A railway station interlocking table external route generation method based on template segment splicing

Abstract

The application provides a railway station interlocking table off-route generation method based on template fragment splicing, and relates to the technical field of intelligent railway stations. The method is used for obtaining an existing route list, a constraint set and a key section set of a railway station, using a fragment extraction and attribute calculation method to obtain a template fragment library and a multi-level index. Based on the template fragment library, a multi-level index is used to recall a candidate fragment set, and based on the key section set and the constraint set, a matching core value of each fragment in the candidate fragment set is calculated. According to the matching core value, the fragments in the candidate fragment set are aligned and spliced to obtain a complete candidate route. Based on the constraint set, the complete candidate route is verified and confidence is calculated to obtain a table off-route list and a traceability evidence chain. The application solves the problems of low recognition accuracy, large enumeration space, lack of unified confidence sorting, high cost of manual template maintenance and lack of traceable evidence chain in the prior art.

Description

Technical Field

[0001] This invention relates to the field of intelligent railway station technology, and in particular to a method for generating external routes for railway station interlocking based on template fragment splicing. Background Technology

[0002] The station's computer interlocking system is based on route data. The compilation and maintenance of traditional interlocking tables and route lists are extremely labor-intensive. While various automated technical approaches have emerged in engineering practice, such as template-based interlocking table generation, generation based on signal plan diagrams or element libraries, blueprint recognition and input of interlocking tables, calibration and verification of interlocking tables, and automatic testing and execution of route lists, these approaches primarily focus on generating or verifying known fields or logical items within the interlocking tables. They are not optimal solutions for outputting only a list of candidate routes outside the interlocking tables at the track segment sequence level.

[0003] Currently, to address potential omissions in railway station interlocking route data during design changes, version iterations, and manual maintenance, existing technologies mainly have the following solutions and their limitations: The first category is machine-generated off-table route schemes based on platform plan diagrams. The typical approach involves importing and binarizing the platform plan, identifying elements such as signals and switches, constructing a binary tree search structure based on track connections, enumerating routes, and comparing them with the interlocking table. While this scheme can quickly generate candidate routes when the drawing quality is high, it is extremely sensitive to drawing quality and element recognition accuracy. Furthermore, the enumeration space grows exponentially with the size of the station, making it difficult to provide a unified and interpretable probability confidence ranking, and easily generating a large number of invalid candidates.

[0004] The second type is an automatic interlocking table generation solution based on template technology. This solution designs a series of interlocking table generation templates and database matching tables, and populates the templates by retrieving equipment information. Its main goal is to output interlocking table fields that conform to a specific format. However, it requires maintaining a complex template library and mapping rules, resulting in high adaptation costs and difficulty in discovering potential paths outside the coverage of existing templates.

[0005] The third category is manual analogy and rule-based retrieval methods. This method relies heavily on human experience, using analogies to existing routes for splicing and verification. Its disadvantages include low efficiency and difficulty in scaling; it can only perform simple searches and cannot generate new routes; it lacks a scoring mechanism deeply bound to complex railway constraints (such as opposing routes and directional protection), resulting in a high false alarm rate; and it lacks a traceable chain of evidence for generation, making manual review extremely costly and prone to missed detections.

[0006] In summary, existing technologies, with only the existing route list as a data source, are insufficient to automatically and efficiently discover and generate high-quality off-table candidate routes while satisfying complex railway constraints. Summary of the Invention

[0007] To address the aforementioned shortcomings in existing technologies, this invention provides a method for generating railway station interlocking routes based on template fragment splicing. This method solves the problems of low accuracy due to reliance on drawings, large enumeration space, lack of unified confidence ranking, high cost of manual template maintenance, and lack of traceable evidence chains in existing technologies.

[0008] To achieve the aforementioned objectives, the technical solution adopted by this invention is: a method for generating railway station interlocking external routes based on template fragment splicing, comprising: S1: Obtain the list of existing routes, constraint set, and key section set for railway stations; S2: Based on the existing route list and key segment set, use segment extraction and attribute calculation methods to obtain a template segment library and multi-level index; S3: Based on the template fragment library, a multi-level index is used to recall the candidate fragment set, and based on the key segment set and constraint set, the matching kernel value of each fragment in the candidate fragment set is calculated; S4: Align and splice the fragments in the candidate fragment set according to the matching kernel value to generate a candidate splicing chain. Then, based on the constraint set and the candidate splicing chain, fill in the gaps caused by splicing to obtain a complete candidate path. S5: Based on the constraint set, the complete candidate routes are verified and confidence is calculated to obtain the list of off-table routes and the traceability evidence chain, thus completing the generation of off-table routes for railway station interlocking.

[0009] Furthermore, S2 specifically includes: Based on the existing route list and key segment set, extract continuous k-segment sequences and key segments containing key segments; Based on k-segment sequences and key segments, the directional labels, key coverage sets, frequency, and weight attributes are calculated to obtain segment labels; Using fragment tags, the signature of the k-fragment sequence is calculated from the k-fragment sequence and key fragments; Based on the signatures of the k-segment sequences, deduplication and aggregation are performed to establish the template segment library; Based on the template fragment library, the endpoint index, anchor point alignment index, and inverted index are constructed using the endpoints, key anchor points, and fragment features of the k-fragment sequence, respectively, to obtain the multi-level index.

[0010] Furthermore, the expression for the fragment weight attribute is: ; in, Indicates the first The weights of k-segments, Indicates the first k-segments, This indicates the frequency of the segment in the existing route list. This represents the weighting coefficient for key coverage levels. This indicates the number of key segments contained in the fragment.

[0011] Further, S3 includes: Based on the template fragment library, a multi-level index is used for recall to obtain a set of candidate fragments; Based on the set of key segments, the weighted coverage of the candidate segments on the target sequence is calculated. Determine the directional consistency between the candidate segment and the target sequence to obtain the directional consistency factor; Based on the constraint set, the number of consecutive triggering rules of candidate fragments is counted, and a nonlinear penalty term is obtained based on the number of consecutive triggering rules. The matching kernel value is calculated based on the weighted coverage, the direction consistency factor, and the nonlinear penalty term.

[0012] Furthermore, the expression for the weighted coverage is: ; in, Indicates candidate fragments For the target sequence Weighted coverage Indicates a sequence The set of k-segments Representing fragments The weights; The expression for the nonlinear penalty term is: ; ; in, This represents the total nonlinear penalty. Indicates constraint rules. Represents a set of constraints. Indicates the rule penalty weight. This indicates the sensitivity to chain-triggered penalties. Representing constraint rules The nonlinear penalty function, Indicates the maximum number of consecutive triggers. This represents the function that takes the maximum value. Indicates the threshold for exemption from punishment; The expression for matching the kernel value is: ; ; in, Indicates matching kernel value, This represents the Sigmoid function. Represents a fixed logarithmic function. Indicates the bias term. Represents the coverage coefficient. This represents the directional consistency coefficient. Indicates the direction consistency factor. Indicates the key coverage factor. This indicates the number of key segments covered by fragment p. This represents the penalty inhibition coefficient. It indicates a tiny amount to prevent dispersion.

[0013] Furthermore, in step S4, the segments in the candidate segment set are aligned and spliced ​​according to the matching kernel value, specifically including: The suffix-prefix overlap length between adjacent segments is calculated based on the candidate segment set. By utilizing the suffix-prefix overlap length, a splicing operator is constructed to generate the spliced ​​segment sequence; Based on the constraint set, the chain trigger count of the trigger rule at the gap in the spliced ​​segment sequence is statistically analyzed; The cost of the gap caused by splicing is calculated using the gap length, direction conflict indication, and the chain trigger count. With the optimization objective of maximizing the sum of fragment matching kernel values ​​and minimizing gap costs, fragments in the candidate fragment set are aligned and spliced ​​to search and generate candidate splicing chains.

[0014] Furthermore, the expression for the gap cost is: ; in, Representing fragments and The cost of the gap between them Indicates the length of the gap. Both represent length coefficients. Indicates the exponential amplification factor. Indicates the chain trigger count, Indicates the threshold for exemption from penalty. Indicates the directional conflict coefficient. Indicates a direction conflict. Indicates the critical coverage missing coefficient. Indicates the degree of critical coverage deficiency.

[0015] Furthermore, in step S4, based on the constraint set and candidate splicing chain, the gaps generated by splicing are filled to obtain complete candidate paths, specifically including: Based on the constraint set, check whether there are gaps between adjacent segments in the candidate splicing chain. If there are gaps, prioritize searching the template segment library for the complete segment that satisfies the endpoint anchor compatibility constraint and has the minimum gap cost. If no complete fragment that meets the conditions is found in the template fragment library, then a restricted search is performed on the allowed segment connection set based on the forbidden segments and forbidden connection rules in the constraint set to generate a complete sequence; The complete fragment, the complete sequence, and the candidate splicing chain are connected to obtain the complete candidate path.

[0016] Further, S5 includes: The complete candidate path is validated based on the constraint set, and paths that do not meet the constraints are eliminated to obtain the validated complete candidate path. Based on the verified complete candidate paths, feature extraction is performed to obtain the feature vectors of the verified complete candidate paths; Based on the feature vector, the fusion confidence is calculated using the preset calibration fusion calculation rules to obtain the off-table route list and the traceability evidence chain, thus completing the generation of off-table routes for railway station interlocking.

[0017] Furthermore, the expression for the fusion confidence is: ; in, Indicate candidate paths fusion confidence, This represents the Sigmoid function. Indicates the bias parameter. Represents a learnable parameter vector. Indicate candidate paths The feature vector includes kernel value accumulation, critical coverage strength, consistency, chain nonlinear penalty accumulation, and gap cost statistics.

[0018] The beneficial effects of this invention are as follows: This invention provides a method for generating off-site routes for railway station interlocking tables based on template fragment splicing. It only requires obtaining the existing route list, constraint set, and key section set of the railway station for implementation. Compared to existing technologies that rely on high-quality drawing recognition or complex database templates, this invention eliminates the need for blueprint recognition or maintaining cumbersome template matching tables, making it more suitable for scenarios with missing existing data or incomplete drawings, thus lowering the implementation threshold.

[0019] By constructing a template fragment library and multi-level index through fragment extraction and attribute calculation, this invention solidifies the structural experience in existing routes into searchable template fragments. Utilizing the multi-level index to recall candidate fragments enables rapid location of potentially usable route fragments, avoiding the combinatorial explosion problem caused by blind full-graph search and significantly improving generation efficiency.

[0020] In calculating the matching kernel value and generating candidate splicing chains, this invention fully considers the influence of the key segment set and the constraint set. By introducing the matching kernel value to filter candidate segments and filling gaps caused by splicing, it ensures that the generated complete candidate paths have high rationality in terms of topology and business rules, effectively reducing the generation of invalid or illegal paths.

[0021] While generating the off-table route list, a traceability evidence chain was also output. This evidence chain records the basis for route generation, providing engineers with clear review clues and ensuring that the generated routes are not black-box results. This significantly reduces the difficulty and workload of manual review and improves the system's credibility.

[0022] By calculating the feature vectors of the verified complete candidate paths and obtaining the fusion confidence score, this invention can quantify and rank the generated off-table paths. This allows engineers to prioritize candidate paths with high confidence scores, further improving review efficiency and reducing the rate of missed detections. Attached Figure Description

[0023] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein: Figure 1 This is an exemplary flowchart illustrating a method for generating railway station interlocking external routes based on template fragment splicing, according to some embodiments of this specification. Detailed Implementation

[0024] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0025] Example Figure 1 This is an exemplary flowchart illustrating a method for generating off-table routes for railway station interlocking based on template fragment splicing, according to some embodiments of this specification. Figure 1As shown, the process includes the following steps. In some embodiments, the process may be executed by a processor.

[0026] S1: Obtain the list of existing routes, constraint set, and key section set for railway stations.

[0027] Existing route list It is a set of existing, legal route data describing railway stations. For example, an existing route list can be represented as... Each of the routes It contains an ordered sequence of orbital segments. It can also include job type tags. (Such as train routes, shunting routes, etc.) to represent different operational attributes. Here, M represents the total number of existing routes; v_1 represents the first track segment in the route sequence, v_2 represents the second track segment, v_{L_m} represents the L_m-th track segment; L_m represents the total number of track segments contained in route r_m; v_k represents the k-th track segment in the route sequence; V represents the set of all track segments in the station topology; l represents the numerical code of a single operational type; and L represents the total number of operational types allowed by the station.

[0028] In some embodiments, the processor can obtain an existing route list by importing external data files, reading a database, or receiving user input, and perform preprocessing such as numbering standardization, deduplication, and consistency checks.

[0029] In some embodiments, the processor may perform preprocessing operations: Numbering and format standardization: unify all input track segment identifiers into a standard format, such as mapping different spellings such as "1DG" and "II G" to a unique segment ID.

[0030] Deduplication and consistency checks: Eliminate duplicate records in the existing route list and detect whether there are broken edges in the route sequence, i.e., adjacent segments are physically unreachable or isolated nodes.

[0031] Illegal segment handling: Mark or remove illegal segments in the segment sequence that do not belong to the current station topology.

[0032] Log generation: Generate data quality logs containing the above processing details for subsequent auditing and traceability.

[0033] A constraint set is a machine-executable set of rules used to restrict path generation and concatenation. For example, constraint sets... This can include restricted sections, restricted connections (such as hostile routes), directional rules (such as one-way traffic restrictions), route length limits (longest / shortest), whether there are loop restrictions, critical section coverage requirements, and endpoint sets, etc.

[0034] In some embodiments, the processor can obtain the constraint set by reading a preset rule configuration file, parsing manually input logical expressions, or mining rules from historical logs. ; Where e_1 and e_2 represent directed connecting edges in the track topology graph; dir(e) represents the function that determines the direction of the directed edges; V_start point represents the set of segments that serve as the starting point of the route, and V_end point represents the set of segments that serve as the ending point of the route.

[0035] The critical segment set is a collection of track segments within a station that have significant topological or operational value. For example, the critical segment set... It can include turnout sections, sections where signals are located, or the core section of throat areas, in the form of... ; where is_key(v_i) represents a Boolean function to determine whether segment v_i is a key segment. If it is true, it means that the segment is a key segment.

[0036] In some embodiments, the processor can obtain a set of key segments by identifying key elements in the site map, reading manually marked configuration files, or automatically identifying them based on the connectivity of the segments.

[0037] S2: Based on the existing route list and key segment set, use segment extraction and attribute calculation methods to obtain a template segment library and multi-level index.

[0038] A template fragment library is a collection of route fragments extracted from and processed from existing routes. For example, a template fragment library... It can contain continuous subsequences (k-segments) after deduplication and aggregation, and key sub-paths. Each segment is accompanied by statistical attributes such as frequency, weight, and coverage.

[0039] In some embodiments, the processor can extract k-fragments by sliding window extraction of each path in the existing path list, calculate fragment labels, and construct a template fragment library by deduplicating and aggregating fragments of the same sequence using signatures.

[0040] In some embodiments, the processor can process each existing path The sliding window method is used to extract a continuous subsequence of length k, i.e., a k-segment. At the same time, key segments containing at least one key section are selected.

[0041] Multilevel indexes are data structures used for fast retrieval of template fragment libraries. For example, a multilevel index may include an endpoint index (based on the fragment start / end point key), an anchor alignment index (based on the key anchor point), and an inverted index (based on the k-fragment key), each mapped to a corresponding fragment set.

[0042] In some embodiments, the processor can obtain a multi-level index by traversing each fragment in the template fragment library, extracting its endpoints, contained anchors, and subsequence features, and registering its ID to the corresponding hash table or search tree structure.

[0043] In some embodiments, attribute calculation includes: direction label: calculating the direction of the fragment. Preferably, the upward direction is denoted as +1, and the downward direction is denoted as -1. Key Coverage: Calculate the set of key segments covered by the segment. Frequency and weight: The frequency of a statistical segment appearing in all existing routes. Segment weights are calculated based on frequency and key coverage. .

[0044] In some embodiments, deduplication aggregation includes: calculating fragment signatures, aggregating fragments with the same signature, and constructing a template fragment library. .

[0045] In some embodiments, multi-level index construction includes: Endpoint index: creating an index that points to the corresponding set of segments, using the start and end points of the segments as keys. Anchor point alignment index: Creates an index that uses key anchor points (key segments) as keys, pointing to segments containing those anchor points and their positions within those segments. Inverted index: Creates an index that uses the k-segment signature as the key and points to a set of long segments containing that subsequence. .

[0046] In some embodiments, the processor may extract continuous k-segment sequences and key segments containing key segments based on the existing route list and key segment set; calculate the direction label, key coverage set, frequency, and segment weight attributes based on the k-segment sequence and key segments to obtain segment labels; calculate the signature of the k-segment sequence using the segment labels; perform deduplication and aggregation based on the signature of the k-segment sequence to establish the template segment library; and construct the endpoint index, anchor alignment index, and inverted index based on the endpoints, key anchors, and segment features of the k-segment sequence to obtain the multi-level index.

[0047] A k-segment sequence is a continuous sequence of orbital segments of length k extracted from the route. For example, for a route... Its k-segment sequence can be represented as ; where j represents the starting segment index position of the k-segment in the original existing route sequence.

[0048] In some embodiments, the processor can perform a sliding window operation on an existing path sequence to obtain a k-segment sequence by setting a window size k.

[0049] A key segment is a sequence of orbital segments that contains at least one key section. For example, a key segment is one that satisfies... A fragment.

[0050] In some embodiments, the processor can filter and obtain key segments by checking whether the extracted k-segment sequence or a specific subpath contains elements from the key segment set.

[0051] Fragment labels are feature data that describe the attributes of template fragments. For example, fragment labels may include direction labels (such as up / down), key coverage sets (key segments contained in the fragment), frequency (the number of times it appears in existing routes), and weight attributes (numerical values ​​calculated based on frequency and coverage).

[0052] In some embodiments, the processor can perform statistical analysis on existing paths to calculate the directional attributes of the fragment, its intersection with the set of key segments, and the frequency of occurrence, and obtain the fragment label by calculating the weight according to the formula.

[0053] In some embodiments, the expression for a fragment tag is: ; in, This represents the key coverage weighting coefficient, used to adjust the weighting of key segments. For example, it can be taken as... ; Representing fragments Frequency of appearance in the existing route list; Representing fragments The number of key segments included. This formula reflects that high-frequency segments that cover key segments have higher reuse value.

[0054] In some embodiments, the expression for the fragment weight attribute is: ; in, Indicates the first The weights of k-segments, Indicates the first k-segments, This indicates the frequency of the segment in the existing route list. This represents the weighting coefficient for key coverage levels. This indicates the number of key segments contained in the fragment.

[0055] The signature of a k-segment sequence is a hash value or feature code used to uniquely identify the content of the segment sequence. For example, the signature can be represented as... ,in, Representing fragments Corresponding orbital segment sequence data; Representing fragments The number of track segments included.

[0056] In some embodiments, the processor can obtain the signature of the k-fragment sequence by concatenating the segment numbers of the k-fragment sequence with strings or by calculating a specific hash algorithm, which is used for fragment deduplication and index construction.

[0057] S3: Based on the template fragment library, a multi-level index is used to recall the candidate fragment set, and based on the key segment set and constraint set, the matching kernel value of each fragment in the candidate fragment set is calculated.

[0058] The candidate fragment set is a subset of fragments with potential matching value retrieved from the template fragment library based on target conditions. For example, the candidate fragment set... This can include recalls via endpoint matching. Recall via anchor point alignment and retrieved through inverted index The union of .

[0059] In some embodiments, the processor can retrieve a set of candidate segments by using multi-level indexes to perform query recall based on the start point, end point, or intermediate anchor point conditions of the route to be generated.

[0060] In some embodiments, the processor constructs a throat prior structure and a recall mask. High-frequency sub-paths to the throat region are extracted from existing routes, and the region is segmented. And construct a set of allowed regional migration sequences. For any fragment Calculate its regional signature ,like and Inconsistency or triggering constraints Then set the recall mask. Otherwise, it is 1, thus achieving pre-pruning. Secondly, multi-level indexing is used for retrieval. The candidate fragment set is calculated, and each fragment is computed. For the target sequence Matching kernel value Among them, zone( ) represents the partitioning function that maps the track segment to the corresponding throat region; Ξ represents the set of allowed region migration sequences based on existing routes; Z(p) represents the calculated region signature feature of segment p; Mask_H represents the recall mask variable used to control whether the segment enters subsequent calculations.

[0061] The matching score is a comprehensive rating that measures the degree to which a candidate fragment matches the target requirement and its safety. For example, the matching score... It is a value between 0 and 1, which combines coverage, directional consistency, critical segment coverage, and rule penalties.

[0062] In some embodiments, the processor can calculate the matching kernel value based on fragment attributes and constraint check results using a formula.

[0063] In some embodiments, the processor can use a multi-level index to recall candidate segments based on a template segment library to obtain a set of candidate segments; calculate the weighted coverage of the candidate segments to the target sequence based on the set of key segments; determine the directional consistency between the candidate segments and the target sequence to obtain a directional consistency factor; count the consecutive number of times the candidate segments trigger rules based on the constraint set, and calculate a nonlinear penalty term based on the consecutive number of times; and calculate the matching kernel value based on the weighted coverage, the directional consistency factor, and the nonlinear penalty term.

[0064] Weighted coverage is an indicator that reflects the ability of a candidate segment to cover a target sequence or potential path. For example, weighted coverage... It can be the proportion of fragment weights in the target-related set, in the form of... .

[0065] In some embodiments, the processor can obtain weighted coverage by calculating the ratio of the sum of the weights of the subsequences contained in the candidate fragment to the total weight.

[0066] The direction consistency factor is a numerical value that characterizes whether the direction of a candidate segment conflicts with the direction of the target path. For example, the direction consistency factor... Desirable These represent directions that are aligned, directions that are unknown, and directions that conflict, respectively.

[0067] In some embodiments, the processor can obtain a direction consistency factor by comparing the direction label of the candidate segment with the running direction (such as up / down) set by the target route.

[0068] A non-linear penalty term is a penalty value that is exponentially amplified for violations of soft constraints. For example, a non-linear penalty term... It can be used to suppress high-risk fragments.

[0069] In some embodiments, the processor can calculate a nonlinear penalty term by substituting the number of consecutive times that various constraint rules (such as prohibition and conflict) are triggered by statistical fragments into the exponential penalty formula.

[0070] In some embodiments, the expression for the weighted coverage is: ; in, Indicates candidate fragments For the target sequence Weighted coverage Indicates a sequence The set of k-segments Representing fragments The weights; The expression for the nonlinear penalty term is: ; ; in, This represents the total nonlinear penalty. Indicates constraint rules. Represents a set of constraints. Indicates the rule penalty weight. This indicates the sensitivity to chain-triggered penalties. Representing constraint rules The nonlinear penalty function, Indicates the maximum number of consecutive triggers. This represents the function that takes the maximum value. Indicates the threshold for exemption from punishment; The expression for matching the kernel value is: ; ; in, Indicates matching kernel value, This represents the Sigmoid function. Represents a fixed logarithmic function. Indicates the bias term. Represents the coverage coefficient. This represents the directional consistency coefficient. Indicates the direction consistency factor. Indicates the key coverage factor. This indicates the number of key segments covered by fragment p. This represents the penalty inhibition coefficient. It indicates a tiny amount to prevent dispersion.

[0071] S4: Align and splice the fragments in the candidate fragment set according to the matching kernel value to generate a candidate splicing chain. Then, based on the constraint set and the candidate splicing chain, fill in the gaps caused by splicing to obtain a complete candidate path.

[0072] A candidate splicing chain is a preliminary path structure that is not yet fully completed, formed by aligning and splicing multiple template fragments. For example, a candidate splicing chain can be represented as an ordered sequence of fragments. And their connections.

[0073] In some embodiments, the processor can construct a candidate splicing chain by searching for the optimal splicing path in the candidate fragment set, maximizing the sum of fragment matching kernel values ​​and minimizing the gap cost.

[0074] In some embodiments, the processor can calculate the suffix-prefix overlap length between adjacent segments based on the candidate segment set; construct a splicing operator using the suffix-prefix overlap length to generate a spliced ​​segment sequence; count the chain triggering count of the spliced ​​segment sequence at the gap based on the constraint set; calculate the gap cost generated by splicing using the gap length, direction conflict indication, and the chain triggering count; and align and splice the segments in the candidate segment set with the optimization objective of maximizing the sum of segment matching kernel values ​​and minimizing the gap cost, thereby searching for and generating candidate splicing chains.

[0075] The suffix-prefix overlap length refers to the maximum length of the sequence of segments that are identical at the junction of two adjacent segments. For example, for a segment... and The overlap length is satisfied end Each section and The beginning The largest number of segments with the same maximum value.

[0076] In some embodiments, the processor can obtain the suffix-prefix overlap length by comparing the suffix of the previous segment with the prefix of the next segment using string matching or sequence alignment algorithms.

[0077] The concatenated sequence is a new list of segments formed by merging two fragments. For example, the concatenated sequence can be represented as follows: That is, retaining the entirety of the previous segment and the content of the next segment after removing the overlapping parts; where l represents the overlap length of the same section between the suffix of the previous segment f_i and the prefix of the next segment f_j.

[0078] In some embodiments, the processor can obtain a spliced ​​segment sequence by executing a splicing operator to perform truncation and concatenation operations on the fragment sequence based on the overlap length.

[0079] Chain trigger count refers to the cumulative number of times a certain type of constraint rule is triggered consecutively during route generation or assembly. For example, chain trigger count... It can represent the number of times a person passes through a restricted area or triggers a hostile route rule.

[0080] In some embodiments, the processor can obtain a chain trigger count by detecting in real time whether the current segment or connection violates the rules in the constraint set and accumulating the count as it advances along the path sequence.

[0081] Gap cost is a numerical value that quantifies the risk of spatial discontinuity or rule violation at the junction of segments. For example, gap cost It can include linear and square terms of gap length, exponential penalty terms for chain-triggered events, and cost of directional conflict.

[0082] In some embodiments, the processor can calculate the gap cost by calculating the topological distance between segments, i.e., the gap length, combined with the chain trigger count and directional consistency, and then substituting it into the cost function formula.

[0083] In some embodiments, the expression for the gap cost is: ; in, Representing fragments and The cost of the gap between them Indicates the length of the gap. Both represent length coefficients. Indicates the exponential amplification factor. Indicates the chain trigger count, Indicates the threshold for exemption from penalty. Indicates the directional conflict coefficient. Indicates a direction conflict. Indicates the critical coverage missing coefficient. Indicates the degree of critical coverage deficiency.

[0084] A complete candidate route is a physically continuous sequence of orbital segments from the start point to the end point after gap filling. For example, a complete candidate route... It is a complete path without breaks, consisting of spliced ​​fragments and complete fragments / sequences.

[0085] In some embodiments, the processor can obtain a complete candidate path by filling each gap in the candidate splicing chain using a template library or a restricted search completion strategy.

[0086] In some embodiments, the processor can check whether there are gaps between adjacent segments in the candidate splicing chain based on the constraint set. If there are gaps, it will first search the template segment library for the complete segment that satisfies the endpoint anchor compatibility constraint and has the minimum gap cost. If no complete segment that meets the conditions is found in the template segment library, it will perform a restricted search on the allowed segment connection set based on the forbidden segments and forbidden connection rules in the constraint set to generate a complete sequence. The complete segment, the complete sequence, and the candidate splicing chain will be connected to obtain the complete candidate path.

[0087] Completion fragments are ready-made template fragments used to fill gaps between adjacent fragments in a candidate splicing chain. For example, completion fragments... It is a fragment in the template fragment library whose two ends are aligned with the fragment anchor points on both sides of the gap.

[0088] In some embodiments, the processor can retrieve complete fragments from the template fragment library by querying the endpoint index or anchor index to find fragments that satisfy the endpoint compatibility constraint.

[0089] A completion sequence is a sequence of segments that connect gaps, generated by a search algorithm when no existing completion fragment can be found. For example, a completion sequence is a list of constrained track segments that connect the start and end points of a gap.

[0090] In some embodiments, the processor can obtain the completion sequence by performing a limited-length depth-first search (DFS) or breadth-first search (BFS) on the station topology map based on a set of constraints (such as prohibition rules).

[0091] In some embodiments, the processor can process candidate fragments. Calculate the longest suffix-prefix overlap length. And generate the spliced ​​sequence. Calculate the gap cost at the splice point. With the goal of generating candidate splicing chains, search for and generate them. This represents the weighting coefficient for the gap cost penalty; a value of 0.8 is recommended. For gaps in the splicing chain, priority should be given to searching the template fragment library for fragments that satisfy endpoint compatibility and... Minimal completion fragment If the sequence cannot be found, a restricted search (BFS / DFS) is performed on the allowed set of segment connections to generate a completion sequence. Chaining counts are also included during the search process to suppress high-risk completions.

[0092] S5: Based on the constraint set, the complete candidate routes are verified and confidence is calculated to obtain the list of off-table routes and the traceability evidence chain, thus completing the generation of off-table routes for railway station interlocking.

[0093] The off-table route list is the final set of routes that are not listed in the existing interlocking tables but are potentially available. For example, the off-table route list can be a list containing multiple complete candidate routes along with their confidence scores and source information.

[0094] In some embodiments, the processor can obtain an off-table list of routes by performing global verification and deduplication on the generated complete candidate routes, sorting them by fusion confidence, and then extracting the Top-N routes.

[0095] A source chain of evidence is a collection of detailed information recording the basis for the generation of each off-table path. For example, a source chain of evidence may include the sequence of fragment IDs that constitute the path, the overlap length at the splice, the source of the gap completion (library retrieval or search), and a summary of the penalty rules that triggered it.

[0096] In some embodiments, the processor records operation logs and decision parameters at each step of path splicing, completion, and verification, and finally summarizes them to obtain the traceability evidence chain.

[0097] In some embodiments, the processor can perform legality verification on the complete candidate routes based on the constraint set, eliminate routes that do not meet the constraints, and obtain complete candidate routes that pass the verification; based on the complete candidate routes that pass the verification, perform feature extraction to obtain feature vectors of the complete candidate routes that pass the verification; based on the feature vectors, calculate the fusion confidence using preset calibration fusion calculation rules to obtain the off-table route list and traceability evidence chain, and complete the generation of off-table routes for railway station interlocking.

[0098] The feature vector of a valid candidate path is a numerical feature representation used to calculate the path confidence. For example, the feature vector... It can include These correspond to the cumulative kernel value, key coverage strength, consistency score, cumulative nonlinear penalty, and gap cost statistics, respectively; among them, For the accumulation of core values, For key coverage strength, For consistency, For the cumulative chain of nonlinear penalties, This is a statistic on the gap cost.

[0099] In some embodiments, the processor can obtain a feature vector by extracting various statistical indicators and scoring components from the path that has passed the global constraint verification.

[0100] The fusion confidence score is the model's final score on the reliability of the generated off-table paths. For example, the fusion confidence score... It is a probability value between 0 and 1, reflecting the likelihood that the route will become a valid interlocking route.

[0101] In some embodiments, the processor can obtain the fusion confidence by inputting the feature vector into the calibration fusion model, performing a weighted summation, and processing it through the Sigmoid activation function.

[0102] In some embodiments, the expression for the fused confidence score is: ; in, Indicate candidate paths fusion confidence, This represents the Sigmoid function. Indicates the bias parameter. Represents a learnable parameter vector. Indicate candidate paths The feature vector includes kernel value accumulation, critical coverage strength, consistency, chain nonlinear penalty accumulation, and gap cost statistics.

[0103] In some embodiments, the processor can apply a set of constraints to the complete candidate routes. Perform verification. If verification fails, output the failure reason code (e.g., F01 prohibition trigger, F02 direction conflict, etc.) and a list of triggering rules, forming part of the traceability evidence chain. Extract feature vectors from paths that pass verification and calculate the fusion confidence score. The output lists the top-N out-of-table routes, along with a chain of evidence including fragment ID sequences, overlap lengths, completion sources, and penalty summaries. Specifically, the processor uses verification tags... For monitoring signals, the following is adopted: The update process adapts the weights to the changes in the distribution of station data, enabling adaptive convergence. (By...) Sort and output a Top-N candidate list, and output the source chain, including but not limited to: fragment ID sequence, overlap length, completion source and penalty summary, etc.

[0104] In some embodiments, the processor can perform a review and reflow update step to achieve adaptive evolution of the system, including: Negative sample parsing: receiving negative samples from manual review and extracting their source chain and violation example structure; Pattern fingerprint generation: generating a pattern fingerprint from the violation example structure. Where AnchorSeq represents the anchor sequence features in the violation example structure, JoinType represents the splicing type features between segments, ZoneSig represents the signature features of the violation region, RuleSet represents the triggered constraint rule features, support(m) represents the support function of the frequency of pattern m in the review record, and confidence(m) represents the confidence function of the probability that pattern m leads to a violation. Gated writing and pruning: calculating pattern support. and confidence level When both exceed the preset threshold, the pattern fingerprint is written to the disabled combination library B. In subsequent recall and assembly, if a structure matching the fingerprint in B is encountered, it is pruned directly. Parameter online update: Based on the review results, the rule penalty weights are updated online.

Claims

1. A method for generating off-site routes for railway station interlocking based on template fragment splicing, characterized in that, include: S1: Obtain the list of existing routes, constraint set, and key section set for railway stations; S2: Based on the existing route list and key segment set, use segment extraction and attribute calculation methods to obtain a template segment library and multi-level index; S3: Based on the template fragment library, a multi-level index is used to recall the candidate fragment set, and based on the key segment set and constraint set, the matching kernel value of each fragment in the candidate fragment set is calculated; S4: Align and splice the fragments in the candidate fragment set according to the matching kernel value to generate a candidate splicing chain. Then, based on the constraint set and the candidate splicing chain, fill in the gaps caused by splicing to obtain a complete candidate path. In step S4, the fragments in the candidate fragment set are aligned and spliced ​​according to the matching kernel value, specifically including: The suffix-prefix overlap length between adjacent segments is calculated based on the candidate segment set. By utilizing the suffix-prefix overlap length, a splicing operator is constructed to generate the spliced ​​segment sequence; Based on the constraint set, the chain trigger count of the trigger rule at the gap in the spliced ​​segment sequence is statistically analyzed; The cost of the gap caused by splicing is calculated using the gap length, direction conflict indication, and the chain trigger count. With the optimization objective of maximizing the sum of fragment matching kernel values ​​and minimizing gap cost, fragments in the candidate fragment set are aligned and spliced ​​to search and generate candidate splicing chains. In step S4, based on the constraint set and candidate splicing chain, the gaps generated by splicing are filled to obtain complete candidate paths, specifically including: Based on the constraint set, check whether there are gaps between adjacent segments in the candidate splicing chain. If there are gaps, prioritize searching the template segment library for the complete segment that satisfies the endpoint anchor compatibility constraint and has the minimum gap cost. If no complete fragment that meets the conditions is found in the template fragment library, then a restricted search is performed on the allowed segment connection set based on the forbidden segments and forbidden connection rules in the constraint set to generate a complete sequence; The complete fragment, the complete sequence, and the candidate splicing chain are connected to obtain the complete candidate path; S5: Based on the constraint set, the complete candidate routes are verified and confidence is calculated to obtain the list of off-table routes and the traceability evidence chain, thus completing the generation of off-table routes for railway station interlocking.

2. The method for generating railway station interlocking external routes based on template fragment splicing according to claim 1, characterized in that, S2 specifically includes: Based on the existing route list and key segment set, extract continuous k-segment sequences and key segments containing key segments; Based on k-segment sequences and key segments, the direction label, key coverage set, frequency, and segment weight attributes are calculated to obtain the segment label; Using fragment tags, the signature of the k-fragment sequence is calculated from the k-fragment sequence and key fragments; Based on the signatures of the k-segment sequences, deduplication and aggregation are performed to establish the template segment library; Based on the template fragment library, the endpoint index, anchor point alignment index, and inverted index are constructed using the endpoints, key anchor points, and fragment features of the k-fragment sequence, respectively, to obtain the multi-level index.

3. The method for generating railway station interlocking external routes based on template fragment splicing according to claim 2, characterized in that, The expression for the fragment weight attribute is: ； in, Indicates the first The weights of k-segments, Indicates the first k-segments, This indicates the frequency of the segment in the existing route list. This represents the weighting coefficient for key coverage levels. This indicates the number of key segments contained in the fragment.

4. The method for generating railway station interlocking external routes based on template fragment splicing according to claim 1, characterized in that, S3 includes: Based on the template fragment library, a multi-level index is used for recall to obtain a set of candidate fragments; Based on the set of key segments, the weighted coverage of the candidate segments on the target sequence is calculated. Determine the directional consistency between the candidate segment and the target sequence to obtain the directional consistency factor; Based on the constraint set, the number of consecutive triggering rules of candidate fragments is counted, and a nonlinear penalty term is obtained based on the number of consecutive triggering rules. The matching kernel value is calculated based on the weighted coverage, the direction consistency factor, and the nonlinear penalty term.

5. The method for generating railway station interlocking external routes based on template fragment splicing according to claim 4, characterized in that, The expression for the weighted coverage is: ； in, Indicates candidate fragments For the target sequence Weighted coverage Indicates a sequence The set of k-segments Representing fragments The weights; The expression for the nonlinear penalty term is: ； ； in, This represents the total nonlinear penalty. Indicates constraint rules. Represents a set of constraints. Indicates the rule penalty weight. This indicates the sensitivity to chain-triggered penalties. Representing constraint rules The nonlinear penalty function, Indicates the maximum number of consecutive triggers. This represents the function that takes the maximum value. Indicates the threshold for exemption from punishment; The expression for matching the kernel value is: ； ； in, Indicates matching kernel value, This represents the Sigmoid function. Represents a fixed logarithmic function. Indicates the bias term. Represents the coverage coefficient. This represents the directional consistency coefficient. Indicates the direction consistency factor. Indicates the key coverage factor. This indicates the number of key segments covered by fragment p. This represents the penalty inhibition coefficient. It indicates a tiny amount to prevent dispersion.

6. The method for generating railway station interlocking external routes based on template fragment splicing according to claim 1, characterized in that, The expression for the gap cost is: ； in, Representing fragments and The cost of the gap between them Indicates the length of the gap. Both represent length coefficients. Indicates the exponential amplification factor. Indicates the chain trigger count, Indicates the threshold for exemption from penalty. Indicates the directional conflict coefficient. Indicates a direction conflict. Indicates the critical coverage missing coefficient. Indicates the degree of critical coverage deficiency.

7. The method for generating railway station interlocking external routes based on template fragment splicing according to claim 1, characterized in that, S5 includes: The complete candidate path is validated based on the constraint set, and paths that do not meet the constraints are eliminated to obtain the validated complete candidate path. Based on the verified complete candidate paths, feature extraction is performed to obtain the feature vectors of the verified complete candidate paths; Based on the feature vector, the fusion confidence is calculated using the preset calibration fusion calculation rules to obtain the off-table route list and the traceability evidence chain, thus completing the generation of off-table routes for railway station interlocking.

8. The method for generating railway station interlocking external routes based on template fragment splicing according to claim 7, characterized in that, The expression for the fusion confidence score is: ； in, Indicate candidate paths fusion confidence, This represents the Sigmoid function. Indicates the bias parameter. Represents a learnable parameter vector. Indicate candidate paths The feature vector includes kernel value accumulation, critical coverage strength, consistency, chain nonlinear penalty accumulation, and gap cost statistics.

Images