Coal trackless rubber-tyred vehicle scheduling method based on space-time diagram neural network

Abstract

The invention provides a coal trackless rubber-tyred vehicle scheduling method based on a space-time diagram neural network, and relates to the technical field of intelligent planning and vehicle scheduling. The scheme comprises the steps of determining a space-time diagram sequence corresponding to a current sliding time window; performing feature extraction on the space-time diagram sequence on the basis of a space-time diagram neural network generated by pre-training, and performing state prediction on the nodes on the basis of the extracted features to obtain a predicted node state at the next moment; determining a target control strategy of the trackless rubber-tyred vehicle according to the node state at the next moment; and carrying out vehicle scheduling on the trackless rubber-tyred vehicle based on the target control strategy. By accurately predicting the state and environment change of the underground coal mine vehicle, potential risk factors can be pre-judged, so that corresponding measures are taken in advance, and the safety of mine operation is guaranteed; a transportation strategy can be dynamically adjusted, and the working efficiency of the vehicle is improved; automation and intellectualization of the coal mine transportation process are achieved, and the risk of human errors is reduced.

Description

Technical Field

[0001] The present application relates to the technical fields of intelligent planning and vehicle scheduling, and in particular to a method for scheduling coal trackless rubber-tyred vehicles based on spatiotemporal graph neural networks. Background Art

[0002] Coal mine auxiliary transportation is a general term for various transportation activities other than coal transportation in the coal mine production process, mainly involving the transportation of personnel, materials, equipment, ores and other items, and is a key component of the coal mine transportation system. Among the existing coal mine auxiliary transportation methods, trackless rubber-tyred vehicles have gradually become a widely used means of transportation in coal mine auxiliary transportation due to their significant advantages such as high transportation efficiency, wide application range, strong climbing performance, large carrying capacity and low transportation cost.

[0003] However, the existing dispatching method of rubber-tyred trackless vehicles still mainly relies on manual experience for management and dispatching. Dispatchers realize the allocation and control of transportation tasks through observation and manual adjustment. Since the manual dispatching method relies heavily on personnel experience and has a certain degree of subjectivity, it often leads to problems such as low transportation efficiency, poor safety, and information asymmetry in the dispatching process. In addition, in complex environments, manual dispatching is prone to misoperation and delays, which further affects the overall efficiency and safety of coal mine production. Summary of the invention

[0004] The present invention provides a method for dispatching a trackless rubber-tyred coal vehicle based on a spatiotemporal graph neural network, so as to at least solve the problem that the dispatching method of trackless rubber-tyred vehicles in the related art still mainly relies on manual experience for management and dispatching, which often leads to low transportation efficiency and poor safety.

[0005] The first aspect of the present application proposes a method for scheduling coal trackless rubber-tyred vehicles based on a spatiotemporal graph neural network, including: determining a spatiotemporal graph sequence corresponding to a current sliding time window, wherein the spatiotemporal graph sequence includes multiple spatiotemporal graphs, the spatiotemporal graph is composed of nodes and edges, the nodes include vehicle nodes corresponding to the trackless rubber-tyred vehicles in a coal mine and environment nodes corresponding to the environment in the coal mine, the edges represent the relationship between the nodes, each node carries node attribute information, and each edge corresponds to a weight coefficient; based on the spatiotemporal graph neural network generated by pre-training, feature extraction is performed on the spatiotemporal graph sequence, and state prediction is performed on the nodes based on the extracted features to obtain the predicted node state at the next moment; the target control strategy of the trackless rubber-tyred vehicle is determined according to the node state at the next moment; and the trackless rubber-tyred vehicle is scheduled based on the target control strategy.

[0006] According to one embodiment of the present application, a target control strategy for a rubber-tyred trackless vehicle is determined based on a node state at the next moment, including: obtaining dynamic parameters corresponding to the coal mine underground at the next moment; inputting the node state and dynamic parameters into a pre-trained reinforcement learning model, and obtaining the target control strategy at the next moment output by the reinforcement learning model.

[0007] According to one embodiment of the present application, the node state and dynamic parameters are input into a pre-trained reinforcement learning model to obtain the target control strategy at the next moment output by the reinforcement learning model, including: using the node state and dynamic parameters at the next moment as the state input of the intelligent agent in the reinforcement learning model, and combining with a preset reward function to calculate the expected reward value of each candidate control strategy at the next moment; selecting the candidate control strategy with the largest expected reward value as the target control strategy at the next moment.

[0008] According to one embodiment of the present application, determining the space-time graph sequence corresponding to the current sliding time window includes: determining multiple moments within the current sliding time window; determining the space-time graph of each of the multiple moments; and arranging the space-time graph of each moment in chronological order to obtain a space-time graph sequence.

[0009] According to one embodiment of the present application, at any moment, a method for generating a space-time graph includes: at any moment, collecting multimodal sensor data of the trackless rubber-tyred vehicle and the surrounding environment of the trackless rubber-tyred vehicle; obtaining the space-time graph of the previous moment of any moment, and updating the space-time graph of the previous moment based on the multimodal sensor data to obtain the space-time graph of any moment.

[0010] According to one embodiment of the present application, multimodal sensor data of a trackless rubber-tyred vehicle and its surrounding environment are collected, including: collecting initial sensor data of the trackless rubber-tyred vehicle and its surrounding environment based on sensors installed on the trackless rubber-tyred vehicle; preprocessing the initial sensor data to obtain multimodal sensor data obtained after the preprocessing, wherein the preprocessing at least includes data cleaning, data normalization and data standardization.

[0011] According to one embodiment of the present application, a method for designing a reward function includes: obtaining a safety reward parameter, an efficiency reward parameter, an energy consumption reward parameter, and a task completion reward parameter of a rubber-tyred trackless vehicle; and performing weighted summation on the safety reward parameter, the efficiency reward parameter, the energy consumption reward parameter, and the task completion reward parameter to obtain a reward function.

[0012] The second aspect of the present application proposes a coal trackless rubber-tyred vehicle scheduling device based on a spatiotemporal graph neural network, including: a first determination module, used to determine the spatiotemporal graph sequence corresponding to the current sliding time window, wherein the spatiotemporal graph sequence includes multiple spatiotemporal graphs, the spatiotemporal graph is composed of nodes and edges, the nodes include vehicle nodes corresponding to the trackless rubber-tyred vehicle in the coal mine and environment nodes corresponding to the environment in the coal mine, the edges represent the relationship between the nodes, each node carries node attribute information, and each edge corresponds to a weight coefficient; a state prediction module, used to extract features of the spatiotemporal graph sequence based on the spatiotemporal graph neural network generated by pre-training, and predict the state of the node based on the extracted features, and obtain the predicted node state at the next moment; a second determination module, used to determine the target control strategy of the trackless rubber-tyred vehicle according to the node state at the next moment; a vehicle scheduling module, used to schedule the trackless rubber-tyred vehicle based on the target control strategy.

[0013] According to one embodiment of the present application, the second determination module is also used to: obtain the corresponding dynamic parameters in the coal mine at the next moment; input the node state and dynamic parameters into the pre-trained reinforcement learning model to obtain the target control strategy of the next moment output by the reinforcement learning model.

[0014] According to one embodiment of the present application, the second determination module is also used to: use the node state and dynamic parameters at the next moment as the state input of the intelligent agent in the reinforcement learning model, and combine with the preset reward function to calculate the expected reward value of each candidate control strategy at the next moment; select the candidate control strategy with the largest expected reward value as the target control strategy at the next moment.

[0015] According to one embodiment of the present application, the first determination module is also used to: determine multiple moments within the current sliding time window; determine the space-time graph of each of the multiple moments; and arrange the space-time graph of each moment in chronological order to obtain a space-time graph sequence.

[0016] According to one embodiment of the present application, the first determination module is also used to: collect multimodal sensor data of the rubber-tyred trackless vehicle and the surrounding environment of the rubber-tyred trackless vehicle at any time; obtain the space-time diagram of the previous moment at any time, and update the space-time diagram of the previous moment based on the multimodal sensor data to obtain the space-time diagram of any moment.

[0017] According to one embodiment of the present application, the first determination module is also used to: collect initial sensor data of the trackless rubber-tyred vehicle and the surrounding environment of the trackless rubber-tyred vehicle based on sensors installed on the trackless rubber-tyred vehicle; preprocess the initial sensor data to obtain multimodal sensor data obtained after preprocessing, and the preprocessing at least includes data cleaning, data normalization and data standardization.

[0018] According to one embodiment of the present application, the second determination module is also used to: obtain the safety reward parameter, efficiency reward parameter, energy consumption reward parameter and task completion reward parameter of the rubber-tyred trackless vehicle; perform weighted summation of the safety reward parameter, efficiency reward parameter, energy consumption reward parameter and task completion reward parameter to obtain a reward function.

[0019] The third aspect embodiment of the present application proposes an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to implement the coal trackless rubber-tyred vehicle scheduling method based on the spatiotemporal graph neural network as described in the first aspect embodiment of the present application.

[0020] The fourth aspect embodiment of the present application proposes a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to implement the coal trackless rubber-tyred vehicle scheduling method based on spatiotemporal graph neural network as described in the first aspect embodiment of the present application.

[0021] The fifth aspect embodiment of the present application proposes a computer program product, including a computer program, which, when executed by a processor, implements the coal trackless rubber-tyred vehicle scheduling method based on the spatiotemporal graph neural network as described in the first aspect embodiment of the present application.

[0022] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects: the present application can predict potential dangerous factors by accurately predicting the status of vehicles and environmental changes in coal mines, so as to take corresponding measures in advance to ensure the safety of mine operations; it can dynamically adjust transportation strategies, reduce unnecessary pauses and waiting, and improve the work efficiency of vehicles; it realizes the automation and intelligence of the coal mine transportation process, does not rely on human intervention, and reduces the risk of human error.

[0023] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0025] Figure 1 It is a schematic diagram of an exemplary implementation of a method for scheduling coal trackless rubber-tyred vehicles based on a spatiotemporal graph neural network, shown in an embodiment of the present application.

[0026] Figure 2 It is a schematic diagram of a time-space diagram sequence shown in an embodiment of the present application.

[0027] Figure 3 It is a schematic diagram of an exemplary implementation of a method for scheduling coal trackless rubber-tyred vehicles based on a spatiotemporal graph neural network, shown in an embodiment of the present application.

[0028] Figure 4 It is a schematic diagram of a coal trackless rubber-tyred vehicle dispatching device based on a spatiotemporal graph neural network shown in an embodiment of the present application.

[0029] Figure 5 It is a schematic diagram of an electronic device shown in one embodiment of the present application. DETAILED DESCRIPTION

[0030] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0031] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application are in compliance with the relevant provisions of national laws and regulations.

[0032] It should be noted that in the embodiments of the present application, certain software, components, models and other existing solutions in the industry may be mentioned, and they should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0033] Figure 1 is a schematic diagram of an exemplary implementation of a method for dispatching a coal trackless rubber-tyred vehicle based on a spatiotemporal graph neural network shown in the present application, such as Figure 1 As shown, the coal trackless rubber-tyred vehicle scheduling method based on spatiotemporal graph neural network includes the following steps:

[0034] S101, determine a space-time graph sequence corresponding to the current sliding time window, wherein the space-time graph sequence includes multiple space-time graphs, and the space-time graph is composed of nodes and edges. The nodes include vehicle nodes corresponding to the rubber-tyred trackless vehicle in the coal mine and environment nodes corresponding to the environment in the coal mine. The edges represent the relationship between the nodes, each node carries node attribute information, and each edge corresponds to a weight coefficient.

[0035] It is not difficult to understand that in order to realize the real-time scheduling of coal trackless rubber-tyred vehicles, it is necessary to determine the space-time graph sequence corresponding to the current sliding time window in real time, and the sliding time window can continue to slide according to the set step size.

[0036] In addition to the vehicle nodes corresponding to the rubber-tyred trackless vehicles in the coal mine, the present application also introduces the environment nodes corresponding to the environment in the coal mine in the space-time graph. For example, the environment nodes may include entity nodes such as tunnel nodes, obstacle nodes, and vent nodes. The environment nodes may establish connections with the vehicle nodes.

[0037] Each node carries node attribute information. For example, a vehicle node can carry attribute information such as vehicle load, speed, type, etc.; a lane node can carry attribute information such as lane width and slope; an obstacle node can carry attribute information such as obstacle size, etc.

[0038] Among them, the edge represents the relationship between nodes. For example, the edge between vehicle nodes can represent the following relationship and obstacle avoidance relationship between vehicles; for example, the edge between vehicle nodes and lane nodes can represent the topological relationship between vehicles and lanes; for example, the edge between vehicle nodes and obstacle nodes can represent the distance relationship between vehicles and obstacles, etc.

[0039] In addition, in order to more accurately reflect the specific impact of different edge relationships on scheduling decisions, the concept of edge weight coefficient is introduced in this application, which is allocated according to the importance and priority of the edge relationship to ensure that more reasonable and efficient decisions can be made based on the actual impact of different edge relationships.

[0040] Figure 2 is a schematic diagram of a time-space diagram sequence shown in this application, such as Figure 2 As shown in the figure, the time series information of vehicle nodes and environment nodes is represented by the evolution of the space-time graph. If one space-time graph is generated for one second and one sliding time window is 100 seconds, then the space-time graph sequence corresponding to one sliding time window contains 100 space-time graphs.

[0041] For ease of understanding, the expression of the space-time graph corresponding to time t is given below:

[0042] X t = {V t ,E t ,W t}

[0043] In the above formula, X t is the space-time diagram corresponding to time t, V t is the node set in the space-time graph corresponding to time t, including entity nodes such as vehicle nodes, obstacle nodes, lane nodes, etc. t is the set of edges in the space-time graph corresponding to time t, where the edges represent the relationship between entity nodes, W t It is the set of weight coefficients of the edges in the space-time graph corresponding to time t, reflecting the influence of different edge relationships on the scheduling decision.

[0044] S102, based on the spatiotemporal graph neural network generated by pre-training, feature extraction is performed on the spatiotemporal graph sequence, and the node state is predicted based on the extracted features to obtain the predicted node state at the next moment.

[0045] In this application, a graph neural network model suitable for processing spatiotemporal graph data is selected and pre-trained to obtain a pre-trained spatiotemporal graph neural network.

[0046] When training the spatiotemporal graph neural network, a large number of spatiotemporal graph sequence samples and sample data of the node state at the next moment to be predicted corresponding to each spatiotemporal graph sequence sample can be collected in advance. Then, the spatiotemporal graph neural network can be iteratively trained based on the spatiotemporal graph sequence samples and the corresponding sample data of the node state at the next moment to be predicted until the training is completed, thereby obtaining the pre-trained spatiotemporal graph neural network.

[0047] In this application, the space-time graph sequence corresponding to the current sliding time window is input into the pre-trained space-time graph neural network, and then based on the pre-trained space-time graph neural network, the features of the nodes and edges in the input space-time graph sequence are extracted, the interactions between the nodes and the spatiotemporal dynamic changes are captured, the nodes are predicted, and the predicted node state at the next moment is obtained (such as the vehicle's position, speed, fault state, changes in the mine environment, etc.).

[0048] The moments in this application refer to the moments when the space-time graph is acquired.

[0049] For example, if 1 second corresponds to the generation of 1 space-time graph, and a sliding time window is 100 seconds, then the space-time graph sequence corresponding to a sliding time window contains 100 space-time graphs. If the space-time graph sequence corresponding to the 1st to the 100th seconds (a total of 100 space-time graphs) is input into the pre-trained space-time graph neural network, the node status of each node of the predicted space-time graph output by the pre-trained space-time graph neural network can be obtained at the 101st second; and so on.

[0050] Among them, the above-mentioned one time-space diagram is generated for one second only for example. In actual applications, it can be set according to the actual situation. For example, in order to reduce the amount of calculation, it can also be set to generate one time-space diagram for every 5 seconds (that is, every 5 seconds is a collection moment). Assuming that a sliding time window is still 100 seconds, the time-space diagram sequence corresponding to a sliding time window contains 20 time-space diagrams (for example, the time-space diagram corresponding to the 1st second, the time-space diagram corresponding to the 6th second, the time-space diagram corresponding to the 11th second, the time-space diagram corresponding to the 16th second... and so on).

[0051] S103, determining a target control strategy for the rubber-tyred trackless vehicle according to the node state at the next moment.

[0052] The next moment here refers to the moment when the next space-time diagram is obtained. For example, if one space-time diagram is generated every 1 second, the next moment refers to the next second of the current sliding time window. If one space-time diagram is generated every 5 seconds, the next moment refers to the moment corresponding to 5 seconds after the current sliding time window.

[0053] Among them, the target control strategy refers to how to determine the scheduling and control measures of the vehicle based on the predicted node status (such as vehicle status and environmental status). For example, based on the predicted vehicle position, speed and environmental changes at the next moment, the control strategy may be to control the vehicle to turn, adjust the vehicle speed, or perform other necessary operations to ensure safe and efficient transportation.

[0054] For example, if it is predicted that the gas concentration in a certain area exceeds a safety threshold, all rubber-tyred trackless vehicles passing through the area can be automatically stopped, or the driving routes of the rubber-tyred trackless vehicles can be adjusted to avoid the dangerous area.

[0055] For example, if it is predicted that a certain lane is about to be congested or have an accident, a new route can be immediately specified for the rubber-tyred trackless vehicle leading to the lane.

[0056] S104, dispatching the rubber-tyred trackless vehicles based on the target control strategy.

[0057] Among them, the target control strategy can be scheduled for execution immediately after it is generated, or it can be executed after the next moment.

[0058] The embodiment of the present application proposes a method for dispatching a trackless rubber-tyred coal vehicle based on a spatiotemporal graph neural network, including: determining a spatiotemporal graph sequence corresponding to the current sliding time window, wherein the spatiotemporal graph sequence includes multiple spatiotemporal graphs, the spatiotemporal graph is composed of nodes and edges, the nodes include vehicle nodes corresponding to the trackless rubber-tyred vehicles in the coal mine and environmental nodes corresponding to the environment in the coal mine, the edges represent the relationship between the nodes, each node carries node attribute information, and each edge corresponds to a weight coefficient; based on the spatiotemporal graph neural network generated by pre-training, feature extraction is performed on the spatiotemporal graph sequence, and the node state is predicted based on the extracted features to obtain the predicted node state at the next moment; the target control strategy of the trackless rubber-tyred vehicle is determined according to the node state at the next moment; and the trackless rubber-tyred vehicle is dispatched based on the target control strategy. The present application can predict potential dangerous factors by accurately predicting the state and environmental changes of vehicles in coal mines, so as to take corresponding measures in advance to ensure the safety of mine operations; it can dynamically adjust the transportation strategy, reduce unnecessary pauses and waiting, and improve the working efficiency of vehicles; it realizes the automation and intelligence of the coal mine transportation process, does not rely on human intervention, and reduces the risk of human error.

[0059] Figure 3 is a schematic diagram of an exemplary implementation of a method for dispatching a coal trackless rubber-tyred vehicle based on a spatiotemporal graph neural network shown in the present application, such as Figure 3 As shown, the coal trackless rubber-tyred vehicle scheduling method based on spatiotemporal graph neural network includes the following steps:

[0060] S301, determining multiple moments in the current sliding time window, and determining a space-time graph of each of the multiple moments.

[0061] The method for generating a space-time graph at any moment includes: collecting multimodal sensor data of a trackless rubber-tyred vehicle and its surrounding environment at any moment; obtaining a space-time graph at the last moment of the any moment, and updating the space-time graph at the last moment based on the multimodal sensor data to obtain the space-time graph at the any moment. The any moment here refers to the moment when the space-time graph needs to be generated, and the last moment here refers to the moment when the last space-time graph is obtained.

[0062] Specifically, when collecting multimodal sensor data of a rubber-tyred trackless vehicle and its surrounding environment, the initial sensor data of the rubber-tyred trackless vehicle and its surrounding environment can be first collected based on sensors installed on the rubber-tyred trackless vehicle (such as cameras, laser radars and other sensors); then the initial sensor data is preprocessed to obtain multimodal sensor data obtained after preprocessing. Among them, preprocessing at least includes processing methods such as data cleaning, data normalization and data standardization.

[0063] Exemplarily, assuming that a space-time graph is generated every second, taking the current time as 8:00:00 as an example, the initial sensor data of the rubber-tyred trackless vehicle and the surrounding environment of the rubber-tyred trackless vehicle at 8:00:00 is collected based on the sensors installed on the trackless rubber-tyred vehicle; then the initial sensor data is preprocessed by data cleaning, data normalization and data standardization to obtain the multimodal sensor data obtained after the preprocessing, and then the space-time graph of the previous time of 8:00:00 is determined, that is, the space-time graph of 7:59:59; finally, the space-time graph of 7:59:59 is updated based on the multimodal sensor data corresponding to 8:00:00 to obtain the space-time graph corresponding to 8:00:00.

[0064] S302, arranging the space-time graph at each moment in chronological order to obtain a space-time graph sequence.

[0065] Arrange the space-time graphs of each moment in the current sliding time window in chronological order to obtain a space-time graph sequence.

[0066] Among them, the space-time graph sequence includes multiple space-time graphs, which are composed of nodes and edges. The nodes include vehicle nodes corresponding to the rubber-tyred trackless vehicles in the coal mine and environment nodes corresponding to the environment in the coal mine. The edges represent the relationship between the nodes. Each node carries node attribute information, and each edge corresponds to a weight coefficient.

[0067] S303, based on the spatiotemporal graph neural network generated by pre-training, feature extraction is performed on the spatiotemporal graph sequence, and the node state is predicted based on the extracted features to obtain the predicted node state at the next moment.

[0068] In this application, the space-time graph sequence corresponding to the current sliding time window is input into the pre-trained space-time graph neural network, and then based on the pre-trained space-time graph neural network, the features of the nodes and edges in the input space-time graph sequence are extracted, the interactions between the nodes and the spatiotemporal dynamic changes are captured, the nodes are predicted, and the predicted node state at the next moment is obtained (such as the vehicle's position, speed, fault state, changes in the mine environment, etc.).

[0069] S304, obtaining the corresponding dynamic parameters in the coal mine at the next moment.

[0070] Since the node state obtained above is the predicted node state at the next moment, in order to make the state input of the subsequent input reinforcement learning model belong to the same moment, it is not difficult to understand that it is necessary to obtain the corresponding dynamic parameters of the coal mine underground at the next moment based on the prediction.

[0071] Among them, dynamic parameters refer to some parameters that may change in coal mines and will affect the scheduling decisions of rubber-tyred trackless vehicles, such as personnel behavior (such as production personnel, maintenance personnel behavior, etc.), environmental changes (such as ventilation conditions, dust concentration, temperature, humidity), emergencies (such as fire, flood, gas explosion, etc.), etc.

[0072] S305, input the node state and dynamic parameters into the pre-trained reinforcement learning model to obtain the target control strategy at the next moment output by the reinforcement learning model.

[0073] The node state and dynamic parameters at the next moment are used as the state input of the agent in the reinforcement learning model, and combined with the preset reward function, the expected reward value of each candidate control strategy at the next moment is calculated; the candidate control strategy with the largest expected reward value is selected as the target control strategy at the next moment. It can be expressed as:

[0074] π=argmax a∈A Q(H t ,a;θ)

[0075] In the above formula, π is the selected target control strategy; A is the set of all possible candidate control strategies; Q(H t ,a;θ) is in state H t is the expected reward value for taking action a; θ is the parameter of the reinforcement learning model.

[0076] Among them, the design of the reward function is mainly to guide the vehicle to move towards the target state and avoid dangerous situations such as collisions.

[0077] The design method of the reward function includes: obtaining the safety reward parameter, efficiency reward parameter, energy consumption reward parameter and task completion reward parameter of the rubber-tyred trackless vehicle; and performing weighted summation on the safety reward parameter, efficiency reward parameter, energy consumption reward parameter and task completion reward parameter to obtain the reward function.

[0078] The formula for the reward function is:

[0079] w1*reward_safety+w2*reward_efficiency+w3*reward_energy+w4*reward_task

[0080] In the above formula, reward_safety represents the safety reward parameter, reward_efficiency represents the efficiency reward parameter, reward_energy represents the energy consumption reward parameter, reward_task represents the task completion reward parameter, and w1~w4 are the weight coefficients corresponding to the safety reward parameter, efficiency reward parameter, energy consumption reward parameter and task completion reward parameter respectively.

[0081] Among them, the safety reward parameters are: a positive reward is given if the vehicle does not collide or have an accident within the predetermined time; a large negative reward is given when the vehicle collides with an obstacle, other vehicles, wall, etc.; a negative reward is given when the vehicle leaves the specified area (such as a lane or work area).

[0082] Among them, the efficiency reward parameter: a positive reward is given when the vehicle reaches the target point, and the size of the reward can be adjusted according to the urgency of the task. The shorter the path, the higher the reward; the shorter the time to complete the task, the higher the reward.

[0083] Among them, energy consumption reward parameter: the less energy the vehicle consumes, the higher the reward.

[0084] Among them, the task completion reward parameters are: the vehicle will be given a positive reward if it successfully completes the task, and a negative reward if it fails to complete the task.

[0085] S306, dispatching the rubber-tyred trackless vehicles based on the target control strategy.

[0086] Since the above determination is for the target control strategy of the next moment, at the next moment, the rubber-tyred trackless vehicles are dispatched based on the target control strategy.

[0087] This application can predict potential dangerous factors by accurately predicting the status and environmental changes of vehicles underground in coal mines, so as to take corresponding measures in advance to ensure the safety of mine operations; it can dynamically adjust transportation strategies, reduce unnecessary pauses and waiting, and improve the work efficiency of vehicles; combined with reinforcement learning models, it realizes the automation and intelligence of the coal mine transportation process, does not rely on human intervention, and reduces the risk of human error.

[0088] Figure 4 is a schematic diagram of a coal trackless rubber-tyred vehicle dispatching device based on a spatiotemporal graph neural network shown in the present application, such as Figure 4 As shown, the coal trackless rubber-tyred vehicle scheduling device 400 based on the spatiotemporal graph neural network includes a first determination module 401, a state prediction module 402, a second determination module 403 and a vehicle scheduling module 404, wherein:

[0089] The first determination module 401 is used to determine the space-time graph sequence corresponding to the current sliding time window, wherein the space-time graph sequence includes multiple space-time graphs, and the space-time graph is composed of nodes and edges. The nodes include vehicle nodes corresponding to the rubber-tyred trackless vehicle in the coal mine and environment nodes corresponding to the environment in the coal mine. The edges represent the relationship between the nodes, each node carries node attribute information, and each edge corresponds to a weight coefficient.

[0090] The state prediction module 402 is used to extract features of the space-time graph sequence based on the pre-trained space-time graph neural network, and predict the state of the node based on the extracted features to obtain the predicted node state at the next moment.

[0091] The second determination module 403 is used to determine the target control strategy of the rubber-tyred trackless vehicle according to the node state at the next moment.

[0092] The vehicle dispatching module 404 is used to dispatch the rubber-tyred trackless vehicles based on the target control strategy.

[0093] This device can accurately predict the status of vehicles and environmental changes in coal mines, and can predict potential dangerous factors, so as to take corresponding measures in advance to ensure the safety of mine operations; it can dynamically adjust transportation strategies, reduce unnecessary pauses and waiting, and improve the work efficiency of vehicles; it realizes the automation and intelligence of the coal mine transportation process, does not rely on human intervention, and reduces the risk of human error.

[0094] Furthermore, the second determination module 403 is also used to: obtain the corresponding dynamic parameters in the coal mine at the next moment; input the node state and dynamic parameters into the pre-trained reinforcement learning model to obtain the target control strategy at the next moment output by the reinforcement learning model.

[0095] Furthermore, the second determination module 403 is also used to: use the node state and dynamic parameters at the next moment as the state input of the intelligent agent in the reinforcement learning model, and combine with the preset reward function to calculate the expected reward value of each candidate control strategy at the next moment; select the candidate control strategy with the largest expected reward value as the target control strategy at the next moment.

[0096] Furthermore, the first determination module 401 is also used to: determine multiple moments in the current sliding time window; determine the space-time graph of each moment in the multiple moments; and arrange the space-time graph of each moment in chronological order to obtain a space-time graph sequence.

[0097] Furthermore, the first determination module 401 is also used to: collect multimodal sensor data of the rubber-tyred trackless vehicle and its surrounding environment at any moment; obtain the space-time diagram of the previous moment at any moment, and update the space-time diagram of the previous moment based on the multimodal sensor data to obtain the space-time diagram of any moment.

[0098] Furthermore, the first determination module 401 is also used to: collect initial sensor data of the trackless rubber-tyred vehicle and the surrounding environment of the trackless rubber-tyred vehicle based on sensors installed on the trackless rubber-tyred vehicle; preprocess the initial sensor data to obtain multimodal sensor data obtained after preprocessing, and the preprocessing at least includes data cleaning, data normalization and data standardization.

[0099] Furthermore, the second determination module 403 is also used to: obtain the safety reward parameter, efficiency reward parameter, energy consumption reward parameter and task completion reward parameter of the rubber-tyred trackless vehicle; and perform weighted summation of the safety reward parameter, efficiency reward parameter, energy consumption reward parameter and task completion reward parameter to obtain a reward function.

[0100] In order to implement the above embodiment, the present application embodiment also proposes an electronic device 500, such as Figure 5 As shown, the electronic device 500 includes: a processor 501 and a memory 502 communicatively connected to the processor, the memory 502 stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor 501 to implement the coal trackless rubber-tyred vehicle scheduling method based on the spatiotemporal graph neural network as shown in the above embodiment.

[0101] In order to implement the above-mentioned embodiments, the embodiments of the present application also propose a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to implement the coal trackless rubber-tyred vehicle scheduling method based on the spatiotemporal graph neural network as shown in the above-mentioned embodiments.

[0102] In order to implement the above embodiments, the embodiments of the present application also propose a computer program product, including a computer program, which, when executed by a processor, implements the coal trackless rubber-tyred vehicle scheduling method based on the spatiotemporal graph neural network as shown in the above embodiments.

[0103] In the description of the present application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.

[0104] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0105] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0106] It should be noted that in the embodiments of the present application, certain software, components, models and other existing solutions in the industry may be mentioned, and they should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0107] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for dispatching coal trackless rubber-tyred vehicles based on spatiotemporal graph neural network, characterized in that: include: Determine a space-time graph sequence corresponding to the current sliding time window, wherein the space-time graph sequence includes multiple space-time graphs, the space-time graph is composed of nodes and edges, the nodes include vehicle nodes corresponding to the rubber-tyred trackless vehicle in the coal mine and environment nodes corresponding to the environment in the coal mine, the edges represent the relationship between the nodes, each of the nodes carries node attribute information, and each of the edges corresponds to a weight coefficient; Based on the spatiotemporal graph neural network generated by pre-training, feature extraction is performed on the spatiotemporal graph sequence, and the state of the node is predicted based on the extracted features to obtain the predicted node state at the next moment; Determining a target control strategy of the rubber-tyred trackless vehicle according to the node state at the next moment; The rubber-tyred trackless vehicle is dispatched based on the target control strategy.

2. The method according to claim 1, characterized in that Determining the target control strategy of the rubber-tyred trackless vehicle according to the node state at the next moment includes: Obtaining the corresponding dynamic parameters of the coal mine underground at the next moment; The node state and the dynamic parameters are input into a pre-trained reinforcement learning model to obtain a target control strategy at the next moment output by the reinforcement learning model.

3. The method according to claim 2, characterized in that The step of inputting the node state and the dynamic parameter into a pre-trained reinforcement learning model to obtain a target control strategy at the next moment output by the reinforcement learning model includes: Using the node state and the dynamic parameters at the next moment as the state input of the agent in the reinforcement learning model, and combining with a preset reward function, calculating the expected reward value of each candidate control strategy at the next moment; The candidate control strategy with the largest expected reward value is selected as the target control strategy at the next moment.

4. The method according to any one of claims 1 to 3, characterized in that The step of determining the space-time graph sequence corresponding to the current sliding time window includes: Determining multiple moments within the current sliding time window; determining a space-time graph for each of the plurality of moments in time; The space-time diagrams at each moment are arranged in chronological order to obtain the space-time diagram sequence.

5. The method according to claim 4, characterized in that For any moment, the method for generating the space-time graph includes: At any one of the above mentioned moments, multimodal sensor data of the rubber-tyred trackless vehicle and the surrounding environment of the rubber-tyred trackless vehicle are collected; The space-time diagram at the last moment of any moment is obtained, and the space-time diagram at the last moment is updated based on the multimodal sensor data to obtain the space-time diagram at any moment.

6. The method according to claim 5, characterized in that The multimodal sensor data of the rubber-tyred trackless vehicle and the surrounding environment of the rubber-tyred trackless vehicle are collected, including: Collecting initial sensor data of the rubber-tyred trackless vehicle and the surrounding environment of the rubber-tyred trackless vehicle based on sensors installed on the trackless rubber-tyred vehicle; The initial sensor data is preprocessed to obtain multimodal sensor data obtained after the preprocessing, wherein the preprocessing at least includes data cleaning, data normalization and data standardization.

7. The method according to claim 3, characterized in that The design method of the reward function includes: Obtaining a safety reward parameter, an efficiency reward parameter, an energy consumption reward parameter, and a task completion reward parameter of the rubber-tyred trackless vehicle; The reward function is obtained by performing a weighted summation on the safety reward parameter, the efficiency reward parameter, the energy consumption reward parameter and the task completion reward parameter.

8. A coal trackless rubber-tyred vehicle dispatching device based on spatiotemporal graph neural network, characterized in that: include: A first determination module is used to determine a spatiotemporal graph sequence corresponding to the current sliding time window, wherein the spatiotemporal graph sequence includes a plurality of spatiotemporal graphs, the spatiotemporal graph is composed of nodes and edges, the nodes include vehicle nodes corresponding to the rubber-tyred trackless vehicle in the coal mine and environment nodes corresponding to the environment in the coal mine, the edges represent the relationship between the nodes, each of the nodes carries node attribute information, and each of the edges corresponds to a weight coefficient; A state prediction module is used to extract features of the space-time graph sequence based on the pre-trained space-time graph neural network, and to predict the state of the node based on the extracted features, and obtain the predicted node state at the next moment; A second determination module is used to determine the target control strategy of the rubber-tyred trackless vehicle according to the node state at the next moment; The vehicle dispatching module is used to dispatch the rubber-tyred trackless vehicle based on the target control strategy.

9. An electronic device, comprising: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-7.

Images