Material dispatching path planning method, electronic equipment and storage medium

Abstract

The invention provides a material scheduling path planning method, electronic equipment and a storage medium, and the method comprises the steps: carrying out the analysis of real-time multi-source data, carrying out the prediction of a demand prediction result in a period of time in the future, and carrying out the planning of a material scheduling path based on the demand prediction results in a plurality of periods of time in the future. And generating a material scheduling path from each demand point to the target supply point. By analyzing the multi-source data, the accuracy of demand prediction can be improved, and more efficient material distribution and scheduling can be realized by generating the scheduling path from the demand point to the target supply point according to the demand prediction results in the plurality of time periods in the future of the current time period. In addition, rolling demand prediction and path generation are carried out in each time period, and timely response to complex demand fluctuation can be realized.

Description

Technical Field

[0001] The present application relates to the field of emergency dispatch technology, and in particular to a material dispatch path planning method, electronic equipment and storage medium. Background Art

[0002] When an emergency occurs, how to quickly and accurately predict the material needs of each demand point and dispatch materials based on the material needs of each demand point and the material inventory of each supply point becomes a problem that needs to be solved.

[0003] At present, the scheduling model is generally used to analyze the material demand data at the demand point and the material inventory data at the supply point at the beginning, so as to generate an overall material scheduling plan.

[0004] However, the existing scheduling model cannot respond and adjust to dynamic demand in a timely manner, and it is difficult to predict the demand of each demand point in the future, resulting in poor flexibility and low efficiency. Summary of the invention

[0005] The purpose of this application is to provide a material scheduling path planning method, electronic device and storage medium to address the deficiencies in the above-mentioned prior art, so as to solve the problems that the scheduling model in the prior art cannot respond and adjust to dynamic demand in a timely manner, and it is difficult to predict the demand for each demand point in the future, resulting in poor flexibility and low efficiency.

[0006] To achieve the above objectives, the technical solutions adopted in this application are as follows:

[0007] In a first aspect, the present application provides a material dispatch path planning method, the method comprising:

[0008] Constructing a real-time graph structure according to actual data and traffic data of the current time period, wherein the real-time graph structure includes a node set and an edge set, wherein the nodes in the node set include: demand points, supply points or transfer points, and the edges in the edge set are used to represent traffic connection information between nodes;

[0009] The real-time graph structure is input into a dynamic graph neural network, and the dynamic graph neural network predicts the demand forecast results of each demand point at multiple future moments in the current time period;

[0010] The demand forecast results and the real-time graph structure are input into a deep reinforcement learning model, and the deep reinforcement learning model predicts the target supply point of each demand point and the scheduling path between each demand point and the corresponding target supply point.

[0011] Optionally, constructing a real-time graph structure according to actual data of the current time period and traffic data includes:

[0012] Extracting at least one demand point, supply point and transfer point from the actual data, and using each of the demand point, supply point and transfer point as a node in the real-time graph structure;

[0013] Based on the location data of each of the demand points, supply points and transfer points, first attribute information of each of the nodes is determined, wherein the first attribute information includes: the population and material demand of each of the demand points, the material inventory of each of the supply points, available vehicles and available personnel;

[0014] Extracting connection paths between the demand points, supply points and transfer points from the traffic data, and generating corresponding edges in the real-time graph structure according to the connection paths;

[0015] The second attribute information of each edge is determined based on the road section condition and road travel time of each connected path, and the second attribute information includes: road travel time, road section condition, and road distance.

[0016] Optionally, the real-time graph structure is input into a dynamic graph neural network, and the dynamic graph neural network predicts demand forecast results of each demand point at multiple future moments in the current time period, including:

[0017] The real-time graph structure is input into the dynamic graph neural network, and the feature extraction layer of the dynamic graph neural network extracts features of the target demand points in the real-time graph structure to obtain node features of each target demand point, and each node feature is input into the convolution layer;

[0018] The convolution layer of the dynamic graph neural network performs time series convolution processing on the node features of each target demand point to obtain a dynamic graph sequence, and inputs the dynamic graph sequence into the output layer;

[0019] The output layer of the dynamic graph neural network predicts the dynamic graph sequence to obtain demand forecast results for each demand point in multiple time periods in the future.

[0020] Optionally, the predicting of the target supply point of each demand point and the scheduling path between each demand point and the corresponding target supply point by the deep reinforcement learning model includes:

[0021] Constructing a state space and an action space according to the real-time graph structure, wherein the state space includes: demand forecast results of nodes corresponding to the demand points, material inventory data of the supply points, and traffic connection information of each edge, and the action space includes: optional paths from the supply points to each of the demand points;

[0022] The deep reinforcement learning model makes predictions based on the current state of each demand point, the state space, and the action space to obtain the target supply point of each demand point and the scheduling path between each demand point and the corresponding target supply point.

[0023] Optionally, the deep reinforcement learning model predicts according to the current state of each demand point, the state space, and the action space to obtain the target supply point of each demand point and the scheduling path between each demand point and the corresponding target supply point, including:

[0024] A. Determine, by the deep reinforcement learning model, a first optional node connected to the starting point according to the current state of the starting point and the action space, where the starting point is any demand point in the real-time graph structure;

[0025] B. Recording the first optional node, and determining whether the first optional node is a target supply point;

[0026] C. If yes, take the first optional node as the target supply point and generate a scheduling path from the starting point to the first optional node; if no, determine a second optional node connected to the first optional node according to the current state of the first optional node and the action space;

[0027] D. Use the second optional node as the new first optional node, determine the current state of the first optional node according to the state space, and repeat steps BC until the first optional node is the target supply point, use the first optional node as the target supply point, and generate a scheduling path from the starting point to the target supply point.

[0028] Optionally, determining, by the deep reinforcement learning model according to a current state of the starting point and the action space, a first optional node connected to the starting point includes:

[0029] Determining, by the deep reinforcement learning model, at least one connected node connected to the starting point according to a current state of the starting point and the action space;

[0030] The first optional node is determined from the at least one connected node according to the reward value of each connected node.

[0031] Optionally, the determining whether the first optional node is a target supply point includes:

[0032] The node attribute of the first optional node is obtained. If the node attribute indicates that the first optional node is a supply point, and the material inventory of the first optional node and the material demand of the starting point meet the preset conditions, the first optional point is used as the target supply point.

[0033] Optionally, generating a scheduling path from the starting point to the target supply point includes:

[0034] Starting from the starting point, the first optional nodes are arranged in the order in which they are recorded, and the target supply point is taken as the last node to obtain a scheduling path from the starting point to the target supply point.

[0035] In the second aspect, an embodiment of the present application also provides an electronic device, comprising: a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium through the bus, and the processor executes the machine-readable instructions to perform the steps of a material scheduling path planning method as described in any one of the first aspects.

[0036] In a third aspect, an embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of a material scheduling path planning method as described in any one of the first aspects are executed.

[0037] The beneficial effects of the present application are as follows: by constructing a real-time graph structure for demand points, supply points, and transfer points, the spatiotemporal relationship between various places can be better characterized. By predicting the real-time graph structure through a dynamic graph neural network, the demand forecast results of each demand point are obtained, and a dynamic evaluation of the material demand of each demand point in the future can be achieved, which helps to optimize the matching of supply and demand. The demand forecast results and the real-time graph structure are predicted by a deep reinforcement learning model to obtain a scheduling path. Even if the road conditions change, the scheduling path can be updated based on the response to environmental changes, thereby ensuring the efficiency and timeliness of the arrival of materials at the demand points. In addition, the present application performs rolling demand forecasting and path generation within each time period, which can achieve timely response and adjustment to dynamic demand, and improve the flexibility of scheduling path planning and the efficiency of material scheduling.

[0038] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 A flow chart of a material dispatch path planning method provided in an embodiment of the present application is shown;

[0041] Figure 2 An example diagram of a real-time graph structure provided by an embodiment of the present application is shown;

[0042] Figure 3 A flowchart of constructing a real-time graph structure provided by an embodiment of the present application is shown;

[0043] Figure 4 A flow chart for generating demand forecast results provided by an embodiment of the present application is shown;

[0044] Figure 5 A flow chart of generating a scheduling path provided by an embodiment of the present application is shown;

[0045] Figure 6 A flowchart of another method of generating a scheduling path provided in an embodiment of the present application is shown;

[0046] Figure 7 A flowchart for determining a first optional node provided by an embodiment of the present application is shown;

[0047] Figure 8 A schematic diagram of the structure of a material dispatching path planning device provided in an embodiment of the present application is shown;

[0048] Fig. 9 A schematic structural diagram of an electronic device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0049] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0050] It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the existence of the features declared thereafter, but does not exclude the addition of other features.

[0051] After an emergency occurs, the existing scheduling models generally collect material demand data from each demand point and material inventory data from each supply point, and after data analysis, generate material scheduling plans and material scheduling paths from the supply point to the demand point.

[0052] However, existing scheduling models generally perform static analysis on a single data source. After an emergency occurs, the material demand at the demand point and the material inventory at the supply point are not static. Therefore, the static solution based on existing technology cannot respond to and adjust to dynamic demand in a timely manner, which will lead to unreasonable resource allocation and difficulty in coping with complex demand fluctuations.

[0053] Based on this, the present application proposes a material dispatch path planning method, which predicts the demand forecast results for a period of time in the future by analyzing real-time multi-source data, and generates material dispatch paths from each demand point to the target supply point based on the demand forecast results for multiple time periods in the future. By analyzing multi-source data, the accuracy of demand forecasts can be improved, and by generating dispatch paths from demand points to target supply points based on demand forecast results for multiple time periods in the future of the current time period, more efficient material distribution and dispatch can be achieved. In addition, the present application performs rolling demand forecasts and path generation within each time period, and can also achieve timely response to complex demand fluctuations.

[0054] First, the application scenario of the present application is introduced. When an emergency occurs in area A, the various places in area A can be divided into demand points, supply points and transfer points according to their demand for materials and material inventory. Area A includes multiple data acquisition devices, such as cameras and sensors arranged in places and traffic arteries. When it is necessary to dispatch materials to area A, the data of each place in area A and the data of each traffic artery can be obtained through the data acquisition device. The electronic equipment analyzes and processes the acquired data based on the method of the present application, and generates a dispatch path between each demand point and the target supply point in area A, so that rescue personnel can dispatch materials according to the dispatch path.

[0055] Next, combine Figure 1 , the material dispatch path planning method of the present application is described, and the execution subject of the method can be an electronic device, such as Figure 1 As shown, the method includes:

[0056] S101. Construct a real-time graph structure based on actual data of the current time period and traffic data. The real-time graph structure includes a node set and an edge set. The nodes in the node set include: demand points, supply points or transfer points. The edges in the edge set are used to represent traffic connection information between nodes.

[0057] The actual data of the current time period includes the actual data of each demand point, supply point and transfer point. The demand point is the place where materials need to be transferred in, the supply point is the place where materials can be supplied, and the transfer point is the place where materials can be stopped when dispatching materials. The actual data of the demand point includes the population, material demand type, material demand intensity and material demand quantity. The actual data of the supply point includes the material inventory, the number of available vehicles and the number of available personnel.

[0058] Traffic data includes the speed, path length, congestion, and path traffic status of the main traffic routes between various places in the area. The path traffic status is used to indicate whether the path is closed.

[0059] In this application, a rolling forecast can be made on the demand of each demand point based on actual data. In each rolling cycle, the latest real-time graph structure is generated according to the latest actual data. The real-time graph structure is used to characterize the real-time status of each demand point, supply point, transfer point and traffic artery in the region.

[0060] like Figure 2 As shown in the figure, it is an example of a real-time graph structure. The graph includes multiple nodes, which represent demand points, supply points, and transfer points. The edges in the graph are used to represent the traffic connection information between the nodes. The traffic connection information includes: road travel time and road section conditions, etc. Among them, Xn represents the demand point, Gn represents the supply point, and Zn represents the transfer point.

[0061] S102: Input the real-time graph structure into the dynamic graph neural network, and use the dynamic graph neural network to predict the demand forecast results of each demand point in multiple future time periods during the current time period.

[0062] Dynamic Graph Neural Network (D-GNN) can model the spatiotemporal dependencies between demand points in the real-time graph structure, thereby predicting the demand of various places in the region in the current time period and obtaining the demand distribution of each demand point in the future.

[0063] Among them, the demand forecast results include: the type of materials and quantity of materials required by each demand point in each time period.

[0064] Exemplarily, after acquiring actual data at time T1 at the beginning of the current time period and constructing a real-time graph structure, the real-time graph structure can be input into a dynamic graph neural network, which can predict demand forecast results for demand points at multiple future times after time T1 within the current time period based on the real-time graph structure.

[0065] S103. Input each demand prediction result and the real-time graph structure into the deep reinforcement learning model, and use the deep reinforcement learning model to predict the target supply point of each demand point and the scheduling path between each demand point and the corresponding target supply point.

[0066] When there are multiple supply points, the Deep Reinforcement Learning (DRL) model can determine the target supply point corresponding to each demand point based on the input demand forecast results and real-time graph structure, and generate a scheduling path from the target supply point to the corresponding demand point.

[0067] When the number of supply points is one, the deep reinforcement learning model can generate a scheduling path from the supply point to each demand point based on the input demand forecast results and the real-time graph structure.

[0068] The deep reinforcement learning model can determine at least one optional supply point that can provide materials to the current demand point based on the demand forecast results of the current demand point and the inventory of the supply point, and based on the traffic connection information between each optional supply point and the current demand point, based on a predefined reward function, determine the target supply point from multiple optional supply points, and simultaneously generate a scheduling path from the current demand point to the target supply point.

[0069] Among them, the reward function can be determined based on the demand satisfaction degree, transportation cost and transportation time of each demand point, ensuring that the model gives priority to meeting the material needs of high-demand points and determines the optimal scheduling path to reduce transportation delays.

[0070] In the embodiment of the present application, by constructing a real-time graph structure for demand points, supply points, and transfer points, the spatiotemporal relationship between various places can be better characterized. By predicting the real-time graph structure through a dynamic graph neural network, the demand forecast results of each demand point are obtained, and a dynamic evaluation of the material demand of each demand point in the future can be achieved, which helps to optimize the matching of supply and demand. The demand forecast results and the real-time graph structure are predicted by a deep reinforcement learning model to obtain a scheduling path. Even if the road conditions change, the scheduling path can be updated based on the response to environmental changes, thereby ensuring the efficiency and timeliness of the arrival of materials at the demand points. In addition, the present application performs rolling demand forecasting and path generation in each time period, which can achieve timely response and adjustment to dynamic demand, and improve the flexibility of scheduling path planning and the efficiency of material scheduling.

[0071] The following is a further explanation of the above-mentioned real-time graph structure constructed based on the actual data of the current time period and traffic data. Figure 3 As shown, the above step S101 includes:

[0072] S301. Extract at least one demand point, supply point and transfer point from actual data, and use each demand point, supply point and transfer point as a node in a real-time graph structure.

[0073] S302. Based on the location data of each demand point, supply point and transfer point, determine the first attribute information of each node, the first attribute information including: the population size and material demand of each demand point, the material inventory of each supply point, available vehicles and available personnel.

[0074] The actual data includes data of various places in the area where material dispatch is required, including demand points where materials need to be transferred in, supply points where materials are provided, and transfer points where materials pass through. The first attribute information of each node can be determined based on the actual data of each place.

[0075] The first attribute information of the demand point includes: the population of the demand point, the material demand of the demand point, and the material inventory of the demand point; the first attribute information of the supply point includes the material inventory, available vehicles, and available personnel of the supply point; the first attribute information of the transfer point may include available vehicles, available personnel, etc. For example, the demand point may be an emergency shelter, a hospital, etc., the supply point may be a warehouse, a rescue site, etc., and the transfer point may be a place for logistics diversion or a place for rest during the journey.

[0076] S303: extracting connection paths between demand points, supply points and transfer points from traffic data, and generating corresponding edges in a real-time graph structure according to each connection path.

[0077] S304. Determine the second attribute information of each edge based on the road section conditions and road travel time of each connected path, the second attribute information including: road travel time, road section conditions, and road distance.

[0078] Traffic data can be obtained through cameras and sensors on traffic arteries. The connection paths between places can be obtained from traffic data. If there is a connection path between place A and place B, an edge between the node of place A and the node of place B can be generated in the real-time graph structure, and the road travel time, road section condition and road distance of each connection path are used as the second attribute information of the edge corresponding to each connection path.

[0079] In the embodiment of the present application, by constructing a real-time graph structure based on actual data and traffic data, the demand changes in various places in the area and the dynamic state of the traffic network can be captured, providing accurate input for subsequent demand forecasting and scheduling decisions.

[0080] The following is a further explanation of the above-mentioned input of the real-time graph structure into the dynamic graph neural network, and the prediction of the demand forecast results of each demand point in multiple time periods in the future by the dynamic graph neural network. Figure 4 As shown, the above step S102 includes:

[0081] S401. Input the real-time graph structure into the dynamic graph neural network. The feature extraction layer of the dynamic graph neural network extracts features of the target demand points in the real-time graph structure to obtain node features of each target demand point, and inputs each node feature into the convolution layer.

[0082] The node feature can be a sequence that characterizes the demand intensity of the node, the node location, and the traffic conditions between the nodes. The dynamic graph neural network can extract the features of the target demand points in the dynamic graph neural network to obtain the node features of each target demand point. The target demand point can be any demand point in the dynamic graph neural network.

[0083] S402: The convolution layer of the dynamic graph neural network performs time series convolution processing on the node features of each target demand point to obtain a dynamic graph sequence, and inputs the dynamic graph sequence into the output layer.

[0084] Optionally, the convolution layer can perform a time series convolution operation on the node features, embed the time series data of each demand point into the node features, and obtain a dynamic graph sequence. The dynamic graph sequence can characterize the change law of the node features in the time dimension. The convolution layer performs time series convolution processing as shown in the following formula (1).

[0085]

[0086] in, is the state of node i at time t, W is the graph convolution weight, and b is the bias. Through multi-layer convolution operations, nodes can perceive changes in demand in neighboring areas.

[0087] S403, the output layer of the dynamic graph neural network predicts the dynamic graph sequence to obtain demand prediction results for each demand point in multiple time periods in the future.

[0088] The output layer can predict the dynamic graph sequence to obtain the demand forecast results of each demand point in multiple future time periods in the current time period, and use the forecast results as input for subsequent scheduling path optimization.

[0089] The following is a further explanation of the target supply point of each demand point predicted by the deep reinforcement learning model and the scheduling path between each demand point and the corresponding target supply point. Figure 5 As shown, the above step S103 includes:

[0090] S501. Construct a state space and an action space according to the real-time graph structure. The state space includes: demand forecast results of demand points, material inventory data of supply points, and traffic connection information of each edge. The action space includes: optional paths from supply points to each demand point.

[0091] Optionally, a state space and an action space can be pre-built based on the real-time graph structure. The state space includes the demand forecast results of the demand point, the material inventory data of the supply point, and the traffic connection information of the path represented by each edge, which is used to determine the current state of each node. The action space is used to represent all optional paths from the supply point to each demand point.

[0092] It should be noted that a reward function may be predefined in the present application, and the definition of the reward function is shown in the following formula (2).

[0093]

[0094] Among them, d j represents the demand, r j Indicates the amount of materials allocated, T ij is the travel time, and α is the dispatch cost weight. The reward function can ensure that high-demand areas are given priority and unnecessary transportation costs are reduced.

[0095] S502. The deep reinforcement learning model makes predictions based on the current state, state space, and action space of each demand point to obtain the target supply point of each demand point and the scheduling path between each demand point and the corresponding target supply point.

[0096] The target supply point of the demand point can be one or more, and the target supply point is a supply point that can meet the material demand of the demand point. The current state of the demand point can include the demand forecast result of the demand point and the traffic connection information of each side.

[0097] In the first implementation, the deep reinforcement learning model can start path planning from the supply point, and based on the reward function of the above formula (2), the scheduling path from the supply point to each demand point is planned, where the demand point with more urgent demand has a larger reward value calculated based on the above reward function, so it can be planned first. After the deep reinforcement model completes the path planning from the supply point to all demand points, the scheduling path between each demand point and the supply point can be output, and materials can be scheduled according to the scheduling path.

[0098] In the second implementation method, the deep reinforcement learning model can start path planning from the demand point. When selecting the demand point, based on the demand urgency and demand volume of each demand point, path planning can be started from the demand point with larger demand volume and higher demand urgency. The reward value of selecting each node is determined based on the above reward function, and nodes with larger reward values ​​are selected first until the currently selected node is the target supply point of the demand point, thereby obtaining the scheduling path between each demand point and the target supply point.

[0099] It should be noted that when there are multiple supply points in the area, the deep reinforcement model in the first implementation method can simultaneously plan paths for each supply point, or start path planning from the supply point closest to the high-demand area. In the process of deep reinforcement learning model planning paths from demand points, the target supply point of the demand point can be one or more, that is, one demand point can correspond to one or more target supply points, and the scheduling path corresponding to the demand point can also be one or more.

[0100] The following is a further explanation of the second implementation method in which the deep reinforcement learning model predicts the target supply point of each demand point and the scheduling path between each demand point and the corresponding target supply point according to the current state, state space and action space of each demand point. Figure 6 As shown, the above step S502 includes:

[0101] S601. The deep reinforcement learning model determines a first optional node connected to the starting point according to the current state of the starting point and the action space, where the starting point is any demand point in the real-time graph structure.

[0102] As a possible implementation method, each demand point can be prioritized according to the urgency of the demand. The priority is determined by the distribution of material demand of each demand point in the future, and the path planning is carried out for each demand point in order from high to low priority.

[0103] The current state of the starting point can be determined based on the state space. The current state of the starting point includes: the material demand of the demand point and the traffic connection information of the paths represented by the edges connected to the demand point. Therefore, based on the current state of the starting point, at least one node connected to the starting point can be determined as the first optional node. Among them, the first optional node can be a supply point, a transfer point or other demand point. The first optional node can be determined based on the above reward function. Specifically, the reward value can be calculated for each node connected to the starting point based on the above formula (2), and the node with the highest reward value is used as the first optional node.

[0104] S602: Record the first selectable node, and determine whether the first selectable node is a target supply point.

[0105] S603: If yes, take the first optional node as the target supply point and generate a scheduling path from the starting point to the first optional node; if no, determine the second optional node connected to the first optional node according to the current state and action space of the first optional node.

[0106] If the first optional node can meet the material demand of the starting point, the first optional node can be determined as the target supply point of the starting point, the first optional point is recorded as the target supply point, and the remaining amount of materials after the target supply point supplies materials to the starting point is recorded. After completing the scheduling path planning of the starting point, the next demand point of the starting point in the priority sorting can be selected as the new starting point, and the above S201 step is re-executed until all demand points have determined the target supply point.

[0107] If the first optional node is a transit point or other demand point, or the first optional node cannot meet the material demand of the starting point, the first optional node will not be used as the target supply point of the starting point, only the first optional node will be recorded, and the search for the next node of the first optional node will continue.

[0108] The second optional node may be a node having the highest reward value calculated based on the above formula (2) among the nodes connected to the first optional node.

[0109] S604. Use the second optional node as the new first optional node, determine the current state of the first optional node according to the state space, and repeat steps S602-S603 until the first optional node is the target supply point. Use the first optional node as the target supply point and generate a scheduling path from the starting point to the target supply point.

[0110] After determining the second optional node, the second optional node can be used as the new first optional node, the new first optional node can be recorded, and it can be determined whether the new first optional node is the target supply point of the starting point. If not, continue to determine the second optional node connected to the new first optional node until the new first optional node is the target supply point.

[0111] In the process of determining the target supply point, the information of each first optional node determined from the starting point is recorded in sequence. After the target supply point is determined, a scheduling path from the starting point to the target supply point can be generated based on all the recorded first optional nodes.

[0112] The following is a further explanation of the above-mentioned determination of the first optional node connected to the demand point by the deep reinforcement learning model according to the current state of the starting point and the action space, such as Figure 7 As shown, the above step S601 includes:

[0113] S701. Determine, by a deep reinforcement learning model, at least one connected node connected to the starting point according to the current state of the starting point and the action space.

[0114] The current state of the starting point may indicate the nodes connected to the starting point. Figure 2 , assuming the starting point is X3, the connected nodes connected to the starting point include Figure 2 Nodes G1, Z3 and X4 in.

[0115] S702: Determine a first optional node from at least one connected node according to the reward value of each connected node.

[0116] Optionally, the reward value of each connected node can be calculated based on the above formula (2), and the node with the highest reward value can be selected from the connected nodes as the first optional node.

[0117] The following is a further explanation of the above determination of whether the first optional node is a target supply point, specifically including:

[0118] The node attribute of the first optional node is obtained. If the node attribute indicates that the first optional node is a supply point, and the material inventory of the first optional node and the material demand of the starting point meet the preset conditions, the first optional point is used as the target supply point.

[0119] Among them, the node attributes of the supply point may include a preset supply identifier. If the node attributes of the first optional node include an attack identifier, it means that the first optional node is a supply point, and the material inventory data of the first optional node meets the material demand of the starting point, then the first optional point can be used as the target supply point.

[0120] The following is a further explanation of the above-mentioned generation of the scheduling path from the starting point to the target supply point, including:

[0121] Starting from the starting point, the first optional nodes are arranged in the order in which they are recorded, and the target supply point is taken as the last node to obtain a scheduling path from the starting point to the target supply point.

[0122] When recording the first optional node, the nodes can be arranged in sequence according to the order of the first optional node to obtain a node sequence of the first optional node, and then the starting point can be inserted into the first position of the node sequence, and the target supply point can be inserted into the last position of the node sequence, and the obtained node sequence can be used as the scheduling path from the starting point to the target supply point.

[0123] Based on the same inventive concept, a material scheduling path planning device corresponding to the material scheduling path planning method is also provided in the embodiment of the present application. Since the principle of solving the problem by the device in the embodiment of the present application is similar to the material scheduling path planning method mentioned above in the embodiment of the present application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0124] Figure 8 A schematic diagram of the structure of a material dispatching path planning device provided in an embodiment of the present application is shown.

[0125] A construction module 801 is used to construct a real-time graph structure according to actual data of the current time period and traffic data. The real-time graph structure includes a node set and an edge set. The nodes in the node set include: demand points, supply points or transfer points. The edges in the edge set are used to represent traffic connection information between nodes.

[0126] The demand forecasting module 802 is used to input the real-time graph structure into the dynamic graph neural network, and the dynamic graph neural network predicts the demand forecast results of each demand point at multiple future moments in the current time period;

[0127] The path prediction module 803 is used to input each demand prediction result and the real-time graph structure into the deep reinforcement learning model, and the deep reinforcement learning model predicts the target supply point of each demand point and the scheduling path between each demand point and the corresponding target supply point.

[0128] Optionally, the construction module 801 is specifically used for:

[0129] Extracting at least one demand point, supply point and transfer point from actual data, and using each demand point, supply point and transfer point as a node in a real-time graph structure;

[0130] Based on the location data of each demand point, supply point and transfer point, determine the first attribute information of each node, the first attribute information includes: the population and material demand of each demand point, the material inventory of each supply point, available vehicles and available personnel;

[0131] Extract the connection paths between each demand point, supply point and transfer point from the traffic data, and generate the corresponding edges in the real-time graph structure according to each connection path;

[0132] The second attribute information of each edge is determined based on the road section conditions and road travel time of each connected path, and the second attribute information includes: road travel time, road section conditions, and road distance.

[0133] Optionally, the demand forecasting module 802 is specifically used for:

[0134] The real-time graph structure is input into the dynamic graph neural network. The feature extraction layer of the dynamic graph neural network extracts features of the target demand points in the real-time graph structure to obtain the node features of each target demand point, and then inputs the features of each node into the convolution layer.

[0135] The convolution layer of the dynamic graph neural network performs time series convolution processing on the node features of each target demand point to obtain a dynamic graph sequence, and inputs the dynamic graph sequence into the output layer;

[0136] The output layer of the dynamic graph neural network predicts the dynamic graph sequence and obtains the demand forecast results of each demand point in multiple time periods in the future.

[0137] Optionally, the path prediction module 803 is specifically used for:

[0138] The state space and action space are constructed based on the real-time graph structure. The state space includes: the demand forecast results of the nodes corresponding to the demand points, the material inventory data of the supply points, and the traffic connection information of each edge. The action space includes: the optional paths from the supply points to each demand point.

[0139] The deep reinforcement learning model makes predictions based on the current state, state space, and action space of each demand point to obtain the target supply point of each demand point and the scheduling path between each demand point and the corresponding target supply point.

[0140] Optionally, the path prediction module 803 is specifically used for:

[0141] A. The deep reinforcement learning model determines the first optional node connected to the starting point according to the current state of the starting point and the action space, where the starting point is any demand point in the real-time graph structure;

[0142] B. Record the first optional node and determine whether the first optional node is a target supply point;

[0143] C. If yes, take the first optional node as the target supply point and generate a scheduling path from the starting point to the first optional node; if no, determine a second optional node connected to the first optional node according to the current state and action space of the first optional node;

[0144] D. Take the second optional node as the new first optional node, determine the current state of the first optional node according to the state space, and repeat steps BC until the first optional node is the target supply point, take the first optional node as the target supply point, and generate a scheduling path from the starting point to the target supply point.

[0145] Optionally, the path prediction module 803 is specifically used for:

[0146] The deep reinforcement learning model determines at least one connected node connected to the starting point according to the current state of the starting point and the action space;

[0147] A first optional node is determined from at least one connected node according to the reward value of each connected node.

[0148] Optionally, the path prediction module 803 is specifically used for:

[0149] The node attribute of the first optional node is obtained. If the node attribute indicates that the first optional node is a supply point, and the material inventory of the first optional node and the material demand of the starting point meet the preset conditions, the first optional point is used as the target supply point.

[0150] Optionally, the path prediction module 803 is specifically used for:

[0151] Starting from the starting point, the first optional nodes are arranged in the order in which they are recorded, and the target supply point is taken as the last node to obtain a scheduling path from the starting point to the target supply point.

[0152] The embodiments of the present application can better characterize the spatiotemporal relationship between various places by constructing a real-time graph structure for demand points, supply points, and transfer points. The real-time graph structure is predicted by a dynamic graph neural network to obtain the demand forecast results of each demand point, which can realize dynamic evaluation of the material demand of each demand point in the future, and help optimize the matching of supply and demand. The demand forecast results and the real-time graph structure are predicted by a deep reinforcement learning model to obtain a scheduling path. Even if the road conditions change, the scheduling path can be updated based on the response to environmental changes, thereby ensuring the efficiency and timeliness of the arrival of materials at the demand points. In addition, the present application performs rolling demand forecasting and path generation in each time period, which can realize timely response and adjustment to dynamic demand, and improve the flexibility of scheduling path planning and the efficiency of material scheduling.

[0153] Fig. 9 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application is shown, including: a processor 901, a storage medium 902 and a bus 903, wherein the storage medium 902 stores machine-readable instructions executable by the processor 901, and when the electronic device runs a material scheduling path planning method such as in the embodiment, the processor 901 communicates with the storage medium 902 through the bus 903, and the processor 901 executes the machine-readable instructions and the preamble of the method item of the processor 901 to execute the steps of the above-mentioned material scheduling path planning method.

[0154] An embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. The computer program is executed when a processor is running, and the processor executes the steps of the above-mentioned material scheduling path planning method.

[0155] In the embodiment of the present application, the computer program can also execute other machine-readable instructions when run by the processor to execute other methods described in the embodiment. For the specific execution method steps and principles, please refer to the description of the embodiment, which will not be repeated here.

[0156] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0157] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0158] In addition, each functional unit in the embodiments provided in the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0159] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0160] It should be noted that similar numbers and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description and are not to be understood as indicating or implying relative importance.

[0161] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application. They should all be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A material dispatching path planning method, characterized in that: include: Constructing a real-time graph structure according to actual data and traffic data of the current time period, wherein the real-time graph structure includes a node set and an edge set, wherein the nodes in the node set include: demand points, supply points or transfer points, and the edges in the edge set are used to represent traffic connection information between nodes; The real-time graph structure is input into a dynamic graph neural network, and the dynamic graph neural network is used to predict the demand forecast results of each demand point at multiple future moments in the current time period; The demand forecast results and the real-time graph structure are input into a deep reinforcement learning model, and the deep reinforcement learning model predicts the target supply point of each demand point and the scheduling path between each demand point and the corresponding target supply point.

2. The method according to claim 1, characterized in that The real-time graph structure is constructed based on the actual data of the current time period and the traffic data, including: Extracting at least one demand point, supply point and transfer point from the actual data, and using each of the demand point, supply point and transfer point as a node in the real-time graph structure; Based on the location data of each of the demand points, supply points and transfer points, first attribute information of each of the nodes is determined, wherein the first attribute information includes: the population and material demand of each of the demand points, the material inventory of each of the supply points, available vehicles and available personnel; Extracting connection paths between the demand points, supply points and transfer points from the traffic data, and generating corresponding edges in the real-time graph structure according to the connection paths; The second attribute information of each edge is determined based on the road section condition and road travel time of each connected path, and the second attribute information includes: road travel time, road section condition, and road distance.

3. The method according to claim 1, characterized in that The real-time graph structure is input into the dynamic graph neural network, and the dynamic graph neural network predicts the demand forecast results of each demand point at multiple future moments in the current time period, including: The real-time graph structure is input into the dynamic graph neural network, and the feature extraction layer of the dynamic graph neural network extracts features of the target demand points in the real-time graph structure to obtain node features of each target demand point, and each node feature is input into the convolution layer; The convolution layer of the dynamic graph neural network performs time series convolution processing on the node features of each target demand point to obtain a dynamic graph sequence, and inputs the dynamic graph sequence into the output layer; The output layer of the dynamic graph neural network predicts the dynamic graph sequence to obtain demand forecast results for each demand point in multiple time periods in the future.

4. The method according to claim 1, characterized in that: The target supply point of each demand point and the scheduling path between each demand point and the corresponding target supply point predicted by the deep reinforcement learning model include: Constructing a state space and an action space according to the real-time graph structure, wherein the state space includes: demand forecast results of nodes corresponding to the demand points, material inventory data of the supply points, and traffic connection information of each edge, and the action space includes: optional paths from the supply points to each of the demand points; The deep reinforcement learning model makes predictions based on the current state of each demand point, the state space, and the action space to obtain the target supply point of each demand point and the scheduling path between each demand point and the corresponding target supply point.

5. The method according to claim 4, characterized in that The deep reinforcement learning model predicts according to the current state of each demand point, the state space and the action space to obtain the target supply point of each demand point and the scheduling path between each demand point and the corresponding target supply point, including: A. Determine, by the deep reinforcement learning model, a first optional node connected to the starting point according to the current state of the starting point and the action space, where the starting point is any demand point in the real-time graph structure; B. Recording the first optional node, and determining whether the first optional node is a target supply point; C. If yes, take the first optional node as the target supply point and generate a scheduling path from the starting point to the first optional node; if no, determine a second optional node connected to the first optional node according to the current state of the first optional node and the action space; D. Use the second optional node as the new first optional node, determine the current state of the first optional node according to the state space, and repeat steps BC until the first optional node is the target supply point, use the first optional node as the target supply point, and generate a scheduling path from the starting point to the target supply point.

6. The method according to claim 5, characterized in that The step of determining, by the deep reinforcement learning model according to the current state of the starting point and the action space, a first optional node connected to the starting point includes: Determining, by the deep reinforcement learning model, at least one connected node connected to the starting point according to a current state of the starting point and the action space; The first optional node is determined from the at least one connected node according to the reward value of each connected node.

7. The method according to claim 5, characterized in that The determining whether the first selectable node is a target supply point includes: The node attribute of the first optional node is obtained. If the node attribute indicates that the first optional node is a supply point, and the material inventory of the first optional node and the material demand of the starting point meet the preset conditions, the first optional point is used as the target supply point.

8. The method according to claim 5, characterized in that The generating a scheduling path from the starting point to the target supply point includes: Starting from the starting point, the first optional nodes are arranged in the order in which they are recorded, and the target supply point is taken as the last node to obtain a scheduling path from the starting point to the target supply point.

9. An electronic device, characterized in that: include: A processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of a material scheduling path planning method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of a material scheduling path planning method as described in any one of claims 1 to 8 are executed.

Images