Dynamic path planning method and device based on time sequence convolutional network traffic prediction and medium

Abstract

The invention relates to a dynamic path planning method and device based on time sequence convolutional network traffic prediction, and a medium. The method comprises the following steps: S1, modeling a to-be-planned regional road network; s2, on the basis of historical traffic data, using a time sequence convolutional network and a gating circulation unit to predict future travel time of a planning area, and forming a dynamic weight table W; and S3, performing global shortest path planning by using an improved Dijkstra algorithm based on the regional road section travel time prediction made in S2. According to the dynamic path planning method for traffic prediction based on the time sequence convolutional network, the traffic prediction model fusing the time sequence convolutional network and the gating circulation unit is used for solving the path planning problem under the dynamic path weight, the efficient time sequence data processing capacity and parallel processing capacity are achieved, and the traffic prediction efficiency is improved. And meanwhile, full-time-domain optimal path planning is realized on the basis of an improved Dijkstra algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic path planning, and specifically to a dynamic path planning method, device and medium based on traffic prediction of a temporal convolutional network. Background Art

[0002] With the acceleration of urbanization and the aggravation of traffic congestion problems, it has become particularly crucial to use traffic prediction to guide path planning. The spatio-temporal characteristics of urban traffic flow are significant, that is, the traffic conditions will fluctuate dynamically with time and location, which makes it difficult for traditional static path planning methods to adapt to this dynamic change. By integrating real-time traffic data, considering the spatio-temporal dynamics of traffic flow, and predicting future traffic trends, we can avoid potential congestion in advance, achieve a more global and forward-looking path planning, thereby improving the accuracy and practicality of path planning. In addition, accurate path planning can reduce unnecessary trips and waiting times, which not only saves energy and time, but also has a positive impact on environmental protection and economic benefits. Therefore, integrating traffic prediction into path planning not only greatly improves the efficiency and effectiveness of urban traffic management, but also plays an increasingly important role in the development of intelligent transportation systems and smart cities.

[0003] The existing Chinese patent (CN202111170295.3) provides a lane-level path planning method and system based on a spatio-temporal traffic model. This method segments the road ahead and receives real-time traffic data, establishes a spatio-temporal traffic model to describe the average speed correlation between lanes, and transforms the path planning problem into a rolling optimization problem to solve the optimal lane-level path. However, its high demand for computing power, high real-time requirements, and the complexity of the model itself may affect the response speed and maintainability of the planning.

[0004] The existing Chinese patent (CN201910278787.0) provides an intelligent transportation refined path planning method based on a grid expansion model. This method uses a convolutional neural network to predict future traffic conditions, and then dynamically expands the grid composed of road units based on these prediction information, the current road network structure and vehicle density, and conducts refined path search within these grids. When the vehicle approaches the grid boundary, the system will re-predict the traffic conditions and plan the path until the vehicle reaches the destination. However, it depends on the accuracy of real-time traffic information and historical data, and in a large-scale road network, dynamically expanding the grid may become complex, affecting timeliness.

[0005] The existing Chinese patent (CN202310527421.9) provides a vehicle path planning method based on traffic flow speed prediction and signal light status. This method collects the historical average driving speed data of each path in the road network, then uses the mWDN-LSTM-ARIMA model to predict the average driving speed in the current time interval, and further predicts the travel time of each path. By combining real-time location information, traffic signal light status, and travel time prediction values, the PPO algorithm is used to train a deep reinforcement learning model to determine the best driving action for the vehicle from the current path to the next path. However, its model has high complexity in parameter selection, and at the same time, the parameter tuning process is complex and time-consuming.

[0006] The existing Chinese patent (CN202410807960.2) provides a vehicle route planning method and system based on the Dijkstra algorithm, which reduces operating costs and improves the passenger travel experience through automation technology. This method obtains the actual traffic route map and bus scheduling model, calculates the shortest path between each station, and uses the improved Dijkstra algorithm to find the path with the lowest cost. The system includes three processing modules: obtaining information, calculating the shortest path, and path planning and optimization. However, it lacks consideration of the future dynamic changes in the road network and cannot effectively achieve the global path optimization. Summary of the Invention

[0007] In order to solve the problems that high computing power requirements, strict real-time requirements, model complexity, dependence on high computing power, and model complexity affect the response speed of planning and the maintainability of the system, and the failure to fully consider the dynamic changes in the road network, resulting in the inability to effectively achieve global optimal path planning, etc., the present invention provides a dynamic path planning method based on traffic prediction of temporal convolutional network, including the following steps:

[0008] S1. Model the road network in the area to be planned, collect the historical traffic data of the area to be planned, and obtain the actual travel time series Y of each road section.

[0009] S2. Use the temporal convolutional network to process the sequence Y and output the feature sequence Y'; use the gated recurrent unit to process Y' to update the hidden state ht and predict the future traffic flow Form a dynamic weight table W, W = {P (1) (i,j), P (2) (i,j), …, P (m) (i,j)}, a total of m moments, where the element P (t) (i,j) represents the weight of the directed path from node i to node j at time t. If such an edge exists, then P (t) (i,j) is equal to the weight, and if not, it is equal to ∞.

[0010] S3. Perform global shortest path planning, including:

[0011] S31. Set the starting node O and the target node D. Set Ti of the remaining nodes except the starting node O to ∞, where Ti is the minimum cumulative travel time from the starting node O to node i. Add the starting node O and the minimum cumulative travel time TO = 0 to the set U, where U is the set of pairs (u, Tu) of nodes u with undetermined minimum cumulative travel times and their corresponding Tu.

[0012] S32. When the set U is not an empty set, traverse the set U to determine the node k with the minimum cumulative travel time and its corresponding cumulative travel time Tk. Remove (k, Tk) from the set U and add (k, Tk) to the set S, where S is the set of pairs (s, Ts) of nodes s with determined minimum cumulative travel times. When k is not the target node D, execute S33; otherwise, obtain the shortest path planning result Path and the minimum cumulative travel time Tk from node O to node D by backtracking through the inheritance relationship Parent of node k, where Parent is the inheritance relationship between two nodes in the shortest path.

[0013] S33. Traverse the neighbor nodes Neighbor_k of node k that are not in the set S. Determine the dynamic weight table W used for travel time retrieval according to the value of Tk. According to Calculate the cumulative minimum travel time TNeighbor_k of Neighbor_k. If TNeighbor_k is less than the original minimum cumulative travel time of node Neighbor_k, update the minimum cumulative travel time of Neighbor_k to TNeighbor_k and update the source node of Neighbor_k to k; otherwise, do not update.

[0014] S34. Loop through steps S32 and S33 until the shortest path Path from node O to node D is obtained.

[0015] Furthermore, in S1, the modeling of the road network in the area to be planned includes: using nodes {1, 2,..., n} to represent the intersections in the area to be planned, and using the adjacency matrix to represent the reachability relationship between intersections in the road network; the element Ai,j of the adjacency matrix represents the connection from node i to node j. If there is a directed edge from node i to node j, the value of Ai,j is 1; otherwise, the value of Ai,j is 0.

[0016] Furthermore, in S1, the time series where each row represents the travel times of all road segments in the planned area at the same moment, and each column represents the travel times of the same road segment in different time periods.

[0017] Further, in S2, processing the sequence Y using a temporal convolutional network to output a feature sequence Y' includes: ensuring that the output at time point t depends only on the current and past inputs through causal convolution, and increasing the receptive field of the convolutional kernel by inserting spaces in the convolutional kernel:

[0018]

[0019] where F(t) is the output feature value at time point t, f(i) is the convolutional kernel, k is the size of the convolutional kernel, d is the dilation factor, and xt - d·i is the input value at time point t - d·i.

[0020] Further, it also includes training the network through residual connections:

[0021]

[0022] where y(t) is the output of the layer, x(t) is the input of the layer, and ReLU is the rectified linear unit activation function.

[0023] Further, in S2, processing Y' using a gated recurrent unit includes: taking the output of the temporal convolutional network as the input of the gated recurrent unit, and the gated recurrent unit controls the information flow through the update gate z t and the reset gate r t to combine the current input and the past state, and update the hidden state ht; the hidden state ht combines the update gate z t and the candidate hidden state to update the final state as follows:

[0024]

[0025] z t = σ(W z · [h t-1 , x t + b z )

[0026] r t = σ(W r · [h t-1 , x t + b r )

[0027]

[0028] where σ is the sigmoid activation function, tanh is the hyperbolic tangent activation function, W and b are the parameters of the model; Wz is the weight matrix of the update gate, bz is the bias term of the update gate; Wr is the weight matrix of the reset gate, br is the bias term of the reset gate; ht - 1 is the hidden state of the previous time step.

[0029] Further, predicting future traffic flow in S2 includes: training using historical data Y, and training the model using a loss function that minimizes the mean squared error between the predicted value and the actual value:

[0030]

[0031] After training is completed, use the trained model to predict a new input sequence and output the prediction result where L is the loss function and N is the number of samples.

[0032] Further, forming the dynamic weight table W in S2 includes: using the generated prediction result to generate the dynamic weight table W; W = {P (1) (i,j), P (2) (i,j), …, P (m) (i,j)} is defined as the dynamic weight table, with a total of m moments; where the element represents the weight of the directed path from node i to node j at time t. If such an edge exists, then P (t) (i,j) is equal to the weight; if not, it is equal to ∞; the matrix is defined as follows:

[0033]

[0034] The present invention also provides a computing device, including a processor and a memory, and the memory stores executable code thereon. When the executable code is executed by the processor, the processor executes the method described above.

[0035] The present invention also provides a non-transitory machine-readable storage medium, which stores executable code thereon. When the executable code is executed by a processor of an electronic device, the processor executes the method described above.

[0036] Compared with traditional static path planning, the present invention can dynamically adapt to real-time changing traffic conditions by using traffic prediction for path planning, avoid congested areas in advance, optimize traffic flow distribution, achieve global path optimality, and effectively improve the overall efficiency and effect of urban traffic management.

[0037] The traffic prediction part of the present invention combines the advantages of the temporal convolutional network and the gated recurrent unit, is more efficient in processing long sequence data, can better avoid the problem of gradient disappearance, and has parallel processing ability, making it faster in training and prediction, and showing good generalization ability. Therefore, the model of the present invention can provide more accurate and efficient prediction performance, especially in capturing complex time dependencies and processing large-scale data sets.

[0038] The improved Dijkstra algorithm used in the present invention can adapt to the dynamic changes of traffic conditions by integrating a dynamic weight table, while maintaining the high efficiency of the algorithm and can be extended to large-scale road networks. And it can be interrupted at any time to provide the current shortest path solution, with high accuracy, efficiency and ease of use. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0040] Figure 1 is the flowchart of the method of the present invention;

[0041] Figure 2 is the schematic diagram of the network structure in step S2;

[0042] Figure 3 is the schematic diagram of path planning under dynamic weights;

[0043] Figure 4 is the flowchart of the improved Dijkstra algorithm of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0045] Embodiment 1:

[0046] See Figure 1 , the dynamic path planning method based on traffic prediction of a temporal convolutional network in this embodiment includes the following steps:

[0047] S1: Model the road network in the area to be planned.

[0048] S11: Model the road network conditions in the area to be planned. Use nodes {1, 2,..., n} to represent the intersections in the area to be planned, and there are a total of n intersections. Use the adjacency matrix to represent the reachability relationship between each intersection in the road network. The adjacency matrix A is an n×n matrix, where n is the total number of nodes in the graph. The element A of the adjacency matrix i，jDenotes the connection from node i to node j. If there is a directed edge from node i to node j, then the value of A i，j is 1, and if there is no such edge, then the value of A i，j is 0. Define the element A i，j of the adjacency matrix A as follows:

[0049]

[0050] S12: Collect historical traffic data of the area to be planned, perform data processing, and obtain the actual travel time series of each road section where each row represents the travel times of all road sections in the planned area at the same moment, and each column is the travel times of the same road section during different time periods.

[0051] S2: Based on the historical traffic data, predict the future travel times of the planned area to form a dynamic weight table W, see Figure 2 .

[0052] S21: Process the sequence Y based on the Temporal Convolutional Network (TCN), capture the features of the time series through dilated convolution and residual connections, and output the feature sequence Y'. During this process, causal convolution is used to ensure that the output at time point t depends only on the current and past inputs:

[0053]

[0054] On this basis, increase the receptive field of the convolutional kernel by inserting spaces in the convolutional kernel without increasing the number of parameters:

[0055]

[0056] where f(i) is the convolutional kernel, k is the size of the convolutional kernel, d is the dilation factor, which determines the interval between elements in the convolutional kernel, and x t-d·i is the input value at time point t - d·i.

[0057] Then, train the network through residual connections to prevent the problem of gradient disappearance:

[0058]

[0059] where y(t) is the output of the layer, x(t) is the input of the layer, and ReLU is the rectified linear unit activation function.

[0060] S22: The Gated Recurrent Unit (GRU) processes the output feature Y' of the TCN. Taking the output of the TCN as the input of the GRU, the GRU controls the information flow through the update gate and the reset gate, combines the current input and the past state, and updates the hidden state h t , and the final hidden state is used to predict the future traffic flow

[0061] Among them, the update gate z t determines how much past information to retain:

[0062] z t = σ(W z ·[h t-1 , x t + b z )

[0063] The reset gate r t determines how much past information to forget:

[0064] r t = σ(W r ·[h t-1 , x t + b r )

[0065] The candidate hidden state combines the information of the current input and the past state:

[0066]

[0067] The final hidden state h t combines the update gate and the candidate hidden state to update the final state:

[0068]

[0069] Among them, σ is the sigmoid activation function, tanh is the hyperbolic tangent activation function, and W and b are the parameters of the model. W z is the weight matrix of the update gate, and b z is the bias term of the update gate. W r is the weight matrix of the reset gate, and b r is the bias term of the reset gate. h t-1 is the hidden state at the previous time step.

[0070] S23: Model training and result output of TCN-GRU.

[0071] Training is performed using historical data Y, and the model is trained using the loss function that minimizes the mean-square error (MSE) between the predicted value and the actual value:

[0072]

[0073] After training is completed, the trained model is used to predict a new input sequence, and the prediction result is output. Where N is the number of samples.

[0074] S24: Generate a dynamic weight table W using the prediction result generated in step S23. W = {P (1) (i, j), P (2) (i, j),..., P (m) (i, j)} is defined as the dynamic weight table, and there are m moments in total. Among them, the element represents the weight of the directed path from node i to node j at time t. If there is such an edge, then P (t) (i, j) is equal to the weight. If not, it is equal to ∞. This matrix is defined as follows:

[0075]

[0076] S3: Based on the regional section travel time prediction made in S2, use the improved Dijkstra algorithm to perform global shortest path planning. See Figure 3 、 Figure 4 .

[0077] S31: Initialization. Obtain the dynamic weight table W according to step S2. Let the starting node of the path be O, the target node be D, and T i be the minimum cumulative travel time from the starting node O to node i. Initially, set the T i of nodes other than node O to ∞. U represents the set of nodes u with undetermined minimum cumulative travel time and T u for (u, T u ), and S represents the set of nodes s with determined minimum cumulative travel time and T s for (s, T s ). Parent represents the inheritance relationship between two nodes in the shortest path. Add the starting node O and the minimum cumulative travel time T o = 0 to the set U.

[0078] S32: Update the set U and the set S. When the set U is not an empty set, traverse the set U to determine the node k with the minimum cumulative travel time and its corresponding cumulative travel time T k , delete (k, T k ) from the set U, and add (k, T k) Add it to the set S. When k is not the target node, continue with S33, otherwise, obtain the shortest path planning result Path and the minimum cumulative travel time T from node O to node D by backtracking through the inheritance relationship of node k k .

[0079] S33: Update the minimum cumulative travel time of neighbor nodes. Traverse the neighbor nodes Neighbor_k of node k that are not in the set S. Determine the dynamic weight table W used for travel time retrieval according to the value of T k , and calculate the cumulative minimum travel time T of Neighbor_k according to Neighbor_k . If T Neighbor_k is less than the original minimum cumulative travel time of node Neighbor_k, then update the minimum cumulative travel time of Neighbor_k to T Neighbor_k , and update the source node of Neighbor_k to k, otherwise, do not update.

[0080] S34: Loop through steps S32 and S33 until the shortest path Path from node O to node D is obtained.

[0081] Embodiment 2:

[0082] This embodiment is a computing device, including a processor and a memory. The memory stores code for executing the method in the above embodiment.

[0083] The processor can be a multi-core processor or can include multiple processors. In some embodiments, the processor can include a general main processor and one or more special coprocessors, such as a graphics processing unit (GPU), a digital signal processor (DSP), etc. In some embodiments, the processor can be implemented using custom circuitry, such as an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA).

[0084] The memory may include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage devices. Among them, the ROM can store static data or instructions required by the processor or other modules of the computer. The permanent storage device can be a read-write storage device. The permanent storage device can be a non-volatile storage device that does not lose the stored instructions and data even when the computer is powered off. In some embodiments, the permanent storage device uses a mass storage device (such as a magnetic or optical disk, flash memory) as the permanent storage device. In some other embodiments, the permanent storage device can be a removable storage device (such as a floppy disk, optical drive). The system memory can be a read-write storage device or a volatile read-write storage device, such as dynamic random access memory. The system memory can store some or all of the instructions and data required by the processor during operation. In addition, the memory can include any combination of computer-readable storage media, including various types of semiconductor storage chips (DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), and magnetic disks and / or optical disks can also be used. In some embodiments, the memory can include a removable storage device that can be read and / or written, such as a compact disc (CD), read-only digital versatile disc (such as DVD-ROM, dual-layer DVD-ROM), read-only Blu-ray disc, super density disc, flash memory card (such as SD card, min SD card, Micro-SD card, etc.), magnetic floppy disk, etc. Computer-readable storage media do not include carrier waves and instantaneous electronic signals transmitted wirelessly or by wire.

[0085] The memory stores executable code thereon, which, when executed by the processor, causes the processor to execute the above method.

[0086] Example 3:

[0087] This embodiment provides a non-transitory machine-readable memory that stores executable code thereon, which, when executed by the processor of an electronic device, causes the processor to execute the method in the above embodiment.

[0088] A non-transitory machine-readable memory (or computer-readable memory, or machine-readable memory) stores executable code (or computer program, or computer instruction code) thereon, which, when executed by the processor of an electronic device (or computing device, server, etc.), causes the processor to execute each step of the above method according to the present invention.

[0089] Those skilled in the art will also understand that the various exemplary logical blocks, modules, circuits, and algorithm steps described in connection with the disclosure herein can be implemented as electronic hardware, computer software, or a combination of both.

[0090] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A dynamic path planning method based on temporal convolutional network traffic prediction, characterized in that: The steps include: S1. Model the road network in the planned area, collect historical traffic data of the planned area, and obtain the actual travel time series Y of each road section; S2, use the temporal convolutional network to process the sequence Y and output the feature sequence Y'; Use the gated recurrent unit to process Y' and update the hidden state h t , predicting future traffic flow Form a dynamic weight table W, W = {P (1) (i,j),P (2) (i,j),…,P (m) (i,j)}, a total of m moments, where the element P (t) (i, j) represents the weight of the directed path from node i to node j at time t. If such an edge exists, then P (t) (i,j) is equal to the weight, if it does not exist, it is equal to ∞; S3. Perform global shortest path planning, including: S31, set the starting node O, the target node D, and set the T of the remaining nodes except the starting node O i Set to ∞, T i is the minimum cumulative travel time from the starting node O to the node i; The starting point O and the minimum cumulative travel time T O = 0 is added to the set U, where U is the node u and T whose minimum cumulative travel time is not determined. u For (u,T u ) S32. When the set U is not an empty set, traverse the set U to determine the node k with the smallest cumulative travel time and its corresponding cumulative travel time T k , delete (k,T from the set U k ), (k,T k ) is added to the set S, where S is the node s and T with the minimum cumulative travel time determined. s For (s,T s ); when k is not the target node D, execute S33, otherwise, backtrack through the inheritance relationship Parent of node k to obtain the shortest path planning result Path and the minimum cumulative travel time T from node O to node D k , Parent is the inheritance relationship between two nodes in the shortest path; S33, traverse the neighbor nodes Neighbor_k of node k that are not in the set S, according to T k The value of determines the dynamic weight table W used for travel time retrieval. Calculate the cumulative minimum travel time T of Neighbor_k Neighbor_k , if T Neighbor_k If it is less than the original minimum cumulative travel time of node Neighbor_k, the minimum cumulative travel time of Neighbor_k is updated to T Neighbor_k , the source node of Neighbor_k is updated to k, otherwise it is not updated; S34, loop steps S32 and S33 until the shortest path from node O to node D is obtained.

2. The dynamic path planning method based on time series convolutional network traffic prediction according to claim 1 is characterized in that: The modeling of the road network in the planned area in S1 includes: using nodes {1, 2, ..., n} to represent the intersections in the planned area, using the adjacency matrix Represents the accessibility relationship between intersections in the road network; the element A of the adjacency matrix i,j represents the connection from node i to node j; if there is a directed edge from node i to node j, then A i,j The value of is 1, otherwise, A i,j The value of is 0.

3. The dynamic path planning method based on time series convolutional network traffic prediction according to claim 2 is characterized in that: In S1, the time series Each row is the travel time for all road sections in the planning area at the same time, and each column is the travel time for the same road section in different time periods.

4. The dynamic path planning method based on time series convolutional network traffic prediction according to claim 1 is characterized in that: In S2, a temporal convolutional network is used to process the sequence Y and output the feature sequence Y', including: ensuring that the output at time point t depends only on the current and past inputs through causal convolution, and increasing the receptive field of the convolution kernel by inserting spaces in the convolution kernel: Among them, F(t) is the output feature value at time point t, f(i) is the convolution kernel, k is the size of the convolution kernel, d is the dilation factor, and x t-d·i is the input value at time point td·i.

5. The dynamic path planning method based on time series convolutional network traffic prediction according to claim 4 is characterized in that: Also included is training the network via residual connections: Here, y(t) is the output of the layer, x(t) is the input of the layer, and ReLU is the rectified linear unit activation function.

6. The dynamic path planning method based on time series convolutional network traffic prediction according to claim 5 is characterized in that: In S2, a gated recurrent unit is used to process Y', including: taking the output of the temporal convolutional network as the input of the gated recurrent unit, and the gated recurrent unit updates the gate z t and reset gate r t Control the flow of information, combine the current input and the past state, and update the hidden state h t ; Hidden state h t Combined with update gate z t and candidate hidden states The final status of the update is as follows: With t =σ(W z ·[h t-1 ,xt]+b z ) r t =σ(W r ·[h t-1 ,x t ]+b r ) Among them, σ is the sigmoid activation function, tanh is the hyperbolic tangent activation function, W and b are the parameters of the model; W z is the weight matrix of the update gate, b z is the bias term of the update gate; W r is the weight matrix of the reset gate, b r is the bias term of the reset gate; h t-1 is the hidden state at the previous time step.

7. The dynamic path planning method based on time series convolutional network traffic prediction according to claim 6 is characterized in that: Predicting future traffic flow in S2 Including: using historical data Y for training, and using the mean square error to minimize the loss function between the predicted value and the actual value for model training: After the training is completed, the trained model is used to predict the new input sequence and output the prediction results Among them, L is the loss function and N is the number of samples.

8. The dynamic path planning method based on time series convolutional network traffic prediction according to claim 7 is characterized in that: The dynamic weight table W formed in S2 includes: using the generated prediction results Generate a dynamic weight table W; W = {P (1) (i,j),P (2) (i,j),…,P (m) (i,j)} is defined as a dynamic weight table, with a total of m moments; the elements represents the weight of the directed path from node i to node j at time t. If such an edge exists, then P (t) (i,j) is equal to the weight, if it does not exist, it is equal to ∞; the matrix is ​​defined as follows:

9. A computing device, characterized in that: The method comprises a processor and a memory, wherein the memory stores executable codes, and when the executable codes are executed by the processor, the processor executes the method according to any one of claims 1 to 8.

10. A non-transitory machine-readable storage medium, characterized in that: An executable code is stored thereon, and when the executable code is executed by a processor of an electronic device, the processor is caused to execute any one of the methods described in claims 1-8.

Images