AI intelligent car rental operation decision-making system

Abstract

The invention relates to the technical field of vehicle scheduling and distribution, and discloses an AI intelligent vehicle renting operation decision-making system, which comprises a data acquisition module used for acquiring regional feature data of sub-regions; the preprocessing module is used for preprocessing the regional feature data to obtain a second feature sequence; the graph structure construction module is used for constructing a region graph structure according to the second feature sequence; the region demand prediction module is used for constructing a region demand prediction model according to the region graph structure; and the distribution scheme generation module is used for generating a distribution scheme according to the regional demand prediction model in combination with a particle swarm algorithm which introduces a t distribution disturbance variation strategy. According to the method, the distribution scheme of the rental vehicles is generated by constructing the region demand prediction model and combining the particle swarm optimization introducing the t distribution disturbance variation strategy, the utilization rate of the rental vehicles is maximized, and idling of the vehicles is avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle dispatching and allocation, and specifically relates to an AI intelligent car rental operation decision-making system. Background Art

[0002] In the modern car rental industry, with the intensification of market competition and the diversification of customer needs, the rationality of vehicle allocation is directly related to the company's operational efficiency and customer experience. How to optimize the allocation of vehicle resources, meet the dynamic needs of different regions, and maximize the utilization of vehicles has become a core issue that major car rental companies need to solve. However, the vehicle allocation plans of most car rental companies still rely mainly on experience and subjective judgment. The limitation of this method is that it cannot accurately capture the dynamic demand fluctuations in various regions, resulting in idle vehicles or failure to meet customer needs. For example, in areas with convenient public transportation, the demand for car rentals is usually low; at tourist attractions on weekends, the demand for car rentals is usually high.

[0003] Therefore, there is an urgent need for an AI intelligent car rental operation decision-making system to solve the above problems. Summary of the invention

[0004] The present invention provides an AI intelligent car rental operation decision system to solve the technical problem that the related technology relies on experience and subjective judgment and cannot accurately capture dynamic needs, resulting in idle vehicles or failure to meet customer needs.

[0005] The present invention provides an AI intelligent car rental operation decision system, comprising:

[0006] A data collection module, which is used to collect regional feature data of M sub-regions in a region G within a preset time period T according to a preset time interval t, and represent the regional feature data at each time point by a first feature sequence;

[0007] The mth sequence unit of the first feature sequence at the nth time point represents the regional feature data of the mth sub-region at the nth time point, where 1≤m≤M, 1≤n≤N, and N=T / t;

[0008] Regional characteristic data include: number of rental orders, number of rental returns, rental duration, regional population density, number of subway stations, and number of bus stops;

[0009] A preprocessing module, which is used to preprocess the first feature sequence to obtain a second feature sequence;

[0010] A graph structure construction module, which is used to construct a regional graph structure according to the second feature sequence, the regional graph structure including nodes, initial features of nodes and edges between nodes; nodes are represented by sub-regions, and the initial features of nodes are represented by regional feature data of sub-regions that establish a mapping relationship with nodes;

[0011] A regional demand prediction module is used to construct a regional demand prediction model based on the regional graph structure. The input of the regional demand prediction model is the regional graph structure, and the output is the number of car rental demands in the sub-region.

[0012] The allocation plan generation module is used to generate an allocation plan based on the regional demand forecasting model combined with the particle swarm algorithm that introduces the t-distribution perturbation mutation strategy. The allocation plan indicates that P vehicles in the area G are allocated to M sub-areas at the second preset time Q.

[0013] Furthermore, T, t, M, m, G, N, n, P, and Q are all custom parameters.

[0014] Furthermore, the pre-processing step includes:

[0015] Step S201, for missing values ​​in the regional feature data, filling processing is performed by taking the average value of the value of the previous sequence unit and the value of the next sequence unit;

[0016] Step S202, replacing the abnormal value in the regional feature data by taking the average value of the values ​​of other sequence units except the abnormal value;

[0017] Step S203, normalizing the regional feature data using a z-score normalization method.

[0018] Furthermore, the edges are constructed in the following way:

[0019] Build edges based on geographic proximity. That is, when two sub-regions are adjacent, build an edge between them. The weight of the edge based on geographic proximity is calculated by Calculated, where δ A，B represents the weight of edge AB based on geographical proximity, H A，B represents the distance between sub-region A and sub-region B, It represents the minimum geographical distance between all sub-regions, and ε represents the geographical proximity adjustment factor, which ranges from 1 to 10;

[0020] The edge is constructed based on the mobility of rental vehicles. That is, when there is a flow of rental vehicles between two sub-areas, an edge is constructed between the two sub-areas. The weight of the edge based on vehicle mobility is calculated by Calculated, where μ A，B represents the weight of edge AB based on vehicle mobility, C A，B represents the number of rental vehicles rented out from sub-area A and returned to sub-area B during the previous preset time interval, S A，Brepresents the number of rental vehicle flows between all sub-areas during the previous preset time interval, max(f) represents the value with the highest flow frequency between all sub-areas during the previous preset time interval, and the flow frequency is expressed by It indicates that α represents the vehicle mobility adjustment factor, and its value range is from 0 to 1;

[0021] Build edges based on public transportation. That is, when there is a direct public transportation connection between two sub-areas, build an edge between the two sub-areas. The public transportation-based weight of the edge is calculated by Calculated, where D A，B represents the weight of edge AB based on public transportation, D A，B represents the number of public transportation lines from sub-area A to sub-area B, L A，B represents the geographical distance between sub-area A and sub-area B, maxD represents the maximum value of public transportation density among all sub-areas, and the public transportation density is expressed by express, It represents the public transportation density adjustment factor, ranging from 1 to 10;

[0022] The combined weight of edge AB is given by W A,B =w1δ A,B +w2μ A,B +w3θ A,B It is obtained by calculation, wherein w1, w2 and w3 represent the first adjustment coefficient, the second adjustment coefficient and the third adjustment coefficient respectively, and the value range is 0 to 1.

[0023] Furthermore, the regional demand forecasting model includes: a first hidden layer, a concatenation layer, a second hidden layer, a reduction layer, and a classifier;

[0024] The first hidden layer includes N first units, the nth first unit inputs the region graph structure and outputs the first sub-region features of M nodes;

[0025] The concatenation layer includes N concatenation units, and the nth concatenation unit is used to concatenate the first updated features output by the N first units into an integrated vector;

[0026] The second hidden layer includes N second units, the nth second unit inputs the integrated vector output by the nth splicing unit, and outputs the second sub-region feature;

[0027] The restoration layer is used to restore the second sub-region feature output by the Nth second unit into M third sub-region features, where each third sub-region feature corresponds to a node update feature of a node;

[0028] The node update features of each node are input into the classifier, and the classification space of the classifier represents the number of car rental demands in each sub-region.

[0029] Furthermore, the nth first unit of the first hidden layer of the regional demand forecasting model is used to process the regional graph structure at the nth time point, and the formula of the nth first unit of the first hidden layer includes:

[0030]

[0031]

[0032]

[0033] in represents the first sub-region feature of the mth node, H m represents the initial feature of the mth node, W m Represents the first weight parameter corresponding to the mth node, represents the weighted adjacency matrix, represents the normalized degree matrix, Represents the degree matrix. The size of the degree matrix is ​​M×M. The elements of the degree matrix represent the number of edges connected to the node. W i，k represents the comprehensive weight of the edge between node i and node k, 1(i, k) represents the indicator function, which is used to determine whether there is an edge between node i and node k, and ReLU represents the ReLU activation function.

[0034] Further, the N second units of the second hidden layer are constructed based on gated recurrent units.

[0035] Furthermore, sample labels of the regional demand forecasting model are obtained through manual labeling.

[0036] Furthermore, the loss function of the regional demand forecasting model is the cross entropy loss function.

[0037] Furthermore, the steps of generating an allocation plan based on the regional demand forecasting model combined with the particle swarm algorithm that introduces the t-distribution perturbation mutation strategy include:

[0038] Step S301, randomly generate an initial particle population with a particle number U that meets the constraint conditions;

[0039] The constraints include: the sum of the number of rental vehicles allocated to each sub-region is P; the number of rental vehicles allocated to each sub-region is less than the number of rental vehicles demanded in this sub-region predicted by the regional demand forecasting model;

[0040] Each particle includes a position code and a velocity code;

[0041] Position encoding is done by vector X = x1…x M represents, where x1…x MThey represent the number of allocated rental vehicles from the 1st sub-area to the Mth sub-area, respectively. The number of allocated rental vehicles is represented by real number coding; the speed coding is represented by vector V = v1…v M Indicates that the value of each dimension of the velocity vector is Z% of the value of each dimension of the position vector, where Z is a custom parameter;

[0042] Step S302, constructing an objective function in combination with a regional demand forecasting model;

[0043] The objective function is: where x m represents the number of allocated rental vehicles in the mth sub-area, y m represents the number of car rental demands in the mth sub-region predicted by the regional demand model;

[0044] Step S303, calculating the fitness values ​​of all particles in the particle swarm according to the objective function;

[0045] Step S304, updating the particle positions and velocities in the next generation particle swarm according to the particle swarm algorithm update strategy and the t-distribution perturbation mutation strategy;

[0046] Step S305, if the fitness value of each particle in the initialized particle swarm with the current iteration number j is determined to be greater than the maximum fitness value of the particle in history, the maximum fitness value of the particle in history is replaced by the fitness value of the particle with the current iteration number j;

[0047] The maximum fitness value in the initialized particle swarm with the current iteration number j is taken as the global optimal solution;

[0048] Step S306, the current number of iterations is increased by 1, and if it is determined that the current number of iterations reaches the maximum number of iterations, the global optimal solution is output as the allocation solution, otherwise, the process returns to step S303, where the maximum number of iterations is a custom parameter.

[0049] The beneficial effects of the present invention are as follows: the regional demand forecasting model constructed by the present invention first updates the node features of the regional graph structure at each time point, and then integrates the vectors of the regional graph structure at N time points to perform overall demand forecasting, fully considering the dynamic changes of the edges caused by time changes, ensuring the accuracy and timeliness of the forecast results, and generating a specific allocation plan based on the forecast results of the regional demand forecasting model combined with the particle swarm algorithm that introduces the t-distribution perturbation mutation strategy, thereby maximizing the utilization rate of rental vehicles and avoiding vehicle idleness. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a module schematic diagram of an AI intelligent car rental operation decision system of the present invention;

[0051] Figure 2 It is a flow chart of the particle swarm algorithm of the present invention that introduces the t-distribution perturbation mutation strategy.

[0052] In the figure: a data collection module 101, a preprocessing module 102, a graph structure building module 103, a regional demand prediction module 104 and an allocation plan generation module 105. DETAILED DESCRIPTION

[0053] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0054] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0055] like Figure 1-Figure 2 As shown, an AI intelligent car rental operation decision system includes:

[0056] A data collection module 101 is used to collect regional feature data of M sub-regions in a region G within a preset time period T according to a preset time interval t, and represent the regional feature data at each time point by a first feature sequence;

[0057] The mth sequence unit of the first feature sequence at the nth time point represents the regional feature data of the mth sub-region at the nth time point, where 1≤m≤M, 1≤n≤N, and N=T / t;

[0058] Regional characteristic data include: number of rental orders, number of rental returns, rental duration, regional population density, number of subway stations, and number of bus stops;

[0059] A preprocessing module 102, which is used to preprocess the first feature sequence to obtain a second feature sequence;

[0060] A graph structure construction module 103, which is used to construct a regional graph structure according to the second feature sequence, wherein the regional graph structure includes nodes, initial features of nodes, and edges between nodes; nodes are represented by sub-regions, and initial features of nodes are represented by regional feature data of sub-regions that establish a mapping relationship with nodes;

[0061] A regional demand prediction module 104 is used to construct a regional demand prediction model according to the regional graph structure, wherein the input of the regional demand prediction model is the regional graph structure, and the output is the number of rental cars required in the sub-region;

[0062] The allocation plan generation module 105 is used to generate an allocation plan based on the regional demand forecasting model combined with the particle swarm algorithm that introduces the t-distribution perturbation mutation strategy. The allocation plan indicates that P vehicles in the area G are allocated to M sub-areas at the second preset time Q.

[0063] In one embodiment of the present invention, T, t, M, m, G, N, n, P and Q are all custom parameters. Preferably, T is set to 21 days and t is set to 1 day.

[0064] In one embodiment of the present invention, the pre-processing step includes:

[0065] Step S201, for missing values ​​in the regional feature data, filling processing is performed by taking the average value of the value of the previous sequence unit and the value of the next sequence unit;

[0066] Step S202, for the outliers in the regional feature data, the average value of the values ​​of other sequence units except the outliers is taken for replacement processing, where the outliers refer to values ​​that are significantly different from the values ​​of other sequence units, such as a negative value for the rental time or a population density of 0;

[0067] Step S203, normalizing the regional feature data using a z-score normalization method.

[0068] In one embodiment of the present invention, edges are constructed in the following manner:

[0069] Build edges based on geographic proximity. That is, when two sub-regions are adjacent, build an edge between them. The weight of the edge based on geographic proximity is calculated by Calculated, where δ A,B represents the weight of edge AB based on geographical proximity, H A,B represents the distance between sub-region A and sub-region B, which is obtained by Euclidean distance, represents the value with the minimum geographical distance between all sub-regions, ε represents the geographical proximity adjustment factor, which is used to amplify the edges with strong geographical proximity, and the value range is 1 to 10. Preferably, ε is set to 2;

[0070] The edge is constructed based on the mobility of rental vehicles. That is, when there is a flow of rental vehicles between two sub-areas, an edge is constructed between the two sub-areas. For example, a rental vehicle is borrowed from sub-area A and returned to sub-area B. The weight of the edge based on vehicle mobility is calculated by Calculated, where μ A,B represents the weight of edge AB based on vehicle mobility, C A,B represents the number of rental vehicles rented out from sub-area A and returned to sub-area B during the previous preset time interval, S A,B represents the number of rental vehicle flows between all sub-areas during the previous preset time interval, maxf represents the value with the highest flow frequency between all sub-areas during the previous preset time interval, and the flow frequency is expressed by α represents the vehicle mobility adjustment factor, which is used to amplify the edges with stronger mobility, and its value range is 0 to 1. Preferably, α is set to 0.5;

[0071] Build edges based on public transportation. That is, when there is a direct public transportation connection between two sub-areas, an edge is built between the two sub-areas, such as subways, bus lines, etc. The public transportation-based weight of the edge is calculated by Calculated, where D A,B represents the weight of edge AB based on public transportation, D A,B represents the number of public transportation lines from sub-area A to sub-area B, L A,B represents the geographical distance between sub-area A and sub-area B, maxD represents the maximum value of public transportation density among all sub-areas, and the public transportation density is expressed by express, represents the public transportation density adjustment factor, which is used to reduce the edge with strong public transportation density. The value range is 1 to 10. Set to 2;

[0072] The combined weight of edge AB is given by W A,B =w1δ A,B +w2μ A,B +w3θ A,B Calculated, where w1, w2 and w3 represent the first adjustment coefficient, the second adjustment coefficient and the third adjustment coefficient respectively, which are used to determine the contribution ratio of each factor to the comprehensive weight, and the value range is 0 to 1.

[0073] In one embodiment of the present invention, the regional demand forecasting model includes: a first hidden layer, a concatenation layer, a second hidden layer, a reduction layer, and a classifier;

[0074] The first hidden layer includes N first units, the nth first unit inputs the region graph structure and outputs the first sub-region features of M nodes;

[0075] The concatenation layer includes N concatenation units, and the nth concatenation unit is used to concatenate the first updated features output by the N first units into an integrated vector;

[0076] The second hidden layer includes N second units, the nth second unit inputs the integrated vector output by the nth splicing unit, and outputs the second sub-region feature;

[0077] The restoration layer is used to restore the second sub-region feature output by the Nth second unit into M third sub-region features, where each third sub-region feature corresponds to a node update feature of a node;

[0078] The node update features of each node are input into the classifier, and the classification space of the classifier represents the number of car rental demands in each sub-region.

[0079] In one embodiment of the present invention, the nth first unit of the first hidden layer of the regional demand forecasting model is used to process the regional graph structure at the nth time point, and the formula of the nth first unit of the first hidden layer includes:

[0080]

[0081]

[0082]

[0083] in represents the first sub-region feature of the mth node, H m represents the initial feature of the mth node, W m Represents the first weight parameter corresponding to the mth node, represents the weighted adjacency matrix, represents the normalized degree matrix, Represents the degree matrix. The size of the degree matrix is ​​M×M. The elements of the degree matrix represent the number of edges connected to the node. W i,k represents the comprehensive weight of the edge between node i and node k, 1i,k represents the indicator function, which is used to determine whether there is an edge between node i and node k, and ReLU represents the ReLU activation function.

[0084] In one embodiment of the present invention, the N second units of the second hidden layer are constructed based on the gated recurrent unit, which is used to process a sequence composed of integrated vectors spliced ​​from the nodes of the regional graph structure of N time points output by the first hidden layer, and obtain a second sub-region feature, which is composed of node update features of M nodes.

[0085] In one embodiment of the present invention, the regional demand forecasting model first updates the node features of the regional graph structure at each time point, and then integrates the vectors of the regional graph structure at N time points to perform overall demand forecasting, fully considering the dynamic changes of edges caused by time changes, and ensuring the accuracy and timeliness of the forecasting results.

[0086] In one embodiment of the present invention, by obtaining the historical number of car rental orders, the number of car rental returns and the length of car rental from the car rental platform business system, the population density, the number of subway stations and the number of bus stops in the area are obtained from public statistical data, and used as sample data for the regional demand forecasting model, the sample labels of the regional demand forecasting model are obtained through manual annotation.

[0087] In one embodiment of the present invention, the loss function of the regional demand forecasting model is a cross entropy loss function.

[0088] In one embodiment of the present invention, Figure 2 As shown, the steps of generating an allocation plan based on the regional demand forecasting model combined with the particle swarm algorithm that introduces the t-distribution perturbation mutation strategy include:

[0089] Step S301, randomly generate an initial particle population with a particle number U that meets the constraint conditions;

[0090] The constraints include: the sum of the number of rental vehicles allocated to each sub-region is P; the number of rental vehicles allocated to each sub-region is less than the number of rental vehicles demanded in this sub-region predicted by the regional demand forecasting model;

[0091] Each particle includes a position code and a velocity code;

[0092] The position encoding is done by a vector X = {x1…x M} means that x1…x M They represent the number of allocated rental vehicles in the 1st sub-area to the Mth sub-area respectively. The number of allocated rental vehicles is represented by a real number code. For example, x4=100 means that the number of allocated rental vehicles in the 4th sub-area is 100.

[0093] The velocity is encoded by the vector V = (v1…v M} indicates that the value of each dimension of the velocity vector is Z% of the value of each dimension of the position vector, where Z is a custom parameter. Preferably, the value range of Z is 10 to 20;

[0094] Step S302, constructing an objective function in combination with a regional demand forecasting model;

[0095] The objective function is: where xm represents the number of allocated rental vehicles in the mth sub-area, y m represents the number of car rental demands in the mth sub-region predicted by the regional demand model;

[0096] Step S303, calculating the fitness values ​​of all particles in the particle swarm according to the objective function;

[0097] Step S304, updating the particle positions and velocities in the next generation particle swarm according to the particle swarm algorithm update strategy and the t-distribution perturbation mutation strategy;

[0098] The particle velocity update formula is: in represents the velocity of the mth particle with iteration number j+1, represents the velocity of the mth particle with the number of iterations j, ω represents the inertia weight, c1 and c2 represent the learning factors, r1 and r2 represent random numbers with a value range of 0 to 1, represents the optimal position of the mth particle in history when the number of iterations is j, g j represents the global optimal position of the entire particle swarm, τ represents the disturbance intensity control parameter, and t γ 0, σ represents a t-distributed random variable with mean 0 and scale σ, and γ represents the degrees of freedom of the t-distribution;

[0099] Step S305, if the fitness value of each particle in the initialized particle swarm with the current iteration number j is determined to be greater than the maximum fitness value of the particle in history, the maximum fitness value of the particle in history is replaced by the fitness value of the particle with the current iteration number j;

[0100] The maximum fitness value in the initialized particle swarm with the current iteration number j is taken as the global optimal solution;

[0101] Step S306, the current number of iterations is increased by 1, and if it is determined that the current number of iterations reaches the maximum number of iterations, the global optimal solution is output as the allocation solution, otherwise the process returns to step S303, wherein the maximum number of iterations is a custom parameter, and preferably, the maximum number of iterations is set to 200.

[0102] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. An AI intelligent car rental operation decision system, characterized in that: include: A data collection module, which is used to collect regional feature data of M sub-regions in a region G within a preset time period T according to a preset time interval t, and represent the regional feature data at each time point by a first feature sequence; The mth sequence unit of the first feature sequence at the nth time point represents the regional feature data of the mth sub-region at the nth time point, where 1≤m≤M, 1≤n≤N, and N=T / t; Regional characteristic data include: number of rental orders, number of rental returns, rental duration, regional population density, number of subway stations, and number of bus stops; A preprocessing module, which is used to preprocess the first feature sequence to obtain a second feature sequence; A graph structure construction module, which is used to construct a regional graph structure according to the second feature sequence, the regional graph structure including nodes, initial features of nodes and edges between nodes; nodes are represented by sub-regions, and the initial features of nodes are represented by regional feature data of sub-regions that establish a mapping relationship with nodes; A regional demand prediction module is used to construct a regional demand prediction model based on the regional graph structure. The input of the regional demand prediction model is the regional graph structure, and the output is the number of car rental demands in the sub-region. The allocation plan generation module is used to generate an allocation plan based on the regional demand forecasting model combined with the particle swarm algorithm that introduces the t-distribution perturbation mutation strategy. The allocation plan indicates that P vehicles in the area G are allocated to M sub-areas at the second preset time Q.

2. According to claim 1, an AI intelligent car rental operation decision system is characterized in that: T, t, M, m, G, N, n, P, and Q are custom parameters.

3. The AI ​​intelligent car rental operation decision system according to claim 1 is characterized in that: The preprocessing steps include: Step S201, for missing values ​​in the regional feature data, filling processing is performed by taking the average value of the value of the previous sequence unit and the value of the next sequence unit; Step S202, replacing the abnormal value in the regional feature data by taking the average value of the values ​​of other sequence units except the abnormal value; Step S203: normalize the regional feature data using a z-score normalization method.

4. The AI ​​intelligent car rental operation decision system according to claim 1 is characterized in that: The edges are constructed in the following way: Build edges based on geographic proximity. That is, when two sub-regions are adjacent, build an edge between them. The weight of the edge based on geographic proximity is calculated by Calculated, where δ A,B represents the weight of edge AB based on geographical proximity, H A,B represents the distance between sub-region A and sub-region B, It represents the minimum geographical distance between all sub-regions, and ε represents the geographical proximity adjustment factor, which ranges from 1 to 10; The edge is constructed based on the mobility of rental vehicles. That is, when there is a flow of rental vehicles between two sub-areas, an edge is constructed between the two sub-areas. The weight of the edge based on vehicle mobility is calculated by Calculated, where μ A,B represents the weight of edge AB based on vehicle mobility, C A,B represents the number of rental vehicles rented out from sub-area A and returned to sub-area B during the previous preset time interval, S A,B represents the number of rental vehicle flows between all sub-areas during the previous preset time interval, max(f) represents the value with the highest flow frequency between all sub-areas during the previous preset time interval, and the flow frequency is expressed by It indicates that α represents the vehicle mobility adjustment factor, and its value range is from 0 to 1; Build edges based on public transportation. That is, when there is a direct public transportation connection between two sub-areas, build an edge between the two sub-areas. The public transportation-based weight of the edge is calculated by Calculated, where D A,B represents the weight of edge AB based on public transportation, D A,B represents the number of public transportation lines from sub-area A to sub-area B, L A,B represents the geographical distance between sub-region A and sub-region B, max(D) represents the maximum value of public transportation density among all sub-regions, and the public transportation density is expressed by express, It represents the public transportation density adjustment factor, ranging from 1 to 10; The combined weight of edge AB is given by W A,B =w1δ A,B +w2μ A,B +w3θ A,B It is obtained by calculation, wherein w1, w2 and w3 represent the first adjustment coefficient, the second adjustment coefficient and the third adjustment coefficient respectively, and the value range is 0 to 1.

5. The AI ​​intelligent car rental operation decision system according to claim 1 is characterized in that: The regional demand forecasting model includes: a first hidden layer, a concatenation layer, a second hidden layer, a reduction layer and a classifier; The first hidden layer includes N first units, the nth first unit inputs the region graph structure and outputs the first sub-region features of M nodes; The concatenation layer includes N concatenation units, and the nth concatenation unit is used to concatenate the first updated features output by the N first units into an integrated vector; The second hidden layer includes N second units, the nth second unit inputs the integrated vector output by the nth splicing unit, and outputs the second sub-region feature; The restoration layer is used to restore the second sub-region feature output by the Nth second unit into M third sub-region features, where each third sub-region feature corresponds to a node update feature of a node; The node update features of each node are input into the classifier, and the classification space of the classifier represents the number of car rental demands in each sub-region.

6. The AI ​​intelligent car rental operation decision system according to claim 5 is characterized in that: The nth first unit of the first hidden layer of the regional demand forecasting model is used to process the regional graph structure at the nth time point. The formula of the nth first unit of the first hidden layer is include: in represents the first sub-region feature of the mth node, H m represents the initial feature of the mth node, W m Represents the first weight parameter corresponding to the mth node, represents the weighted adjacency matrix, represents the normalized degree matrix, Represents the degree matrix. The size of the degree matrix is ​​M×M. The elements of the degree matrix represent the number of edges connected to the node. W i,k represents the comprehensive weight of the edge between node i and node k, 1(i, k) represents the indicator function, which is used to determine whether there is an edge between node i and node k, and ReLU represents the ReLU activation function.

7. The AI ​​intelligent car rental operation decision system according to claim 5 is characterized in that: The N second units of the second hidden layer are constructed based on gated recurrent units.

8. The AI ​​intelligent car rental operation decision system according to claim 1 is characterized in that: The sample labels of the regional demand forecasting model are obtained through manual labeling.

9. The AI ​​intelligent car rental operation decision system according to claim 1 is characterized in that: The loss function of the regional demand forecasting model is the cross entropy loss function.

10. The AI ​​intelligent car rental operation decision system according to claim 1 is characterized in that: The steps of generating an allocation plan based on the regional demand forecasting model combined with the particle swarm algorithm that introduces the t-distribution perturbation mutation strategy include: Step S301, randomly generate an initial particle population with a particle number U that meets the constraint conditions; The constraints include: the sum of the number of rental vehicles allocated to each sub-region is P; the number of rental vehicles allocated to each sub-region is less than the number of rental vehicles demanded in this sub-region predicted by the regional demand forecasting model; Each particle includes a position code and a velocity code; The position encoding is done by a vector X = {x1…x M } means that x1…x M The number of allocated rental vehicles from the 1st sub-area to the Mth sub-area is represented by real number coding; the speed coding is represented by vector V = {v1…v M } means that the value of each dimension of the velocity vector is Z% of the value of each dimension of the position vector, where Z is a custom parameter; Step S302, constructing an objective function in combination with a regional demand forecasting model; The objective function is: where x m represents the number of allocated rental vehicles in the mth sub-area, y m represents the number of car rental demands in the mth sub-region predicted by the regional demand model; Step S303, calculating the fitness values ​​of all particles in the particle swarm according to the objective function; Step S304, updating the particle positions and velocities in the next generation particle swarm according to the particle swarm algorithm update strategy and the t-distribution perturbation mutation strategy; Step S305, if the fitness value of each particle in the initialized particle swarm with the current iteration number j is determined to be greater than the maximum fitness value of the particle in history, the maximum fitness value of the particle in history is replaced by the fitness value of the particle with the current iteration number j; The maximum fitness value in the initialized particle swarm with the current iteration number j is taken as the global optimal solution; Step S306, the current number of iterations is increased by 1, and if it is determined that the current number of iterations reaches the maximum number of iterations, the global optimal solution is output as the allocation solution, otherwise, the process returns to step S303, where the maximum number of iterations is a custom parameter.

Images