A multi-modal based unmanned vehicle dispatching method and system

Abstract

The application provides a multi-modal intermodal unmanned carrier scheduling method and system, which comprises the following steps: creating a multi-source information collection network and transmitting historical transportation data collected by the multi-source information collection network in real time to a preset edge computing node; standardizing the historical transportation data by the preset edge computing node to generate corresponding standard transportation data in real time, and obtaining a plurality of vehicle models corresponding to the carrier in real time; training a corresponding deep learning scheduling model according to the standard transportation data and a preset neural network in real time, and extracting the position change characteristics and the operation law characteristics corresponding to each vehicle model by the deep learning scheduling model in real time; generating a scheduling priority sequence corresponding to the carrier according to the position change characteristics and the operation law characteristics in real time, and sending a corresponding scheduling instruction to each carrier according to the scheduling priority sequence. The application can improve the use efficiency of the carrier and the transportation efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of logistics transportation, and in particular relates to a method and system for scheduling unmanned transport vehicles based on multimodal transport. BACKGROUND

[0002] With the progress of science and technology and the rapid development of the times, people have made significant progress in the field of logistics technology, and have developed a multimodal transport transfer mode for unmanned transport vehicles, that is, the corresponding transportation is completed by means of public transit or transit.

[0003] Among them, under the background of public transit and transit transport structure adjustment, the logistics transportation pressure increases significantly. Specifically, public transit relies on short-distance transport vehicles to solve railway freight distribution, and transit needs transport vehicles to complete end distribution, both of which increase the intensity of transport vehicle operation. At the same time, logistics enterprises are faced with the dual constraints of limited cost and high efficiency, and the efficiency of transport vehicle operation has become the key to improving competitiveness.

[0004] Further, in the process of actual application, most of the prior art completes the scheduling of unmanned transport vehicles by artificial means. However, this scheduling method completely relies on the experience of artificial means, and thus cannot integrate the dynamic information such as the working state, position and operation type of the transport vehicle in real time and accurately, resulting in problems such as order dispatch delay and invalid order dispatch, and the optimal transport vehicle cannot be selected, which reduces the transportation efficiency. SUMMARY

[0005] Therefore, the purpose of the present application is to provide a method and system for scheduling unmanned transport vehicles based on multimodal transport, so as to solve the problem that the prior art cannot integrate the dynamic information such as the working state, position and operation type of the transport vehicle in real time and accurately, resulting in problems such as order dispatch delay and invalid order dispatch.

[0006] The first aspect of the embodiment of the present application provides:

[0007] A method for scheduling unmanned transport vehicles based on multimodal transport, wherein the method comprises:

[0008] creating a corresponding multi-source information collection network in real time according to a plurality of preset transfer nodes, and transmitting historical transportation data collected by the multi-source information collection network in real time to a preset edge computing node;

[0009] standardizing the historical transportation data by the preset edge computing node to generate corresponding standard transportation data in real time, and obtaining a plurality of vehicle models corresponding to the transport vehicles in real time;

[0010] According to the standard transportation data and a preset neural network, a corresponding deep learning scheduling model is trained in real time, and position change characteristics and operation rule characteristics corresponding to each vehicle model are extracted in real time through the deep learning scheduling model;

[0011] According to the position change characteristics and the operation rule characteristics, a scheduling priority order corresponding to the carrying vehicle is generated in real time, and a corresponding scheduling instruction is sent to each carrying vehicle according to the scheduling priority order.

[0012] The beneficial effects of the present application are: through the multi-source information collection network created in real time, the historical transportation data corresponding to various carrying vehicles in the actual working process can be collected, based on this, in order to facilitate subsequent analysis, at this time, the corresponding standard transportation data is converted, and the position change characteristics and operation rule characteristics corresponding to each type of carrying vehicle are extracted in real time through the trained deep learning scheduling model, based on this, the final scheduling priority order is generated in real time according to the current two characteristics, and the corresponding scheduling instruction is sent according to the order, so as to avoid the problems of order dispatching delay and invalid order dispatching, and improve the use efficiency of the carrying vehicle, eliminate the influence of human factors, and improve the transportation efficiency.

[0013] Further, the step of standardizing the historical transportation data through the preset edge computing node to generate corresponding standard transportation data in real time comprises:

[0014] According to a plurality of transfer nodes, the historical transportation data is divided into a plurality of data subsets, wherein each data subset corresponds to a unique road segment identifier;

[0015] According to the road segment identifier, the text type transportation state description in each data subset is subjected to word segmentation processing, and the word segmentation result is mapped to a preset standardized state code library;

[0016] The deviation coefficient of data collected by the same transfer node in the same time period is calculated, and the data in the preset standardized state code library that exceeds the threshold range is gradient corrected according to the deviation coefficient, so as to correspondingly generate the standard transportation data.

[0017] Further, the step of calculating the deviation coefficient of data collected by the same transfer node in the same time period, and gradient correcting the data in the preset standardized state code library that exceeds the threshold range according to the deviation coefficient to correspondingly generate the standard transportation data comprises:

[0018] The transport data set generated by the same transport node in several historical same time periods is collected, and a corresponding normal distribution model is constructed in real time according to the mean and variance of the transport data set;

[0019] The collection data generated by the transport node at present is input into the normal distribution model, so as to match the corresponding deviation coefficient in real time;

[0020] A gradient correction function is constructed in real time according to the difference between the deviation coefficient and a preset coefficient threshold, and the preset standardization state code library is gradient corrected through the gradient correction function, so as to generate the standard transport data.

[0021] Further, the step of extracting the position change feature and the operation rule feature corresponding to each vehicle model in real time through the deep learning scheduling model comprises:

[0022] The position record field associated with the vehicle model in the standard transport data is time stamped and aligned, and a plurality of data segments are divided according to a preset time window, and the latitude and longitude coordinate fluctuation range of the vehicle in each data segment is counted in real time to form a corresponding original position data sequence;

[0023] The loading and unloading record, driving time record and stop node record associated with the vehicle model in the standard transport data are classified and summarized, and the operation frequency and operation time distribution of the same vehicle model in different transport nodes are counted to form a corresponding original operation data set;

[0024] The original position data sequence and the original operation data set are analyzed and processed to correspondingly generate the position change feature and the operation rule feature.

[0025] Further, the step of analyzing and processing the original position data sequence and the original operation data set to correspondingly generate the position change feature and the operation rule feature comprises:

[0026] The original position data sequence is input into the convolution layer of the deep learning scheduling model, the original position data sequence is filtered by a 3*3 size convolution kernel, and the position offset change trend in adjacent time windows is extracted to generate a primary position change feature vector;

[0027] The original operation data set is input into the recurrent neural network layer of the deep learning scheduling model, the operation data in continuous periods is analyzed by a gated recurrent unit, and the operation interval rule and operation time fluctuation rule of the vehicle model in different transport nodes are extracted to generate a primary operation rule feature vector;

[0028] The primary position change feature vector and the primary operation rule feature vector are nonlinearly mapped by an activation function, and feature dimensions with a weight higher than a preset weight threshold are screened out to form the position change feature and the operation rule feature.

[0029] Further, the step of generating a scheduling priority order corresponding to the carrying vehicle in real time according to the position change feature and the operation rule feature comprises:

[0030] The real-time relative azimuth angle and the road network topological distance between the carrying vehicle and each transfer node are extracted from the position change feature, and the Floyd-Warshall algorithm is used to calculate the shortest transfer path, and the shortest transfer path is converted into a distance score value;

[0031] Task execution data corresponding to the same vehicle model generated within a preset time are extracted from the operation rule feature, the task execution data including node stay duration standard deviation, cross-mode transfer success rate, and daily operation frequency, and the task execution data is reduced to a single operation ability factor by principal component analysis;

[0032] The distance score value and the operation ability factor are weighted and summed according to a weight of 4:6 to obtain an initial priority score value, and after being arranged in descending order of score value, the sorting result is locally optimized by a preset optimization algorithm to generate the scheduling priority order.

[0033] Further, the step of generating the scheduling priority order by locally optimizing the sorting result by the preset optimization algorithm comprises:

[0034] A preset particle swarm optimization algorithm is called out, and the initial descending order arrangement is taken as the initial position of particles;

[0035] A particle fitness function adapted to each vehicle model is created in real time, and the speed and position of the particles are iteratively updated by the particle fitness function;

[0036] The particle position corresponding to the global optimal fitness is retained, and after iteration to a maximum number of times, the global optimal particle position is taken as the final scheduling priority order.

[0037] The second aspect of the embodiment of the application provides:

[0038] An unmanned carrying vehicle scheduling system based on multimodal transport, wherein the system comprises:

[0039] A creating module is configured to create a corresponding multi-source information collection network in real time according to a preset number of transfer nodes, and transmit historical transportation data collected by the multi-source information collection network in real time to a preset edge computing node;

[0040] A processing module is configured to standardize the historical transportation data by the preset edge computing node to generate corresponding standard transportation data in real time, and obtain a number of vehicle models corresponding to the carrying vehicle in real time;

[0041] An extracting module is configured to train a corresponding deep learning scheduling model in real time according to the standard transportation data and a preset neural network, and extract position change characteristics and work rule characteristics corresponding to each vehicle model in real time through the deep learning scheduling model;

[0042] An ordering module is configured to generate a scheduling priority order corresponding to the carrying vehicle in real time according to the position change characteristics and the work rule characteristics, and send a corresponding scheduling instruction to each carrying vehicle according to the scheduling priority order.

[0043] Further, the processing module is specifically configured to:

[0044] divide the historical transportation data into a number of data subsets according to a number of transfer nodes, wherein each data subset corresponds to a unique road segment identifier;

[0045] perform word segmentation processing on text type transportation state descriptions in each data subset according to the road segment identifier, and map the word segmentation results to a preset standardized state code library;

[0046] calculate a deviation coefficient of data collected by the same transfer node in the same time period, and perform gradient correction on data in the preset standardized state code library that exceeds a threshold range according to the deviation coefficient to generate the standard transportation data correspondingly.

[0047] Further, the processing module is specifically configured to:

[0048] combine transportation data sets generated by the same transfer node in a number of historical same time periods, and construct a corresponding normal distribution model in real time according to a mean and a variance of the transportation data sets;

[0049] input collected data generated by the transfer node at present into the normal distribution model to match a corresponding deviation coefficient in real time;

[0050] construct a corresponding gradient correction function in real time according to a difference between the deviation coefficient and a preset coefficient threshold, and perform gradient correction on the preset standardized state code library through the gradient correction function to generate the standard transportation data.

[0051] Further, the extraction module is specifically configured to:

[0052] timestamp-align the position record field associated with the vehicle model in the standard transportation data, divide a plurality of data segments according to a preset time window, and statistically calculate the latitude and longitude coordinate fluctuation range of the carrying vehicle in each data segment to form a corresponding original position data sequence;

[0053] classify and summarize the loading and unloading record, driving time record and stop node record associated with the vehicle model in the standard transportation data, and statistically calculate the operation frequency and operation time distribution of the same vehicle model at different transfer nodes to form a corresponding original operation data set;

[0054] analyze and process the original position data sequence and the original operation data set to correspondingly generate the position change feature and the operation regularity feature.

[0055] Further, the extraction module is specifically configured to:

[0056] input the original position data sequence into the convolution layer of the deep learning scheduling model, slide filter the original position data sequence through a 3x3 size convolution kernel, and extract the position offset change trend in adjacent time windows to generate a primary position change feature vector;

[0057] input the original operation data set into the recurrent neural network layer of the deep learning scheduling model, perform time series correlation analysis on the operation data in the continuous period through the gated recurrent unit, and extract the operation interval regularity and operation time fluctuation regularity of the vehicle model at different transfer nodes to generate a primary operation regularity feature vector;

[0058] nonlinearly map the primary position change feature vector and the primary operation regularity feature vector through an activation function, and filter out feature dimensions with an influence weight on scheduling decision higher than a preset weight threshold to correspondingly form the position change feature and the operation regularity feature.

[0059] Further, the sorting module is specifically configured to:

[0060] extract the real-time relative azimuth angle and road network topological distance between the carrying vehicle and each transfer node from the position change feature, calculate the shortest transfer path by using the Floyd-Warshall algorithm, and convert the shortest transfer path into a distance score value;

[0061] Extract task execution data corresponding to the same vehicle model generated within a preset time from the job law characteristics, the task execution data including node stay duration standard deviation, cross-mode transfer success rate, and daily operation frequency, and reduce the task execution data to a single job capacity factor by principal component analysis;

[0062] Sum the distance score value and the job capacity factor by weighting 4:6 to obtain an initial priority score, arrange in descending order of score, and perform local optimization on the sorted result by a preset optimization algorithm to correspondingly generate the scheduling priority order.

[0063] Further, the sorting module is specifically used for:

[0064] Call a preset particle swarm optimization algorithm, and take the initial descending order as the initial position of the particle;

[0065] Real-time create a particle fitness function adapted to each vehicle model, and update the speed and position of the particle through the particle fitness function;

[0066] Keep the particle position corresponding to the global optimal fitness, and after iteration to the maximum number of times, take the global optimal particle position as the final scheduling priority order.

[0067] The third aspect of the embodiment of the present application provides:

[0068] A computer comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the multi-modal transport-based unmanned carrier vehicle scheduling method as described above when executing the computer program.

[0069] The fourth aspect of the embodiment of the present application provides:

[0070] A readable storage medium having a computer program stored thereon, wherein the program is executed by a processor to implement the multi-modal transport-based unmanned carrier vehicle scheduling method as described above.

[0071] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0072] Figure 1 A flowchart of the multi-modal transport-based unmanned carrier vehicle scheduling method provided by the first embodiment of the present application;

[0073] Figure 2 A structural block diagram of the multi-modal transport-based unmanned carrier vehicle scheduling system provided by the third embodiment of the present application.

[0074] The following detailed description will further explain the present application with reference to the above mentioned drawings. DETAILED DESCRIPTION

[0075] For the purpose of promoting the understanding and facilitating appreciation of the application, the application will be described in further detail below with reference to the attached drawings. Several embodiments of the application are shown in the drawings. It should be noted, however, that the application can be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and fully convey the scope of the application to those skilled in the art.

[0076] It should be noted that when an element is referred to as being "on" another element, it can be directly on the other element or intervening elements can also be present. When an element is referred to as being "connected" to another element, it can be directly connected to the other element or intervening elements can also be present. As used herein the terms "vertical", "horizontal", "left", "right" and the like are merely used for the purpose of illustration and are not intended to be limiting.

[0077] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. The terms "comprises", "comprising", "including" and "having" are intended to be inclusive and mean that there can be additional elements other than the listed elements.

[0078] Referring to Figure 1 , a multi-modal transport-based unmanned carrier vehicle dispatching method provided by a first embodiment of the present application is shown. The multi-modal transport-based unmanned carrier vehicle dispatching method provided by the embodiment can accurately and effectively complete real-time dispatching of each carrier vehicle, improve the use efficiency of the carrier vehicle, and correspondingly improve the transportation efficiency.

[0079] Specifically, the embodiment provides:

[0080] A multi-modal transport-based unmanned carrier vehicle dispatching method, specifically comprising the following steps:

[0081] Step S10, a corresponding multi-source information collection network is created in real time according to a plurality of preset transfer nodes, and historical transportation data collected by the multi-source information collection network in real time is transmitted to a preset edge computing node in real time.

[0082] It should be noted that in order to accurately schedule each carrier, the required transportation data needs to be accurately obtained, and based on this, subsequent analysis is carried out. Specifically, the application first constructs a multi-source information collection network covering all nodes according to a plurality of preset transfer nodes (such as ports, railway stations, logistics parks and other key nodes of intermodal transportation), which integrates monitoring devices, sensors, RFID data collection terminals and other data collection terminals in the nodes; then the historical transportation data (including carrier driving trajectory, loading and unloading time, node stay duration, etc.) collected by the network in real time is transmitted to the preset edge computing node (deployed on the edge server close to the data source to reduce data transmission delay). It should be noted that the multi-source information collection network ensures the comprehensiveness and real-time of the transportation data, and the deployment of the edge computing node provides low-delay and high-reliability hardware support for subsequent data processing, laying the foundation for the timeliness of scheduling decisions, so as to facilitate subsequent processing.

[0083] In step S20, the historical transportation data is standardized by the preset edge computing node to generate corresponding standard transportation data in real time, and the corresponding vehicle models of the carrier are obtained in real time;

[0084] It should be noted that the edge computing node standardizes the received historical transportation data (including data cleaning, format unification, and abnormal value correction), and converts heterogeneous data (such as different format data of different nodes) into standard transportation data with consistent structure; at the same time, the vehicle management system is used to obtain the vehicle models (such as container trucks, cold chain transport vehicles, etc.) corresponding to all participating carriers in real time. Through the above-mentioned manner, the data format difference can be standardized and eliminated, and the input consistency of subsequent model training is ensured, so as to facilitate subsequent processing.

[0085] In step S30, a deep learning scheduling model corresponding to each vehicle model is trained in real time according to the standard transportation data and a preset neural network, and the position change feature and the operation rule feature corresponding to each vehicle model are extracted in real time by the deep learning scheduling model.

[0086] It should be noted that the standard transportation data is used as a training sample, and a deep learning scheduling model is trained by using a preset neural network (such as a hybrid network combining CNN and RNN). After the model training is completed, the feature extraction module of the model is used to extract the position change feature (such as driving route preference and speed change rule) and the operation rule feature (such as loading and unloading efficiency and node stay habit) corresponding to each vehicle model from the historical data in real time. It should be noted that the original data can be converted into structured features that can be used for decision-making through feature extraction, which provides a basis for subsequent priority sorting, so as to facilitate subsequent processing.

[0087] Step S40, generating a scheduling priority corresponding to the carrier according to the position change feature and the operation regularity feature in real time, and sending a corresponding scheduling instruction to each carrier according to the scheduling priority.

[0088] It should be noted that, based on the extracted position change feature (reflecting vehicle accessibility) and operation regularity feature (reflecting vehicle operation ability), the scheduling priority of each carrier is comprehensively evaluated (vehicles with high priority are preferentially assigned tasks); finally, according to the priority ranking, the scheduling instruction (such as going to a node to load goods, transfer path planning, etc.) is sent to the corresponding carrier. Based on this, the scheduling of various types of carriers can be accurately and effectively completed, and the use efficiency of the carrier is improved, which corresponds to improving the transportation efficiency.

[0089] Second embodiment

[0090] Further, the step of standardizing the historical transportation data by the preset edge computing node to generate corresponding standard transportation data in real time comprises:

[0091] According to a plurality of transfer nodes, the historical transportation data is divided into a plurality of data subsets, wherein each data subset corresponds to a unique road segment identifier;

[0092] According to the road segment identifier, the text type transportation state description in each data subset is processed by word segmentation, and the word segmentation result is mapped to a preset standardized state code library;

[0093] Calculate the deviation coefficient of the same transfer node in the same time period, and according to the deviation coefficient, the data in the preset standardized state code library that exceeds the threshold range is gradient corrected to correspondingly generate the standard transportation data.

[0094] It should be noted that the historical transportation data is divided into several data subsets according to the geographical boundaries and functional attributes of the transfer nodes, each subset corresponds to a unique road segment identifier (such as "Port A - Railway Station B" road segment, "Railway Station B - Logistics Park C" road segment), and the data is associated with the specific transportation road segment. Based on this, for the text type transportation state description (such as congestion, normal, loading and unloading, etc. Unstructured text) in each data subset, a word segmentation algorithm (such as jieba word segmentation) is used for word segmentation processing to extract key information; then the word segmentation result is mapped to a preset standardized state code library (such as "congestion" corresponds to code 001, "normal" corresponds to code 002), realizing the digital conversion of text information. Based on this, the deviation coefficient (reflecting the degree of data fluctuation) of the same transfer node in the same time period (such as 8:00-10:00 every day) is calculated; if the data in the standardized state code library exceeds the preset threshold range (such as deviating from the historical mean by 3 times the standard deviation), it is gradient corrected according to the deviation coefficient (such as gradually adjusting to a reasonable range according to the deviation ratio), and finally the standard transportation data is generated. Through standardization processing of the data, the subsequent data processing time can be correspondingly shortened, and the data processing amount can be reduced to facilitate subsequent processing.

[0095] Further, the step of calculating the deviation coefficient of the same transfer node in the same time period and gradient correcting the data in the preset standardized state code library that exceeds the threshold range according to the deviation coefficient to correspondingly generate the standard transportation data includes:

[0096] Collecting transportation data generated by the same transfer node in several historical same time periods, and constructing a corresponding normal distribution model in real time according to the mean and variance of the transportation data set;

[0097] Inputting the collected data generated by the transfer node into the normal distribution model to match the corresponding deviation coefficient in real time;

[0098] Constructing a gradient correction function in real time according to the difference between the deviation coefficient and the preset coefficient threshold, and gradient correcting the preset standardized state code library through the gradient correction function to generate the standard transportation data.

[0099] It should be noted that the transport data generated by the same transfer node in several historical same time periods (such as 9:00-11:00 every day in the past 30 days) is collected to form a transport data set; the mean and variance of the data are calculated based on the set to construct a corresponding normal distribution model (reflecting the normal data distribution law of the node in the time period). Based on this, the data collected by the transfer node is input into the constructed normal distribution model, and the deviation of the data from the mean is calculated to match the corresponding deviation coefficient (such as Z-score value, reflecting the position of the data in the normal distribution). On this basis, according to the difference between the deviation coefficient and the preset coefficient threshold (such as Z-score=3), a gradient correction function (such as a linear correction function, the larger the deviation, the larger the correction amplitude) is constructed; the data in the standardized state code library that exceeds the threshold is gradually corrected through the function (to avoid data distortion caused by one-time correction), and finally the standard transport data is generated. Through the above-mentioned manner, the data obtained in real time can be adjusted to a reasonable range, so as to facilitate subsequent processing.

[0100] Further, the step of extracting the position change feature and the operation rule feature corresponding to each vehicle model through the deep learning scheduling model in real time comprises:

[0101] Aligning the timestamp of the position record field associated with the vehicle model in the standard transport data, and dividing a plurality of data segments according to a preset time window, and real-time statistics of the latitude and longitude coordinate fluctuation range of the vehicle in each data segment to form a corresponding original position data sequence;

[0102] Classifying and summarizing the loading and unloading record, driving time record and stop node record associated with the vehicle model in the standard transport data, and counting the operation frequency and operation time distribution of the same vehicle model in different transfer nodes to form a corresponding original operation data set;

[0103] Analyzing and processing the original position data sequence and the original operation data set to correspondingly generate the position change feature and the operation rule feature.

[0104] It should be noted that the position record field (such as GPS longitude and latitude) associated with the vehicle model in the standard transportation data is time-stamped (ensuring synchronization of position data at different time points); the data is divided into several data segments according to a preset time window (such as 10 minutes / window), and the longitude and latitude coordinate fluctuation range (such as maximum longitude and minimum latitude) of the carrying vehicle in each segment is counted to form the original position data sequence (reflecting the spatial movement trajectory of the vehicle). At the same time, the operation-related data such as loading and unloading records, driving time records, and stop node records associated with the vehicle model are selected from the standard transportation data, and are classified and summarized according to the transfer node and time period; the operation frequency (such as the number of loading and unloading times at port A per day) and operation time length distribution (such as the mean and standard deviation of loading and unloading time) of the same vehicle model at different transfer nodes are counted to form the original operation data set. Based on this, only the current original position data sequence and the original operation data set need to be analyzed and processed in real time to obtain the required position change characteristics and operation rule characteristics for subsequent processing.

[0105] Further, the step of analyzing and processing the original position data sequence and the original operation data set to correspondingly generate the position change characteristics and the operation rule characteristics comprises:

[0106] inputting the original position data sequence into the convolution layer of the deep learning scheduling model, sliding filtering the original position data sequence through a 3x3 convolution kernel, and extracting the position offset change trend in adjacent time windows to generate a primary position change characteristic vector;

[0107] inputting the original operation data set into the recurrent neural network layer of the deep learning scheduling model, performing time series correlation analysis on the operation data in consecutive periods through a gated recurrent unit, and extracting the operation interval rule and operation time length fluctuation rule of the vehicle model at different transfer nodes to generate a primary operation rule characteristic vector;

[0108] performing nonlinear mapping on the primary position change characteristic vector and the primary operation rule characteristic vector through an activation function, and screening out feature dimensions with an influence weight on scheduling decision higher than a preset weight threshold to correspondingly form the position change characteristics and the operation rule characteristics.

[0109] It is necessary to note that the original position data sequence is input into the convolution layer of the deep learning scheduling model, and a 3x3 size convolution kernel is used to slide filter the data (smooth noise and retain key position changes); by calculating the position offset change trend (such as the change rate of longitude difference and latitude difference) in the adjacent time window, a primary position change feature vector (containing the direction and speed of position movement) is generated. Similarly, the original job data set is input into the recurrent neural network layer (such as LSTM or GRU) of the deep learning scheduling model, and the gate recurrent unit is used to perform time series correlation analysis on the job data in the continuous period (such as daily job records in a week); the job interval rule (such as the time difference distribution of two jobs) and the job time length fluctuation rule (such as the coefficient of variation of loading and unloading time) of the vehicle model at different transfer nodes are extracted, and a primary job rule feature vector is generated. Finally, the primary position change feature vector and the primary job rule feature vector are nonlinearly mapped (enhance the expression ability of the features) through an activation function (such as ReLU); using a feature importance evaluation algorithm (such as random forest feature importance), the feature dimensions with an impact weight higher than a preset threshold (such as weight>0.1) are selected (redundant features are removed), and finally the position change features and the job rule features are formed. The features obtained in the above manner are more representative, thereby improving the accuracy of the subsequent generated ranking table, facilitating subsequent processing.

[0110] Further, the step of generating a scheduling priority ranking corresponding to the transport vehicle according to the position change feature and the job rule feature comprises:

[0111] extracting the real-time relative azimuth angle and road network topological distance between the transport vehicle and each transfer node from the position change feature, and calculating the shortest transfer path using the Floyd-Warshall algorithm, and converting the shortest transfer path into a distance score value;

[0112] extracting the task execution data corresponding to the same vehicle model within a preset time from the job rule feature, the task execution data including node stay time standard deviation, cross-mode transfer success rate, and daily average job number, and reducing the task execution data to a single job capability factor by principal component analysis;

[0113] weighting and summing the distance score value and the job capability factor according to a weight of 4:6 to obtain an initial priority score, and arranging the ranking results in descending order of score, and then locally optimizing the ranking results by a preset optimization algorithm to generate the scheduling priority ranking.

[0114] It should be noted that the real-time relative azimuth angle (reflecting the direction relationship) and the road network topological distance (considering the distance of actual road connection, not straight-line distance) of the carrying vehicle and each transfer node are extracted from the position change feature; the Floyd-Warshall (multi-source shortest path) algorithm (suitable for solving multi-source shortest path) is used to calculate the shortest transfer path of the carrying vehicle to the target node; and the shortest path length is converted into a distance score value (the shorter the path, the higher the score). At the same time, the task execution data of the same vehicle model within a preset time is extracted from the operation rule feature, including node stay time standard deviation (reflecting operation stability, the smaller the standard deviation, the more stable), cross-mode transfer success rate (such as the success rate of transferring from highway to railway, reflecting the multimodal transport adaptation ability), and daily operation frequency (reflecting operation efficiency); the multi-dimensional data is reduced to a single operation ability factor (comprehensively evaluates the operation ability of the vehicle, the higher the value, the stronger the ability) through principal component analysis. Based on this, the distance score value (weight 40%, emphasizing accessibility) and the operation ability factor (weight 60%, emphasizing operation efficiency) are weighted and summed to obtain an initial priority score; the preliminary sorting is obtained in descending order of the score; and the sorting result is locally optimized (such as adjusting the priority of adjacent vehicles to avoid resource conflicts) through a preset optimization algorithm (such as particle swarm optimization) to finally generate a scheduling priority sorting. Through the above-mentioned manner, the applicability of the final sorting list can be ensured, so as to facilitate subsequent processing.

[0115] Further, the step of generating the scheduling priority sorting by locally optimizing the sorting result through a preset optimization algorithm includes:

[0116] The preset particle swarm optimization algorithm is called out, and the initial descending order sorting is taken as the initial position of the particle;

[0117] A particle fitness function adapted to each vehicle model is created in real time, and the speed and position of the particle are iteratively updated through the particle fitness function;

[0118] The particle position corresponding to the global optimal fitness is retained, and after iteration to the maximum number of times, the global optimal particle position is taken as the final scheduling priority sorting.

[0119] It should be noted that the preset particle swarm optimization algorithm (a swarm intelligence optimization algorithm simulating the foraging behavior of bird flocks) is called, and the initial descending order is taken as the initial position of each particle in the particle swarm (each particle represents a possible priority order scheme). A particle fitness function (an index for evaluating the pros and cons of the order scheme, such as comprehensively considering task completion time, resource utilization, etc., and the higher the value, the better the scheme) is constructed for each vehicle model; the fitness value of each particle is calculated according to the fitness function, and the speed and position of the particle are iteratively updated (i.e., the priority order is adjusted) through the speed-position update formula of the particle swarm algorithm (combined with the individual optimal and global optimal positions). Finally, in the iteration process, the particle position corresponding to the global optimal fitness value (i.e., the current optimal order scheme) is retained; when the number of iterations reaches a preset maximum value (such as 100 times) or the fitness value no longer changes significantly, the iteration is stopped, and the global optimal particle position is taken as the final scheduling priority order. Through the above method, the sorting list of each carrier vehicle can be effectively formulated, thereby eliminating the process of human participation and improving the use efficiency of the carrier vehicle, thereby improving the transportation efficiency.

[0120] Please refer to Figure 2 The third embodiment of the present application provides:

[0121] An unmanned carrier vehicle scheduling system based on multimodal transport, wherein the system comprises:

[0122] A creation module for creating a corresponding multi-source information collection network in real time according to a preset number of transfer nodes, and transmitting historical transportation data collected by the multi-source information collection network in real time to a preset edge computing node;

[0123] A processing module for standardizing the historical transportation data through the preset edge computing node to generate corresponding standard transportation data in real time, and obtaining a number of vehicle models corresponding to the carrier vehicle in real time;

[0124] An extraction module for training a corresponding deep learning scheduling model in real time according to the standard transportation data and a preset neural network, and extracting position change features and work regularity features corresponding to each vehicle model in real time through the deep learning scheduling model;

[0125] An ordering module for generating a scheduling priority order corresponding to the carrier vehicle in real time according to the position change features and the work regularity features, and sending a corresponding scheduling instruction to each carrier vehicle according to the scheduling priority order.

[0126] Further, the processing module is specifically configured to:

[0127] According to a plurality of the transfer nodes, the historical transportation data is divided into a plurality of data subsets, wherein each of the data subsets corresponds to a unique road section identifier;

[0128] According to the road section identifier, a word segmentation processing is performed on the text type transportation state description in each of the data subsets, and the word segmentation result is mapped to a preset standardized state code library;

[0129] A deviation coefficient of the same transfer node collecting data in the same time period is calculated, and the data in the preset standardized state code library that exceeds a threshold range is gradient corrected according to the deviation coefficient, so as to correspondingly generate the standard transportation data.

[0130] Further, the processing module is specifically used for:

[0131] A transportation data set generated by the same transfer node in a plurality of historical same time periods is collected, and a corresponding normal distribution model is constructed in real time according to the mean and variance of the transportation data set;

[0132] The collection data currently generated by the transfer node is input into the normal distribution model to match a corresponding deviation coefficient in real time;

[0133] A gradient correction function is constructed in real time according to the difference between the deviation coefficient and a preset coefficient threshold, and the preset standardized state code library is gradient corrected by the gradient correction function to generate the standard transportation data.

[0134] Further, the extraction module is specifically used for:

[0135] The position record field associated with the vehicle model in the standard transportation data is time stamped and aligned, and a plurality of data segments are divided according to a preset time window, and the latitude and longitude coordinate fluctuation range of the vehicle in each of the data segments is statistically calculated in real time to form a corresponding original position data sequence;

[0136] The loading and unloading record, the driving time record and the stop node record associated with the vehicle model in the standard transportation data are classified and summarized, and the operation frequency and operation time distribution of the same vehicle model at different transfer nodes are statistically calculated to form a corresponding original operation data set;

[0137] The original position data sequence and the original operation data set are parsed to correspondingly generate the position change feature and the operation regularity feature.

[0138] Further, the extraction module is specifically used for:

[0139] input the original position data sequence into the convolution layer of the deep learning scheduling model, slide filter the original position data sequence through a convolution kernel of 3*3 size, and extract the position offset change trend in the adjacent time window to generate a primary position change feature vector;

[0140] input the original job data set into the recurrent neural network layer of the deep learning scheduling model, perform time series correlation analysis on the job data in the continuous period through the gated recurrent unit, and extract the job interval law and job duration fluctuation law of the vehicle model at different transfer nodes to generate a primary job law feature vector;

[0141] nonlinearly map the primary position change feature vector and the primary job law feature vector through an activation function, and filter out feature dimensions with a weight higher than a preset weight threshold to form the position change feature and the job law feature correspondingly.

[0142] Further, the sorting module is specifically configured to:

[0143] extract the real-time relative azimuth angle and road network topological distance between the carrying vehicle and each transfer node from the position change feature, and calculate the shortest transfer path by using the Floyd-Warshall algorithm, and convert the shortest transfer path into a distance score value;

[0144] extract the task execution data corresponding to the same vehicle model within a preset time from the job law feature, the task execution data including node stay duration standard deviation, cross-mode transfer success rate, and daily average job number, and reduce the task execution data to a single job capability factor by using principal component analysis;

[0145] weight the distance score value and the job capability factor by 4:6 to obtain an initial priority score, arrange the initial priority score in descending order, and locally optimize the sorting result by using a preset optimization algorithm to generate the scheduling priority sorting correspondingly.

[0146] Further, the sorting module is specifically configured to:

[0147] call a preset particle swarm optimization algorithm, and set the initial descending order as the initial position of the particle;

[0148] create a particle fitness function suitable for each vehicle model in real time, and update the speed and position of the particle through the particle fitness function;

[0149] retain the particle position corresponding to the global optimal fitness, and after iteration to the maximum number of times, set the global optimal particle position as the final scheduling priority sorting.

[0150] The fourth embodiment of the present application provides a computer, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the intermodal freight transport based unmanned vehicle dispatching method as described above when executing the computer program.

[0151] The fifth embodiment of the present application provides a readable storage medium, which stores a computer program, wherein the program implements the intermodal freight transport based unmanned vehicle dispatching method as described above when executed by a processor.

[0152] In summary, the intermodal freight transport based unmanned vehicle dispatching method and system provided by the above embodiments of the present application can accurately and effectively complete the real-time dispatching of each vehicle, improve the use efficiency of the vehicle and the transportation efficiency.

[0153] It should be noted that each of the above modules can be a functional module or a program module, which can be implemented by software or hardware. For the modules implemented by hardware, each of the above modules can be located in the same processor, or each of the above modules can be located in different processors in any combination.

[0154] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a list of executable instructions for implementing logic functions, which 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, a system including a processor or other system that can fetch instructions from the instruction execution system, apparatus or device and execute the instructions, or in conjunction with these instruction execution systems, apparatus or devices. For the purpose of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate or transport a program for use by or in connection with an instruction execution system, apparatus or device, or in conjunction with these instruction execution systems, apparatus or devices.

[0155] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that is then employable by a computer. Examples of computer-readable media that are further within the spirit of the present application are a computer program product, a computer readable storage medium, and a computer.

[0156] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the embodiments described above, various steps or methods can be implemented, for example, by software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following techniques can be used to implement the hardware: discrete logic circuits having logic gates for implementing logic functions upon data signals, application specific integrated circuits having logic gates for implementing logic functions upon data signals, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like.

[0157] In the description of the specification, the description of the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" and the like means that the specific feature, structure, material or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the application. In the specification, the illustrative expressions of the above terms do 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 more embodiments or examples in a suitable manner.

[0158] The above-described embodiments are merely some embodiments of the present application, which are described in detail and specifically, but should not be understood as limiting the scope of the present application. It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A method for scheduling unmanned vehicles based on multimodal transportation, characterized in that, The method comprises: According to a plurality of preset transfer nodes, a corresponding multi-source information collection network is created in real time, and historical transportation data collected by the multi-source information collection network is transmitted to a preset edge computing node in real time; The historical transportation data is standardized by the preset edge computing node to generate corresponding standard transportation data in real time, and a plurality of vehicle models corresponding to the carrying vehicle are obtained in real time; According to the standard transportation data and a preset neural network, a corresponding deep learning scheduling model is trained in real time, and the position change characteristics and the work regularity characteristics corresponding to each vehicle model are extracted in real time by the deep learning scheduling model; According to the position change characteristics and the work regularity characteristics, a scheduling priority order corresponding to the carrying vehicle is generated in real time, and a corresponding scheduling instruction is sent to each carrying vehicle according to the scheduling priority order.

2. The method of claim 1, wherein: The step of standardizing the historical transportation data by the preset edge computing node to generate corresponding standard transportation data in real time comprises: According to a plurality of transfer nodes, the historical transportation data is divided into a plurality of data subsets, wherein each data subset corresponds to a unique road segment identifier; According to the road segment identifier, the text type transportation state description in each data subset is processed, and the word segmentation result is mapped to a preset standardized state code library; The deviation coefficient of the data collected by the same transfer node in the same time period is calculated, and the data in the preset standardized state code library that exceeds the threshold range is gradient corrected according to the deviation coefficient, so as to generate the standard transportation data.

3. The method of claim 2, wherein: The step of calculating the deviation coefficient of the data collected by the same transfer node in the same time period, and gradient correcting the data in the preset standardized state code library that exceeds the threshold range according to the deviation coefficient to generate the standard transportation data comprises: The transportation data generated by the same transfer node in a plurality of historical same time periods is collected, and a corresponding normal distribution model is constructed in real time according to the mean and variance of the transportation data set; The collected data generated by the transfer node at present is input into the normal distribution model to match the corresponding deviation coefficient in real time; According to the difference between the deviation coefficient and the preset coefficient threshold, a gradient correction function is constructed in real time, and the preset standardized state code library is gradient corrected by the gradient correction function to generate the standard transportation data.

4. The method of claim 1, wherein: The step of extracting the position change characteristics and the work regularity characteristics corresponding to each vehicle model in real time by the deep learning scheduling model comprises: The position record field associated with the vehicle model in the standard transportation data is timestamped, a plurality of data segments are divided according to a preset time window, and the latitude and longitude coordinate fluctuation range of the carrying vehicle in each data segment is calculated in real time to form a corresponding original position data sequence; Classifying and summarizing the loading and unloading records, the driving duration records and the stop node records associated with the vehicle model in the standard transportation data, and counting the operation frequency and the operation duration distribution of the same vehicle model at different transfer nodes to form a corresponding original operation data set; Analyzing the original position data sequence and the original operation data set to correspondingly generate the position change feature and the operation rule feature.

5. The method of claim 4, wherein: The step of analyzing the original position data sequence and the original operation data set to correspondingly generate the position change feature and the operation rule feature comprises: inputting the original position data sequence into the convolution layer of the deep learning scheduling model, performing sliding filtering on the original position data sequence through a 3×3 size convolution kernel, and extracting the position offset change trend in the adjacent time window to generate a primary position change feature vector; inputting the original operation data set into the recurrent neural network layer of the deep learning scheduling model, performing time series correlation analysis on the operation data in the continuous period through the gated recurrent unit, and extracting the operation interval rule and the operation duration fluctuation rule of the vehicle model at different transfer nodes to generate a primary operation rule feature vector; performing nonlinear mapping on the primary position change feature vector and the primary operation rule feature vector through an activation function, and screening out feature dimensions with an influence weight on scheduling decision higher than a preset weight threshold to correspondingly form the position change feature and the operation rule feature.

6. The multi-modal based driverless vehicle dispatching method of claim 1, wherein: The step of generating the scheduling priority order corresponding to the carrying vehicle in real time according to the position change feature and the operation rule feature comprises: extracting the real-time relative azimuth angle and the road network topological distance between the carrying vehicle and each transfer node from the position change feature, and calculating the shortest transfer path by using the Floyd-Warshall algorithm, and converting the shortest transfer path into a distance score value; extracting the task execution data corresponding to the same vehicle model within a preset time from the operation rule feature, the task execution data including the node stay duration standard deviation, the cross-mode transfer success rate and the daily operation frequency, and reducing the task execution data to a single operation ability factor by using the principal component analysis method; performing weighted summation on the distance score value and the operation ability factor according to a weight of 4:6 to obtain an initial priority score, arranging in descending order of score, and performing local optimization on the sorting result by using a preset optimization algorithm to correspondingly generate the scheduling priority order.

7. The multi-modal based driverless vehicle dispatching method of claim 6, wherein: The step of performing local optimization on the sorting result by using a preset optimization algorithm to correspondingly generate the scheduling priority order comprises: calling a preset particle swarm optimization algorithm, and taking the initial descending order as the initial position of particles; creating a particle fitness function adapted to each vehicle model in real time, and updating the speed and position of particles through the particle fitness function; keeping the particle position corresponding to the global optimal fitness, and after iteration to the maximum number of times, taking the global optimal particle position as the final scheduling priority order.

8. A multi-modal based unmanned vehicle dispatching system, characterized by, The system comprises: a creating module, configured to create a corresponding multi-source information collection network in real time according to a preset number of transfer nodes, and transmit historical transportation data collected by the multi-source information collection network in real time to a preset edge computing node; a processing module, configured to perform standardization processing on the historical transportation data by the preset edge computing node, to generate corresponding standard transportation data in real time, and to obtain a number of vehicle models corresponding to a carrying vehicle in real time; an extracting module, configured to train a corresponding deep learning scheduling model in real time according to the standard transportation data and a preset neural network, and to extract position change characteristics and work regularity characteristics corresponding to each vehicle model in real time by the deep learning scheduling model; a sorting module, configured to generate a scheduling priority order corresponding to the carrying vehicle in real time according to the position change characteristics and the work regularity characteristics, and to send a corresponding scheduling instruction to each carrying vehicle according to the scheduling priority order.

9. A computer comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the multi-modal transport-based unmanned carrying vehicle scheduling method in any one of claims 1 to 7.

10. A readable storage medium, having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the multi-modal transport-based unmanned carrying vehicle scheduling method in any one of claims 1 to 7.

Images