A multi-floor fire patrol path planning method based on a semantic weighted navigation graph

Abstract

The application discloses a multi-floor fire-fighting inspection path planning method based on a semantic weighted navigation graph, and belongs to the technical field of building fire-fighting inspection path planning and indoor navigation graph processing. In order to solve the problem that the multi-floor fire-fighting inspection path is difficult to consider the target semantic value, the must-arrive coverage, the quota sampling, the cross-floor connection and the explainable execution, the application generates the inspection path with single line, no branch and audit basis through target layering, risk calibration edge weight, target closure graph and open must-arrive path optimization.

Description

Technical Field

[0001] This invention relates to the field of building fire inspection route planning and indoor navigation map processing technology, and in particular to a multi-story fire inspection route planning method based on semantically weighted navigation maps. Background Technology

[0002] As large public buildings, underground parking garages, commercial complexes, and multi-story industrial buildings expand in scale, the number, types, and spatial distribution of fire safety inspection targets are constantly increasing. Targets such as safety exits, fire pump rooms, power distribution rooms, fan rooms, smoke extraction rooms, and fire control rooms are typically distributed across different floors and areas, requiring inspection routes to simultaneously meet the requirements of target coverage, route continuity, and on-site execution sequence.

[0003] Existing route planning methods based on architectural drawings or indoor navigation maps primarily rely on geometric distance and accessibility, typically determining the shortest path on the navigation map. These methods struggle to uniformly translate requirements such as full inspection of safety exits, quota sampling of equipment rooms, optional target gains, cross-floor connectivity, necessary backoff, and route interpretability and auditability into map-based route optimization constraints, leading to discrepancies between the generated routes and actual fire inspection rules.

[0004] Therefore, a multi-story fire inspection route planning method is needed to address the shortcomings of existing technologies. Summary of the Invention

[0005] One objective of this invention is to propose a multi-floor fire inspection route planning method based on a semantically weighted navigation graph. Addressing the limitations of existing technologies in simultaneously satisfying requirements such as semantic value of fire safety targets, mandatory full inspection, quota-based random inspection of equipment rooms, optional target gain, cross-floor connectivity, open-ended execution without returning to the starting point, necessary backtracking, path interpretability, and auditability, this invention proposes a technical solution centered on semantically weighted navigation graphs, target closure graphs, and open-ended mandatory path optimization. This invention achieves the technical effect of ensuring that inspection routes balance accessibility, rule fulfillment, and audit interpretability.

[0006] This invention provides a multi-floor fire inspection route planning method based on semantically weighted navigation graphs, including: S1. Obtain the corrected multi-story building navigation map, fire inspection object set, floor connectivity, navigation edge passability probability, navigation edge uncertainty, inspection start node and inspection business rules, divide the fire inspection object set into mandatory full inspection targets, quota sampling targets and optional gain targets, and generate a target semantic attribute table. S2. Calculate the risk calibration edge weights based on the geometric distance, edge type, floor affiliation, passability probability, and uncertainty of the navigation edges in the multi-story building navigation map, and construct a semantically weighted navigation map by combining the target semantic attribute table and cross-floor cost. S3. Calculate the weighted shortest reachable paths from the inspection start node to each inspection target and between each inspection target on the semantically weighted navigation graph, construct the target closure graph, and save the backtracking index between the edges between closure nodes in the target closure graph and the paths in the original navigation graph. S4. Determine the set of must-reach targets based on the target semantic attribute table and inspection business rules, and generate an open initial inspection sequence on the target closure graph according to the short-hop nearest neighbor strategy. S5. Perform two-swap optimization, upper limit constraint on the number of jumps, and target insertion judgment on the open initial inspection sequence to obtain the optimized target access sequence. Based on the backtracking index, output the multi-floor fire inspection path with the inspection start node as the starting point, single line without branches, no forced return to the starting point, and allowed to backtrack along the existing passage.

[0007] Optionally, S1 includes: Write the object type, floor identifier, area identifier, and inspection rule identifier of the fire inspection object into the object attribute record; Safety exits will be designated as mandatory full inspection targets. At least one of the following equipment rooms—fire pump room, power distribution room, fan room, smoke control room, and fire control room—is designated as quota inspection targets, and the quota inspection quantity or quota inspection ratio is recorded for each type of quota inspection target. Objects that are not subject to mandatory full inspection targets or quota sampling targets but have pre-set inspection value are identified as optional gain targets; The target semantic attribute table is generated from the object attribute records, target category, and quota rules.

[0008] Optionally, S2 includes: Risk calibration edge weights are determined as follows: for each navigation edge, the geometric distance, edge type, floor affiliation, passability probability, and uncertainty are read; The lower confidence bound is determined based on the difference between the passability probability and the product of the preset confidence coefficient and uncertainty, and the lower confidence bound is restricted to a value range of 0 to 1. Find the basic penalty coefficient based on the edge type, and find the cross-floor cost based on the floor affiliation and floor connectivity. The risk penalty amount is determined by multiplying the preset risk coefficient by the difference between 1 and the confidence lower bound; Read the distance weight, edge type weight, cross-layer weight and risk weight from the preset edge weight parameter table, and add the product of distance weight and geometric distance, edge type weight and basic penalty coefficient, cross-layer weight and cross-layer cost, and risk weight and risk penalty amount to obtain the risk calibration edge weight. Furthermore, the target semantic attribute table and the cross-floor cost construct semantic weighted navigation map include: the target semantic attribute table includes the target basic semantic value, the number of neighboring objects, the category of neighboring objects, the quota allocation field, and the target floor field; Among them, the target basic semantic value is determined according to the target category mapping table, the number of neighboring objects and the category of neighboring objects are obtained by statistics based on the neighborhood range centered on the navigation node where the target is located and the radius is determined by the building scale, the quota attribution field is used to identify the floor, area or equipment category to which the target belongs, and the target floor field is used to participate in cross-floor cost calculation. Write a node benefit field and a node constraint field on the navigation node corresponding to the inspection target in the multi-story building navigation map. The node benefit field includes the target basic semantic value and the number of neighboring objects. The node constraint field includes a quota allocation field and a target floor field. The navigation nodes with written node benefit fields and node constraint fields, along with the navigation edges with the aforementioned risk calibration edge weights, are combined to generate the semantically weighted navigation graph, which simultaneously stores path search cost, target benefit basis, and target constraint basis.

[0009] Optionally, S3 includes: The navigation nodes corresponding to the inspection start node, the mandatory full inspection target, the quota sampling target, and the optional gain target are used as closure nodes. Using the risk-calibrated edge weights as the path search cost, calculate the weighted shortest reachable path between any two closure nodes; When there is a reachable path between two closure nodes, an edge is established between the closure nodes in the target closure graph, and the weight of the edge between the closure nodes is set to the cumulative risk calibration edge weight of the corresponding weighted shortest reachable path. When there is no reachable path between the closure node and the inspection start node, or when there is no reachable path between the closure node and all existing closure nodes, an unreachable target identifier is generated for the corresponding closure node. The original navigation node sequence and original navigation edge sequence traversed by the corresponding weighted shortest reachable path are saved as the backtracking index, and the unreachable target identifier is saved to the abnormal node field of the target closure graph.

[0010] Optionally, S4 includes: The set of achievable targets is determined as follows: mandatory full-inspection targets that have not generated unachievable target identifiers are added to the set of achievable targets, and mandatory full-inspection targets that have generated unachievable target identifiers are written into the set of targets to be reviewed; For each type of quota sampling target, the remaining quota gap is calculated based on the quota inspection quantity or quota inspection ratio in the inspection business rules and the number of similar targets that have been added to the must-achieve target set. When the remaining quota gap is greater than zero, quota sampling targets that have not generated unreachable target identifiers and have a reachable path from the inspection start node to the target node are selected as quota candidate targets, and quota sampling targets that have generated unreachable target identifiers are written into the target set to be reviewed. When the number of quota candidate targets is zero, a quota unmet status is generated and normalization and sorting of such quota sampling targets are stopped. When the number of quota candidate targets is greater than zero, the candidate set composed of the current quota candidate targets is used as the normalization benchmark, and the target basic semantic value, the number of neighboring objects and the target closure graph cost from the inspection start node to the target node are mapped to the interval between 0 and 1 respectively. Read the quota semantic weight, quota proximity weight, and quota cost weight from the preset target sorting parameter table. Then, subtract the product of the quota cost weight and the target closure graph cost normalization value from the sum of the product of the quota semantic weight and the target basic semantic value normalization value, and the product of the quota proximity weight and the neighbor number normalization value. This will give you the quota candidate comprehensive score. When the number of quota candidate targets is less than the remaining quota gap, all quota candidate targets are added to the must-achieve target set and a quota unmet status is generated; When the number of quota candidate targets is not less than the remaining quota gap, quota sampling targets that meet the remaining quota gap are selected and added to the must-achieve target set in descending order of the comprehensive score of quota candidates. When two quota sampling targets have the same comprehensive score, the target with the highest target closure graph cost will be selected first. Furthermore, the short-hop nearest neighbor strategy includes: using the inspection start node as the current node; Among the unvisited reachable targets, the candidate set consisting of the currently unvisited reachable targets is used as the normalization benchmark. The target basic semantic value of the candidate target, the number of neighboring objects, the target closure graph cost from the current node to the candidate target, and the cross-floor cost between the floor where the candidate target is located and the floor where the current node is located are mapped to the interval between 0 and 1 respectively. Read the semantic weight of the next target, the proximity weight of the next target, the path cost weight of the next target, and the cross-layer cost weight of the next target from the preset target sorting parameter table; The next target selection score is obtained by subtracting the product of the next target semantic weight and the normalized value of the target basic semantic value, the sum of the product of the next target proximity weight and the normalized value of the number of neighboring objects, the product of the next target path cost weight and the normalized value of the target closure graph cost, and the product of the next target cross-layer cost weight and the normalized value of the cross-layer cost. Each time, candidate targets are selected as the next access targets in descending order of their next target selection scores. When two candidate targets have the same next target selection score, the candidate target with the higher target closure graph cost is selected first, and the closure node corresponding to the next access target is updated to the current node, until all targets in the target set are added to the open initial inspection sequence.

[0011] Optionally, S5 includes: The two-swap optimization and hop count upper limit constraints include: performing a swap trial calculation on two non-adjacent access segments in the open initial inspection sequence; Calculate the total path cost before and after the swap, the number of consecutive cross-floor jumps, and the number of cross-floor jump segments exceeding the preset jump limit; If the number of consecutive cross-floor jumps before the exchange does not exceed the preset jump limit, then among the exchange results where the number of consecutive cross-floor jumps does not exceed the preset jump limit, the exchange result with the total path cost less than the total path cost before the exchange and the total path cost sorted first in ascending order will be accepted. When the number of consecutive cross-floor jumps before the swap exceeds the preset jump limit, if the number of cross-floor jump segments exceeding the preset jump limit is less than the swap result before the swap, the swap result that is ranked first is accepted in order of the number of cross-floor jump segments exceeding the preset jump limit from the lowest to the highest and the total path cost from the lowest to the highest. If no exchange result that meets the above conditions exists, retain the current inspection order; Recalculate the number of consecutive cross-floor jumps for the retained or updated inspection sequence, and output a jump constraint not met only when the recalculated number of consecutive cross-floor jumps exceeds the preset jump limit; Furthermore, the target insertion judgment includes: taking quota sampling targets and optional gain targets that have not entered the current inspection sequence, have not generated unreachable target identifiers, and have target closure graph edge weights and backtracking indexes with the previous and next accessed targets of the proposed insertion position as insertion candidate targets; targets that do not meet the above conditions are not subject to insertion evaluation and the unreachable target identifier is retained. When a candidate target is inserted between adjacent access targets A and B in the current inspection sequence, the edge weights of the target closure graph from A to the candidate target are added to the edge weights of the target closure graph from the candidate target to B, and then the edge weights of the target closure graph from A to B are subtracted to obtain the cost of the new risk calibration path. Using the candidate set consisting of the currently inserted candidate targets as the normalization benchmark, the target basic semantic value, the number of neighboring objects and the remaining quota gap are mapped to the interval between 0 and 1 respectively. When there is no remaining quota gap, the normalized value of the remaining quota gap is zero. The insertion semantic weight, insertion neighbor weight, and insertion quota weight are read from the preset insertion parameter table. The target marginal revenue is obtained by adding the product of the insertion semantic weight and the normalized value of the target basic semantic value, the product of the insertion neighbor weight and the normalized value of the number of neighboring objects, and the product of the insertion quota weight and the normalized value of the remaining quota gap. When the cost of the new risk calibration path is zero, the preset lower limit of cost greater than zero is used as the denominator of the insertion evaluation. When the cost of the new risk calibration path is greater than zero, the cost of the new risk calibration path is used as the denominator of the insertion evaluation, and the insertion evaluation value is generated based on the ratio of the target marginal benefit to the insertion evaluation denominator. When there is a remaining quota gap in the quota category to which the target belongs, the first threshold correction amount is calculated based on the product of the remaining quota gap and the preset quota correction coefficient; otherwise, the first threshold correction amount is zero. When the inserted path contains navigation edges whose confidence lower bound is less than the preset passage threshold, the confidence lower bounds of each navigation edge in the inserted path are sorted in ascending order, and the second threshold correction amount is calculated based on the difference between the preset passage threshold and the first sorted confidence lower bound and the preset risk correction coefficient; otherwise, the second threshold correction amount is zero. Subtract the first threshold correction amount from the preset insertion threshold and add the second threshold correction amount to obtain the corrected insertion threshold; When the inserted evaluation value reaches the corrected insertion threshold, the corresponding target will be inserted into the position where the cost of the newly added risk calibration path is sorted from smallest to largest. Furthermore, when outputting the multi-floor fire inspection path, the path audit record is output simultaneously. The path audit record includes the original navigation node sequence, original navigation edge sequence, risk calibration edge weight, covered inspection target, target category, quota satisfaction status, target insertion basis, unreachable target identifier, hop count constraint not satisfied identifier, and backtracking path identifier for each path segment. The backtrack path identifier is determined based on the navigation edges that appear repeatedly in the original navigation edge sequence between two adjacent access targets, and is used to identify the path segment that allows backtracking along the existing channel.

[0012] The beneficial effects of this invention are: 1. By dividing fire inspection targets into mandatory full inspection targets, quota sampling targets, and optional gain targets, and generating a target semantic attribute table, the business rules of different inspection targets can be transformed into target constraints that path planning can handle, so that the path generation process no longer depends solely on geometric distance.

[0013] 2. By calculating risk calibration edge weights based on navigation edge geometric distance, edge type, floor affiliation, passability probability, and uncertainty, spatial distance and passability reliability information can be used simultaneously during the path search phase, so that the edges between targets in the target closure graph have clear passability cost basis.

[0014] 3. By using the short-hop nearest neighbor strategy, two-swap optimization, hop count limit constraint and target insertion judgment to generate an open inspection sequence, and simultaneously generate path audit records when outputting the path, it can form a multi-floor inspection path with no branches, no forced return to the starting point, allow necessary backtracking, and facilitate verification of coverage basis and edge weight basis. Attached Figure Description

[0015] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a multi-story fire inspection route planning method based on semantically weighted navigation graphs.

[0016] Figure 2 The flowchart for constructing the semantically weighted navigation graph in step S2 of the present invention is as follows. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figures 1-2 A multi-story fire inspection route planning method based on semantically weighted navigation graphs includes: S1. Obtain the corrected multi-story building navigation map, fire inspection object set, floor connectivity, navigation edge passability probability, navigation edge uncertainty, inspection start node and inspection business rules, divide the fire inspection object set into mandatory full inspection targets, quota sampling targets and optional gain targets, and generate a target semantic attribute table. S2. Calculate the risk calibration edge weights based on the geometric distance, edge type, floor affiliation, passability probability, and uncertainty of the navigation edges in the multi-story building navigation map, and construct a semantically weighted navigation map by combining the target semantic attribute table and cross-floor cost. S3. Calculate the weighted shortest reachable paths from the inspection start node to each inspection target and between each inspection target on the semantically weighted navigation graph, construct the target closure graph, and save the backtracking index between the edges between closure nodes in the target closure graph and the paths in the original navigation graph. S4. Determine the set of must-reach targets based on the target semantic attribute table and inspection business rules, and generate an open initial inspection sequence on the target closure graph according to the short-hop nearest neighbor strategy. S5. Perform two-swap optimization, upper limit constraint on the number of jumps, and target insertion judgment on the open initial inspection sequence to obtain the optimized target access sequence. Based on the backtracking index, output the multi-floor fire inspection path with the inspection start node as the starting point, single line without branches, no forced return to the starting point, and allowed to backtrack along the existing passage.

[0019] In this specific embodiment, S1 includes: The system synchronously reads the fire inspection object set from the building fire protection facility ledger database and the building space information database and establishes an object primary key index. Each object record in the fire inspection object set contains object identifier, object type, spatial location and inspection rule information. The spatial location is defined by the floor identifier and the area identifier and establishes a one-to-one correspondence with the navigation nodes in the corrected multi-story building navigation map. The one-to-one correspondence is achieved by writing the corresponding navigation node identifier of the object in the object record. The inspection rule information is associated with the inspection business rule table through the inspection rule identifier. The inspection business rule table configures an inspection category field and a quota rule field for each object type. The inspection category field is used to indicate whether the object belongs to a mandatory full inspection target, a quota sampling target or an optional gain target. The quota rule field is used to indicate the quota inspection quantity or quota inspection ratio. Object attribute records are written and stored in a structured field format. The object attribute records include at least an object identifier field, an object type field, a floor identifier field, a region identifier field, and an inspection rule identifier field. The object identifier field is used to uniquely identify the inspection object globally. The object type field is used to find the business category of the object in a preset object type enumeration table. The object type enumeration table includes safety exits, fire pump rooms, power distribution rooms, fan rooms, smoke prevention and exhaust fan rooms, and fire control rooms, as well as other object types not mentioned above. The floor identifier field is used to identify the floor where the object is located and is consistent with the floor number of the floor connection relationship. The region identifier field is used to identify the region where the object is located and is consistent with the building zoning code. The inspection rule identifier field is used to locate the applicable inspection category field and quota rule field of the object in the inspection business rule table. When the inspection business rule table configures a quota inspection ratio for a certain object type, the system converts the quota inspection ratio into a definite quota inspection quantity in step S1 and writes it into the quota rule field. This allows subsequent steps to perform quota fulfillment calculations without relying on floating-point ratios. The conversion is calculated using the following formula: ; in, This index represents the equipment room category of the quota sampling target and corresponds to one of the following in the object type field: "fire pump room, power distribution room, fan room, smoke control room, fire control room". The index indicates that the equipment room category is The quota check quantity must be a non-negative integer. The index indicates that the equipment room category is The quota inspection ratio and the range of values ​​are This indicates that the object type within the current building scope belongs to the category index. The total number of fire inspection targets is a positive integer. The round-up operator is used to convert a ratio result into an integer quantity; After completing the writing of object attribute records and determining quota rules, the system performs target segmentation on the fire inspection object set based on the object type field and the inspection business rule table, and generates a target semantic attribute table. The target segmentation rules are as follows: Objects classified as safety exits are identified as mandatory full inspection targets and added to the mandatory full inspection target set. Objects classified as fire pump rooms, power distribution rooms, fan rooms, smoke control rooms, or fire control rooms are identified as quota sampling inspection targets and added to the quota sampling inspection target set, with their quota rule field containing the corresponding category's quota inspection quantity. Alternatively, the quota inspection quantity can be directly configured. The remaining objects are marked as having a preset inspection value in the inspection value configuration table and determined as optional gain targets, and written into the optional gain target set. The inspection value configuration table uses object type as the key and Boolean inspection value identifier as the value, and is configured by the system administrator during deployment. The target semantic attribute table is generated by concatenating object attribute records, target category fields, and quota rule fields, and includes at least target identifier, target category, object type, floor identifier, area identifier, inspection rule identifier, quota inspection quantity, and quota inspection ratio fields. The quota inspection quantity field and quota inspection ratio field of non-quota sampling targets are written with zero values ​​to maintain field consistency and for unified reading in subsequent rule calculations.

[0020] In this specific embodiment, S2 includes: The system uses a modified multi-story building navigation map as the input graph structure. This modified map consists of a set of navigation nodes and a set of navigation edges. Each navigation node carries a floor identifier field and a planar coordinate field, with the planar coordinate field represented in a metric coordinate system. Each navigation edge connects two navigation nodes and carries a geometric distance field, an edge type field, a floor affiliation field, a passability probability field, and an uncertainty field. The geometric distance field is calculated from the metric planar coordinates of the navigation nodes at both ends of the edge and written into the edge attributes. The edge type field is one of corridor edge, door edge, stair edge, or elevator edge. The floor affiliation field is either the same-floor identifier or the cross-floor identifier from the floor identifiers of the connected navigation nodes. The passability probability field is denoted as... And the range of values ​​is The uncertainty field is denoted as And the range of values ​​is ,in This represents the navigation edge index and is used to uniquely locate a navigation edge within the navigation edge set. The system calculates the risk calibration edge weight for each navigation edge and writes this risk calibration edge weight into the path search cost field of the navigation edge. The risk calibration edge weight is denoted as... The The parameters are obtained by linearly superimposing distance, edge type, cross-floor, and risk terms. The parameters used are provided by a preset edge weight parameter table and written as fixed constants during system deployment. The preset edge weight parameter table includes distance weights. Edge type weight Cross-layer weights With risk weight At the same time, the system has a fixed preset confidence coefficient. With preset risk coefficient The lower confidence bound Depend on minus and After obtaining the original confidence lower bound by the product, interval truncation is performed to obtain the result, satisfying the condition that when the original confidence lower bound is less than 0, let... And when the original confidence lower bound is greater than 1, let The basic penalty coefficient is obtained from the edge type field in the basic penalty coefficient table and recorded as follows. ,in Indicates navigation edge The edge type and the basic penalty coefficient table are fixed as the corresponding corridor edges. Corresponding to the door Corresponding to the staircase Elevator side The cost across floors is obtained from the floor affiliation field and the floor connectivity relationship and recorded as follows. When the navigation side When the edge is on the same level When navigation side When the edge spans multiple floors and its type is stair edge, it is considered seasonal. When navigation side When the edge spans multiple floors and its type is elevator edge, it is considered a seasonal edge. and the At the same time, write the cross-floor cost field of the navigation edge for subsequent cross-floor cost reading; The risk calibration edge weights are calculated using the following formula and written as navigation edges. Risk calibration margin: ; in Indicates navigation edge Risk calibration edge weights are used as the cost of path search. Indicates distance weight, Indicates navigation edge The geometric distance, expressed in meters. Indicates the edge type weight. Indicates edge type The base penalty coefficient was found. Indicates navigation edge edge type, Indicates cross-layer weights, Indicates navigation edge The cost of crossing floors, Indicates risk weight. This indicates the preset risk coefficient. Indicates navigation edge The confidence lower bound and the passability probability With uncertainty Determined according to the above truncation rules, in the formula... Indicates the unreliability of passage and is used to form the risk penalty amount; After completing the risk calibration edge weight writing for all navigation edges, the system writes the target semantic attributes into the navigation nodes corresponding to the inspection targets based on the target semantic attribute table to construct a semantically weighted navigation graph. The target semantic attribute table includes at least the target basic semantic value field, the neighboring object quantity field, the neighboring object category field, the quota allocation field, and the target floor field. The target basic semantic value field is obtained by solidifying the mapping from the target category mapping table and written as a fixed value. The mapping rule is solidified as 1.0 for mandatory full inspection targets, 0.7 for quota sampling targets, and 0.4 for optional gain targets. The neighboring object quantity field is obtained through neighborhood statistics and generates a neighborhood range with a radius of 8 meters centered on the metric plane coordinates of the target's corresponding navigation node. The number of fire inspection objects falling within this neighborhood range and whose floor identification field matches the target floor field is counted. The neighboring object category field records the set of object types appearing within the neighborhood range in the form of a set of object type fields and writes it as a deduplicated object type code list. The quota allocation field is expressed as the string "Quota Category: Write the identifier in the form of "identifier" and write its object type as the quota category and its object type code as the identifier for quota sampling targets. Write "none: 0" for non-quota sampling targets. Write the floor identifier field of the corresponding navigation node directly into the target floor field. The system writes a node benefit field and a node constraint field to each navigation node corresponding to the inspection target in the corrected multi-story building navigation map, forming a semantically weighted navigation map. The node benefit field consists of a target basic semantic value field and a neighboring object quantity field, and is used to provide benefit basis when selecting and inserting targets in subsequent evaluations. The node constraint field consists of a quota allocation field and a target floor field, and is used to provide constraint basis when judging quota satisfaction and calculating cross-floor costs in subsequent evaluations. The semantically weighted navigation map is generated by combining the set of navigation nodes with the node benefit field and node constraint field written in it, and the set of navigation edges with the risk calibration edge weight written in it and the cross-floor cost field retained, and is persistently stored in the form of a graph data structure, so that the same graph structure can simultaneously save path search costs, target benefit basis, and target constraint basis.

[0021] In this specific embodiment, S3 includes: The system uses a semantically weighted navigation graph as the base graph for path search and reads the risk calibration edge weight of each navigation edge in it. As a cost of path search, This represents the navigation edge index and is used to uniquely locate a navigation edge. Indicates navigation edge The risk calibration edge weight is a non-negative real number and is consistent with the path search cost field written in step S2; The system constructs a set of closure nodes and writes them into a closure node table. The set of closure nodes consists of an inspection start node and navigation nodes corresponding to mandatory full inspection targets, quota sampling targets, and optional gain targets. The closure node table records the closure node identifier, the original navigation node identifier, the target category field, and the floor identifier field for each closure node. The target category field of the inspection start node is written with the starting point identifier to distinguish the inspection target. The system processes each closure node in ascending order of the closure node identifier in the closure node table and constructs a target closure graph. The target closure graph is composed of closure nodes as nodes and edges between closure nodes as edges. Each edge between closure nodes stores an edge weight field and a backtracking index pointer field. The system initializes the established closure node set and only includes the closure node corresponding to the inspection start node, which is used to implement the rules for determining unreachable target identifiers. When processing the current closure node At that time, the system The corresponding original navigation node is used as the source node to perform a priority queue-based Dijkstra's shortest path search on the semantically weighted navigation graph to obtain the path from the source node. The minimum cumulative cost value of all original navigation nodes is obtained, and the predecessor navigation node and predecessor navigation edge of each relaxed node are recorded to form a backtrackable predecessor index table. The system uses each closure node in the established set of closure nodes. The corresponding original navigation node is used as the endpoint to read its minimum cumulative cost and determine reachability. Arrival When a reachable path exists, the system establishes edges between closure nodes in the target closure graph. The system writes the edge weights as the cumulative risk calibration edge weights for the reachable path. Simultaneously, based on the predecessor index table, it backtracks from the endpoint original navigation node to the source original navigation node to recover the original navigation node sequence and the original navigation edge sequence. The original navigation node sequence is stored as an ordered list according to the actual travel direction and includes start and end nodes. The original navigation edge sequence is stored as an ordered list according to the actual travel direction and includes segmented navigation edge indices. The system writes the original navigation node sequence and the original navigation edge sequence as backtracking indices into the backtracking index table and performs edge mapping between closure nodes. The index number of the backtracking index table is written into the backtracking index pointer field to achieve a unique mapping from the closure edge to the original navigation graph path; The edges between closure nodes The edge weight is denoted as and satisfy ,in Represents a closure node With closure node The edge weights between closure nodes are non-negative real numbers. This indicates that the index of the closure node in the set of closure nodes has been established. This indicates the index of the closure node currently being processed. This indicates the starting point determined by the backtracking index. arrive The set of original navigation edge index sequences traversed by the weighted shortest reachable path. Represents a set One of the original navigation edge indices, Represents the original navigation edge Risk calibration of boundary weights; When processing the current closure node If no reachable path exists between the current closure node and any of the existing closure nodes in the set, the system generates a corresponding closure node. The unreachable target identifier is written into the abnormal node field of the target closure graph. At the same time, the closure node is excluded from the subsequent edge establishment process between closure nodes to avoid generating invalid closure edges. The unreachable target identifier records the unreachable reason field as unreachable and records the corresponding original navigation node identifier for subsequent review. When the current closure node The system will [implement a new closure] if there is a reachable path to at least one closure node in the existing set of closure nodes. The established closure node set is added to ensure that the reachability judgment of subsequent closure nodes is performed based on the connected closure node set, thereby completing the construction of the target closure graph and ensuring that each edge between closure nodes in the target closure graph has an original navigation node sequence and an original navigation edge sequence that can be traced back to the original navigation graph, and that each unreachable inspection target has an unreachable target identifier in the abnormal node field.

[0022] In this specific embodiment, S4 includes: The system reads the target closure graph and parses the abnormal node fields to obtain the unreachable target identifier. The system establishes a set of reachable targets and a set of targets to be reviewed and initializes them as empty sets. At the same time, the inspection start node is recorded as the starting closure node and used as the starting point for the subsequent open initial inspection sequence. The system iterates through the closure nodes corresponding to the mandatory full inspection target and performs reachability filtering. When the mandatory full inspection target does not generate an unreachable target identifier, it is added to the must-reach target set. When the mandatory full inspection target generates an unreachable target identifier, it is added to the target set to be reviewed and the target identifier and unreachable reason field of the target are recorded in the target set to be reviewed. The system then groups the quota sampling targets according to the quota attribution field. The quota attribution field is the quota attribution field written in the constraint field of the navigation node in step S2, and the object type is used as the quota category identifier. The system then processes each quota category. Read the quota check count written into the target semantic attribute table in step S1 and record it as . ,in This represents a quota category index that corresponds one-to-one with the object type. Indicates quota category The number of quota checks must be a non-negative integer; The system counts the number of similar quota sampling targets already included in the guaranteed target set and records them as follows. ,in Indicates quota category The number of selected targets in the target set that must be reached, and these numbers are non-negative integers; The system calculates the remaining quota gap and records it as... And according to Confirmed, among which Indicates quota category The remaining quota amount that needs to be supplemented must be a non-negative integer. This function represents the larger of the two values. when When the value is greater than 0, the system constructs a quota candidate target set. The quota candidate target set consists of quota sampling targets that simultaneously meet the conditions of not generating an unreachable target identifier and having an edge between the inspection start node and the target's closure node in the target closure graph. The system writes the quota sampling targets that generate an unreachable target identifier into the target set to be reviewed and records their quota category identifier. When the quota candidate target set is empty and When the value is greater than 0, the system generates a quota not satisfied status, writes the status to the quota category status table, and stops performing normalization and sorting on the quota category. When the quota candidate target set is not empty, the system uses the quota candidate target set as the normalization benchmark to read the target basic semantic value, the number of neighboring objects, and the target closure graph cost from the inspection start node to the candidate target for each candidate target. The target basic semantic value and the number of neighboring objects are from the node benefit field of the semantic weighted navigation graph and correspond to the target basic semantic value field and the number of neighboring objects field written in step S2, respectively. The target closure graph cost is the edge weight between the closure nodes from the inspection start node to the candidate target in the target closure graph and is consistent with the closure edge weight defined in step S3. The system uses minimum-maximum normalization to map the above three types of values ​​to the interval between 0 and 1, and writes the normalization result of the values ​​of that type as 0 when the maximum value is equal to the minimum value, in order to ensure the determinism of the calculation; The system reads the quota semantic weight, quota proximity weight, and quota cost weight from the preset target sorting parameter table and solidifies them. ,in Indicates the semantic weight of the quota. Indicates the adjacent weight of the quota. Indicates the quota cost weight; The system calculates the comprehensive score for each quota candidate target and determines it using the following formula: ; in Indicate candidate target The selected score is used for sorting. Indicates the candidate target index. Indicate candidate target The target basic semantic value normalized value, Indicate candidate target The normalized value of the number of neighboring objects. Indicate candidate target The normalized value of the target closure graph cost. Indicate candidate target The normalized value of cross-floor cost. This represents the cross-level cost weight and is used in the calculation of the comprehensive score for quota candidates. And ensure that cross-floor cost items are not included in the overall score for quota candidates; The system will sort the quota candidate targets by Sort by size from largest to smallest and When the costs are the same, they are sorted from smallest to largest according to the target closure graph cost from the inspection start node to the target. If the number of quota candidate targets is less than [a certain value], [the remaining targets will be considered]. Then, all quota candidate targets are added to the guaranteed target set and a quota unmet status is generated and written to the quota category status table. If the number of quota candidate targets is not less than Then select the one that appears first in the sort. Each quota candidate target is added to the must-achieve target set; After completing all quota category processing, the system generates an open initial inspection sequence on the target closure graph. The system takes the inspection start node as the current node and establishes an unvisited must-reach target set. The unvisited must-reach target set consists of all must-reach targets in the must-reach target set except for the inspection start node. In each round of selection, the system reads the basic semantic value of each candidate target and the number of neighboring objects based on the current set of unvisited reachable targets as the normalization benchmark, and reads the target closure graph cost from the current node to the candidate target. The system also reads the floor where the candidate target is located and the floor where the current node is located, and retrieves the cross-floor cost from the cross-floor cost table accordingly. The cross-floor cost table uses floor identifier pairs as keys and cross-floor cost values ​​as values, with 0 for the same floor, 1 for jumps to adjacent floors, and 2 for crossing two floors, and increases linearly according to the absolute value of the floor difference. The system performs minimum-maximum normalization on the above four types of values, and writes the normalization result of the value of the above type to 0 when the maximum value is equal to the minimum value. The system reads the semantic weight of the next target, the proximity weight of the next target, the path cost weight of the next target, and the cross-layer cost weight of the next target from the preset target sorting parameter table and fixes them into a fixed value. ,in Indicates the semantic weight of the next target. Indicates the neighbor weights of the next target. Indicates the cost weight of the next target path. Indicates the cross-level cost weight of the next objective; The system calculates the next target selection score for each candidate target using the same scoring formula described above. and according to Sort from largest to smallest, in If the target closure graph costs from the current node to the candidate target are the same, they are sorted in ascending order. The candidate target with the highest score is determined as the next access target and added to the open initial inspection order. Then, the next access target is removed from the set of unvisited guaranteed targets and the current node is updated to the closure node corresponding to the next access target. This process continues until the set of unvisited guaranteed targets is empty, thus obtaining an open initial inspection order that starts from the inspection start node and does not force a return to the start point. The quota unmet status and the set of targets to be reviewed are written into the output data structure of step S4 for further processing in step S5.

[0023] In this specific embodiment, S5 includes: The system reads the open initial inspection sequence and represents it as an ordered sequence of closure nodes. Simultaneously, it reads the edge weights between adjacent closure nodes in the target closure graph and the corresponding backtracking index pointers. The edge weights between closure nodes are the steps... The cumulative risk calibration edge weight written is denoted as and Let represent the preceding and following closure nodes in the open initial inspection sequence, respectively. The system defines the total path cost of the current inspection sequence as the cost of all adjacent closure node pairs in the sequence. The sum is used as the evaluation criterion for two-way exchange optimization; The system has a fixed preset maximum number of jumps. ,in This indicates the maximum number of consecutive cross-floor jumps allowed. Consecutive cross-floor jumps refer to adjacent segments with different floor identifiers that appear consecutively in the inspection sequence. The system calculates the number of consecutive cross-floor jumps for the current inspection sequence and the number exceeding this limit. The number of cross-floor jump segments is recorded and written to a temporary statistics field. When performing two-swap optimization, the system employs a deterministic traversal and update strategy. The system enumerates two non-adjacent access segments based on the closure node's position index in the inspection sequence, from smallest to largest, and performs a swap trial for each segment. The swap trial only replaces the affected adjacent closure node pairs on the target closure graph and recalculates the total path cost after the swap, as well as the number of consecutive cross-floor jumps and other related costs. The number of cross-floor jump segments; When the number of consecutive cross-floor jumps before the exchange does not exceed The number of consecutive cross-floor jumps after all exchanges should also not exceed [a certain number]. From the exchange results, filter the exchange results whose total path cost is less than the total path cost before the exchange, and select the exchange result with the smallest total path cost to update the inspection order. When the number of consecutive cross-floor jumps before the exchange exceeds The system exceeds [a certain value] after all exchanges. The number of cross-floor jump segments is less than the number of segments in the exchange result before the exchange, according to the number of segments exceeding the number of jump segments. The system updates the inspection order by selecting the exchange results that are sorted first, based on the number of cross-floor jump segments from few to many and the total path cost from small to large. If there are no exchange results that meet the above conditions, the system keeps the current inspection order unchanged. The system fixes the number of iterations for the two-swap optimization as follows: ,in This represents the maximum number of iterations in a two-commutation optimization, where the system accepts only one update in each iteration and reaches [the maximum number of iterations]. Or stop the two-swap optimization if no update occurs in two consecutive iterations; After the two-way exchange optimization is completed, the system recalculates the number of consecutive cross-floor jumps in the updated inspection sequence, and if the number still exceeds this limit... The system outputs a "hop count constraint not met" flag and writes it into the hop count constraint field. The "hop count constraint not met" flag is used in subsequent audit records to indicate that although the inspection sequence has been optimized, there is still a risk of frequent jumps across floors. The system then performs a target insertion decision to select quota sampling targets that have not entered the current inspection sequence and optional gain targets as candidate supplementary targets. The selection of candidate targets for insertion also meets the following conditions: no unreachable target identifier has been generated and the target is the previous access target in the target closure graph at the proposed insertion position. and the next visit target All nodes have edges between closure nodes and have backtracking index pointers, among which and This indicates two adjacent selected access targets in the current inspection sequence. The system retains the unreachable target identifier for targets that do not meet the above edge and backtracking index conditions and does not enter the insertion evaluation. The system accesses each pair of adjacent targets and Construct a candidate set for insertion at this position and use this set as a normalization benchmark to perform minimum-maximum normalization mapping on the target basic semantic value, the number of neighboring objects, and the remaining quota gap of the candidate insertion target to the interval of 0 to 1. The target basic semantic value and the number of neighboring objects are from the node revenue field of the semantic weighted navigation graph, and the remaining quota gap is from the quota category status table output in step S4. When there is no remaining quota gap in the quota category to which the target belongs, the normalized value of the remaining quota gap is written as 0. The system reads the insertion semantic weight, insertion proximity weight, and insertion quota weight from the preset insertion parameter table and stores them as... ,in Indicates the semantic weight of insertion. This indicates the insertion of neighboring weights. This represents the weight of the inserted quota, which the system uses to calculate each candidate insertion target. Target marginal revenue And Defined as Normalized value of the target basic semantic value The product of Normalized value of the number of neighboring objects The product of Normalized value of remaining quota gap The sum of the products of the three, where Indicate target The target basic semantic value normalized value, Indicate target The normalized value of the number of neighboring objects. Indicate target The normalized value of the remaining quota gap; The system performs insertion on each candidate target Calculate the cost of adding a risk calibration path And Defined as ,in Represents the target closure graph arrive The edge weights between closure nodes Represents the target closure graph arrive The edge weights between closure nodes Represents the target closure graph arrive The edge weights between closure nodes are fixed by the system with a pre-defined lower cost limit. ,in Indicates when It is used to avoid the lower limit of positive numbers where the denominator is zero; The system performs insertion on each candidate target Calculate the interpolated evaluation value and determine it using the following formula: ; in Indicates the insertion of candidate targets Inserted evaluation value, Indicate target The target marginal return, Indicate target Insert adjacent access target and The additional risk calibration path costs generated between them This indicates the preset lower limit of cost. This function represents the larger of the two values. The system has a fixed preset insertion threshold. ,in This represents the insertion threshold before correction; the system has a fixed preset quota correction coefficient. With preset risk correction coefficient ,in This is used to convert the remaining quota gap of the target quota category into a first threshold correction amount and then adjust it according to the remaining quota gap. The product is determined. This is used to convert the risk difference of the low-confidence lower bound edge in the insertion path into a second threshold correction amount, and then multiply it by the difference between the preset passage threshold and the lowest confidence lower bound. Confirmed, the system has a fixed preset access threshold. ,in This represents the threshold used to determine whether the inserted path contains a navigation edge with a low confidence lower bound; The system retrieves the backtracking index pointer from the target closure graph when calculating the first threshold correction. arrive as well as arrive The corresponding original navigation edge sequence is read edge by edge, and the confidence lower bound determined in step S2 is obtained. ,in Represents the original navigation edge If the confidence lower bound is between 0 and 1, the system will insert all values ​​in the path that satisfy the condition. navigation side Sort the values ​​in ascending order and take the minimum value at the top of the sort as the lowest confidence lower bound of the insertion path to calculate the second threshold correction amount. The system defines the corrected insertion threshold as... Subtract the first threshold correction amount from the base value and add the second threshold correction amount, then insert the corrected value into the threshold value and denot it as . The system accesses each pair of adjacent targets. and The filter for all inserted candidate targets satisfies The goal is to calibrate the path cost according to the added risks. Sort by size from smallest to largest, and insert the first item in the sorted list. and The system updates the inspection order synchronously between the two points. The system performs the insertion judgment pair by pair from the starting point to the ending point in the order of the inspection order until the traversal is completed, thereby ensuring that the insertion process is definite and reproducible. After obtaining the final target access order, the system expands the edges between the closure nodes of each pair of adjacent access targets into the original navigation node sequence and the original navigation edge sequence according to the backtracking index pointer in the target closure graph, and splices them together to output a multi-story fire inspection path. The multi-story fire inspection path takes the inspection start node as the starting point and does not force a return to the starting point. It is output in the form of a linear node sequence to meet the single-line no-branch requirement. At the same time, the system allows repeated navigation edges in the original navigation edge sequence to indicate backtracking along the existing channel. The system outputs multi-floor fire inspection routes and route audit records simultaneously, generating audit entries based on route segments. A route segment refers to the travel segment corresponding to the original navigation edge sequence obtained by expanding the sequence between two adjacent access targets in the inspection sequence. The audit entry fields are fixed and include: route segment identifier, route segment start and end targets, original navigation node sequence, original navigation edge sequence, cumulative risk calibration edge weight of the route segment, list of inspection targets covered by the route segment, category of each inspection target, quota fulfillment status, target insertion basis, unreachable target identifier, hop count constraint not met identifier, and backtracking path identifier. The cumulative risk calibration edge weight of the route segment is obtained by reading the risk calibration edge weights edge by edge from the original navigation edge sequence of that route segment. The summation is consistent with the edge weights between the closure nodes. The target insertion is then written into the field corresponding to that target. With the corrected insertion threshold and the adopted and To support verification, the fallback path identifier is determined by detecting whether there are duplicate navigation edge indices in the original navigation edge sequence of the same path segment. If duplicates are found, the fallback identifier is written as true, thus clearly identifying the path segment that allows fallback along the existing channel.

[0024] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0025] This invention incorporates fire inspection rules, target semantic attributes, and navigation edge access reliability into a unified graph structure through a semantically weighted navigation graph. This creates a continuous technical processing chain between target selection, inter-target path calculation, and final route splicing, thereby addressing the problem of difficulty in mapping rule constraints in multi-story fire inspection path planning.

[0026] This invention introduces an uncertainty calibration and quota gap linkage mechanism in the edge cost and target insertion stages, so that risk calibration edge weight, quota gap benefit and candidate insertion ranking jointly participate in path optimization, thereby enabling the generated path to simultaneously take into account the reliability of passage, the satisfaction of sampling rules and the inspection benefit.

Claims

1. A multi-story fire inspection route planning method based on semantically weighted navigation graphs, characterized in that, include: S1. Obtain the corrected multi-story building navigation map, fire inspection object set, floor connectivity, navigation edge passability probability, navigation edge uncertainty, inspection start node and inspection business rules, divide the fire inspection object set into mandatory full inspection targets, quota sampling targets and optional gain targets, and generate a target semantic attribute table. S2. Calculate the risk calibration edge weights based on the geometric distance, edge type, floor affiliation, passability probability, and uncertainty of the navigation edges in the multi-story building navigation map, and construct a semantically weighted navigation map by combining the target semantic attribute table and cross-floor cost. S3. Calculate the weighted shortest reachable paths from the inspection start node to each inspection target and between each inspection target on the semantically weighted navigation graph, construct the target closure graph, and save the backtracking index between the edges between closure nodes in the target closure graph and the paths in the original navigation graph. S4. Determine the set of must-reach targets based on the target semantic attribute table and inspection business rules, and generate an open initial inspection sequence on the target closure graph according to the short-hop nearest neighbor strategy. S5. Perform two-swap optimization, upper limit constraint on the number of jumps, and target insertion judgment on the open initial inspection sequence to obtain the optimized target access sequence. Based on the backtracking index, output the multi-floor fire inspection path with the inspection start node as the starting point, single line without branches, no forced return to the starting point, and allowed to backtrack along the existing passage.

2. The multi-story fire inspection route planning method based on semantically weighted navigation graphs according to claim 1, characterized in that, S1 includes: Write the object type, floor identifier, area identifier, and inspection rule identifier of the fire inspection object into the object attribute record; Safety exits will be designated as mandatory full inspection targets. At least one of the following equipment rooms—fire pump room, power distribution room, fan room, smoke control room, and fire control room—is designated as quota inspection targets, and the quota inspection quantity or quota inspection ratio is recorded for each type of quota inspection target. Objects that are not subject to mandatory full inspection targets or quota sampling targets but have pre-set inspection value are identified as optional gain targets; The target semantic attribute table is generated from the object attribute records, target category, and quota rules.

3. The multi-story fire inspection route planning method based on semantically weighted navigation graphs according to claim 2, characterized in that, The risk calibration edge weights in step S2 are determined as follows: for each navigation edge, the geometric distance, edge type, floor affiliation, passability probability, and uncertainty are read; The lower confidence bound is determined based on the difference between the passability probability and the product of the preset confidence coefficient and uncertainty, and the lower confidence bound is restricted to a value range of 0 to 1. Find the basic penalty coefficient based on the edge type, and find the cross-floor cost based on the floor affiliation and floor connectivity. The risk penalty amount is determined by multiplying the preset risk coefficient by the difference between 1 and the confidence lower bound; Read the distance weight, edge type weight, cross-layer weight, and risk weight from the preset edge weight parameter table, and add the product of distance weight and geometric distance, edge type weight and basic penalty coefficient, cross-layer weight and cross-layer cost, and risk weight and risk penalty amount to obtain the risk calibration edge weight.

4. The multi-story fire inspection route planning method based on semantically weighted navigation graphs according to claim 3, characterized in that, Step S2, which combines the target semantic attribute table and cross-floor cost to construct a semantically weighted navigation graph, includes: The target semantic attribute table includes the target basic semantic value, the number of neighboring objects, the category of neighboring objects, the quota allocation field, and the target floor field; Among them, the target basic semantic value is determined according to the target category mapping table, the number of neighboring objects and the category of neighboring objects are obtained by statistics based on the neighborhood range centered on the navigation node where the target is located and the radius is determined by the building scale, the quota attribution field is used to identify the floor, area or equipment category to which the target belongs, and the target floor field is used to participate in cross-floor cost calculation. Write a node benefit field and a node constraint field on the navigation node corresponding to the inspection target in the multi-story building navigation map. The node benefit field includes the target basic semantic value and the number of neighboring objects. The node constraint field includes a quota allocation field and a target floor field. The navigation nodes with written node benefit fields and node constraint fields, along with the navigation edges with the aforementioned risk calibration edge weights, are combined to generate the semantically weighted navigation graph, which simultaneously stores path search cost, target benefit basis, and target constraint basis.

5. A multi-story fire inspection route planning method based on semantically weighted navigation graphs according to claim 4, characterized in that, S3 includes: The navigation nodes corresponding to the inspection start node, the mandatory full inspection target, the quota sampling target, and the optional gain target are used as closure nodes. Using the risk-calibrated edge weights as the path search cost, calculate the weighted shortest reachable path between any two closure nodes; When there is a reachable path between two closure nodes, an edge is established between the closure nodes in the target closure graph, and the weight of the edge between the closure nodes is set to the cumulative risk calibration edge weight of the corresponding weighted shortest reachable path. When there is no reachable path between the closure node and the inspection start node, or when there is no reachable path between the closure node and all existing closure nodes, an unreachable target identifier is generated for the corresponding closure node. The original navigation node sequence and original navigation edge sequence traversed by the corresponding weighted shortest reachable path are saved as the backtracking index, and the unreachable target identifier is saved to the abnormal node field of the target closure graph.

6. The multi-story fire inspection route planning method based on semantically weighted navigation graphs according to claim 5, characterized in that, The set of achievable targets in step S4 is determined as follows: mandatory full-inspection targets that have not generated unachievable target identifiers are added to the set of achievable targets, and mandatory full-inspection targets that have generated unachievable target identifiers are written into the set of targets to be reviewed. For each type of quota sampling target, the remaining quota gap is calculated based on the quota inspection quantity or quota inspection ratio in the inspection business rules and the number of similar targets that have been added to the must-achieve target set. When the remaining quota gap is greater than zero, quota sampling targets that have not generated unreachable target identifiers and have a reachable path from the inspection start node to the target node are selected as quota candidate targets, and quota sampling targets that have generated unreachable target identifiers are written into the target set to be reviewed. When the number of quota candidate targets is zero, a quota unmet status is generated and normalization and sorting of such quota sampling targets are stopped. When the number of quota candidate targets is greater than zero, the candidate set composed of the current quota candidate targets is used as the normalization benchmark, and the target basic semantic value, the number of neighboring objects and the target closure graph cost from the inspection start node to the target node are mapped to the interval between 0 and 1 respectively. Read the quota semantic weight, quota proximity weight, and quota cost weight from the preset target sorting parameter table. Then, subtract the product of the quota cost weight and the target closure graph cost normalization value from the sum of the product of the quota semantic weight and the target basic semantic value normalization value, and the product of the quota proximity weight and the neighbor number normalization value. This will give you the quota candidate comprehensive score. When the number of quota candidate targets is less than the remaining quota gap, all quota candidate targets are added to the must-achieve target set and a quota unmet status is generated; When the number of quota candidate targets is not less than the remaining quota gap, quota sampling targets that meet the remaining quota gap are selected and added to the must-achieve target set in descending order of the comprehensive score of quota candidates. When two quota sampling targets have the same comprehensive score, the target with the highest target closure graph cost will be selected first.

7. A multi-story fire inspection route planning method based on semantically weighted navigation graphs according to claim 6, characterized in that, The short-hop nearest neighbor strategy in step S4 includes: The inspection start node is taken as the current node; Among the unvisited reachable targets, the candidate set consisting of the currently unvisited reachable targets is used as the normalization benchmark. The target basic semantic value of the candidate target, the number of neighboring objects, the target closure graph cost from the current node to the candidate target, and the cross-floor cost between the floor where the candidate target is located and the floor where the current node is located are mapped to the interval between 0 and 1 respectively. Read the semantic weight of the next target, the proximity weight of the next target, the path cost weight of the next target, and the cross-layer cost weight of the next target from the preset target sorting parameter table; The next target selection score is obtained by subtracting the product of the next target semantic weight and the normalized value of the target basic semantic value, the sum of the product of the next target proximity weight and the normalized value of the number of neighboring objects, the product of the next target path cost weight and the normalized value of the target closure graph cost, and the product of the next target cross-layer cost weight and the normalized value of the cross-layer cost. Each time, candidate targets are selected as the next access targets in descending order of their next target selection scores. When two candidate targets have the same next target selection score, the candidate target with the higher target closure graph cost is selected first, and the closure node corresponding to the next access target is updated to the current node, until all targets in the target set are added to the open initial inspection sequence.

8. A multi-story fire inspection route planning method based on semantically weighted navigation graphs according to claim 7, characterized in that, The two-swap optimization and hop count upper limit constraints in step S5 include: A trial exchange is performed on two non-adjacent access segments in the open initial inspection sequence. Calculate the total path cost before and after the swap, the number of consecutive cross-floor jumps, and the number of cross-floor jump segments exceeding the preset jump limit; If the number of consecutive cross-floor jumps before the exchange does not exceed the preset jump limit, then among the exchange results where the number of consecutive cross-floor jumps does not exceed the preset jump limit, the exchange result with the total path cost less than the total path cost before the exchange and the total path cost sorted first in ascending order will be accepted. When the number of consecutive cross-floor jumps before the swap exceeds the preset jump limit, if the number of cross-floor jump segments exceeding the preset jump limit is less than the swap result before the swap, the swap result that is ranked first is accepted in order of the number of cross-floor jump segments exceeding the preset jump limit from the lowest to the highest and the total path cost from the lowest to the highest. If no exchange result that meets the above conditions exists, retain the current inspection order; The number of consecutive cross-floor jumps is recalculated for the retained or updated inspection sequence, and a jump constraint failure flag is output only when the recalculated number of consecutive cross-floor jumps exceeds the preset jump limit.

9. A multi-story fire inspection route planning method based on semantically weighted navigation graphs according to claim 8, characterized in that, The target insertion judgment in step S5 includes: Quota sampling targets and optional gain targets that have not entered the current inspection sequence, have not generated unreachable target identifiers, and have target closure graph edge weights and backtracking indexes with the previous and next access targets of the proposed insertion position are regarded as insertion candidate targets. Targets that do not meet the above conditions will not be evaluated for insertion and will retain the unreachable target identifier. When a candidate target is inserted between adjacent access targets A and B in the current inspection sequence, the edge weights of the target closure graph from A to the candidate target are added to the edge weights of the target closure graph from the candidate target to B, and then the edge weights of the target closure graph from A to B are subtracted to obtain the cost of the new risk calibration path. Using the candidate set consisting of the currently inserted candidate targets as the normalization benchmark, the target basic semantic value, the number of neighboring objects and the remaining quota gap are mapped to the interval between 0 and 1 respectively. When there is no remaining quota gap, the normalized value of the remaining quota gap is zero. The insertion semantic weight, insertion neighbor weight, and insertion quota weight are read from the preset insertion parameter table. The target marginal revenue is obtained by adding the product of the insertion semantic weight and the normalized value of the target basic semantic value, the product of the insertion neighbor weight and the normalized value of the number of neighboring objects, and the product of the insertion quota weight and the normalized value of the remaining quota gap. When the cost of the new risk calibration path is zero, the preset lower limit of cost greater than zero is used as the denominator of the insertion evaluation. When the cost of the new risk calibration path is greater than zero, the cost of the new risk calibration path is used as the denominator of the insertion evaluation, and the insertion evaluation value is generated based on the ratio of the target marginal benefit to the insertion evaluation denominator. When there is a remaining quota gap in the quota category to which the target belongs, the first threshold correction amount is calculated based on the product of the remaining quota gap and the preset quota correction coefficient; otherwise, the first threshold correction amount is zero. When the inserted path contains navigation edges whose confidence lower bound is less than the preset passage threshold, the confidence lower bounds of each navigation edge in the inserted path are sorted in ascending order, and the second threshold correction amount is calculated based on the difference between the preset passage threshold and the first sorted confidence lower bound and the preset risk correction coefficient; otherwise, the second threshold correction amount is zero. Subtract the first threshold correction amount from the preset insertion threshold and add the second threshold correction amount to obtain the corrected insertion threshold; When the inserted evaluation value reaches the corrected insertion threshold, the corresponding target will be inserted into the position where the cost of the newly added risk calibration path is sorted from smallest to largest.

10. A multi-story fire inspection route planning method based on semantically weighted navigation graphs according to claim 9, characterized in that, When outputting the multi-floor fire inspection path in step S5, a path audit record is output simultaneously. The path audit record includes the original navigation node sequence, original navigation edge sequence, risk calibration edge weight, covered inspection target, target category, quota satisfaction status, target insertion basis, unreachable target identifier, hop count constraint not satisfied identifier, and backtracking path identifier for each path segment. The backtracking path identifier is determined based on the navigation edge that appears repeatedly in the original navigation edge sequence between two adjacent access targets and is used to identify the path segment that allows backtracking along the existing channel.

Images