Multi-intention conflict resolution oriented tourism multi-agent collaborative planning system

Abstract

This invention relates to the field of smart tourism technology and discloses a multi-agent collaborative planning system for tourism aimed at resolving multi-intention conflicts. The system includes: constructing a cabin time series table based on real-time operating speed and cabin spacing; calculating equipment availability by combining environmental interference data; extracting the tour willingness level reflecting individual experience demands and the maintenance pressure value reflecting mechanical evacuation bottlenecks; extrapolating the descent probability value to obtain the predicted load for future time and space; calculating the tour deviation value and the carrying capacity deviation value based on these, and constructing a conflict assessment value measuring the degree of deviation using the adjustment direction of both; solving the joint adjustment quantity from the deviation direction to collaboratively update the current itinerary value; and finally outputting a closed-loop execution plan that minimizes the sum of all assessment items. This invention effectively balances the cross-time-period contradiction between tourists' subjective tour demands and the safe operation of mechanical equipment.

Description

Technical Field

[0001] This invention relates to the field of smart tourism technology, and more specifically, to a multi-agent collaborative planning system for tourism aimed at resolving multi-intention conflicts. Background Technology

[0002] In the daily operation and maintenance of mountainous scenic areas, the circulating passenger cableway is the core link for tourists to ascend and descend the mountain. However, unlike the continuous and highly tolerant flat terrain transportation network, the capacity of the cableway system is extremely limited and consists of discrete cabins. Constrained by the passenger boarding and alighting rhythm within the station, the interval between cable car departures, and complex environmental factors such as high-altitude crosswinds and low-temperature ice and snow, the actual carrying capacity of the cableway is highly time-varying. In the event of severe operating conditions, it may be forced to reduce speed or even temporarily suspend operation. Currently, mainstream smart tourism platforms, when providing route guidance, typically treat the cable car as just a regular passageway, primarily focusing on planning the best current tour route for tourists to satisfy their sightseeing or photo-taking desires.

[0003] This one-way recommendation logic has problems in actual operation: when the platform guides tourists to high-altitude areas in the morning, it effectively locks in an implicit transportation liability for these tourists who must take the cable car down the mountain in the afternoon or evening. As time goes on, tourists' desire to stay and watch the sunset or sea of ​​clouds clashes sharply with the scenic area's operational and evacuation efforts to clear queues before dark or when strong winds intensify. The existing one-dimensional recommendation system lacks the ability to anticipate future trends, failing to project the current uphill passenger flow into future downhill pressure, and unable to dynamically balance tourists' subjective preferences with the real-time operating conditions of the cable car machinery. This easily leads to severe passenger congestion and safety risks at the cable car stations during the evening peak hours. Summary of the Invention

[0004] This invention provides a multi-agent collaborative planning system for tourism aimed at resolving multi-intention conflict, thus solving the technical problems mentioned in the background art.

[0005] This invention provides a multi-agent collaborative planning system for tourism aimed at resolving multi-intent conflicts, applied to a circular passenger cableway system in a mountainous scenic area. The circular passenger cableway system includes upper and lower stations, drive wheels, slewing wheels, traction cables, load-bearing cables, grips, gondolas, towers, in-station deceleration tracks, in-station acceleration tracks, turnstiles, and platform queuing areas. The system is configured to execute: A cabin timing table is constructed using real-time operating speed and cabin spacing to extract the physical supply timing that characterizes discrete mechanical transport capacity. Mechanical operating parameters are obtained to calculate the capacity reduction rate, which characterizes the degree of environmental disturbance, and the capacity reduction rate is used as the equipment availability. Based on tourist preference values ​​and attraction status values, a visit intention value reflecting individual experience demands is generated; combined with regional passenger flow and current queuing volume, an operation and maintenance pressure value reflecting mechanical load constraints is generated. Based on the distribution law of stay, the probability value of going down the mountain is deduced, and then the return flow prediction value is extracted to characterize the future spatiotemporal liabilities; The tour deviation value is calculated based on the tour intention value, the carrying capacity deviation value is calculated based on the operation and maintenance pressure value and the return flow prediction value, the adjustment direction of the tour deviation value and the adjustment direction of the carrying capacity deviation value are extracted, and the conflict assessment value that measures the degree of deviation from intention is constructed using the adjustment direction of the tour deviation value and the adjustment direction of the carrying capacity deviation value. The joint adjustment amount is calculated using the adjustment direction amount of the tour deviation value and the adjustment direction amount of the load deviation value, so as to update the current travel value in a coordinated manner; The final itinerary is output by selecting the scheme that minimizes the sum of the tour deviation value, the carrying capacity deviation value, and the conflict assessment value to achieve a compromise under physical constraints, and the descent probability value is corrected by using real-time observation data in a closed loop.

[0006] The beneficial effects of this invention include: by directly mapping real-time environmental wind load, gondola spacing and other mechanical operating conditions to the future dynamic capacity boundary, and by converting the current uphill trajectory of tourists into the future downhill load in advance, this invention enables tourist experience orientation and equipment operation and maintenance orientation to compromise and correct each other on the same time axis; this effectively avoids congestion caused by sudden weather slowdown or peak tourist season overlap, and maximizes the preservation of tourists' willingness to play while ensuring the safety boundary of mechanical operation. Attached Figure Description

[0007] Figure 1 This is a flowchart of the multi-agent collaborative planning system for tourism aimed at resolving multi-intention conflicts, as described in this invention. Detailed Implementation

[0008] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0009] like Figure 1As shown, a multi-agent collaborative planning system for tourism aimed at resolving multi-intention conflicts is applied to a circular passenger cableway system in a mountainous scenic area. This system includes upper and lower stations, drive wheels, slewing wheels, traction cables, load-bearing cables, grips, gondolas, towers, in-station deceleration tracks, in-station acceleration tracks, turnstiles, and platform queuing areas. The system is configured to execute the following: A cabin timing table is constructed using real-time operating speed and cabin spacing to extract the physical supply timing that characterizes discrete mechanical transport capacity. Mechanical operating parameters are obtained to calculate the capacity reduction rate, which characterizes the degree of environmental disturbance, and the capacity reduction rate is used as the equipment availability. Based on tourist preference values ​​and attraction status values, a visit intention value reflecting individual experience demands is generated; combined with regional passenger flow and current queuing volume, an operation and maintenance pressure value reflecting mechanical load constraints is generated. Based on the distribution law of stay, the probability value of going down the mountain is deduced, and then the return flow prediction value is extracted to characterize the future spatiotemporal liabilities; The tour deviation value is calculated based on the tour intention value, the carrying capacity deviation value is calculated based on the operation and maintenance pressure value and the return flow prediction value, the adjustment direction of the tour deviation value and the adjustment direction of the carrying capacity deviation value are extracted, and the conflict assessment value that measures the degree of deviation from intention is constructed using the adjustment direction of the tour deviation value and the adjustment direction of the carrying capacity deviation value. The joint adjustment amount is calculated using the adjustment direction amount of the tour deviation value and the adjustment direction amount of the load deviation value, so as to update the current travel value in a coordinated manner; The final itinerary is output by selecting the scheme that minimizes the sum of the tour deviation value, the carrying capacity deviation value, and the conflict assessment value to achieve a compromise under physical constraints, and the descent probability value is corrected by using real-time observation data in a closed loop.

[0010] This embodiment provides a multi-agent collaborative planning system for tourism aimed at resolving multi-intent conflicts, applied to a circular passenger cableway system in a mountainous scenic area. The circular passenger cableway system includes upper and lower stations, drive wheels, slewing wheels, traction cables, load-bearing cables, grips, gondolas, towers, in-station deceleration tracks, in-station acceleration tracks, turnstiles, and platform queuing areas. The system comprises three types of agents: tourist agents, cableway agents, and a global coordination agent. The tourist agent is deployed on tourist mobile terminals, responsible for collecting tourist preference data, generating initial itineraries, receiving adjustment instructions, and updating itineraries. The cableway agent is deployed at the cableway control center, responsible for collecting cableway operating parameters, calculating equipment availability, monitoring queuing volume, and reporting maintenance pressure values. The global coordination agent is deployed on the scenic area's cloud server, responsible for integrating multi-source data, calculating return flow prediction values ​​and conflict assessment values, generating joint adjustment quantities, and issuing itinerary update instructions. The agents communicate using HTTP / HTTPS protocols, with data transmission in JSON format, and a communication frequency of once per minute. The system updates data every minute, and all core calculations are performed every minute.

[0011] S201: Based on the ratio between real-time operating speed and cabin spacing, time-domain mapping is accumulated within the historical operating period to extract cabin numbers that represent the physical cabin sequence.

[0012] The calculation formula is: ; in, The current time is expressed in seconds. The initial moment is defined as the moment when the cableway is first started and reaches its rated operating speed on the same day, and is synchronized in real time by the cableway control center system. This is the time-domain mapping integral variable for the historical runtime period, in seconds. The real-time operating speed of the historical mapping is measured in meters per second and is collected by the encoder of the cableway drive wheel at a sampling period of 1 second. The distance between gondolas is in meters and is a fixed parameter of the cableway equipment, determined by the manufacturer's instruction manual. The cabin class number is a non-negative integer. The integration operation uses the trapezoidal integration method, with a sampling period of 1 second. The cabin class number is obtained by rounding down the integration result using the floor function.

[0013] Integral variable From the initial moment up to the current moment The continuous historical time markers are used to traverse all historical periods of cableway operation, enabling continuous time-domain accumulation of the number of gondolas passing through. The trapezoidal integration method for integration is specifically implemented as follows: the average real-time operating speed of two adjacent sampling times is multiplied by the sampling period of 1 second to obtain the increment of gondola passing distance within that period, which is then divided by the gondola spacing to obtain the increment of the number of gondolas. This is accumulated over time and then rounded down.

[0014] S202: Combining proportional relationships, single-cabin passenger capacity, and equipment availability, perform flux integration in the time domain dimension within the time sequence space corresponding to the cabin number to quantify the total transport capacity during a specific physical time phase when transport capacity is in a contracted state.

[0015] The calculation formula is: ; in, It is the phase number in the time sequence space, which is a non-negative integer and increments from 0. A time period within a time sequence space is defined as starting from Instant arrival Seconds, the instantaneous phase granularity is 15 minutes. This represents the current real-time operating speed, expressed in meters per second. This refers to the passenger capacity of a single cabin, which is the rated passenger capacity of the cableway. It is determined by the factory parameters of the equipment and is usually set to 6, 8, or 10 people. This represents equipment availability, with a value ranging from 0 to 1. This represents the total transport capacity for the time period, expressed in person. The integration operation uses the trapezoidal integration method, with a sampling period of 1 second, and maintains timing alignment with S201.

[0016] S301: Extract crosswind speed, gondola swing angle, swing angle change rate, cable span, cable tension, and cable linear density.

[0017] Crosswind speed is collected by ultrasonic anemometers installed at the top of each tower and at each station platform, with a sampling period of 1 second. The maximum value of all sensor measurements is taken as the current crosswind speed. The gondola swing angle and rate of change of swing angle are collected by MEMS tilt sensors installed at the center of the top of each gondola, with a sampling period of 1 second. The maximum value of all gondola measurements is taken as the current gondola swing angle and rate of change of swing angle. Cable span is the fixed distance between adjacent towers, in meters. For multi-span cableways, the maximum span of the currently operating section is taken as the cable span value. Cable tension is collected by tension sensors installed at the drive wheels, with a sampling period of 1 second, in Newtons (N). Cable linear density is the actual mass per unit length of the cable containing grease, in kilograms per meter, determined by the equipment's factory parameters.

[0018] S302: Construct an aerodynamic response mapping model to isolate the mechanical disturbance values ​​caused by environmental wind loads, and introduce cable tension and cable linear density into the calculation to eliminate cable resonance interference.

[0019] The calculation formula is: ; in, Crosswind speed, measured in meters per second. The swing angle of the gondola is expressed in radians. This represents the rate of change of the pendulum angle, expressed in radians per second. This refers to the cable span, in meters. The tension is measured in Newtons (N). This refers to the cable linear density, expressed in kilograms per meter. , , , For each term, there are response mapping coefficients, where The dimension is square seconds per square meter. Dimensionless For square seconds, It is dimensionless. The mechanical disturbance value is represented by a dimensionless parameter. The response mapping coefficient is obtained by performing multiple linear regression fitting between historical cableway operation data and capacity attenuation data under corresponding operating conditions. The training dataset has no fewer than 10,000 samples, covering the entire operating range of wind speed (0-25 m / s), gondola swing angle (0-15 degrees), and swing angle change rate (0-5 degrees / s). The last term of the formula introduces a parameter related to the cable's natural frequency to correct for the influence of cable resonance on the mechanical disturbance calculation and eliminate errors caused by resonance interference.

[0020] The multiple linear regression training of the response mapping coefficients uses mean squared error as the loss function, and the optimization algorithm employs gradient descent. The initial learning rate is 0.01, and the number of iterations is 1000. Training is terminated early when the change in the loss function is less than 1e-6. The input features of the regression model are the square of the crosswind speed, the square of the gondola's swing angle, the square of the rate of change of the swing angle, and the square of the cable's natural frequency. The output label is the negative of the natural logarithm of the ratio of the actual carrying capacity to the rated carrying capacity under the corresponding operating condition, maintaining consistency with the exponential mapping relationship of the carrying capacity reduction rate.

[0021] S303: Establish an exponential decay mapping relationship between mechanical disturbance values ​​and system standard disturbance values ​​to extract the energy reduction rate that characterizes the degree of limitation of system power output under severe working conditions.

[0022] The calculation formula is: ; in, The standard disturbance value of the system is a dimensionless parameter, defined as the average value of mechanical disturbance values ​​collected after the cableway has been running unloaded and at its rated operating speed for 30 minutes in a windless environment. It is determined through on-site commissioning. This is the capacity reduction rate, ranging from 0 to 1. A smaller value indicates more severe environmental interference and a lower actual usable capacity of the cableway. This capacity reduction rate represents the equipment availability. When the capacity reduction rate is less than 0.2, the cableway system triggers a mandatory shutdown warning.

[0023] S401: Map tourist preference values ​​to attraction status values ​​using a spatial vector, and remove the tour attenuation components constrained by time tolerance, relative time consumption, fatigue tolerance, and relative fatigue to obtain the tour willingness degree that reflects individual subjective tendencies.

[0024] The calculation formula is: ; in, The tourist preference values ​​are represented by a 5-dimensional column vector, with each dimension corresponding to natural scenery, cultural landscapes, entertainment projects, catering services, and shopping experiences. Each dimension ranges from 0 to 1 and is generated using a collaborative filtering algorithm based on tourists' historical browsing, collection, and review data on the scenic area's official APP. New users have a default value of 0.5 for each dimension. The status value of the attraction is a 5-dimensional row vector, with each dimension corresponding to a visitor preference value. Each dimension ranges from 0 to 1. The dimensions of natural scenery and cultural landscape are determined by the attraction rating, while the dimensions of entertainment, catering services, and shopping experience are determined by the real-time business status and queuing situation. This is the time tolerance, ranging from 0.5 to 2. It is obtained by fitting the average of the ratios of actual tour time to estimated tour time in the tourist's historical itinerary. The default value for new users is 1. The relative time spent is the ratio of the estimated tour time to the average tour time of the attraction. The estimated tour time is obtained by multiplying the average tour time of the attraction by the congestion correction factor. The congestion correction factor ranges from 1 to 3, with the factor increasing as the congestion level increases. The fatigue tolerance value ranges from 0.5 to 2 and is obtained by fitting the average of the ratios of actual walking distance to average walking distance in the tourist's historical itinerary. The default value for new users is 1. The relative fatigue level is the ratio of the expected physical exertion to the average physical exertion at the scenic spot. The expected physical exertion is calculated by combining the altitude difference, walking distance, and crowding level of the scenic spot. Physical exertion increases by 0.1 units for every 100-meter increase in altitude, 0.05 units for every 100-meter increase in walking distance, and 0.05 units for every 0.1 increase in crowding level. The value represents the willingness to visit the attraction, ranging from negative infinity to positive infinity. The larger the value, the stronger the tourist's willingness to visit the attraction.

[0025] Tourist preference values ​​are generated using a user-based collaborative filtering algorithm. The input is a matrix of attraction ratings from all users on the scenic area's official app, with ratings ranging from 1 to 5. By calculating the cosine similarity between users, the ratings of the 20 users with the highest similarity are selected and weighted averaged to obtain a 5-dimensional preference vector for the target user. The congestion dimension of the attraction status value is obtained by dividing the regional visitor flow by the attraction's maximum carrying capacity, which is determined by the scenic area's management department.

[0026] S402: Based on regional passenger flow and current queuing volume, combined with the total capacity of the corresponding cabin number for the time period, calculate the platform congestion load, collect full-domain features along the candidate cableway topology network, and extract the operation and maintenance pressure value reflecting the bottleneck of mechanical evacuation.

[0027] The calculation formula is: ; in, The regional visitor flow, measured in individuals, is calculated using a Kalman filter algorithm by fusing data from the scenic area's access control system, video surveillance, and mobile phone signaling. The access control system data has a weight of 0.5, the video surveillance data has a weight of 0.3, and the mobile phone signaling data has a weight of 0.2. The current queue length is expressed in person and is obtained by averaging the counts from the cable car station turnstiles and the video passenger flow statistics. This refers to the cabin number. This represents the total transport capacity for the corresponding time period, expressed in person. The candidate cableway topology network is defined as the set of all cableways whose walking time from attraction v is no more than 60 minutes. This represents the operational pressure value, ranging from 0 to positive infinity. A larger value indicates greater mechanical evacuation pressure in the area. The overall feature aggregation directly sums the platform congestion loads of all cableways, without considering cableway priority.

[0028] In the Kalman filter fusion algorithm for regional passenger flow, the state variable is the actual regional passenger flow, and the state equation is: ,in The process noise has a covariance of 0.1. The observation equation is: ,in The observation matrix has coefficients of 0.5, 0.3, and 0.2 for the three observations. To observe noise, the noise covariance for access control data is 0.05, for video surveillance data it is 0.1, and for mobile phone signaling data it is 0.15.

[0029] S501: Based on the total walking time and cable car time, the time domain is phase-stripped to extract the remaining time representing the effective tour window.

[0030] The total walking time is the time required for a tourist to walk from their current location to the target cable car station, in seconds. This is achieved through improved A... The path planning algorithm calculates the walking speed by considering road gradient and congestion. For every 5-degree increase in gradient, walking speed decreases by 10%; for every 0.1 degree increase in congestion, walking speed decreases by 5%. The cable car travel time, measured in seconds, is the time taken to travel from the upper station to the lower station, and is equal to the distance between the upper and lower stations divided by the real-time operating speed. The cable car's closing time is dynamically adjusted. Under normal circumstances, it follows the closing time announced by the scenic area. If the crosswind speed exceeds 15 meters per second and lasts for more than 30 minutes, the closing time will be moved forward to 2 hours after the current time. If the queue at the cable car platform exceeds 80% of the maximum capacity and lasts for more than 1 hour, the closing time will be postponed by 30 minutes, up to a maximum of 1 hour before dark. The remaining time of the valid tour window is equal to the cable car closing time minus the current time, plus the total walking time and cable car time, in seconds.

[0031] S502: Input the remaining time into the stay distribution law, combine it with the individual tour rhythm ratio to perform multimodal probability distribution aggregation, and deduce the probability value of descending the mountain that represents the tourist's future stay characteristics.

[0032] The calculation formula is: ; in, This is a time-domain variable, with the unit being seconds. The extreme value of the ascent time is defined as the actual time it takes for a tourist to ride the cable car up the mountain, in seconds, and is obtained from the cable car gate passage record. The time taken for the cable car ride is in seconds. The total time taken for the walk is in seconds. To determine the duration distribution law, a log-normal distribution was adopted, with each attraction configured separately. The mean and variance of the distribution were obtained by fitting the actual visitor stay time data of that attraction over the past 3 months. The tour rhythm is categorized into three levels, with values ​​of 1, 2, and 3, corresponding to fast, medium, and slow rhythms, respectively. The corresponding tour pace percentage ranges from 0 to 1, with the sum of the percentages for the three categories being 1. This percentage is determined using K-means clustering based on the average walking speed, average dwell time, and number of attractions visited in the tourist's historical itinerary. The cluster centers are: Fast pace (average walking speed > 4 km / h, average dwell time < 15 minutes per attraction, > 8 attractions visited per day); Medium pace (average walking speed 2-4 km / h, average dwell time 15-30 minutes per attraction, 5-8 attractions visited per day); Slow pace (average walking speed < 2 km / h, average dwell time > 30 minutes per attraction, < 5 attractions visited per day). For new tourists without historical data, the default tour pace percentage is 0.3 for fast pace, 0.5 for medium pace, and 0.2 for slow pace. These are candidate itineraries. Let be the probability value for descending the mountain, and be the probability density function, expressed in reciprocals per second. Multimodal probability distribution aggregation weights and sums the stay distributions corresponding to different tour paces according to their proportions, yielding the comprehensive probability distribution for tourists descending the mountain.

[0033] The log-normal probability density function used in the dwell distribution law is: in, The time spent by tourists is expressed in seconds. The logarithmic mean is... The two parameters, denoted as logarithmic standard deviation, were obtained through maximum likelihood estimation of historical dwell time data at the attractions. K-means clustering of the tour rhythm used Euclidean distance as a similarity metric. Initial cluster centers were selected using the k-means++ algorithm, with 100 cluster iterations. The process terminated when the change in cluster centers was less than 1e-3.

[0034] S503: The probability value of going downhill is accumulated in segments within the time-series space to quantify and extract the return flow prediction value of a single tourist for the implicit occupation of future downhill capacity, and then the total return flow prediction is obtained by global aggregation.

[0035] The calculation formula is: ; ; in, The time segments within the time space are perfectly aligned with the phase of S202, meaning each segment is 15 minutes long. Let be the predicted return flow value for a single tourist, and be a dimensionless parameter representing the probability that the tourist will take the cable car r down the mountain at the k-th time phase. Identify and label tourists with numbers. The total return flow prediction, in units of people, represents the expected total number of downhill passengers for cable car r at the k-th time phase. The passenger range for the return flow prediction excludes pedestrians descending the mountain. Pedestrians are identified as those who have been moving downhill continuously for more than 30 minutes, whose movement path matches the downhill walking route by more than 80%, and whose movement speed is between 2 and 5 kilometers per hour. The integration calculation uses the trapezoidal integration method with an integration step size of 1 minute.

[0036] S601: Perform inverse gradient mapping on the travel intention degree associated with candidate itinerary nodes to extract travel deviation values.

[0037] The calculation formula is: ; in, is the spatiotemporal identifier associated with the candidate itinerary node, where v is the attraction number and t is the node arrival time in seconds. To indicate willingness to visit. This represents the travel deviation value, a dimensionless parameter. A larger value indicates a greater deviation between the candidate itinerary and the tourist's subjective travel intentions. The summation range includes all scenic spots in the candidate itinerary, but excludes cable car rides.

[0038] S602: Based on the return flow forecast, total return flow forecast and expected queuing volume, construct downlink load characteristics, and perform spatiotemporal conversion limited by the total capacity of the time period to output the load deviation value.

[0039] The calculation formula is: ; in, This represents the predicted return flow value for a single tourist. For total return flow forecast, the unit is person. The estimated queue size, in units of people, is obtained using the LSTM time series prediction algorithm. The input features are the queue size over the past 24 hours, current passenger flow, weather conditions, and holiday information. The output is the estimated queue size for each 15-minute time phase in the next 4 hours. Total capacity for a given period, in units of people. The load-bearing deviation value is a dimensionless parameter; a larger value indicates a greater occupancy of the cableway's capacity by the candidate route. The summation range is limited to the time phases of the cableway accessible to tourists and the next 4 hours.

[0040] The LSTM time series prediction model for estimated queue length consists of one input layer, two hidden layers, and one output layer. The input layer has a dimension of 4, corresponding to queue length data for the past four 15-minute time periods. Each hidden layer contains 64 neurons, using ReLU activation. The output layer has a dimension of 16, corresponding to the estimated queue length for 16 15-minute time periods within the next four hours. The model training dataset includes cable car queue length data, passenger flow data, weather data, and holiday data from the past year. The training batch size is 32, the number of iterations is 50, and the Adam optimization algorithm is used with a learning rate of 0.001.

[0041] S603: Extract the spatial geometric direction correlation features between the adjustment direction of the tour deviation value and the adjustment direction of the load deviation value, eliminate consistency intention interference by amplitude attenuation, and couple the divergent component with the load deviation value to output a conflict assessment value that measures the risk of mechanical overload.

[0042] The calculation formula is: ; in, is the adjustment direction of the tour deviation value, and is the gradient vector of the tour deviation value with respect to the candidate itinerary time parameters. Each dimension corresponds to the arrival time of a itinerary node, in reciprocals of seconds. is the adjustment direction of the carrying deviation value, and is the gradient vector of the carrying deviation value with respect to the candidate travel time parameter. Its dimension is the same as the adjustment direction of the travel deviation value, and its unit is the reciprocal of seconds. The gradient calculation uses the central difference method, with a finite difference step size of 60 seconds. When the magnitude of the gradient vector is less than 1e-6, it is set to 1e-6 to avoid numerical instability. The value represents the conflict assessment and is a dimensionless parameter. A larger value indicates a more intense conflict between tourists' willingness to visit and the cableway's carrying capacity, and a higher risk of mechanical overload. In the formula, the numerator is the dot product of two gradient vectors, the denominator is the product of two moduli, and the ratio is the cosine similarity of the two vectors, ranging from -1 to 1. When the cosine similarity is negative, it indicates that the two adjustment directions are opposite, and there is an intentional conflict.

[0043] S701: Calculate the modal distribution of the difference between the adjustment direction of the tour deviation value and the adjustment direction of the load deviation value in the parameter space, and use the threshold truncation operation to constrain it to a predetermined normalization space to eliminate extreme value divergence and generate a dynamic adjustment ratio.

[0044] The calculation formula is: ; in, This refers to the adjustment direction of the bearing deviation value. This refers to the adjustment direction of the tour deviation value. This is a dynamic adjustment ratio, with a value ranging from 0 to 1. When the denominator... When the value is less than 1e-12, the dynamic adjustment ratio is set to 0.5. At this point, the two adjustment directions are completely consistent, and the weights are evenly distributed. The threshold truncation operation restricts the calculation results to between 0 and 1, ensuring the validity of the physical meaning of the adjustment weights.

[0045] S702: Using the dynamic adjustment ratio as the allocation weight, the adjustment direction of the tour deviation value and the adjustment direction of the bearing deviation value are superimposed by a linear convex combination to eliminate the trap of unilateral extreme value programming and output the joint adjustment amount of the fusion dimension.

[0046] The calculation formula is: ; in, The joint adjustment quantity has the same dimension as the adjustment direction quantity, and the unit is the reciprocal of the second. The linear convex combination superposition balances the weights of the two adjustment directions through a dynamic adjustment ratio. When the dynamic adjustment ratio is close to 1, priority is given to optimizing the tourist experience; when the dynamic adjustment ratio is close to 0, priority is given to ensuring the cableway's load-bearing safety. This method avoids extreme cases caused by single-objective optimization and achieves a dynamic trade-off between the two objectives.

[0047] S703: Utilizes the coordinated driving step size of single adjustment and joint adjustment to update the spatiotemporal state of the node sequence for the current travel value.

[0048] The calculation formula is: ; in, The updated current travel value is a vector, where each element corresponds to the arrival time of a travel node, in seconds, starting from the initial time. Start timing. This is the current trip value before the update. This is a single adjustment amount, in seconds, with an initial value of 600 seconds. It decays to 0.9 times its original value every 10 iterations. . This is the joint adjustment variable. The iteration termination condition is: the maximum number of iterations reaches 50, or the change in the global cost function is less than 1e-3.

[0049] The node sequence of the current itinerary value follows geographical logic constraints, allowing only geographically adjacent or directly accessible attraction nodes to exchange their order. Exchanges of non-adjacent nodes are considered infeasible. During itinerary updates, if adjusting the arrival time of a node results in the arrival time of subsequent nodes being earlier than the departure time of the current node, the arrival time of the subsequent nodes will be automatically postponed to after the departure time of the current node. The departure time of a node is equal to its arrival time plus the estimated tour time for that node.

[0050] S801: Construct a global cost penalty function that includes tour deviation value, carrying deviation value and conflict assessment value, and iterate through the candidate trips associated with the updated current trip value to select the trips mapped to the cost minimum point as the final trips.

[0051] The calculation formula is: ; in, This is the final itinerary. The traversal optimization operation represents the search for the minimum cost. The global cost penalty function comprehensively considers tourist experience, cableway capacity, and the conflict between the two. The traversal optimization uses a greedy algorithm combined with a local search strategy. The greedy algorithm selects attractions from high to low visitor willingness, prioritizing attractions with high visitor willingness, while also considering the cableway's capacity pressure. The neighborhood of the local search is defined as follows: the arrival time of each trip node can be adjusted within ±1800 seconds, and the order of two adjacent nodes can be swapped. The candidate itinerary is generated within the neighborhood of the initial itinerary, i.e., all feasible itineraries generated after the above adjustments to the initial itinerary. The final itinerary is pushed to tourists through the scenic area's official APP. Tourists can accept or reject itinerary adjustments. If they reject, the system will generate a suboptimal solution for selection. Tourists' satisfaction ratings for the itinerary will be used to subsequently optimize the tourist preference model.

[0052] The method for generating the suboptimal solution is as follows: Based on the optimal solution, the arrival time of each attraction is adjusted by ±900 seconds to generate 10 candidate solutions. The global cost function value of each solution is calculated, and the solution with the second smallest cost function value and a difference of no more than 5% from the optimal solution is selected as the suboptimal solution. Tourist satisfaction ratings are based on a 5-point scale, and the rating results will be used as update data for the collaborative filtering algorithm, updating tourist preference values ​​every 24 hours.

[0053] S802: Collect real-time observation data of tourist displacement status to construct measured matching degree conditions. Use these conditions as prior intervention to perform posterior probability filtering and time-domain normalization correction on the updated downhill probability value in the entire time domain, so as to dynamically eliminate the interference of environmental noise on the inference of the updated downhill probability value.

[0054] The calculation formula is: ; in, To monitor data in real time, including tourists' mobile phone GPS coordinates, scenic area access control card swipe records, and cable car turnstile passage records, the GPS coordinate sampling period is 60 seconds, and access control and turnstile records are uploaded in real time. To measure the matching degree, the Gaussian kernel function was used for calculation. The bandwidth of the kernel function was determined by cross-validation. 1000 samples were randomly selected from the tourist displacement data of the past 7 days for cross-validation, and the bandwidth value that minimized the mean square error was selected. and All values ​​are the downhill probability values ​​before the update. is the independent variable of the posterior filter integral over the entire time domain, in seconds. This represents the updated probability of descending the mountain. The time range of the discrete integral is from the current moment to the time the cable car stops operating, with an integration step size of 60 seconds. Time-domain normalization correction ensures that the integral sum of the updated probability distribution over the entire time domain is 1, eliminating probability distribution distortion caused by noise. When the deviation between the tourist's actual location and the recommended location exceeds 1 kilometer, or the deviation between the actual arrival time and the recommended arrival time exceeds 1800 seconds, the system immediately triggers a replanning process to recalculate the tourist's optimal itinerary.

[0055] It should be noted that when handling conflicts in multiple tourist itineraries, the global coordination agent employs a priority-weighted adjustment mechanism. Tourist priorities are divided into three levels: Level 1 prioritizes seniors over 65, children under 12, and people with disabilities; Level 2 prioritizes group tourists; and Level 3 prioritizes individual tourists. When the cableway's load exceeds 90%, the dynamic adjustment ratio will be multiplied by a priority coefficient: 1.2 for Level 1, 1.0 for Level 2, and 0.8 for Level 3, prioritizing the tour experience of high-priority tourists.

[0056] It should be noted that when the wind speed sensor malfunctions, the wind speed data from the nearest weather station within 5 kilometers of the scenic area is multiplied by a correction factor of 1.2 as a substitute value. When the gondola tilt sensor malfunctions, the average tilt angle of other normal gondolas on the same cableway is used as a substitute value. When mobile phone signaling data is missing for more than 30 minutes, the Kalman filter algorithm automatically adjusts the weight of mobile phone signaling data to 0, the weight of access control data to 0.6, and the weight of video surveillance data to 0.4.

[0057] It should be noted that the steps for the tourist agent to generate the initial itinerary are as follows: First, select the attractions that are open, sort them from high to low according to the tourist's preference value, and add the attractions to the itinerary in turn. At the same time, calculate the cumulative tour time and walking distance. When the cumulative tour time reaches 80% of the opening time of the scenic spot or the cumulative walking distance reaches the maximum walking distance corresponding to the tourist's fatigue tolerance, stop adding attractions. Finally, adjust the itinerary order according to the shortest path between attractions to generate the initial itinerary.

[0058] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A multi-agent collaborative planning system for tourism aimed at resolving multi-intention conflicts, applied to a circular passenger cableway system in a mountainous scenic area. The circular passenger cableway system includes upper and lower stations, drive wheels, slewing wheels, traction cables, load-bearing cables, grips, gondolas, towers, in-station deceleration tracks, in-station acceleration tracks, turnstiles, and platform queuing areas. Its features include... Configured for execution: A cabin timing table is constructed using real-time operating speed and cabin spacing to extract the physical supply timing that characterizes discrete mechanical transport capacity. Mechanical operating parameters are obtained to calculate the capacity reduction rate, which characterizes the degree of environmental disturbance, and the capacity reduction rate is used as the equipment availability. Based on tourist preference values ​​and attraction status values, a visit intention value reflecting individual experience demands is generated; combined with regional passenger flow and current queuing volume, an operation and maintenance pressure value reflecting mechanical load constraints is generated. Based on the distribution law of stay, the probability value of going down the mountain is deduced, and then the return flow prediction value is extracted to characterize the future spatiotemporal liabilities; The tour deviation value is calculated based on the tour intention value, the carrying capacity deviation value is calculated based on the operation and maintenance pressure value and the return flow prediction value, the adjustment direction of the tour deviation value and the adjustment direction of the carrying capacity deviation value are extracted, and the conflict assessment value that measures the degree of deviation from intention is constructed using the adjustment direction of the tour deviation value and the adjustment direction of the carrying capacity deviation value. The joint adjustment amount is calculated using the adjustment direction amount of the tour deviation value and the adjustment direction amount of the load deviation value, so as to update the current travel value in a coordinated manner; The final itinerary is output by selecting the scheme that minimizes the sum of the tour deviation value, the carrying capacity deviation value, and the conflict assessment value to achieve a compromise under physical constraints, and the descent probability value is corrected by using real-time observation data in a closed loop.

2. The multi-agent collaborative planning system for tourism aimed at resolving multi-intent conflicts according to claim 1, characterized in that, The method of constructing a cabin timing table using real-time operating speed and cabin spacing includes: Based on the ratio between the real-time operating speed and the cabin spacing, time-domain mapping accumulation is performed within the historical operating period to extract the cabin number representing the physical cabin sequence; Combining the aforementioned proportional relationship, single-cabin passenger capacity, and equipment availability, a time-domain flux integral is performed within the time-series space corresponding to the cabin number to quantify the total transport capacity during a specific physical time phase when transport capacity is in a contracted state.

3. The multi-agent collaborative planning system for tourism aimed at resolving multi-intent conflicts according to claim 2, characterized in that, The process of obtaining mechanical operating parameters to calculate the capacity reduction rate used to characterize the degree of environmental disturbance includes: Extract crosswind speed, gondola swing angle, swing angle change rate, cable span, cable tension, and cable linear density; An aerodynamic response mapping model is constructed to isolate the mechanical disturbance values ​​caused by environmental wind loads, and the cable tension and the cable linear density are introduced into the calculation to eliminate cable resonance interference. An exponential decay mapping relationship is established between the mechanical disturbance value and the system standard disturbance value to extract the energy reduction rate, which characterizes the degree of limitation of system power output under severe working conditions.

4. The multi-agent collaborative planning system for tourism aimed at resolving multi-intent conflicts according to claim 3, characterized in that, The process of generating a visitor willingness value based on tourist preference values ​​and attraction status values, reflecting individual experience needs, and generating an operation and maintenance pressure value reflecting mechanical load constraints by combining regional visitor flow and current queue length, includes: The tourist preference values ​​and the attraction status values ​​are mapped using a spatial vector, and the tour attenuation components constrained by time consumption tolerance, relative time consumption, fatigue tolerance, and relative fatigue are removed to obtain the tour willingness degree that reflects the individual's subjective inclination. Based on the regional passenger flow and the current queue volume, the platform congestion load is calculated by combining the total transport capacity of the time period corresponding to the cabin number. The entire region features are collected along the candidate cableway topology network to extract the operation and maintenance pressure value that reflects the bottleneck of mechanical evacuation.

5. The multi-agent collaborative planning system for tourism aimed at resolving multi-intent conflicts according to claim 4, characterized in that, The process of deriving the probability value of descending the mountain based on the dwell distribution law, and then extracting the return flow prediction value to characterize future spatiotemporal liabilities, includes: Phase separation is performed on the time domain based on the total walking time and cable car time to extract the remaining time representing the effective tour window; The remaining time is input into the stay distribution law, and multimodal probability distribution aggregation is performed in combination with the individual tour rhythm ratio to deduce the probability value of going down the mountain that represents the tourist's future stay characteristics; The probability value of going downhill is accumulated in segments within the time-series space to quantify and extract the return flow prediction value of a single tourist's implicit occupation of future downhill capacity, and then globally aggregated to obtain the total return flow prediction.

6. The multi-agent collaborative planning system for tourism aimed at resolving multi-intent conflicts according to claim 5, characterized in that, The process involves calculating a tour deviation value based on the tour intention level, calculating a carrying capacity deviation value based on the operation and maintenance pressure value and the return flow prediction value, extracting the adjustment direction of the tour deviation value and the adjustment direction of the carrying capacity deviation value, and constructing a conflict assessment value to measure the degree of deviation from intention using the adjustment direction of the tour deviation value and the adjustment direction of the carrying capacity deviation value, including: Perform inverse gradient mapping on the travel intention degree associated with the candidate itinerary node to extract the travel deviation value; Based on the backflow prediction value, the total backflow prediction and the expected queuing volume, the downlink load characteristics are constructed, and spatiotemporal conversion is performed within the constraints of the total transport capacity of the time period to output the carrying capacity deviation value; Extract the spatial geometric direction correlation features between the adjustment direction of the tour deviation value and the adjustment direction of the load deviation value, eliminate consistency intention interference by amplitude attenuation, and couple the divergent component with the load deviation value to output the conflict assessment value that measures the risk of mechanical overload.

7. The multi-agent collaborative planning system for tourism aimed at resolving multi-intent conflicts according to claim 6, characterized in that, The step of using the adjustment direction of the tour deviation value and the adjustment direction of the load deviation value to solve for the joint adjustment amount, in order to coordinately update the current travel value, includes: The difference mode distribution between the adjustment direction of the tour deviation value and the adjustment direction of the bearing deviation value in the parameter space is calculated, and the threshold truncation operation is used to constrain it to a predetermined normalization space to eliminate extreme value divergence and generate a dynamic adjustment ratio. Using the dynamic adjustment ratio as the allocation weight, a nonlinear convex combination superposition is performed on the adjustment direction of the tour deviation value and the adjustment direction of the bearing deviation value to eliminate the trap of unilateral extreme value programming and output the joint adjustment amount of the fusion dimension. The spatiotemporal state of the node sequence is updated by using the coordinated driving step size of the single adjustment amount and the joint adjustment amount.

8. The multi-agent collaborative planning system for tourism aimed at resolving multi-intent conflicts according to claim 7, characterized in that, The process of selecting the scheme that minimizes the sum of the tour deviation value, the carrying capacity deviation value, and the conflict assessment value outputs the final itinerary to achieve a compromise under physical constraints, and uses real-time observation data to close-loop correct the downhill probability value, including: Construct a global cost penalty function that includes the tour deviation value, the carrying deviation value, and the conflict assessment value; iterate through the candidate trips associated with the updated current trip value to filter the trips mapped by the cost minimum point as the final trip. The real-time observation data of tourist displacement status is collected to construct the measured matching degree condition. This condition is used as a priori intervention to perform posterior probability filtering and time-domain normalization correction on the updated downhill probability value in the entire time domain, so as to dynamically eliminate the interference of environmental noise on the inference of the updated downhill probability value.

Images