An integrated geographic information automatic driving taxi operation scheduling method

Abstract

The application provides an integrated geographic information automatic driving taxi operation scheduling method, which relates to the technical field of automatic driving taxi scheduling, and through real-time fusion of multi-level geographic information such as traffic situation, road topology and POI demand hot area, a dynamic weight matrix is generated by adopting a space-time convolutional neural network to construct a space-time coupled traffic state feature map; when responding to a passenger request, candidate vehicles are screened based on a four-dimensional space-time reachable domain, and an optimal path is solved by using a multi-objective optimization model to cooperatively optimize passenger waiting time, vehicle energy consumption and transport capacity balance; an elastic anchor point mechanism is introduced, replaceable nodes are pre-embedded in the path, and local re-planning is triggered in combination with an insertion cost model to dynamically update passenger ETA. The application effectively solves the problems of low multi-source data fusion efficiency, insufficient dynamic response and multi-passenger path conflict, and improves the scheduling robustness in a complex urban traffic scene.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation technology and relates to a multi-objective operation scheduling method for autonomous taxis that integrates geographic information data, which is applicable to the efficient operation of city-level autonomous taxi networks. Background Technology

[0002] With the rapid development of autonomous driving technology, the commercial operation of autonomous taxis (AT) has become an important direction in the field of intelligent transportation. Especially in densely populated cities, multi-passenger carpooling services are of significant value in reducing operating costs and alleviating traffic congestion. However, existing technological solutions face multiple technical bottlenecks in complex urban environments, specifically in the following aspects:

[0003] Dispatch systems based on static road networks typically use fixed topology maps for route planning. While this can achieve basic carpooling matching, it doesn't consider the impact of real-time traffic flow changes on route feasibility. For example, sudden congestion during peak hours can lead to excessively high estimated arrival times, causing continuous delays to subsequent orders.

[0004] While the greedy strategy of quickly matching the nearest vehicle with passengers reduces computational complexity, its discrete geographic information processing makes it impossible to guarantee global capacity balance. When demand surges in a business district during the morning rush hour, vehicles in adjacent residential areas cannot respond in time, leading to a widening capacity gap.

[0005] When dealing with multi-passenger route planning, a fixed priority ranking method is often used to process carpooling requests, without quantifying the correlation between route overlap and passenger experience.

[0006] Therefore, how to provide an autonomous taxi operation scheduling method that integrates dynamic geographic information to achieve optimal global resource allocation is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] In view of this, the present invention proposes an autonomous driving taxi operation scheduling method integrating geographic information. By deeply integrating dynamic geographic information, a spatiotemporally coupled decision model is constructed, which achieves optimal global resource allocation while ensuring real-time computation.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] This invention discloses a method for scheduling autonomous taxi operations that integrates geographic information, comprising the following steps:

[0010] S1: Real-time acquisition of multi-source geographic information data, including traffic situation layer data, road topology layer data and POI demand hot zone layer data, to construct a multi-level real-time geographic information data network;

[0011] S2: A dynamic weight matrix is ​​generated through a spatiotemporal convolutional neural network to adaptively fuse the data between the layers of the multi-level geographic information real-time data network and output a spatiotemporally coupled traffic state feature map.

[0012] S3: Respond to passenger carpooling requests, extract the POI attributes and demand priority associated with the carpooling request, generate a spatiotemporal reachability domain based on the traffic state feature map, and filter a set of candidate vehicles that meet the preset response time threshold.

[0013] S4: Construct a set of decision variables including vehicle-passenger matching relationships, path sequences, and ETA, and solve for the optimal scheduling path based on a multi-objective optimization model;

[0014] S5: Pre-embed replaceable anchor points in the optimal scheduling path. When a new carpooling request arrives, calculate the path insertion cost, trigger the replanning of the optimal scheduling path based on the path insertion cost, and update the ETA of the carpooling passengers.

[0015] Preferably, the multi-level geographic information real-time data network structure includes:

[0016] The traffic situation layer acquires the floating satellite positioning trajectory of autonomous taxis and microwave radar data uploaded in real time by roadside RSUs, and connects to the urban traffic network to generate the spatiotemporal tensor of autonomous taxis.

[0017] The road topology layer acquires the urban road network and marks the coverage radius of charging stations and the geofence restrictions of traffic control areas;

[0018] The POI demand hot zone layer calculates the passenger transfer probability between POIs based on the spatiotemporal matrix of historical orders of autonomous taxis, and calculates the temporal correlation of demand fluctuations in conjunction with the urban POI functional classification map. Based on the temporal correlation of demand fluctuations, the POI attributes of the demand hot zone are marked on the urban POI functional classification map.

[0019] Preferably, step S1 further includes: predicting the distribution of future demand hotspots to drive idle autonomous taxis to move in the direction of the demand gradient.

[0020] Obtain the spatiotemporal tensor T of the autonomous taxi and its own state V, and map the vehicle state to a spatiotemporal grid to obtain spatiotemporal fusion features. :

[0021] ;

[0022] In the formula, is the number of spatiotemporal grids, ⊕ represents feature concatenation, and Pool represents the max pooling operation;

[0023] For spatiotemporal fusion features Each spatiotemporal node n in the model captures causal relationships through cross-modal attention weights;

[0024] A hierarchical masking mechanism is constructed, and cross-modal attention weights are corrected using temporal causal masks and spatial reachability masks to obtain corrected causal relationships. The reachable spatiotemporal nodes with causal relationships are dynamically connected to obtain a causal reachability graph.

[0025] Based on the corrected cross-modal attention weights and causal reachability graphs, the activation probability of reachable spatiotemporal nodes is calculated, and dispatch instructions for autonomous taxis to predicted demand hotspots are generated.

[0026] Preferably, step S2 includes the following steps:

[0027] We employ 3D dilated convolution to capture multi-scale spatial relationships between data layers and process temporal dependencies based on axial Transformer.

[0028] The average traffic situation, POI feature entropy, and topological complexity are calculated based on the multi-level geographic information real-time data network and input into the meta-learning weight generator to obtain the dynamic range of the weights.

[0029] Based on the dynamic range of the weights, the traffic situation layer data, road topology layer data, and POI demand hotspot layer data are adaptively fused to output a spatiotemporally coupled traffic state feature map.

[0030] Preferably, step S3 includes the following steps:

[0031] In response to a passenger's carpooling request, the associated POI attributes, including POI category and service area, are extracted from the traffic state feature map based on the origin and destination locations.

[0032] Based on the POI category and service radius, the historical demand average of the corresponding POI category and service range is extracted from the traffic state feature map according to the historical orders of autonomous taxis.

[0033] Preferably, step S3 includes the following steps:

[0034] Based on the preset response time threshold and traffic state feature map in the passenger carpooling request, the spatial reachability range is combined with the time window to construct a four-dimensional dynamic reachability domain {spatial latitude and longitude, time window, driving speed, vehicle driving constraints}, and the reachability domain boundary is adjusted according to the demand priority. Based on the four-dimensional dynamic reachability domain, all possible locations that autonomous taxis can reach within a given time are obtained.

[0035] The system acquires the real-time status of autonomous taxis, maps the real-time status parameters to a four-dimensional dynamic reachability domain, and uses a hybrid index structure to filter a set of candidate vehicles that meet a preset response time threshold.

[0036] Preferably, step S4 includes the step of generating a path sequence for the candidate vehicle set:

[0037] For each candidate vehicle, based on the starting and ending points in the passenger carpooling request, a hierarchical path planning algorithm is used to generate multiple feasible paths to form a candidate path set, including: direct paths, detour paths, and combined paths;

[0038] The ETA probability distribution and confidence interval for each path are generated through Monte Carlo sampling;

[0039] A spatiotemporal conflict matrix is ​​constructed based on the candidate path set and ETA probability distribution, and the time overlap risk coefficient of different vehicle paths at preset key nodes is used as a decision variable.

[0040] Preferably, the step of solving for the optimal scheduling path based on the multi-objective optimization model in S4 includes:

[0041] Structured matching relation variables, path sequence variables, and ETA variables;

[0042] Construct a multi-objective function that considers the resolution of risks and conflicts arising from time overlap;

[0043] The optimal scheduling path after conflict resolution is generated using the NSGA-II multi-objective optimization algorithm.

[0044] Preferably, step S5 includes the step of determining replaceable anchor points:

[0045] The optimal scheduling path is divided into discrete segments by fixed distance or time window, and the historical time overlap risk conflict probability of the discrete segments is obtained.

[0046] Search for and mark the preset key nodes within the radiation range of the discrete segment in the traffic state feature map, and obtain the congestion index of the marked nodes in the traffic state feature map.

[0047] Based on the probability of risk conflict due to historical time overlap and the congestion index, a dynamic feasibility score is assigned to each node;

[0048] Nodes with dynamic feasibility scores higher than a preset threshold are retained as replaceable anchor points.

[0049] Preferably, S5 includes:

[0050] For each replaceable anchor point, calculate the overall cost after insertion:

[0051] ;

[0052] In the formula, This is the increment of the total travel time. This is the increment of the detour distance. This represents the variance of passenger waiting time.

[0053] Replaceable anchors are sorted in ascending order of overall cost, with priority given to the best option. The smallest number of replaceable anchor points are used to replan the optimal scheduling path and update the ETA of the shared passengers.

[0054] As can be seen from the above technical solution, compared with the prior art, the beneficial effects of the present invention include:

[0055] By fusing multi-source geographic information (traffic status, road topology, POI hotspots) in real time, the system can dynamically perceive changes in urban traffic conditions. The adaptive weight matrix generated by the spatiotemporal convolutional neural network (ST-CNN) significantly improves the accuracy of data fusion.

[0056] Based on a multi-objective optimization model, and under the conditions of resolving time overlap risk conflicts, a coordinated optimization of passenger experience, energy consumption economy, and capacity balance is achieved.

[0057] By pre-embedding replaceable anchor points that meet the dynamic feasibility scoring requirements in the optimal path, local replanning can be quickly triggered when a new order is added.

[0058] In summary, this invention significantly improves the operational efficiency of autonomous taxis through multimodal data fusion, causal reasoning decision-making, and flexible resource scheduling, providing core technical support for city-level intelligent transportation systems. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0060] Figure 1 A flowchart illustrating an autonomous taxi operation scheduling method integrating geographic information, provided as an embodiment of the present invention;

[0061] Figure 2 This is a schematic diagram illustrating the generation principle of the candidate vehicle set provided in an embodiment of the present invention.

[0062] Figure 3 This is a schematic diagram illustrating the principle of determining replaceable anchor points according to an embodiment of the present invention. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.

[0064] This invention provides a method for scheduling autonomous taxi operations that integrates geographic information, such as... Figure 1 As shown, it includes the following steps:

[0065] S1: Real-time acquisition of multi-source geographic information data, including traffic situation layer data, road topology layer data and POI demand hot zone layer data, to construct a multi-level real-time geographic information data network;

[0066] S2: A dynamic weight matrix is ​​generated through a spatiotemporal convolutional neural network to adaptively fuse data between layers of a multi-level geographic information real-time data network and output a spatiotemporally coupled traffic state feature map.

[0067] S3: Respond to passenger carpooling requests, extract the POI attributes and demand priority associated with the carpooling request, generate a spatiotemporal reachability domain based on the traffic state feature map, and filter a set of candidate vehicles that meet the preset response time threshold; where, the POI attribute (Point of Interest) represents a discrete location point on the map with a specific function, representing a specific entity location;

[0068] S4: Construct a set of decision variables including vehicle-passenger matching relationships, route sequences, and ETA, and solve for the optimal scheduling route based on a multi-objective optimization model; where ETA (Estimated Time of Arrival) represents the estimated arrival time.

[0069] S5: Pre-embed replaceable anchor points in the optimal scheduling path. When a new carpooling request arrives, calculate the path insertion cost, trigger the replanning of the optimal scheduling path based on the path insertion cost, and update the ETA of the carpooling passengers.

[0070] In one embodiment, a multi-level geographic information real-time data network organization includes:

[0071] The traffic situation layer acquires the floating satellite positioning trajectory of autonomous taxis and microwave radar data uploaded in real time by roadside RSUs, and connects to the urban traffic network to generate the spatiotemporal tensor of autonomous taxis.

[0072] The road topology layer acquires the urban road network and marks the coverage radius of charging stations and the geofence restrictions of traffic control areas;

[0073] The POI demand hot zone layer calculates the passenger transfer probability between POIs based on the spatiotemporal matrix of historical orders of autonomous taxis, and calculates the temporal correlation of demand fluctuations in conjunction with the urban POI functional classification map. Based on the temporal correlation of demand fluctuations, the POI attributes of demand hot zones are marked on the urban POI functional classification map.

[0074] It should be noted that a spatiotemporal tensor is a structured multidimensional data vector that stores dynamic geographic information. The spatiotemporal tensor T∈R H×W×C×T Where H×W is the spatial grid, such as dividing a city into 100m×100m; C is the number of feature channels, such as vehicle speed, traffic flow, accident markers, etc.; T is the time slice, such as a slice every 5 minutes, storing 24 hours of data.

[0075] In this embodiment, S1 further includes the step of: predicting the distribution of future demand hot zones to drive idle autonomous taxis to move according to the demand gradient direction.

[0076] Obtain the spatiotemporal tensor T of the autonomous taxi and its own state V, and map the vehicle state to a spatiotemporal grid to obtain spatiotemporal fusion features. :

[0077] ;

[0078] In the formula, is the number of spatiotemporal grids, ⊕ represents feature concatenation, and Pool represents the max pooling operation;

[0079] For spatiotemporal fusion features Each spatiotemporal node n in the model captures causal relationships through cross-modal attention weights. :

[0080] ;

[0081] In the formula, W Q W K d represents the learnable query and key transformation matrix; d represents the dimension of the query and key transformation matrix.

[0082] The constraint condition is that Reachable(n) is only valid if i →n j )=1 to calculate attention weights.

[0083] A hierarchical masking mechanism is constructed, utilizing temporal causality masks and spatial reachability masks to correct cross-modal attention weights, thereby obtaining corrected causal relationships. Dynamically connecting reachable spatiotemporal nodes with causal relationships yields a causal reachability graph. :

[0084] ;

[0085] In the formula, The corrected cross-modal attention weights;

[0086] The activation probability of reachable spatiotemporal nodes is calculated based on the modified cross-modal attention weights and causal reachability graphs. And generate dispatch instructions for autonomous taxis to predicted demand hotspots:

[0087] ;

[0088] In the formula, For the Sigmoid function, The transformation matrix is ​​a value transformation matrix.

[0089] In one embodiment, S2 includes the following steps:

[0090] We employ 3D dilated convolution to capture multi-scale spatial relationships between data layers and process temporal dependencies based on axial Transformer.

[0091] The average traffic situation, POI feature entropy, and topological complexity are calculated based on the multi-level geographic information real-time data network and input into the meta-learning weight generator to obtain the dynamic range of the weights.

[0092] Based on the dynamic range of weights, the traffic situation layer data, road topology layer data, and POI demand hot zone layer data are adaptively fused to output a spatiotemporally coupled traffic state feature map.

[0093] In one embodiment, S3 includes the following steps:

[0094] In response to passenger carpooling requests, the associated POI attributes are extracted from the traffic state feature map based on the origin and destination locations, including POI category and service area.

[0095] Based on POI category and service radius, the historical demand average of the corresponding POI category and service range is extracted from the traffic state feature map according to the historical orders of autonomous taxis.

[0096] In one embodiment, S3 includes the following steps:

[0097] Based on the preset response time threshold and traffic state feature map in the passenger carpooling request, the spatial reachability is combined with the time window to construct a four-dimensional dynamic reachability domain. The reachability domain boundary is adjusted according to the demand priority. Based on the four-dimensional dynamic reachability domain, all possible locations that autonomous taxis can reach within a given time are obtained.

[0098] The system acquires the real-time status of autonomous taxis, maps the real-time status parameters to a four-dimensional dynamic reachability domain, and uses a hybrid index structure to filter a set of candidate vehicles that meet a preset response time threshold.

[0099] In this implementation, the four-dimensional dynamic reachable domain is constructed as follows: (spatial latitude and longitude, time window, driving speed, vehicle driving constraints), where vehicle driving constraints may include: speed constraints, remaining battery power constraints, etc.

[0100] In one embodiment, S4 includes the step of generating a path sequence for the candidate vehicle set:

[0101] For each candidate vehicle, based on the starting and ending points in the passenger carpooling request, a hierarchical path planning algorithm is used to generate multiple feasible paths to form a candidate path set, including: direct paths, detour paths, and combined paths;

[0102] The ETA probability distribution and confidence interval for each path are generated through Monte Carlo sampling;

[0103] A spatiotemporal conflict matrix is ​​constructed based on the candidate path set and ETA probability distribution, and the time overlap risk coefficient of different vehicle paths at preset key nodes is used as a decision variable.

[0104] In this embodiment, the step of solving the optimal scheduling path based on the multi-objective optimization model in S4 includes:

[0105] Structured matching relation variable x v,p ∈{0,1} indicates whether vehicle v is carrying passenger p;

[0106] Structured path sequence variable S v ={s1,s2,...,s C ,d i,j} represents the sequence of path nodes s and the sequence of time overlap risk coefficients d for vehicle v. i,j C represents the total number of vehicles;

[0107] Structured ETA variables t p , representing the estimated arrival time of passenger p;

[0108] Constructing a multi-objective function:

[0109] ;

[0110] ;

[0111] ;

[0112] ;

[0113] ;

[0114] In the formula, K represents the preset response time threshold for passenger carpooling requests; K is the number of time window segments for route planning. Let i be the path state of vehicle i in the k-th time window; This is an indicator function, which is 1 if vehicles i and j overlap in time and space within the k-th window, and 0 otherwise; For time overlap risk coefficient;

[0115] The optimal scheduling path after conflict resolution is generated using the NSGA-II multi-objective optimization algorithm.

[0116] In one embodiment, S5 includes the step of determining alternative anchor points:

[0117] The optimal scheduling path is divided into discrete segments by fixed distance or time window, and the historical time overlap risk conflict probability of the discrete segments is obtained.

[0118] Search for and mark the preset key nodes within the radiation range of the discrete segment in the traffic state feature map, and obtain the congestion index of the marked nodes in the traffic state feature map.

[0119] Based on the probability of risk conflict due to historical time overlap and the congestion index, a dynamic feasibility score is assigned to each node;

[0120] Nodes with dynamic feasibility scores higher than a preset threshold are retained as replaceable anchor points.

[0121] In one embodiment, S5 includes:

[0122] For each replaceable anchor point, calculate the overall cost after insertion:

[0123] ;

[0124] In the formula, This is the increment of the total travel time. This is the increment of the detour distance. This represents the variance of passenger waiting time.

[0125] Replaceable anchors are sorted in ascending order of overall cost, with priority given to the best option. The smallest number of replaceable anchor points are used to replan the optimal scheduling path and update the ETA of shared passengers.

[0126] The above provides a detailed description of the autonomous taxi operation and dispatching method integrating geographic information provided by the present invention. Specific examples are used in this embodiment to illustrate the principle and implementation of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation and application scope based on the idea of ​​the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

[0127] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined in these embodiments may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for scheduling autonomous taxi operations using integrated geographic information, characterized in that, Includes the following steps: S1: Real-time acquisition of multi-source geographic information data, including traffic situation layer data, road topology layer data, and POI demand hotspot layer data, to construct a multi-level real-time geographic information data network, including: The traffic situation layer acquires the floating satellite positioning trajectory of autonomous taxis and microwave radar data uploaded in real time by roadside RSUs, and connects to the urban traffic network to generate the spatiotemporal tensor of autonomous taxis. The road topology layer acquires the urban road network and marks the coverage radius of charging stations and the geofence restrictions of traffic control areas; The POI demand hot zone layer calculates the passenger transfer probability between POIs based on the spatiotemporal matrix of historical orders of autonomous taxis, and calculates the temporal correlation of demand fluctuations in combination with the urban POI functional classification map. Based on the temporal correlation of demand fluctuations, the POI attributes of the demand hot zone are marked on the urban POI functional classification map. S2: A dynamic weight matrix is ​​generated through a spatiotemporal convolutional neural network to adaptively fuse the data between the layers of the multi-level geographic information real-time data network and output a spatiotemporally coupled traffic state feature map. S3: Respond to passenger carpooling requests, extract the POI attributes and demand priority associated with the carpooling request, generate a spatiotemporal reachability domain based on the traffic state feature map, and filter a set of candidate vehicles that meet the preset response time threshold. S4: Construct a set of decision variables including vehicle-passenger matching relationships, route sequences, and ETAs, and solve for the optimal scheduling path based on a multi-objective optimization model; this includes the step of generating route sequences for the candidate vehicle set: For each candidate vehicle, based on the starting and ending points in the passenger carpooling request, a hierarchical path planning algorithm is used to generate multiple feasible paths to form a candidate path set, including: direct paths, detour paths, and combined paths; The ETA probability distribution and confidence interval for each path are generated through Monte Carlo sampling; A spatiotemporal conflict matrix is ​​constructed based on the candidate path set and ETA probability distribution, and the risk coefficient of time overlap of different vehicle paths at preset key nodes is identified as a decision variable. The steps for finding the optimal scheduling path based on a multi-objective optimization model include: Structured matching relation variables, path sequence variables, and ETA variables; Construct a multi-objective function that considers the resolution of risks and conflicts arising from time overlap; The optimal scheduling path after conflict resolution is generated using the NSGA-II multi-objective optimization algorithm; S5: Pre-embed replaceable anchor points in the optimal scheduling path. When a new carpooling request arrives, calculate the path insertion cost, trigger the replanning of the optimal scheduling path based on the path insertion cost, and update the ETA of the carpooling passengers.

2. The autonomous driving taxi operation scheduling method integrating geographic information according to claim 1, characterized in that, S1 further includes the step of: predicting the distribution of future demand hot zones to drive idle autonomous taxis to move in the direction of the demand gradient. Obtain the spatiotemporal tensor of autonomous taxis T and the vehicle's own condition V Vehicle states are mapped to a spatiotemporal grid to obtain spatiotemporal fusion features. : ； In the formula, ⊕ represents the number of spatiotemporal grids, and ⊕ represents feature concatenation. Pool This is a max pooling operation; For spatiotemporal fusion features Each spatiotemporal node in n Causal relationships are captured through cross-modal attention weights; A hierarchical masking mechanism is constructed, and cross-modal attention weights are corrected using temporal causal masks and spatial reachability masks to obtain corrected causal relationships. The reachable spatiotemporal nodes with causal relationships are dynamically connected to obtain a causal reachability graph. Based on the corrected cross-modal attention weights and causal reachability graphs, the activation probability of reachable spatiotemporal nodes is calculated, and dispatch instructions for autonomous taxis to predicted demand hotspots are generated.

3. The autonomous driving taxi operation scheduling method integrating geographic information according to claim 1, characterized in that, S2 includes the following steps: We employ 3D dilated convolution to capture multi-scale spatial relationships between data layers and process temporal dependencies based on axial Transformer. The average traffic situation, POI feature entropy, and topological complexity are calculated based on the multi-level geographic information real-time data network and input into the meta-learning weight generator to obtain the dynamic range of the weights. Based on the dynamic range of the weights, the traffic situation layer data, road topology layer data, and POI demand hotspot layer data are adaptively fused to output a spatiotemporally coupled traffic state feature map.

4. The autonomous driving taxi operation scheduling method integrating geographic information according to claim 3, characterized in that, S3 includes the following steps: In response to a passenger's carpooling request, the associated POI attributes, including POI category and service area, are extracted from the traffic state feature map based on the origin and destination locations. Based on the POI category and service radius, the historical demand average of the corresponding POI category and service range is extracted from the traffic state feature map according to the historical orders of autonomous taxis.

5. The autonomous driving taxi operation scheduling method integrating geographic information according to claim 1, characterized in that, S3 includes the following steps: Based on the preset response time threshold and traffic state feature map in the passenger carpooling request, the spatial reachability is combined with the time window to construct a four-dimensional dynamic reachability domain, including: spatial latitude and longitude, time window, driving speed, and vehicle driving constraints; and the reachability domain boundary is adjusted according to the demand priority. Based on the four-dimensional dynamic reachability domain, all possible locations that autonomous taxis can reach within a given time are obtained. The system acquires the real-time status of autonomous taxis, maps the real-time status parameters to a four-dimensional dynamic reachability domain, and uses a hybrid index structure to filter a set of candidate vehicles that meet a preset response time threshold.

6. The autonomous driving taxi operation scheduling method integrating geographic information according to claim 1, characterized in that, S5 includes the step of determining replaceable anchor points: The optimal scheduling path is divided into discrete segments by fixed distance or time window, and the historical time overlap risk conflict probability of the discrete segments is obtained. Search for and mark the preset key nodes within the radiation range of the discrete segment in the traffic state feature map, and obtain the congestion index of the marked nodes in the traffic state feature map. Based on the probability of risk conflict due to historical time overlap and the congestion index, a dynamic feasibility score is assigned to each node; Nodes with dynamic feasibility scores higher than a preset threshold are retained as replaceable anchor points.

7. The autonomous driving taxi operation scheduling method integrating geographic information according to claim 1, characterized in that, S5 includes: For each replaceable anchor point, calculate the overall cost after insertion: ； In the formula, This is the increment of the total travel time. This is the increment of the detour distance. This represents the variance of passenger waiting time. Replaceable anchors are sorted in ascending order of overall cost, with priority given to the best option. The smallest number of replaceable anchor points are used to replan the optimal scheduling path and update the ETA of the shared passengers.

Images