current construction event Whether the topological mask value is valid for edge class.

[0220] It should be noted that when an edge is ineffective in the current construction event, the penalty term increases, causing the attention of that edge to approach zero. Therefore, the model can automatically select effective control relationships, disturbance relationships, measurement relationships, and risk transmission relationships based on the construction event.

[0221] Next, the attention weights are recursively calculated using the event-gated time recursion layer to obtain the node time memory state.

[0222] The aforementioned node temporal memory state is used to characterize the historical state retention results and current event-driven update results of the construction node during continuous construction. It is obtained recursively by the event-gated temporal recursive layer based on the current node characteristics, neighborhood attention aggregation characteristics, the memory state of the previous time step, and the current construction event state.

[0223] First, regarding the first For each node, the features of its effective neighboring nodes are aggregated based on the event-constrained attention weights to obtain the neighborhood aggregated features, which satisfy the following relationship:

[0224] in, This is a neighborhood aggregation feature. For the current construction event The set of valid neighboring nodes of a node For nodes With nodes Event-constrained attention weights For nodes At any moment The node feature vectors.

[0225] Second, the current node features, neighborhood aggregation features, the node's temporal memory state from the previous time step, and the current event embedding vector are all input into the event gating function to obtain the memory update gate, which satisfies the following relationship:

[0226] in, For the first Each node at time... The memory update gate It is the Sigmoid activation function. For trainable parameters, For nodes At any moment The node feature vectors, This is a neighborhood aggregation feature. This is the node time memory state of the previous moment. This is the event embedding vector corresponding to the current event status code.

[0227] Furthermore, the temporal memory states of candidate nodes are generated, satisfying the following relationship:

[0228] in, The time-memory state of candidate nodes. For trainable parameters, For nodes At any moment The node feature vectors, This is a neighborhood aggregation feature. This is the node time memory state of the previous moment. This is the event embedding vector corresponding to the current event status code.

[0229] Third, by fusing the memory state of the previous time step and the candidate memory state through a gating recursive relation, the node time memory state of the current time step is obtained, satisfying the following relationship:

[0230] in, The node is used to remember the state at a specific time. For the first Each node at time... The memory update gate The time-memory state of candidate nodes. This is the node time memory state of the previous moment. This represents element-wise multiplication of vectors.

[0231] In an optional embodiment, when a node is not activated by the event topology mask under the current construction event, or when the node observation data is invalid, a conservative update method is adopted, satisfying the following relationship:

[0232] in, The node is used to remember the state at a specific time. This is the node time memory state of the previous moment.

[0233] Through the above steps, the event-gated time recursion layer can make the node time memory state simultaneously reflect the node's historical evolution, the influence of neighborhood associations, and event-driven changes, based on the degree of retention of historical information of the current construction event control node and the degree of update of the current state.

[0234] Subsequently, based on the node time memory state, the residual gating coefficients of the mechanism residual gating layer are obtained, satisfying the following relationship:

[0235] in, For residual gating coefficients, It is the Sigmoid activation function. This is the weight matrix for the gating coefficients. The node is used to remember the state at a specific time. For nodes The mechanistic baseline prediction value at the next time step. This is the measured value at the current moment. This is the baseline prediction value for the mechanism at the current moment. Embed vectors for construction events. This is the construction control command vector. This is the bias term for the gating coefficient.

[0236] Finally, based on the residual gating coefficients, the predicted residual values ​​are determined through the uncertainty output layer, satisfying the following relationship:

[0237] in, For nodes The predicted residual value at the next time step. For residual gating coefficients, This is the weight matrix for residual prediction. The node is used to remember the state at a specific time. This is the measured value at the current moment. This is the baseline prediction value for the mechanism at the current moment.

[0238] In this embodiment, the final predicted value is obtained by summing the residual predicted value and the mechanistic baseline predicted value, which satisfies the following relationship:

[0239] in, For nodes The final predicted value at the next moment, For nodes The mechanistic baseline prediction value at the next time step. For nodes The predicted residual value at the next time step.

[0240] It should be noted that the event-gated mechanism residual spatiotemporal graph network does not directly predict the final construction state, but learns the residual between the mechanism baseline prediction value and the measured construction state, and then superimposes the mechanism baseline prediction value and the residual prediction value to obtain the final prediction value. This design realizes the joint prediction mechanism of "mechanism baseline prediction + residual spatiotemporal graph correction", so that the model has both engineering mechanism constraints and can learn the deviation of on-site measurement.

[0241] Please see Figure 5The diagram illustrates the joint prediction process of the six-DOF mechanistic baseline prediction model and the event-gated mechanistic residual spatiotemporal graph network. The process consists of two parallel input branches. The first is the mechanistic baseline prediction branch: pipe section configuration, ballast, tugboat traction, and hydrodynamic parameters are input into the six-DOF mechanistic baseline prediction model, which outputs the mechanistic baseline prediction value. This value is fed down into the final prediction value calculation module and also fed up into the event-gated mechanistic residual spatiotemporal graph network. The second is the spatiotemporal graph network input branch: event constraint dynamic scene graph, event status code, and construction status history sequence are simultaneously input into the event-gated mechanistic residual spatiotemporal graph network. After network computation, the residual prediction value is output. The residual prediction value is fused with the aforementioned mechanistic baseline prediction value to obtain the final prediction value, which includes four types of construction prediction results: attitude prediction, trajectory prediction, end face gap prediction, and construction risk prediction.

[0242] S6. Calculate the construction risk value based on the final predicted value. When the construction risk value exceeds the set threshold, use the construction control parameters as decision variables and solve for the construction adjustment scheme.

[0243] In this embodiment, the pipe section position deviation, attitude deviation, sinking trajectory deviation, end face gap risk, and abnormal pressure risk of the waterstop are calculated based on the final predicted values. The construction risk value is calculated by using the risk weights, standardized exceedance quantities, and exceedance probabilities of the predicted indicators, combined with the current construction hazard coefficient, and satisfies the following relationship:

[0244] in, For the future Construction risk value at each time step. The current construction risk level is [not specified]. For the first Risk weights of each predictive indicator For the first The standardization of each indicator exceeds the limit. This is the penalty coefficient for the probability of exceeding the limit. For the first The probability of exceeding the limit for each indicator.

[0245] It should be noted that the time interval between any two time steps is .

[0246] The aforementioned current construction hazard coefficient is determined based on the construction event status, stage risk level, and intensity of abnormal disturbance, and satisfies the following relationship:

[0247] in, The current construction hazard factor, For the amplitude limiting function, The foundation risk factor corresponding to the current construction event status code. This is the amplification factor for abnormal disturbances. The intensity of the abnormal disturbance. This is the lower limit of the risk factor. This represents the upper limit of the risk factor.

[0248] The basic hazard factor is determined according to the risk level of the construction stage, as shown in Table 1: Table 1

[0249] For example, during the near-bottom sinking stage, Higher values ​​should be selected, and the risk weights for indicators such as lateral tilt deviation, longitudinal tilt deviation, and vertical sinking speed should be increased; during the docking and connection phase, Similarly, higher values ​​are used, and the risk weights for indicators such as end face normal clearance, misalignment, and waterstop pressure are increased; during towing, routine positioning, or low-risk monitoring phases, Choose a lower value to avoid the system overreacting to normal fluctuations.

[0250] It should be noted that the construction risk value comprehensively considers the predicted indicators exceeding the limit, the probability of exceeding the limit, and the current construction event hazard coefficient; the predicted indicators include the deviation of the pipe section plane position, vertical elevation deviation, transverse tilt angle deviation, longitudinal tilt angle deviation, bow and stern direction angle deviation, sinking trajectory deviation, normal gap of the docking end face, and abnormal pressure risk of the waterstop; this risk calculation method can distinguish between general deviation, trend risk, and controllable risk.

[0251] In this embodiment, when the construction risk value exceeds the set threshold, the construction parameter rolling optimization model is used, and the decision variables are solved by combining the rolling optimization objective function to obtain the construction adjustment plan.

[0252] It should be noted that the decision variables are construction control parameters, including ballast adjustment amount, ballast change rate, tugboat traction correction amount, tugboat direction angle correction amount, sinking speed correction amount, and measurement verification frequency.

[0253] In this embodiment, a rolling optimization model for construction parameters is initiated, including optimization variables and a rolling optimization objective function.

[0254] The above optimization variables satisfy the following relationship:

[0255] in, To optimize variables, Adjust the water volume for each ballast tank. Adjustments for the ballast change rate of each ballast tank. Adjustments to the traction force of each tugboat. For the azimuth adjustment amount of each tugboat, Adjustment amount for sinking speed, To measure and verify the frequency adjustment amount.

[0256] Furthermore, the rolling optimization objective function takes the target position, target attitude, end face gap limit, and comprehensive construction risk value of the pipe section as the objective, construction control parameters as decision variables, and upper and lower limits of ballast water volume, upper limit of ballast change rate, upper and lower limits of tugboat traction force, upper and lower limits of single adjustment of tugboat direction angle, upper limit of sinking speed, upper and lower limits of measurement and verification frequency, and construction window period constraints as constraints to solve the construction adjustment scheme; it can transform the prediction results into executable construction parameters, forming a closed-loop feedback from digital scenario construction, state prediction, risk assessment to construction adjustment.

[0257] The above rolling optimization objective function satisfies the following relationship:

[0258] in, For the variable to be solved, To predict the length of the time domain, For time step index, This is the weighting coefficient for the positional deviation. For the target location, Let the target position vector of the pipe segment be _____. This is the weighting coefficient for attitude deviation. To predict attitude, For the target posture, This is the weighting coefficient for gap risk. To predict the gap between the mating ends, The end face clearance control limit, This is a weighting coefficient for overall risk. For the future Construction risk value at each time step. The weighting coefficients are used to control the amplitude. For the future Control increment vector for each time step Weighting coefficients to control smoothness, For the future The control increment vector for each time step.

[0259] Specifically, when the system predicts that the lateral tilt angle will exceed the control limit within a certain number of time steps during the near-bottom sinking stage, the construction parameter rolling optimization module 7 can output the following construction adjustment schemes: reduce the sinking speed, adjust the water volume in the left or right ballast tanks, correct the traction direction of the tugboat, and increase the measurement verification frequency. When the system predicts that the end face gap is less than the safety limit or the waterstop pressure rises abnormally during the docking stage, the system can output suggestions to pause docking, reduce the approach speed, perform attitude fine-tuning, and increase the retest frequency.

[0260] S7. Based on the final predicted value and the construction adjustment plan, dynamic updates are implemented to construct a digital construction scenario for the immersed tunnel.

[0261] In this embodiment, the final predicted value and construction adjustment plan are written into the digital construction scenario, and the corresponding nodes, edges, risk areas and control instructions are dynamically updated.

[0262] Specifically, the optimization results are written into the digital construction scenario through the digital scenario update module 8, enabling construction managers to view the closed-loop chain of "prediction deviation - risk source - recommended adjustment amount - expected adjustment effect". In this way, closed-loop control is realized from construction event identification, dynamic scenario diagram construction, mechanism baseline prediction, residual correction, risk judgment to construction parameter reverse calculation.

[0263] Please see Figure 6 The diagram illustrates a closed-loop feedback mechanism where prediction results drive rolling optimization of construction parameters. The final predicted value is obtained based on the event-gated mechanism residual spatiotemporal graph network. After calculating the construction risk value, it is input into the rolling optimization model of construction parameters. The model output includes adjustments to construction parameters such as ballast adjustment, tugboat traction correction, sinking speed correction, and measurement verification frequency. These adjustments are then distributed to the construction control execution stage for implementation. After execution, real-time construction status data is generated to update the digital scene. The updated digital scene information is then fed back to the event-gated mechanism residual spatiotemporal graph network, forming a continuously iterative closed-loop feedback chain of construction prediction, optimization, execution, and update.

[0264] In an optional embodiment, when the error between the measured construction state and the predicted construction state exceeds the adaptive threshold corresponding to the current construction event, the system marks the sample as a difficult case and uses it for incremental updates of the event gating mechanism residual spatiotemporal graph network. For high-risk events such as docking and near-bottom immersion, a smaller adaptive threshold is used; for general floating events, a larger adaptive threshold is used. In this way, the model can gradually adapt to the specific hydrological environment, equipment coordination methods, and measurement error characteristics on site, improving the adaptability of subsequent pipe section construction predictions.

[0265] It should be noted that this invention addresses the construction processes of immersed tunnel segments, including undocking, floating, positioning, sinking, near-bottom attitude adjustment, docking, and re-measurement control. By constructing construction event status codes, event topology masks, event-constrained dynamic scene diagrams, six-degree-of-freedom mechanism baseline prediction models, event-gated mechanism residual spatiotemporal graph networks, and construction parameter rolling optimization models, it achieves state representation, risk prediction, and feedback of construction control parameters during the immersed tunnel construction process. It goes beyond simply providing 3D visualization or monitoring data management of the immersed tunnel construction process; it further integrates construction events, segment attitude, hydrodynamic environment, ballast control, tugboat traction, docking gap, waterstop stress, and near-bottom effect of the foundation bed into a unified, computable construction scenario. Through event-constrained machine learning prediction methods, it predicts key state parameters in advance during the immersed tunnel construction process and transforms the prediction results into executable construction adjustment suggestions such as ballast water volume, tugboat traction force, sinking speed, and measurement verification frequency. This improves the reliability of state prediction, the ability to identify risks in advance, and the construction feedback capability during the floating, sinking, and docking processes of the immersed tunnel.

[0266] Please see Figure 7 In one optional embodiment, the present invention provides a digital scene construction system for immersed tunnel construction process. The system includes an input device, an output device, a processor, and a memory, all interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to invoke the program instructions to execute specific steps as described in the relevant embodiments of the digital scene construction method for immersed tunnel construction process provided by the present invention. The digital scene construction system for immersed tunnel construction process provided by the present invention has a complete structure, is objective and stable, and improves the overall applicability and practical application capability of the present invention.

[0267] Please see Figure 8The diagram illustrates a modular system for constructing a digital scene during immersed tunnel construction. It includes: a construction event identification module 1, used to identify construction event status codes based on construction status data, construction control instructions, and construction log information; a data credibility fusion module 2, used to perform credibility-weighted fusion of multi-source construction status data; an event-constrained dynamic scene graph construction module 3, used to construct a dynamic scene graph based on construction event status codes and event topology masks; a six-degree-of-freedom mechanism baseline prediction module 4, used to calculate the predicted mechanism baseline value based on tunnel segment movement, ballast adjustment, tugboat traction, hydrodynamic disturbance, and near-bottom effect; and an event-gated mechanism residual spatiotemporal graph network. The network prediction module 5 is used to predict the residual between the baseline prediction value and the actual construction state and obtain the final prediction value; the risk calculation module 6 is used to calculate the attitude deviation, trajectory deviation, docking gap risk and comprehensive construction risk value; the construction parameter rolling optimization module 7 is used to back-calculate the ballast adjustment amount, tugboat traction correction amount, sinking speed correction amount and measurement verification frequency based on the prediction results; the digital scene update module 8 is used to write the prediction results, risk results and construction adjustment plan into the digital construction scene; through the collaboration of the above modules, the prediction of construction status, risk identification and construction parameter feedback during the floating, sinking and docking process of the immersed tunnel are realized.

[0268] In summary, the present invention provides a method and system for constructing a digital scene of immersed tunnel construction process; it establishes a set of construction events according to the construction event chain of immersed tunnel, and configures an event status code and an event topology mask for each construction event; it collects multi-source construction status data related to the floating, sinking, attitude adjustment and docking processes of immersed tunnel segments; it constructs an event-constrained dynamic scene diagram based on the construction event status codes; it establishes a mechanism baseline prediction model based on the six-degree-of-freedom motion of the tunnel segments, ballast adjustment, tugboat traction, hydrodynamic disturbance and near-bottom effect; it inputs the event-constrained dynamic scene diagram, construction event status codes, construction status history sequence and mechanism baseline prediction values ​​into an event-gated mechanism residual spatiotemporal graph network, predicts the residual between the mechanism baseline prediction value and the measured construction status, and obtains the tunnel segment attitude, sinking trajectory, docking end face gap and construction risk prediction results in the future prediction time domain; when the risk value exceeds a set threshold, it uses a construction parameter rolling optimization model to back-calculate the ballast adjustment amount, tugboat traction correction amount, sinking speed correction amount and measurement verification frequency, and writes the prediction results and construction adjustment plan into the digital construction scene. The method of this invention is easy to understand, simple to calculate, requires less work, and is convenient for engineering applications, providing a theoretical basis and technical support for the further development of the field of intelligent geotechnical engineering.

[0269] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for constructing a digital scene of immersed tunnel construction process, characterized in that, Includes the following steps: Establish a set of construction events based on the event chain of immersed tunnel construction, and configure event status codes and event topology masks; Acquire construction status data, construction control instructions, and construction log information to identify the event status code of the current construction and determine the event topology mask; The construction object is converted into a node, the object relationship is converted into an edge, and the event constraint dynamic scene graph is constructed by combining the event status code of the current construction and the event topology mask. A six-degree-of-freedom mechanism baseline prediction model is established, and the mechanism baseline prediction value is obtained based on the six-degree-of-freedom mechanism baseline prediction model; Based on the event status code of the current construction, the event constraint dynamic scene diagram, the mechanism baseline prediction value, and the construction status history sequence, the residual prediction value is obtained to obtain the final prediction value; The construction risk value is calculated based on the final predicted value. When the construction risk value exceeds the set threshold, the construction control parameters are used as decision variables, and the construction adjustment plan is obtained by solving the problem. The final predicted value and the construction adjustment plan are dynamically updated to construct a digital construction scenario for immersed tunnels.

2. The method for constructing a digital scene of immersed tunnel construction process according to claim 1, characterized in that, The process of establishing a set of construction events based on the immersed tunnel construction event chain, and configuring event status codes and event topology masks, includes: The construction events in the set of construction events include: pipe section launch, floating, arrival at the sinking area, commencement of sinking, attitude adjustment, near-bottom sinking, connection and docking, and retesting and locking. The event status code is determined based on the event's implicit state, classification weight, and bias. Establish an event topology mask library, and determine the event topology mask for the construction event based on the event status code.

3. The method for constructing a digital scene of immersed tunnel construction process according to claim 1, characterized in that, The step of acquiring construction status data, construction control instructions, and construction log information to identify the event status code of the current construction and determine the event topology mask includes: Collect the current construction status data, including pipe section movement parameters and docking control parameters; Obtain the construction control commands currently in operation, including ballast control commands, tugboat traction commands, sinking speed control commands, and measurement verification commands; The construction log characteristics of the current construction are obtained through a log encoding function, and the construction log characteristics are used as the construction log information. A construction event recognition machine learning model is established. Based on the construction status data, the construction control instructions, and the construction log information, an event recognition input feature vector is constructed as input to output the event status code and call the corresponding event topology mask.

4. The method for constructing a digital scene of immersed tunnel construction process according to claim 1, characterized in that, The process of converting construction objects into nodes, object relationships into edges, and constructing an event-constrained dynamic scene graph by combining the current construction event status code and the event topology mask includes: The construction objects include pipe section control points, ballast tanks, tugboats, sinking equipment, measuring targets, docking faces, local areas of the foundation bed, and areas affected by hydrodynamics, in order to obtain the nodes; The object relationships include traction control relationships, ballast control relationships, hydrodynamic disturbance relationships, docking constraint relationships, measurement binding relationships, spatial proximity relationships, and risk transmission relationships, in order to obtain the edges; The node set and edge set are determined by effectively filtering the nodes and edges using the event topology mask. Obtain the node attribute matrix of the node set, the edge attribute matrix and the adjacency weight matrix of the edge set, and construct the event-constrained dynamic scene graph by combining the event status code.

5. The method for constructing a digital scene of immersed tunnel construction process according to claim 1, characterized in that, The establishment of the six-degree-of-freedom mechanism baseline prediction model includes: The generalized configuration vector and the generalized velocity vector of the pipe section are used as state variables; Based on the aforementioned state variables, a six-degree-of-freedom mechanistic motion equation is established by introducing the tugboat traction generalized force, ballast adjustment generalized force, hydrodynamic disturbance generalized force, near-bottom effect generalized force, and auxiliary control generalized force. The six-degree-of-freedom mechanism motion equations are used as the baseline prediction model for the six-degree-of-freedom mechanism.

6. The method for constructing a digital scene of immersed tunnel construction process according to claim 1, characterized in that, The process of obtaining the mechanistic baseline prediction value based on the six-degree-of-freedom mechanistic baseline prediction model includes: Under the discrete prediction step size, the predicted value of the mechanism baseline at each time point in the future prediction time domain is obtained by using the six-degree-of-freedom mechanism baseline prediction model. The predicted mechanism baseline values ​​include the predicted mechanism baseline values ​​for pipe section configuration and the predicted mechanism baseline values ​​for pipe section velocity.

7. The method for constructing a digital scene of immersed tunnel construction process according to claim 1, characterized in that, The process of obtaining residual prediction values ​​based on the event status code of the current construction, the event constraint dynamic scene diagram, the mechanism baseline prediction value, and the construction status history sequence to obtain the final prediction value includes: A spatiotemporal graph network for event-gated mechanism residuals is established, comprising an event embedding layer, a multi-relation graph attention layer, an event-gated time recursion layer, a mechanism residual gating layer, and an uncertainty output layer; The event status code, the event constraint dynamic scene diagram, the mechanism baseline prediction value, and the construction status history sequence are input into the event gate mechanism residual spatiotemporal graph network to output the residual prediction value. The final predicted value is obtained by summing the residual predicted value and the mechanism baseline predicted value.

8. The method for constructing a digital scene of immersed tunnel construction process according to claim 7, characterized in that, The step of inputting the event status code, the event constraint dynamic scene graph, the mechanism baseline prediction value, and the construction status history sequence into the event gating mechanism residual spatiotemporal graph network to output the residual prediction value includes: Based on the event status code, the construction event embedding vector is obtained through the event embedding layer, and the initial implicit features of the node are constructed by combining node features, spatial location coding, the mechanism baseline prediction value and the construction status history sequence. In the event-constrained dynamic scene graph, attention weights are obtained by utilizing the multi-relationship graph attention layer and combining the initial implicit features of the nodes. The node time memory state is obtained by performing time recursion on the attention weights based on the event gating time recursion layer. Based on the node time memory state, obtain the residual gating coefficients of the mechanism residual gating layer; Based on the residual gating coefficient, the predicted residual value is determined through the uncertainty output layer.

9. The method for constructing a digital scene of immersed tunnel construction process according to claim 1, characterized in that, The step of calculating the construction risk value based on the final predicted value, and when the construction risk value exceeds a set threshold, using construction control parameters as decision variables and solving for a construction adjustment plan, includes: Based on the final predicted value, the construction risk value is calculated using the risk weights, standardized excess quantities and excess probabilities of the predicted indicators, and in combination with the current construction hazard coefficient. When the construction risk value exceeds the set threshold, the decision variables are solved by the construction parameter rolling optimization model to obtain the construction adjustment plan.

10. A digital scene construction system for immersed tunnel construction process, characterized in that, The system includes an input device, an output device, a processor, and a memory, which are interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to invoke the program instructions to execute the digital scene construction method for immersed tunnel construction as described in any one of claims 1-9.

Images