A method and device for staggered scheduling of a rolling time domain for a scenic spot bus and a medium

Abstract

The application discloses a kind of rolling time domain peak-shaving scheduling method, device and medium for scenic spot bus, comprising: obtaining the operation data and external influence data of scenic spot bus system before current decision time, according to the transfer relationship of scenic spot bus line structure, road connection relationship and tourist between different stations, construct scenic spot bus network diagram, construct space-time input sample according to operation data and external influence data;Scenic spot bus network diagram and space-time input sample are input into space-time diagram neural network passenger flow prediction submodel, obtain the passenger flow prediction result in future rolling window;According to passenger flow prediction result, build departure matrix, input rolling time domain departure matrix optimization submodel to solve, obtain the optimal departure matrix of each line in future rolling window under current decision time, and extract the departure decision corresponding to current time slice in optimal departure matrix as actual execution scheme.

Description

Technical Field

[0001] This invention relates to the field of rolling time-domain off-peak scheduling technology for scenic area buses, and more specifically, to a rolling time-domain off-peak scheduling method, device, and medium for scenic area buses. Background Technology

[0002] Existing rolling time-domain off-peak scheduling methods for public transportation to scenic areas mostly rely on historical passenger flow statistics, fixed departure intervals, or on-site scheduling experience to formulate departure plans. Their responsiveness to real-time factors such as high-speed rail arrival schedules, scenic area reservation numbers, and weather changes is limited. Some existing technologies that have incorporated passenger flow forecasting primarily rely on single-site time-series forecasting or single-route passenger flow forecasting, failing to adequately consider the spatial relationships between high-speed rail stations, bus stations, scenic area passenger transport centers, and scenic area connecting stations, making it difficult to depict the overall passenger flow evolution of the scenic area's public transportation system. Furthermore, departure plans for each route are often... The existing scheduling system cannot achieve coordinated optimization among multiple routes, time slots, and upstream and downstream transport capacity. When passenger flow arrives in large numbers at upstream source stations, if buses are dispatched intensively to the scenic area's passenger transport center, it can easily cause congestion in the transfer and waiting areas. If the frequency of departures is reduced, it may lead to an increase in the number of people waiting at the source stations, affecting the travel experience of tourists. In addition, the existing scheduling system lacks the ability to identify and proactively respond to sudden factors such as the concentrated arrival of high-speed trains, the large-scale arrival of tour groups, and weather changes. The scheduling decision-making is lagging and it is difficult to adapt to the distribution pressure on the scenic area's passenger transport center during peak passenger flow periods.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This invention provides a rolling time-domain off-peak scheduling method, device, and medium for scenic area public transportation, which at least solves the technical problems of existing rolling time-domain off-peak scheduling methods for scenic area public transportation that are difficult to dynamically schedule according to the overall passenger flow evolution of the scenic area public transportation network, cannot coordinate and optimize the capacity configuration of upstream and downstream multi-line routes, and lack the ability to proactively respond to future passenger flow changes.

[0005] According to one aspect of the present invention, in order to achieve the above-mentioned objective, a rolling time-domain off-peak scheduling method for scenic area public transportation is provided, comprising:

[0006] Obtain operational data and external impact data of the scenic area's public transportation system prior to the current decision-making time. Construct a scenic area public transportation network map based on the scenic area's public transportation route structure, road connectivity, and tourist transfer relationships between different stops. Construct spatiotemporal input samples based on the operational data and external impact data.

[0007] Input the scenic area bus network map and spatiotemporal input samples into the spatiotemporal graph neural network passenger flow prediction sub-model to obtain the passenger flow prediction results within the future rolling window.

[0008] Based on the passenger flow forecast results, a departure matrix is ​​constructed. The departure matrix is ​​then input into the rolling time-domain departure matrix optimization sub-model for solution. The optimal departure matrix for each route within the future rolling window at the current decision time is obtained, and the departure decision corresponding to the current time slice in the optimal departure matrix is ​​extracted as the actual execution plan.

[0009] Furthermore, obtain operational data and external impact data of the scenic area's public transportation system prior to the current decision-making moment, including:

[0010] The system acquires data on the number of tourists arriving, waiting, boarding and alighting, number of passengers stranded at each station, number of vehicle arrivals and departures, and route operating time over multiple consecutive time periods. It also acquires data on weather changes, holidays, number of reservations at scenic spots, high-speed rail arrival plans, tour bus arrival plans, and large-scale event arrangements as external impact data to obtain operational data and external impact data.

[0011] Furthermore, based on the scenic area's bus route structure, road connections, and tourist transfer patterns between different stops, a scenic area bus network map is constructed, including:

[0012] The high-speed rail station, bus station, tourist distribution center, scenic area passenger transport center, and scenic area shuttle station are abstracted as nodes in the graph. The bus route connections, road accessibility relationships, or stable passenger flow transfer relationships between the stations are abstracted as edges in the graph. The edge weight construction method is selected according to the data completeness to assign values ​​to the edges in the graph. The weighted adjacency matrix after assignment is normalized to obtain the scenic area bus network graph.

[0013] Furthermore, based on the completeness of the data, the corresponding edge weight construction method is selected to assign values ​​to the edges in the graph, including:

[0014] When stable historical passenger transfer data or passenger flow transfer data between stations can be obtained, the historical passenger flow transfer intensity is used as the edge weight; when reliable passenger flow transfer data is lacking but vehicle operation records can be obtained, the average running time between stations is used as the edge weight; when only basic road network data is available, the road distance between stations is used as the edge weight.

[0015] Furthermore, the scenic area's public transportation network map and spatiotemporal input samples are input into the spatiotemporal graph neural network passenger flow prediction sub-model to obtain the passenger flow prediction results within the future rolling window, including:

[0016] For each node in the scenic area's public transportation network, the passenger flow and operational status features of that node in multiple consecutive time slices are extracted. Combined with external influencing factors, feature representations of each node in different time slices are formed and organized into a spatiotemporal input tensor in chronological order.

[0017] The spatiotemporal input tensor is input into the spatiotemporal graph neural network passenger flow prediction sub-model. The spatial correlation features between stations are extracted using a graph convolutional network, and the dynamic features of passenger flow changes over time are extracted using a gated recurrent unit. The passenger flow prediction results for multiple future time slices are output, and the passenger flow prediction results within the future rolling window are obtained.

[0018] Furthermore, a departure matrix is ​​constructed based on passenger flow forecast results, including:

[0019] Construct a departure matrix with bus routes as rows and discrete time slices within a future rolling window as columns. Each element in the departure matrix corresponds to the planned number of departures for a route within a future time slice, thus obtaining the departure matrix.

[0020] Based on passenger flow forecast results, a passenger flow state evolution process is established, so that the future waiting state of the source station is jointly determined by the number of existing waiting passengers, the number of new arrivals in the future, and the transportation capacity provided by the route departures, and the future congestion state of the scenic area passenger transport center is jointly determined by the input passenger flow formed by the upstream transport lines, the output passenger flow formed by the downstream connecting lines, and the congestion state at the current moment.

[0021] Furthermore, the departure matrix is ​​input into the rolling time-domain departure matrix optimization sub-model for solution, to obtain the optimal departure matrix for each route within the future rolling window at the current decision time, including:

[0022] With the goals of reducing tourist waiting congestion, alleviating congestion at scenic area passenger transport centers, controlling public transport operating costs, and smoothing the input flow to scenic area passenger transport centers, the rolling time-domain departure matrix optimization sub-model is solved by inputting passenger flow forecast results, passenger flow state evolution process, and actual operational constraints together to obtain the optimal departure matrix for each route within the future rolling window at the current decision time.

[0023] Furthermore, the actual operational constraints include at least the following: the maximum number of departures allowed on a single route within a single time slot, the minimum departure interval between adjacent departures on the same route, the number of vehicles that can be called up within a time slot, the capacity constraints of the scenic area passenger transport center, and the vehicle turnaround constraint that vehicles must undergo a turnaround period after completing a departure task before they can be put into a new task.

[0024] According to one embodiment of the present invention, a rolling time-domain off-peak scheduling device for scenic area buses is also provided, comprising:

[0025] The acquisition module is used to acquire the operational data and external impact data of the scenic area's public transportation system before the current decision-making time. Based on the scenic area's public transportation route structure, road connection relationships, and the transfer relationships of tourists between different stations, a scenic area public transportation network map is constructed. Spatiotemporal input samples are constructed based on the operational data and external impact data.

[0026] The prediction module is used to input the scenic area bus network map and spatiotemporal input samples into the spatiotemporal graph neural network passenger flow prediction sub-model to obtain the passenger flow prediction results within the future rolling window;

[0027] The optimization module is used to construct a departure matrix based on passenger flow forecast results. The departure matrix is ​​input into the rolling time-domain departure matrix optimization sub-model for solution to obtain the optimal departure matrix for each route in the future rolling window at the current decision time. The departure decision corresponding to the current time slice in the optimal departure matrix is ​​extracted as the actual execution plan.

[0028] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is executed, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of the present invention.

[0029] In this embodiment of the invention, all stations of the scenic area's public transportation network are uniformly modeled and a spatiotemporal graph neural network is used for passenger flow prediction. This fully utilizes the spatial relationships between stations and the dynamic changes in passenger flow, effectively improving the accuracy of short-term passenger flow prediction. The future congestion status of the scenic area's passenger transport center is incorporated into the departure optimization decision, enabling the scheduling plan to simultaneously consider the connection needs of the source stations and the capacity pressure of the passenger transport center. By actively adjusting the upstream transport rhythm and downstream evacuation capacity, staggered arrival times for tourists at the scenic area's passenger transport center are achieved, reducing the risk of short-term passenger flow concentration at the passenger transport center. A rolling time-domain optimization method is adopted, executing the departure plan only for the current time slot and re-predicting and optimizing based on the latest data in subsequent time slots. This system enables the scheduling scheme to be dynamically updated according to changes in high-speed rail arrival times, weather conditions, and actual passenger flow deviations. This improves the real-time performance and adaptability of the rolling time-domain off-peak scheduling for scenic area public transportation. At the same time, it optimizes the waiting costs at source stations, congestion penalties at passenger transport centers, operating costs, and fluctuations in input flow through a unified objective function. This enhances the overall stability of the scenic area public transportation connection system while ensuring the travel experience of tourists and the economic efficiency of operations. In turn, it solves the technical problems of existing rolling time-domain off-peak scheduling methods for scenic area public transportation, such as the inability to dynamically schedule according to the overall passenger flow evolution of the scenic area public transportation network, the inability to coordinate and optimize the capacity configuration of upstream and downstream multi-line routes, and the lack of proactive response capabilities to future passenger flow changes. Attached Figure Description

[0030] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0031] Figure 1 This is a flowchart of a rolling time-domain off-peak scheduling method for scenic area buses according to one embodiment of the present invention;

[0032] Figure 2 This is a structural block diagram of a rolling time-domain off-peak scheduling device for scenic area buses according to one embodiment of the present invention. Detailed Implementation

[0033] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0034] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0035] According to an embodiment of the present invention, a rolling time-domain off-peak scheduling method for scenic area buses is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0036] This method embodiment can be executed in an electronic device or similar computing device that includes a memory and a processor. Taking operation on a terminal as an example, the terminal may include one or more processors (processors may include, but are not limited to, central processing units (CPUs), graphics processing units (GPUs), digital signal processing (DSP) chips, microcontroller units (MCUs), field-programmable gate arrays (FPGAs), neural network processors (NPUs), tensor processors (TPUs), artificial intelligence (AI) type processors, etc.) and a memory for storing data. Optionally, the terminal may also include transmission devices, input / output devices, and display devices for communication functions. Those skilled in the art will understand that the above structural description is merely illustrative and does not limit the structure of the terminal. For example, the terminal may include more or fewer components than described above, or have a different configuration than described above.

[0037] The memory can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the rolling time-domain off-peak scheduling method for scenic area buses in this embodiment of the invention. The processor executes various functional applications and data processing by running the computer program stored in the memory, thereby realizing the aforementioned rolling time-domain off-peak scheduling method for scenic area buses. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the mobile terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0038] The transmission device is used to receive or send data via a network. Specific examples of the network mentioned above may include a wireless network provided by the mobile terminal's communication provider. In one example, the transmission device includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0039] Display devices can be, for example, touchscreen liquid crystal displays (LCDs) and touch displays (also referred to as "touchscreens" or "touch displays"). The LCD allows users to interact with the user interface of the mobile terminal. In some embodiments, the mobile terminal has a graphical user interface (GUI), which allows users to interact with the GUI through finger contact and / or gestures on a touch-sensitive surface. Optional human-computer interaction functions include: creating web pages, drawing, word processing, creating electronic documents, playing games, video conferencing, instant messaging, sending and receiving emails, call interfaces, playing digital video, playing digital music, and / or web browsing, etc. Executable instructions for performing the above human-computer interaction functions are configured / stored in one or more processor-executable computer program products or readable storage media.

[0040] Figure 1 This is a flowchart of a rolling time-domain off-peak scheduling method for scenic area buses according to one embodiment of the present invention, as follows: Figure 1 As shown, the method includes the following steps:

[0041] Step S110: Obtain the operational data and external impact data of the scenic area's public transportation system prior to the current decision-making time. Based on the scenic area's public transportation route structure, road connectivity, and tourist transfer relationships between different stops, construct a scenic area public transportation network map. Construct spatiotemporal input samples based on the operational data and external impact data. The specific content is as follows:

[0042] In step S110, the operational data and external impact data of the scenic area's public transportation system prior to the current decision-making time are acquired. The operational data includes at least: the number of tourists arriving, waiting, boarding and alighting, and stranded at each station in the scenic area's public transportation network over multiple consecutive time slices; the number of vehicle arrivals and departures; and the operating time of each route. The external impact data includes at least: weather change information, holiday types, number of reservations for the scenic area, high-speed rail arrival plans, tour bus arrival plans, and large-scale event arrangements. In the actual system, the number of passengers boarding and alighting can be obtained from bus card swipe records, electronic ticketing records, onboard passenger flow counting devices, platform video recognition devices, gate data, or boarding records in the rolling time-domain off-peak scheduling system for scenic area public transportation; high-speed rail arrival plans and tour bus arrival plans are used to characterize the potential concentrated arrival of large numbers of tourists at upstream source stations; the number of reservations for the scenic area and the type of holiday are used to characterize the changing trend of overall demand intensity in the scenic area direction.

[0043] Based on the scenic area's bus route structure, road connections, and tourist transfer relationships between different stations, a scenic area bus network map is constructed. Specifically, high-speed rail stations, bus stations, tourist distribution centers, scenic area passenger transport centers, and scenic area shuttle stations are abstracted as nodes in the graph, and the bus route connections, road accessibility relationships, or stable passenger flow transfer relationships between stations are abstracted as edges in the graph. When determining edge weights, the corresponding edge weight construction method is selected according to the completeness of the actual data: if stable historical boarding and alighting data or passenger flow transfer data between stations are available, historical passenger flow transfer intensity is used as the edge weight, i.e., the proportion of the number of tourists transferring from one station to another via the bus system within the statistical period to the total number of tourists transferring out of that station by bus is assigned a value; if reliable passenger flow transfer data is lacking but vehicle operation records are available, the average travel time between stations is used as the edge weight; if only basic road network data is available, the road distance between stations is used as the edge weight. When using average travel time or road distance as edge weights, the edge weights are negatively correlated with travel time or road distance. After assigning edge weights, the resulting weighted adjacency matrix and self-connection matrix are merged and degree normalized to obtain a normalized adjacency matrix, thus forming a scenic area bus network graph that can be used by spatiotemporal graph neural networks.

[0044] Spatiotemporal input samples are constructed based on the collected operational data and external influence data. For each node in the scenic area's public transportation network, passenger flow and operational status features are extracted across multiple consecutive time slices, including the number of arriving tourists, waiting passengers, and boarding / departing passengers at that station. Simultaneously, external influencing factors such as weather, holidays, scenic area reservations, and high-speed rail arrival schedules are combined with the station features to form a complete feature representation of each node under different time slices. Specifically, the input feature vector for each node consists of the number of newly arriving tourists, the number of waiting passengers, the number of people who board and leave the station via the scenic area's public transportation system, and external features. The number of waiting passengers at the tourist origin station represents the number of tourists waiting to go to the scenic area's passenger transport center, while at the scenic area's passenger transport center, it represents the number of tourists waiting to be evacuated to various scenic areas. Subsequently, the node features from multiple consecutive time slices are organized into a spatiotemporal input tensor in chronological order. In constructing input samples, for continuous time series data such as the number of tourists arriving, waiting, and boarding / alighting, if a single time slice is missing, the valid value from the previous time slice is used for forward completion; if multiple consecutive time slices are missing, the historical average of the same type of date and time slice is used for completion; for outliers that significantly deviate from the historical normal range, correction is performed by combining vehicle operation logs, ticketing records, or video recognition results; if correction is not possible, the historical quantile range is used for truncation; continuous numerical features are standardized using the mean and standard deviation obtained from training sample statistics; and categorical features such as weather type, holiday type, and large-scale event type are converted into inputtable numerical features using one-hot encoding, thus enabling data from different time periods, different stations, and different units to be uniformly input into the spatiotemporal graph neural network.

[0045] Step S120: Input the scenic area bus network map and spatiotemporal input samples into the spatiotemporal graph neural network passenger flow prediction sub-model to obtain the passenger flow prediction results within the future rolling window. The specific content is as follows:

[0046] In step S120, the constructed scenic area bus network map and the corresponding spatiotemporal input samples are input into the spatiotemporal graph neural network passenger flow prediction sub-model. The spatiotemporal graph neural network passenger flow prediction sub-model adopts a spatiotemporal graph neural network structure combining a graph convolutional network and a gated recurrent unit. The graph convolutional network is used to extract the spatial correlation features between stations in the scenic area bus network, while the gated recurrent unit is used to extract the dynamic features of passenger flow changes over time. Internally, the feature vectors of each node in the spatiotemporal input samples are first subjected to graph convolution with the normalized adjacency matrix of the scenic area bus network map. This allows the feature representation of each station to incorporate the passenger flow status information of its neighboring stations, thereby depicting the spatial coupling relationship formed between high-speed rail stations, bus stations, tourist distribution centers, scenic area passenger transport centers, and scenic area connecting stations due to bus route connections and tourist transfer behavior. Subsequently, the hidden state sequence output by the graph convolution is input into the gated recurrent unit to capture the evolution pattern of passenger flow over continuous time slices, enabling the model to perceive the upward trend, downward trend, and periodic fluctuation patterns of passenger flow demand. Through the joint modeling of the spatial and temporal dimensions, the model can comprehensively predict the passenger flow status of each time slice and each station within the future rolling window.

[0047] During the prediction process, the model input is a spatiotemporal input tensor of multiple consecutive time slices prior to the current decision moment. Each node in this tensor contains features including at least the number of newly arriving tourists, the number of people waiting for transportation, the number of people who boarded and left the station via the scenic area's public transportation system, and external features such as weather, holidays, high-speed rail arrival schedules, and scenic area reservations. For tourist origin stations, the model output prediction results reflect the changing trends in the number of newly arriving tourists and the cumulative number of people waiting for transportation in future time slices. For scenic area passenger transport centers, the model output prediction results reflect the changing trends in the scale of arriving passenger flow brought by upstream transportation lines and the number of people stranded in the passenger transport centers in future time slices. For scenic area connecting stations, the model output prediction results reflect the tourist evacuation needs from various scenic area directions. During the processing, the gated recurrent units of the graph convolutional network recursively calculate the time dimension of the input tensor in turn. Each recursion uses a normalized adjacency matrix to aggregate the spatial information of adjacent stations. Finally, the hidden state is converted into the passenger flow prediction value of each time slice in the future rolling window through a fully connected mapping layer, thereby obtaining a complete passenger flow prediction result that includes the waiting status of each tourist origin station, the congestion status of the scenic area passenger transport center, and the transfer demand of each scenic area direction.

[0048] The spatiotemporal graph neural network passenger flow prediction sub-model needs to be pre-trained before deployment. During training, the actual passenger flow status in historical operational data is used as the supervision label, and the mean squared error loss function is used to optimize the model parameters so that the predicted values ​​output by the model are as close as possible to the actual observed values. After training, in the actual scheduling process, the model runs in forward inference mode, directly outputting the passenger flow prediction results within the future rolling window based on the real-time data collected at the current moment, without the need to update the network parameters online, thus ensuring the real-time performance of the prediction and computational efficiency.

[0049] Step S140: Construct a departure matrix based on passenger flow forecast results. Input the departure matrix into the rolling time-domain departure matrix optimization sub-model for solution to obtain the optimal departure matrix for each route within the future rolling window at the current decision time. Extract the departure decision corresponding to the current time slice from the optimal departure matrix as the actual execution plan. The specific content is as follows:

[0050] In step S140, a departure matrix is ​​constructed based on the passenger flow forecast results within the future rolling window. The departure matrix uses bus routes as rows and discrete time slices within the future rolling window as columns. Each element in the matrix corresponds to the planned number of departures for a route within a given future time slice. This matrix representation integrates upstream transport routes from tourist origin stations to the scenic area passenger transport center and downstream evacuation routes from the scenic area passenger transport center to various scenic area connecting stations into a unified scheduling framework. When constructing the departure matrix, a vehicle allocation variable is introduced to describe whether each vehicle is assigned to perform departure tasks for a particular route in each future time slice. Therefore, the planned number of departures for each route in each time slice is equal to the sum of the total number of vehicles assigned to perform departure tasks for that route in that time slice, thus establishing a clear correspondence between the departure matrix and vehicle resources. Multiplying the single-shift capacity of each route by the planned number of departures for that route yields the total capacity provided by that route within that time slice.

[0051] The completed departure matrix is ​​input into the rolling time-domain departure matrix optimization sub-model for solution. Before optimization, the evolution process of passenger flow status is first established based on passenger flow forecast results, so that a clear correspondence is formed between the departure plan and passenger flow changes. For tourist origin stations, their future waiting status is jointly determined by the number of people currently waiting, the number of new arrivals, and the capacity provided by upstream routes. That is, the number of people waiting in the next time slot is equal to the sum of the current number of people waiting and the number of new arrivals minus the number of people actually transported away by public transport capacity in that time slot. For scenic area passenger transport centers, their future lingering status is jointly determined by the input passenger flow formed by upstream transport routes, the output passenger flow formed by downstream connecting routes, and the current lingering status. That is, the number of lingering people in the next time slot is equal to the sum of the current number of lingering people and the upstream input passenger flow minus the downstream output passenger flow. Among them, the upstream input passenger flow is determined by the travel time and single-shift capacity of each route departing from the origin station. After the vehicles depart from the origin station, they arrive at the scenic area passenger transport center after the corresponding travel time, forming the input passenger flow to the scenic area passenger transport center in that time slot. Through the above-described state evolution process, the impact of upstream route departure plans on the inflow to the scenic area passenger transport center and the impact of downstream route departure plans on the evacuation capacity of the passenger transport center are uniformly described.

[0052] Based on the established passenger flow state evolution process, the future passenger flow forecast results, state evolution relationships, and actual operational constraints are jointly input into the rolling time-domain departure matrix optimization sub-model for solution. The rolling time-domain departure matrix optimization sub-model aims to reduce passenger waiting congestion at source stations, alleviate congestion at scenic area passenger transport centers, control public transport operating costs, and smooth the input flow to scenic area passenger transport centers, constructing a unified objective function. Specifically, the first term of the objective function is the weighted sum of the number of passengers waiting at each source station within the future rolling window, reflecting the passenger waiting cost; the second term is the sum of squares of the number of passengers exceeding the capacity threshold at the scenic area passenger transport center, used to suppress overload operation of the passenger transport center; the third term is the weighted sum of the number of departures from each route, used to control operating costs; and the fourth term is the sum of squares of the differences in input flow between adjacent time slots at the scenic area passenger transport center, used to smooth the upstream input rhythm and avoid a large number of tourists arriving at the passenger transport center in a short period. The weight coefficients can be adjusted according to actual operational needs to reflect the emphasis on different optimization objectives.

[0053] The rolling time-domain departure matrix optimization sub-model needs to satisfy several practical operational constraints during the solution process. The constraint on the maximum number of departures allowed per route within a single time slice limits the departure intensity within a unit time slice, primarily determined by platform berth capacity, road capacity, vehicle organization capacity, and dispatch management capacity. The constraint on the minimum departure interval between adjacent departures on the same route avoids overly dense departures, specifically ensuring that the cumulative number of departures over multiple consecutive time slices does not exceed the maximum allowable value. The constraint on the number of vehicles available within a time slice ensures that the total number of vehicles scheduled to perform departure tasks within any time slice does not exceed the number of available vehicles in the system for that time slice. The constraint on the capacity of the scenic area passenger transport center limits or penalizes situations where the number of passengers overstaying at the passenger transport center exceeds safety or service level requirements. To ensure the physical feasibility of the departure scheme output by the model, the original departure matrix is ​​projected onto the feasible region to obtain a modified departure matrix. The feasible region includes at least vehicle availability constraints, vehicle turnaround constraints, and departure origin matching constraints. The vehicle turnaround constraint ensures that after a vehicle completes a departure task on a certain route, it must undergo route travel time, passenger pick-up and drop-off time, and necessary preparation time before it can be assigned to a new departure task. The departure origin matching constraint ensures that when a vehicle is assigned to a route, its location should be consistent with the departure origin of that route. Through projection of the feasible region, the final output departure plan is matched with the actual vehicle resource location and status.

[0054] After completing the above optimization solution, the optimal departure matrix for each route within the future rolling window at the current decision time is obtained. Under the rolling time-domain control method, the entire departure plan for the entire rolling window is not executed all at once. Instead, only the departure decision corresponding to the current time slice in the optimal departure matrix is ​​extracted as the actual execution plan; the departure arrangements for other future time slices are only retained as reference plans generated under the current prediction conditions. When entering the next time slice, the latest passenger flow data, vehicle operating status, and external influence information are collected again, new input samples are constructed, and the passenger flow prediction and departure matrix optimization process is repeated, thereby achieving continuous rolling updates of the rolling time-domain off-peak scheduling scheme for scenic area public transportation. The actual departure times of each bus route within the current time slice are output, along with the reference departure matrix within the future rolling window, the predicted waiting status of each station, the predicted congestion status of the scenic area passenger transport center, and the predicted transfer demand for each direction within the scenic area. This information is used to guide the allocation of capacity for scenic area public transportation between different time periods and different routes, enabling tourists to arrive at the scenic area passenger transport center during off-peak hours.

[0055] Based on steps S110 to S140 above, in this embodiment of the invention, all stations of the scenic area bus network are uniformly modeled and a spatiotemporal graph neural network is used for passenger flow prediction. This fully utilizes the spatial correlation between stations and the dynamic changes in passenger flow, effectively improving the accuracy of short-term passenger flow prediction. The future congestion status of the scenic area passenger transport center is incorporated into the departure optimization decision, enabling the scheduling scheme to simultaneously consider the connection needs of the source stations and the capacity pressure of the passenger transport center. By actively adjusting the upstream transport rhythm and downstream evacuation capacity, staggered arrival of tourists at the scenic area passenger transport center is achieved, reducing the risk of short-term passenger flow concentration at the passenger transport center. A rolling time-domain optimization method is adopted, executing only the departure scheme for the current time slice and adjusting it based on the latest data in subsequent time slices. By re-predicting and optimizing, the scheduling scheme can be dynamically updated according to changes in high-speed rail arrivals, weather changes, and actual passenger flow deviations. This improves the real-time performance and adaptability of the rolling time-domain off-peak scheduling of scenic area buses. At the same time, by using an objective function to uniformly optimize the waiting costs at source stations, congestion penalties at passenger transport centers, operating costs, and fluctuations in input flow, the overall stability of the scenic area bus connection system is improved while ensuring the travel experience of tourists and the economic efficiency of operation. This solves the technical problems of the existing rolling time-domain off-peak scheduling method for scenic area buses, which is unable to dynamically schedule according to the overall passenger flow evolution of the scenic area bus network, cannot coordinate and optimize the capacity configuration of upstream and downstream multi-line routes, and lacks the ability to proactively respond to future passenger flow changes.

[0056] The rolling time-domain off-peak scheduling method for scenic area public transportation in this invention acquires operational data and external influence data of the scenic area public transportation system prior to the current decision-making time. This includes acquiring data on the number of arriving tourists, waiting passengers, boarding and alighting passengers, stranded passengers at each station, vehicle arrival and departure times, and route operating times for each station over multiple consecutive time slices. It also acquires data on weather changes, holidays, number of reservations at the scenic area, high-speed rail arrival plans, tour bus arrival plans, and large-scale event arrangements as external influence data, thus obtaining both operational data and external influence data. By fusing multi-source heterogeneous data, a comprehensive understanding of the internal operational status and external influencing factors of the scenic area public transportation system is achieved, providing a complete data foundation for subsequent networked passenger flow prediction and collaborative scheduling optimization.

[0057] Furthermore, based on the scenic area's bus route structure, road connectivity, and tourist transfer relationships between different stations, a scenic area bus network graph is constructed. This includes: abstracting high-speed rail stations, bus stations, tourist distribution centers, scenic area passenger transport centers, and scenic area shuttle stations as nodes in the graph; abstracting the bus route connections, road accessibility relationships, or stable passenger flow transfer relationships between stations as edges in the graph; assigning values ​​to the edges in the graph according to the data completeness using the appropriate edge weight construction method; and normalizing the weighted adjacency matrix after assignment to obtain the scenic area bus network graph. By uniformly abstracting multiple types of stations into network nodes and adaptively selecting edge weight construction methods based on data completeness, a flexible and computable graph modeling of the spatial association structure of the scenic area bus system is achieved, laying a structural foundation for the subsequent effective extraction of passenger flow coupling relationships between stations by spatiotemporal graph neural networks.

[0058] Furthermore, the edge weights in the graph are assigned values ​​based on the data completeness, including: when stable historical passenger flow data or passenger transfer data between stations is available, historical passenger flow intensity is used as the edge weight; when reliable passenger flow data is lacking but vehicle operation records are available, the average travel time between stations is used as the edge weight; and when only basic road network data is available, the road distance between stations is used as the edge weight. By adaptively selecting the edge weight assignment method based on the data completeness, a scenic area public transport network map reflecting the relationships between stations can be effectively constructed under different data conditions, ensuring the applicability and robustness of the model under different data collection levels.

[0059] Furthermore, the scenic area's public transport network map and spatiotemporal input samples are input into a spatiotemporal graph neural network passenger flow prediction sub-model to obtain passenger flow prediction results within the future rolling window. This includes: for each node in the scenic area's public transport network, extracting passenger flow and operational status features of that node across multiple consecutive time slices, combining external influencing factors to form feature representations of each node under different time slices, and organizing them into a spatiotemporal input tensor in chronological order; inputting the spatiotemporal input tensor into the spatiotemporal graph neural network passenger flow prediction sub-model, using a graph convolutional network to extract spatial correlation features between stations, and using a gated recurrent unit to extract dynamic features of passenger flow changes over time, outputting passenger flow prediction results for multiple future time slices, thus obtaining passenger flow prediction results within the future rolling window. By utilizing graph convolutional networks and gated recurrent units to extract spatial correlation features between stations and dynamic temporal features of passenger flow respectively, and combining external influencing factors, the spatiotemporal evolution law of passenger flow in the scenic area's public transport network is effectively captured, thereby significantly improving the accuracy and reliability of passenger flow prediction within the future rolling window.

[0060] Furthermore, a departure matrix is ​​constructed based on passenger flow forecast results. This includes: constructing a departure matrix with bus routes as rows and discrete time slices within the future rolling window as columns. Each element in the departure matrix corresponds to the planned number of departures for a route within a future time slice, resulting in the departure matrix. A passenger flow state evolution process is established based on the passenger flow forecast results, ensuring that the future waiting status at the source station is jointly determined by the existing number of waiting passengers, the number of future arrivals, and the capacity provided by the route departures. Similarly, the future congestion status at the scenic area's passenger transport center is jointly determined by the input passenger flow from upstream transport routes, the output passenger flow from downstream connecting routes, and the current congestion status. By unifying upstream transport routes and downstream evacuation routes into the same matrix-based scheduling framework and establishing a quantitative evolutionary relationship between departure plans and passenger flow status, a closed-loop feedback loop is formed between the waiting demand at the source station and the congestion status at the scenic area's passenger transport center during the optimization process. This provides a quantifiable decision-making basis for subsequent collaborative solutions to achieve staggered arrival times for tourists.

[0061] Furthermore, the departure matrix is ​​input into the rolling time-domain departure matrix optimization sub-model for solution, yielding the optimal departure matrix for each route within the future rolling window at the current decision time. This includes: with the objectives of reducing passenger waiting congestion, alleviating congestion at the scenic area passenger transport center, controlling public transport operating costs, and smoothing the input flow to the scenic area passenger transport center, the passenger flow forecast results, passenger flow state evolution process, and actual operational constraints are jointly input into the rolling time-domain departure matrix optimization sub-model for solution, obtaining the optimal departure matrix for each route within the future rolling window at the current decision time. By unifying the passenger flow forecast results, passenger flow state evolution process, and actual operational constraints into a rolling time-domain solution framework with multiple optimization objectives such as waiting congestion, operating costs, and input flow fluctuations, coordinated optimization of upstream transportation and downstream evacuation capacity is achieved, thereby enabling proactive adjustment of the input rhythm of the scenic area passenger transport center while ensuring the actual feasibility of the scheduling plan.

[0062] Furthermore, the actual operational constraints include at least: the maximum number of departures allowed on a single route within a single time slice; the minimum departure interval between adjacent departures on the same route; the number of vehicles available for deployment within a time slice; the capacity constraint of the scenic area passenger transport center; and the vehicle turnaround constraint, which requires a turnaround time after a vehicle completes a departure task before it can be deployed for a new task. By introducing a multi-dimensional constraint system covering departure intensity, departure interval, vehicle resources, center capacity, and vehicle turnaround, the optimal departure matrix generated through optimization is ensured to be fully executable in actual operation, avoiding the problem of a disconnect between theoretical solutions and physical scheduling capabilities.

[0063] According to another embodiment of the present invention, a rolling time-domain off-peak scheduling method for scenic area public transportation includes the following:

[0064] Step S201: Obtain the operational data and external impact data of the scenic area's public transportation system prior to the current decision-making moment.

[0065] The acquired data primarily includes the number of tourists arriving, waiting, boarding and alighting, and stranded at each station over multiple consecutive time periods, as well as the number of vehicle arrivals, departures, and route operating times. It also incorporates external factors such as weather changes, holidays, number of reservations at the scenic area, high-speed rail arrival schedules, tour bus arrival schedules, and large-scale event arrangements to provide a unified description of the scenic area's public transportation needs. This data collection forms the foundation for raw data reflecting the changing trends in visitor flow and vehicle operation status within the scenic area. The collected data and the model input variables have the following correspondence:

[0066] (1)

[0067] in, To increase the number of arriving tourists at the station, For the number of people waiting for the train, To indicate the site In time slice The number of people who board and leave the station via the scenic area's public transportation system does not include those who leave the station on foot, by private car, taxi, ride-hailing service, or other non-scenic area public transportation methods. In the actual system, Passenger data can be obtained from bus card swipe records, electronic ticketing records, onboard passenger flow counting devices, platform video recognition devices, turnstile data, or boarding records in a rolling time-domain off-peak scheduling system for buses serving scenic areas. When multiple bus routes operate at the same stop... This represents the total number of passengers actually boarding and departing from all relevant lines at that station within that time frame. External characteristics comprised of weather, holidays, high-speed rail arrival schedules, and scenic spot reservations.

[0068] The number of arriving tourists corresponds to the number of newly arriving tourists at the station in formula (1); the number of waiting passengers and the number of stranded passengers at the station correspond to the number of waiting passengers or the stranded status in formula (1), where the variable at the tourist origin station represents the number of tourists waiting to go to the scenic area passenger transport center, and the variable at the scenic area passenger transport center represents the number of tourists waiting to be evacuated to various scenic areas; the number of passengers boarding is used to count the number of passengers leaving the station in formula (1), and the number of passengers alighting is used to describe the input passenger flow formed to the corresponding station after the vehicle arrives. The number of vehicle arrivals, the number of vehicle departures, and the route running time are used to determine the number of available vehicles, route capacity, travel time, vehicle turnaround time, and departure constraint parameters in the departure matrix optimization sub-model.

[0069] Step S202: Construct a public transportation network map of the scenic area based on the structure of the scenic area's bus routes, road connections, and the transfer relationships of tourists between different stops.

[0070] High-speed rail stations, bus stations, tourist distribution centers, scenic area passenger transport centers, and scenic area shuttle stations are abstracted as nodes in the graph. The connections between stations via bus routes, road accessibility, or stable passenger flow transfers are abstracted as edges in the graph. When determining edge weights, different edge weight construction methods can be selected based on the completeness of the actual data. If the scenic area bus system has stable historical alighting and arrival data or passenger flow transfer data between stations, historical passenger flow transfer intensity is prioritized as the edge weight. If reliable passenger flow transfer data is lacking, but vehicle operation records are available, the average travel time between stations is used as the edge weight. If only basic road network data is available, the road distance between stations is used as the edge weight. When using historical passenger flow transfer intensity as the edge weight, the weight is determined by the number of passengers traveling between stations within the statistical period. Transferred to the station via public transportation system The number of tourists accounts for the site The proportion of all bus-transferring tourists is used as the edge weight; when average travel time or road distance is used as the edge weight, the edge weight is negatively correlated with travel time or road distance. After assigning edge weights, the resulting weighted adjacency matrix and self-join matrix are merged and degree-normalized to form the normalized adjacency matrix used. Thus, a scenic area bus map structure that can be used by a spatiotemporal graph neural network is obtained.

[0071] Specifically, the scenic area's public transportation network is represented as... ,in For site collection, A collection of site connections; For tourists to gather at their source stations, As a node of the scenic area's passenger transport center, For the collection of scenic area stations, For the collection of bus routes, For a collection of vehicles. Discretize the operating time into segments of length . Time slices at each decision moment Above, for the future Optimize the rolling process using time slices, defining the time window set as follows: Introduce vehicle assignment variables: .in, Indicates vehicle In the future Each time slice is allocated to execute the circuit. The departure task, This indicates that it has not been assigned.

[0072] Let the input dimension be Then the past The input tensor consists of several time slices:

[0073] (2)

[0074] The short-term passenger flow prediction sub-model employs a spatiotemporal graph neural network structure combining a graph convolutional network and a gated recurrent unit (GRU) to output the passenger flow status within the future rolling window. For any time slice... It can be represented as:

[0075] (3)

[0076] (4)

[0077] (5)

[0078] (6)

[0079] in, Indicates the first The node features after the time slice undergoes the first layer of graph convolution; Indicates the first Node features after the time slice undergoes a second layer of graph convolution; This represents the temporal hidden representation obtained by GRU after performing temporal modeling on features from multiple consecutive time slices; Represents the future predicted by the model Passenger flow status at different times; This is the normalized adjacency matrix; The Xavier uniform initialization method is used; Zero initialization is used; The Xavier uniform initialization method is used; Zero initialization is used; and These are the hidden feature dimensions of the first and second layer graph convolutions, respectively; It is a vector of all 1s, used to extend the bias term to all nodes; The ReLU function is used; in a specific embodiment, the value is... , .

[0080] Step S203: Construct input samples for the spatiotemporal graph neural network based on the collected historical and real-time data.

[0081] For each node in the scenic area's public transportation network, passenger flow and operational status features across multiple consecutive time slices are extracted. External influencing factors such as weather, holidays, scenic area reservations, and high-speed rail arrival schedules are combined with these station features to form a feature representation of the node in different time slices. Subsequently, the node features from multiple consecutive time slices are organized into spatiotemporal input samples according to chronological order, and data preprocessing is performed before inputting them into the model. For continuous time series data such as visitor arrivals, waiting numbers, boarding and alighting numbers, station congestion, vehicle arrival times, vehicle departure times, and route operating times, if data from a single time slice is missing, it is imputed using valid values ​​from the previous time slice; if data from multiple consecutive time slices is missing, it is imputed using the historical average of the same date and time slice. For outliers that significantly deviate from the historical normal range, corrections are made using vehicle operation logs, ticketing records, or video recognition results; if corrections are not possible, historical quantile ranges are used for truncation. Continuous numerical features are standardized using the mean and standard deviation obtained from training sample statistics, and the same standardized parameters are used in the validation, testing, and actual prediction stages. For category features such as weather type, holiday type, and large-scale event type, one-hot encoding is used to convert them into numerical features that can be input into the model. Through the above processing, data from different time periods, different stations, and different units can be uniformly input into the spatiotemporal input tensor described in formula (2).

[0082] Specifically, during model training, real passenger flow data from historical samples are used as supervision labels. Let the training sample set be... The sample size is For the sample ,node , the future No. The first predicted time slice, the... Class predictor variable, model predicted value The actual observed value is Then the mean squared error loss function is used:

[0083] (7)

[0084] in, , , These represent the number of nodes, the number of time slices, and the number of types of predictive variables, respectively.

[0085] Step S204 involves inputting the scenic area's public transport network map and corresponding spatiotemporal input samples into a spatiotemporal graph neural network passenger flow prediction model. The model utilizes graph convolutional structures to extract spatial correlation features between stations and time-series modeling structures to extract dynamic features of passenger flow changes over time. External factors such as weather, holidays, high-speed rail arrival schedules, and the number of reservations at the scenic area are integrated into the prediction process to obtain passenger flow prediction results within the future rolling window. These prediction results reflect the new arrivals at each tourist origin station in future time slots, the waiting status at each station, and the congestion status at the scenic area's passenger transport center in future time slots. They also characterize the transfer demand from the scenic area's passenger transport center to various scenic areas, providing input for subsequent departure matrix optimization.

[0086] Step S205: After obtaining the passenger flow forecast results within the future rolling window, a departure matrix is ​​constructed to describe the planned departure times of different bus routes in each future time slot. This departure matrix uses routes as rows and discrete time slots within the future rolling window as columns. Each element in the matrix corresponds to the planned number of departures for a route in a given future time slot. This matrix representation allows for the unified consideration of upstream transport routes from tourist origin stations to the scenic area passenger transport center and downstream evacuation routes from the scenic area passenger transport center to various scenic area shuttle stations within the same scheduling framework.

[0087] Specifically, in the future The decision matrix is ​​composed of the number of departures on each route within a given time period:

[0088] (8)

[0089] in, Indicates the line In the future The planned number of departures within a time slot meets the requirements. .

[0090] Step S206: After constructing the departure matrix, a passenger flow state evolution process is established based on future passenger flow forecasts, creating a clear correspondence between the departure plan and passenger flow changes. For tourist origin stations, their future waiting status is jointly determined by the existing number of waiting passengers, the number of new arrivals, and the capacity provided by the route departures. For scenic area passenger transport centers, their future congestion status is jointly determined by the input passenger flow from upstream transport routes, the output passenger flow from downstream connecting routes, and the current congestion status. Through this state evolution process, the impact of upstream route departure plans on the input flow of scenic area passenger transport centers and the impact of downstream route departure plans on the evacuation capacity of passenger transport centers can be uniformly described, enabling the optimized departure matrix results to directly affect congestion control at scenic area passenger transport centers.

[0091] Specifically, source sites The evolution equation for the number of people waiting for the train is shown below:

[0092] (9)

[0093] in, For the line Single-shift capacity, To serve the site The passenger transport center for entering the scenic area is a collection point for routes.

[0094] The scenic area's passenger transport center will be available in the future. The input stream, output stream, and number of people remaining satisfy the following:

[0095] (10)

[0096] (11)

[0097] (12)

[0098] in, For the line The number of time segments corresponding to the travel time from the source station to the scenic area's passenger transport center. For the set of input lines, This is a collection of routes that depart from the scenic area's passenger transport center to various scenic areas.

[0099] Step S207: After establishing the passenger flow state evolution relationship, the future passenger flow forecast results, state evolution relationship, and actual operational constraints are jointly input into the rolling time-domain departure matrix optimization model for solution. This optimization process aims to reduce tourist waiting congestion, alleviate congestion at the scenic area passenger transport center, control public transport operating costs, and smooth the input flow to the scenic area passenger transport center. It comprehensively considers actual constraints such as the number of available vehicles, the maximum departure limit for a single route, departure interval requirements, the capacity limit of the scenic area passenger transport center, and the minimum evacuation demand in the scenic area direction, thereby obtaining the optimal departure matrix for each route within the future rolling window at the current decision time. Because this optimization result simultaneously considers the coupling relationship between upstream tourist input and downstream scenic area evacuation, it can make the tourist arrival process at the scenic area passenger transport center more stable in the future time period.

[0100] Specifically, the objective function of the departure matrix optimization sub-model is defined as:

[0101] (13)

[0102] in, The capacity threshold for the scenic area's passenger transport center is determined based on its physical space capacity, safety management requirements, and service level requirements. In practice, the physical capacity can be determined first by considering the effective area of ​​the waiting area, transfer area, ticket check area, and other areas where passengers can stay, combined with the allowed number of passengers per unit area. Then, the safe capacity can be determined based on fire evacuation, safety control, or regulations of the scenic area management department. Finally, the service capacity can be determined based on passenger comfort, queue length, transfer efficiency, or service level evaluation requirements. For the line The unit dispatch cost, , , , These are the weighting coefficients. The objective function corresponds to the waiting cost at the source station, the congestion penalty at the scenic area passenger transport center, the operating cost, and the input flow fluctuation penalty, thereby achieving proactive control over off-peak arrivals at the passenger transport center.

[0103] The model further satisfies the following constraints:

[0104] (14)

[0105] (15)

[0106] (16)

[0107] (17)

[0108] in, For the line The maximum number of departures allowed within a single time slice is used to limit the departure intensity within a unit of time slice; For time slices The number of vehicles that can be called up internally. This is a capacity control threshold for scenic area passenger transport centers, used to limit or penalize situations where the number of people staying at the passenger transport center exceeds safety or service level requirements; , For the line Minimum departure interval.

[0109] The constraints on the maximum number of departures in a single time slice and the minimum departure interval apply to different levels. The former limits the maximum number of departures that can be scheduled on the same route within a discrete time slice, and is mainly determined by platform berths, road capacity, vehicle organization capacity, and dispatching management capacity; the latter limits the shortest time interval between adjacent departures on the same route, and is mainly used to avoid excessively dense departures.

[0110] When setting specific parameters, if the time slice length is long, theoretically, a single time slice may accommodate multiple departures that meet the minimum interval requirement. In this case, it is still necessary to... Limit the maximum number of departures within the time slice; if the time slice length is short, the minimum departure interval constraint mainly limits the departure density across multiple consecutive time slices, while It is still used to constrain the maximum number of departures within a single time slice. Both types of constraints must be satisfied simultaneously, and the parameter settings should ensure that there are no factors affecting departures. Unfeasible departure plans due to excessively large or excessively short minimum departure intervals.

[0111] To ensure the model output is executable, the original departure matrix will be used. Project to feasible region The corrected departure matrix is ​​obtained:

[0112] (18)

[0113] Among them, feasible region It should include at least vehicle availability constraints, vehicle turnover constraints, and departure point matching constraints.

[0114] Vehicle availability constraints are used to ensure that the number of vehicles scheduled to perform departure tasks within any given time slice does not exceed the number of vehicles available for use in the system during that time slice. Let... Indicates the future number The number of vehicles that can be called up within a time slice is:

[0115] (19)

[0116] Vehicle turnaround constraints are used to ensure that after a vehicle completes a departure task on a certain route, it must undergo the required route travel time, passenger pick-up and drop-off time, and necessary preparation time before it can be put into a new departure task. Let the route... The number of time slices corresponding to the vehicle turnaround time is:

[0117] (20)

[0118] in, For the line runtime For passenger pick-up and drop-off times, For scheduling buffer time, This is the time slice length. If the vehicle... In time slice The internal route was assigned for execution. Then the vehicle's next... No other line tasks may be scheduled during this time slice. This constraint can be expressed as:

[0119] (twenty one)

[0120] The departure origin matching constraint is used to ensure that when a vehicle is assigned to execute a certain route, its location should be consistent with the departure origin of that route. Let... Indicates the line The starting point of the train, Indicates vehicle In the future The station at the start of each time slice, defined parameters:

[0121] (twenty two)

[0122] The departure point matching constraint can then be expressed as:

[0123] (twenty three)

[0124] Step S208: Under the rolling time-domain control method, the entire departure plan within the rolling window is not executed all at once. Instead, only the departure decision corresponding to the current time slot in the optimal departure matrix is ​​extracted as the actual execution plan. The departure arrangements for other future time slots are only retained as reference plans generated under the current prediction conditions. When entering the next time slot, the latest passenger flow data, vehicle operating status, and external influence information are collected again. New input samples are constructed again, and the passenger flow prediction and departure matrix optimization process is repeated, thereby realizing the continuous rolling update of the rolling time-domain off-peak scheduling plan for scenic area public transportation.

[0125] Step S209: After the above-described rolling prediction and optimization process, the actual departure times of each bus route within the current time slice are output. Simultaneously, the reference departure matrix for the future rolling window, the predicted waiting status at each station, the predicted congestion status at the scenic area's passenger transport center, and the predicted transfer demand for each scenic area direction are also output. These output results can be directly used to guide the allocation of transport capacity among different routes and at different times within the scenic area, thereby enabling tourists to arrive at the scenic area's passenger transport center during off-peak hours and improving the overall scheduling efficiency and service stability during the scenic area's bus connection process.

[0126] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0127] This invention also provides a rolling time-domain off-peak scheduling device for scenic area buses. This device is used to implement the above embodiments and preferred embodiments, and will not be repeated for details already described. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0128] Figure 2 According to one embodiment of the present invention, a rolling time-domain off-peak scheduling device for scenic area buses includes:

[0129] The acquisition module 301 is used to acquire the operating data and external impact data of the scenic area bus system before the current decision time. Based on the scenic area bus route structure, road connection relationship and the transfer relationship of tourists between different stations, a scenic area bus network map is constructed, and a spatiotemporal input sample is constructed based on the operating data and external impact data.

[0130] Prediction module 302 is used to input the scenic area bus network map and spatiotemporal input samples into the spatiotemporal graph neural network passenger flow prediction sub-model to obtain the passenger flow prediction results within the future rolling window;

[0131] The optimization module 303 is used to construct a departure matrix based on the passenger flow forecast results, input the departure matrix into the rolling time domain departure matrix optimization sub-model for solution, obtain the optimal departure matrix of each route in the future rolling window at the current decision time, and extract the departure decision corresponding to the current time slice in the optimal departure matrix as the actual execution plan.

[0132] It should be noted that the above modules can be implemented by software or hardware. For the latter, they can be implemented in the following ways, but are not limited to: all the above modules are located in the same processor; or, the above modules are located in different processors in any combination.

[0133] According to one embodiment of the present invention, an electronic device is also provided, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the above-described rolling time-domain off-peak scheduling method for scenic area buses.

[0134] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:

[0135] Step S1: Obtain the operational data and external impact data of the scenic area's public transportation system before the current decision-making time. Based on the scenic area's public transportation route structure, road connection relationships, and the transfer relationships of tourists between different stations, construct a scenic area public transportation network map. Construct spatiotemporal input samples based on the operational data and external impact data.

[0136] Step S2: Input the scenic area bus network map and spatiotemporal input samples into the spatiotemporal graph neural network passenger flow prediction sub-model to obtain the passenger flow prediction results within the future rolling window;

[0137] Step S3: Construct a departure matrix based on passenger flow forecast results, input the departure matrix into the rolling time domain departure matrix optimization sub-model for solution, obtain the optimal departure matrix for each route in the future rolling window at the current decision time, and extract the departure decision corresponding to the current time slice from the optimal departure matrix as the actual execution plan.

[0138] According to one embodiment of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the storage medium is located to execute the above-described rolling time-domain off-peak scheduling method for scenic area buses.

[0139] Optionally, in this embodiment, the storage medium may be configured to store a computer program for performing the following steps:

[0140] Step S1: Obtain the operational data and external impact data of the scenic area's public transportation system before the current decision-making time. Based on the scenic area's public transportation route structure, road connection relationships, and the transfer relationships of tourists between different stations, construct a scenic area public transportation network map. Construct spatiotemporal input samples based on the operational data and external impact data.

[0141] Step S2: Input the scenic area bus network map and spatiotemporal input samples into the spatiotemporal graph neural network passenger flow prediction sub-model to obtain the passenger flow prediction results within the future rolling window;

[0142] Step S3: Construct a departure matrix based on passenger flow forecast results, input the departure matrix into the rolling time domain departure matrix optimization sub-model for solution, obtain the optimal departure matrix for each route in the future rolling window at the current decision time, and extract the departure decision corresponding to the current time slice from the optimal departure matrix as the actual execution plan.

[0143] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0144] According to one embodiment of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the above-described rolling time-domain off-peak scheduling method for scenic area buses.

[0145] Optionally, in this embodiment, the above-mentioned computer program product can be configured as a computer program that performs the following steps:

[0146] Step S1: Obtain the operational data and external impact data of the scenic area's public transportation system before the current decision-making time. Based on the scenic area's public transportation route structure, road connection relationships, and the transfer relationships of tourists between different stations, construct a scenic area public transportation network map. Construct spatiotemporal input samples based on the operational data and external impact data.

[0147] Step S2: Input the scenic area bus network map and spatiotemporal input samples into the spatiotemporal graph neural network passenger flow prediction sub-model to obtain the passenger flow prediction results within the future rolling window;

[0148] Step S3: Construct a departure matrix based on passenger flow forecast results, input the departure matrix into the rolling time domain departure matrix optimization sub-model for solution, obtain the optimal departure matrix for each route in the future rolling window at the current decision time, and extract the departure decision corresponding to the current time slice from the optimal departure matrix as the actual execution plan.

[0149] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.

[0150] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0151] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between units or modules may be electrical or other forms.

[0152] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0153] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0154] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0155] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A rolling time-domain off-peak scheduling method for public transportation in scenic areas, characterized in that, include: Obtain operational data and external impact data of the scenic area's public transportation system prior to the current decision-making time. Construct a scenic area public transportation network map based on the scenic area's public transportation route structure, road connection relationships, and tourist transfer relationships between different stops. Construct spatiotemporal input samples based on the operational data and external impact data. The scenic area bus network map and the spatiotemporal input samples are input into the spatiotemporal graph neural network passenger flow prediction sub-model to obtain the passenger flow prediction results within the future rolling window. Based on the passenger flow prediction results, a departure matrix is ​​constructed. The departure matrix is ​​then input into the rolling time-domain departure matrix optimization sub-model for solution. The optimal departure matrix for each route within the future rolling window at the current decision time is obtained. The departure decision corresponding to the current time slice in the optimal departure matrix is ​​then extracted as the actual execution plan.

2. The rolling time-domain off-peak scheduling method for scenic area buses according to claim 1, characterized in that, Obtain operational data and external impact data of the scenic area's public transportation system prior to the current decision-making moment, including: The system acquires the number of tourists arriving, waiting, boarding and alighting, stranded at each station, vehicle arrivals, vehicle departures, and route operating times over multiple consecutive time segments. It also acquires weather changes, holiday information, number of reservations at scenic spots, high-speed rail arrival plans, tour bus arrival plans, and large-scale event arrangements as external impact data to obtain the operational data and the external impact data.

3. The rolling time-domain off-peak scheduling method for scenic area buses according to claim 1, characterized in that, Based on the scenic area's bus route structure, road connections, and tourist transfer patterns between different stops, a scenic area bus network map is constructed, including: The high-speed rail station, bus station, tourist distribution center, scenic area passenger transport center, and scenic area shuttle station are abstracted as nodes in the graph. The bus route connections, road accessibility relationships, or stable passenger flow transfer relationships between the stations are abstracted as edges in the graph. The edge weight construction method is selected according to the data completeness to assign values ​​to the edges in the graph. The weighted adjacency matrix after assignment is normalized to obtain the scenic area bus network graph.

4. The rolling time-domain off-peak scheduling method for scenic area buses according to claim 3, characterized in that, The edge weights in the graph are assigned values ​​based on the completeness of the data, using the appropriate edge weight construction method. When stable historical passenger transfer data or passenger flow transfer data between stations can be obtained, the historical passenger flow transfer intensity is used as the edge weight; when reliable passenger flow transfer data is lacking but vehicle operation records can be obtained, the average running time between stations is used as the edge weight; when only basic road network data is available, the road distance between stations is used as the edge weight.

5. The rolling time-domain off-peak scheduling method for scenic area buses according to claim 1, characterized in that, The scenic area bus network map and the spatiotemporal input samples are input into the spatiotemporal graph neural network passenger flow prediction sub-model to obtain the passenger flow prediction results within the future rolling window, including: For each node in the scenic area's public transportation network, the passenger flow and operational status features of that node in multiple consecutive time slices are extracted. Combined with external influencing factors, feature representations of each node in different time slices are formed and organized into a spatiotemporal input tensor in chronological order. The spatiotemporal input tensor is input into the spatiotemporal graph neural network passenger flow prediction sub-model. The spatial correlation features between stations are extracted using a graph convolutional network, and the dynamic features of passenger flow changes over time are extracted using a gated recurrent unit. The passenger flow prediction results for multiple future time slices are output to obtain the passenger flow prediction results within the future rolling window.

6. The rolling time-domain off-peak scheduling method for scenic area buses according to claim 1, characterized in that, Based on the passenger flow forecast results, a departure matrix is ​​constructed, including: The departure matrix is ​​constructed by using bus routes as rows and discrete time slices within a future rolling window as columns. Each element in the departure matrix corresponds to the planned number of departures for a route within a future time slice, thus obtaining the departure matrix. Based on the passenger flow forecast results, a passenger flow state evolution process is established, so that the future waiting state of the source station is jointly determined by the number of existing waiting passengers, the number of new arrivals in the future, and the transportation capacity provided by the route departures, and the future stagnation state of the scenic area passenger transport center is jointly determined by the input passenger flow formed by the upstream transport lines, the output passenger flow formed by the downstream connecting lines, and the stagnation state at the current moment.

7. The rolling time-domain off-peak scheduling method for scenic area buses according to claim 6, characterized in that, The departure matrix is ​​input into the rolling time-domain departure matrix optimization sub-model for solution, to obtain the optimal departure matrix for each route within the future rolling window at the current decision time, including: With the goals of reducing tourist waiting congestion, alleviating congestion at scenic area passenger transport centers, controlling public transport operating costs, and smoothing the input flow to scenic area passenger transport centers, the passenger flow prediction results, the passenger flow state evolution process, and actual operational constraints are jointly input into the rolling time-domain departure matrix optimization sub-model for solution, so as to obtain the optimal departure matrix of each route within the future rolling window at the current decision time.

8. The rolling time-domain off-peak scheduling method for scenic area buses according to claim 7, characterized in that, The actual operational constraints include at least the following: the maximum number of departures allowed on a single route within a single time slice; the minimum departure interval between adjacent departures on the same route; the number of vehicles that can be called up within a time slice; the capacity constraints of the scenic area passenger transport center; and the vehicle turnaround constraint, which requires a turnaround time after a vehicle has completed a departure task before it can be put into a new task.

9. A rolling time-domain off-peak scheduling device for scenic area buses, characterized in that, include: The acquisition module is used to acquire the operational data and external impact data of the scenic area's public transportation system before the current decision-making time. Based on the scenic area's public transportation route structure, road connection relationships, and the transfer relationships of tourists between different stations, it constructs a scenic area public transportation network map and constructs spatiotemporal input samples based on the operational data and the external impact data. The prediction module is used to input the scenic area bus network map and the spatiotemporal input samples into the spatiotemporal graph neural network passenger flow prediction sub-model to obtain the passenger flow prediction results within the future rolling window; The optimization module is used to construct a departure matrix based on the passenger flow prediction results, input the departure matrix into the rolling time domain departure matrix optimization sub-model for solution, obtain the optimal departure matrix of each route in the future rolling window at the current decision time, and extract the departure decision corresponding to the current time slice in the optimal departure matrix as the actual execution plan.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, the device on which the storage medium is located executes the rolling time-domain off-peak scheduling method for scenic area buses according to any one of claims 1 to 8.

Images