A tourism route recommendation system based on prospective information

Abstract

The application relates to a tourism route recommendation system based on prospective information, which comprises a scenic spot future passenger flow prediction module, which is used for predicting scenic spot future passenger flow by using an LSTM model based on scenic spot historical passenger flow; a personalized route design module, which is used for generating and optimizing a tourism route by using an ant colony algorithm to construct a heuristic method based on certainty information including tourist preferences, time budgets and scenic spot opening hours and based on prospective information of future scenic spot congestion prediction, and outputting the best tourism route; the future scenic spot congestion degree is determined by the ratio of the predicted scenic spot future passenger flow and the carrying capacity of the scenic spot. The prediction module can predict the scenic spot future passenger flow, and then predict the future congestion degree of the scenic spot, thereby providing prospective information for subsequent route design; the personalized route design module can design and optimize the route according to the tourist demand and the reality constraint, output the best tourism route, and improve the travel experience and satisfaction of tourists.

Description

Technical Field

[0001] The present invention relates to the technical field of tourist routes, and more particularly to a tourist route recommendation system based on forward-looking information. Background Art

[0002] The modern tourism market is undergoing significant changes. On the one hand, the market structure is shifting from a mass market to an independent travel market. On the other hand, the behavior of independent travelers has undergone significant changes. Tourists are increasingly choosing personalized itineraries rather than standard tour packages. Consequently, travel recommendation systems have garnered increasing attention. However, due to various potential emergencies, such as traffic congestion and weather changes, itineraries designed by recommendation systems may become suboptimal or even unrealistic during the actual travel process. Therefore, how to proactively mitigate these potential uncertainties during itinerary design has become a common concern for many tourism practitioners and researchers.

[0003] Existing research or solutions have the following shortcomings: (1) Most studies only focus on the design of travel routes in a deterministic environment, ignoring the negative impact of various uncertain factors on the established travel routes; (2) Existing studies mainly use probability theory or fuzzy optimization methods to deal with uncertainty, but the former relies on a huge amount of data, and the latter will lose a lot of useful information within the feasible domain; (3) Existing studies mainly focus on the uncertainty of transfer time, the research problems are relatively fixed, and the system is not very scalable. Summary of the Invention

[0004] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a travel route recommendation system based on forward-looking information. The system can infer various uncertain events that may occur during the journey based on relevant historical data, so as to avoid these events in advance when designing the route, thereby improving the travel experience and satisfaction of tourists. At the same time, the route design comprehensively considers the personalized requirements of tourists and various practical constraints, and has high practicality.

[0005] The technical solutions of the present invention are as follows:

[0006] On the one hand, a travel route recommendation system based on forward-looking information includes:

[0007] The module for predicting the future visitor flow of scenic spots is used to predict the future visitor flow of scenic spots based on the historical visitor flow of scenic spots using the LSTM model;

[0008] The personalized route design module is used to generate and optimize tourist routes based on deterministic information including tourist preferences, time budgets and attraction opening hours, as well as forward-looking information based on future attraction congestion predictions, using an ant colony algorithm to construct a heuristic method to output the best tourist route; the future attraction congestion is determined by the ratio of the predicted future attraction passenger flow to the attraction's carrying capacity.

[0009] Preferably, the personalized route design module specifically includes:

[0010] The initialization unit is used to encode the movement of tourists based on the spatiotemporal structure of the route and generate the initialization route design;

[0011] The solution set construction unit is used to place all tourists at the starting point before each iteration, and the tourist code will be initialized to an empty chromosome. Subsequently, each tourist will move independently according to the deposited pheromone and the actual constraints. When all tourists reach the end node, an iteration will be completed, and Q feasible routes will be generated. Among them, each movement of the tourist is based on the transition probability, which is calculated based on the deposited pheromone, the potential utility value of the next attraction, and the transition time between attractions.

[0012] The local search unit is used to select the best route S in the current iteration from Q feasible routes after each iteration. cbest ; for S cbest Perform a local search to test whether there is a neighborhood solution with a better utility value. If so, replace S with the neighborhood solution. cbest ; The global optimal solution S so far gbest Will be S gbest and S cbest It is replaced by the one with better utility value;

[0013] The pheromone update unit is used to update the pheromone after the local search and enter the next round of iteration; the units in the personalized route design module are repeatedly executed until the set number of iterations is reached, and the optimal travel route is finally output.

[0014] Preferably, in the initialization unit, the upper layer of the double layer represents the selection and visiting order of scenic spots, and the lower layer of the double layer represents the visiting duration of the corresponding scenic spots.

[0015] Preferably, in the solution set construction unit, from the scenic spot v i Move to attraction v j The transition probability p k The calculation method of (i,j) is as follows:

[0016]

[0017] Among them, τ(i,j) represents the scenic spot v i and attractions j The pheromone concentration between, i, j∈[1, N], N represents the number of scenic spots; allowed l represents the set of unvisited attractions; u j (t) represents the number of tourists arriving at the scenic spot v at time t j The utility value obtained; t(v i ,v j ) indicates scenic spots v i and attractions j The transfer time between them.

[0018] Preferably, tourists from scenic spot v at time t j The utility value u obtained j (t), is expressed as follows:

[0019] u j (t) = MS j (t)·p j ·sc j (t)

[0020] Among them, MS j (t) is a marginal decreasing function; p j is a value between 0 and 1; sc j (t) is a nonlinear function of the congestion degree of attraction j at time t.

[0021] Preferably, in the pheromone updating unit, the updating rule adopted is as follows:

[0022]

[0023] Among them, τ g (i, j) represents the pheromone of route (i, j) at the g-th iteration; ρ represents the rate of increase used to control pheromone; i, j∈[1, N], N represents the number of scenic spots; represents the pheromone increase value brought by tourist l to route (i, j) at the g-th iteration; represents the pheromone increase value brought by the global best tourist to route (i, j) at the g-th iteration; The total utility value U obtained by tourist l from the route at the gth iteration l and the number of paths in the route P l decided.

[0024] Preferably, the total utility value U l , which is expressed as follows:

[0025]

[0026] Where D represents the budgeted number of travel days; It represents the utility value obtained by tourists in the mth stage on the kth day. Each time a tourist visits a scenic spot, it is recorded as a stage. Assume that the number of times a tourist visits a scenic spot during the kth day of travel is The entire tourism process can be divided into i, j∈[1, N], N represents the number of scenic spots; represents the starting time of the mth stage of the kth day, which is determined by the maximum value of the tourist’s arrival time and the opening hours of the scenic spot; It means the end of the visit time; is a marginal decreasing function; p i is a value between 0 and 1; Indicates whether to visit attraction i in the mth stage on the kth day. If yes, then otherwise It is a nonlinear function of the congestion degree of attraction i at time t on day k.

[0027] Preferably, the nonlinear function of the crowdedness of attraction i at time t on day k is It is expressed as follows:

[0028]

[0029] Among them, cap iα Indicates the crowding threshold that affects the tourist experience; cap iβ represents the cutoff point where crowding leads to a decrease in satisfaction; γ represents the sensitivity of tourists to crowding, and the smaller the value, the more sensitive tourists are to crowding; It represents the crowdedness of scenic spot i at time t on day k, which is determined by the number of tourists visiting the scenic spot in the future. and the carrying capacity of scenic spots i The ratio is determined.

[0030] Preferably, the future visitor flow prediction module for a scenic spot includes:

[0031] A data preprocessing unit is used to preprocess the collected historical passenger flow data, including data standardization and division into training and test sets;

[0032] The LSTM training unit is used to train the LSTM model based on the processing results of the data preprocessing unit. During the model training process, the batch size, learning rate, and number of hidden layer units are optimized.

[0033] The prediction generation unit is used to predict the future visitor flow of attractions based on the trained LSTM model.

[0034] The beneficial effects of the present invention are as follows:

[0035] The present invention provides a forward-looking tourist route recommendation system that takes into account the problem of tourist route design under the influence of uncertain events and solves this problem with the help of prediction technology. Compared with existing tourist route recommendation systems, the system proposed by the present invention can effectively avoid the negative impact of various uncertain events (such as traffic congestion, weather changes, etc.) on tourist itineraries, thereby improving tourists' travel experience and satisfaction. At the algorithm design level, LSTM is introduced to generate predictions for uncertain events. This model can effectively balance prediction accuracy and efficiency; based on the ant colony algorithm, it integrates variable neighborhood search and differential evolution algorithms to optimize tourist routes in time and space, and at the same time improves optimization efficiency by improving the search structure; it has high scalability and provides a reference for solving the negative impact of other common uncertain events during travel.

[0036] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. However, the travel route recommendation system based on forward-looking information of the present invention is not limited to the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a system framework diagram of an embodiment of the present invention;

[0038] Figure 2 1 is a structural diagram of an LSTM unit according to an embodiment of the present invention;

[0039] Figure 3 This is a flowchart of the operation of the personalized route design module according to an embodiment of the present invention;

[0040] Figure 4 is a schematic diagram of route coding according to an embodiment of the present invention;

[0041] Figure 5 Schematic diagram of the exchange and insertion operations of an embodiment of the present invention; wherein, (a) is an itinerary diagram for exchanging any two days, (b) is a sequence diagram for exchanging any two scenic spots in the same day's itinerary, (c) is an itinerary diagram for exchanging any two scenic spots in any two days' itinerary, (d) is an itinerary diagram for inserting an unvisited scenic spot at any position, and (e) is a route diagram after inserting an unvisited scenic spot based on (d). DETAILED DESCRIPTION

[0042] The present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0043] like Figure 1 As shown, the present invention provides a travel route recommendation system based on forward-looking information, comprising:

[0044] The future visitor flow prediction module 1 of the scenic spot is used to predict the future visitor flow of the scenic spot based on the historical visitor flow of the scenic spot using the LSTM model;

[0045] The personalized route design module 2 is used to generate and optimize the travel route by using an ant colony algorithm based on deterministic information including tourist preferences, time budgets and opening hours of scenic spots, as well as forward-looking information based on future scenic spot congestion predictions, and output the best travel route; the future scenic spot congestion is determined by the ratio of the predicted future tourist flow of the scenic spot to the carrying capacity of the scenic spot.

[0046] The future visitor flow prediction module 1 of the scenic spot includes:

[0047] A data preprocessing unit 10 is used to preprocess the collected historical passenger flow data, including data standardization and division into training sets and test sets;

[0048] An LSTM training unit 11 is used to perform LSTM model training based on the processing results of the data preprocessing unit 10, and optimize the batch size, learning rate, and number of hidden layer units during the model training process;

[0049] The prediction generation unit 12 is used to predict the future passenger flow of the scenic spot based on the trained LSTM model.

[0050] The personalized route design module 2 specifically includes:

[0051] An initialization unit 20 is used to encode the movement of tourists based on the spatiotemporal structure of the route and generate an initialized route design by using a double-layer variable chromosome;

[0052] The solution set construction unit 21 is used to place all tourists at the starting point before each iteration, and the tourist codes will be initialized to empty chromosomes; then, each tourist will move independently according to the deposited pheromone and the actual constraints; when all tourists reach the end node, one iteration will be completed, and Q feasible routes will be generated; wherein, each movement of the tourist is based on the transition probability, which is calculated based on the deposited pheromone, the potential utility value of the next attraction, and the transition time between attractions;

[0053] The local search unit 22 is used to select the best route S in the current iteration from Q feasible routes after each iteration. cbest ; for S cbest Perform a local search to check whether there is a neighborhood solution with a higher utility value. If so, replace S with the neighborhood solution. cbest ; The global optimal solution S so far gbest Will be S gbest and S cbest It is replaced by the one with higher utility value;

[0054] The pheromone updating unit 23 is used to update the pheromone after the local search and to carry out the next round of iteration; repeatedly execute each unit in the personalized route design module 2 until the set number of iterations is reached, and finally output the best travel route.

[0055] As described above, the forward-looking information-based travel route recommendation system of the present invention can infer various uncertain events that may occur during a trip based on relevant historical data, thereby proactively avoiding these events during itinerary design, thereby improving tourists' travel experience and satisfaction. This invention comprehensively considers tourists' individual requirements and various practical constraints, resulting in highly practical itineraries. The system will be described in detail below, focusing on congestion avoidance during travel.

[0056] The route design problem is first described using a mathematical model as follows. Equation (1) represents the objective function of route design, which is to enable tourists to obtain the highest overall utility value U in a D-day journey, where It represents the utility value obtained by tourists in the mth stage on the kth day. Each time a tourist visits a scenic spot, it is recorded as a stage. Assume that the number of times a tourist visits a scenic spot during the kth day of travel is The entire tourism process can be divided into N represents the number of scenic spots.

[0057]

[0058] Tourists are in the mth stage on the kth day The utility value obtained and tourists’ preference value for the attraction (p i ), visit time in this stage Marginal subjective perception and tourists' satisfaction with crowding Therefore, in The utility value obtained in the stage can be expressed as formula (2):

[0059]

[0060] Where, represents the starting time of the mth stage of the kth day, which is determined by the maximum value of the tourist’s arrival time and the opening hours of the scenic spot; It means the end of the visit time; is a marginal decreasing function; p i is a value between 0 and 1; Indicates whether to visit attraction i in the mth stage on the kth day. If yes, then otherwise

[0061] is the congestion degree of attraction i at time t on day k The nonlinear function is defined as follows:

[0062]

[0063] Among them, cap iα Refers to the crowding threshold that affects the visitor experience; cap iβ It refers to the cut-off point where crowding leads to a decrease in satisfaction; γ represents the sensitivity of tourists to crowding, and the smaller the value, the more sensitive tourists are to crowding; It represents the crowdedness of scenic spot i at time t on day k, which is determined by the number of tourists at that time. and the carrying capacity of the scenic spot (cap i ), the future passenger flow of the scenic spot (the number of tourists in the future of the scenic spot) is an uncertain variable (i.e., currently unknown information). The present invention adopts the future passenger flow prediction module 1 of the scenic spot to process this variable.

[0064] The following equations (4)-(10) are the realistic constraints for route design. Equation (4) indicates that tourists start from a starting point (V S ) and reaches an end point (V) on the last day (D) F ); Equations (5) and (6) indicate that except for the last day, the rest point (V R ) is also the starting point of the second day’s journey; Formula (7) indicates that the visiting time of the scenic spot should be within its opening time window, and Respectively represent attractions v i Opening and closing times, V A is a set of scenic spots; Formula (8) indicates that only one scenic spot can be visited in each stage; Formulas (9) and (10) are used to ensure the connectivity of the route, t(Λ j ,Λ j+1 ) represents the transfer time between two stages, represents the arrival time of the m+1th stage on the kth day, It indicates whether to visit attraction j immediately after visiting attraction i on day k. Indicates yes, Indicates no.

[0065]

[0066] Among them, v i is the starting set (V S ) in a node. v j is the set of end points (V F ) in a node.

[0067] Indicates the first stage of visiting scenic spot i on the first day (i.e. v i , v i Here, it means that the point visited in the first stage of the first day is a point in the starting point set. Indicates the Dth day and the Mth day D Stage (M D Stage D is the last stage of the day) visit the scenic spot j (ie v j , v j is another point in the set of ending points). Indicates the Mth day on the kth day k Whether the stage visits the node i. Indicates the Mth day on the kth day k stage, It means the first stage on the k+1th day.

[0068] In addition to the above-mentioned realistic constraints that must be met, the present invention also provides some personalized options to meet the specific needs and preferences of tourists, as defined in equations (11)-(14). Specifically, tourist destinations often contain a large number of attractions, and tourists often reserve some must-see attractions and some must-not-see attractions. If the designed itinerary does not include must-see attractions or involves must-not-see attractions, the tourist's travel experience will be significantly affected. To this end, equations (11) and (12) constrain this problem, where S V Indicates a collection of must-see attractions, S A represents the set of attractions that must be avoided. Formula (13) constrains the time spent on daily travel to be within the tourist's time budget. represents the waiting time of the mth stage on the kth day, which is caused by the arrival of tourists earlier than the opening time of the scenic spot; τ k represents the starting time of the k-th day. Finally, for multi-day tours, some tourists may not want to change their resting place (hotel) every day during the trip. Therefore, Equation (14) constrains tourists' willingness to change hotels.

[0069]

[0070] In order to solve the above problems, the present invention proposes a tourist route recommendation system including a future tourist flow prediction module 1 and a personalized route design module 2. The future tourist flow prediction module 1 is used to predict the future based on historical data (such as predicting the future congestion level of tourist attractions), thereby providing forward-looking information for subsequent route design. The personalized route design module 2 is used to design and optimize routes based on tourist needs and actual constraints. Considering that tourist route design is an NP-hard problem, it is difficult to obtain its exact solution through a polynomial algorithm in a limited time, especially in the context of urban tourism. The spatiotemporal correlation between attractions and the variability of the urban environment further complicate the difficulty of solving the problem.

[0071] For the future visitor flow prediction module 1 of the scenic spot, the present invention uses a long short-term memory neural network (LSTM) to generate future predictions. The operating principle of LSTM is as follows Figure 2 As shown, its ability comes from its unique "gating" mechanism, which is mainly the input gate (i t ), forget gate (f t ) and the output gate (o t ). Assume that the historical passenger flow sequence of the scenic spot is X=[x1,x2,…,x t ], then its propagation in the LSTM unit mainly includes the following four steps. First, using f t Determine the state of the previous unit (c t-1 ) should be removed. t will decide what new information will be kept in the current cell state (c t ), in this step, a new candidate variable is generated Furthermore, with the help of c t-1 and Generate c t Finally, o t Will determine what information can be output.

[0072] LSTM is a deep learning model with excellent prediction accuracy and computational efficiency. Furthermore, it effectively extracts temporal dependencies and effectively avoids problems such as vanishing and exploding gradients. Generating predictions using LSTM requires three processing units: a data preprocessing unit 10, which preprocesses the collected historical passenger flow data, including data standardization and partitioning into training and test sets; an LSTM training unit 11, which optimizes key parameters such as batch size, learning rate, and number of hidden layer units; and a prediction generation unit 12.

[0073] The present invention utilizes a future visitor flow prediction module 1 to provide forward-looking information for route planning. Specifically, this module takes historical visitor flow data for each attraction as input and outputs predictions of future visitor flows for these attractions. Visitor flow predictions are forward-looking information (i.e., information not currently known). Incorporating this information into subsequent route planning can help tourists avoid crowded visitor times, thereby enhancing their travel experience.

[0074] For the personalized route design module 2, the present invention constructs a heuristic method based on the ant colony algorithm to generate and optimize the travel route. The personalized route design module 2 includes four parts: an initialization unit 20, a solution set construction unit 21, a local search unit 22, and a pheromone update unit 23. The initialization unit 20 will preliminarily set some important parameters of the ant colony algorithm and use a double-layer variable chromosome to directly encode the initial solution; the solution set construction unit 21 will calculate the transfer probability of the ant colony based on the deposited pheromones and realistic constraints, thereby constructing a feasible route set; the local search unit 22 integrates variable neighborhood search and differential evolution algorithms to optimize the spatiotemporal structure of the route; the pheromone update unit 23 updates the deposited pheromone according to the best route in the ant colony. The operation flow chart of the personalized route design module 2 is as follows: Figure 3 shown.

[0075] (1) Initialization unit

[0076] In this unit, we first need to preliminarily set the important parameters of the ant colony algorithm, including the number of iterations (G), population size (Q), pheromone importance (α), heuristic factor importance (β) and pheromone volatility (ρ). In the context of tourist route design, the coding rules of the ant colony algorithm can regard each ant as a tourist, and the tourist's movement route is the tourist route. Taking into account the time (visit time) and space (attraction selection and visiting order) structure of the route, the present invention uses a double-layer, variable chromosome to encode the movement of tourists. Figure 4 The diagram shows the structure of a route code, where the top layer represents the selection and order of attractions, and the bottom layer represents the duration of visits to the corresponding attractions. Thus, the schematic route code indicates that on the first day, a tourist visited attractions v4, v7, and v1 consecutively, spending 35, 30, and 60 minutes at each of these attractions, respectively.

[0077] (2) Solution set construction unit

[0078] Before each iteration, all tourists will be placed at the starting point and their codes will be initialized to empty chromosomes. Subsequently, each tourist moves independently according to the deposited pheromones and real-world constraints. When all tourists reach the end node, an iteration is completed and Q feasible routes are generated. Each movement of a tourist is mainly based on the transition probability, which is calculated based on the deposited pheromones, the potential utility value of the next attraction, and the transfer time between attractions. For example, for tourist l, it moves from attraction v to attraction v. i Move to attraction v j The probability p k The calculation method of (i, j) is shown in formulas (11) and (12). i and attractions j Pheromone concentration between allowed l A collection of unvisited attractions; u j (t) represents the number of tourists arriving at the scenic spot v at time t j The utility value obtained; t(v i ,v j ) indicates scenic spots v i and attractions j The transfer time between them.

[0079]

[0080]

[0081] Tourists at time t from scenic spot v j The utility value u obtained j (t), is expressed as follows:

[0082] u j (t) = MS j (t)·p j ·sc j (t) (13)

[0083] Among them, MS j (t) is a marginal decreasing function; p j is a value between 0 and 1; sc j (t) is a nonlinear function of the congestion degree of attraction j at time t.

[0084] (3) Local search unit

[0085] In this unit, the present invention introduces a local search process to further enhance the optimization capability of the ant colony algorithm. Specifically, after each iteration, the best route S in the current iteration is selected from Q feasible routes. cbest . Then, S cbestPerform a local search to check whether there is a better neighborhood solution. If so, replace S with that solution. cbest Finally, the global optimal solution (S gbest ) will be S gbest and S cbest The above local search is mainly based on variable neighborhood search and differential evolution algorithm. Among them, variable neighborhood search seeks a better solution by changing the neighborhood structure of the system, which is mainly used to optimize the spatial structure of the route. Taking into account the problem context and the quality and efficiency of the solution, the present invention designs two neighborhood structures: exchange operation and insertion operation. The structure is shown in the figure. Figure 5 shown. Figure 5 (a) To exchange any two days of itinerary, Figure 5 (b) To exchange the order of any two attractions in the same day’s itinerary, Figure 5 (c) To exchange any two attractions in any two-day itinerary, Figure 5 (d) To insert an unvisited attraction at any point in the itinerary, Figure 5 (e) Figure 5 (d) Based on the example above, we insert a route after we've visited an unvisited attraction. All newly constructed routes must meet real-world constraints before they are saved. After optimizing the spatial structure, we use the differential evolution algorithm to optimize the temporal structure of the route. The differential evolution algorithm uses operations such as mutation, crossover, and selection to converge the initial solution to the global optimal solution.

[0086] (4) Pheromone update unit

[0087] After the local search, the pheromone will be updated. The existing technology mainly adopts two update strategies: ranking-based ant system or elite ant system. The update rule adopted by the present invention is defined as formula (14), where τ g (i, j) represents the pheromone of route (i, j) at the g-th iteration; It represents the pheromone added value brought by tourist l (globally optimal tourist) to route (i, j) at the g-th iteration. As shown in formula (15), is the total utility value (U l ) and the number of paths in the route (P l ). ρ is used to control the rate of pheromone increase, thereby reducing the probability of local optimality. In addition, by incorporating the global optimal visitor, this method can effectively accelerate the aggregation speed.

[0088]

[0089] In summary, Personalized Route Design Module 2 is responsible for generating route recommendations based on various information and real-world constraints. Specifically, its inputs include deterministic information such as tourist preferences, time budget, and attraction opening hours, as well as forward-looking information such as future attraction crowding predictions. Based on this information, Personalized Route Design Module 2 first generates an initial route design, then performs a series of complex optimizations on the spatiotemporal structure of the initial route, ultimately outputting the optimal travel route.

[0090] The above embodiments are only used to illustrate the present invention and are not intended to limit the present invention. As long as they are based on the technical essence of the present invention, any changes or modifications to the above embodiments will fall within the scope of the claims of the present invention.

Claims

1. A travel route recommendation system based on forward-looking information, characterized in that: include: The module for predicting the future visitor flow of scenic spots is used to predict the future visitor flow of scenic spots based on the historical visitor flow of scenic spots using the LSTM model; A personalized route design module is used to generate and optimize tourist routes using an ant colony algorithm based on deterministic information such as tourist preferences, time budgets, and attraction opening hours, as well as forward-looking information based on future attraction crowding predictions. The module then outputs the optimal route. The future crowdedness of the scenic spot is determined by the ratio of the predicted future passenger flow of the scenic spot to the carrying capacity of the scenic spot; The personalized route design module specifically includes: The initialization unit is used to encode the movement of tourists based on the spatiotemporal structure of the route and generate the initialization route design; The solution set construction unit is used to place all tourists at the starting point before each iteration, and the tourist code will be initialized to an empty chromosome. Subsequently, each tourist will move independently according to the deposited pheromone and the actual constraints. When all tourists reach the end node, an iteration will be completed, and Q feasible routes will be generated. Among them, each movement of the tourist is based on the transition probability, which is calculated based on the deposited pheromone, the potential utility value of the next attraction, and the transition time between attractions. The local search unit is used to select the best route S in the current iteration from Q feasible routes after each iteration. cbest ; for S cbest Perform a local search to test whether there is a neighborhood solution with a better utility value. If so, replace S with the neighborhood solution. cbest ; The global optimal solution S so far gbest Will be S gbest and S cbest It is replaced by the one with better utility value; The pheromone update unit is used to update the pheromone after the local search and enter the next round of iteration; the units in the personalized route design module are repeatedly executed until the set number of iterations is reached, and the optimal travel route is finally output; In the pheromone updating unit, the updating rule adopted is as follows: Among them, τ g (i, j) represents the pheromone of route (i, j) at the g-th iteration; ρ represents the rate of increase used to control pheromone; i, j∈[1, N], N represents the number of scenic spots; represents the pheromone increase value brought by tourist l to route (i, j) at the g-th iteration; represents the pheromone increase value brought by the global best tourist to route (i, j) at the g-th iteration; The total utility value U obtained by tourist l from the route at the gth iteration l and the number of paths in the route P l decided; Total utility value U l , which is expressed as follows: Where D represents the budgeted number of travel days; It represents the utility value obtained by tourists in the mth stage on the kth day. Each time a tourist visits a scenic spot, it is recorded as a stage. The number of times a tourist visits a scenic spot during the kth day of travel is The entire travel process is divided into i, j∈[1, N], N represents the number of scenic spots; represents the starting time of the mth stage of the kth day, which is determined by the maximum value of the tourist’s arrival time and the opening hours of the scenic spot; It means the end of the visit time; is a marginal decreasing function; p i is a value between 0 and 1; Indicates whether to visit attraction i in the mth stage on the kth day. If yes, then otherwise It is a nonlinear function of the congestion degree of attraction i at time t on day k; The nonlinear function of the crowdedness of attraction i at time t on day k It is expressed as follows: Among them, cap iα Indicates the crowding threshold that affects the tourist experience; cap iβ represents the cutoff point where crowding leads to a decrease in satisfaction; γ represents the sensitivity of tourists to crowding, and the smaller the value, the more sensitive tourists are to crowding; It represents the crowdedness of scenic spot i at time t on day k, which is determined by the number of tourists visiting the scenic spot in the future. and the carrying capacity of scenic spots i The ratio is determined.

2. The travel route recommendation system based on forward-looking information according to claim 1, characterized in that: In the initialization unit, the upper layer of the double layer represents the selection and visiting order of scenic spots, and the lower layer of the double layer represents the visiting duration of the corresponding scenic spots.

3. The travel route recommendation system based on forward-looking information according to claim 1, characterized in that: In the solution set construction unit, from the scenic spot v i Move to attraction v j The transition probability p k The calculation method of (i,j) is as follows: Among them, τ(i,j) represents the scenic spot v i and attractions j The pheromone concentration between, i, j∈[1, N], N represents the number of scenic spots; allowed l represents the set of unvisited attractions; u j (t) represents the number of tourists arriving at the scenic spot v at time t j The utility value obtained; t(v i ,v j ) indicates scenic spots v i and attractions j The transfer time between them.

4. The travel route recommendation system based on forward-looking information according to claim 3, characterized in that: Tourists at time t from scenic spot v j The utility value u obtained j (t), is expressed as follows: u j (t)=MS j (t)·p j ·sc j (t) Among them, MS j (t) is a marginal decreasing function; p j is a value between 0 and 1; sc j (t) is a nonlinear function of the congestion degree of attraction j at time t.

5. The travel route recommendation system based on forward-looking information according to claim 1, characterized in that: The future visitor flow prediction module for a scenic spot includes: A data preprocessing unit is used to preprocess the collected historical passenger flow data, including data standardization and division into training and test sets; The LSTM training unit is used to train the LSTM model based on the processing results of the data preprocessing unit. During the model training process, the batch size, learning rate, and number of hidden layer units are optimized. The prediction generation unit is used to predict the future visitor flow of attractions based on the trained LSTM model.

Images