A flight service management system, method and storage medium

Abstract

The invention belongs to the technical field of flight service management, and discloses a flight service management system, method and storage medium; the method comprises: collecting behavior data of m passengers; constructing a dynamic model of passenger distribution in a terminal hall based on the behavior data; obtaining possible passenger stay locations and passenger movement probabilities in a future y time period by using the dynamic model of passenger distribution; grouping passengers based on the behavior data to obtain n passenger classes; constructing a weighted directed graph for each passenger class based on possible passenger stay locations and passenger movement probabilities in a future y time period; solving the weighted directed graph in layers to obtain the shortest path and required time from the current position to the boarding gate for the corresponding passenger class; issuing boarding reminders to corresponding passengers according to the obtained shortest path and required time; and a scientific crowd flow management scheme improves the waiting environment and enhances the waiting comfort.

Description

Technical Field

[0001] The present invention relates to the technical field of flight service management, and more specifically, to a flight service management system, method and storage medium. Background Art

[0002] The patent application with the publication number CN117973758A discloses a method and device for dispatching transportation hub capacity based on passenger portrait clustering; the method includes using the FCM clustering algorithm to cluster passengers based on historical travel data to form multiple types of passenger portraits, and according to the transfer methods of passengers in each type of passenger portrait, the probability of passengers in each type of passenger portrait choosing various transfer methods is obtained, and the future capacity values ​​of various transfer methods are calculated. For passengers expected to arrive, the passenger count of the passenger choosing various transfer methods is obtained, and the passenger count of all passengers in a certain time period is superimposed according to the transfer method to obtain the total number of passengers for each transfer method, and according to the difference between the total number of passengers for each transfer method in the corresponding time period and the future capacity values ​​of the various transfer methods, the transportation hub capacity for the time period is warned or dispatched; the problem of untimely evacuation of passengers by the hub is solved, the reasonable evacuation of passengers is achieved, and the hub pressure is relieved.

[0003] However, in terms of flight services, due to the lack of in-depth analysis of passenger behavior, service plans are too unified and unable to meet personalized needs; in emergencies such as flight delays, service strategies cannot be adjusted in a timely manner, resulting in a large number of passengers stranded in the terminal hall, causing congestion and complaints; the traditional broadcast boarding reminder method disrupts the order of the terminal hall and is not effective, and passengers often cannot find the boarding gate or miss the boarding time; the layout of facilities and service settings in the terminal hall lack scientific basis and fail to fully consider the actual behavior habits of passengers, resulting in overcrowding in some areas and idle resources in other areas; during peak passenger flow periods such as holidays, the inability to accurately predict and manage the distribution of passenger flow can easily lead to safety hazards; the manpower allocation of service personnel lacks specificity, which not only causes a waste of human resources, but also fails to respond to passenger needs in a timely manner; the lack of analysis of the behavioral characteristics of different passenger groups leads to unreasonable site selection of commercial service facilities, affecting operating efficiency; the simple and extensive management method reduces passenger satisfaction and affects the service quality and competitiveness of the airport.

[0004] In view of this, the present invention proposes a flight service management system, method and storage medium to solve the above problems. Summary of the invention

[0005] In order to overcome the above defects of the prior art and to achieve the above objectives, the present invention provides the following technical solutions: A flight service management system, comprising: a data collection module, for collecting behavior data of m passengers;

[0006] The model building prediction module builds a dynamic model of passenger distribution in the terminal hall based on behavioral data; the dynamic model of passenger distribution is used to obtain the possible location of passengers and the probability of passenger movement in the future y time period;

[0007] The grouping module groups passengers based on behavioral data to obtain n passenger classes;

[0008] A graph construction module constructs a weighted directed graph for each passenger class based on the possible locations of passengers and the probability of passenger movement in the future y time periods;

[0009] The hierarchical solution module push module is used to hierarchically solve the weighted directed graph to obtain the shortest path and required time from the current location to the boarding gate for the corresponding passenger class; boarding reminders are issued to the corresponding passengers based on the obtained shortest path and required time; and each module is connected by wired and / or wireless means.

[0010] Furthermore, the behavior data includes the passenger's movement trajectory, stay time and stay location in the terminal hall.

[0011] Furthermore, the passenger distribution dynamic model is constructed in the following manner:

[0012] The area of ​​the terminal hall is divided into several cell grids, each of which represents a small area in the terminal hall; each cell grid includes several cells;

[0013] According to the behavior data, the state of each cell in the cell grid is initialized and the cell state transition rule is defined. The formula is:

[0014] c(i,j,t+1)=(1-a)×c(i,j,t)+a×[β×g(N(i,j,t),α,O)+(1-β)×h(N(i,j,t),R)];

[0015] Among them, c(i,j,t+1) is the state of the (i,j)th cell at time t+1; N(i,j,t) represents the state of the neighboring cells of the (i,j)th cell at time t, i is the horizontal index of the cell grid, and j is the vertical index of the cell grid; c(i,j,t) is the state of the (i,j)th cell at time t; a is a smoothing parameter, β is the proportional coefficient of the preference weight and the time weight, O is the distribution matrix of obstacles; R is the time factor function, g() is the state probability transfer function; h() is the neighborhood state transfer function;

[0016] Update the state of the cell according to the cell state transition rule. For each time t, traverse all cells, calculate the state N(i,j,t) of the neighboring cells of the (i,j)th cell, calculate the state c(i,j,t+1) of the (i,j)th cell at time t+1 according to the formula of the cell state transition rule, and update the state of the (i,j)th cell to c(i,j,t+1); for each time t, record the state of all cells, and combine these states in the order of time steps to form a dynamic model of passenger distribution.

[0017] Furthermore, the method of initializing the state of each cell in the cell grid includes:

[0018] Define a binary state variable for each cell, determine the position of the passenger at each time step according to the timestamp in the movement trajectory; set the binary state variable of the cell corresponding to the passenger's position to 1; determine the corresponding cell according to the stop position; determine the corresponding cell where the passenger stays at the time step according to the start and end timestamps of the stop time, and set the binary state variables of the corresponding cells at these time steps to 1; for cells without passengers, set their binary state variables to 0;

[0019] Preset the change probability p, traverse all cells, and generate a random number r between 0 and 1 for each cell. If r is less than p, the probability of a cell with a binary state variable of 1 changing to 0 is defined as p, and the probability of a cell with a binary state variable of 0 changing to 1 is defined as p.

[0020] Define a smoothing radius r; traverse all cells, and for each cell c, determine the set N of all cells in the neighborhood with radius r centered on c; calculate the weighted average s' of the binary state variables of all cells in N;

[0021] The weighted average s' is used to update the binary state variable of cell c. The update formula is:

[0022] K_new(c)=w(t)×K_old(c)+(1-w(t))×(α T ×s'+f(c,t)); where K_new(c) is the binary state variable of cell c after update, K_old(c) is the binary state variable of cell c before update; w(t) is the dynamic smoothing weight, t is time, α T is the transpose of the passenger attribute weight vector α; f(c,t) is the environmental impact function;

[0023] ; where w1, w2 and w3 are the corresponding weight coefficients; d(c,ga) is the distance from cell c to gate ga; s(c) is an indicator function; A(v,t) is the state of cell v other than cell c at time t; d(v,c) is the Euclidean distance between cell c and cell v, θ(v,c) is the angle between the direction of cell v pointing to cell c and the normal vector of cell c; ar(c) is the area of ​​cell c, is the decay rate parameter, Greater than 0; t0 is the reference time point.

[0024] Furthermore, the possible staying locations of passengers and the probability of passenger movement in the future y time period are obtained in the following manner:

[0025] Define the predicted time period y in the future, extract the passenger distribution state at the current time point as the initial state from the passenger distribution dynamic model, traverse all cells for each time step, calculate the state of the cell at the next time step according to the cell state transfer rule for each cell, and update the calculated next state to the corresponding cell. In the process of advancement, record the state of all cells at each time step, combine the recorded cell states according to the time step, and obtain the dynamic distribution of passengers in the terminal hall in the simulated future time period t. From the simulated dynamic distribution in the terminal hall, directly read out the possible location of passengers at each time step in the future; that is, the possible location of passengers in the future time period y;

[0026] For each pair of adjacent time steps, compare the changes in cell states. If the state of cell c changes from 0 to 1, a passenger moves from other locations to cell c. Count the number of times the state of cell c changes from 0 to 1 and divide it by the total number of time steps to get the probability of a passenger moving from other locations to cell c. Similarly, calculate the probability of a passenger moving from cell c to other locations, that is, get the passenger movement probability. The passenger movement probability is the probability of a passenger moving from one location to another.

[0027] Furthermore, the method of grouping passengers includes:

[0028] Extract the comprehensive action features of passengers in the terminal hall from the movement trajectory, which include the movement path, movement speed and movement direction; extract the stay features of passengers in different areas from the stay time and stay location; the stay features include the stay duration and stay frequency; standardize or normalize the comprehensive action features and stay features, and perform vector-level splicing to obtain cluster features; all passengers and the corresponding cluster features constitute a data set; the passengers in the data set and the corresponding cluster features are used as samples;

[0029] Define the initial number of clusters \(n_0\), the search range \([n_{min}, n_{max}]\), and the update step size \(z\); where \(n_{min}\) is the lower limit of the search range and \(n_{max}\) is the upper limit of the search range; within the neighborhood \([n_0 - z, n_0 + z]\) of the initial number of clusters \(n_0\), define the candidate number of clusters \(n_1\) for clustering and calculate the comprehensive evaluation index value \(EP(n_1)\); find the number of clusters \(n_{op}\) with the minimum comprehensive evaluation index value; if the comprehensive evaluation index value \(EP(n_{op})\) of \(n_{op}\) is less than the comprehensive evaluation index value \(EP(n_0)\) of the initial number of clusters \(n_0\), then take \(n_{op}\) as the new current number of clusters;

[0030] If the search range \([n_{min}, n_{max}]\) is completely covered, or the evaluation index value does not change in consecutive several iterations, then terminate the update of the number of clusters;

[0031] During the update process of the number of clusters, if \(n_{op}\) is within the range \([n_0 - z, n_0]\), then update the search range to \([n_{min}, n_0 + z]\); if \(n_{op}\) is within the range \((n_0, n_0 + z]\), then update the search range to \([n_0 - z, n_{max}]\); output the final number of clusters as the optimal number of clusters;

[0032] Randomly select several samples from the dataset as the initial cluster centers. For each sample, calculate its distances to each cluster center and assign it to the nearest cluster; for each cluster, recalculate the cluster center and repeat until the optimal number of clusters is reached; obtain the preliminary clustering;

[0033] Take the preliminary clustering as the initial solution \(S_0\), calculate the total sum of squared errors \(E(S_0)\) of the initial solution \(S_0\); and preset the maximum number of neighborhood search times \(maxn\); for each sample \(x\), remove \(x\) from the cluster it belongs to and then reassign it to other clusters to obtain a new solution \(S'\); all these new solutions \(S'\) constitute the neighborhood \(N(S_0)\) of the initial solution \(S_0\); initialize a counter and set its count to 0; traverse each new solution \(S'\) in the neighborhood \(N(S_0)\) of the initial solution \(S_0\); calculate the total sum of squared errors \(E(S')\) of the new solution \(S'\), if \(E(S') < E(S_0)\), then take \(S'\) as the new current solution, i.e., \(S_0 = S'\), and reset the count of the counter to 0; otherwise, increment the count of the counter by 1; if the count of the counter is greater than or equal to the maximum number of neighborhood search times \(maxn\), then terminate, otherwise, continue to search in the neighborhood of the new current solution \(S\). When terminated, output the current solution \(S\) as the optimal solution \(S_{best}\), and the clustering corresponding to the optimal solution \(S_{best}\) is the final clustering;

[0034] The total sum of squared errors of the initial solution \(S_0\)

[0035] ; Where I is the index of the cluster, C_I is the I-th cluster, d(x,m_I) is the distance from sample x to the cluster center m_I of the I-th cluster, L(x,C_I) is a function that measures the degree of abnormality between sample x and its cluster C_I; Q(x,t) is the time factor function; a1, a2 and a3 are the weights of the corresponding items;

[0036] ; Where δ is a positive parameter that controls the curvature of the function, and dist(x,C_I) is the average distance from sample x to the Ith cluster C_I; is a comprehensive factor function; according to the cluster center of the final clustering and the attribution of the samples, the passengers are divided into n categories according to their movement trajectory, stay time and stay location, that is, n passenger classes.

[0037] Furthermore, the method of constructing a weighted directed graph includes:

[0038] Initialize a directed graph, traverse each time step in the future y time period, for each time step, obtain the set of possible passenger stay locations at the time step, add these possible passenger stay locations as nodes to the directed graph, for each pair of adjacent times t and t+1, obtain the set S_t of possible passenger stay locations at time t, and the set S_t+1 of possible passenger stay locations at time t+1; for each possible passenger stay location q in S_t, for each possible passenger stay location p in St+1; if q is not equal to p, add a directed edge from the possible passenger stay location q to the possible passenger stay location p in the directed graph, and the weight of the directed edge is set to the probability of the passenger moving from location q to location p;

[0039] For each directed edge from node b1 to node b2, find the probability of a passenger moving from the corresponding position of node b1 to the corresponding position of node b2 based on the obtained passenger movement probability; use the probability as the weight of the directed edge;

[0040] If at a certain time t, there are several new possible locations for passengers to stay, add these new possible locations for passengers to stay as new nodes to the directed graph. For each new node k, add a directed edge pointing to the new node k from all nodes at time t-1, and the weight of the directed edge is set to the corresponding passenger movement probability; if at a certain time t, there are several possible locations for passengers to stay that are no longer the locations where passengers stay, remove the nodes corresponding to these possible locations for passengers from the directed graph, and remove all directed edges pointing to or from these nodes; add the corresponding boarding gates to the directed graph as nodes therein; that is, complete the construction of the weighted directed graph.

[0041] Furthermore, the method for solving the shortest path and required time from the current location to the boarding gate for the corresponding passenger class includes:

[0042] Divide the weighted directed graph into several levels, each level contains several adjacent nodes, representing a set of regions or locations, adjacent levels are connected by edges, and the weight of the edge represents the distance or movement cost between levels; construct a hierarchical graph, the nodes in the hierarchical graph represent the levels, and the edges represent the connections between the levels; use the Floyd algorithm to calculate the shortest path and the corresponding shortest distance between any two points in the level; store the shortest path and the corresponding shortest distance between any two points in a preset lookup table;

[0043] The level where the node corresponding to the passenger's current location is located is determined as the starting level, and the level where the node corresponding to the boarding gate is located is determined as the target level;

[0044] The A* algorithm or the Dijkstra algorithm is used on the hierarchical graph to find the shortest path from the starting level to the target level. During the search process, the lookup table is used to obtain the shortest distance between nodes in the level and the corresponding shortest path. The shortest path found on the hierarchical graph is split by level. For each level, the lookup table is used to obtain the nodes of the shortest path in the level and arrange them into a sequence, which is recorded as a node sequence. The node sequences of all levels are spliced ​​to obtain a refined node sequence of the entire shortest path. Two adjacent nodes in the refined node sequence of the shortest path are traversed, and the weights of the edges between the two adjacent nodes are accumulated. The accumulated value obtained is the time required for the shortest path from the current location to the boarding gate for the passenger class.

[0045] A flight service management method, which is implemented based on the flight service management system, comprises:

[0046] S1. Collect behavioral data of m passengers;

[0047] S2. Based on the behavior data, a dynamic model of passenger distribution in the terminal hall is constructed; the dynamic model of passenger distribution is used to obtain the possible location of passengers and the probability of passenger movement in the future y time period;

[0048] S3. Based on the behavior data, passengers are grouped to obtain n passenger classes;

[0049] S4. Construct a weighted directed graph for each passenger class based on the possible locations of passengers and the passenger movement probabilities in the future y time periods;

[0050] S5. hierarchically solve the weighted directed graph to obtain the shortest path and required time from the current location to the boarding gate for the corresponding passenger class; and issue a boarding reminder to the corresponding passenger based on the obtained shortest path and required time.

[0051] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the flight service management method is implemented.

[0052] Technical effects and advantages of a flight service management system, method and storage medium of the present invention:

[0053] The present invention realizes the precision and personalization of waiting services by accurately grasping the behavioral characteristics and movement patterns of passengers, greatly improving passenger satisfaction; the system can accurately predict and manage the distribution of passenger flow in the terminal hall, effectively alleviate congestion, and optimize the utilization efficiency of venue resources; differentiated service strategies for different types of passengers meet the personalized needs of special passenger groups and improve service quality; intelligent route planning and timely reminder mechanisms significantly reduce the risk of passengers missing flights and improve boarding efficiency; intelligent operation reduces airport human resource investment and improves operational efficiency and economic benefits; dynamically adjusted service strategies enhance the flexibility and adaptability of flight services and improve the ability to respond to emergencies; personalized service push improves passenger experience and enhances the competitiveness of airport services; scientific passenger flow management solutions improve the waiting environment and enhance the comfort of waiting; the improvement of the intelligence level has driven the overall upgrade of airport services. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A schematic diagram of a flight service management system of the present invention;

[0055] Figure 2 A schematic diagram of a flight service management method of the present invention;

[0056] Figure 3 Schematic diagram of the storage medium of the present invention. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] Example 1

[0059] See also Figure 1 As shown, a flight service management system described in this embodiment includes:

[0060] The data collection module is used to collect the behavior data of m passengers; m is an integer greater than 1;

[0061] The model building prediction module builds a dynamic model of passenger distribution in the terminal hall based on behavioral data; the dynamic model of passenger distribution is used to obtain the possible location of passengers and the probability of passenger movement in the future y time period;

[0062] The grouping module groups passengers based on the behavior data to obtain n passenger classes; n is an integer greater than 1;

[0063] A graph construction module constructs a weighted directed graph for each passenger class based on the possible locations of passengers and the probability of passenger movement in the future y time periods;

[0064] The hierarchical solution module push module is used to hierarchically solve the weighted directed graph to obtain the shortest path and required time from the current location to the boarding gate for the corresponding passenger class; personalized boarding reminders are issued to the corresponding passengers based on the obtained shortest path and required time; and the modules are connected by wire and / or wireless means to realize data transmission between modules.

[0065] Behavioral data includes passengers' movement trajectory, stay time and stay location in the terminal hall; the airport's internal video surveillance system is used to analyze the video stream through computer vision technology to detect and track passengers' movement trajectory; based on the above-obtained movement trajectory, when it is judged that the passenger's stay time in a certain area exceeds a certain threshold, the area is considered to be the passenger's stay location, and the stay time is the stay time, or the video surveillance is used to detect and analyze the passenger's stationary state to obtain his / her stay location and stay time.

[0066] At the same time, the acquired original behavior data is preprocessed, denoised, and trajectory reconstructed to obtain more accurate behavior data.

[0067] The construction methods of passenger distribution dynamic model include:

[0068] The area of ​​the terminal hall is divided into several cell grids, each of which represents a small area in the terminal hall; the cell grid includes several cells; according to the behavior data, the state of each cell in the cell grid is initialized; specifically, a binary state variable is defined for each cell, when the binary state variable is 1, it indicates that the state of the cell is that there is a passenger in it, and when the binary state variable is 0, it indicates that the state of the cell is that there is no passenger in it; according to the timestamp in the movement trajectory, the position of the passenger at each time step is determined; the binary state variable of the cell corresponding to the passenger's position is set to 1; according to the stop position, the corresponding cell is determined; according to the start and end timestamps of the stop time, the corresponding cell where the passenger stays at the time step is determined, and the binary state variables of the corresponding cells at these time steps are set to 1; for cells without passengers, their binary state variables are set to 0.

[0069] Preset the change probability p, traverse all cells, and generate a random number r between 0 and 1 for each cell. If r is less than p, perturb the binary state variable of the cell, that is, define the probability that the cell with a binary state variable of 1 has p to 0, and the probability that the cell with a binary state variable of 0 has p to 1;

[0070] Define a smoothing radius r; traverse all cells, and for each cell c, determine the set N of all cells in the neighborhood with radius r centered on c; calculate the weighted average s' of the binary state variables of all cells in N;

[0071] The weighted average s' is used to update the binary state variable of cell c. The update formula is:

[0072] K_new(c)=w(t)×K_old(c)+(1-w(t))×(α T ×s'+f(c,t)); where K_new(c) is the binary state variable of cell c after update, K_old(c) is the binary state variable of cell c before update; w(t) is the dynamic smoothing weight, t is time, α T is the transpose of the passenger attribute weight vector α; f(c,t) is the environmental impact function.

[0073] The passenger attribute weight vector is obtained by:

[0074] Supervised learning algorithms such as logistic regression and decision trees are used, with passenger attributes as features and passenger movement trajectories as labels, so as to learn the weight coefficients of each attribute, including age, gender, whether or not a passenger has a disability or special needs, whether or not a passenger is carrying children, whether or not the passenger is a frequent flyer, flight delay duration, remaining check-in / boarding time, and flight cabin class.

[0075] ; Among them, w1, w2 and w3 are corresponding weight coefficients, which are adjusted according to actual conditions. They reflect the relative importance of distance to the boarding gate, store and congestion level to passenger movement; d(c,ga) is the distance from cell c to the boarding gate ga; s(c) is an indicator function, if cell c has a store, then s(c)=1, otherwise s(c)=0; A(v,t) is the state of cell v other than cell c at time t, which is 1 if there is a passenger, otherwise 0; d(v,c) is the Euclidean distance between cell c and cell v, θ(v,c) is the angle between the direction of cell v pointing to cell c and the normal vector of cell c; ar(c) is the area of ​​cell c, is the decay rate parameter, Greater than 0, it determines the speed of attenuation; t0 is the reference time point. The farther away from the reference time, the smaller the impact of the past congestion state on the current one.

[0076] Define the cell state transition rules to simulate the movement of passengers in the terminal hall. Specifically, define the cell state transition rules, and the formula is:

[0077] c(i,j,t+1)=(1-a)×c(i,j,t)+a×[β×g(N(i,j,t),α,O)+(1-β)×h(N(i,j,t),R)];

[0078] Among them, c(i,j,t+1) is the state of the (i,j)th cell at time t+1 (discrete into the value of a binary state variable); N(i,j,t) represents the state of the neighboring cells of the (i,j)th cell at time t, i is the horizontal index of the cell grid, and j is the vertical index of the cell grid; c(i,j,t) is the state of the (i,j)th cell at time t; a is a smoothing parameter that controls the weight of the new state and the old state, β is the proportional coefficient of the preference weight and the time weight, O is the obstacle distribution matrix, which represents the distribution of obstacles, such as the position of seats and pillars; R is the time factor function, which describes the impact of time factors on passenger movement, g() is the state probability transfer function; h() is the neighborhood state transfer function; the values ​​of the state probability transfer function and the neighborhood state transfer function are obtained by processing the weighted sum of their internal variables with the sigmoid function; the weight of the weighted sum is obtained by learning the training data.

[0079] Update the state of the cell according to the cell state transition rule. For each time t, traverse all cells, calculate the state N(i,j,t) of the neighboring cells of the (i,j)th cell, calculate the state c(i,j,t+1) of the (i,j)th cell at time t+1 according to the formula of the cell state transition rule, and update the state of the (i,j)th cell to c(i,j,t+1); for each time t, record the state of all cells, combine these states in the order of time steps to form a dynamic passenger distribution model, that is, the passenger distribution dynamic model; simulate the complex movement behavior considering the passenger's movement preference, crowding effect, obstacle influence and time factor, and finally obtain the dynamic passenger distribution model; this model not only contains the static distribution of each time step, but also contains the dynamic trend of passenger movement, which can be used to predict the distribution in future time periods.

[0080] Define the predicted time period y in the future, extract the passenger distribution state at the current time point as the initial state from the passenger distribution dynamic model, use the cell state transition rule, advance step by step according to the time step, and simulate the movement behavior of passengers in the terminal hall in the future t time period; specifically, for each time step, traverse all cells, for each cell, calculate the state of the cell in the next time step according to the cell state transition rule, and update the calculated next state to the corresponding cell. In the process of advancement, record the state of all cells at each time step, and combine the recorded cell states according to the time step to obtain the dynamic distribution of passengers in the terminal hall in the simulated future t time period. From the simulated dynamic distribution in the terminal hall, directly read out the possible location of passengers at each future time step (corresponding to the cell with state 1); that is, the possible location of passengers in the future y time period.

[0081] For each pair of adjacent time steps, compare the changes in cell states. If the state of cell c changes from 0 to 1, a passenger moves from other locations to cell c. Count the number of times the state of cell c changes from 0 to 1 and divide it by the total number of time steps to get the probability of a passenger moving from other locations to cell c. Similarly, calculate the probability of a passenger moving from cell c to other locations, that is, get the passenger movement probability. The passenger movement probability is the probability of a passenger moving from one location to another.

[0082] The possible staying locations of passengers in the future y time period and the probability of passengers moving between different locations are obtained from the passenger distribution dynamic model; it provides basic data support for the subsequent construction of weighted directed graphs and finding the shortest path.

[0083] Ways to group passengers include:

[0084] The comprehensive action characteristics of passengers in the terminal hall are extracted from the movement trajectory, and the comprehensive action characteristics include the movement path, movement speed and movement direction; the stay characteristics of passengers in different areas are extracted from the stay time and stay location; the stay characteristics include stay duration and stay frequency; the comprehensive action characteristics and stay characteristics are standardized or normalized, and vector-level splicing is performed to obtain clustering characteristics; all passengers and the corresponding clustering characteristics constitute a data set; the passengers in the data set and the corresponding clustering characteristics are used as samples.

[0085] Define the initial number of clusters n0, the search range [nmin, nmax] and the update step size; where nmin is the lower limit of the search range and nmax is the upper limit of the search range; the search range [nmin, nmax] is used to determine the range of change in the number of clusters, and the update step size z is used to control the amplitude of change in the number of clusters at each iteration; within the neighborhood [n0-z, n0+z] of the initial number of clusters n0, define the candidate number n1 for clustering, and calculate the comprehensive evaluation index value EP(n1); find the number of clusters nop with the smallest comprehensive evaluation index value; if the comprehensive evaluation index value EP(nop) of nop is less than the comprehensive evaluation index value EP(n0) of the initial number of clusters n0, then use nop as the new current number of clusters.

[0086] The comprehensive evaluation index value is obtained as follows:

[0087] Define several evaluation indicators (such as the maximum distance from the sample to the cluster center, the minimum distance between cluster centers, the maximum density span, the number of isolated points, the consistency of sample attribution, etc.), fuzzify the value of each evaluation indicator, and correspond to several language values ​​(such as very high, relatively high, medium, relatively low, and very low). Build a fuzzy rule base, define the comprehensive evaluation index values ​​corresponding to different combinations of language values, and perform fuzzy reasoning in the fuzzy rule base for the new number of clusters according to its various evaluation indicators to obtain the final comprehensive evaluation index value.

[0088] If the search range [nmin, nmax] is completely covered, or the evaluation index value does not change in several consecutive iterations, the update of the number of clusters is terminated; in the process of updating the number of clusters, if nop is in the range of [n0-z, n0], the search range is updated to [nmin, n0+z]; if nop is in the range of (n0, n0+z], the search range is updated to [n0-z, nmax], and the search range will gradually shrink and eventually gather near an optimal solution; the final number of clusters is output as the optimal number of clusters.

[0089] Randomly select several samples from the data set as the initial cluster centers. For each sample, calculate its distance to each cluster center (using Euclidean distance or other distance metrics) and assign it to the nearest cluster. For each cluster, recalculate the cluster center, that is, the mean vector of the clustering features of all samples in the cluster. Repeat until the optimal number of clusters is reached to obtain preliminary clustering.

[0090] Take the initial clustering as the initial solution S0, and calculate the total sum of squared errors E(S0) of the initial solution S0; and preset the maximum number of neighborhood search times maxn; for each sample x, remove x from the cluster it belongs to, and then reassign it to other clusters to obtain a new solution S'; all these new solutions S' constitute the neighborhood N(S0) of the initial solution S0.

[0091] Initialize a counter and set its count to 0; traverse each new solution S' in the neighborhood N(S0) of the initial solution S0; calculate the total sum of squared errors E(S') of the new solution S', if E(S') < E(S0), then take S' as the new current solution, that is, S0 = S', and reset the count of the counter to 0; otherwise, increment the count of the counter by 1; if the count of the counter is greater than or equal to the maximum number of neighborhood search times maxn, then terminate, otherwise, continue to search in the neighborhood of the new current solution S, when terminated, output the current solution S as the optimal solution Sbest, and the clustering corresponding to the optimal solution Sbest is the final clustering.

[0092] The total sum of squared errors of the initial solution S0

[0093] ; where I is the index of the cluster, C_I is the I-th cluster, d(x, m_I) is the distance from the sample x to the cluster center m_I of the I-th cluster, and L(x, C_I) is a function that measures the abnormality degree of the sample x and its belonging cluster C_I; Q(x, t) is the time factor function; a1, a2, and a3 are the weights of the corresponding terms, which are adjusted according to the actual situation.

[0094] ; where δ is a positive parameter that controls the curvature of the function, and dist(x, C_I) is the average distance from the sample x to the I-th cluster C_I; is the comprehensive factor function, which is used to consider other factors that affect the abnormality degree of the sample; specifically, during airport waiting, there are multiple factors that affect the abnormality degree of the sample, such as passenger age, whether there is a disability, whether carrying a child, flight delay duration, etc.; introducing the comprehensive factor function can comprehensively consider these influencing factors, so as to more accurately measure the abnormality degree of the sample; the comprehensive factor function is the weighted sum of multiple influencing factor functions, and each influencing factor function corresponds to a specific influencing factor, and different function forms are designed according to the actual situation; for example, one of the influencing factor functions is an indicator function, which indicates whether the passenger has a disability; one of the influencing factor functions is a continuous function, which represents the influence degree of the flight delay duration on the abnormality degree; the weight coefficients of the weighted sum can control the relative importance of different factors in the comprehensive abnormality degree, and the weight coefficients are learned from historical data through machine learning algorithms.

[0095] According to the cluster center of the final clustering and the attribution of the samples, the passengers are divided into n categories according to their movement trajectories, stay time and stay location, that is, n passenger classes.

[0096] Ways to construct a weighted directed graph include:

[0097] Initialize a directed graph, traverse each time step in the future y time period, for each time step, obtain the set of possible passenger stay locations at that time step, add these possible passenger stay locations as nodes to the directed graph, for each pair of adjacent times t and t+1, obtain the set S_t of possible passenger stay locations at time t, and the set S_t+1 of possible passenger stay locations at time t+1; for each possible passenger stay location q in S_t, for each possible passenger stay location p in St+1; if q is not equal to p, add a directed edge from the possible passenger stay location q (node) to the possible passenger stay location p (node) in the directed graph, and the weight of the directed edge is set to the probability of the passenger moving from location q to location p (the passenger movement probability is obtained).

[0098] For each directed edge from node b1 to node b2, find the probability of a passenger moving from the corresponding position of node b1 to the corresponding position of node b2 based on the obtained passenger movement probability; and use the probability as the weight of the directed edge.

[0099] If at a certain time t, there are several new possible locations for passengers to stay (which have not appeared in the previous time step), these new possible locations for passengers to stay are added as new nodes to the directed graph. For each new node k, a directed edge pointing to the new node k is added from all nodes at time t-1, and the weight of the directed edge is set to the corresponding passenger movement probability (if it does not exist in the data, it can be set to a smaller default value); if at a certain time t, there are several possible locations for passengers to stay that are no longer the locations where passengers stay, the nodes corresponding to these possible locations for passengers to stay are removed from the directed graph, and all directed edges pointing to or from these nodes are removed; the corresponding boarding gates are added to the directed graph as nodes therein; that is, the construction of the weighted directed graph is completed; the constructed weighted directed graph, in which the nodes represent the possible locations where passengers may stay in the future y time period, the edges represent the movement of passengers between different locations, and the weights of the edges are the corresponding movement probabilities; this weighted directed graph provides the basic data structure for subsequent operations such as finding the shortest path.

[0100] It should be noted that nodes and edges need to be encoded to facilitate storage and calculation; at the same time, graph optimization and compression should be considered, such as merging adjacent position nodes, pruning low-probability edges, etc., to improve efficiency.

[0101] The solutions for the shortest path and time required from the current location to the boarding gate for the corresponding passenger class include:

[0102] The weighted directed graph is divided into several levels, each level contains several adjacent nodes, representing a set of areas or locations, and adjacent levels are connected by edges, and the weight of the edge represents the distance or movement cost between levels; a hierarchical graph is constructed, in which the nodes represent the levels and the edges represent the connections between the levels; the nodes in each level are preprocessed, and the Floyd algorithm is used to calculate the shortest path and the corresponding shortest distance between any two points in the level; the shortest path and the corresponding shortest distance between any two points are stored in a preset lookup table, and the key-value pair format of the lookup table is <starting point node, ending point node>: <shortest distance, shortest path node sequence>; the shortest path node sequence records all the passing nodes on the shortest path from the starting point node to the ending point.

[0103] The level where the node corresponding to the passenger's current position is located is determined as the starting level, and the level where the node corresponding to the boarding gate is located is determined as the target level; the A* algorithm or Dijkstra algorithm is used to find the shortest path from the starting level to the target level on the hierarchical graph. During the search process, the shortest distance between nodes in the level and the corresponding shortest path are obtained using a lookup table. The shortest path found on the hierarchical graph is split by level. For each level, the nodes of the shortest path in the level are obtained using a lookup table, and are arranged into a sequence, which is recorded as a node sequence; the node sequences of all levels are spliced ​​to obtain a refined node sequence of the entire shortest path, and two adjacent nodes in the refined node sequence of the shortest path are traversed, and the weights of the edges between the two adjacent nodes are accumulated (the inverse of the movement probability can be regarded as the time cost), and the accumulated value obtained is the required time corresponding to the shortest path from the current position of the passenger class to the boarding gate.

[0104] Based on the shortest path and required time for each type of passenger, a reasonable advance time (such as 30 minutes) is set for each type of passenger. When the time required for the shortest path is added to the advance time, a personalized boarding reminder is issued to the passenger of that type. Specifically, the content of the boarding reminder includes basic information, path guidance, estimated time, warm reminders and additional services. Basic information includes flight number, boarding gate, destination, boarding time, etc. Path guidance is to give text / image instructions based on the shortest path to guide passengers to walk from their current location to the boarding gate. Estimated time is to remind passengers of the time required to walk along the shortest path. Warm reminders are to give corresponding precautions for special circumstances such as disabilities or carrying children. Additional services are promotional information for catering, shopping and other services in the airport (targeted push according to passenger types).

[0105] Boarding reminders are provided in the form of SMS reminders and app push notifications. The reminder time is dynamically adjusted based on the real-time passenger location, flight status, etc. If the flight is delayed, the reminder is dynamically adjusted and resent.

[0106] This embodiment achieves precision and personalization of waiting services by accurately grasping the behavioral characteristics and movement patterns of passengers, greatly improving passenger satisfaction; the system can accurately predict and manage the distribution of passenger flow in the terminal hall, effectively alleviate congestion, and optimize the utilization efficiency of venue resources; differentiated service strategies for different types of passengers meet the personalized needs of special passenger groups and improve service quality; intelligent route planning and timely reminder mechanisms significantly reduce the risk of passengers missing flights and improve boarding efficiency; intelligent operations reduce airport human resource investment and improve operational efficiency and economic benefits; dynamically adjusted service strategies enhance the flexibility and adaptability of flight services and improve the ability to respond to emergencies; personalized service push improves passenger experience and enhances the competitiveness of airport services; scientific passenger flow management solutions improve the waiting environment and improve the comfort of waiting; the improvement of the level of intelligence has driven the overall upgrade of airport services.

[0107] Example 2

[0108] See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, which provides a flight service management method, including:

[0109] S1. Collect the behavior data of m passengers; m is an integer greater than 1;

[0110] S2. Based on the behavior data, a dynamic model of passenger distribution in the terminal hall is constructed; the dynamic model of passenger distribution is used to obtain the possible location of passengers and the probability of passenger movement in the future y time period;

[0111] S3. Based on the behavior data, the passengers are grouped to obtain n passenger classes; n is an integer greater than 1;

[0112] S4. Construct a weighted directed graph for each passenger class based on the possible locations of passengers and the passenger movement probabilities in the future y time periods;

[0113] S5. Layeredly solve the weighted directed graph to obtain the shortest path and required time from the current location to the boarding gate for the corresponding passenger class; and send a personalized boarding reminder to the corresponding passenger based on the obtained shortest path and required time.

[0114] Example 3

[0115] See also Figure 3As shown, one embodiment of the present application discloses a computer-readable storage medium 600. Computer-readable instructions are stored on the computer-readable storage medium 600. When the computer-readable instructions are executed by the processor, a flight service management method according to an embodiment of the present application described with reference to the above figures can be executed. The storage medium 600 includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0116] In addition, according to the embodiments of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium, which stores machine-readable instructions, and the machine-readable instructions can be executed by a processor to execute instructions corresponding to the method steps provided by the present application, for example: a flight service management method. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are executed.

[0117] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0118] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A flight service management system, characterized in that: include: A data collection module is used to collect the behavior data of m passengers; The model building prediction module builds a dynamic model of passenger distribution in the terminal hall based on behavioral data; the dynamic model of passenger distribution is used to obtain the possible location of passengers and the probability of passenger movement in the future y time period; The grouping module groups passengers based on behavioral data to obtain n passenger classes; A graph construction module constructs a weighted directed graph for each passenger class based on the possible locations of passengers and the probability of passenger movement in the future y time periods; The layered solution module push module is used to layer-solve the weighted directed graph to obtain the shortest path and required time from the current location to the boarding gate for the corresponding passenger class; and send a boarding reminder to the corresponding passenger based on the obtained shortest path and required time; Each module is connected via wired and / or wireless means; The passenger distribution dynamic model is constructed in the following manner: The area of ​​the terminal hall is divided into several cell grids, each of which represents a small area in the terminal hall; each cell grid includes several cells; According to the behavior data, the state of each cell in the cell grid is initialized and the cell state transition rule is defined. The formula is: in, For in time , located at The status of each cell; Indicates time At that time, The state of the neighboring cells of a cell, is the horizontal index of the cell grid, is the vertical index of the cell grid; For in time , located at The status of each cell; is a smoothing parameter, is the proportional coefficient of preference weight and time weight, is the obstacle distribution matrix; is a function of time factor, is the state probability transfer function; is the neighborhood state transfer function; Update the state of the cell according to the cell state transition rule. For each time t, traverse all cells and calculate the The state of the neighboring cells of a cell , according to the formula of cell state transition rule, the time , located at The state of the cell , update The state of the cell is ; For each time , record the states of all cells, and combine these states in time step order to form a dynamic model of passenger distribution.

2. A flight service management system according to claim 1, characterized in that: The behavior data includes the passenger's movement trajectory, stay time and stay location in the terminal hall.

3. A flight service management system according to claim 2, characterized in that: The method of initializing the state of each cell in the cell grid includes: Define a binary state variable for each cell, determine the position of the passenger at each time step according to the timestamp in the movement trajectory; set the binary state variable of the cell corresponding to the passenger's position to 1; determine the corresponding cell according to the stop position; determine the corresponding cell where the passenger stays at the time step according to the start and end timestamps of the stop time, and set the binary state variables of the corresponding cells at these time steps to 1; for cells without passengers, set their binary state variables to 0; Preset the change probability p, traverse all cells, and generate a random number r between 0 and 1 for each cell. If r is less than p, the probability of a cell with a binary state variable of 1 changing to 0 is defined as p, and the probability of a cell with a binary state variable of 0 changing to 1 is defined as p. Define a smoothing radius r; traverse all cells, for each cell , determine the set N of all cells in the neighborhood with radius r and center c; calculate the weighted average of the binary state variables of all cells in N ; Using weighted average Renewing cells The binary state variable of is updated as follows: ;in, For the updated cells The binary state variable, Cells before update A binary state variable of ; is the dynamic smoothing weight, For time, is the passenger attribute weight vector The transpose of is the environmental impact function; in, , and is the corresponding weight coefficient; For cells To the boarding gate distance; is an indicator function; It is cell removal Cells other than In time Status; For cells and cells The Euclidean distance between For cells Pointing to cells Direction and cells The normal vector angle of It's cells The area of is the decay rate parameter, Greater than 0; is the reference time point.

4. A flight service management system according to claim 3, characterized in that: The method for obtaining the possible stay location of the passenger and the passenger movement probability in the future y time period includes: Define the predicted time period y in the future, extract the passenger distribution state at the current time point as the initial state from the passenger distribution dynamic model, traverse all cells for each time step, calculate the state of the cell at the next time step according to the cell state transfer rule for each cell, and update the calculated next state to the corresponding cell. In the process of advancement, record the state of all cells at each time step, combine the recorded cell states according to the time step, and obtain the dynamic distribution of passengers in the terminal hall in the simulated future time period t. From the simulated dynamic distribution in the terminal hall, directly read out the possible location of passengers at each time step in the future; that is, the possible location of passengers in the future time period y; For each pair of adjacent time steps, compare the changes in cell states. The state changes from 0 to 1, if a passenger moves from another location to the cell , count cells The number of times the state changes from 0 to 1 is divided by the total number of time steps to get the number of passengers moving from other locations to the cell. The probability of a passenger from a cell The probability of moving to other locations is the statistically obtained passenger movement probability; the passenger movement probability is the probability that a passenger moves from one location to another.

5. A flight service management system according to claim 4, characterized in that: The method for solving the shortest path and required time from the current location to the boarding gate of the corresponding passenger class includes: Divide the weighted directed graph into several levels, each level contains several adjacent nodes, representing a set of regions or locations, adjacent levels are connected by edges, and the weight of the edge represents the distance or movement cost between levels; construct a hierarchical graph, the nodes in the hierarchical graph represent the levels, and the edges represent the connections between the levels; use the Floyd algorithm to calculate the shortest path and the corresponding shortest distance between any two points in the level; store the shortest path and the corresponding shortest distance between any two points in a preset lookup table; The level where the node corresponding to the passenger's current location is located is determined as the starting level, and the level where the node corresponding to the boarding gate is located is determined as the target level; Using the hierarchical graph The algorithm or Dijkstra algorithm finds the shortest path from the starting level to the target level. During the search process, the lookup table is used to obtain the shortest distance between nodes in the level and the corresponding shortest path. The shortest path found on the level graph is split by level. For each level, the lookup table is used to obtain the nodes of the shortest path in the level and arrange them into a sequence, which is recorded as a node sequence. The node sequences of all levels are spliced ​​to obtain a refined node sequence of the entire shortest path. The two adjacent nodes in the refined node sequence of the shortest path are traversed, and the weights of the edges between the two adjacent nodes are accumulated. The accumulated value obtained is the time required for the shortest path from the current location to the boarding gate for the passenger class.

6. A flight service management method, which is implemented based on a flight service management system according to any one of claims 1 to 5, characterized in that: include: S1. Collect behavioral data of m passengers; S2. Based on the behavior data, a dynamic model of passenger distribution in the terminal hall is constructed; the dynamic model of passenger distribution is used to obtain the possible location of passengers and the probability of passenger movement in the future y time period; S3. Based on the behavior data, passengers are grouped to obtain n passenger classes; S4. Construct a weighted directed graph for each passenger class based on the possible locations of passengers and the passenger movement probabilities in the future y time periods; S5. hierarchically solve the weighted directed graph to obtain the shortest path and required time from the current location to the boarding gate for the corresponding passenger class; and issue a boarding reminder to the corresponding passenger based on the obtained shortest path and required time.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the flight service management method according to claim 6 is implemented.

Images