Intelligent fine adjustment method and system for slab-type ballastless track based on multi-source perception fusion

Abstract

The application provides a slab ballastless track intelligent fine adjustment method and system based on multi-source perception fusion, and relates to the technical field of track construction. The method comprises the following steps: establishing a fastener node network topology based on a slab ballastless track discrete fastener; performing node attribute analysis on the fastener node network topology to obtain multiple node attribute characteristics; obtaining a key node sub-network topology; collecting a multi-source perception data set, mapping the multi-source perception data set by analyzing the topological relationship between key nodes and non-key nodes; and fusing two groups of multi-source perception data sets and sending them to a track fine adjustment control terminal. The application solves the technical problems of high cost and complexity of full node sensor deployment in the prior art, and the high data collection cost and low slab ballastless track fine adjustment operation efficiency caused by the perception blind area caused by the simple reduction of sensors. The application realizes high-precision global perception of the full track section with limited key nodes, and improves the slab ballastless track fine adjustment operation efficiency.

Description

Technical Field

[0001] This application relates to the field of track construction technology, specifically to a method and system for intelligent fine-tuning of slab track based on multi-source sensing fusion. Background Technology

[0002] Currently, in the intelligent fine-tuning construction and operation and maintenance of slab track, multi-source sensors are typically deployed at each fastener node to directly acquire parameters of all fastener nodes along the entire line, thereby guiding fine-tuning decisions. While this full-node sensing mode ensures fine-tuning quality to some extent, it also exposes practical problems such as a massive number of sensors, complex on-site wiring, bloated data acquisition and transmission systems, and high long-term maintenance costs, making it difficult to achieve an ideal balance between cost and fine-tuning accuracy. Because the track structure exhibits significant continuity and correlation along the track direction, the state changes of adjacent fastener nodes are not entirely independent. Existing technologies lack effective utilization of spatial, topological, and historical correlation information between nodes, resulting in the collection of a large amount of redundant data without a significant increase in key information, further affecting the fine-tuning accuracy of slab track.

[0003] In summary, existing technologies suffer from the following technical problems: the deployment of sensors at all nodes is costly and complex, while simply reducing the number of sensors can lead to blind spots, resulting in high data acquisition costs and low efficiency in fine-tuning of slab track, which further affects the accuracy of track fine-tuning. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for intelligent fine-tuning of slab track based on multi-source sensing fusion, in order to solve the technical problems in the prior art where the deployment of sensors at all nodes is costly and complex, while simply reducing the number of sensors leads to sensing blind spots, resulting in high data acquisition costs and low efficiency of slab track fine-tuning operations, which further affects the track fine-tuning accuracy.

[0005] To achieve the above objectives, this application provides a method and system for intelligent fine-tuning of slab track based on multi-source sensing fusion.

[0006] Firstly, this application provides an intelligent fine-tuning method for slab track based on multi-source sensing fusion. This method is implemented through an intelligent fine-tuning system for slab track based on multi-source sensing fusion. The method includes: establishing a fastener node network topology based on discrete fasteners of the slab track; performing node attribute analysis on the fastener node network topology to obtain multiple node attribute features, including node location importance indicators, node coverage neighborhood indicators, and node structural vulnerability indicators. Based on the aforementioned multiple node attribute features, a key node sub-network topology, including a subset of key nodes, is obtained. This key node subset is acquired through a greedy algorithm using iterative optimization analysis of the marginal benefit function of the fastener node network topology. A first set of multi-source sensing datasets of the key node sub-network topology is collected. By analyzing the topological relationship between key nodes and non-key nodes, the first set of multi-source sensing datasets is mapped to obtain a second set of multi-source sensing datasets. The first set of multi-source sensing datasets and the second set of multi-source sensing datasets are fused, and the resulting fused multi-source sensing dataset is sent to the track fine-tuning control terminal.

[0007] Optionally, initialize an empty set of key nodes; traverse each candidate node in the fastener node network topology, and calculate the marginal benefit index for adding the current candidate node using the marginal benefit function. The marginal benefit index includes a fixed-term benefit index based on the normalized weighted fitting of the multiple node attribute features; select the first candidate node according to the marginal benefit index and add it to the empty set of key nodes for multiple rounds of iterative selection until the number of nodes in the key node subset reaches a value of K, and output the key node subset.

[0008] Optionally, the marginal revenue indicator may further include an incentive revenue indicator based on the coverage incentive factor; wherein the coverage incentive factor is used to perform an analysis of the contribution of uncovered nodes to the current candidate node, including evaluating the candidate node's ability to supplement the coverage of nodes that are not yet effectively covered by the subset of key nodes.

[0009] Optionally, the fastener node network topology is updated with node status based on the first candidate node added, including covered status labels and uncovered status labels; the cumulative coverage of the fastener node network topology is cumulatively updated during multiple iterations.

[0010] Optionally, the system receives a fine-tuning control task from the track fine-tuning control terminal; analyzes the set of fastener nodes for the track section to be adjusted according to the fine-tuning control task, determines K non-critical nodes in the set of fastener nodes, and establishes K topological relationships between the K non-critical nodes and the sub-network topology of the critical nodes; maps K multi-source sensing mapping datasets corresponding to the K non-critical nodes according to the first set of multi-source sensing datasets and the K topological relationships; and outputs the K multi-source sensing mapping datasets as the second set of multi-source sensing datasets.

[0011] Optionally, for any one of the K non-critical nodes, a local topology search is performed on the sub-network topology of the critical node to obtain an initial association set whose topological distance is less than a preset distance threshold; the relationship strength set between the initial association set and the corresponding non-critical node is calculated, wherein each relationship strength is obtained by weighted averaging of spatial distance decay factor, topological path dependency factor, and historical data correlation factor; the top M critical nodes of the relationship strength set are selected as the final association set, and a topological relationship description between each non-critical node and the corresponding top M critical nodes is established to obtain K topological relationships.

[0012] Optionally, the top M key nodes of the relationship strength set are selected as the final association set, wherein the value of M is dynamically adjusted in a positive correlation with the adjustment accuracy index of the fine-tuning control task.

[0013] Optionally, based on K topological relationships, M multi-source sensing datasets for each non-critical node based on the corresponding topological relationship are extracted from the first group of multi-source sensing datasets; the M multi-source sensing datasets are weighted and fused to obtain the multi-source sensing mapping dataset fused for each non-critical node, and so on to obtain the K multi-source sensing mapping datasets corresponding to the K non-critical nodes.

[0014] Optionally, the first set of multi-source sensing datasets and the second set of multi-source sensing datasets are fused to obtain an initial fused multi-source sensing dataset; when the difference between the fusion results of adjacent time steps of the fused multi-source sensing dataset exceeds a preset threshold, the fused multi-source sensing dataset is smoothed and corrected.

[0015] Secondly, this application also provides an intelligent fine-tuning system for slab track based on multi-source sensing fusion, used to execute the intelligent fine-tuning method for slab track based on multi-source sensing fusion as described in the first aspect. The intelligent fine-tuning system for slab track based on multi-source sensing fusion includes: a fastener node network topology establishment module, used to establish a fastener node network topology based on discrete fasteners of slab track; a node attribute analysis module, used to perform node attribute analysis on the fastener node network topology to obtain multiple node attribute features, including node location importance indicators, node coverage neighborhood indicators, and node structural vulnerability indicators; and a key node sub-network topology identification module. The system comprises the following modules: a separation module for obtaining a key node sub-network topology, including a subset of key nodes, based on the multiple node attribute features. The key node subset is obtained by iterative optimization analysis of the marginal benefit function of the fastener node network topology using a greedy algorithm. A data mapping module for collecting a first set of multi-source sensing datasets of the key node sub-network topology and mapping the first set of multi-source sensing datasets by analyzing the topological relationship between key nodes and non-key nodes to obtain a second set of multi-source sensing datasets. A data fusion module for fusing the first set of multi-source sensing datasets and the second set of multi-source sensing datasets and sending the resulting fused multi-source sensing dataset to the track fine-tuning control terminal.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: Based on the distribution of discrete fasteners of slab track, a fastener node network topology is established, the importance of each node is analyzed, a greedy algorithm is used to select a subset of the most representative key nodes for sensor deployment, and the topological relationship between non-key nodes and key nodes is used to extrapolate the data of a large number of non-key nodes based on the data of a small number of key nodes. The real data and the mapped data are fused to form a multi-source sensing dataset for the entire track section, which is sent to the fine-tuning control terminal. This achieves fine-tuning sensing capabilities that are similar to or even better than full-node deployment with low cost and low complexity, thereby improving the fine-tuning efficiency and accuracy of slab track.

[0017] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the intelligent fine-tuning method for slab track based on multi-source sensing fusion proposed in this application.

[0020] Figure 2 This is a schematic diagram of the intelligent fine-tuning system for slab track based on multi-source sensing fusion, as described in this application.

[0021] Figure labeling: 11 Fastener node network topology establishment module, 12 Node attribute analysis module, 13 Key node sub-network topology identification module, 14 Data mapping module, 15 Data fusion module. Detailed Implementation

[0022] This application provides a method and system for intelligent fine-tuning of slab track based on multi-source sensing fusion. It addresses the technical problems in existing technologies where the deployment of sensors at all nodes is costly and complex, while simply reducing sensors leads to blind spots, resulting in high data acquisition costs, low efficiency in slab track fine-tuning, and further impacting track fine-tuning accuracy. Based on the distribution of discrete fasteners in the slab track, a fastener node network topology is established. The importance of each node is analyzed, and a greedy algorithm is used to select a subset of the most representative key nodes for sensor deployment. Utilizing the topological relationships between non-key and key nodes, data from a large number of non-key nodes is extrapolated from data from a small number of key nodes. The real data and the mapped data are fused to form a multi-source sensing dataset for the entire track section, which is then sent to the fine-tuning control terminal. This achieves fine-tuning sensing capabilities similar to or even better than full-node deployment at low cost and low complexity, thereby improving the efficiency and accuracy of slab track fine-tuning.

[0023] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0024] Example 1, please refer to the appendix. Figure 1This application provides a method for intelligent fine-tuning of slab track based on multi-source sensing fusion. The method is applied to an intelligent fine-tuning system for slab track based on multi-source sensing fusion. The method specifically includes the following steps: Establish a fastener node network topology based on discrete fasteners for slab track.

[0025] Specifically, slab track is a highly stable track structure composed of precast concrete track slabs, a cement asphalt mortar adjustment layer, and a concrete base. Its sleepers are cast integrally with the track bed slab, and the track geometry is primarily adjusted via fasteners. Discrete fasteners are independent components installed at fixed intervals on the track slab, used to lock the rails, provide elasticity, and perform geometric adjustments.

[0026] Extract the design position coordinates of all fasteners along the entire line from the track design drawings. For newly built lines, the theoretical three-dimensional spatial coordinates of each fastener can be calculated based on the line's horizontal and vertical profile design parameters and fastener spacing. For existing lines, the actual spatial coordinates of each fastener can be measured using a total station.

[0027] Each fastener is assigned a unique number, such as starting from the line origin and numbering in ascending order of mileage, from 1 to N. Along the rail extension direction, according to the actual installation sequence, each fastener is connected to its preceding and following fasteners; these connections are the edges in the network. Each edge is assigned a weight to quantify the strength of the association between two fasteners; weights include spatial distance, topological path dependency factors, etc. All fastener nodes and their connecting edges are organized into a graph structure, commonly using an adjacency list. Each node records its number, 3D coordinates, and the numbers of all its connected neighboring nodes and the weights of their corresponding edges. The system checks for isolated nodes (fasteners without any edges); if any are found, it checks whether the design or measured data has omitted any connections. Simultaneously, it checks the connection logic in complex sections such as turnout areas to ensure the network fully reflects the continuous transmission relationships of track geometry.

[0028] The fastener node network topology is analyzed to obtain multiple node attribute features, including node location importance index, node coverage neighborhood index, and node structural vulnerability index.

[0029] Specifically, the track parameters for each fastener's location are obtained, including the horizontal curve radius, superelevation, longitudinal slope, vertical curve radius, whether it is located in a turnout area, or whether it is located in a bridge-tunnel transition section. For each fastener, an initial location importance score is assigned based on its location characteristics. For example, fasteners located on curves with a radius less than 2000m have a base importance score of 0.8; fasteners located on curves with a radius between 2000m and 4000m have a base score of 0.5; and fasteners located on curves or straight sections with a radius greater than 4000m have a base score of 0.2. Fasteners located in turnout areas have an additional base score of 0.5. Fasteners located within ±20m of bridge-tunnel transition sections or road-bridge transition sections have an additional base score of 0.3. Fasteners located on vertical curves have an additional base score of 0.2. The location importance values ​​of all fasteners are normalized by dividing the original value of each node by the maximum value for the entire line. This ensures that the location importance index values ​​for all nodes are distributed between 0 and 1, with higher values ​​indicating greater location importance. The node location importance index measures the sensitivity of a track section where a fastener is located to track geometry smoothness. For example, fasteners located in special sections such as curves, turnouts, bridge-tunnel transition sections, and gradient change points have a greater impact on train safety and smoothness due to changes in their condition, thus their location importance is higher.

[0030] Based on the shortest path distance in the network topology, if the shortest path length from a non-critical node to a target node is less than or equal to a preset neighborhood radius, the non-critical node is considered to be covered by the target node. For each fastener node, a breadth-first search is performed in the network topology, starting from this node, to record all nodes whose topological distance does not exceed the neighborhood radius. The topological distance is calculated as the number of hops on the edge. The total number of these covered nodes is counted and recorded as the coverage count. The larger the coverage count, the larger the coverage neighborhood of the node. Nodes at different distances should contribute differently to the coverage, and a weighted coverage neighborhood value is calculated. For example, the contribution weight of a node at a distance of 0 is 1.0; the contribution weight of a node at a distance of 1 is 0.8; the contribution weight of a node at a distance of 2 is 0.5; and the contribution weight of a node at a distance of 3 is 0.3. The weighted coverage value is normalized across all nodes on the entire line to obtain the coverage neighborhood index of each node, ranging from 0 to 1. The node coverage neighborhood index is used to measure the ability of a fastener node to influence how many other fastener nodes around it. In a track structure, the deformation or adjustment of a fastener will affect fasteners within a certain range adjacent to it through the continuity of the track slab and the rail.

[0031] Calculate the degree centrality of each node. The degree of a node refers to the number of its directly connected neighboring nodes. For standard lines, most fastener nodes have a degree of 2, but nodes in turnout areas may have a degree of 3 or more. Divide the node's degree by the maximum degree of the entire line to obtain the normalized degree centrality. Calculate the betweenness centrality of each node. Betweenness centrality measures the degree to which a node acts as a bridge in the network; that is, the proportion of paths passing through that node in the shortest paths between all pairs of nodes. A higher betweenness centrality indicates a more vulnerable node; its failure will sever connections between many pairs of nodes. Combine the degree centrality and betweenness centrality with weights to obtain the original structural vulnerability value of the node. The weights can be set based on engineering experience, such as a degree centrality weight of 0.3 and a betweenness centrality weight of 0.7, as betweenness centrality better reflects structural vulnerability. Normalize the original structural vulnerability value to the range of 0 to 1 across the entire line to obtain the structural vulnerability index for each node. The node structural vulnerability index is used to measure the vulnerability of a fastener node at the network structure level. If a node is in a critical connection position in the network, or if the node itself is prone to failure or deformation in the past, its structural vulnerability is high and needs to be prioritized for detection and monitoring.

[0032] The three indicators for each fastener node—node location importance, node coverage neighborhood, and node structural vulnerability—are stored as vectors, serving as the node's attribute features. Through quantitative calculations on the abstract network model, the structural behavior of each fastener node is transformed into comparable numerical features. The location importance indicator reflects engineering sensitivity, the coverage neighborhood indicator reflects information representativeness, and the structural vulnerability indicator reflects network criticality. These three dimensions complement each other, comprehensively characterizing the overall value of a fastener node in the perception network and avoiding selection bias caused by a single indicator.

[0033] Based on the aforementioned multiple node attribute features, a key node sub-network topology including a subset of key nodes is obtained. The key node subset is obtained by using a greedy algorithm to perform iterative optimization analysis of the marginal benefit function of the fastener node network topology.

[0034] Furthermore, this application also includes the following steps: initializing the empty set of key nodes; calculating the marginal benefit index for adding the current candidate node by traversing each candidate node in the fastener node network topology and using the marginal benefit function, wherein the marginal benefit index includes a fixed-term benefit index based on the normalized weighted fitting of the multiple node attribute features; selecting the first candidate node according to the marginal benefit index and adding it to the empty set of key nodes for multiple rounds of iterative selection until the number of nodes in the key node subset reaches a value of K, and outputting the key node subset.

[0035] Furthermore, this application also includes the following steps: the marginal revenue indicator further includes an incentive revenue indicator based on the coverage incentive factor; wherein, the coverage incentive factor is used to perform an analysis of the contribution of uncovered nodes to the current candidate node, including evaluating the candidate node's ability to supplement the coverage of nodes that are not yet effectively covered by the subset of key nodes.

[0036] Specifically, an empty set of critical nodes is initialized, and the set of nodes not fully covered is initially set to all component nodes, since no nodes are covered at the beginning. A coverage threshold is set to determine whether a node is fully covered. For example, when the sum of the cumulative coverage contributions of the existing subset of critical nodes to a certain node reaches 0.8 or higher, the node is considered to be fully covered, and it is removed from the set of nodes not fully covered.

[0037] For each candidate node, a weighted fit is performed using the node location importance index, the node coverage neighborhood index, and the node structural vulnerability index. Let the weights be ω1, ω2, and ω3, where ω1 + ω2 + ω3 = 1. The weights can be adjusted according to engineering requirements, such as a node location importance weight of 0.4, a node coverage neighborhood weight of 0.3, and a node structural vulnerability weight of 0.3. The fixed-item benefit index is calculated as: Fixed-item benefit = ω1 × Location importance index + ω2 × Coverage neighborhood index + ω3 × Structural vulnerability index. Since the node location importance index, node coverage neighborhood index, and node structural vulnerability index have been normalized to between 0 and 1, the fixed-item benefit index naturally also falls between 0 and 1.

[0038] For candidate node i and uncovered node j, define the coverage contribution φ. ij Let be the coverage capability of i to j, and its value varies with the topological distance d between i and j. ij (Measured by the number of hops on an edge) It increases and then decays. An exponential decay function φ is used. ij =exp(-α×d ij ), where α is the attenuation coefficient, typically taken as 0.5 to 1.0. When d ij When φ is 0, ij It is 1.0; when d ij When φ is 1, ij Approximately 0.606; when d ij When φ is 2, ij Approximately 0.367; when d ij When φ is 3, ij Approximately 0.223; when d ij When ≥5, φ ij This can be disregarded. In each iteration, for candidate node i, calculate its coverage contribution φ to each node j in the current set of insufficiently covered nodes. ij Then sum them up. Simultaneously, an incentive factor I is introduced. coverWhen there are nodes in the insufficiently covered node set that are not effectively covered by any of the selected key nodes, I cover Set the value to 1.0; otherwise, set it to 0.0. A higher value for the coverage incentive revenue indicator indicates that the candidate node can supplement and cover more currently uncovered nodes, demonstrating its pioneering value. Adding the fixed revenue indicator to the coverage incentive revenue indicator yields the marginal revenue of candidate node i after it is added to the current selected set, i.e., marginal revenue = fixed revenue + coverage incentive revenue. The coverage incentive revenue depends on the set of currently under-covered nodes and changes dynamically with iteration; therefore, the marginal revenue function is iteration-dependent. The fixed revenue does not change with iteration.

[0039] Set the target size K value for the key node subset. The determination of K needs to comprehensively consider factors such as sensor budget, line length, and accuracy requirements. For example, for a test section with 500 nodes, K=30 can be set. Iterate through all candidate nodes not yet selected for the key node subset; for each candidate node, calculate its marginal benefit based on the current set of insufficiently covered nodes; select the candidate node with the largest marginal benefit and add it to the key node subset; update the set of insufficiently covered nodes. For each node j in the updated set of insufficiently covered nodes, calculate the cumulative coverage contribution of the current key node subset to it, i.e., the sum of the coverage contributions of all nodes in the current key node subset to each node j; if the cumulative coverage contribution is greater than or equal to the coverage threshold, remove node j from the set of insufficiently covered nodes. Repeat this process until the iteration ends, i.e., the number of nodes in the key node subset reaches the preset K value, and obtain the key node subset at this point. Record the node number, coordinates, and attribute indicators of each node in the key node subset at this point. The corresponding sub-network topology consists of the nodes in the key node subset and the edges between them in the original network.

[0040] The critical node subset is a finite number of nodes selected from all fastener nodes using a greedy algorithm. These nodes serve as the locations for actual sensor deployment, and the data collected will be used to infer the state of all non-critical nodes in the entire network. The greedy algorithm is a stepwise optimization strategy. In each step, the node that brings the maximum marginal benefit in the current state is selected from the remaining candidate nodes and added to the result set. This process is repeated until the target number is reached. While the greedy algorithm cannot guarantee global absolute optimality, it can provide a near-optimal approximate solution when the marginal benefit function satisfies submodularity, and it is computationally efficient. The marginal benefit function measures the overall benefit added by adding a candidate node to the currently selected critical node subset. It includes a fixed benefit metric and a coverage incentive benefit metric. The fixed benefit metric is based on three attribute characteristics of the node itself, while the coverage incentive benefit metric is based on the candidate node's ability to supplement the coverage of nodes that have not yet been effectively covered. The set of insufficiently covered nodes is the set of nodes that have not yet been effectively covered by the selected critical node subset during the current greedy iteration. As the selected critical node subset expands, the set of insufficiently covered nodes gradually shrinks. The K value is a preset budget or resource constraint constant that represents the number of critical nodes that need to be selected in the end.

[0041] For example, the number of nodes is set to 500, the fixed item revenue weights are ω1=0.4, ω2=0.3, ω3=0.3, the coverage decay coefficient is α=0.6, the coverage threshold is θ=0.8, and the incentive factor is I. cover The value is dynamically used during the iteration process, taking 1.0 when the set of insufficiently covered nodes U is not empty, and 0.0 when U is empty. Based on the node attribute data, the fixed item revenue for each node is calculated: the fixed item revenue for node 100 is 0.3202, for node 200 it is 0.6742, for node 250 it is 0.92, and for node 300 it is 0.5037. In the first iteration: the initial key node S is empty, and the set of insufficiently covered nodes U contains all 500 nodes. For all candidate nodes, the marginal revenue is calculated as: Marginal Revenue = Fixed Item Revenue + Coverage Incentive Revenue. In the Coverage Incentive Revenue, since U is not empty, I... cover =1.0. For node 250, its coverage contribution is summed over a large number of nodes in the insufficiently covered node set U. Node 250 is located at a turnout branch point, with approximately 3 nodes within 1 hop of the topological distance, approximately 7 nodes within 2 hops, and approximately 15 nodes within 3 hops. A rough calculation of Σφ is performed. ij The marginal benefit is 7.2343. The marginal benefit is 8.1543. The Σφ of node 200. ijThe marginal benefit is approximately 5.2, with a marginal return of 5.8742. Node 250 has the highest marginal return, therefore it is selected to join S in the first round. The set of insufficiently covered nodes U is updated. Node 250 and its surrounding nodes within 3 hops have a cumulative coverage contribution exceeding 0.8, so it is removed from the set of insufficiently covered nodes U, reducing the number of nodes in U by approximately 15. Second iteration: Currently, S = {250}, and U has approximately 485 nodes. The marginal returns of the remaining candidate nodes are calculated. The coverage incentive term return for node 200 is recalculated because there are still many nodes in U, and its Σφ... ij The coverage incentive for node 100 is approximately 5.0, with a marginal benefit of 5.6742. The coverage incentive for node 100 is approximately 4.0, with a marginal benefit of 4.3202. The marginal benefit for node 201 may be higher. After calculation, it is assumed that node 201 is selected. As the initial key node S expands, the set of insufficiently covered nodes U gradually shrinks, and the coverage incentive gradually decreases. When U is empty, meaning all nodes are fully covered by the existing S, the coverage incentive is 0, and the marginal benefit is entirely determined by the fixed benefit. At this point, the greedy algorithm will supplement the remaining key nodes according to the fixed benefit from high to low, until K=30 is reached. After 30 iterations, the selected key node subset S contains 30 nodes, distributed in key locations such as turnout areas, curve midpoints, curve beginnings and ends, and track slab joints, with minimal overlap between them, achieving maximum information coverage with a limited number of sensors. The average coverage contribution of the key node subset S to all 500 nodes is calculated. The cumulative coverage contribution of each node j is the sum of φ for all nodes in the key node subset S. The calculations show that the minimum coverage contribution is 0.62, the maximum is 1.85, and the average is 1.12, all greater than the threshold of 0.8, indicating that the key node subset S can effectively cover the entire network.

[0042] The marginal revenue metric includes not only a fixed-term revenue metric based on a weighted fit of multiple node attribute features, but also an incentive-term revenue metric based on coverage incentive factors. This is used to analyze the contribution of uncovered nodes to the current candidate nodes, including assessing the candidate nodes' ability to supplement the coverage of nodes not yet effectively covered by the key node subset, and to cover nodes that currently have no effective coverage from any key node. In each iteration, the marginal revenue is calculated for all candidate nodes, and the node with the highest marginal revenue is added to the key node subset. Due to the existence of incentive-term revenue, candidate nodes that can cover more nodes in the insufficiently covered node set will receive higher marginal revenue, and may be selected even if their fixed-term revenue is not the highest, thus preventing sensors from concentrating entirely on the nodes with the highest fixed-term revenue while ignoring other areas. As iterations progress, the set of insufficiently covered nodes gradually shrinks, and the impact of incentive-term revenue on point selection gradually weakens. When the set of insufficiently covered nodes is empty, incentive-term revenue disappears, and subsequent point selection is based entirely on fixed-term revenue from high to low to replenish the number of key nodes to K.

[0043] From thousands of fastener nodes, a subset of key nodes with the highest overall value is selected quantitatively, avoiding the arbitrariness and subjectivity of experience-based point selection. Fixed-term benefits ensure the importance of the selected nodes, while coverage incentive benefits ensure complementary coverage, avoiding redundancy. Given a budget of K for the number of sensors, the greedy algorithm can provide a near-optimal point selection scheme in polynomial time, enabling the deployment of a small number of sensors to achieve near-full-node sensing coverage.

[0044] Furthermore, this application also includes the following steps: updating the node status of the fastener node network topology based on the first candidate node added, including covered status labels and uncovered status labels; and cumulatively updating the cumulative coverage of the fastener node network topology during multiple iterations.

[0045] Specifically, in each iteration of the greedy algorithm, after selecting a candidate node with the highest marginal benefit and adding it to the set of selected key nodes, the node states of the entire network topology need to be updated immediately to prepare for the next iteration. Define the coverage determination rules. Clarify the range of other nodes that a selected key node can cover. Typically, an effective coverage radius is defined based on the network topology; for example, all nodes whose network distance (shortest path hop count) is less than or equal to 2 are considered to be covered by the key node. Next, update the covered / uncovered status labels. For the newly added key node, find all nodes in the network topology whose distance is within its effective coverage radius and set their status labels to covered. Simultaneously, check the other existing key nodes in the key node subset S to ensure that the node labels within their coverage range are also maintained as covered. All nodes not covered by any key node retain their status labels as uncovered. Calculate and update the cumulative coverage, maintaining a counter for each node in the network with an initial value of 0. During each state update, for each newly added key node, the cumulative coverage counter of that node is incremented by a coverage contribution value for each node within its effective coverage radius. This contribution value is typically inversely proportional to the distance between the two nodes, e.g., contribution value = 1 / (network distance). Simultaneously, the coverage contributions of existing nodes in the key node subset S to other nodes in the network, already accumulated in previous iterations, remain unchanged. Cumulative coverage reflects the overall enrichment of perceived resources for a node. The updated state is used to calculate the incentive factor for the next round. When calculating the collaborative coverage gain of candidate nodes in the next iteration, the incentive factor directly depends on the node's latest state. A candidate node receives a high incentive if it can cover an uncovered node; however, if it can cover a node with already high cumulative coverage (i.e., it is fully perceived), the incentive will be very low or even zero. Thus, through dynamic state labels and cumulative coverage, the greedy algorithm can determine which areas of the network are fully perceived and which remain blind spots, guiding subsequent choices to explore and cover those weakly perceived areas.

[0046] A first set of multi-source sensing datasets is collected from the sub-network topology of the key nodes. The first set of multi-source sensing datasets is then mapped by analyzing the topological relationship between the key nodes and non-key nodes to obtain a second set of multi-source sensing datasets.

[0047] Furthermore, this application also includes the following steps: receiving a fine-tuning control task issued by the track fine-tuning control terminal; analyzing the set of fastener nodes for the track section to be adjusted according to the fine-tuning control task, determining K non-critical nodes in the set of fastener nodes, and establishing K topological relationships between the K non-critical nodes and the sub-network topology of the critical nodes; mapping K multi-source sensing mapping datasets corresponding to the K non-critical nodes according to the first set of multi-source sensing datasets and the K topological relationships; and outputting the K multi-source sensing mapping datasets as the second set of multi-source sensing datasets.

[0048] Furthermore, this application also includes the following steps: for any non-critical node among the K non-critical nodes, a local topology search is performed on the sub-network topology of the critical node to obtain an initial association set whose topological distance is less than a preset distance threshold; the relationship strength set between the initial association set and the corresponding non-critical node is calculated, wherein each relationship strength is obtained by weighted averaging of spatial distance decay factor, topological path dependency factor and historical data correlation factor; the top M critical nodes of the relationship strength set are selected as the final association set, and a topological relationship description between each non-critical node and the corresponding top M critical nodes is established to obtain K topological relationships.

[0049] Furthermore, this application also includes the following steps: selecting the top M key nodes of the relationship strength set as the final association set, wherein the value of M is dynamically adjusted in a positive correlation through the adjustment accuracy index of the fine-tuning control task.

[0050] Furthermore, this application also includes the following steps: based on K topological relationships, extract M multi-source sensing datasets for each non-critical node based on the corresponding topological relationship from the first group of multi-source sensing datasets; perform weighted fusion on the M multi-source sensing datasets to obtain the fused multi-source sensing mapping dataset for each non-critical node, and so on to obtain K multi-source sensing mapping datasets corresponding to the K non-critical nodes.

[0051] Specifically, the track fine-tuning control terminal issues a fine-tuning control task, including the start and end mileage of the track section requiring fine-tuning, a list of fastener nodes that need adjustment within that section, the target adjustment amount for each node, and the adjustment accuracy index for this fine-tuning operation. The system receives the fine-tuning control task from the track fine-tuning control terminal and extracts all fastener nodes within the task section.

[0052] Based on the set of fastener nodes within the task segment, compare it with the existing critical node subset. Nodes belonging to the critical node subset are retained as critical nodes; nodes not belonging to the critical node subset are non-critical nodes. Let there be P fastener nodes in the segment, with Q being critical nodes. Then the number of non-critical nodes, K, is equal to P minus Q. Record the numbers of these non-critical nodes.

[0053] For each non-critical node, a correlation analysis subprocess is initiated. Centered on this non-critical node, a breadth-first search is performed within the entire fastener node network topology. The search depth is controlled by a preset topological distance threshold, and the target node type is limited to critical nodes. All critical nodes found within this search range constitute the initial correlation set for that non-critical node. For each critical node in the initial correlation set, the quantitative relationship strength between it and the current non-critical node is calculated. The strength value is obtained by a weighted comprehensive calculation of three core factors: First, the spatial distance decay factor, usually calculated using an exponential decay function; the closer the physical distance, the larger the factor value. Second, the topological path dependence factor, obtained by calculating the reciprocal of the weights of all edges on the shortest path between two nodes in the network; the stronger the connection and the fewer the hops on the path, the larger the factor value. Third, the historical data correlation factor, obtained by analyzing the Pearson correlation coefficient between the two in historical monitoring data sequences, such as the elevation change over the past 30 days. Preset weight coefficients are assigned to these three factors, and a weighted sum is performed to obtain the final relationship strength value. After completing the strength calculations for all initial associated nodes, the value of parameter M is dynamically determined based on the adjustment accuracy index specified in the fine-tuning control task. The spatial distance attenuation factor is the physical straight-line distance or track distance between critical and non-critical nodes; the topology path dependence factor is the complexity of the topological path between two nodes; the historical data correlation factor is based on the statistical correlation between historical measurement data of critical and non-critical nodes over a past period.

[0054] Based on this M value, the top M key nodes in terms of relationship strength are selected from the initial association set to form the final association set. For each non-key node, the unique identifiers of the M key nodes in its final association set and their corresponding relationship strength values ​​are recorded, thus obtaining a topological relationship description. When this process has been completed for all K non-key nodes, K such relationship descriptions are obtained, defining the mapping rules from key node perceived data to overall scene state data.

[0055] The value of M is dynamically determined based on the adjustment accuracy index in the fine-tuning control task. The adjustment accuracy index is typically expressed as an allowable error range; for example, M is 5 when the error does not exceed ±0.3mm, M is 3 when the error does not exceed ±0.5mm, and M is 2 when the error does not exceed ±1.0mm. The higher the accuracy requirement, the larger the value of M. The key nodes in the initial association set are sorted in descending order of relationship strength, and the top M key nodes are selected as the final association set for the non-key node. The relationship strength values ​​corresponding to these M key nodes are recorded and used as fusion weights later.

[0056] Before the actual fine-tuning operation begins or during real-time operation, multi-source sensors deployed on key nodes collect multi-source status parameters of each key node. The data is then packaged, timestamped, and transmitted to the central processing unit to form a structured first set of multi-source sensing datasets. This is typically a data table, with each row corresponding to a key node and each column corresponding to a sensing parameter. The first set of multi-source sensing datasets consists of real-time data collected by multi-source sensors actually deployed on the sub-network topology of the key nodes. Each key node may be equipped with various sensors, such as displacement sensors measuring the lateral and vertical displacement of fasteners, strain sensors measuring the force on fasteners, and temperature sensors measuring ambient temperature and track slab temperature.

[0057] The process loads K pre-stored topological relationships for each non-critical node. For each non-critical node, based on the M critical node numbers in its topological relationship description, it extracts the corresponding multi-source sensing data from the first set of multi-source sensing datasets. The data for each critical node may contain multiple sensor dimensions, resulting in M ​​vectors, each with a dimension equal to the number of sensor types. For each non-critical node, the extracted sensing data from the M critical nodes is averaged using the corresponding relationship strength as weights. The average is calculated for each type of sensor data. After calculating the average for all sensor dimensions, a fused multi-source sensing data vector is obtained, which is the mapped sensing dataset for that non-critical node. This process is repeated for each of the K non-critical nodes to obtain the K multi-source sensing mapping datasets corresponding to the K non-critical nodes. The multi-source sensing mapping dataset represents the inferred state of a non-critical node, generated by weighted fusion of the sensing data from its associated M critical nodes. It contains parameters of the same type as the original critical nodes, but these parameters are calculated rather than directly measured. Once the mapping data for all K non-critical nodes has been calculated, they are aggregated in order of node ID to generate a complete second set of multi-source sensing datasets.

[0058] Specifically, the first set of multi-source sensing datasets stored in a database or memory, along with K pre-calculated and stored topological relationship descriptions, are loaded into the processing engine. An iterative fusion loop is initiated, iterating through all non-critical node indices k from 1 to K. For the k-th non-critical node, its topological relationship description is read, obtaining a list of IDs of its M associated critical nodes and a corresponding list of relationship strength weights. Using these M critical node IDs as search keys, a query operation is performed in the first set of multi-source sensing datasets to precisely extract the measurement values ​​of all P parameters corresponding to these M critical nodes. The result is an M-row... A temporary data matrix of P columns and an M-dimensional weight vector are used. For each sensing parameter, such as elevation and lateral displacement, a fusion calculation is performed. For each sensing parameter, the corresponding M measurements are extracted from the temporary data matrix to form an M-dimensional vector. The inferred value for the sensing parameter at this non-critical node is calculated; the inferred value is the average of the M-dimensional vector. This calculation is performed sequentially for all parameters. The calculated P inferred values ​​are organized according to parameter order to form the multi-source sensing mapping dataset for the k-th non-critical node, and bound to the node's unique ID, stored in a result container. After completing the calculation for one node, the process is repeated for the next non-critical node until all K non-critical nodes have been processed. When the loop ends, the result container has stored the complete mapping data for the K non-critical nodes in sequence. This data is then organized and packaged to form a structured multi-source sensing mapping dataset of K rows, typically output as a K-row dataset. The data matrix in column P has rows ordered in the same order as the non-critical node IDs, and columns ordered in the same order as the perception parameter types.

[0059] The second set of multi-source sensing datasets is structurally identical to the first set and serves as the final output of track state sensing, provided to the track fine-tuning control terminal. By deploying real sensors at key nodes, sensing data for all nodes can be obtained through topological mapping, significantly reducing hardware costs and maintenance complexity. The calculation of relationship strength incorporates various physical and statistical information by introducing three factors: spatial distance attenuation, topological path dependence, and historical correlation, ensuring the confidence level of the mapping. The mapping result for each non-critical node can be traced back to its M associated critical nodes and their corresponding relationship strengths, facilitating verification and debugging by engineers.

[0060] The first set of multi-source sensing datasets and the second set of multi-source sensing datasets are fused, and the resulting fused multi-source sensing dataset is sent to the track fine-tuning control terminal.

[0061] Furthermore, this application also includes the following steps: fusing the first group of multi-source sensing datasets and the second group of multi-source sensing datasets to obtain an initial fused multi-source sensing dataset; when the difference between the fusion results of adjacent time steps of the fused multi-source sensing dataset exceeds a preset threshold, smoothing correction is performed on the fused multi-source sensing dataset.

[0062] Specifically, since the two datasets have the same parameter columns and the node IDs are unique and ordered in their respective sets, all rows of the first dataset and all rows of the second dataset are merged and sorted according to the actual spatial order of the nodes on the track to generate an initial fused multi-source sensing dataset containing all N nodes of the target segment.

[0063] A time dimension is introduced into the initial fused multi-source sensing dataset, and a cache is maintained to store the fused dataset from the previous processing cycle. After generating the initial fused dataset for the current time t, for each node and each key parameter in the dataset, the absolute difference between its current value and the cached value from the previous time step is calculated. This difference is compared with a preset threshold for that parameter. If the difference is less than or equal to the preset threshold, the current value is considered reasonable and reliable, and is directly retained. If the difference is greater than the preset threshold, the value is determined to be an abnormal jump, triggering a smoothing correction algorithm, such as a first-order exponential smoothing algorithm with a length of 5 cycles, and a correction value Vt1=a. Vt+(1-a) V(t-1), where a is a smoothing factor, such as 0.3, V(t-1) is the smoothed value of the node parameter at the previous time step, Vt is the node parameter, and Vt1 is the corrected parameter of the node parameter. The calculated smoothed correction value Vt1 replaces the original value Vt in the initial fused dataset. After performing this threshold judgment and possible smoothing correction on all nodes and parameters in the dataset, the final reliable fused multi-source sensing dataset is obtained.

[0064] The final dataset, after time-domain smoothing, along with metadata such as timestamps, segment identifiers, and data quality identifiers, is encapsulated according to the communication protocol agreed upon with the track fine-tuning control terminal and sent to the terminal via the network. Upon receiving the dataset, the track fine-tuning control terminal triggers the subsequent intelligent analysis process.

[0065] In summary, the intelligent fine-tuning method for slab track based on multi-source sensing fusion provided in this application has the following technical effects: Based on the distribution of discrete fasteners of the slab track, a fastener node network topology is established, the importance of each node is analyzed, and a greedy algorithm is used to select a subset of the most representative key nodes for sensor deployment. Utilizing the topological relationship between non-key and key nodes, data from a large number of non-key nodes is inferred from data from a small number of key nodes. The real data and the mapped data are fused to form a multi-source sensing dataset for the entire track section, which is then sent to the fine-tuning control terminal. This achieves fine-tuning sensing capabilities similar to or even better than full-node deployment with low cost and low complexity, thereby improving the efficiency and accuracy of slab track fine-tuning.

[0066] Example 2: Based on the same inventive concept as the intelligent fine-tuning method for slab track based on multi-source sensing fusion in Example 1, this application also provides an intelligent fine-tuning system for slab track based on multi-source sensing fusion. Please refer to the appendix. Figure 2 The intelligent fine-tuning system for slab track based on multi-source sensing fusion includes: The fastener node network topology establishment module 11 is used to establish a fastener node network topology based on discrete fasteners of slab track; the node attribute analysis module 12 is used to perform node attribute analysis on the fastener node network topology and obtain multiple node attribute features, including node location importance index, node coverage neighborhood index, and node structural vulnerability index; the key node sub-network topology identification module 13 is used to obtain a key node sub-network topology including a subset of key nodes based on the multiple node attribute features, wherein the key node subset is obtained by iterative optimization analysis of the marginal benefit function of the fastener node network topology using a greedy algorithm; the data mapping module 14 is used to collect a first set of multi-source sensing datasets of the key node sub-network topology, and map the first set of multi-source sensing datasets by analyzing the topological relationship between key nodes and non-key nodes to obtain a second set of multi-source sensing datasets; the data fusion module 15 is used to fuse the first set of multi-source sensing datasets and the second set of multi-source sensing datasets, and send the obtained fused multi-source sensing dataset to the track fine-tuning control terminal.

[0067] Furthermore, the node attribute analysis module 12 in the intelligent fine-tuning system for slab track based on multi-source perception fusion is also used for: initializing the empty set of key nodes; calculating the marginal benefit index for adding the current candidate node by traversing each candidate node in the fastener node network topology using the marginal benefit function, wherein the marginal benefit index includes a fixed-term benefit index based on the normalized weighted fitting of the multiple node attribute features; selecting the first candidate node according to the marginal benefit index and adding it to the empty set of key nodes for multiple rounds of iterative selection until the number of nodes in the key node subset reaches a value of K, and outputting the key node subset.

[0068] Furthermore, the node attribute analysis module 12 in the intelligent fine-tuning system for slab track based on multi-source perception fusion is also used for: the marginal benefit index further includes an incentive benefit index based on the coverage incentive factor; wherein, the coverage incentive factor is used to perform an analysis of the contribution of uncovered nodes to the current candidate nodes, including evaluating the supplementary coverage capability of the candidate nodes to nodes that are not yet effectively covered by the subset of key nodes.

[0069] Furthermore, the node attribute analysis module 12 in the intelligent fine-tuning system for slab track based on multi-source perception fusion is also used to: update the node status of the fastener node network topology according to the added first candidate node, including covered status labels and uncovered status labels; and cumulatively update the cumulative coverage of the fastener node network topology when multiple iterations are performed.

[0070] Furthermore, the data mapping module 14 in the intelligent fine-tuning system for slab track based on multi-source sensing fusion is also used for: receiving fine-tuning control tasks issued by the track fine-tuning control terminal; analyzing the set of fastener nodes according to the track section that needs to be adjusted according to the fine-tuning control task, determining K non-critical nodes in the set of fastener nodes, and establishing K topological relationships between the K non-critical nodes and the sub-network topology of the critical nodes; mapping K multi-source sensing mapping datasets corresponding to the K non-critical nodes according to the first set of multi-source sensing datasets and the K topological relationships; and outputting the K multi-source sensing mapping datasets as the second set of multi-source sensing datasets.

[0071] Furthermore, the data mapping module 14 in the intelligent fine-tuning system for slab track based on multi-source sensing fusion is also used for: performing a local topology search on the sub-network topology of the key node for any non-key node among the K non-key nodes to obtain an initial association set whose topological distance is less than a preset distance threshold; calculating the relationship strength set between the initial association set and the corresponding non-key node, wherein each relationship strength is obtained by weighted averaging of spatial distance attenuation factor, topological path dependence factor and historical data correlation factor; selecting the top M key nodes of the relationship strength set as the final association set, establishing a topological relationship description between each non-key node and the corresponding top M key nodes, and obtaining K topological relationships.

[0072] Furthermore, the data mapping module 14 in the intelligent fine-tuning system for slab track based on multi-source perception fusion is also used to: filter the top M key nodes of the relationship strength set as the final association set, wherein the value of M is dynamically adjusted in a positive correlation with the adjustment accuracy index of the fine-tuning control task.

[0073] Furthermore, the data mapping module 14 in the intelligent fine-tuning system for slab track based on multi-source sensing fusion is also used to: extract M multi-source sensing datasets for each non-critical node based on the corresponding topological relationship from the first group of multi-source sensing datasets according to K topological relationships; perform weighted fusion on the M multi-source sensing datasets to obtain the fused multi-source sensing mapping dataset for each non-critical node, and so on to obtain the K multi-source sensing mapping datasets corresponding to the K non-critical nodes.

[0074] Furthermore, the data fusion module 15 in the intelligent fine-tuning system for slab track based on multi-source sensing fusion is also used to: fuse the first set of multi-source sensing datasets and the second set of multi-source sensing datasets to obtain an initial fused multi-source sensing dataset; and when the difference between the fusion results of adjacent times of the fused multi-source sensing dataset exceeds a preset threshold, to perform smooth correction on the fused multi-source sensing dataset.

[0075] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The intelligent fine-tuning method and specific examples of slab track based on multi-source sensing fusion in the aforementioned embodiment 1 are also applicable to the intelligent fine-tuning system of slab track based on multi-source sensing fusion in this embodiment. Through the foregoing detailed description of the intelligent fine-tuning method of slab track based on multi-source sensing fusion, those skilled in the art can clearly understand the intelligent fine-tuning system of slab track based on multi-source sensing fusion in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0076] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0077] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for intelligent fine-tuning of slab track based on multi-source sensing fusion, characterized in that, include: Establish a fastener node network topology based on discrete fasteners for slab track; Node attribute analysis is performed on the fastener node network topology to obtain multiple node attribute features, including node location importance index, node coverage neighborhood index, and node structural vulnerability index. Based on the aforementioned multiple node attribute features, a key node sub-network topology including a subset of key nodes is obtained. The key node subset is obtained by iterative optimization analysis of the marginal revenue function of the fastener node network topology using a greedy algorithm. A first set of multi-source sensing datasets is collected from the sub-network topology of the key nodes. The first set of multi-source sensing datasets is mapped by analyzing the topological relationship between key nodes and non-key nodes to obtain a second set of multi-source sensing datasets. The first set of multi-source sensing datasets and the second set of multi-source sensing datasets are fused, and the resulting fused multi-source sensing dataset is sent to the track fine-tuning control terminal.

2. The intelligent fine-tuning method for slab track based on multi-source sensing fusion as described in claim 1, characterized in that, The first set of multi-source sensing datasets is mapped by analyzing the topological relationships between key nodes and non-key nodes. The method includes: Receive fine-tuning control tasks from the track fine-tuning control terminal; According to the track section fastener node set that needs to be adjusted according to the fine-tuning control task, determine K non-critical nodes in the fastener node set, and establish K topological relationships between the K non-critical nodes and the sub-network topology of the critical nodes. Based on the first set of multi-source sensing datasets and the K topological relationships, K multi-source sensing mapping datasets corresponding to the K non-critical nodes are obtained; The K multi-source sensing mapping datasets are output as the second set of multi-source sensing datasets.

3. The intelligent fine-tuning method for slab track based on multi-source sensing fusion as described in claim 2, characterized in that, The method for establishing K topological relationships between the K non-critical nodes and the sub-network topology of the critical nodes includes: For any one of the K non-critical nodes, a local topology search is performed on the sub-network topology of the critical node to obtain an initial association set whose topological distance is less than a preset distance threshold; Calculate the set of relationship strengths between the initial association set and the corresponding non-critical nodes, wherein each relationship strength is obtained by weighted averaging of spatial distance decay factor, topological path dependency factor and historical data correlation factor; The top M key nodes of the relationship strength set are selected as the final association set, and a topological relationship description between each non-key node and the corresponding top M key nodes is established to obtain K topological relationships.

4. The intelligent fine-tuning method for slab track based on multi-source sensing fusion as described in claim 3, characterized in that, The top M key nodes of the relationship strength set are selected as the final association set, wherein the value of M is dynamically adjusted in a positive correlation with the adjustment accuracy index of the fine-tuning control task.

5. The intelligent fine-tuning method for slab track based on multi-source sensing fusion as described in claim 3, characterized in that, Based on the first set of multi-source sensing datasets and the K topological relationships, K multi-source sensing mapping datasets corresponding to the K non-critical nodes are obtained through mapping, the method including: Based on K topological relationships, extract M multi-source sensing datasets for each non-critical node from the first set of multi-source sensing datasets based on the corresponding topological relationships; The M multi-source sensing datasets are weighted and fused to obtain the multi-source sensing mapping dataset after fusion for each non-critical node, and so on to obtain the K multi-source sensing mapping datasets corresponding to the K non-critical nodes.

6. The intelligent fine-tuning method for slab track based on multi-source sensing fusion as described in claim 1, characterized in that, The subset of key nodes is obtained through a greedy algorithm that iteratively optimizes the marginal revenue function of the fastener node network topology. The method includes: Initialize the empty set of critical nodes; By traversing each candidate node in the fastener node network topology, the marginal revenue index for adding the current candidate node is calculated using the marginal revenue function. The marginal revenue index includes a fixed-term revenue index based on the normalized weighted fitting of the multiple node attribute features. The first candidate node is selected according to the marginal revenue index and added to the empty set of key nodes for multiple rounds of iterative selection until the number of nodes in the key node subset reaches K value, and then the key node subset is output.

7. The intelligent fine-tuning method for slab track based on multi-source sensing fusion as described in claim 6, characterized in that, The marginal revenue metric also includes incentive item revenue metrics based on the coverage incentive factors; The coverage incentive factor is used to perform contribution analysis of uncovered nodes on the current candidate nodes, including evaluating the candidate nodes' ability to supplement the coverage of nodes that are not yet effectively covered by the subset of key nodes.

8. The intelligent fine-tuning method for slab track based on multi-source sensing fusion as described in claim 7, characterized in that, After selecting the first candidate node according to the marginal revenue index and adding it to the empty set of key nodes, the method further includes: The fastener node network topology is updated based on the first candidate node added, including covered status labels and uncovered status labels. The cumulative coverage of the fastener node network topology is updated cumulatively during multiple iterations.

9. The intelligent fine-tuning method for slab track based on multi-source sensing fusion as described in claim 1, characterized in that, The method for fusing the first set of multi-source sensing datasets and the second set of multi-source sensing datasets includes: The first set of multi-source sensing datasets and the second set of multi-source sensing datasets are fused to obtain an initial fused multi-source sensing dataset. When the difference between adjacent time-series fusion results of the fused multi-source sensing dataset exceeds a preset threshold, the fused multi-source sensing dataset is smoothed and corrected.

10. A smart fine-tuning system for slab track based on multi-source sensing fusion, characterized in that, The steps for implementing the intelligent fine-tuning method for slab track based on multi-source sensing fusion as described in any one of claims 1 to 9, wherein the intelligent fine-tuning system for slab track based on multi-source sensing fusion comprises: The fastener node network topology establishment module is used to establish the fastener node network topology based on discrete fasteners for slab track; The node attribute analysis module is used to perform node attribute analysis on the fastener node network topology and obtain multiple node attribute features, including node location importance index, node coverage neighborhood index, and node structural vulnerability index. The key node sub-network topology identification module is used to obtain the key node sub-network topology, including a key node subset, based on the multiple node attribute features. The key node subset is obtained by iterative optimization analysis of the marginal benefit function of the fastener node network topology using a greedy algorithm. The data mapping module is used to collect the first set of multi-source sensing datasets of the key node sub-network topology, and to map the first set of multi-source sensing datasets by analyzing the topological relationship between key nodes and non-key nodes to obtain the second set of multi-source sensing datasets. The data fusion module is used to fuse the first set of multi-source sensing datasets and the second set of multi-source sensing datasets, and send the resulting fused multi-source sensing dataset to the track fine-tuning control terminal.

Images