Intelligent dispatching method and device for unmanned taxi

Abstract

The application relates to the technical field of intelligent transportation, in particular to an intelligent scheduling method and device for unmanned taxis, wherein the method comprises the following steps: based on received vehicle data, passenger data and environment data, a spatial topology structure for connecting target unmanned taxis, passengers and charging stations is constructed; spatial dependency relationships among target unmanned taxi nodes, passenger nodes and charging station nodes are captured to generate spatial coding features and time coding features, and the spatial coding features and the time coding features are fused by combining corresponding weights to generate space-time features of the target unmanned taxis, so that passenger matching results and / or path planning and / or charging decisions of the target unmanned taxis are generated based on the space-time features. Therefore, the problems that traditional unmanned taxi scheduling methods do not fully consider complex factors in actual driving scenarios, scheduling decisions are still mostly single-target oriented, lack of dynamic adaptability and cannot meet the growing travel demands of passengers and the like are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent transportation, in particular to an intelligent scheduling method and device for unmanned taxis. BACKGROUND

[0002] With the development of automatic driving technology, unmanned taxis have become an important direction for future travel, but there are many problems in the existing scheduling of unmanned taxi vehicles.

[0003] In related technologies, the scheduling of unmanned taxi vehicles still mostly adopts traditional scheduling methods, that is, all unmanned taxis are scheduled based on simple rules.

[0004] However, in related technologies, the traditional scheduling method for unmanned taxis does not fully consider the complex factors in the actual driving scene, resulting in long response time for calling a car, ineffective optimization of empty car mileage, uneven service level, and other problems, and the scheduling decision is still mostly single-target oriented, lacks dynamic adaptability, and cannot meet the growing travel demand of passengers, which needs to be solved urgently. SUMMARY

[0005] The present application provides an intelligent scheduling method and device for unmanned taxis to solve the problems in related technologies that the traditional scheduling method for unmanned taxis does not fully consider the complex factors in the actual driving scene, resulting in long response time for calling a car, ineffective optimization of empty car mileage, uneven service level, and other problems, and the scheduling decision is still mostly single-target oriented, lacks dynamic adaptability, and cannot meet the growing travel demand of passengers.

[0006] The first aspect of the present application provides an intelligent scheduling method for unmanned taxis, comprising the following steps: based on the received vehicle data of the target unmanned taxi, the passenger data and the environmental data of the target unmanned taxi allowed driving area, a spatial topology structure for connecting the target unmanned taxi, the passenger and the charging station is constructed; based on the spatial topology structure, the spatial dependence relationship between the target unmanned taxi node, the passenger node and the charging station node in the spatial topology structure is captured, to generate the spatial coding feature and the time coding feature between the target unmanned taxi, the passenger and the charging station in combination with the spatial dependence relationship and the time point of capturing the spatial dependence relationship; the weight of the spatial coding feature and the weight of the time coding feature are calculated, to fuse the spatial coding feature and the time coding feature according to the weight of the spatial coding feature and the weight of the time coding feature, generate the spatiotemporal feature of the target unmanned taxi, and generate the passenger matching result and / or path planning and / or charging decision of the target unmanned taxi based on the spatiotemporal feature.

[0007] Optionally, in an embodiment of the present application, the spatial topology structure further comprises edges connecting the target unmanned taxi node and the passenger node and edges connecting the target unmanned taxi node and the charging station node; the target unmanned taxi node comprises at least one of the vehicle coordinates, the vehicle speed, the remaining power and the passenger carrying state of the target unmanned taxi; the passenger node comprises at least one of the starting point coordinates, the end point coordinates, the order creation time, the waiting time and the ride demand of the passenger; and the charging station node comprises at least one of the coordinates, the available pile number and the queuing time of the charging station.

[0008] Optionally, in an embodiment of the present application, the generation of the spatial encoding features and the time encoding features between the target unmanned taxi, the passenger and the charging station in combination with the spatial dependence and the time point of capturing the spatial dependence comprises: calculating the similarity of the node features between the target unmanned taxi node, the passenger node and the charging station node; and extracting the spatial encoding features according to the similarity of the node features.

[0009] Optionally, in an embodiment of the present application, the generation of the spatial encoding features and the time encoding features between the target unmanned taxi, the passenger and the charging station in combination with the spatial dependence and the time point of capturing the spatial dependence comprises: obtaining the abnormal order information of the passenger; and generating the time encoding features in combination with the abnormal order information and the historical memory information corresponding to the time point.

[0010] Optionally, in an embodiment of the present application, before capturing the spatial dependence between the target unmanned taxi node, the passenger node and the charging station node in the spatial topology structure, the method further comprises: obtaining the task target of the target unmanned taxi, and determining a matching loss, a path loss, a charging loss and a regularization loss and their corresponding weights for training a preset joint model according to the task target; and constructing the preset joint model by using the matching loss, the path loss, the charging loss, the regularization loss and their corresponding weights, so as to capture the spatial dependence by using the preset joint model.

[0011] The second aspect embodiment of the present application provides a kind of unmanned taxi intelligent scheduling device, comprising: construction module, for based on the vehicle data of received target unmanned taxi, passenger data and the environmental data of the target unmanned taxi allowed driving area, constructs the space topology for connecting the target unmanned taxi, passenger and charging station;Generation module, for based on the space topology, capture the space dependence between the target unmanned taxi node, passenger node and charging station node in the space topology, to combine the space dependence and the time point of capturing the space dependence, generate the space coding feature and time coding feature between the target unmanned taxi, the passenger and the charging station;Fusion module, for calculating the weight of the space coding feature and the weight of the time coding feature, to fuse the space coding feature and the time coding feature according to the weight of the space coding feature and the weight of the time coding feature, generate the space-time feature of the target unmanned taxi, to generate the passenger matching result and / or path planning and / or charging decision of the target unmanned taxi based on the space-time feature.

[0012] Optionally, in one embodiment of the present application, further comprising: the space topology includes the edge connecting the target unmanned taxi node and the passenger node and the edge connecting the target unmanned taxi node and the charging station node;The target unmanned taxi node includes at least one of the vehicle coordinates, vehicle speed, remaining power and passenger carrying state of the target unmanned taxi;The passenger node includes at least one of the starting point coordinates, terminal point coordinates, order creation time, waiting time, ride demand of the passenger;The charging station node includes at least one of the coordinates, available pile number, queue time of the charging station.

[0013] Optionally, in one embodiment of the present application, the generation module comprises: a calculation unit for calculating the similarity of node features between the target unmanned taxi node, the passenger node and the charging station node;Generation unit, for extracting the space coding feature according to the similarity of node features.

[0014] Optionally, in one embodiment of the present application, the generation module comprises: an acquisition unit for acquiring abnormal order information of the passenger;Combination unit, for combining the time coding feature and the historical memory information corresponding to the time point to generate the time coding feature.

[0015] Optionally, in an embodiment of the present application, further comprising: an acquisition module, configured to acquire a task target of the target unmanned taxi before capturing the spatial dependency among the target unmanned taxi node, the passenger node and the charging station node in the spatial topology, and determine a matching loss, a path loss, a charging loss, a regularization loss and their corresponding weights for training a preset joint model according to the task target; and a capturing module, configured to construct the preset joint model by using the matching loss, the path loss, the charging loss, the regularization loss and their corresponding weights, so as to capture the spatial dependency by using the preset joint model.

[0016] An embodiment of the third aspect of the present application provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the intelligent scheduling method of the unmanned taxi as described in the above embodiments.

[0017] An embodiment of the fourth aspect of the present application provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the intelligent scheduling method of the unmanned taxi as described above.

[0018] An embodiment of the fifth aspect of the present application provides a computer program product, comprising a computer program, and the computer program is executed to implement the intelligent scheduling method of the unmanned taxi as described above.

[0019] The embodiment of the present application can capture the spatial dependence relationship among the target unmanned taxi, the passenger and the charging station, generate spatial coding features and time coding features and perform weighted fusion to generate spatio-temporal features, so as to generate the passenger matching result and / or path planning and / or charging decision of the target unmanned taxi based on the spatio-temporal features. In this way, the multi-dimensional data such as vehicle features (position, speed, power, etc.), passenger features (position, order demand, etc.), environment features (road condition, charging station state, etc.) are combined, the spatio-temporal features of the target unmanned taxi are effectively captured through the joint model, the spatial topological structure and the time dynamic law are deeply coded and fused, and the problem of one-sided scheduling decision caused by the spatio-temporal information fragmentation of the traditional model is solved. At the same time, the vehicle-passenger matching result, the path planning result and the charging scheduling result of the target unmanned taxi are synchronously output by using the model, the multi-service demand can be effectively balanced, the problem of insufficient scheduling strategy adaptation in complex scenarios is solved, and the efficient, real-time and intelligent scheduling of large-scale unmanned taxi fleet in complex urban environment is helpful. In this way, the problems in the related art that the traditional unmanned taxi scheduling method does not fully consider the complex factors in the actual driving scene, resulting in long car calling response time, non-effective optimization of empty car mileage, uneven service level and other problems, and the scheduling decision is still single-target oriented, lacks dynamic adaptability and cannot meet the growing travel demand of passengers are solved.

[0020] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0021] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, taken in conjunction with the accompanying drawings, in which:

[0022] Figure 1 A flowchart of an intelligent scheduling method of an unmanned taxi according to an embodiment of the present application is shown in the figure;

[0023] Figure 2 A schematic diagram of an ST-GCN-GRU joint model architecture of an embodiment of the present application is shown in the figure;

[0024] Figure 3 A flowchart of model output decision of an embodiment of the present application is shown in the figure;

[0025] Figure 4 A structural schematic diagram of an intelligent scheduling device of an unmanned taxi according to an embodiment of the present application is shown in the figure;

[0026] Figure 5 A structural schematic diagram of an electronic device according to an embodiment of the present application is shown in the figure.

[0027] Reference Signs:

[0028] 10 - intelligent dispatching device of unmanned taxi; 100 - construction module, 200 - generation module and 300 - fusion module; 501 - memory, 502 - processor and 503 - communication interface. DETAILED DESCRIPTION

[0029] Embodiments of the present application are described in detail below with reference to the accompanying drawings, in which the same or similar reference signs represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by reference to the accompanying drawings are exemplary and are intended to explain the present application, and cannot be understood as a limitation of the present application.

[0030] The intelligent dispatching method and device of unmanned taxi of the embodiments of the present application are described below with reference to the accompanying drawings. In view of the problems in the prior art that the traditional unmanned taxi dispatching method does not fully consider the complex factors in the actual driving scene, resulting in long car calling response time, ineffective optimization of empty car mileage, uneven service level, etc., and the dispatching decision is still mostly single-target oriented, lacking dynamic adaptability, and unable to meet the growing travel demand of passengers, the present application provides an intelligent dispatching method of unmanned taxi, in which the spatial dependency between the target unmanned taxi, the passenger and the charging station can be captured, the spatial coding features and the time coding features are generated and weighted and fused to generate spatio-temporal features, and the passenger matching result and / or path planning and / or charging decision of the target unmanned taxi are generated based on the spatio-temporal features. Thus, by combining vehicle features (position, speed, power, etc.), passenger features (position, order demand, etc.), environmental features (road conditions, charging station status, etc.), etc., through a joint model, the spatio-temporal features of the target unmanned taxi are effectively captured, the spatial topological structure and the time dynamic law are deeply coded and fused, and the problem of one-sided dispatching decision caused by the split of spatio-temporal information in traditional models is solved. At the same time, the vehicle-passenger matching result, the path planning result, and the charging dispatching result of the target unmanned taxi are simultaneously output by the model, which can effectively balance multiple business demands and solve the problem of insufficient adaptation of dispatching strategies in complex scenarios, helping to achieve efficient, real-time, and intelligent dispatching of large-scale unmanned taxi fleets in complex urban environments. Thus, the problems in the related art that the traditional unmanned taxi dispatching method does not fully consider the complex factors in the actual driving scene, resulting in long car calling response time, ineffective optimization of empty car mileage, uneven service level, etc., and the dispatching decision is still mostly single-target oriented, lacking dynamic adaptability, and unable to meet the growing travel demand of passengers, etc. are solved.

[0031] Specifically, Figure 1A flowchart of an intelligent scheduling method of an unmanned taxi provided by an embodiment of the present application.

[0032] As shown in the figure, the intelligent scheduling method of the unmanned taxi comprises the following steps: Figure 1

[0033] In step S101, based on the received vehicle data of the target unmanned taxi, passenger data and environmental data of the target unmanned taxi allowed driving area, a spatial topology structure for connecting the target unmanned taxi, the passenger and the charging station is constructed.

[0034] In some embodiments, the present application can construct a certain spatial topology structure based on the received vehicle data of the target unmanned taxi, passenger data and environmental data of the target unmanned taxi allowed driving area, connect the target unmanned taxi (node), the passenger (node) and the charging station in the target unmanned taxi allowed driving area, and facilitate understanding the relationship between the target unmanned taxi and the passenger and the charging station.

[0035] Among them, the target unmanned taxi here can be understood as an unmanned taxi car as a scheduling and management object in a certain scenario, which has automatic driving capability and can realize autonomous navigation, obstacle avoidance, order receiving and passenger pickup and other functions through vehicle-mounted sensors, positioning systems, intelligent algorithms and other ways without driver operation. For example, a batch of unmanned taxis running in a certain city area, each of which can be understood as a target unmanned taxi and become an individual of scheduling.

[0036] The target unmanned taxi allowed driving area here refers to a specific geographical range approved or set by the system, which allows the target unmanned taxi to perform a series of driving and passenger pickup activities. In this area, the unmanned taxi can operate according to the scheduling instructions and certain rules, but it may be restricted from driving outside the area. For example, the core area of the city center, a certain industrial park or the area around the airport may be the allowed driving area of a batch of target unmanned taxis.

[0037] The vehicle data of the target unmanned taxi, the passenger data and the environmental data of the target unmanned taxi allowed driving area received by the embodiment of the present application can be but not limited to from multiple data sources or multiple clouds and terminals.

[0038] ​For example, in order to guarantee the accuracy and convenience of data sending, transmission and reception, the application can, but is not limited to, build a certain "cloud-edge-terminal" three-level collaborative architecture. The cloud can rely on a GPU cluster and a certain computer platform to carry out model training, global scheduling and big data storage, and process cross-regional complex decisions; the edge can be set as a lightweight server deployed in the core area of the city and the charging station, which is convenient for realizing regional real-time scheduling, data caching and charging optimization, and reducing the pressure on the cloud; the terminal can be set as a vehicle-mounted intelligent terminal (integrating sensors and lightweight inference modules), a passenger APP (Application) and a management platform, etc.

[0039] Among them, the terminal device and the edge can, but are not limited to, transmit real-time data through 5G, V2X (Vehicle-to-Everything), Bluetooth and other communication technologies; the edge and the cloud can, but are not limited to, use MQTT (Message Queuing Telemetry Transport), HTTP (HyperText Transfer Protocol) / HTTPS (HyperText Transfer Protocol Secure) protocol for data synchronization to guarantee the real-time and reliability of data transmission; for critical scheduling instructions (such as emergency order scheduling), priority queue transmission can be used to ensure priority processing. Through the "cloud-edge-terminal" three-level collaborative architecture, the application can process data collection, instruction execution and operation monitoring in a process, and build an end-to-end closed loop.

[0040] In addition, real-time data (such as vehicle location, order status) is temporarily stored in the edge Redis (RemoteDictionary Server, a high-performance open-source in-memory database) cache for fast query and real-time decision-making; historical data (such as vehicle historical trajectory, order historical record) is uploaded to the cloud data lake, stored by time partitioning and topic classification, providing long-term data support for model training and business analysis.

[0041] Further, on the basis of the "cloud-edge-terminal" three-level collaborative architecture, the application embodiment can, but is not limited to, set a certain "collection-processing-transmission-application" multi-source data closed loop process in the process of receiving vehicle data, passenger data and environmental data of the target unmanned taxi allowed driving area of the target unmanned taxi and before and after the process of receiving.

[0042] For example, the application can first collect these data through the terminal in the three-level collaborative architecture:

[0043] Vehicle data: through the sensors of the vehicle terminal and the CAN (Controller Area Network) bus, vehicle position (latitude and longitude), speed, acceleration, power, cruising range, fault code and other information are collected at a sampling frequency of 10 Hz to reflect the real-time vehicle running state.

[0044] Passenger data: order information is collected by the passenger APP, mini program and other passenger use terminals, including but not limited to the passenger's pick-up and drop-off location, call time, number of passengers, special needs (such as barrier-free vehicles), etc., and the real-time location of the passenger can be obtained through the positioning module built-in the passenger use terminal, which is helpful for assisting the dispatch vehicle to accurately pick up;

[0045] Environmental data: real-time traffic from various map third-party platforms, including but not limited to congestion level, road construction, accident information, etc.; charging station distribution, including but not limited to charging pile location, charging pile number, idle state, electricity price, etc.; weather, such as rainfall, snowfall, temperature, etc., so as to provide environmental constraints for dispatch decision.

[0046] The embodiment of the present application can build a "cloud-edge-terminal" three-level cooperative deployment architecture, which is convenient for realizing cross-regional complex decision and regional real-time scheduling cooperation, solving the problems of traditional scheduling computing power pressure and response lag, supporting large-scale vehicle fleet efficient scheduling; at the same time, by using the closed loop of multi-source data "collection-processing-transmission-application", the full-factor data is cleaned and fused, and the hierarchical storage ensures real-time decision and model training, which can effectively improve the scheduling timeliness and accuracy.

[0047] Optionally, in an embodiment of the present application, the spatial topology structure includes edges connecting the target unmanned taxi node and the passenger node and edges connecting the target unmanned taxi node and the charging station node; the target unmanned taxi node includes at least one of the vehicle coordinates, the vehicle speed, the remaining power and the passenger carrying state of the target unmanned taxi; the passenger node includes at least one of the starting point coordinates, the end point coordinates, the order creation time, the waiting time, the riding demand of the passenger; the charging station node includes at least one of the coordinates, the available pile number and the queuing time of the charging station.

[0048] As can be understood based on the related description in other embodiments, the spatial topology structure in the present application can connect the target unmanned taxi, the passenger and the charging station, that is, it can represent the complex relationship between the target unmanned taxi, the passenger and the charging station.

[0049] Based on this, in the actual implementation process, the spatial topology in the application includes but is not limited to target unmanned taxi nodes, passenger nodes and charging station nodes, and edges connecting the target unmanned taxi nodes and the passenger nodes and edges connecting the vehicle and the charging station nodes.

[0050] The target unmanned taxi node includes but is not limited to at least one of the dynamic attributes of the target unmanned taxi, such as the vehicle coordinates, the vehicle speed, the remaining power and the passenger carrying state, and can be updated once every 3 seconds according to the vehicle data of the target unmanned taxi, the passenger data and the environmental data of the target unmanned taxi driving area.

[0051] The passenger node includes but is not limited to at least one of the information of the passenger, such as the starting point coordinates, the end point coordinates, the order creation time, the waiting time and the riding demand (special demand), and the passenger node can be created when the passenger's order is generated.

[0052] The charging station node includes at least one of the information of the charging station, such as the coordinates, the available pile number and the queuing time, and can be updated once every 30 seconds according to the vehicle data of the target unmanned taxi, the passenger data and the environmental data of the target unmanned taxi driving area.

[0053] In addition, the embodiment of the application can also but not limited to determine the weight (adaptation weight) of the edge connecting the target unmanned taxi node and the passenger node according to the Euclidean distance between the vehicle coordinates and the starting point coordinates, the estimated arrival time of the passenger's order and the power threshold of the target unmanned taxi, and the mathematical expression can be but not limited to represented as:

[0054] ,

[0055] wherein, is the Euclidean distance between the vehicle and the passenger starting point, is the estimated arrival time, is the power threshold (such as 20% remaining power), is the adaptation weight (the adaptation weight of the edge between the target unmanned taxi node and the passenger node, used to measure the matching degree of the two, the larger the value, the better the adaptation), , and are the weight coefficients related to the Euclidean distance, the estimated arrival time and the power difference (which can be set and optimized according to the actual operation scene, the scheduling strategy, etc., such as increasing the value of if the distance between the vehicle and the passenger is more important), , and respectively, are decay coefficients of the Euclidean distance, the predicted arrival time and the power difference (used to control the decay rate of the influence of the corresponding factor on the fitness weight, for example The greater the Euclidean distance The influence of on the fitness decays faster with the increase of the distance, which means that the factor quickly reduces its role in the fitness calculation when the distance is slightly far away, is a natural constant.

[0056] In addition, the embodiment of the present application can also, but is not limited to, determining the weight (charging necessity weight) of the edge connecting the target unmanned taxi node and the charging station node according to the remaining power and the available number of piles, and the mathematical expression can be, but is not limited to, represented as:

[0057] ,

[0058] The formula quantifies the charging necessity of the vehicle to the charging station from two dimensions of vehicle power demand and charging station pile resources, balances the contradiction between the urgent need for charging of the vehicle and the sufficient pile resources of the charging station through the dynamic configuration of and , so that the scheduling strategy can guarantee the vehicle endurance and avoid idle / congestion of charging piles.

[0059] wherein, represents the charging necessity weight, and respectively represent the power-related weight coefficient (used to adjust the influence degree of the remaining power of the vehicle on the charging necessity), represents a natural constant, represents the available number of charging piles.

[0060] The weight of the edge connecting the target unmanned taxi node and the charging station node is inversely proportional to the remaining power of the vehicle and inversely proportional to the available number of charging piles.

[0061] The embodiment of the present application can construct a spatial topology structure containing vehicle information, passenger information and charging station information of multiple target unmanned taxis, which is helpful to clearly and concisely understand the complex spatial dependency relationship between the target unmanned taxis, passengers and charging stations and various data contained therein.

[0062] Optionally, in an embodiment of the present application, based on the spatial topology, the spatial dependence relationship among the target unmanned taxi node, the passenger node and the charging station node in the spatial topology is captured, including: preprocessing the vehicle data, the passenger data and the environmental data, and associating the preprocessed vehicle data, the passenger data and the environmental data according to the time stamp and the spatial position, to generate a time-spatial dimension scenario data set; extracting at least one spatio-temporal feature and / or at least one statistical feature and / or at least one correlation feature in the scenario data set, to generate vehicle features, passenger features and environmental features based on at least one spatio-temporal feature and / or at least one statistical feature and / or at least one correlation feature in the scenario data set and the preprocessed vehicle data, passenger data and environmental data; and combining the vehicle features, the passenger features and the environmental features with the spatial topology, to capture the spatial dependence relationship.

[0063] In certain embodiments, after receiving the vehicle data, the passenger data and the environmental data of the target unmanned taxi in the allowed driving area of the target unmanned taxi, considering that data from various channels have different data dimensions, the present application can first preprocess the vehicle data, the passenger data and the environmental data to ensure that all data are in the same data dimension, and then associate the preprocessed vehicle data, the passenger data and the environmental data according to the time stamp and the spatial position, to generate a time-spatial dimension scenario data set.

[0064] For example, Table 1 is a data information table of an embodiment of the present application, which can but is not limited to represent as follows:

[0065] Table 1

[0066]

[0067] Firstly, the embodiment of the present application can perform data cleaning on the received vehicle data, passenger data and environmental data, including but not limited to: (1) removing all abnormal values in the data, for example, a value in the data is obviously too different from other values (such as most speed values in the vehicle data are in the range of 40-60 km / h, and a value is suddenly 90 km / h, which is abnormal and should be removed) or obviously does not conform to the actual situation (such as vehicle speed exceeding 200 km / h, passenger position coordinates exceeding the city range) and the like; (2) missing value filling, for example, using mean value method or interpolation method to fill the missing positions in the data, to ensure the quality of the data.

[0068] Then, the embodiment of the present application can associate the pre-processed vehicle data, passenger data and environment data according to timestamps and spatial positions, and construct a scenario data set with "time-space" as the dimension, i.e. associate these data according to the time dimension and the space dimension, such as the associated data of the distribution of the autonomous taxi, passenger demand, road conditions and charging station status in a certain area at a certain time, etc.

[0069] Then, the embodiment of the present application can extract at least one spatio-temporal feature and / or at least one statistical feature and / or at least one correlation feature (at least one spatio-temporal feature, or at least one statistical feature, or at least one correlation feature, or at least one spatio-temporal feature and at least one statistical feature, or at least one spatio-temporal feature and at least one correlation feature, or at least one statistical feature and at least one correlation feature, or at least one spatio-temporal feature, at least one statistical feature and at least one correlation feature) from the scenario data set, so as to capture the spatial dependence relationship between multiple nodes of the spatial topology based on at least one spatio-temporal feature and / or at least one statistical feature and / or at least one correlation feature.

[0070] Here, the spatio-temporal feature can be understood as simultaneously associating time information (when it occurs) and spatial information (where it occurs), mining the rules, trends or distribution patterns of data in the spatio-temporal dimension, which reflects the behavior characteristics of things in a specific time and space. The statistical feature can be understood as a mathematical calculation and summary of data in a specific dimension (single dimension or multiple dimensions), and the overall condition, central tendency, dispersion degree or dynamic change of data expressed by a specific numerical value, which is an intuitive description of the data volume. The correlation feature can be understood as mining the correlation between different objects (such as vehicle-passenger, vehicle-charging station) or different dimensions (such as distance-time, power-order), and quantifying the strength, distance or influence degree of such relationship by a numerical value.

[0071] For example, the present application can extract the spatio-temporal pattern of vehicle historical trajectory, the regional heat of passenger call position, etc. from the scenario data set, extract the statistical features such as the number of orders per unit time in a certain area and the average speed of vehicles, and extract the correlation features such as the distance between vehicles and passengers and the path time consumption of vehicles to charging stations.

[0072] Then, the embodiment of the present application can combine the extracted spatio-temporal features, statistical features and correlation features, and combine the pre-processed vehicle data, passenger data and environment data to generate vehicle features, passenger features and environment features, so as to capture the spatial dependence relationship between the target autonomous taxi node, passenger node and charging station node according to the vehicle features, passenger features and environment features.

[0073] Wherein, the vehicle features include but are not limited to the features such as the position, speed, power of the vehicle, the passenger features include but are not limited to the features such as the position and order demand of the passenger, and the environment features include but are not limited to the features such as the road condition and charging station state.

[0074] And, for these vehicle features, passenger features and environment features, the embodiment of the application needs to normalize (such as normalizing the vehicle speed to the interval [0, 1]) and / or encode (such as One-Hot encoding of the special needs of the passenger) according to the actual situation of the features, so as to generate a feature vector according to the normalized and encoded features, and then perform subsequent data processing.

[0075] Specifically, for the spatial coordinates involved in the features, the embodiment of the application can perform normalization processing (longitude and latitude processing), for example, the embodiment of the application can but is not limited to taking a certain place in the area or city where the target driverless taxi is located as the origin, and then calculating the relative coordinates of the spatial coordinates as the vector elements of the feature vector. Wherein, (x, y) is the relative coordinate, (x0, y0) is the coordinate origin, and (x, y) is the actual spatial coordinate of the feature.

[0076] For the distance involved in the features, the embodiment of the application can but is not limited to calculating the relative distance by using the earth radius, and then performing standardization processing, for example, the calculation formula of the relative distance in the embodiment of the application can but is not limited to being expressed as:

[0077]

[0078] Wherein, dlon is the longitude distance, dlat is the latitude distance, lon1 and lon2 are the longitudes of two points, lat1 and lat2 are the latitudes of the two points, lat0 is the reference latitude for dimension correction, and R is the earth radius.

[0079] Then, the embodiment of the application can use Min-Max to scale the relative distance to [-1, 1] (standardization), wherein the scaling formula of the relative distance can but is not limited to being expressed as:

[0080] ,​​​​​​​​​​​​​​​​​

[0081] wherein, , is a relative distance obtained after scaling (normalization), is a minimum value that the distance can take, is a maximum value that the distance can take, is a minimum value that the distance can take, is a maximum value that the distance can take.

[0082] Additionally, for dynamic characteristics of the vehicle speed, the vehicle power, etc. of the target unmanned taxi, embodiments of the present application can, but are not limited to, use Z-Score standardization and Min-Max scaling respectively to process them. For example, for the speed, embodiments of the present application can, but are not limited to, use Z-Score standardization processing, and the Z-Score standardization processing formula can, but is not limited to, be expressed as: wherein, is a standardized value, is an original data, is a mean value of the feature, is a standard deviation of the feature; for the power, embodiments of the present application can, but are not limited to, use Min-Max scaling processing, and the Min-Max scaling processing formula can, but is not limited to, be expressed as: wherein, is a scaled value, is a feature value to be scaled, is a minimum value of the feature, is a maximum value of the feature.

[0083] Embodiments of the present application can generate vehicle features, passenger features and environment features based on at least one spatio-temporal feature and / or at least one statistical feature and / or at least one correlation feature of the acquired scenario data and perform normalization or encoding, convert multiple features into structured data, and then combine the spatial topology to capture the spatial dependency between the target unmanned taxi node, the passenger node and the charging station node in the spatial topology, and prepare for the intelligent scheduling of the target unmanned taxi.

[0084] In step S102, based on the spatial topology, the spatial dependency between the target unmanned taxi node, the passenger node and the charging station node in the spatial topology is captured, and the spatial coding features and the time coding features between the target unmanned taxi, the passenger and the charging station are generated in combination with the spatial dependency and the time point of capturing the spatial dependency.

[0085] In some embodiments, after obtaining the vehicle data of the target unmanned taxi, the passenger data, the environmental data of the target unmanned taxi allowed driving area, and constructing the spatial topology for connecting the target unmanned taxi, the passenger, and the charging station, the embodiment of the application can capture the spatial dependency relationship between the target unmanned taxi node, the passenger node, and the charging station node in the spatial topology based on the spatial topology.

[0086] Here, capturing the spatial dependency relationship between the target unmanned taxi node, the passenger node, and the charging station node in the spatial topology can be understood as analyzing how the target unmanned taxi, the passenger, and the charging station influence and connect with each other at each time slice (for example, at this moment in the last minute).

[0087] After obtaining the spatial dependency relationship between the target unmanned taxi node, the passenger node, and the charging station node, the embodiment of the application can generate the spatial encoding feature and the time encoding feature between the target unmanned taxi, the passenger, and the charging station based on the spatial dependency relationship and the time point when the spatial dependency relationship is captured.

[0088] The embodiment of the application can capture the spatial dependency relationship between the target unmanned taxi node, the passenger node, and the charging station node to determine how the target unmanned taxi, the passenger, and the charging station influence and connect with each other, so that the influence of all aspects can be considered when scheduling the target unmanned taxi, and the optimization degree of the final scheduling scheme can be improved.

[0089] Optionally, in an embodiment of the application, before capturing the spatial dependency relationship between the target unmanned taxi node, the passenger node, and the charging station node in the spatial topology, the method further comprises: obtaining a task target of the target unmanned taxi, and determining a matching loss, a path loss, a charging loss, a regularization loss, and their corresponding weights for training a preset joint model according to the task target; and constructing the preset joint model by using the matching loss, the path loss, the charging loss, the regularization loss, and their corresponding weights, so as to capture the spatial dependency relationship by using the preset joint model.

[0090] As a possible implementation manner, when capturing the spatial dependency relationship between the target unmanned taxi node, the passenger node, and the charging station node, and generating the spatial encoding feature and the time encoding feature, the embodiment of the application can but is not limited to using the preset joint model.

[0091] The preset joint model can be understood as a space-time fusion model (ST-GCN-GRU) that deeply encodes and fuses the space topology structure and the time dynamic law.

[0092] The space encoding layer (ST-GCN layer) is mainly used to establish the space topology relationship of vehicles, passengers and facilities based on the space topology structure (for example, the accessibility of vehicles to passengers and the distance to charging stations), and then dynamically capture node dependence through graph convolution (for example, the response priority of vehicles to passenger demand during peak hours).

[0093] The time encoding layer (GRU layer) is mainly used to mine the time sequence dynamic law (for example, the periodicity of traffic flow and the trend of power change), and then memorize the key historical state through the gating mechanism (for example, the influence of early morning peak congestion on subsequent scheduling).

[0094] The space-time fusion layer (attention layer) is mainly used to dynamically allocate the space-time feature weight (for example, to enhance the time feature during peak hours and to focus on the spatial distribution during off-peak hours), and to output the precise fusion decision basis.

[0095] The output layer is mainly used to synchronously generate vehicle-passenger matching, path planning and charging scheduling results, and to realize the collaborative decision of "demand response-path optimization-range guarantee".

[0096] For the most important data set in the model training process, the embodiment of the application can select the historical order data, vehicle operation data and environment data of the target unmanned taxi from the cloud data, and then divide the training set (80%), the verification set (10%) and the test set (10%) according to time, to ensure that each set covers different scenes such as weekdays / weekends / holidays. Then, the embodiment of the application can perform enhancement processing on the data, adopt time window sliding (sliding step 15 minutes), space region rotation (±30° to simulate different driving directions), noise injection (to simulate sensor errors) and other methods to expand the data, so as to improve the generalization ability of the model.

[0097] Then, the embodiment of the application can obtain what the task targets of the target unmanned taxi are. For example, in the embodiment of the application, the task targets of the target unmanned taxi include but are not limited to the matching task of the target unmanned taxi and the passenger, the path task and the charging task.

[0098] Because in the actual scheduling process, the adaptability of these task targets to the current target unmanned taxi and the actual situation of the joint model need to be comprehensively considered, therefore, the embodiment of the application constructs a matching loss function for training the basic joint model , a path loss function , a charging loss function , and a regularization loss function (for preventing overfitting of the joint model) for measuring the vehicle matching accuracy, the path accuracy, the charging strategy accuracy, and whether the joint model is overfitted, respectively.

[0099] Further, the embodiment of the application also sets different weight coefficients according to different task targets, so that in the actual application process, the multi-target weighted loss function can be used to balance the matching, path, and charging three task targets by dynamically weighting these loss functions. Specifically, the embodiment of the application can dynamically determine the importance of each task target according to the actual operation stage of the target unmanned taxi, dynamically adjust the weights of matching ( ), the weights of path ( ), the weights of charging ( ), and the weights of model fitting ( ), such as increasing the weight of matching between the target unmanned taxi and the passenger in the peak period to shorten the waiting time of the passenger, and increasing the weight of preferential optimization of charging when the vehicle power is tight, so as to control the influence degree of each sub-task on the total loss.

[0100] Finally, the embodiment of the application can construct a certain joint model by using the matching loss, the path loss, the charging loss, the regularization loss, and the corresponding weights, so as to capture the spatial dependency between the target unmanned taxi node, the passenger node, and the charging station node by using the joint model.

[0101] In the embodiment of the application, the default weight configuration is : : : =5:3:2:1, and the multi-target weighted loss function can be but is not limited to represented as:

[0102]

[0103] wherein, is the total loss of the model, is the weight of matching, is the weight of path, is the weight of charging, is the weight of model fitting, is the matching loss function, is the path loss function, is a charging loss function, is a regularization loss function.

[0104] is a matching loss, which is a binary cross-entropy loss, can measure the deviation of the predicted matching probability from the true label, and penalize the deviation of the "matching probability" predicted by the joint model from the true result, and its mathematical expression can be but is not limited to represented as:

[0105]

[0106] wherein, is the number of samples (how many groups of "vehicle-passenger" matching pairs need to be judged), is a true label, indicating whether the first group of matching pairs is really successful (1=success, 0=failure), represents the matching probability predicted by the joint model, that is, the probability that the joint model considers the first group of "vehicle-passenger" successful matching (range 0~1).

[0107] and the logic of this loss function is: if the real matching ( =1), but the joint model prediction probability is very low→ the loss will be very large (penalize the joint model); vice versa. Thus, the embodiment of the present application can ensure that the joint model has the ability to learn "high probability matching high priority order" first, such as the matching probability of an emergency order needs to be ≥0.9.

[0108] is a path loss, which is a mean square error (MSE), used to calculate the coordinate deviation of the predicted path and the true path, and its mathematical expression can be but is not limited to represented as:

[0109]

[0110] wherein, represents the number of path points (on a planned path, how many coordinate points are sampled, such as taking 1 point every 500 meters), represents the path point predicted by the joint model (such as the predicted coordinate of the first sampled point), represents the true optimal path point (such as the actual coordinate of the first sampled point (such as the path point planned by Gaode map)), represents the square of the Euclidean distance, used to calculate the straight-line distance between the predicted point and the true point, and then square the error to make the deviation of the point more severe, so as to constrain the joint model to generate a path that conforms to the traffic rules (such as avoiding reverse driving and illegal lane changing). It should be noted that the error should be ≤50 meters to ensure navigation accuracy.

[0111] charging loss (L ) is a categorical cross-entropy loss used to optimize the accuracy of charging station selection, and its mathematical expression can be but is not limited to represented as:

[0112]

[0113] wherein, represents the number of vehicles to be charged, represents the total number of charging stations, represents the real charging selection, i.e., whether the first vehicle has really selected the first charging station (1 = selected, 0 = not selected), represents the charging probability predicted by the model, and the model considers the probability (range 0~1) of the first vehicle selecting the first charging station. Thus, it can be ensured that low-battery vehicles (battery level ≤ 20%) are preferentially dispatched to the nearest available charging station, reducing order loss due to insufficient endurance. Regularization loss (L ) is L2 regularization used to constrain the parameter norm of the joint model to prevent overfitting of the joint model, and its mathematical expression can be but is not limited to represented as:

[0114]

[0115]

[0116]

[0117] wherein, represents the regularization strength, which is a manually set coefficient used to control the degree of punishment, and the greater the coefficient, the stricter the punishment; represents all trainable parameters of the model, such as ( represents the weight matrix of the "vehicle-passenger matching task" in the model, which belongs to ), the ( represents the weight matrix of the "path planning task" in the model, etc. represents the L2 norm square of the parameter, i.e., the sum of squares of each value in the parameter, and the greater the parameter, the greater this part of the loss.

[0118] Further, Figure 2 is a schematic diagram of the ST-GCN-GRU joint model architecture of one embodiment of the present application, as shown in Figure 2

[0119] ​​​​​(1) Input layer: receives processed vehicle features (location, speed, power), passenger features (location, order demand), environmental features (road conditions, charging station status), etc. Multimodal data is organized in (B, T, N, F) dimensions, where B = batch size, T = time step, N = node number, and F = feature dimension.

[0120] (2) ST-GCN module: Preliminary association of spatial encoding features and time dimension, output (B, T, N, D) dimensional spatial encoding features, where D = hidden layer dimension.

[0121] Graph convolution layer: extract spatial dependencies;

[0122] Spatial attention: dynamically adjust neighbor node weights;

[0123] Space-time fusion: integrate spatial features and time information;

[0124] (3) GRU module: combine the outputs of update gate and reset gate to generate the current time encoding candidate state, output (B, T, N, D) dimensional time encoding features, and ST-GCN spatial features;

[0125] Update gate (z): control the retention proportion of historical information;

[0126] Reset gate (r): control the forgetting proportion of historical information;

[0127] Candidate state ( ): generate the current time candidate hidden state;

[0128] (4) Attention fusion layer: dynamically weighted fusion of spatial and temporal encoding features through attention mechanism, output (B, N, D) dimensional decision basis features, providing accurate input for multi-task output;

[0129] Spatial attention weight: weigh the importance of ST-GCN output;

[0130] Temporal attention weight: weigh the importance of GRU output;

[0131] (5) Output layer:

[0132] Matching decision: vehicle and passenger matching probability;

[0133] Path planning: optimal driving path prediction;

[0134] Charging decision: charging station selection and priority;

[0135] Comprehensive scheduling: integrate multi-dimensional decision results.

[0136] For example, the embodiment of the present application can first use the constructed joint model to organize the vehicle features (position, speed, power, etc.), passenger features (position, order demand, etc.), environmental features (road conditions, charging station status, etc.) and other features with time step T, node number N and feature dimension F to form an input tensor . Wherein T is the length of the time window (such as taking the last 10 minutes of data, the time step is 1 minute), N is the total number of nodes such as vehicles, passengers, charging stations in the region, and F is the number of features of each node.

[0137] Then, the input layer of the joint model can construct an adjacency matrix based on the input tensor, combined with the spatial relationship between the target unmanned taxi node, passenger node and charging station node in the spatial topology (such as the accessibility of the vehicle and the passenger, the distance between the vehicle and the charging station) , so as to quantify the connection strength between the nodes and provide topology structure information for the subsequent spatial coding generation.

[0138] Finally, the graph coding layer of the joint model can capture the spatial dependency relationship between the target unmanned taxi node, passenger node and charging station node based on the graph convolution network (GCN) in the spatial coding layer of the joint model.

[0139] The embodiment of the present application can construct a certain joint model to capture the spatial dependency relationship between the target unmanned taxi node, passenger node and charging station node, dynamically capture the spatial dependency and time sequence law, which is helpful for multi-task scheduling decision of the target unmanned taxi, solves the problem of spatio-temporal fragmentation of data, and realizes the demand, path and endurance collaborative optimization of the target unmanned taxi.

[0140] Optionally, in an embodiment of the present application, the spatial coding features and time coding features between the target unmanned taxi, passenger and charging station are generated in combination with the spatial dependency relationship and the time point of capturing the spatial dependency relationship, including: calculating the similarity of the node features between the target unmanned taxi node, passenger node and charging station node; extracting the spatial coding features according to the similarity of the node features.

[0141] In some embodiments, when the embodiment of the present application generates the spatial coding features of the target unmanned taxi, passenger and charging station according to the spatial dependency relationship and the time point of capturing the spatial dependency relationship, it is necessary to calculate the feature similarity between the target unmanned taxi node, passenger node and charging station node, so as to generate the spatial coding features according to the feature similarity.

[0142] Specifically, after the spatial coding layer captures the spatial dependency relationship between the nodes based on the graph convolution network (GCN), certain spatial features can be generated according to the captured spatial dependency relationship, wherein the calculation of the kth layer graph convolution can be but not limited to represented as follows:

[0143]

[0144] wherein, is a node at the i-th layer; spatial feature representation of the node is a neighbor node of the current node is a neighbor set of the node ; is an attention coefficient, is a learnable weight matrix; is an activation function, is a node at the i-th layer; spatial feature representation of the node .

[0145] wherein, the attention coefficient needs to be calculated by the feature similarity between the target unmanned taxi node, the passenger node and the charging station node (i.e. the similarity between the vehicle feature, the passenger feature and the environment feature), so as to dynamically adjust the influence weight of the neighbor node on the current node, the calculation formula can be but not limited to expressed as follows:

[0146]

[0147] wherein, is a linear transformation parameter of the attention score, is a feature transformation matrix (a learnable parameter), is a feature vector of the neighbor node u , is a feature vector of the current node , is a feature vector of the neighbor node, is any one neighbor node in the neighbor node set of the current node v, is a neighbor set of the node ; is a natural constant.

[0148] Space-time fusion preliminary processing: after multi-layer graph convolution, the spatial coding layer can output spatial coding features , wherein D is the dimension of the hidden layer, and the feature f combines the spatial dependency relationship between nodes and the initial feature in time sequence, that is, the captured spatial dependency relationship contains the time information when the spatial dependency relationship is captured. Further, the spatial coding feature can be used as the input of the time coding layer.

[0149] The embodiments of the present invention can capture the spatial dependency relationship between the target driverless taxi node, passenger node and charging station node to generate spatial features, and then combine the similarity between node features to generate spatial coding features, thereby effectively reflecting the actual spatial dependency relationship between the target driverless taxi, passenger and charging station through the spatial coding features.

[0150] Optionally, in one embodiment of the present invention, spatial coding features and temporal coding features between the target driverless taxi, passengers and charging stations are generated by combining spatial dependencies and the time points for capturing spatial dependencies, including: obtaining abnormal order information of passengers; and generating temporal coding features by combining abnormal order information and historical memory information corresponding to the time points.

[0151] In other embodiments, when generating the time-coded features of the target driverless taxi, passengers, and charging stations, the present invention needs to comprehensively consider the abnormal order information of passengers and the historical memory information corresponding to the time point when capturing spatial dependencies.

[0152] In this context, the historical memory information corresponding to the time point when capturing spatial dependencies can be understood as the event information (on a daily basis) that occurred at the same historical time point as the time point when capturing spatial dependencies, as remembered by the joint model. For example, the morning rush hour at 7-8 o'clock every morning causes road congestion.

[0153] Specifically, embodiments of the present invention can utilize the time coding layer in the joint model and employ a gated recurrent unit (GRU) to capture the temporal dynamic changes of the feature sequence to generate time-coded features.

[0154] The GRU computation process involves updating the gates. Obtain passenger's abnormal order information and reset the door. Controlling the retention and forgetting of historical memory information, and combining the two to generate time-encoded features, can be expressed, but is not limited to, as follows:

[0155]

[0156]

[0157]

[0158]

[0159] in, for Spatial coding layer output features at time step; This represents the hidden state of the GRU in the previous moment; To retain a certain percentage of historical information, For activation function, is a forgetting ratio of historical information, is a candidate hidden state of the GRU unit at the time point t, is a hidden state of the GRU unit at the time point t, is a forgetting ratio of historical information, is a hidden state of the GRU unit at the time point t, is an element-wise multiplication; , and are learnable weight matrices; , and are bias terms.

[0160] The embodiment of the present application can capture the time sequence law among the target unmanned taxi, the passenger and the charging station by combining the abnormal order information of the passenger and the historical memory information, for example, the eight o'clock morning rush hour congestion causes the vehicle to be stuck, and then causes the passenger to be late, or the passenger cancels the order because of waiting for a long time without a vehicle, and the like, which is helpful to plan the driving path of the target unmanned taxi and the matching result between the target unmanned taxi and the passenger.

[0161] In step S103, the weight of the spatial coding feature and the weight of the temporal coding feature are calculated to fuse the spatial coding feature and the temporal coding feature according to the weight of the spatial coding feature and the weight of the temporal coding feature, to generate the spatiotemporal feature of the target unmanned taxi, to generate the passenger matching result and / or the path planning and / or the charging decision of the target unmanned taxi based on the spatiotemporal feature.

[0162] As a possible implementation manner, the embodiment of the present application can calculate the weight of the spatial coding feature and the weight of the temporal coding feature respectively, to fuse the spatial coding feature and the temporal coding feature according to the weight of the spatial coding feature and the weight of the temporal coding feature, to generate the spatiotemporal feature of the target unmanned taxi, and finally generate the passenger matching result and / or the path planning and / or the charging decision of the target unmanned taxi based on the spatiotemporal feature (the passenger matching result, or the path planning, or the charging decision, or the passenger matching result and the path planning, or the path planning and the charging decision, or the passenger matching result and the path planning and the charging decision).

[0163] For example, the present application can utilize the attention mechanism of the spatiotemporal attention fusion layer of the joint model to dynamically adjust the weight of the spatial coding feature and the temporal coding feature, to realize the deep fusion of the spatiotemporal feature.

[0164] Firstly, the embodiment of the present application can calculate the spatial attention weight (the weight of the spatial coding feature) and the temporal attention weight (the weight of the temporal coding feature) , and the calculation formula can be but is not limited to represented as follows:

[0165]

[0166]

[0167] wherein, , are the final output states of the spatial and temporal encoding layers respectively; , is a learnable weight matrix; is a function for normalizing the weights.

[0168] Then, the spatial encoding feature and the temporal encoding feature are fused according to the spatial attention weight and the temporal attention weight to obtain the spatio-temporal feature of the target unmanned taxi, and the calculation formula can be but is not limited to expressed as:

[0169]

[0170] wherein, is the last time step feature of the spatial encoding layer; is the fused spatio-temporal feature.

[0171] The fused spatio-temporal feature contains both key spatial information (such as high demand area, charging station) and key temporal information (such as historical order peak, traffic flow trend), that is, the spatio-temporal feature not only retains the spatial dependency relationship but also contains integrated time series dynamic information, which can provide a basis for the final decision of the target unmanned taxi.

[0172] For example, the output layer of the joint model can generate the scheduling decision of the target unmanned taxi through a multilayer perceptron (MLP) based on the fused spatio-temporal feature, including but not limited to vehicle-passenger matching, path planning, charging decision, etc. Figure 3 is the model output decision flowchart of an embodiment of the present application, as shown in Figure 3 .

[0173] wherein, the matching of the vehicle and the passenger needs to calculate the matching probability of the vehicle and the passenger, and the calculation formula of the matching probability can be but is not limited to expressed as follows:

[0174]

[0175] wherein, is the matching probability of the vehicle and the passenger ; is a matching task weight matrix; is a bias term; The sigmoid activation function maps the output to the [0,1] interval, and determines the matching relationship based on a probability threshold (e.g., 0.7, meaning a match is possible if the value is greater than 0.7). These are the spatiotemporal characteristics after fusion.

[0176] Once the matching relationship is determined, the target drone-driven taxi is assigned a corresponding passenger and the passenger is notified.

[0177] Path planning: The output layer outputs a sequence of path point coordinates through an MLP. In conjunction with real-time road conditions and traffic rules, the system generates the optimal driving route from the target drone taxi's current location to the passenger's location (or charging station), and generates navigation instructions based on this route to send to the target drone taxi terminal for navigation.

[0178] Charging Decision: Output the charging priority and charging station selection for the target drone taxi. The expression can be, but is not limited to, the following:

[0179]

[0180] Among them, C( v ) for vehicles The charging decision distribution; Weight matrix for charging tasks; This is a bias term; the softmax function is used to normalize charging priorities, ensuring that the target drone-driven taxi is dispatched to the charging station according to priority, thus guaranteeing the target drone-driven taxi's range. These are the spatiotemporal characteristics after fusion.

[0181] The embodiments of the present invention can integrate multi-dimensional decision-making results and simultaneously output the vehicle-passenger matching, route planning, and charging scheduling results of the target drone-driven taxi. This can effectively balance the multiple business needs of the target drone-driven taxi, overcome the problem of insufficient adaptation of scheduling strategies in complex scenarios, and realize efficient, real-time, and intelligent scheduling of large-scale target drone-driven taxi fleets.

[0182] The intelligent scheduling method of the unmanned taxi according to the embodiment of the present application can capture the spatial dependency relationship among the target unmanned taxi, the passenger and the charging station, generate spatial coding features and time coding features and perform weighted fusion to generate spatiotemporal features, so as to generate the passenger matching result and / or the path planning and / or the charging decision of the target unmanned taxi based on the spatiotemporal features. In this way, the multi-dimensional data such as the vehicle features (position, speed, power, etc.), the passenger features (position, order demand, etc.), the environmental features (road conditions, charging station status, etc.) are combined, and the spatiotemporal features of the target unmanned taxi are effectively captured through the joint model. The spatial topological structure and the time dynamic law are deeply coded and fused, and the problem of one-sided scheduling decision caused by the spatiotemporal information fragmentation of the traditional model is solved. At the same time, the vehicle-passenger matching result, the path planning result and the charging scheduling result of the target unmanned taxi are synchronously output by using the model, which can effectively balance the multi-business demand and solve the problem of insufficient scheduling strategy adaptation in complex scenarios, and is helpful to realize the efficient, real-time and intelligent scheduling of large-scale unmanned taxi fleet in complex urban environment. In this way, the problems in the related art that the traditional unmanned taxi scheduling method does not fully consider the complex factors in the actual driving scenario, resulting in long response time for calling a car, non-optimized empty car mileage, uneven service level and other problems, and the scheduling decision is still single-target oriented, lacks dynamic adaptability and cannot meet the growing travel demand of passengers are solved.

[0183] Secondly, the intelligent scheduling device of the unmanned taxi according to the embodiment of the present application is described with reference to the accompanying drawings.

[0184] Figure 4 is a structural schematic diagram of the intelligent scheduling device of the unmanned taxi according to the embodiment of the present application.

[0185] As shown in Figure 4 , the intelligent scheduling device 10 of the unmanned taxi includes a construction module 100, a generation module 200 and a fusion module 300.

[0186] The construction module 100 is configured to construct a spatial topological structure for connecting the target unmanned taxi, the passenger and the charging station based on the received vehicle data of the target unmanned taxi, the passenger data and the environmental data of the target unmanned taxi driving area.

[0187] The generation module 200 is configured to capture the spatial dependency relationship among the target unmanned taxi node, the passenger node and the charging station node in the spatial topological structure based on the spatial topological structure, so as to generate the spatial coding features and the time coding features among the target unmanned taxi, the passenger and the charging station in combination with the spatial dependency relationship and the time point of capturing the spatial dependency relationship.

[0188] The fusion module 300 is configured to calculate the weight of the spatial encoding feature and the weight of the time encoding feature, fuse the spatial encoding feature and the time encoding feature according to the weight of the spatial encoding feature and the weight of the time encoding feature, and generate the spatio-temporal feature of the target unmanned taxi, so as to generate the passenger matching result and / or the path planning and / or the charging decision of the target unmanned taxi based on the spatio-temporal feature.

[0189] Optionally, in an embodiment of the present application, further comprising:

[0190] The spatial topology structure includes edges connecting the target unmanned taxi nodes and the passenger nodes and edges connecting the target unmanned taxi nodes and the charging station nodes.

[0191] The target unmanned taxi node includes at least one of the vehicle coordinates, the vehicle speed, the remaining power and the passenger carrying state of the target unmanned taxi.

[0192] The passenger node includes at least one of the starting point coordinates, the ending point coordinates, the order creation time, the waiting time, and the ride demand of the passenger.

[0193] The charging station node includes at least one of the coordinates, the available pile number and the queuing time of the charging station.

[0194] Optionally, in an embodiment of the present application, the generation module 200 includes a calculation unit and a generation unit.

[0195] The calculation unit is configured to calculate the similarity of the node features between the target unmanned taxi nodes, the passenger nodes and the charging station nodes.

[0196] The generation unit is configured to extract the spatial encoding feature according to the similarity of the node features.

[0197] Optionally, in an embodiment of the present application, the generation module 200 includes an acquisition unit and a combination unit.

[0198] The acquisition unit is configured to acquire the abnormal order information of the passenger.

[0199] The combination unit is configured to combine the abnormal order information and the historical memory information corresponding to the time point to generate the time encoding feature.

[0200] Optionally, in an embodiment of the present application, further comprising an acquisition module and a capture module.

[0201] The obtaining module is configured to obtain a task target of the target unmanned taxi before capturing a spatial dependency relationship among the target unmanned taxi node, the passenger node and the charging station node in the spatial topology structure, and determine a matching loss, a path loss, a charging loss, a regularization loss and corresponding weights for training a preset joint model according to the task target.

[0202] The capturing module is configured to construct the preset joint model by using the matching loss, the path loss, the charging loss, the regularization loss and the corresponding weights, and capture the spatial dependency relationship by using the preset joint model.

[0203] It should be noted that the foregoing explanation of the embodiment of the intelligent scheduling method of the unmanned taxi is also applicable to the intelligent scheduling device 10 of the unmanned taxi of the embodiment, and will not be described here again.

[0204] The intelligent scheduling device 10 of the unmanned taxi according to the embodiment of the present application can capture the spatial dependency relationship among the target unmanned taxi, the passenger and the charging station, generate spatial coding features and temporal coding features, and perform weighted fusion to generate spatiotemporal features, so as to generate a passenger matching result and / or a path planning and / or a charging decision of the target unmanned taxi based on the spatiotemporal features. Thus, the multi-dimensional data such as vehicle features (position, speed, power, etc.), passenger features (position, order demand, etc.), and environment features (road conditions, charging station status, etc.) are combined, and the spatiotemporal features of the target unmanned taxi are effectively captured through the joint model. The spatial topology structure and the temporal dynamic law are deeply coded and fused, and the problem of one-sided scheduling decision caused by the split of spatiotemporal information in the traditional model is solved. At the same time, the vehicle-passenger matching result, the path planning result and the charging scheduling result of the target unmanned taxi are simultaneously output by using the model, which can effectively balance the multi-service demand and solve the problem of insufficient scheduling strategy adaptation in complex scenarios, and is helpful to realize efficient, real-time and intelligent scheduling of large-scale unmanned taxi fleet in complex urban environment. Thus, the problems in the related art that the traditional unmanned taxi scheduling method does not fully consider the complex factors in the actual driving scenario, resulting in long response time for calling a car, non-optimized empty car mileage, uneven service level, and the like, and the scheduling decision is still single-target oriented, lacks dynamic adaptability, and cannot meet the growing travel demand of passengers, and the like are solved.

[0205] Figure 5 The electronic device provided by the embodiment of the present application is shown in the structural schematic diagram of the electronic device. The electronic device can include:

[0206] The memory 501, the processor 502, and the computer program stored in the memory 501 and executable on the processor 502.

[0207] The processor 502 implements the intelligent dispatching method of the unmanned taxi provided in the above embodiments when executing a program.

[0208] Further, the electronic device further comprises:

[0209] The communication interface 503 is configured to communicate between the memory 501 and the processor 502.

[0210] The memory 501 is configured to store a computer program executable on the processor 502.

[0211] The memory 501 can include a high-speed RAM memory, and can further include a non-volatile memory, for example, at least one disk memory.

[0212] If the memory 501, the processor 502 and the communication interface 503 are implemented independently, the communication interface 503, the memory 501 and the processor 502 can be connected to each other through a bus and complete communication between each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, Figure 5 Only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0213] Optionally, in a specific implementation, if the memory 501, the processor 502 and the communication interface 503 are integrated on a chip, the memory 501, the processor 502 and the communication interface 503 can complete communication between each other through an internal interface.

[0214] The processor 502 can be a Central Processing Unit (CPU), or an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0215] The embodiments of the present application further provide a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the intelligent dispatching method of the unmanned taxi as above.

[0216] The embodiment of the present application further provides a computer program product comprising a computer program which can run computer instructions, and the computer instructions are executed by a processor to realize the intelligent scheduling method of unmanned taxi provided by the embodiment of the present application.

[0217] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples without contradiction.

[0218] In addition, the terms "first", "second" are only used for description purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "N" is at least two, for example, two, three, etc., unless otherwise specifically limited.

[0219] Any process or method descriptions in flow charts or otherwise described herein represent embodiments which can be understood as a module, segment, or portion of code which comprises one or N executable instructions for implementing the specified logic function or process. The scope of the preferred embodiments of the present application includes additional implementation in which the functions described in the illustrated or discussed order are performed in a different order, including substantially simultaneously, or in reverse order, according to the function involved, which should be understood by those skilled in the art to which the embodiments of the present application belong.

[0220] The logic and / or steps represented in flow diagrams or otherwise described herein, for example, can be considered as a sequence of executable instructions, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For purposes of this specification, a "computer-readable medium" can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be a product of the manufacturing and / or processing. The computer-readable medium can include, but is not limited to, the following: an electronic connection (an electronic device with one or N wires), a portable computer diskette (a magnetic device), a RAM (random access memory), a ROM (read-only memory), an EPROM (erasable programmable ROM) or a Flash memory, an optical fiber, and a portable CD ROM. In addition, the computer-readable medium can even be paper or other suitable medium upon which the program can be printed, as the program can be electronically captured, via the optically scanning of the paper or other medium, then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and stored in the computer memory.

[0221] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, the N steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented in hardware and in another embodiment, the hardware can be implemented using any or a combination of the following technologies, which are all well known in the art: a discrete logic circuit having logic gates for implementing logic functions upon data signals, an application specific integrated circuit having appropriate combinational logic gates, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0222] Those of skill in the art would understand that the steps carried out in the above-mentioned embodiment methods can be carried out wholly or partly by programs instructing relevant hardware, and the programs can be stored in a computer-readable storage medium. When the programs are executed, they include one or a combination of the steps of the method embodiments.

[0223] In addition, each function unit in each embodiment of the present application can be integrated in one processing module, or each unit can be physically present separately, or two or more units can be integrated in one module. The integrated module can be realized in the form of hardware or in the form of a software function module. When the integrated module is realized in the form of a software function module and sold or used as an independent product, it can also be stored in a computer readable storage medium.

[0224] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A method for intelligent dispatching of driverless taxis, characterized in that, The method comprises the following steps: Based on the received vehicle data, passenger data and environmental data of the target unmanned taxi, the space topology structure for connecting the target unmanned taxi, passenger and charging station is constructed; Based on the space topology structure, the spatial dependency relationship between the target unmanned taxi node, passenger node and charging station node in the space topology structure is captured to generate the space coding feature and time coding feature between the target unmanned taxi, the passenger and the charging station in combination with the spatial dependency relationship and the time point of capturing the spatial dependency relationship; The weight of the space coding feature and the weight of the time coding feature are calculated to fuse the space coding feature and the time coding feature according to the weight of the space coding feature and the weight of the time coding feature, generate the space-time feature of the target unmanned taxi, and generate the passenger matching result and / or path planning and / or charging decision of the target unmanned taxi based on the space-time feature.

2. The intelligent dispatching method of driverless taxis according to claim 1, characterized in that, Further comprising: The space topology structure comprises edges connecting the target unmanned taxi node and the passenger node and edges connecting the target unmanned taxi node and the charging station node; The target unmanned taxi node comprises at least one of the vehicle coordinates, vehicle speed, remaining power and passenger carrying state of the target unmanned taxi; The passenger node comprises at least one of the starting point coordinates, end point coordinates, order creation time, waiting time, ride demand of the passenger; The charging station node comprises at least one of the coordinates, available pile number and queuing time of the charging station.

3. The intelligent dispatching method of driverless taxis according to claim 1, wherein, The combination of the spatial dependency relationship and the time point of capturing the spatial dependency relationship to generate the space coding feature and the time coding feature between the target unmanned taxi, the passenger and the charging station comprises: Calculate the similarity of node features between the target unmanned taxi node, the passenger node and the charging station node; According to the similarity of the node features, the space coding feature is extracted.

4. The intelligent dispatching method of driverless taxis according to claim 1, wherein, The combination of the spatial dependency relationship and the time point of capturing the spatial dependency relationship to generate the space coding feature and the time coding feature between the target unmanned taxi, the passenger and the charging station comprises: Obtain the abnormal order information of the passenger; Combine the abnormal order information and the historical memory information corresponding to the time point to generate the time coding feature.

5. The intelligent dispatching method of driverless taxis according to claim 1, wherein, Before capturing the spatial dependency relationship between the target unmanned taxi node, the passenger node and the charging station node in the space topology structure, further comprising: Obtain the task target of the target unmanned taxi, and determine the matching loss, path loss, charging loss, regularization loss and their corresponding weights for training the preset joint model according to the task target; Construct the preset joint model by using the matching loss, the path loss, the charging loss, the regularization loss and their corresponding weights to capture the spatial dependency relationship by using the preset joint model.

6. An intelligent dispatching device for unmanned taxicabs, characterized in that, The method comprises the following steps: a construction module is used to construct a spatial topology for connecting the target unmanned taxi, a passenger and a charging station based on the received vehicle data of the target unmanned taxi, passenger data and environmental data of a driving area allowed by the target unmanned taxi; a generation module is used to capture spatial dependency between target unmanned taxi nodes, passenger nodes and charging station nodes in the spatial topology based on the spatial topology, to generate spatial coding features and time coding features between the target unmanned taxi, the passenger and the charging station in combination with the spatial dependency and a time point at which the spatial dependency is captured; a fusion module is used to calculate weights of the spatial coding features and weights of the time coding features, to generate spatiotemporal features of the target unmanned taxi by fusing the spatial coding features and the time coding features according to the weights of the spatial coding features and the weights of the time coding features, and to generate a passenger matching result and / or path planning and / or charging decision of the target unmanned taxi based on the spatiotemporal features.

7. The intelligent dispatching apparatus for unmanned taxicabs according to claim 6, wherein, Further comprising: the spatial topology comprises edges connecting the target unmanned taxi nodes and the passenger nodes and edges connecting the target unmanned taxi nodes and the charging station nodes; the target unmanned taxi nodes comprise at least one of vehicle coordinates, vehicle speed, remaining power and passenger carrying status of the target unmanned taxi; the passenger nodes comprise at least one of starting point coordinates, ending point coordinates, order creation time, waiting time and ride demand of the passenger; the charging station nodes comprise at least one of coordinates, available pile number and queue time of the charging station.

8. An electronic device, comprising: The method comprises the following steps: a memory, a processor and a computer program stored in the memory and executable on the processor, the processor executes the program to implement the intelligent scheduling method of the unmanned taxi according to any one of claims 1-5.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the intelligent scheduling method of the unmanned taxi according to any one of claims 1-5.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed to implement the intelligent scheduling method of the unmanned taxi according to any one of claims 1-5.

Images