Hierarchical semantic-guided tourism path optimization method

Abstract

The present application belongs to the technical field of intelligent tourism service, tourism path planning and personalized recommendation, and particularly relates to a layered semantic-guided tourism path optimization method. The method acquires tourism planning data such as accommodation nodes, candidate scenic spots, opening time windows, node-to-node travel time, tourist portraits, budget constraints and variable time window events; establishes a random stay duration and random travel time model, performs forward time sequence deduction and depicts time window cascading effects; models catering, resting, replenishment and reservation activities as variable time window events; measures tail execution risks based on Monte Carlo deduction and CVaR index measurement; and generates and updates a tourism path through macroscopic layer semantic screening, structural layer semantic correction and microscopic layer action selection. The method can improve path semantic matching degree, structural coordination, dynamic event response capability and tail risk control capability under the condition of meeting time window, budget, date and event constraints.

Description

Technical Field

[0001] This invention belongs to the field of dynamic path optimization and risk perception decision-making technology, and particularly relates to a hierarchical semantic guidance method for tourism path optimization. Background Technology

[0002] Tourism route planning typically requires generating visit sequences under constraints such as limited time, attraction opening hours, spatial movement, budget, and tourist preferences. Existing methods can mostly complete attraction selection and visit ordering under static data conditions, but in real-world implementation, flexible events such as fluctuations in travel time, shifts in dwell time, and changes in food, rest, and resupply, as well as changes in tourist preferences, can occur. Local time deviations can compress the feasible visit intervals of subsequent nodes, flexible events can alter the original timeline, and semantic preferences also need to participate in the dynamic correction process. Therefore, existing technologies still have room for improvement in characterizing cascading risks of time windows, inserting variable time window events, coordinating route structures, and selecting hierarchical semantic actions.

[0003] Specifically, the existing technology also includes the following problems:

[0004] The problem of time deviation cascading propagation: fluctuations in travel time and deviations in dwell time can propagate between consecutively accessed nodes, leading to time window conflicts and increased tail execution risk for subsequent nodes.

[0005] The problem of dynamically inserting flexible events: When flexible needs such as catering, rest, replenishment and reserved activities are included in the existing schedule timeline, it can easily cause time sequence reordering, budget changes and access node conflicts.

[0006] The problem of insufficient dynamic participation of semantic information: Tourist profiles and attraction semantic tags are usually used for initial recommendations, but it is difficult to further participate in path structure coordination, risk correction and local action selection.

[0007] The problem of insufficient uniformity in multi-objective evaluation: A unified evaluation standard is needed among path benefits, semantic matching, budget consumption, time utilization, structural coordination and risk control.

[0008] The problem of insufficient local repair mechanism: Under dynamic disturbances, the path needs to be able to quickly select actions such as event insertion, dwell compression, node replacement and node skipping to improve the executability and dynamic adaptability of the solution. Summary of the Invention

[0009] This invention discloses a hierarchical semantic-guided tourism route optimization method, which includes the following steps:

[0010] Step 1: Obtain tourism planning data, which includes accommodation nodes, candidate attraction sets, attraction ratings, access costs, suggested stay duration, opening time windows, attraction themes, attraction tags, baseline travel time between nodes, tourist profiles, budget constraints, planning date set, daily activity periods, and variable time window event set;

[0011] Step 2: Establish a node set and path basic model, unify accommodation nodes and candidate attraction nodes into accessible nodes, and form a path representation that includes the order of accessing nodes, access date, and event insertion position;

[0012] Step 3: The recommended dwell time and the baseline travel time between nodes are used to generate the actual dwell time and the actual travel time between nodes, and to generate basic timing status parameters, including the arrival time, start time, waiting time and departure time of each access node.

[0013] Step 4: Insert variable time window events into the basic timing state parameters to generate reconstructed timing parameters;

[0014] Step 5: Based on the basic time series state parameters and the reconstructed time series parameters, calculate the time window cascading risk, and obtain the tail execution risk of the candidate path through Monte Carlo deduction and CVaR risk measurement to generate a weighted optimization function;

[0015] Step 6: Determine the dominant theme for the candidate attraction set based on visitor profiles, attraction themes, and attraction tags;

[0016] Step 7: Penalize and correct the weighted optimization function to obtain the corrected weighted optimization function;

[0017] Step 8: Generate local action results based on the dominant theme and the modified weighted optimization function;

[0018] Step 9: Update the candidate paths according to the results of local actions, and repeat steps 3 to 8 until the termination condition is met, and output the tourism route optimization solution.

[0019] Furthermore, in step 1, let the set of candidate attractions be... , Indicates the first Candidate attractions This represents the total number of candidate attractions, and the set of accommodation nodes is... , Indicates the accommodation point at the start of each day or trip. The destination accommodation node is represented by the set of all nodes. Any candidate attraction The attribute is represented as:

[0020] ;

[0021] in, Indicate candidate attractions The attributes, Indicates the rating of the attraction. Indicates access cost, Indicates the suggested duration of stay. Indicates the open time window, This indicates semantic attributes including attraction theme and attraction tag.

[0022] Furthermore, in step 2, the path is represented as:

[0023] ;

[0024] This represents the number of attractions actually visited along the route. For ease of calculating the route travel time, let's denote it as:

[0025]

[0026] The path-based model satisfies the following constraints: access uniqueness, budget, open time window, date range, monotonic access date, and daily activity period.

[0027] Furthermore, in step 3, the actual duration of stay at the attraction is modeled using a log-normal distribution:

[0028] ;

[0029] It follows a log-normal distribution. This refers to the actual duration of stay. and The recommended length of stay can be determined by the attraction. and coefficient of variation Confirmed; specifically:

[0030] ;

[0031] The actual travel time between nodes consists of the base travel time and traffic disturbance:

[0032] , ;

[0033] in, This refers to the actual travel time. Indicates the baseline travel time. Indicates traffic disruption. Indicates the variance of traffic disturbances;

[0034] No. The arrival time, start time, waiting time, and departure time of each access node are recursively calculated in the following manner:

[0035] ,

[0036] ,

[0037] ,

[0038] ;

[0039] Let m-1 be the departure time of node m-1. For nodes The actual travel time. These are the opening hours of the attraction;

[0040] Due to the The departure time of each access node Directly affects the next Arrival time of each access node Local time deviations will propagate backward along the access sequence.

[0041] Furthermore, in step 4, the variable time window event set includes drinking, rest, resupply, scheduled activities, and temporary experiences:

[0042] , ;

[0043] in, For events with variable time windows, For the event Attribute vectors, Indicates the duration of the event. This represents the range from the allowed start time to the allowed end time of the event. This represents the set of dates on which the event is allowed to occur. Indicates the event type;

[0044] The event duration relationship and time window constraints are as follows:

[0045] ;

[0046] The end time of the event. The start time of the event. The date the event occurred;

[0047] Set nodes The cumulative time occupied by events before access is The cumulative time occupied by the event after the access is ;

[0048] After the event is inserted, the node sequence is updated as follows:

[0049] ,

[0050] ,

[0051] ,

[0052] ;

[0053] Event insertion actions include gap absorption, forward avoidance, and delayed execution.

[0054] Furthermore, in step 5, average attraction ratings, visit size, time utilization, total visit cost, and tail execution risk are used as the basic objectives:

[0055] ,

[0056] ,

[0057] ;

[0058] Indicates the rating of the attraction. For planning the set of dates, For planning date set Any date in, For date The end time of the day's activities, For a period of time The start time of the day's activities, Indicates access cost;

[0059] node Cumulative time deviation at the location The recursive representation is as follows:

[0060] ;

[0061] For suggested duration of stay, Time absorption margin:

[0062] ;

[0063] For the date of access The corresponding end time of the day's activities, This is the buffer coefficient; Internal buffer size of the node:

[0064] .

[0065] These are the closing times for the attractions. These are the opening hours of the attraction;

[0066] For path conduct The second independent randomized simulation, the first The second simulation yielded the terminal time deviation. Confidence level The quantile value below is denoted as The CVaR risk indicator is:

[0067] ;

[0068] Will to Normalization to Then, construct the weighted optimization function:

[0069] ,

[0070] ;

[0071] in, Let the weight of the i-th item be , For the i-th normalization objective, This represents the feasible region, which is composed of time window constraints, event insertion constraints, budget constraints, node uniqueness constraints, and date constraints.

[0072] Furthermore, in step 6, the tourist profile consists of a set of preferred topics, a set of preferred subtopics, a set of preferred tags, a set of avoidance tags, a set of must-see attractions, and a crowd avoidance coefficient; a macro-semantic matching score is calculated for any candidate attraction.

[0073] , .

[0074] in, A collection of must-see attractions For preferred topic sets, For preferred subtopic sets, and These are respectively the labeled hit items and the avoided hit items. To normalize semantic weights, For indicator functions, and They are divided into primary and secondary themes for the scenic spot. A collection of attraction tags. For a set of preference labels, To avoid tag sets, This is the congestion avoidance coefficient. Due to the crowding at the attractions, and These are respectively marked as disturbance-friendly and supply-friendly. All are macro-level semantic weights;

[0075] Filtered semantic candidate set for:

[0076] ;

[0077] This is the semantic filtering threshold;

[0078] Main topic definition for:

[0079] ;

[0080] A collection of candidate attractions by theme Any topic in This is a collection of candidate scenic spot themes.

[0081] Furthermore, in step 7, the local adjacency coordination degree of adjacent attractions is defined. :

[0082] ;

[0083] and These are the weights for topic compatibility and tag similarity, respectively. For actual tourist attractions and Theme compatibility, For actual tourist attractions and Tag similarity;

[0084] ;

[0085] ;

[0086] in, For the first One candidate scenic spot as a primary theme. For the first A collection of attraction tags, This represents a set of compatible themes that have a natural connection to the tour. This refers to the cross-topic compatibility coefficient.

[0087] Path adjacency uncoordination item for:

[0088] ;

[0089] Definition of the first Daily time load Normalized load and multi-day average load :

[0090] ,

[0091] ;

[0092] Cross-day load imbalance term is :

[0093] ;

[0094] The structural layer embeds adjacency miscoordination and cross-day load imbalance terms into the path evaluation. The corrected time efficiency and risk indicators are as follows:

[0095] ,

[0096] ;

[0097] in, To correct the lower limit parameter, ~ Structural penalty weights; modified time efficiency index Used to replace the time utilization rate indicator in step 5 Revised risk indicators Used to replace the tail execution risk indicator in step 5 and participate in the comprehensive evaluation of subsequent paths; , , , and After normalization, construct the weighted optimization function with modified structure:

[0098] ;

[0099] These are the weights of each objective.

[0100] Furthermore, in step 8, node cascading risk... It consists of local delay pressure, the degree of subsequent time window compression, and local loss terms:

[0101] ;

[0102] in, This indicates localized delay pressure. Indicates the degree of compression of the subsequent time window. This indicates a local semantic quality loss. Risk portfolio weights;

[0103] Local action triggering variables for:

[0104] ;

[0105] As the risk trigger threshold, For the current node The set of feasible event candidates in the vicinity that meet the constraints of time window, date, and single-day activity period. It is an empty set;

[0106] when At that time, construct a set of actionable actions. :

[0107] ,

[0108] in, , , , These represent event insertion, dwell compression, node replacement, and node skipping, respectively.

[0109] Macro-level score of candidate actions Structural layer score and micro-level scores They are represented as follows:

[0110] ,

[0111] ,

[0112] ;

[0113] in, This represents the average semantic value of nodes in a local segment after the action is executed. The main theme is consistent with the following items:

[0114] ;

[0115] in, To execute candidate actions The set of attractions retained or introduced in the current local segment. Indicates tourist attractions The first-level topic, It is an indicator function;

[0116] and This refers to the incremental structural degradation caused by the action. To mitigate the benefits of cascading risk, For event requirement fulfillment items, The cost of delaying the timing and Weights for structural degradation penalties; ~ As a weight for the micro-level score;

[0117] Hierarchical semantic comprehensive score of candidate actions for:

[0118] ;

[0119] in, To normalize residual risk, ~ The final selected local action result is based on the overall weighting.

[0120] ;

[0121] This is the result of a local action.

[0122] The beneficial effects achieved by this invention are:

[0123] This invention can integrate tourist profile semantics, path structure semantics, and dynamic execution semantics into the tourism route optimization process; it can characterize the cascading risks of time windows caused by disturbances in stay duration and travel time; it can handle variable time window events such as dining, rest, resupply, and reserved activities; it can achieve dynamic route repair through local action selection; and it can improve the personalization, structural coordination, executability, and risk control capabilities of tourism route plans. Attached Figure Description

[0124] Figure 1 This is a diagram illustrating the overall framework of the method of this invention, used to demonstrate the overall relationship between input scene construction, dynamic optimization search, macro-level semantic filtering, structural-level semantic correction, micro-level action selection, and output results.

[0125] Figure 2 It is a schematic diagram of tourism route planning scenarios, event insertion, and partial timelines, used to show the relationships between accommodation nodes, attraction nodes, event nodes, traffic relationships, time deviation propagation, opening time windows, and event time windows.

[0126] Figure 3 This is a schematic diagram of the probability density of random dwell time and cascading risk regions. It is used to illustrate that when the dwell time is modeled using a log-normal distribution, long-tail dwell regions may trigger cascading risks in subsequent time windows.

[0127] Figure 4 It is a comparison diagram of the path structure and cross-day load before and after the semantic correction of the structural layer, which is used to show the impact of adjacency coordination constraints and cross-day load balancing constraints on the final access sequence, topic connection and two-day load distribution.

[0128] Figure 5 This is a comparison chart of the overall effects of micro-level action selection strategies. Detailed Implementation

[0129] The present invention will be further described below with reference to specific embodiments, and the advantages and features of the present invention will become clearer as a result. However, these embodiments are merely exemplary and do not constitute any limitation on the scope of the present invention. Those skilled in the art should understand that modifications or substitutions can be made to the details and form of the technical solutions of the present invention without departing from the spirit and scope of the present invention, but all such modifications and substitutions fall within the protection scope of the present invention.

[0130] like Figure 1 As shown, this invention provides a hierarchical semantic-guided tourism route optimization method. This method takes tourism planning data as input and integrates time window cascading effect assessment, Monte Carlo-CVaR risk measurement, variable time window event insertion, hierarchical semantic guidance, and local repair actions into the tourism route search and update process. This ensures that the route plan, during its generation, simultaneously considers tourism benefits, temporal feasibility, risk control, semantic matching, and dynamic adaptability.

[0131] Step 1: Obtain tourism planning data and build a node model. This includes accommodation nodes, candidate attraction sets, attraction ratings, access costs, suggested stay duration, opening time windows, attraction themes, attraction tags, baseline travel time between nodes, tourist profiles, budget constraints, planning date set, daily activity periods, and variable time window event set.

[0132] Let the set of candidate attractions be The accommodation node set is The set of all nodes is For any candidate attraction Its attributes are represented as:

[0133] ,

[0134] in, Indicates the rating of the attraction. Indicates access cost, Indicates the suggested duration of stay. Indicates the opening hours of the attraction. It represents semantic attributes such as attraction theme, function, or preference tags.

[0135] The main symbols and parameters include: A list of candidate attractions Indicates the first Candidate attractions Indicates the total number of candidate attractions; Accommodation nodes collection, Indicates the accommodation point at the start of each day or trip. Indicates the final accommodation node; The entire node set consists of candidate attraction nodes and accommodation nodes; Scenic spot attribute vector Rate the attractions For access costs, This refers to the actual duration of stay. For the opening hours of the attraction, For the semantic attributes of scenic spots; Tourist routes, This indicates the actual number of attraction nodes visited, with no duplicate attraction nodes in the path; For nodes Date of visit, For planning the set of dates, and For the first Start and end times of daily activities; For suggested duration of stay, The parameters are those for the log-normal residence time distribution. For nodes To node The actual travel time. Based on the standard travel time, For traffic disruption, For traffic disturbance variance; They are nodes Arrival time, start time, waiting time, and departure time; For a variable time window set of events, For the first One event, For duration, For the allowed time interval, For the allowed date set, Event type; They are nodes The cumulative time consumed by event insertion before and after the access; To accumulate time deviation, To allow for time margin, This refers to the internal buffer size of the node. These represent average attraction rating, visit size, time utilization, total visit cost, and CVaR tail execution risk, respectively. For the target weight, For the normalization goal, To optimize the function, This is the feasible region; These are the macro-semantic matching score, the semantic filtering threshold, and the set of semantic candidate attractions after filtering, respectively. These are: a set of attraction tags, a set of preference tags, a set of avoidance tags, tag-matched items, and avoidance-matched items. These are respectively the coordination degree of adjacent attractions, the adjacency miscoordination item, and the cross-day load imbalance item; These are node cascade risk, local action triggering variables, and risk triggering thresholds, respectively. The four types of actions are: action set and its event insertion, dwell compression, node replacement, and node skipping. These are the macro-level score, structural level score, micro-level score, and comprehensive action score for the candidate action, respectively.

[0136] Tourism route planning scenarios such as Figure 2 As shown, the scenario includes accommodation nodes, candidate attractions, access relationships between nodes, attraction opening time windows, budget constraints, tourist profiles, and flexible events that can be dynamically inserted.

[0137] Step 2: Establish a basic model of node set and path, unifying accommodation nodes and candidate attraction nodes into accessible nodes, and forming a path representation that includes the order of accessed nodes, access date, and event insertion position:

[0138] ;

[0139] The path-based model satisfies the following constraints: access uniqueness, budget, open time window, date range, monotonic access date, and daily activity period.

[0140] Step 3: Establish a model for random dwell time and random passage time. The actual dwell time at the attraction is modeled using a log-normal distribution:

[0141] ,

[0142]

[0143] It follows a log-normal distribution. This refers to the actual duration of stay. and The recommended length of stay can be determined by the attraction. and coefficient of variation Confirmed; specifically:

[0144] .

[0145] The actual travel time between nodes consists of the base travel time and traffic disturbance:

[0146] ,

[0147] .

[0148] in, Indicates the baseline travel time. Indicates traffic disruption. This represents the variance of traffic disturbances. Therefore, the dual randomness of tourist attraction dwelling behavior and traffic conditions is uniformly incorporated into the route execution process.

[0149] Random dwell time probability density and cascading risk areas, such as Figure 3 As shown.

[0150] Perform forward timeline extrapolation and characterize the cascading effect of time windows. For the first... For each node, calculate its arrival time, actual start time, waiting time, and departure time:

[0151] ,

[0152] ,

[0153] ,

[0154] .

[0155] because Direct impact Local time deviations will propagate backward along the access order. Each node also needs to meet the constraints of the open time window, date range, and daily activity boundary:

[0156] ,

[0157] .

[0158] Step 4: Establish a variable time window event insertion model. Catering, rest, resupply, reserved activities, and temporary experience projects are uniformly represented as a set of variable time window events:

[0159] ,

[0160] .

[0161] in, Indicates the duration of the event. Indicates the allowed time interval for the event to occur. This represents the set of dates on which the event is allowed to occur. Indicates the event type. The event duration and time window constraints are as follows:

[0162] .

[0163] Set nodes The cumulative time occupied by events before access is The cumulative time occupied by the event after the access is After the event is inserted, the node sequence is updated as follows:

[0164] ,

[0165] ,

[0166] ,

[0167] .

[0168] Event insertion actions include gap absorption, look-ahead avoidance, and deferred execution. Gap absorption utilizes the waiting period before a node opens; look-ahead avoidance prioritizes the execution of events just before the event window closes; and deferred execution schedules events if the event time window is still met after a node has been accessed.

[0169] Step 5: Construct a multi-objective evaluation function and a time window cascaded risk indicator.

[0170] The basic objectives are average attraction rating, visit size, time utilization, total visit cost, and tail execution risk.

[0171] ,

[0172] ,

[0173] ;

[0174] node Cumulative time deviation at the location The recursive representation is as follows:

[0175]

[0176] The time absorption margin is defined as:

[0177] ,

[0178] .

[0179] For path conduct The second independent randomized simulation, the first The second simulation yielded the terminal time deviation. Confidence level The quantile value below is denoted as The CVaR risk indicator is:

[0180] ;

[0181] Will to Normalization to Then, construct the weighted optimization function:

[0182] ,

[0183] .

[0184] in, This represents the feasible region, which is composed of time window constraints, event insertion constraints, budget constraints, node uniqueness constraints, and date constraints.

[0185] Step 6: Perform macro-level semantic filtering. The tourist macro-preference profile consists of a set of preference topics, a set of preference subtopics, a set of preference tags, a set of avoidance tags, a set of must-see attractions, and a crowd avoidance coefficient. Calculate the macro-level semantic matching score for any candidate attraction:

[0186] ,

[0187] .

[0188] in, To normalize semantic weights, For indicator functions, and They are divided into primary and secondary themes for the scenic spot. A collection of attraction tags. This is the congestion avoidance coefficient. Due to the crowding at the attractions, and These are respectively marked as disturbance-friendly and supply-friendly. This refers to macroscopic semantic weights.

[0189] The filtered semantic candidate set is as follows:

[0190] .

[0191] The dominant theme is defined as:

[0192] .

[0193] Step 7: Perform structural layer semantic correction. Define the local adjacency coordination degree of adjacent attractions:

[0194] ,

[0195] ,

[0196] .

[0197] in, This represents a set of compatible themes that have a natural connection to the tour. This is the cross-topic compatibility coefficient. The path adjacency miscoordination term is:

[0198] .

[0199] Definition of the first Daily time load, normalized load, and multi-day average load:

[0200] ,

[0201] .

[0202] The cross-day load imbalance item is:

[0203] .

[0204] The structural layer embeds adjacency mismatch and cross-day load imbalance terms into the path evaluation. The revised time efficiency and risk indicators are as follows:

[0205] ,

[0206] .

[0207] in, To correct the lower limit parameter, to The structural penalty weight.

[0208] Revised time efficiency index Used to replace the time utilization rate indicator in step 5 Revised risk indicators Used to replace the tail execution risk indicator in step 5 and participate in the comprehensive evaluation of subsequent paths; , , , and After normalization, construct the weighted optimization function with modified structure:

[0209] ;

[0210] These are the weights of each objective.

[0211] Comparison of path structure and cross-day load before and after semantic correction at the structural layer. Figure 4 As shown.

[0212] Step 8: Perform micro-level action triggering and action scoring. Node cascading risk consists of local delay pressure, the degree of subsequent time window compression, and local loss terms.

[0213] .

[0214] in, This indicates localized delay pressure. Indicates the degree of compression of the subsequent time window. This indicates a local semantic quality loss. For risk portfolio weights. Local action trigger variables are:

[0215] .

[0216] when At that time, construct a set of possible actions:

[0217] ,

[0218] in, , , , These represent event insertion, dwell compression, node replacement, and node skipping, respectively.

[0219] The scores for the macroscopic, structural, and microscopic levels of the candidate action are represented as follows:

[0220] ,

[0221] ,

[0222] .

[0223] in, This represents the average semantic value of nodes in a local segment after the action is executed. As the dominant theme consistent item, and The incremental structural degradation caused by the action. To mitigate the benefits of cascading risk, For event requirement fulfillment items, The cost of delaying the timing.

[0224] Specifically:

[0225] ;

[0226] in, To execute candidate actions The set of attractions retained or introduced in the current local segment. Indicates tourist attractions The first-level topic, It is an indicator function;

[0227] The hierarchical semantic comprehensive score of the candidate action is:

[0228] .

[0229] in, To normalize residual risk, to The weights are calculated based on the overall criteria. The final selected local action result is as follows:

[0230] .

[0231] Step 9: Output the path optimization plan and update it dynamically. The output includes the daily visit sequence, arrival time, start time, departure time, event insertion position, budget consumption, semantic matching score, time utilization, structural coordination indicators, and risk indicators. If real-time traffic, attraction congestion, tourist demand, or reservation status changes, the system will re-execute the time sequence simulation, risk calculation, and local action selection.

[0232] The core parameters and example values ​​in the embodiments are shown in Table 1.

[0233] Table 1 Parameter Setting Table for Example

[0234]

[0235] In one implementation, the present invention can be deployed in a tourism recommendation platform, a smart scenic spot guide system, a travel agency itinerary planning system, an urban cultural tourism service platform, or a mobile personalized itinerary assistant. The system can run on a general-purpose computer or cloud server with a processor of at least four cores and at least 8GB of memory; the software environment can use a Windows or Linux operating system, and be configured with a Python runtime environment, database access interface, map service interface, and scenic spot reservation service interface.

[0236] In one embodiment, the system uses 35 candidate attractions, a 2-day planning cycle, accommodation node start and end constraints, daily activity time boundaries, and budget caps as input data. Visitor profiles are categorized into four types: historical and cultural study tours, low-disturbance mountain and ecological tours, urban nighttime leisure tours, and mixed preferences. The system reads each attraction's primary theme, secondary theme, tags, rating, cost, suggested stay duration, and opening hours, as well as variable time window events such as dining, rest, resupply, and reserved activities.

[0237] The system first establishes a candidate attraction attribute table and a set of path nodes; it then performs random modeling of dwell time and travel time; it recursively calculates the arrival, waiting, start, and departure times of nodes; it inserts flexible events into the path timeline; it calculates CVaR risk through Monte Carlo simulation; it performs macro-level semantic screening, structural-level semantic correction, and micro-level action scoring respectively; and finally, it outputs the optimized visit sequence and path evaluation results.

[0238] Regarding parameter settings, the number of Monte Carlo simulations can be set to 500 to 1000, the CVaR confidence level can be set to 0.90 to 0.95, the semantic filtering threshold can be set according to the size of the candidate attractions, and the node cascading risk trigger threshold can be set according to the destination congestion level, travel time fluctuation level, and user itinerary flexibility. For multi-day travel scenarios, the cross-day load imbalance term is used to constrain the intensity of daily activities and avoid activities being concentrated on a single date.

[0239] When the actual queuing time at a certain attraction increases in the morning, the system calculates the arrival time of subsequent nodes based on the updated departure time and compares the time absorption margin of the subsequent nodes. If the allowed time interval for the lunch event conflicts with the opening time window of subsequent attractions, the system prioritizes attempting to insert the lunch event during the waiting period; if the waiting period is insufficient, it attempts to reduce the stay duration at the previous attraction or replace attractions with low semantic matching. The system calculates semantic benefits, structural coordination benefits, micro-execution benefits, and residual risks for each candidate action and selects the action with the highest comprehensive score.

[0240] In terms of the effectiveness of macro-level semantic filtering, after enabling macro-level semantics, the average semantic matching score of the final path under the four user profiles all improved significantly. Among them, the historical and cultural study tour type improved by 24.14%, the landscape and ecological low-disturbance type improved by 87.02%, the urban night tour and leisure type improved by 83.59%, and the parent-child study tour and exhibition type improved by 118.87%. The coverage rate of preferred themes improved by up to 250.00%, the hit rate of preferred tags improved by up to 160.00%, and the suppression rate of avoidance tags improved by up to 15.38%.

[0241] Regarding the effect of semantic constraints at the structural layer, such as Figure 4As shown, with the structural layer constraints disabled, the relative optimal performance index for adjacency coordination is 51.9%, and the relative optimal performance index for cross-day equilibrium is 1.4%. After enabling the full structural layer, the relative optimal performance index for adjacency coordination increases to 75.7%, and the relative optimal performance index for cross-day equilibrium increases to 15.3%. In the representative path, without enabling the structural layer, the two-day load times are 203 min and 658 min, respectively, with a load difference of 455 min. After enabling the structural layer, the two-day load times are 482 min and 671 min, respectively, with the load difference reduced to 189 min.

[0242] Regarding the effect of action selection at the micro level, such as Figure 5 As shown, the post-action risk of the fixed-priority strategy is 0.4242, the post-action risk of the risk-greedy strategy is 0.2613, and the post-action risk is reduced to 0.2581 after applying the micro-level action scoring of this invention; the event response rate reaches 0.83, approximately 83.33%; compared with the fixed-priority strategy, the CVaR risk is reduced by 56.20%. These results demonstrate that this invention can improve flexible event response capabilities and reduce tail execution risk while maintaining path access scale and time efficiency.

[0243] Without altering the core concept of this invention, the macro-level semantic filtering threshold, structural layer correction weights, CVaR confidence level, Monte Carlo simulation count, event insertion rules, and micro-level action set can all be adjusted based on destination size, real-time data quality, and user preference intensity. All of the above adjustments are equivalent embodiments of the technical solution of this invention.

[0244] The above are merely preferred embodiments of the present invention and do not constitute any limitation on the scope of protection of the present invention; all technical solutions formed by equivalent transformations or equivalent substitutions fall within the scope of protection of the present invention; the parts of the present invention not described in detail are well-known technologies to those skilled in the art.

Claims

1. A hierarchical semantic-guided tourism route optimization method, characterized in that, The hierarchical semantic-guided tourism route optimization method includes the following steps: Step 1: Obtain tourism planning data, which includes accommodation nodes, candidate attraction sets, attraction ratings, access costs, suggested stay duration, opening time windows, attraction themes, attraction tags, baseline travel time between nodes, tourist profiles, budget constraints, planning date set, daily activity periods, and variable time window event set; Step 2: Establish a node set and path basic model, unify accommodation nodes and candidate attraction nodes into accessible nodes, and form a path representation that includes the order of accessing nodes, access date, and event insertion position; Step 3: The recommended dwell time and the baseline travel time between nodes are used to generate the actual dwell time and the actual travel time between nodes, and to generate basic timing status parameters, including the arrival time, start time, waiting time and departure time of each access node. Step 4: Insert variable time window events into the basic timing state parameters to generate reconstructed timing parameters; Step 5: Based on the basic time series state parameters and the reconstructed time series parameters, calculate the time window cascading risk, and obtain the tail execution risk of the candidate path through Monte Carlo deduction and CVaR risk measurement to generate a weighted optimization function; Step 6: Determine the dominant theme for the candidate attraction set based on visitor profiles, attraction themes, and attraction tags; Step 7: Penalize and correct the weighted optimization function to obtain the corrected weighted optimization function; Step 8: Generate local action results based on the dominant theme and the modified weighted optimization function; Step 9: Update the candidate paths according to the results of local actions, and repeat steps 3 to 8 until the termination condition is met, and output the tourism route optimization solution.

2. The hierarchical semantic guidance-based tourism route optimization method according to claim 1, characterized in that, In step 1, let the set of candidate attractions be... , Indicates the first Candidate attractions This represents the total number of candidate attractions, and the set of accommodation nodes is... , Indicates the accommodation point at the start of each day or trip. The destination accommodation node is represented by the set of all nodes. Any candidate attraction The attribute is represented as: ； in, Indicate candidate attractions The attributes, Indicates the rating of the attraction. Indicates access cost, Indicates the suggested duration of stay. Indicates the open time window, This indicates semantic attributes including attraction theme and attraction tag.

3. The hierarchical semantic guidance-based tourism route optimization method according to claim 1, characterized in that, In step 2, the path is represented as: ； This represents the number of attractions actually visited along the route. For ease of calculating the route travel time, let's denote it as: 。 4. The path basic model satisfies the following constraints: access uniqueness constraint, budget constraint, open time window constraint, date range constraint, access date monotonicity constraint, and daily activity period constraint.

5. The hierarchical semantic guidance-based tourism route optimization method according to claim 1, characterized in that, In step 3, the actual duration of stay at the attraction is modeled using a log-normal distribution: ； It follows a log-normal distribution. This refers to the actual duration of stay. and The recommended length of stay can be determined by the attraction. and coefficient of variation Confirmed; specifically: ； The actual travel time between nodes consists of the base travel time and traffic disturbance: ， ； in, This refers to the actual travel time. Indicates the baseline travel time. Indicates traffic disruption. Indicates the variance of traffic disturbances; No. The arrival time, start time, waiting time, and departure time of each access node are recursively calculated in the following manner: , , , ； Let m-1 be the departure time of node m-1. For nodes The actual travel time. These are the opening hours of the attraction; Due to the The departure time of each access node Directly affects the next Arrival time of each access node Local time deviations will propagate backward along the access sequence.

6. The hierarchical semantic guidance-based tourism route optimization method according to claim 1, characterized in that, In step 4, the variable time window event set includes drinking, rest, resupply, scheduled activities, and temporary experience items: ， ； in, For events with variable time windows, For the event Attribute vectors, Indicates the duration of the event. This represents the range from the allowed start time to the allowed end time of the event. This represents the set of dates on which the event is allowed to occur. Indicates the event type; The event duration relationship and time window constraints are as follows: ； The end time of the event. The start time of the event. The date the event occurred; Set nodes The cumulative time occupied by events before access is The cumulative time occupied by the event after the access is ; After the event is inserted, the node sequence is updated as follows: ， ， ， ； Event insertion actions include gap absorption, forward avoidance, and delayed execution.

7. The hierarchical semantic guidance-based tourism route optimization method according to claim 1, characterized in that, In step 5, the basic objectives are average attraction rating, visit size, time utilization, total visit cost, and tail execution risk. ， , ； Indicates the rating of the attraction. For planning the set of dates, For planning date set Any date in, For date The end time of the day's activities, For a period of time The start time of the day's activities, Indicates access cost; node Cumulative time deviation at the location The recursive representation is as follows: ； For suggested duration of stay, Time absorption margin: ； For the date of access The corresponding end time of the day's activities, This is the buffer coefficient; Internal buffer size of the node: ； These are the closing times for the attractions. These are the opening hours of the attraction; For path conduct The second independent randomized simulation, the first The second simulation yielded the terminal time deviation. Confidence level The quantile value below is denoted as The CVaR risk indicator is: ； Will to Normalization to Then, construct the weighted optimization function: ， ； in, Let the weight of the i-th item be , For the i-th normalization objective, This represents the feasible region, which is composed of time window constraints, event insertion constraints, budget constraints, node uniqueness constraints, and date constraints.

8. The hierarchical semantic guidance-based tourism route optimization method according to claim 1, characterized in that, In step 6, the tourist profile consists of a set of preferred topics, a set of preferred subtopics, a set of preferred tags, a set of avoidance tags, a set of must-see attractions, and a crowd avoidance coefficient; a macro-semantic matching score is calculated for any candidate attraction. ， ； in, A collection of must-see attractions For preferred topic sets, For preferred subtopic sets, and These are respectively the labeled hit items and the avoided hit items. To normalize semantic weights, For indicator functions, and They are divided into primary and secondary themes for the scenic spot. A collection of attraction tags. For a set of preference labels, To avoid tag sets, This is the congestion avoidance coefficient. Due to the crowding at the attractions, and These are respectively marked as disturbance-friendly and supply-friendly. All are macro-level semantic weights; Filtered semantic candidate set for: ； This is the semantic filtering threshold; Main topic definition for: ； A collection of candidate attractions by theme Any topic in This is a collection of candidate scenic spot themes.

9. The hierarchical semantic guidance-based tourism route optimization method according to claim 1, characterized in that, In step 7, the local adjacency coordination degree of adjacent attractions is defined. : ； and These are the weights for topic compatibility and tag similarity, respectively. For actual tourist attractions and Theme compatibility, For actual tourist attractions and Tag similarity; ； ； in, For the first One candidate scenic spot as a primary theme. For the first A collection of attraction tags, This represents a set of compatible themes that have a natural connection to the tour. This refers to the cross-topic compatibility coefficient. Path adjacency uncoordination item for: ； Definition of the first Daily time load Normalized load and multi-day average load : ， ； Cross-day load imbalance term is : ； The structural layer embeds adjacency miscoordination and cross-day load imbalance terms into the path evaluation. The corrected time efficiency and risk indicators are as follows: ， ； in, To correct the lower limit parameter, ~ Structural penalty weights; modified time efficiency index Used to replace the time utilization rate indicator in step 5 Revised risk indicators Used to replace the tail execution risk indicator in step 5 and participate in the comprehensive evaluation of subsequent paths; , , , and After normalization, construct the weighted optimization function with modified structure: ； These are the weights of each objective.

10. The hierarchical semantic guidance-based tourism route optimization method according to claim 1, characterized in that, In step 8, node cascading risk It consists of local delay pressure, the degree of subsequent time window compression, and local loss terms: ； in, This indicates localized delay pressure. Indicates the degree of compression of the subsequent time window. This indicates a local semantic quality loss. Risk portfolio weights; Local action triggering variables for: ； As the risk trigger threshold, For the current node The set of feasible event candidates in the vicinity that meet the constraints of time window, date, and single-day activity period. It is an empty set; when At that time, construct a set of actionable actions. : ， in, , , , These represent event insertion, dwell compression, node replacement, and node skipping, respectively. Macro-level score of candidate actions Structural layer score and micro-level scores They are represented as follows: ， ， ； in, This represents the average semantic value of nodes in a local segment after the action is executed. The main theme is consistent with the following items: ； in, To execute candidate actions The set of attractions retained or introduced in the current local segment. Indicates tourist attractions The first-level topic, It is an indicator function; and The incremental structural degradation caused by the action. To mitigate the benefits of cascading risk, For event requirement fulfillment items, The cost of delaying the timing and Weights for structural degradation penalties; ~ As a weight for the micro-level score; Hierarchical semantic comprehensive score of candidate actions for: ； in, To normalize residual risk, ~ The final selected local action result is based on the overall weighting. ； This is the result of a local action.

Images