Railway transfer center dispatching system and railway transfer center dispatching method

Abstract

The invention provides a railway transfer center dispatching system and a railway transfer center dispatching method, and relates to the technical field of dispatching control, and the railway transfer center dispatching system comprises an information sensing system which is used for obtaining target information related to railway transfer dispatching; the control system is used for acquiring the target information and obtaining a target control instruction related to railway transfer scheduling by adopting a deep Q network DQN algorithm according to the target information; and the execution system is used for obtaining the target control instruction and controlling the transfer device related to railway transfer scheduling to execute the target control instruction. Through mutual cooperation of the information sensing system, the control system and the execution system, the flexibility and adaptability of the railway transfer center dispatching system can be improved, timely and reasonable response can be made to a dynamically changing environment, especially in a complex scene, the resource advantages of a railway transfer center can be fully played, and the service life of the railway transfer center is prolonged. And the overall operation efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of dispatching control, and particularly relates to a dispatching system and a dispatching method for a railway transshipment center. Background Art

[0002] With the continuous growth of railway transportation business volume and the increasing demand for efficient transportation in the logistics industry, the importance of the dispatching work of railway transfer centers, as key hubs, has become increasingly prominent. However, current railway transfer centers are facing numerous challenges. Against the backdrop of the gradual popularization of Internet of Things technology, although there are some applications in the field of railway transportation, they are mostly scattered and single-functional deployments, failing to form a comprehensive and efficient data collection and fusion system. In terms of algorithm applications, traditional railway transfer dispatching mostly relies on simple rules or traditional algorithms and is difficult to adapt to the complex and ever-changing actual operation environment. Specifically, in the traditional dispatching mode of railway transfer centers, the rule-based method pre-establishes a series of fixed dispatching rules, such as arranging vehicle loading and unloading according to the arrival time sequence of trains or determining the entry and exit of vehicles based on the idle situation of the yard, and the entry, exit, and allocation of transfer vehicles within the transfer center mainly rely on the experience judgment of dispatchers. This method lacks comprehensive judgment of multi-dimensional information such as real-time traffic flow, cargo loading and unloading progress, and vehicle location, resulting in excessive waiting times for trains and loading and unloading vehicles, low transfer efficiency, lack of flexibility, inability to make dynamic adjustments according to real-time changes, and difficulty in playing an effective role in complex actual scenarios. Moreover, it is prone to dispatching errors due to human negligence; for the judgment of whether a train enters the station, manual observation or simple track circuit detection methods are mostly used. However, manual observation is limited by factors such as weather and vision, and its accuracy is difficult to guarantee, while simple track circuit detection can only provide rough information on whether a train is approaching and cannot achieve refined management. Although some railway transfer centers have introduced simple automated monitoring devices, such as installing cameras beside train tracks to assist in manually judging the entry of trains and installing Global Positioning System (GPS) positioning devices on vehicles so that dispatchers can understand the vehicle location in real time, these devices only provide information in isolation and are not deeply integrated with the dispatching system, unable to achieve automated dispatching decisions. Therefore, in terms of yard access management, there is a lack of an automated control mechanism, which is likely to cause yard traffic chaos during train entry, affecting operation safety and efficiency; in the cargo transfer link, the verification of train car numbers often relies on manual records and comparisons. This process is not only time-consuming and laborious but also prone to errors. At the same time, the arrangement of transport vehicles to pick-up points lacks scientific planning and cannot be rationally allocated according to the arrival sequence of goods, resulting in possible long-term backlogs of goods that arrive first and affecting the overall transfer efficiency. Although there are some traditional logistics management systems applied to railway transfer centers, the logistics management systems mainly focus on functions such as cargo information entry, query, and statistics. These systems have played a certain role in cargo information management, but in the dispatching link, they still rely on manual operations and cannot achieve intelligent dispatching optimization;For the situation of goods tied to cars, the traditional monitoring method mainly relies on manual inspection, which is difficult to achieve real-time and comprehensive monitoring. This leads to frequent safety accidents caused by loose cargo tying during transportation, which brings great safety hazards to cargo transportation; the operating environment of the railway transshipment center is complex and changeable. Severe weather such as heavy rain, heavy snow, and strong winds will have a serious impact on transportation equipment and operating procedures. Traditional dispatching systems often fail to fully consider severe weather factors when designing and lack corresponding response mechanisms. Once encountering severe weather, dispatching work will fall into chaos, resulting in transportation delays or even interruptions; some studies have tried to apply a single algorithm to a local link of the railway transshipment center, such as using genetic algorithms to optimize the path planning of transport vehicles, or using linear programming algorithms to solve the resource allocation problem of cargo loading and unloading. Although these algorithms have achieved certain results in their respective application scenarios, they cannot fundamentally solve the complexity of railway transshipment center scheduling due to the lack of overall consideration of the entire dispatching system. ;

[0003] In summary, the existing railway transfer center dispatching model carries out dispatching work according to pre-set fixed rules. However, the actual railway transfer scenario is extremely complex, and factors such as transportation demand, cargo type and quantity, and equipment status are in dynamic change at any time. For example, when encountering special transportation tasks, sudden failures of loading and unloading equipment in a certain area, etc., fixed rules are difficult to cope with, resulting in the dispatching plan being out of touch with actual needs, causing problems such as unreasonable extension of vehicle waiting time, idle resources or excessive concentration of resources. Therefore, the existing railway transfer center dispatching model lacks flexibility and adaptability. It relies entirely on established rules and cannot make timely and reasonable responses to dynamically changing environments. As a result, it is difficult to give full play to the resource advantages of the railway transfer center in complex scenarios, which seriously restricts the improvement of overall operational efficiency. Summary of the invention

[0004] The invention provides a railway transfer center dispatching system and a railway transfer center dispatching method, which are used to solve the problem that the existing railway transfer center dispatching mode lacks flexibility and adaptability.

[0005] In order to solve the above technical problems, the embodiments of the present invention provide the following technical solutions:

[0006] In a first aspect, an embodiment of the present invention provides a railway transfer center dispatching system, the system comprising:

[0007] Information perception system, used to obtain target information related to railway transshipment scheduling;

[0008] A control system, used to obtain the target information, and according to the target information, use a deep Q network DQN algorithm to obtain target control instructions related to railway transport scheduling;

[0009] An execution system for obtaining the target control instruction and controlling a transfer device related to railway transfer scheduling to execute the target control instruction.

[0010] Optionally, the target information includes at least one of the following: the position information of the transfer vehicle, the speed information of the transfer vehicle, the relevant images of cargo transfer, the information related to the cargo transfer time of the transferred cargo, and the meteorological information during the transfer process;

[0011] The information perception system includes at least one of the following:

[0012] A vehicle positioning module for obtaining the position information and speed information of the transfer vehicle related to railway transfer scheduling;

[0013] A camera module for obtaining relevant images of the freight train related to railway transfer scheduling;

[0014] A meteorological monitoring module for obtaining meteorological information related to railway transfer scheduling;

[0015] A cargo monitoring module for obtaining information related to the cargo transfer time of the transferred cargo.

[0016] Optionally, the camera module includes:

[0017] An image acquisition unit for obtaining the inbound image, outbound image, docking area image, cargo loading and unloading area image of the freight train related to railway transfer scheduling, and the image of the cargo tied on the transfer vehicle;

[0018] An image recognition unit for using the optical character recognition (OCR) technology combined with the convolutional neural network (CNN) to respectively recognize the inbound image, the docking area image, the cargo loading and unloading area image, and the image of the cargo tied on the transfer vehicle, so as to obtain the inbound information of the freight train, the outbound information of the freight train, the docking area information of the freight train, the cargo loading and unloading area information of the freight train, and the compliance information of the cargo tied on the transfer vehicle after the cargo on the freight train is transported to the transfer vehicle.

[0019] Optionally, the control system includes:

[0020] A network training module for obtaining historical information and historical control instructions related to railway transfer scheduling, and performing DQN algorithm training according to the historical information and the historical control instructions to obtain a DQN network;

[0021] An instruction generation module for inputting the target information into the DQN network to obtain the target control instruction output by the DQN network.

[0022] Optionally, the network training module is specifically used for:

[0023] Obtain the historical information and the historical control instructions related to railway transshipment scheduling;

[0024] Execute a first operation; wherein, the first operation includes: using an evaluation network to predict a transshipment device related to railway transshipment scheduling, and executing a first Q value of the historical control instruction according to the historical information; using a target network to obtain a second Q value according to a reward function and the first Q value; obtain an error value between the second Q value and a preset target Q value;

[0025] In the case where the error value is greater than a preset value, update the prediction parameters in the evaluation network, and loop to execute the first operation until the error value is less than or equal to the preset value;

[0026] Use the evaluation network as the DQN network.

[0027] Optionally, the target control instruction includes at least one of the following:

[0028] A vehicle scheduling instruction, which is an instruction related to the scheduling of transshipment vehicles;

[0029] A yard management instruction, which is an instruction related to the arrival and departure of freight trains at the yard;

[0030] A cargo transshipment instruction, which is an instruction related to cargo transshipment;

[0031] An area cleaning instruction, which is an instruction for cleaning the scheduling area;

[0032] A safety monitoring instruction, which is an instruction related to the safe lashing of goods.

[0033] Optionally, the instruction generation module is specifically used for at least one of the following:

[0034] Input the position information and speed information of the transshipment vehicles in the target information into the DQN network, and obtain a first instruction in the vehicle scheduling instruction output by the DQN network; wherein, the first instruction is used to indicate that when the number of transshipment vehicles in the cargo loading and unloading area is greater than a first preset number, transfer the transshipment vehicles in the cargo loading and unloading area to the vehicle waiting area;

[0035] Input the arrival information of the freight train in the target information into the DQN network, and obtain a second instruction and a third instruction in the yard management instruction output by the DQN network; wherein, the second instruction is used to indicate the arrival of the freight train, and the third instruction is used to indicate the closing of the access control system;

[0036] Input the departure information of the freight train in the target information into the DQN network to obtain the fourth instruction and the fifth instruction in the yard management instruction output by the DQN network; wherein, the fourth instruction is used to instruct the access control system to open, and the fifth instruction is used to instruct the freight train to depart;

[0037] Input the docking area information of the freight train in the target information into the DQN network to obtain the sixth instruction in the vehicle scheduling instruction output by the DQN network; wherein, the sixth instruction is used to instruct to dispatch a second preset number of transfer vehicles to the target area where the freight train is docked;

[0038] Input the cargo handling information of the freight train in the target information into the DQN network to obtain the seventh instruction in the cargo transfer instruction output by the DQN network; wherein, the seventh instruction is used to instruct the transfer vehicle to transport the cargo after the cargo unloading is completed;

[0039] Input the compliance information of tying the cargo on the transfer vehicle after the cargo on the freight train in the target information is carried to the transfer vehicle into the DQN network to obtain the eighth instruction in the safety monitoring instruction output by the DQN network; wherein, the eighth instruction is used to instruct that the tying of the cargo on the transfer vehicle is non-compliant and needs to be re-tied;

[0040] Input the information related to the cargo transfer time of the transferred cargo in the target information into the DQN network to obtain the ninth instruction in the cargo transfer instruction output by the DQN network; wherein, the ninth instruction is used to instruct the transfer vehicle to transfer the cargo out of the station within a preset duration or before a preset time point;

[0041] Input the meteorological information in the target information into the DQN network to obtain the tenth instruction in the vehicle scheduling instruction and the eleventh instruction in the area cleaning instruction output by the DQN network; wherein, the tenth instruction is used to instruct the transfer vehicle to transfer the cargo to a sheltered area, and the eleventh instruction is used to instruct the cleaning equipment to clean all areas related to railway transfer scheduling.

[0042] Optionally, the execution system is specifically used for at least one of the following:

[0043] Send the first instruction to the transfer vehicle, and the transfer vehicle is used to be scheduled from the cargo handling area to the vehicle waiting area according to the first instruction;

[0044] Send the second instruction to the freight train and send the third instruction to the access control system. The freight train is used to enter the station according to the second instruction, and the access control system is used to close according to the third instruction;

[0045] Send the fourth instruction to the access control system and send the fifth instruction to the freight train. The access control system is used to open according to the fourth instruction, and the freight train is used to depart according to the fifth instruction;

[0046] Send the sixth instruction to the transfer vehicle. The transfer vehicle is used to be dispatched to the target area where the freight train is docked according to the sixth instruction;

[0047] Send the seventh instruction to the transfer vehicle. The transfer vehicle is used to transport the goods after the goods unloading is completed according to the seventh instruction;

[0048] Send the eighth instruction to the safety supervision device. The safety supervision device is used to re-tie the goods tied on the transfer vehicle according to the eighth instruction;

[0049] Send the ninth instruction to the transfer vehicle. The transfer vehicle is used to transfer the goods out of the station within a preset duration or before a preset time point according to the ninth instruction;

[0050] Send the tenth instruction to the transfer vehicle and send the eleventh instruction to the cleaning equipment. The transfer vehicle is used to transfer the goods to the shelter area according to the eighth instruction, and the cleaning equipment is used to clean all areas related to the railway transfer scheduling according to the ninth instruction.

[0051] In a second aspect, an embodiment of the present invention further provides a railway transfer center scheduling method, and the method includes:

[0052] Obtain the target information related to the railway transfer scheduling sent by the information perception system;

[0053] According to the target information, use the deep Q-network DQN algorithm to obtain the target control instruction related to the railway transfer scheduling;

[0054] Send the target control instruction to the execution system. The execution system is used to control the transfer device related to the railway transfer scheduling to execute the target control instruction.

[0055] The beneficial effects of the present invention are:

[0056] The railway transfer center dispatching system provided by the solution of the present invention includes an information perception system, which is used to obtain target information related to railway transfer dispatching, and also includes a control system, which is used to obtain the target information, and according to the target information, uses the Deep Q-Network (DQN) algorithm to obtain a target control instruction related to railway transfer dispatching. It also includes an execution system, which is used to obtain the target control instruction and control the transfer device related to railway transfer dispatching to execute the target control instruction. Through the mutual cooperation among the above information perception system, control system and execution system, the flexibility and adaptability of the railway transfer center dispatching system can be increased, and a timely and reasonable response can be made to the dynamically changing environment. Especially in complex scenarios, the resource advantages of the railway transfer center can be fully utilized to improve the overall operation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a schematic structural diagram of the railway transfer center dispatching system provided by an embodiment of the present invention;

[0058] Figure 2 It is a specific structural schematic diagram of the railway transfer center dispatching system provided by an embodiment of the present invention;

[0059] Figure 3 It is a schematic structural diagram of the data acquisition layer and the central control layer in the railway transfer center dispatching system provided by an embodiment of the present invention;

[0060] Figure 4 It is a schematic structural diagram of the execution layer in the railway transfer center dispatching system provided by an embodiment of the present invention;

[0061] Figure 5 It is a flowchart of the railway transfer center dispatching method provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0062] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0063] Aiming at the problem that the existing dispatching mode of railway transfer centers lacks flexibility and adaptability, the present invention provides a railway transfer center dispatching system and a railway transfer center dispatching method.

[0064] As Figure 1 shown, an embodiment of the present invention provides a railway transfer center dispatching system, and the system includes:

[0065] An information perception system 101, which is used to obtain target information related to railway transfer dispatching.

[0066] In the embodiment of the present invention, the information perception system 101 collects and stores in real time the target information related to railway transfer scheduling.

[0067] It should be noted that in the real scenario, the railway transfer scenario is extremely complex, and elements such as transportation demand, types and quantities of goods, and equipment status are constantly changing dynamically. Therefore, it is necessary to collect in real time the target information related to railway transfer scheduling. By considering the target information related to railway transfer scheduling and then performing subsequent scheduling, the tight correlation between all links of the entire scheduling system can be ensured.

[0068] Among them, the target information includes at least one of the following: the position information of the transfer vehicle, the speed information of the transfer vehicle, the relevant images of goods transfer, the information related to the goods transfer time of the transferred goods, and the meteorological information during the transfer process.

[0069] In addition, it should also be noted that in the railway transfer center scheduling system provided by the embodiment of the present invention, a comprehensive Internet of Things system is constructed, covering a vehicle positioning management system, a camera monitoring system, and a meteorological monitoring system. Each system collects data in real time and transmits it to the central control system through the Internet of Things to achieve the efficient integration of multi-source data.

[0070] It can be understood that among them, the information perception system can also be called the data acquisition layer or the perception layer, and this data acquisition layer is composed of multiple modules. The information perception system includes at least one of the following:

[0071] A vehicle positioning module, which is used to obtain the position information and speed information of the transfer vehicle (or called transportation vehicle) related to railway transfer scheduling, that is, the vehicle positioning module collects in real time the vehicle position and driving speed and other information of the transfer vehicle related to railway transfer scheduling. Among them, this vehicle positioning module can also be called a vehicle positioning management system, and specifically includes: a positioning device and an Internet of Things (IoT) communication module. The positioning device is used to obtain the vehicle position and driving speed in real time, and the IoT communication module is used to transmit the information obtained by the positioning device to the control system 102;

[0072] A camera module, which is used to obtain the relevant images of the freight train related to railway transfer scheduling. This camera module can be a camera monitoring system, which is used to collect the image data of the freight train's arrival status at the station, the car body number of the train, and the situation of goods tied to the vehicle related to railway transfer scheduling;

[0073] A meteorological monitoring module for obtaining meteorological information related to railway transfer scheduling. Optionally, the meteorological information may be severe weather information, such as rain and snow weather, wind speed information, or weather information such as heavy rain, heavy snow, strong wind, etc., or weather intensity data such as heavy rain, heavy snow, strong wind, etc. The meteorological monitoring module includes a meteorological monitor and an IoT communication module. The meteorological monitor is used to collect severe weather information in real time, such as weather types and intensity data of heavy rain, heavy snow, strong wind, etc. The IoT communication module is used to transmit the information collected by the meteorological monitor to the control system 102;

[0074] A cargo monitoring module for obtaining information related to the cargo transfer time of the transferred cargo. The information related to the cargo transfer time includes the time when the cargo enters the station, the time when the cargo is required to leave the station, the time when the cargo enters the cargo storage area, and the residence time of the cargo in the cargo storage area.

[0075] The control system 102 is used to obtain the target information and, based on the target information, use the Deep Q-Network (DQN) algorithm to obtain a target control instruction related to railway transfer scheduling.

[0076] In the embodiment of the present invention, the information perception system 101 sends the target information to the control system 102, and the control system 102 receives the target information, that is, receives the multi-source data of the data acquisition layer.

[0077] The role of the data acquisition layer is to collect the target information related to railway transfer scheduling and provide data for the control system 102.

[0078] Among them, the control system 102 can also be called the central control layer. The core of the central control layer is the Deep Q-Network (DQN) algorithm module. The DQN algorithm module receives multi-source data from the data acquisition layer, generates a scheduling decision through learning and operation. The central control layer also includes a scheduling decision generation module. The DQN algorithm module sends the scheduling decision to the scheduling decision generation module, and the scheduling decision generation module generates a target control instruction related to railway transfer scheduling according to the scheduling decision output by the DQN algorithm module.

[0079] The control system 102 is used to receive and integrate various types of data transmitted from the data acquisition layer, perform data analysis and processing and algorithm decision-making, and then generate specific execution instructions and send them to the execution system 103. These multi-source data of the target information are converged to the central control system in real time through the Internet of Things, constructing a comprehensive and dynamic data model of the railway transfer center, providing a rich and accurate data basis for subsequent intelligent decision-making.

[0080] The execution system 103 is used to obtain the target control instruction and control the transfer device related to railway transfer scheduling to execute the target control instruction.

[0081] In an embodiment of the present invention, the control system 102 sends a target control instruction to the execution system 103. The execution system 103 receives the target control instruction sent by the control system 102, and the execution system 103 executes specific operations according to the scheduling decision corresponding to the target control instruction.

[0082] The execution layer is used to receive the instruction issued by the central control layer and convert it into a specific operation instruction to achieve the actual scheduling and control of the railway transfer center.

[0083] In summary, through the mutual cooperation among the information perception system 101, the control system 102, and the execution system 103 in the embodiment of the present invention, the flexibility and adaptability of the scheduling system of the railway transfer center can be increased, and a timely and reasonable response can be made to the dynamically changing environment. Especially in complex scenarios, the resource advantages of the railway transfer center can be fully utilized to improve the overall operation efficiency.

[0084] In an optional embodiment, the camera module includes:

[0085] An image acquisition unit, configured to acquire the inbound image of the freight train related to railway transfer scheduling, the outbound image of the freight train, the image of the docking area, the image of the cargo loading and unloading area, and the image of the goods tied on the transfer vehicle;

[0086] An image recognition unit, configured to use optical character recognition (OCR) technology in combination with convolutional neural networks (CNN) to respectively recognize the inbound image, the docking area image, the cargo loading and unloading area image, and the image of the goods tied on the transfer vehicle, so as to obtain the inbound information of the freight train, the outbound information of the freight train, the docking area information of the freight train, the cargo loading and unloading area information of the freight train, and the compliance information of the goods tied on the transfer vehicle after the goods on the freight train are transported to the transfer vehicle.

[0087] Specifically, the image acquisition unit includes: high-definition cameras, which are respectively located at the train inbound port, the train docking area, and the cargo loading and unloading area, and are used to respectively acquire images of the freight train from the above areas. The high-definition cameras also correspond to an image recognition server, which uses technologies such as OCR and CNN to judge the inbound state and outbound state of the train, identify the car body number, the positioning number of the cargo loading and unloading location (i.e., the docking area information of the freight train), and judge the compliance of the goods tying (i.e., the compliance information of the goods tied on the transfer vehicle after the goods on the freight train are transported to the transfer vehicle).

[0088] It should also be noted that first, it is necessary to build a railway transfer center dispatching system. The building process includes the deployment of hardware devices and the building of software systems. Among them, in the deployment of hardware devices, satellite positioning devices are installed on transport vehicles, and their positioning accuracy can reach the meter level, which can real-time feedback the vehicle position, speed, and cargo details, and transmit the data to the central control system at high speed through the 5G communication module. At the inbound port of the transfer center, each train stopping platform, and the cargo loading and unloading area, high-definition intelligent cameras are installed. These cameras have characteristics such as low illuminance and wide dynamic range, and can clearly collect images in complex light environments. A complete set of meteorological monitoring stations are deployed, including high-precision rain gauges, wind speed and direction sensors, etc., which can real-time monitor the surrounding bad weather conditions.

[0089] In the building of software systems, the central control system software is developed and deployed, integrating a data processing engine, which can quickly integrate data from various hardware devices. At the same time, a deep Q-network (DQN) algorithm module is embedded to build an image recognition software based on a convolutional neural network (CNN) for processing the images collected by the cameras to accurately identify the train arrival status, car body number, and the situation of goods tied on the vehicle. A distributed data storage and management platform is built to store a large amount of real-time and historical data, supporting second-level data query and analysis.

[0090] The high-definition monitoring cameras deployed in the cargo storage area are used to continuously collect image information of the cargo area. Through image recognition technology, the collected images are analyzed in real time. First, the You Only Look Once (YOLO) series of algorithms based on deep learning are used to identify the cargo boxes in the images and determine their position coordinates. At the same time, a unique identifier is assigned to each identified cargo box (associated with the cargo box number in the database through image feature matching).

[0091] The control system integrates data from the vehicle positioning management system, camera monitoring system, meteorological monitoring system, and the newly added cargo box residence time sensors. A real-time updated database is established to associate and store the unique identifier of the cargo box (such as the cargo box number) with the corresponding residence time to ensure data consistency and traceability. At the same time, the data in the database is updated at a certain time interval (such as every minute) to reflect the real-time change of the cargo box residence time.

[0092] In an optional embodiment, the control system 102 includes:

[0093] A network training module, which is used to obtain historical information and historical control instructions related to railway transfer dispatching, perform DQN algorithm training according to the historical information and the historical control instructions, and obtain a DQN network;

[0094] An instruction generation module, configured to input the target information into the DQN network to obtain the target control instruction output by the DQN network.

[0095] Wherein, the historical information includes at least one of the following: historical position information of the transfer vehicle, historical speed information of the transfer vehicle, historical relevant images of cargo transfer, historical cargo transfer time-related information of the transferred cargo, and historical meteorological information during the transfer process;

[0096] Wherein, the historical control instructions include at least one of the following:

[0097] Historical vehicle scheduling instructions, which are historical instructions related to the scheduling of the transfer vehicle;

[0098] Historical yard management instructions, which are historical instructions related to the arrival and departure of freight trains;

[0099] Historical cargo transfer instructions, which are historical instructions related to cargo transfer;

[0100] Historical area cleaning instructions, which are historical instructions for cleaning the scheduling area;

[0101] Historical safety monitoring instructions, which are historical instructions related to the safe lashing of goods.

[0102] Further, the network training module is specifically configured to:

[0103] Obtain the historical information and the historical control instructions related to railway transfer scheduling;

[0104] Execute a first operation; wherein, the first operation includes: using an evaluation network to predict a transfer device related to railway transfer scheduling, and obtaining a first Q value of executing the historical control instruction according to the historical information; using a target network to obtain a second Q value according to a reward function and the first Q value; obtaining an error value between the second Q value and a preset target Q value;

[0105] In the case that the error value is greater than a preset value, update the prediction parameters in the evaluation network, and loop to execute the first operation until the error value is less than or equal to the preset value;

[0106] Use the evaluation network as the DQN network.

[0107] Specifically, during the training process, first, data preprocessing is performed on the historical information. For example, noise data generated due to signal interference is removed through a data cleaning algorithm, and then the data with different dimensions is unified in scale using a normalization method to facilitate subsequent processing. At the same time, accurate labels are added to each training data, such as decision labels for whether a vehicle should enter the station or wait outside, result labels for whether the comparison of car body numbers is correct, etc.

[0108] Initialize the evaluation network and the target network. The evaluation network is set to include 3 hidden layers, with 128 neurons in each layer, and is used to predict the Q-values (i.e., the first Q-values) of various actions such as vehicle scheduling and yard management in different states; the structure of the target network is the same as that of the evaluation network. Define the reward function. Exemplarily, if efficient vehicle scheduling is successfully achieved and the vehicle waiting time is reduced by more than 10%, a positive reward of 10 points is given; if the car body number is accurately identified, a reward of 5 points is given; if vehicles are reasonably scheduled in rainy weather to avoid goods getting wet, a reward of 15 points is given. Conversely, if vehicle scheduling errors lead to congestion, 20 points are deducted; if the car body number is misidentified, 10 points are deducted. In the simulation environment, the agent selects actions according to the output of the evaluation network. For example, it decides that a certain vehicle enters the station. After executing the action, it observes the new state and the obtained reward, and stores the data in the experience replay pool. When the system can reasonably arrange a transport vehicle to transport the cargo box before the warning time threshold of the cargo box residence time is reached, a relatively high positive reward (such as +10 points) is given; when the cargo box residence time reaches the warning time threshold but is transported away before the processing time threshold, a certain positive reward (such as +5 points) is given; if the cargo box residence time exceeds the processing time threshold and is still not transported away, a severe negative reward (such as -20 points) is given. Through this reward mechanism, the DQN algorithm is guided to learn a reasonable cargo box residence time management strategy. During training, 128 pieces of data are randomly sampled from the experience replay pool, and the mean squared error (i.e., the error value) between the predicted Q-value of the evaluation network and the target Q-value (i.e., the second Q-value) calculated by the target network is calculated, and the parameters of the evaluation network are updated using the Adam optimizer. Thus, the trained DQN network is obtained.

[0109] Update the training data in the experience replay pool and add experience samples related to cargo box residence time management. These samples include the state (including the cargo box residence time), actions (such as the decision to arrange a vehicle to transport the cargo box), rewards (corresponding rewards are given according to whether the cargo box is transported away on time), and new states. By continuously training these new experience samples, the DQN algorithm can better adapt to the requirements of cargo box residence time management and optimize the scheduling decision.

[0110] Algorithm Optimization and Verification: Conduct 1000 tests of different scenarios in the simulation environment every week, and analyze indicators such as the average waiting time of vehicles and the cargo transfer efficiency. According to the test results, if it is found that the algorithm performs poorly in complex cargo transfer scenarios, appropriately increase the number of neurons in the hidden layer of the evaluation network to 160, and adjust the learning rate from 0.001 to 0.0008. Select a specific area of the transfer center every month for a 3-day algorithm verification test, compare the actual scheduling results with the algorithm prediction results, and further optimize the algorithm.

[0111] The specific training process of the DQN network is that the evaluation network is responsible for predicting the Q value of the action based on the current state, the target network provides a stable target Q value for the evaluation network, and the experience replay pool is used to store and randomly sample the experiences of the agent interacting with the environment. The three cooperate to jointly realize the effective application of the DQN algorithm in the railway transfer center scheduling system.

[0112] Network Structure Design:

[0113] The evaluation network includes an input layer, a convolutional layer, a fully connected layer, and an output layer. In the railway transfer center scheduling system, since it involves image data (such as images collected by cameras) and numerical data (such as vehicle positions and meteorological information), a convolutional neural network (CNN) structure is adopted here. For the image data part, the convolutional layer is used for feature extraction; for the numerical data part, it is directly input into the fully connected layer.

[0114] (1) Input Layer

[0115] Receive various state information from the perception layer, including vehicle positions, train arrival status, cargo information, real-time vehicle flow, cargo loading and unloading progress, and severe weather information, etc. After encoding this information, it is used as the input of the input layer.

[0116] (2) Convolutional Layer (for Image Data)

[0117] For the image information contained in the input, such as the image of the train car number collected by the camera or the image of the goods tied on the vehicle, the convolutional layer is used for feature extraction. The convolutional layer consists of multiple convolutional kernels, which extract local features of the image through convolutional operations, and then use the ReLU function provided by the deep learning framework PyTorch for non-linear transformation.

[0118] (3) Fully Connected Layer

[0119] Flatten the feature map output by the convolutional layer and input it into the fully connected layer together with the numerical data. The fully connected layer consists of multiple neurons, each neuron is connected to all neurons in the previous layer, and calculations are performed through weighted summation and activation functions to gradually extract higher-level features.

[0120] (4) Output Layer

[0121] Output the Q-values of all possible actions. The actions include the vehicle entering the station or waiting outside, controlling the access to the station yard, arranging the transport vehicle to a specific pick-up point, etc. The number of neurons in the output layer is equal to the size of the action space.

[0122] Forward propagation process:

[0123] The forward propagation process of the evaluation network is to calculate the input state information through the layers of the network, and finally obtain the Q-values of each action. The specific steps are as follows:

[0124] (1) Input the input state information into the input layer of the network.

[0125] (2) If there is a convolutional layer, perform convolutional operations and activation function processing on the image data to extract image features.

[0126] (3) Flatten the feature map output by the convolutional layer and input it into the fully connected layer together with the numerical data for weighted summation and activation function calculation.

[0127] (4) In the output layer, obtain the Q-values of each action through linear transformation.

[0128] Training process:

[0129] The training objective of the evaluation network is to make the predicted Q-values as close as possible to the target Q-values. During the training process, randomly sample a batch of data from the experience replay pool, predict the Q-values according to the current state, compare them with the target Q-values calculated by the target network, and update the parameters of the evaluation network by minimizing the mean square error between the two.

[0130] The specific steps are as follows:

[0131] 1. Randomly sample a batch of data from the experience replay pool, including states, actions, rewards, and new states.

[0132] 2. For each sample, use the evaluation network to predict the Q-values of all actions according to the current state.

[0133] 3. Calculate the target Q-values according to the target network. The calculation method of the target Q-values is as follows: If the current state is not a terminal state, the target Q-value is equal to the reward plus the discount factor multiplied by the maximum Q-value in the new state; if the current state is a terminal state, the target Q-value is equal to the reward.

[0134] 4. Calculate the mean square error between the predicted Q-values and the target Q-values.

[0135] 5. Use the backpropagation algorithm to calculate the gradient of the error with respect to the parameters of the evaluation network.

[0136] 6. Update the parameters of the evaluation network according to the gradient, and use the optimization algorithm Adam for parameter update.

[0137] Target network:

[0138] The structure of the target network is the same as that of the evaluation network to ensure that the target network can calculate the same type of Q value as the evaluation network, that is, where r is the reward obtained by executing the action in the current state, γ is the discount factor, is the next state, is the optimal action in the next state, and is the parameter of the target network. By stabilizing the training process of the target network, it is avoided that the evaluation network is updated too frequently during the training process, resulting in unstable training. The parameters of the target network are periodically copied from the evaluation network. Every certain number of time steps (such as 1000 steps), the parameters of the evaluation network are copied to the target network. So that the target network remains relatively stable for a period of time, providing a relatively stable target Q value for the evaluation network to achieve training convergence.

[0139] Experience replay pool:

[0140] The purpose of setting up the experience replay pool is to improve the utilization rate of data, so that the same experience sample can be used multiple times in different training steps, to make full use of the experience of the interaction between the agent and the environment, reduce the number of samples required for training, and speed up the training speed.

[0141] Data storage:

[0142] Establish a sample for storing the interaction between the agent and the environment. Each time the agent executes an action in the environment, a new state and reward will be obtained. The current state, action, reward, and new state are stored as an experience sample in the experience replay pool. Each experience sample contains a quadruple, where s is the current state, is the executed action, r is the reward obtained after executing the action, and is the next state after executing the action. The experience replay pool is implemented in the form of a queue or a list. When the experience replay pool reaches the maximum capacity, the earliest stored sample will be discarded to ensure that the size of the experience replay pool remains unchanged.

[0143] Random sampling:

[0144] When training the evaluation network, a batch of data is randomly selected from the experience replay pool for training to break the correlation between samples, make the training data more independent and identically distributed, and improve the efficiency and stability of training.

[0145] Optionally, the target control instruction includes at least one of the following:

[0146] Vehicle scheduling instruction, and the vehicle scheduling instruction is an instruction related to the scheduling of transfer vehicles;

[0147] A yard management instruction, where the yard management instruction is an instruction related to the arrival and departure of freight trains;

[0148] A cargo transfer instruction, where the cargo transfer instruction is an instruction related to cargo transfer;

[0149] An area cleaning instruction, where the area cleaning instruction is an instruction for cleaning the dispatching area;

[0150] A safety monitoring instruction, where the safety monitoring instruction is an instruction related to the safe lashing of goods.

[0151] That is, the railway transfer center dispatching system provided by the embodiments of the present invention constructs a railway transfer center dispatching system based on the Internet of Things and the Deep Q-Network (DQN) algorithm, realizes the real-time collection and deep fusion of multi-source data, and uses the powerful learning and decision-making ability of the DQN algorithm to intelligently and collaboratively manage various operations of the railway transfer center. By comprehensively integrating operations such as vehicle dispatching, train arrival management, cargo transfer, safety monitoring, and cargo residence time management, an efficient, intelligent, and safe railway transfer center operation system is created.

[0152] Optionally, the instruction generation module is specifically used for at least one of the following:

[0153] First, input the position information and speed information of the transfer vehicle in the target information into the DQN network to obtain a first instruction in the vehicle dispatching instruction output by the DQN network; where the first instruction is used to indicate that when the number of transfer vehicles in the cargo loading and unloading area is greater than a first preset number, transfer the transfer vehicles in the cargo loading and unloading area to the vehicle waiting area;

[0154] If the DQN network determines that the number of transfer vehicles in a certain cargo loading and unloading area is greater than the first preset number based on the position information and speed information of the transfer vehicle, a first instruction in the vehicle dispatching instruction is generated, and this first dispatching instruction is used to transfer a preset number of transfer vehicles in the cargo loading and unloading area to the vehicle waiting area. For example, transfer 10 transfer vehicles to the vehicle waiting area.

[0155] Second, input the arrival information of the freight train in the target information into the DQN network to obtain a second instruction and a third instruction in the yard management instruction output by the DQN network; where the second instruction is used to indicate the arrival of the freight train, and the third instruction is used to indicate the closing of the access control system;

[0156] When the DQN network recognizes that a train is approaching the station based on the approaching information of the freight train, it generates a second instruction and a third instruction. The second instruction is used to indicate the approach of the train, and the third instruction controls the access control system to close through the yard access control system, prohibiting access to the relevant area where related trains are prohibited from entering.

[0157] Furthermore, if the DQN network recognizes that the car body number of the freight train does not match the approaching area of the freight train, it generates an alarm instruction. This alarm instruction is used to indicate by the alarm system that the car body number of the train does not match the approaching area of the freight train and notify the staff.

[0158] III. Input the departure information of the freight train in the target information into the DQN network to obtain a fourth instruction and a fifth instruction in the yard management instruction output by the DQN network; among them, the fourth instruction is used to indicate the opening of the access control system, and the fifth instruction is used to indicate the departure of the freight train.

[0159] When the DQN network determines that the train is about to depart based on the departure information of the freight train, it generates a fourth instruction and a fifth instruction in the yard management instruction. Among them, the fourth instruction is used to control the access control system to open through the yard access control system, and the fifth instruction is used to indicate the departure of the train.

[0160] IV. Input the parking area information of the freight train in the target information into the DQN network to obtain a sixth instruction in the vehicle scheduling instruction output by the DQN network; among them, the sixth instruction is used to indicate scheduling a second preset number of transfer vehicles to the target area where the freight train is parked;

[0161] Based on the parking area information of the freight train, the DQN network generates a sixth instruction in the vehicle scheduling instruction. This sixth instruction can be used to schedule a second preset number of transfer vehicles to the target area where the freight train is parked in advance. For example, arrange 30 transfer vehicles to go to the target area to pick up goods to ensure that the goods that arrive first are given priority for transfer.

[0162] V. Input the goods loading and unloading information of the freight train in the target information into the DQN network to obtain a seventh instruction in the goods transfer instruction output by the DQN network; among them, the seventh instruction is used to indicate that after the goods are unloaded, the transfer vehicle transports the goods;

[0163] When the DQN network determines that the goods on the freight train are being transferred to the transfer train based on the goods loading and unloading information of the freight train, it generates a seventh instruction in the goods transfer instruction. This seventh instruction indicates that after the goods are unloaded, the transfer vehicle transports the goods.

[0164] VI. Input the compliance information of the goods tied on the transfer vehicle after the goods on the freight train in the target information are transported to the transfer vehicle into the DQN network to obtain the eighth instruction in the safety monitoring instruction output by the DQN network; wherein, the eighth instruction is used to indicate that the tying of the goods on the transfer vehicle is non-compliant and needs to be re-tied.

[0165] The DQN network is used to monitor the situation of goods tying. If the tying of the goods on the transfer vehicle is non-compliant and needs to be re-tied, an alarm is given to ensure the accuracy of goods transportation.

[0166] VII. Input the information related to the goods transfer time of the transferred goods in the target information into the DQN network to obtain the ninth instruction in the goods transfer instruction output by the DQN network; wherein, the ninth instruction is used to instruct the transfer vehicle to transfer the goods out of the station within a preset duration or before a preset time point.

[0167] Among them, the information related to the goods transfer time includes at least one of the goods arrival time, the required departure time of the goods, the time when the goods enter the goods storage area, and the residence time of the goods in the goods storage area. Therefore, the DQN network can determine the preset time point when the goods need to be transferred out of the station or the preset duration of the goods staying in the storage area according to the information related to the goods transfer time, and thus generate the ninth instruction, which is used to instruct the transfer vehicle to transfer the goods out of the station within a preset duration or before a preset time point.

[0168] Optionally, develop a special time management and early warning module in the control system. This module is responsible for reading the data of the residence time of the cargo box in the database (i.e., the information related to the goods transfer time) and comparing it with the preset time threshold. Set two key time thresholds, namely the early warning time threshold (20 hours) and the processing time threshold (24 hours). When the residence time of the cargo box reaches the early warning time threshold (20 hours), the module automatically generates an early warning message; when the residence time exceeds the processing time threshold (24 hours), the module immediately issues an alarm message.

[0169] VIII. Input the meteorological information in the target information into the DQN network to obtain the tenth instruction in the vehicle scheduling instruction and the eleventh instruction in the area cleaning instruction output by the DQN network; wherein, the tenth instruction is used to instruct the transfer vehicle to transfer the goods to a sheltered area, and the eleventh instruction is used to instruct the cleaning equipment to clean all areas related to the railway transfer scheduling.

[0170] The DQN network incorporates meteorological information into the scheduling decision-making process and formulates corresponding strategies for different severe weather conditions. That is, based on the meteorological information, the DQN network generates the tenth instruction, which is used to indicate that during heavy rain, the transfer vehicle transfers the goods to a sheltered area to prevent the goods from getting wet. Also, the eleventh instruction is used to indicate that during heavy snow, the cleaning equipment cleans all areas related to railway transfer scheduling, such as cleaning the tracks and the roads in the yard. This comprehensively ensures the safe operation of the railway transfer center.

[0171] It should also be noted that the early warning and alarm information are pushed in multiple ways. On the one hand, the early warning and alarm information are displayed on the operation interface of the control system in a prominent color and prompt box to attract the attention of the dispatcher. On the other hand, notifications are sent to the relevant responsible persons via text messages to ensure that the information can be conveyed in a timely manner. After receiving the information, the relevant responsible persons arrange for the transport vehicles to take away the cargo containers as soon as possible.

[0172] In the embodiment of the present invention, the control system introduces the Deep Q-Network (DQN) algorithm as the core decision-making engine. The central control system inputs the integrated multi-source data into the DQN algorithm module. The evaluation network in this module predicts the Q-values of all possible actions based on the current state, providing a basis for decision-making. Through a carefully designed reward function, the DQN algorithm is guided to continuously learn and optimize. If the system achieves reasonable vehicle scheduling, accurately identifies the train car numbers, effectively responds to severe weather to ensure transportation, etc., a positive reward is given; conversely, if there are scheduling mistakes, identification errors, or improper scheduling during severe weather leading to accidents, a negative reward is given. During the training process, the experience replay pool stores the state, action, reward, and new state data. The evaluation network predicts the Q-value based on the current state, and the target network calculates the target Q-value based on the new state. By minimizing the mean square error between the two, the parameters of the evaluation network are updated, enabling the DQN algorithm to gradually learn the optimal scheduling strategy under different complex states, achieving reasonable allocation of vehicle entry and waiting, scientific arrangement of transport vehicles to the pick-up points, and precise control of yard access and exit.

[0173] Optionally, the execution system is specifically used for at least one of the following:

[0174] Sending the first instruction to the transfer vehicle (or to the transfer vehicle control system), and the transfer vehicle is used to be scheduled from the cargo loading and unloading area to the vehicle waiting area according to the first instruction;

[0175] Sending the second instruction to the freight train and sending the third instruction to the access control system. The freight train is used to enter the station according to the second instruction, and the access control system is used to close according to the third instruction;

[0176] Send the fourth instruction to the access control system (or the station access control system) and send the fifth instruction to the freight train. The access control system is used to open according to the fourth instruction, and the freight train is used to leave the station according to the fifth instruction;

[0177] Send the sixth instruction to the transfer vehicle. The transfer vehicle is used to be dispatched to the target area where the freight train is parked according to the sixth instruction;

[0178] Send the seventh instruction to the transfer vehicle. The transfer vehicle is used to transport the goods after the goods unloading is completed according to the seventh instruction;

[0179] Send the eighth instruction to the safety supervision device. The safety supervision device is used to re-tie the goods tied on the transfer vehicle according to the eighth instruction;

[0180] Send the ninth instruction to the transfer vehicle. The transfer vehicle is used to transfer the goods out of the station within a preset duration or before a preset time point according to the ninth instruction;

[0181] Send the tenth instruction to the transfer vehicle and send the eleventh instruction to the cleaning equipment. The transfer vehicle is used to transfer the goods to the sheltered area according to the eighth instruction, and the cleaning equipment is used to clean all areas related to the railway transfer scheduling according to the ninth instruction.

[0182] Send an alarm instruction to the alarm system. The alarm instruction is used to indicate that the train car number does not match the inbound area of the freight train and notify the staff.

[0183] Moreover, in the embodiments of the present invention, the scheduling result data, equipment operation data, etc. generated by the system are collected every day. It is found through analysis that in a specific period of each month, due to the concentration of goods types, the efficiency of the existing scheduling strategy is reduced. Therefore, the training data of the DQN algorithm is adjusted, the special scenario data in this period is increased, and the scheduling strategy is optimized. When a new type of equipment failure is found, the fault diagnosis and processing mechanism of the system is improved.

[0184] The above process can be Figure 2 executed by the railway transfer center scheduling system. Specifically, the railway transfer center scheduling system includes a sensing layer, a central control layer, and an execution layer; where:

[0185] The sensing layer includes: a vehicle positioning management system, a camera monitoring system, and a meteorological monitoring system;

[0186] The central control layer includes:

[0187] The data reception and preprocessing module is responsible for receiving and integrating various types of data transmitted from the perception layer, including vehicle information, train status information, meteorological information, etc. It identifies different states such as approaching the station, having left the station, and having entered the station through the vehicle's position and driving speed information, records information such as the train has come to a stop, is in motion, arrival time, departure time, etc. through the train status information, and records the time information related to cargo transfer, saves information including rainfall, wind speed, fog, etc. through the meteorological information, and then stores the above analysis results in the background database;

[0188] The DQN algorithm module is used to execute the evaluation network, target network, experience replay pool, and training operations. Among them, the evaluation network is used to predict the action Q value based on the input state, the target network is used to calculate the target Q value, the experience replay pool is used to store state, action, reward, and new state data, and the training module is used to update the evaluation network parameters by minimizing the mean square error;

[0189] The decision execution module is used to issue execution instructions including controlling the access to the station yard, arranging the pick-up points for vehicles, and issuing warnings according to the decision results of the DQN algorithm.

[0190] The execution layer includes:

[0191] The station yard access control system is used to control the opening and closing of the access control system, control the entrances and exits of the station yard, and decide whether the vehicle enters the station or waits outside;

[0192] The transportation vehicle scheduling system is used to arrange the pick-up points for vehicles, send the number of the pick-up point to the driver via text message, and the driver drives to the target pick-up point to pick up the goods according to the number;

[0193] The warning system is used to notify the relevant processing personnel via text message for the warning information received from the central control layer.

[0194] The following combines Figure 3 and Figure 4 to illustrate the functions of each module in the railway transfer center scheduling system, Figure 3 is the structural schematic diagram of the data acquisition layer and the central control layer in the railway transfer center scheduling system, Figure 4 is the structural schematic diagram of the execution layer in the railway transfer center scheduling system.

[0195] The data acquisition layer collects the above target information through the vehicle positioning management system, camera monitoring system, and meteorological monitoring system, and transmits the target information to the central control layer. The scheduling decision module of the central control layer distributes the scheduling decisions to various functional modules such as vehicle scheduling, train arrival management, and cargo transfer arrangement according to the output of the DQN algorithm module, realizing the intelligent regulation of various operations in the railway transfer center.

[0196] The vehicle scheduling module in the execution layer commands the actions of the transfer vehicles, deciding whether they enter the station, leave the station, or go to a specific pick-up point; the train arrival management module manages the relevant operations of the railway yard. When a train arrives, access to the yard is prohibited and the ban is lifted after the train docks; the cargo transfer arrangement module coordinates the transfer vehicles and the railway yard to ensure efficient cargo transfer; the cargo residence time management module coordinates the residence time of the cargo in the cargo storage area.

[0197] In summary, the railway transfer center scheduling system provided by the embodiments of the present invention embeds the Deep Q-Network (DQN) algorithm in the central control system. This algorithm learns and makes decisions based on a large amount of historical and real-time data. According to multi-source information such as the train arrival status, vehicle position, cargo information, real-time vehicle flow, cargo loading and unloading progress, bad weather information, equipment operation parameters, energy consumption data, personnel information, and cargo residence time, an optimal scheduling decision is generated; multiple important services such as cargo residence time management are incorporated into a unified scheduling system. By setting a cargo residence time threshold and including it in the DQN algorithm decision consideration, timely cargo transfer is ensured. The camera monitoring system is used to monitor the lashing of goods on the vehicle in real time and comprehensively. Once any non-compliance is found, an alarm is immediately issued. Combining with the equipment operation monitoring system to monitor the key equipment in real time, potential faults are warned in advance. Before the arrival of bad weather, according to the DQN algorithm, protective measures for vehicles and cargo are arranged in advance to effectively reduce safety risks.

[0198] With the help of the vehicle positioning management system and the central control system based on the Internet of Things and DQN algorithm, the present invention can collect information such as vehicle position and train entry and exit in real time. After being calculated by the DQN algorithm, it automatically and accurately arranges the vehicles to enter the station or wait outside, improving the vehicle scheduling efficiency.

[0199] The present invention deploys a camera monitoring system. Using image recognition technology, it automatically and accurately judges the train arrival status and controls the access to the yard; at the same time, it automatically takes pictures of the train car body number and compares it with the plan. If there is a discrepancy, an alarm is issued, improving the accuracy and automation level of judgment and recognition.

[0200] Based on the train arrival and cargo situation, the central control system of the present invention uses the DQN algorithm to generate a scientific scheduling plan, automatically arranging the transport vehicles to specific pick-up points to ensure that the goods that arrive first are given priority for transfer and optimizing the cargo transfer process.

[0201] The present invention takes pictures of the lashing of goods on the vehicle through the camera monitoring system, uses image recognition technology to judge whether it meets the requirements, and issues an alarm if it does not, effectively reducing the transport safety risk.

[0202] The present invention introduces a meteorological monitoring system to collect severe weather information in real time. The DQN algorithm incorporates this information into the scheduling decision-making process and formulates corresponding scheduling strategies for different severe weather conditions. For example, during heavy rain, vehicles are arranged to safe areas, and during heavy snow, snow removal equipment is dispatched, enhancing the system's ability to handle severe weather and ensuring transportation safety and efficiency.

[0203] A complete scheduling system architecture for the railway transfer center is constructed, where various hardware devices, software systems, and algorithm modules cooperate with each other. The hardware devices are responsible for data collection, the software systems perform data processing, algorithm hosting, and decision execution, and all parts work closely together to achieve the full-process intelligent operation of the railway transfer center from information collection, analysis to scheduling decision execution.

[0204] The system processes a large amount of historical data every day, realizes policy iteration through the experience replay pool, and can continuously improve the decision accuracy in complex scenarios. By constructing a three-dimensional safety protection system through car body number image recognition, cargo lashing visual inspection, and meteorological data fusion analysis, the efficiency is significantly improved compared with traditional manual verification.

[0205] As Figure 5 shown, an embodiment of the present invention also provides a scheduling method for a railway transfer center, and the method includes:

[0206] Step 501: Obtain target information related to railway transfer scheduling sent by the information perception system;

[0207] Step 502: According to the target information, use the deep Q-network DQN algorithm to obtain a target control instruction related to railway transfer scheduling;

[0208] Step 503: Send the target control instruction to the execution system, and the execution system is used to control the transfer device related to railway transfer scheduling to execute the target control instruction.

[0209] Among them, the scheduling method for the railway transfer center is executed by the above control system.

[0210] Optionally, the target information includes at least one of the following: the position information of the transfer vehicle, the speed information of the transfer vehicle, the relevant images of cargo transfer, the information related to the cargo transfer time of the transferred cargo, and the meteorological information during the transfer process.

[0211] Optionally, according to the target information, using the deep Q-network DQN algorithm to obtain a target control instruction related to railway transfer scheduling includes:

[0212] Obtain historical information and historical control instructions related to railway transfer scheduling;

[0213] Perform DQN algorithm training according to the historical information and the historical control instructions to obtain a DQN network;

[0214] Input the target information into the DQN network to obtain the target control instruction output by the DQN network.

[0215] Optionally, perform DQN algorithm training based on the historical information and the historical control instruction to obtain a DQN network, including:

[0216] Obtain the historical information and the historical control instruction related to railway transshipment scheduling;

[0217] Execute a first operation; wherein, the first operation includes: using an evaluation network to predict a transshipment device related to railway transshipment scheduling, and predicting a first Q value of executing the historical control instruction according to the historical information; using a target network to obtain a second Q value according to a reward function and the first Q value; obtaining an error value between the second Q value and a preset target Q value;

[0218] In the case where the error value is greater than a preset value, update the prediction parameters in the evaluation network, and loop to execute the first operation until the error value is less than or equal to the preset value;

[0219] Use the evaluation network as the DQN network.

[0220] Optionally, the target control instruction includes at least one of the following:

[0221] A vehicle scheduling instruction, which is an instruction related to the scheduling of transshipment vehicles;

[0222] A yard management instruction, which is an instruction related to the arrival and departure of freight trains at the yard;

[0223] A cargo transshipment instruction, which is an instruction related to cargo transshipment;

[0224] An area cleaning instruction, which is an instruction for cleaning the scheduling area;

[0225] A safety monitoring instruction, which is an instruction related to the safe lashing of goods.

[0226] Optionally, inputting the target information into the DQN network to obtain the target control instruction output by the DQN network includes at least one of the following:

[0227] Input the location information and speed information of the transfer vehicle in the target information into the DQN network to obtain the first instruction in the vehicle scheduling instruction output by the DQN network; wherein, when the number of transfer vehicles in the cargo loading and unloading area indicated by the first instruction is greater than a first preset number, the transfer vehicles in the cargo loading and unloading area are scheduled to the vehicle waiting area.

[0228] Input the inbound information of the freight train in the target information into the DQN network to obtain the second instruction and the third instruction in the yard management instruction output by the DQN network; wherein, the second instruction is used to indicate the inbound of the freight train, and the third instruction is used to indicate the closing of the access control system.

[0229] Input the departure information of the freight train in the target information into the DQN network to obtain the fourth instruction and the fifth instruction in the yard management instruction output by the DQN network; wherein, the fourth instruction is used to indicate the opening of the access control system, and the fifth instruction is used to indicate the departure of the freight train.

[0230] Input the docking area information of the freight train in the target information into the DQN network to obtain the sixth instruction in the vehicle scheduling instruction output by the DQN network; wherein, the sixth instruction is used to indicate scheduling a second preset number of transfer vehicles to the target area where the freight train is docked.

[0231] Input the cargo handling information of the freight train in the target information into the DQN network to obtain the seventh instruction in the cargo transfer instruction output by the DQN network; wherein, the seventh instruction is used to indicate that after the goods are unloaded, the transfer vehicle transports the goods.

[0232] Input the compliance information of tying the goods on the transfer vehicle after the goods on the freight train are carried onto the transfer vehicle in the target information into the DQN network to obtain the eighth instruction in the safety monitoring instruction output by the DQN network; wherein, the eighth instruction is used to indicate that the tying of the goods on the transfer vehicle is non-compliant and needs to be re-tied.

[0233] Input the information related to the cargo transfer time of the transferred cargo in the target information into the DQN network to obtain the ninth instruction in the cargo transfer instruction output by the DQN network; wherein, the ninth instruction is used to indicate that the transfer vehicle transfers the goods out of the station within a preset duration or before a preset time point.

[0234] Input the meteorological information in the target information into the DQN network to obtain the tenth instruction in the vehicle scheduling instruction and the eleventh instruction in the area cleaning instruction output by the DQN network; wherein, the tenth instruction is used to instruct the transfer vehicle to transfer the goods to the shelter area, and the eleventh instruction is used to instruct the cleaning equipment to clean all areas related to the railway transfer scheduling.

[0235] It should be noted that the railway transfer center scheduling method provided in the embodiments of the present invention is executed by the control system in the railway transfer center scheduling system. Therefore, all embodiments of the control system in the above-mentioned railway transfer center scheduling system are applicable to this railway transfer center scheduling method and can achieve the same or similar technical effects.

[0236] The above are the preferred embodiments of the present invention. It should be pointed out that for those of ordinary skill in the art, several improvements and refinements can be made without departing from the principles described in the present invention, and these improvements and refinements are also within the protection scope of the present invention.

Claims

1. A railway transfer center dispatching system, characterized in that, The system includes: An information perception system, configured to obtain target information related to railway transfer scheduling; A control system, configured to obtain the target information, and based on the target information, use the deep Q-network (DQN) algorithm to obtain a target control instruction related to railway transfer scheduling; An execution system, configured to obtain the target control instruction and control a transfer device related to railway transfer scheduling to execute the target control instruction.

2. The system according to claim 1, characterized in that The target information includes at least one of the following: the position information of the transfer vehicle, the speed information of the transfer vehicle, relevant images of cargo transfer, information related to the cargo transfer time of the transferred cargo, and meteorological information during the transfer process; The information perception system includes at least one of the following: A vehicle positioning module, configured to obtain the position information and speed information of the transfer vehicle related to railway transfer scheduling; A camera module, configured to obtain relevant images of the freight train related to railway transfer scheduling; A meteorological monitoring module, configured to obtain meteorological information related to railway transfer scheduling; A cargo monitoring module, configured to obtain information related to the cargo transfer time of the transferred cargo.

3. The system according to claim 2, wherein The camera module includes: An image acquisition unit, configured to obtain the inbound image, outbound image, docking area image, cargo loading and unloading area image of the freight train related to railway transfer scheduling, and the image of the cargo tied on the transfer vehicle; An image recognition unit, configured to use optical character recognition (OCR) technology combined with a convolutional neural network (CNN) to respectively recognize the inbound image, the docking area image, the cargo loading and unloading area image, and the image of the cargo tied on the transfer vehicle, and obtain the inbound information of the freight train, the outbound information of the freight train, the docking area information of the freight train, the cargo loading and unloading area information of the freight train, and the compliance information of the cargo tied on the transfer vehicle after the cargo on the freight train is transferred to the transfer vehicle.

4. The system according to claim 1, wherein The control system includes: A network training module, configured to obtain historical information and historical control instructions related to railway transfer scheduling, and perform DQN algorithm training based on the historical information and the historical control instructions to obtain a DQN network; An instruction generation module, configured to input the target information into the DQN network to obtain the target control instruction output by the DQN network.

5. The system according to claim 4, characterized in that, Specifically, the network training module is configured to: Obtain the historical information and the historical control instructions related to railway transfer scheduling; Perform a first operation; wherein, the first operation includes: using an evaluation network to predict a transfer device related to railway transfer scheduling, and obtaining a first Q value of the historical control instruction executed according to the historical information; using a target network to obtain a second Q value according to a reward function and the first Q value; obtaining an error value between the second Q value and a preset target Q value; In the case where the error value is greater than a preset value, update the prediction parameters in the evaluation network, and loop to perform the first operation until the error value is less than or equal to the preset value; Use the evaluation network as the DQN network.

6. The system according to claim 5, characterized in that, The target control instruction includes at least one of the following: A vehicle scheduling instruction, where the vehicle scheduling instruction is an instruction related to the scheduling of the transfer vehicle; The yard management instruction, where the yard management instruction is an instruction related to the arrival and departure of freight trains; The cargo transfer instruction, where the cargo transfer instruction is an instruction related to cargo transfer; The area cleaning instruction, where the area cleaning instruction is an instruction for cleaning the dispatching area; The safety monitoring instruction, where the safety monitoring instruction is an instruction related to the secure lashing of cargo.

7. The system according to claim 6, wherein The instruction generation module is specifically used for at least one of the following: Input the location information and speed information of the transfer vehicle in the target information into the DQN network to obtain the first instruction in the vehicle dispatching instruction output by the DQN network; where the first instruction is used to indicate that when the number of transfer vehicles in the cargo loading and unloading area is greater than a first preset number, dispatch the transfer vehicles in the cargo loading and unloading area to the vehicle waiting area; Input the arrival information of the freight train in the target information into the DQN network to obtain the second instruction and the third instruction in the yard management instruction output by the DQN network; where the second instruction is used to indicate the arrival of the freight train, and the third instruction is used to indicate the closing of the access control system; Input the departure information of the freight train in the target information into the DQN network to obtain the fourth instruction and the fifth instruction in the yard management instruction output by the DQN network; where the fourth instruction is used to indicate the opening of the access control system, and the fifth instruction is used to indicate the departure of the freight train; Input the docking area information of the freight train in the target information into the DQN network to obtain the sixth instruction in the vehicle dispatching instruction output by the DQN network; where the sixth instruction is used to indicate dispatching a second preset number of transfer vehicles to the target area where the freight train is docked; Input the cargo handling information of the freight train in the target information into the DQN network to obtain the seventh instruction in the cargo transfer instruction output by the DQN network; where the seventh instruction is used to indicate that after the cargo is unloaded, the transfer vehicle transports the cargo; Input the compliance information of lashing the cargo on the transfer vehicle after the cargo on the freight train is carried onto the transfer vehicle in the target information into the DQN network to obtain the eighth instruction in the safety monitoring instruction output by the DQN network; where the eighth instruction is used to indicate that the lashing of the cargo on the transfer vehicle is non-compliant and needs to be re-lashed; Input the information related to the cargo transfer time of the transferred cargo in the target information into the DQN network to obtain the ninth instruction in the cargo transfer instruction output by the DQN network; where the ninth instruction is used to indicate that the transfer vehicle transfers the cargo out of the station within a preset duration or before a preset time point; Input the meteorological information in the target information into the DQN network to obtain the tenth instruction in the vehicle scheduling instruction and the eleventh instruction in the area cleaning instruction output by the DQN network; wherein, the tenth instruction is used to instruct the transfer vehicle to transfer the goods to the shelter area, and the eleventh instruction is used to instruct the cleaning equipment to clean all areas related to railway transfer scheduling.

8. The system according to claim 7, wherein The execution system is specifically used for at least one of the following: Send the first instruction to the transfer vehicle, and the transfer vehicle is used to be scheduled from the goods loading and unloading area to the vehicle waiting area according to the first instruction; Send the second instruction to the freight train and send the third instruction to the access control system. The freight train is used to enter the station according to the second instruction, and the access control system is used to close according to the third instruction; Send the fourth instruction to the access control system and send the fifth instruction to the freight train. The access control system is used to open according to the fourth instruction, and the freight train is used to leave the station according to the fifth instruction; Send the sixth instruction to the transfer vehicle, and the transfer vehicle is used to be scheduled to the target area where the freight train is parked according to the sixth instruction; Send the seventh instruction to the transfer vehicle, and the transfer vehicle is used to transport the goods after the goods unloading is completed according to the seventh instruction; Send the eighth instruction to the safety supervision device, and the safety supervision device is used to re-tie the goods tied to the transfer vehicle according to the eighth instruction; Send the ninth instruction to the transfer vehicle, and the transfer vehicle is used to transfer the goods out of the station within a preset time period or before a preset time point according to the ninth instruction; Send the tenth instruction to the transfer vehicle and send the eleventh instruction to the cleaning equipment. The transfer vehicle is used to transfer the goods to the shelter area according to the eighth instruction, and the cleaning equipment is used to clean all areas related to railway transfer scheduling according to the ninth instruction.

9. A dispatching method for a railway transfer center, characterized in that, The method includes: Obtain the target information related to railway transfer scheduling sent by the information perception system; According to the target information, use the deep Q-network DQN algorithm to obtain the target control instruction related to railway transfer scheduling; Send the target control instruction to the execution system, and the execution system is used to control the transfer device related to railway transfer scheduling to execute the target control instruction.

Images