Time-sensitive network delay control method and system for autonomous driving

Abstract

The application provides an automatic driving-oriented time-sensitive network delay control method and system, relates to the technical field of network communication, and comprises the following steps: obtaining data packet transmission delay information and network topology information, generating a candidate transmission path set based on node connection degree layering and state space search; collecting historical delay data, performing prediction calculation and weight distribution; iteratively updating delay weights based on hierarchical time scales, and selecting a path with the highest comprehensive score to forward data packets. The application can effectively reduce the network transmission delay of the automatic driving system and improve the stability and reliability of data transmission.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of network communication, and in particular to a time-sensitive network delay control method and system for automatic driving. BACKGROUND

[0002] With the rapid development of automatic driving technology, the timeliness of data transmission of vehicle network communication system is put forward higher requirements. As a deterministic network technology, time-sensitive network can provide low-delay and high-reliability data transmission guarantee for automatic driving, which is crucial for the safety and performance of automatic driving.

[0003] In the automatic driving scene, vehicles need to receive and process a large amount of data from the surrounding environment in real time, such as sensor data, traffic information and control instructions, etc. The transmission delay of these data directly affects the decision efficiency and safety performance of the automatic driving system. At present, the common time-sensitive network technology mainly reduces the network transmission delay through optimizing network scheduling algorithm, traffic shaping and preemption mechanism, etc. However, in the complex and changeable automatic driving environment, the existing technology still has many shortcomings.

[0004] The existing automatic driving network delay control technology also has some defects. The existing technology lacks adaptive ability to the dynamic changes of network topology in the aspect of route selection, and cannot flexibly adjust the transmission path according to the real-time network state, resulting in that the timely and reliable transmission of data packets cannot be guaranteed in the case of network congestion or node failure. The existing technology lacks accuracy in delay prediction, and mostly uses simple historical data average value as the basis for prediction, which cannot effectively cope with the delay fluctuation caused by network traffic burst change, affecting the real-time performance of the automatic driving system. The existing technology lacks comprehensive consideration of the correlation between network node load state and delay, and cannot develop differentiated transmission strategies for different node characteristics, resulting in uneven distribution of network resources and low overall transmission efficiency. SUMMARY

[0005] The embodiment of the present application provides a time-sensitive network delay control method and system for automatic driving, which can solve the problems in the prior art.

[0006] In a first aspect of the embodiment of the present application, a time-sensitive network delay control method for automatic driving is provided, comprising:

[0007] Obtaining the data packet transmission delay information and network topology information of the automatic driving vehicle, and calculating the node delay evaluation parameter according to the transmission delay information;

[0008] Generating a network node connection relationship graph according to the network topology information, layering the nodes in the network node connection relationship graph based on the connection degree values of the nodes, performing state space search operation to generate node state transition sequence, and obtaining a candidate transmission path set;

[0009] Collecting historical delay data of each candidate transmission path, performing mean calculation based on a preset time window, and obtaining predicted delay values of each candidate transmission path through weighted calculation;

[0010] Based on the comparison result of the delay difference value between the candidate transmission paths, the delay weight coefficients of each candidate transmission path are determined through subset division, and a delay weight distribution sequence is generated;

[0011] Based on the hierarchical time scale, the delay weight distribution sequence is iteratively updated, the candidate transmission paths are scored according to the node load rate data and the node delay evaluation parameters, and the candidate transmission path with the highest comprehensive score is selected as the optimal transmission path;

[0012] Through the optimal transmission path to forward data packets, when detecting that the real-time delay change value exceeds the preset delay change range, the delay weight coefficients of each path are recalculated using the latest collected historical delay data and node delay evaluation parameters.

[0013] In an optional embodiment, the nodes are layered based on the connection degree values of the nodes in the network node connection relationship graph, a state space search operation is performed to generate a node state transition sequence, and a candidate transmission path set is obtained, including:

[0014] The number of connection edges of each node in the network node connection relationship graph is counted to obtain the node connection degree value, and the nodes in the network node connection relationship graph are divided into multiple connection layers according to a preset first connection degree threshold and a preset second connection degree threshold;

[0015] Obtain the geographic coordinate information of each node, perform regional clustering operation on the nodes in each connection layer, calculate the Euclidean distance between nodes, and generate a node region distribution sequence;

[0016] Calculate the Euclidean distance between each node, obtain the connection degree ratio of each node, record the historical transmission success times and total transmission times between each node, and calculate the transmission state transition value through weighted calculation;

[0017] Determine the transition probability between nodes according to the transmission state transition value, perform a state space search operation to generate a node state transition sequence based on a dynamic search depth value, calculate the generation value of each transition, and optimize the node state transition sequence based on the generation value;

[0018] Use a random sampling method to perform multiple traversal searches on the state space, select a state transition path whose generation value satisfies a preset generation threshold, map the state transition path to an actual node access sequence, and generate a candidate transmission path set.

[0019] In an alternative embodiment, the inter-node transition probability is determined according to the transition value of the transmission state, the state space search operation is performed based on the dynamic search depth value to generate a node state transition sequence, the generation cost of each transition is calculated, and the node state transition sequence is optimized based on the generation cost, including:

[0020] The inter-node transition probability is determined according to the transition value of the transmission state.

[0021] The starting node is taken as the search starting point, the state space search operation is performed, the starting node is set as the current search node, and the dynamic search depth value is initialized.

[0022] The next hop node of the current search node is selected based on the inter-node transition probability, the next hop node is recorded to the visited node set, the historical access frequency of the next hop node is counted, and when the historical access frequency is greater than a preset access threshold, the inter-node transition probability of the next hop node is multiplied by a preset attenuation coefficient.

[0023] When the next hop node is the target node, the current search path is recorded as a node state transition sequence; when the next hop node is not the target node and is not in the visited node set, the next hop node is set as the current search node, the dynamic search depth value is updated based on a preset depth adjustment coefficient, and the state space search operation is continued; when the next hop node is in the visited node set, the previous node is returned to reselect the next hop node; when the search depth reaches the dynamic search depth value, the path cost is calculated based on the generation cost of each node in the visited node set and the distance from the target node, and when the path cost is greater than a preset path cost threshold, the previous node is returned to reselect the next hop node.

[0024] The generation cost of each transition in the node state transition sequence is calculated.

[0025] The node state transition sequence is optimized based on the generation cost, and the node state transition sequence with a generation cost less than a preset generation cost threshold and a node hop count less than a preset hop count threshold is selected as the final node state transition sequence.

[0026] In an alternative embodiment, historical delay data of each candidate transmission path is collected, mean value calculation is performed based on a preset time window, and a predicted delay value of each candidate transmission path is obtained through weighted calculation, including:

[0027] The historical delay data of each candidate transmission path is collected, the historical delay data is periodically sampled according to a preset sampling time interval, and a delay sampling sequence is generated.

[0028] perform data partitioning on the time delay sampling sequence according to a preset first time window and a preset second time window, wherein the second time window is greater than the first time window;

[0029] perform mean value calculation on the sampling data in the first time window and the sampling data in the second time window, and determine a short-term time delay mean value and a long-term time delay mean value respectively;

[0030] calculate a deviation rate of the short-term time delay mean value and the long-term time delay mean value, and determine a time delay prediction weight coefficient according to the deviation rate;

[0031] perform weighted calculation on the time delay prediction weight coefficient and a time delay sampling value at a current time, to obtain a predicted time delay value of each candidate transmission path;

[0032] statistically determine a prediction accuracy of the predicted time delay value, and when the prediction accuracy is lower than a preset accuracy threshold, update the time delay prediction weight coefficient based on a preset adjustment step, and recalculate the predicted time delay value.

[0033] In an optional embodiment, based on a comparison result of time delay difference values between candidate transmission paths, a time delay weight coefficient of each candidate transmission path is determined through subset division, and a time delay weight distribution sequence is generated, which includes:

[0034] calculate a time delay difference value between each candidate transmission path and all other candidate transmission paths, and establish a time delay difference value set for each transmission path;

[0035] compare the positive and negative attributes of each time delay difference value in the time delay difference value set of each candidate transmission path, and classify the time delay difference values greater than zero into a positive subset and the time delay difference values less than zero into a negative subset;

[0036] accumulate the time delay difference values in the positive subset of each candidate transmission path to obtain a time delay disadvantage value, and accumulate the time delay difference values in the negative subset to obtain a time delay advantage value;

[0037] determine a weight coefficient of the corresponding candidate transmission path through exponential operation based on the ratio relationship between the time delay disadvantage value and the time delay advantage value, and perform normalization calculation on the weight coefficients of all candidate transmission paths to determine a time delay weight distribution sequence.

[0038] In an optional embodiment, the time delay weight distribution sequence is iteratively updated based on hierarchical time scales, candidate transmission paths are scored according to node load rate data and node time delay evaluation parameters, and the candidate transmission path with the highest comprehensive score is selected as the optimal transmission path, which includes:

[0039] obtain a time delay weight distribution sequence of a candidate transmission path, and combine the time delay weight distribution sequence with a current path time delay state to form a state parameter;

[0040] The weight learning is performed on each candidate transmission path, the current weight value is calculated according to the time delay weight distribution sequence of the candidate transmission path, the ratio of the historical transmission success rate of the candidate transmission path to the weight fluctuation rate is taken as a weight prediction coefficient, and the current weight value is corrected based on the weight prediction coefficient to determine an expected weight value;

[0041] A weight updating mechanism of a hierarchical time scale is constructed, the current weight value is updated based on path state changes in a first time scale, and the calculation parameters of the expected weight value are updated based on path stability in a second time scale, and the weight value after iteration is obtained by weighted summation of the weight mean value of the first time scale and the weight prediction value of the second time scale;

[0042] The load rate data of each node on the candidate transmission path is collected, and the node time delay evaluation parameters are calculated based on the data transmission rate and the data packet round trip time between nodes;

[0043] The load balancing score is calculated according to the load rate data, the time delay score is calculated based on the node time delay evaluation parameters, and the weighted sum of the load balancing score and the time delay score is taken as the path basic score;

[0044] The path comprehensive score is obtained by multiplying the path basic score by the weight value after iteration, and the candidate transmission path with the highest path comprehensive score and greater than a preset score threshold is selected as the optimal transmission path.

[0045] The second aspect of the embodiment of the application provides a time-sensitive network time delay control system for automatic driving, comprising:

[0046] The first unit is used for acquiring data packet transmission time delay information and network topology information of the automatic driving vehicle, and calculating node time delay evaluation parameters according to the transmission time delay information;

[0047] The second unit is used for generating a network node connection relationship graph according to the network topology information, layering nodes based on the connection degree values of the nodes in the network node connection relationship graph, performing state space search operation to generate a node state transition sequence, and obtaining a candidate transmission path set;

[0048] The third unit is used for collecting historical time delay data of each candidate transmission path, performing mean value calculation based on a preset time window, and obtaining predicted time delay values of each candidate transmission path through weighted calculation;

[0049] The fourth unit is used for determining time delay weight coefficients of each candidate transmission path and generating a time delay weight distribution sequence based on the comparison results of the time delay difference values between the candidate transmission paths through subset division;

[0050] The fifth unit is configured to iteratively update the time delay weight distribution sequence based on a hierarchical time scale, score the candidate transmission paths according to the node load rate data and the node time delay evaluation parameters, and select the candidate transmission path with the highest comprehensive score as the optimal transmission path.

[0051] The sixth unit is configured to forward the data packet through the optimal transmission path, and when detecting that the real-time time delay change value exceeds the preset time delay change range, recompute the time delay weight coefficients of the paths by using the latest collected historical time delay data and the node time delay evaluation parameters.

[0052] In a third aspect, an electronic device is provided, including:

[0053] a processor;

[0054] a memory for storing processor-executable instructions;

[0055] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0056] In a fourth aspect, a computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0057] In the embodiment of the application, the candidate transmission path set is constructed by combining node layering and state space search, the historical time delay data is used for prediction and dynamic allocation of time delay weight, the optimal transmission path in the network can be accurately identified, the data transmission time delay in the automatic driving system is effectively reduced, and the transmission efficiency of the data packet is improved; the iterative updating mechanism based on the hierarchical time scale is adopted, the node load rate and the time delay evaluation parameters are combined for comprehensive scoring, the path selection is more adaptive, the transmission strategy can be adjusted in a timely manner according to the network state change, the network congestion and fluctuation are effectively coped with, and the reliability and stability of the automatic driving network are enhanced; the real-time monitoring and rapid response of the network time delay are realized, the optimal path can be immediately recomputed when detecting that the time delay change exceeds the preset range, the dynamic time delay control mechanism is particularly suitable for the application scenarios with high real-time requirements such as automatic driving, and the adaptability of the system to the network environment change and the transmission quality of the time-sensitive data are significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 FIG. 1 is a flowchart of a time-sensitive network time delay control method for automatic driving according to an embodiment of the application;

[0059] Figure 2 FIG. 4 is a flowchart of dynamic probability state transition path search and optimization. DETAILED DESCRIPTION

[0060] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to 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 the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the protection scope of the present application.

[0061] The technical solutions of the present application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and some embodiments can not be described again for the same or similar concepts or processes.

[0062] Figure 1 The flowchart of the time-sensitive network delay control method for autonomous driving of the embodiments of the present application is shown in FIG. 1, which comprises the following steps. Figure 1

[0063] Obtaining the data packet transmission delay information and network topology information of the autonomous driving vehicle, and calculating the node delay evaluation parameter according to the transmission delay information;

[0064] Generating a network node connection relationship graph according to the network topology information, layering the nodes in the network node connection relationship graph based on the connection degree values of the nodes, performing state space search operation to generate a node state transition sequence, and obtaining a candidate transmission path set;

[0065] Collecting the historical delay data of each candidate transmission path, performing mean value calculation based on a preset time window, and obtaining the predicted delay value of each candidate transmission path through weighted calculation;

[0066] Based on the delay difference comparison result between the candidate transmission paths, determining the delay weight coefficients of each candidate transmission path through subset division, and generating a delay weight distribution sequence;

[0067] Iteratively updating the delay weight distribution sequence based on the hierarchical time scale, scoring the candidate transmission paths according to the node load rate data and the node delay evaluation parameter, and selecting the candidate transmission path with the highest comprehensive score as the optimal transmission path;

[0068] Forwarding the data packet through the optimal transmission path, and when detecting that the real-time delay change value exceeds the preset delay change range, recalculating the delay weight coefficients of each path using the latest collected historical delay data and the node delay evaluation parameter.

[0069] ​In one specific implementation, in a time-sensitive network environment for autonomous driving, the reliability and real-time performance of data packet transmission directly affect the safety control and decision-making of vehicles. Obtaining the data packet transmission delay information and network topology information of autonomous vehicles is a basic step for path selection. Through the vehicle-mounted communication module, the communication quality with roadside units, cloud servers, and surrounding vehicles is continuously monitored, and end-to-end delay data of data packet transmission is collected. Through the network discovery protocol, the current available network node information is obtained, including node identification, connection relationship, and transmission capacity, etc. For example, in a typical urban road scenario, an autonomous vehicle may be connected to multiple roadside units, mobile base stations, and edge computing nodes, forming a network topology containing 15 nodes. By analyzing the timestamp information in the network layer protocol message, the transmission delay of data packets from sending to receiving is measured, and the processing delay of each node is recorded. Statistical analysis of these delay data calculates the average processing delay, delay fluctuation range, and packet loss rate of each node, etc., to comprehensively form the node delay evaluation parameters. Taking an edge computing node as an example, its average processing delay is 3.5 milliseconds, the delay fluctuation range is 1.2 milliseconds, and the packet loss rate is 0.03%, and accordingly the node delay evaluation parameter value is calculated to be 4.2.

[0070] Based on the obtained network topology information, a network node connection relationship diagram is generated, which is a prerequisite for candidate transmission path generation. Each network node is represented as a vertex in the graph, and the communication link between nodes is represented as an edge, constructing an undirected connection graph. The connectivity of each node in the graph is calculated, that is, the number of other nodes directly connected to the node. Based on the connectivity, the nodes are layered, with high connectivity nodes assigned to higher levels and low connectivity nodes assigned to lower levels. In the above 15-node network, analysis shows that the connectivity of node 2, node 7, and node 11 is 6, 5, and 4 respectively, and they are classified into the first layer; the connectivity of node 1, node 4, node 8, and node 13 is 3, and they are classified into the second layer; the connectivity of the remaining nodes is 1 or 2, and they are classified into the third layer. Based on the layering results, taking the node where the vehicle-mounted communication module is located as the starting point and the target service node as the end point, a search algorithm is executed for state space search. In the search process, nodes with high levels are preferentially selected for state expansion, and the distance between nodes and the estimated transmission delay are considered as heuristic information. Through state space search, a series of node state transition sequences from the starting point to the end point are generated, each sequence corresponding to a possible transmission path. After screening and optimization, 8 candidate transmission paths are finally obtained, constituting the candidate transmission path set.

[0071] Collecting historical delay data of each candidate transmission path and making delay prediction is a key step for evaluating path performance. For each candidate transmission path, by periodically sending probe data packets, the end-to-end transmission delay is recorded to establish a delay history database. Based on a preset time window, for example, the delay data in the last 10 minutes, the average delay of each path is calculated. In the calculation process, the weighted average method is used for the data in the time window, and the closer to the current time, the greater the weight of the delay data, in order to reflect the latest changes in network state. For example, for the third candidate path, assuming that a total of 60 sets of delay data are collected in the last 10 minutes, the average delay in the last 1 minute is 22 milliseconds, the average delay in the first 1-5 minutes is 25 milliseconds, and the average delay in the first 6-10 minutes is 27 milliseconds, then through weighted calculation (weight ratio is 5:3:2), the predicted delay value of the path is 23.8 milliseconds. Similarly, the predicted delay values of the other 7 candidate paths are calculated, which are 18.5 milliseconds, 21.2 milliseconds, 25.6 milliseconds, 19.7 milliseconds, 24.3 milliseconds, 20.1 milliseconds and 22.4 milliseconds respectively.

[0072] Based on the comparison results of the delay difference values between the candidate transmission paths, the weight coefficients of each path are determined, which is an important basis for resource allocation. For each pair of candidate transmission paths i and j, the delay difference value Dij is calculated, which is the predicted delay of path i minus the predicted delay of path j. According to the calculation results, a delay difference value set is established for each path. For the first path (predicted delay 18.5 milliseconds), its delay difference value set is {-2.7, -7.1, -1.2, -5.8, -1.6, -3.9}, all of which are negative, indicating that the delay performance of this path is better than that of all other paths. The delay difference value set of each path is divided into subsets, and the positive value difference is assigned to the positive value subset, and the negative value difference is assigned to the negative value subset. The sum of the positive value subset (delay disadvantage value) and the sum of the negative value subset (delay advantage value) of each path are calculated. For the second path (predicted delay 21.2 milliseconds), its delay difference value set is {2.7, -4.4, 1.5, -3.1, 1.1, -1.2}, the sum of the positive value subset is 5.3, and the sum of the negative value subset is -8.7. Based on the proportional relationship between the absolute values of the delay disadvantage value and the delay advantage value, the weight coefficient is determined by exponential operation. When the delay advantage value of the path is zero, the weight coefficient is set to the minimum value 0.01; otherwise, the weight coefficient is equal to the power operation result with the natural logarithm e as the base and the negative number of the ratio of the delay disadvantage value to the absolute value of the delay advantage value as the exponent. The weight coefficients of the 8 paths are calculated to be 1.0, 0.543, 0.257, 0.073, 0.721, 0.124, 0.682 and 0.387 respectively. These weight coefficients are normalized to obtain the delay weight distribution sequence {0.264, 0.143, 0.068, 0.019, 0.190, 0.033, 0.180, 0.102}.

[0073] The iterative update of the delay weight distribution sequence based on hierarchical time scales is the key mechanism to adapt to network dynamic changes. Three time scales are set: short-term (1 minute) and long-term (15 minutes). In the short-term scale, the real-time delay data of each path is updated every 10 seconds; in the long-term scale, the complete path search and evaluation process is re-executed every 5 minutes. Through this hierarchical time strategy, both the rapid response to network state changes and the system overhead caused by frequent path switching are avoided. At the same time, combined with node load rate data and node delay evaluation parameters, a comprehensive score is given to each candidate transmission path. The score calculation considers three factors: path predicted delay, average load rate of nodes on the path, and node delay evaluation parameter. Taking the first path as an example, its predicted delay is 18.5 milliseconds, the average load rate of nodes on the path is 65%, and the average value of the node delay evaluation parameter is 3.8, and the comprehensive score is 87.3 after weighted calculation. Similarly, the comprehensive scores of other paths are calculated, and the first path with the highest score is finally selected as the optimal transmission path.

[0074] The data packet is forwarded through the optimal transmission path, and the actual transmission delay change is continuously monitored. When the real-time delay change value is detected to be out of the preset delay change range, the path reselection mechanism is triggered. For example, if the actual transmission delay of the first path increases from 18.5 milliseconds to 23.5 milliseconds, which exceeds the preset range, the latest collected historical delay data and updated node delay evaluation parameters are used to recalculate the delay weight coefficients and comprehensive scores of each path. In a certain reselection process, due to the sharp increase of the load rate of a key node on the first path to 95%, the comprehensive score of the path decreases to 75.8, while the comprehensive score of the fifth path increases to 83.5, and the data transmission is automatically switched to the fifth path to ensure the continuous and efficient communication.

[0075] Through the above method, the time-sensitive network for automatic driving can continuously optimize the transmission path of the data packet according to the dynamic changes of the network state, ensure the reliable transmission of critical information under strict delay constraints, provide stable network support for the automatic driving system, and effectively improve the real-time performance and reliability of vehicle control, environment perception and decision-making.

[0076] In an optional implementation, the nodes are layered based on the connection degree values of the nodes in the network node connection relationship graph, the state space search operation is performed to generate the node state transition sequence, and the candidate transmission path set is obtained, including:

[0077] The number of connection edges of each node in the network node connection relationship graph is counted to obtain a node connection degree value, and the nodes in the network node connection relationship graph are divided into multiple connection layers according to a preset first connection degree threshold and a preset second connection degree threshold;

[0078] Geographic coordinate information of each node is obtained, and regional clustering operations are performed on the nodes in each connection layer, the Euclidean distance between nodes is calculated, and a node region distribution sequence is generated;

[0079] The Euclidean distance between each node is calculated, the connection degree ratio of each node is obtained, the historical number of successful transmissions and the total number of transmissions between each node are recorded, and a transmission state transition value is calculated by weighting;

[0080] The transition probability between nodes is determined according to the transmission state transition value, a state space search operation is performed based on a dynamic search depth value to generate a node state transition sequence, the generation value of each transition is calculated, and the node state transition sequence is optimized based on the generation value.

[0081] The state space is searched multiple times using a random sampling method, a state transition path whose generation value meets a preset generation threshold is selected, the state transition path is mapped to an actual node access order, and a candidate transmission path set is generated.

[0082] In a specific embodiment, the number of connection edges of each node in the network node connection relationship graph is counted to obtain a node connection degree value. For example, in a network composed of 50 nodes, node A is connected to 8 edges, so its connection degree is 8; node B is connected to 3 edges, so its connection degree is 3. According to the pre-set first connection degree threshold 5 and the second connection degree threshold 10, the nodes are divided into three connection layers: the nodes with a connection degree less than 5 are classified into the first connection layer (such as node B), the nodes with a connection degree between 5 and 10 are classified into the second connection layer (such as node A), and the nodes with a connection degree greater than 10 are classified into the third connection layer. This hierarchical division helps to prioritize nodes with better connectivity in subsequent path selection.

[0083] After obtaining the geographic coordinate information of each node, regional clustering operations are performed on the nodes in each connection layer. Taking the second connection layer as an example, assume that this layer contains 20 nodes, each node has its geographic coordinates (x, y). The Euclidean distance between any two nodes is calculated, such as the Euclidean distance between node A (10, 15) and node C (13, 17) is 5. Set the clustering radius to 10, and nodes with a distance less than this value are classified into a regional cluster. Through clustering operations, this connection layer can be divided into 3 regional clusters, containing 7, 8 and 5 nodes respectively. Similar operations are performed on all connection layers to finally generate a node region distribution sequence.

[0084] The Euclidean distance between nodes in the computing network is calculated, such as the distance between node D (20, 30) and node E (25, 35) is 7.07. The connection degree ratio of each node is obtained, for example, the connection degree of node F is 7, and the connection degree of node G is 14, so the connection degree ratio of F to G is 0.5. The historical transmission between each node is recorded, such as node H to node I has successfully transmitted 85 times in the past 100 transmissions, and the historical transmission success rate is 0.85. Based on the three indicators, the transmission state transition value is calculated. Assuming that the distance weight is 0.3, the connection degree ratio weight is 0.3, and the historical transmission success rate weight is 0.4, the transmission state transition value between nodes can be calculated. For example, the transmission state transition value of node pair (H, I) may be 0.75.

[0085] The transition probability between nodes is determined according to the transmission state transition value, and the higher the transition probability, the more superior the link. The dynamic search depth value is set, which is initially 3 and can be dynamically adjusted to 5 or 7 according to the network size. The state space search operation is performed to generate a node state transition sequence. For example, starting from the source node S, a state transition sequence [S→A→C→D→T] can be generated, where T is the target node. The total generation value of this sequence is calculated, which is assumed to be 28.5. Multiple searches are performed to generate multiple state transition sequences, such as [S→B→E→T] with a generation value of 32.1 and [S→F→G→H→T] with a generation value of 26.8. Based on the generation value, the node state transition sequence is optimized, and the sequence with the lowest generation value is selected as the optimal path.

[0086] The state space is searched multiple times using a random sampling method. In each search, the next hop node is randomly selected with a certain probability to avoid falling into a local optimal solution. For example, 100 random sampling searches are performed to obtain 100 possible paths. A preset generation cost threshold of 30 is set, and paths with a generation value less than 30 are selected, such as [S→A→C→D→T] and [S→F→G→H→T]. These state transition paths are mapped to actual node access sequences to generate a candidate transmission path set.

[0087] In this embodiment, the connection degree of the network nodes is layered, and regional clustering is performed in combination with node geographic coordinate information, so that the organization structure of the network nodes is more clear, and the complexity of path search is reduced; the connection degree ratio, Euclidean distance and historical transmission success rate of the nodes are considered in the path selection process, which can comprehensively reflect the actual transmission capacity between nodes, thereby improving the accuracy of path evaluation; state space search is performed based on the transmission state transition value and dynamic search depth, and optimization is performed based on the generation value, which is helpful to realize dynamic optimization of the transmission path; multiple iterations of the state space are performed by the random sampling method to avoid falling into a local optimal solution, thereby improving the selection probability of the globally optimal path.

[0088] In an alternative embodiment, the inter-node transition probability is determined according to the transition value of the transmission state, the state space search operation is performed based on the dynamic search depth value to generate a node state transition sequence, the generation cost of each transition is calculated, and the node state transition sequence is optimized based on the generation cost, including:

[0089] determining the inter-node transition probability according to the transition value of the transmission state;

[0090] taking the starting node as the search starting point, performing the state space search operation, setting the starting node as the current search node, and initializing the dynamic search depth value;

[0091] selecting the next hop node of the current search node based on the inter-node transition probability, recording the next hop node to the visited node set, and counting the historical access frequency of the next hop node; when the historical access frequency is greater than a preset access threshold, multiplying the inter-node transition probability of the next hop node by a preset attenuation coefficient;

[0092] when the next hop node is the target node, recording the current search path as a node state transition sequence; when the next hop node is not the target node and is not in the visited node set, setting the next hop node as the current search node, updating the dynamic search depth value based on a preset depth adjustment coefficient, and continuing to perform the state space search operation; when the next hop node is in the visited node set, returning to the previous node to reselect the next hop node; when the search depth reaches the dynamic search depth value, calculating the path cost based on the generation cost of each node in the visited node set and the distance from the target node; when the path cost is greater than a preset path cost threshold, returning to the previous node to reselect the next hop node;

[0093] calculating the generation cost of each transition in the node state transition sequence;

[0094] optimizing the node state transition sequence based on the generation cost, and selecting the node state transition sequence with a generation cost less than a preset generation cost threshold and a node hop count less than a preset hop count threshold as the final node state transition sequence.

[0095] In a specific embodiment, in network transmission, the state transition between nodes is usually dependent on various network conditions such as bandwidth, delay, and packet loss rate. By quantifying these conditions, the transmission state transition value can be calculated. Specifically, assume that there are nodes A, B, C, D, and E in the network, where A is the starting node and E is the target node. The transmission state transition value between nodes can be obtained through network detection, for example, the transmission state transition value from node A to B is 0.8, from A to C is 0.6, from B to D is 0.7, from C to D is 0.5, and from D to E is 0.9.

[0096] The determination of the inter-node transition probability is directly based on the transmission state transition value, which can be obtained by normalizing the transition value. For node A, the transition probabilities of its adjacent nodes B and C are 0.8 / (0.8+0.6)=0.57 and 0.6 / (0.8+0.6)=0.43, respectively.

[0097] The state space search operation is executed from the starting node A. The dynamic search depth value is initialized to 3, indicating the maximum depth allowed for the initial search. The preset depth adjustment coefficient is 0.8, the preset access threshold is 2, the preset decay coefficient is 0.5, the preset path cost threshold is 5.0, the preset cost threshold is 10.0, and the preset hop threshold is 4.

[0098] During the search process, the current search node is initially set to the starting node A. Based on the inter-node transition probability, the next hop node is selected from the adjacent nodes B and C of A. Assuming that the value generated by the random number generator is 0.3, which is less than the transition probability 0.57 of B, B is selected as the next hop node. B is recorded in the visited node set {B}, and the historical access frequency of B is counted as 1, which does not exceed the preset access threshold 2, so the transition probability does not need to be adjusted.

[0099] Since B is not the target node E and is not in the visited node set (at this time the visited node set is {B}), B is set as the current search node. The dynamic search depth value is updated to 3*0.8=2.4, which is rounded to 2. The state space search is continued, and the next hop node D is selected from the adjacent nodes of B. D is added to the visited node set {B, D}, and the historical access frequency of D is counted as 1.

[0100] D is not the target node and is not in the visited node set, so D is set as the current search node. The dynamic search depth value is updated to 2*0.8=1.6, which is rounded to 2. The search is continued, and the adjacent node E of D is selected as the next hop node. E is the target node, so the current search path A→B→D→E is recorded as a node state transition sequence.

[0101] Other possible paths are also explored. For example, starting from A, C is selected as the next hop node, then D is selected, and finally E is reached, forming the path A→C→D→E.

[0102] For each node state transition sequence, the cost value of each transition is calculated. The cost value can be based on various factors such as transmission delay, energy consumption, etc. Assuming that the cost of A→B is 2.0, the cost of B→D is 3.0, and the cost of D→E is 1.5, the total cost of the path A→B→D→E is 6.5. While the cost of A→C is 1.5, the cost of C→D is 2.5, and the cost of D→E is 1.5, the total cost of the path A→C→D→E is 5.5.

[0103] In the search process, if the historical access frequency of a node exceeds the preset access threshold, the transition probability of the node needs to be adjusted. For example, if the node D is accessed again, its historical access frequency increases to 3, which exceeds the preset access threshold 2, then the transition probability of D is multiplied by the preset decay coefficient 0.5, reducing the possibility of repeated access.

[0104] When the search depth reaches the dynamic search depth value, the cost of the current path is evaluated. For example, if the search depth is 3 when starting from A and passing through B and C to reach D, which is equal to the dynamic search depth value. At this time, the path cost is calculated as the cost of A→B 2.0 plus the cost of B→C 2.2 plus the cost of C→D 2.5, which is 6.7, greater than the preset path cost threshold 5.0, so the last node is returned to reselect the next hop node.

[0105] Based on the value of the generation, all node state transition sequences are optimized. The total cost of the path A→C→D→E is 5.5, which is less than the preset cost threshold 10.0, and the node hop count is 3, which is less than the preset hop count threshold 4, while the total cost of the path A→B→D→E is 6.5, which also meets the conditions. Comparing the two paths, the cost of A→C→D→E is smaller, so A→C→D→E is selected as the final node state transition sequence.

[0106] As shown in Figure 2 The dynamic probability state transition path search and optimization flowchart is shown.

[0107] In this embodiment, by determining the transition probability between nodes based on the transmission state transition value, and combining the dynamic search depth to search the state space, the search complexity can be reduced while ensuring the sufficiency of the search; by using the statistical results of the visited node set and the historical access frequency, and introducing the probability decay mechanism when the preset access threshold is exceeded, repeated access and invalid search are effectively avoided; during the search process, the search depth is dynamically updated according to the access situation and the preset depth adjustment coefficient, making the search process adaptive and flexible; the path cost is evaluated by combining the node generation value and the distance to the target node, and the optimization of the node state transition sequence is realized through the double constraints of the cost threshold and the hop count threshold, thereby ensuring the high quality and rationality of the selected path.

[0108] In an optional implementation, historical delay data of each candidate transmission path is collected, mean value calculation is performed based on a preset time window, and a predicted delay value of each candidate transmission path is obtained through weighted calculation, including:

[0109] The historical delay data of each candidate transmission path is collected, and the historical delay data is periodically sampled according to a preset sampling time interval to generate a delay sampling sequence;

[0110] perform data partitioning on the time delay sampling sequence according to a preset first time window and a preset second time window, where the second time window is greater than the first time window;

[0111] perform mean value calculation on the sampling data in the first time window and the sampling data in the second time window, and determine a short-term time delay mean value and a long-term time delay mean value respectively;

[0112] calculate a deviation rate of the short-term time delay mean value and the long-term time delay mean value, and determine a time delay prediction weight coefficient according to the deviation rate;

[0113] perform weighted calculation on the time delay prediction weight coefficient and a time delay sampling value at a current moment, to obtain a predicted time delay value of each candidate transmission path;

[0114] statistically determine a prediction accuracy of the predicted time delay value, and when the prediction accuracy is lower than a preset accuracy threshold, update the time delay prediction weight coefficient based on a preset adjustment step, and recalculate the predicted time delay value.

[0115] In a specific embodiment, to collect historical time delay data of each candidate transmission path, a plurality of probe nodes can be deployed in the network, which periodically send probe packets to the target server and record the round-trip time. The sampling time interval can be set to 500 milliseconds, i.e., time delay data is collected every 500 milliseconds. For each candidate path, a time delay data record table is maintained, which contains a timestamp and a corresponding time delay value, forming a time delay sampling sequence. For example, part of the sampling data of a certain path may be: [2023-10-01 08:00:00, 25ms], [2023-10-01 08:00:00.5, 27ms], [2023-10-01 08:00:01, 26ms], etc.

[0116] After the time delay sampling sequence is generated, data is partitioned based on preset time windows. The first time window (short-term window) can be set to 10 seconds, and the second time window (long-term window) can be set to 60 seconds. The short-term window reflects the immediate changes in network state, and the long-term window reflects the overall stability of the network. In a specific implementation, a sliding window mechanism is maintained, and as time passes, the data in the window is constantly updated. For example, if the current time is 08:01:00, the short-term window contains data from 08:00:50 to 08:01:00 (a total of 20 sampling points), and the long-term window contains data from 08:00:00 to 08:01:00 (a total of 120 sampling points).

[0117] When calculating the mean value of the divided data, the arithmetic mean value of all sampling points in the short-term window and the long-term window is calculated respectively. Assuming that the total of the time delay values of 20 sampling points in the short-term window is 520 ms, the short-term time delay mean value is 26 ms; the total of the time delay values of 120 sampling points in the long-term window is 2880 ms, and the long-term time delay mean value is 24 ms.

[0118] The deviation rate of the short-term time delay mean value and the long-term time delay mean value is a key step for predicting the time delay. The deviation rate is calculated by the difference between the short-term mean value and the long-term mean value divided by the long-term mean value, and then taking the absolute value. Continuing the above example, the deviation rate is |26-24| / 24=0.0833, about 8.33%. The deviation rate reflects the trend of the network state, and the greater the deviation rate, the greater the network state fluctuation.

[0119] When determining the time delay prediction weight coefficient based on the deviation rate, an adaptive weight distribution mechanism is adopted. When the deviation rate is small (such as less than 5%), a higher weight (such as 0.7) is given to the current sampling value, and a lower weight (such as 0.3) is given to the historical mean value; when the deviation rate is large (such as greater than 20%), a higher weight (such as 0.8) is given to the historical mean value, and a lower weight (such as 0.2) is given to the current sampling value; when the deviation rate is in the middle area, the weight coefficient changes linearly. In the above example, the deviation rate is 8.33%, which is in the middle area, and the system may assign a weight of 0.6 to the current sampling value and a weight of 0.4 to the historical mean value.

[0120] When performing weighted calculation, the time delay sampling value at the current time and the historical mean value are combined according to the weight. Assuming that the current sampling value is 28 ms and the weight is 0.6; the historical mean value (which can be the long-term time delay mean value) is 24 ms and the weight is 0.4, then the predicted time delay value is 28×0.6+24×0.4=26.4 ms. The predicted time delay is calculated for each candidate path, and the result is stored in the predicted time delay table.

[0121] Prediction accuracy evaluation is an important link to ensure the prediction quality. The difference between each prediction value and the actual observation value is recorded, and the prediction error rate is calculated. For example, the predicted time delay is 26.4 ms, and the actual observed time delay at the next time is 25 ms, then the prediction error rate is |26.4-25| / 25=0.056, about 5.6%. A sliding window (such as the last 100 predictions) is maintained to calculate the average error rate, which is used as an evaluation index of the prediction accuracy.

[0122] When the prediction accuracy is lower than the preset threshold (e.g., the average error rate exceeds 10%), a weight adjustment mechanism is started. The adjustment step can be set to 0.05, and the weight coefficient is adjusted according to the direction of the prediction error. If the predicted value is generally higher than the actual value, the weight of the current sampling value is reduced, and the weight of the historical average value is increased; otherwise, the weight of the current sampling value is increased, and the weight of the historical average value is reduced. After adjustment, the system recalculates the prediction delay using the new weight coefficient and continues to monitor the prediction accuracy, forming a closed-loop optimization mechanism.

[0123] Through the above method, the delay of each candidate transmission path can be accurately predicted, providing a reliable basis for network transmission path selection, effectively improving network transmission efficiency and user experience. The method has strong adaptability and can cope with delay changes in various complex network environments, and has wide application prospects.

[0124] In an optional embodiment, based on the comparison result of the delay difference values between the candidate transmission paths, the delay weight coefficients of each candidate transmission path are determined by subset division, and a delay weight distribution sequence is generated, which includes:

[0125] The delay difference values between each candidate transmission path and all other candidate transmission paths are calculated, and a delay difference value set is established for each transmission path;

[0126] The positive and negative properties of each delay difference value in the delay difference value set of each candidate transmission path are compared, and the delay difference values greater than zero are classified into a positive subset, and the delay difference values less than zero are classified into a negative subset;

[0127] The delay difference values in the positive subset of each candidate transmission path are accumulated to obtain a delay disadvantage value, and the delay difference values in the negative subset are accumulated to obtain a delay advantage value;

[0128] Based on the ratio relationship between the delay disadvantage value and the delay advantage value, the weight coefficient of the corresponding candidate transmission path is determined by exponential operation; the weight coefficients of all candidate transmission paths are normalized to determine the delay weight distribution sequence.

[0129] In a specific embodiment, calculating the delay difference values between the candidate transmission paths is the basic step of the method. For n candidate transmission paths in an automatic driving network, the end-to-end delay values of each path are measured and denoted as T1, T2,..., Tn. To evaluate the delay performance of each path relative to other paths, the delay difference value Dij between path i and all other paths j needs to be calculated. The calculation method of Dij is Ti minus Tj, which represents the delay difference of path i relative to path j. For example, in the case of 5 candidate paths, the delay of path 1 is 25 milliseconds, and the delay of path 2 is 30 milliseconds. Then D12 is -5 milliseconds, indicating that path 1 is 5 milliseconds faster than path 2. In this way, a delay difference value set containing n-1 elements is established for each candidate transmission path.

[0130] The classification of the set of delay difference values is an important step for evaluating the performance of the paths. For each candidate transmission path i, its set of delay difference values {Di1, Di2,..., Din} (excluding Dii) is classified according to the positive and negative properties. The delay difference values greater than zero are classified into the positive subset Pi, and the delay difference values less than zero are classified into the negative subset Ni. If the delay difference value is positive, it indicates that the path is longer than the compared path; if it is negative, it indicates that the path is shorter. In practical applications, for example, the set of delay difference values of path 3 is {5, -3, -2, 4}, then the positive subset P3 is {5, 4} and the negative subset N3 is {-3, -2}.

[0131] The calculation of the delay advantage and disadvantage values directly affects the evaluation results of the paths. For each candidate transmission path i, the sum of all delay difference values in its positive subset Pi is calculated to obtain the delay disadvantage value PSumi, which represents the cumulative time that the path is slower than other paths; the sum of all delay difference values in its negative subset Ni is calculated to obtain the delay advantage value NSumi, which represents the cumulative time that the path is faster than other paths. Taking path 3 as an example, its delay disadvantage value PSum3 is 9 milliseconds (5+4), and its delay advantage value NSum3 is -5 milliseconds (-3+(-2)). The smaller the delay disadvantage value and the larger the absolute value of the delay advantage value, the better the performance of the path.

[0132] The determination of the weight coefficient uses an exponential operation method, which reflects the nonlinear relationship between the path performance and the weight allocation. For the candidate transmission path i, based on its delay disadvantage value PSumi and delay advantage value NSumi, the weight coefficient Wi is calculated. When the delay advantage value NSumi is 0, Wi=1; when NSumi is not 0, Wi is equal to the power operation result with e as the base and the negative number of the ratio of PSumi to |NSumi| as the exponent. For example, for path 3, its weight coefficient W3=e (-(9 / 5)) ≈0.166. This exponential operation method ensures that the path with better delay performance obtains a higher weight coefficient.

[0133] The weight normalization processing ensures the rationality of resource allocation. The sum Sum_W of all candidate transmission path weight coefficients is calculated, and then the weight coefficient of each path is divided by Sum_W to obtain the normalized weight value NWi. In this way, the sum of the normalized weights of all paths is equal to 1, forming the delay weight distribution sequence {NW1, NW2,..., NWn}. In a network containing 5 paths, assuming that the calculated weight coefficients are W1=0.368, W2=0.223, W3=0.166, W4=0.135, and W5=0.108, then Sum_W=1.0, and the normalized weight distribution sequence is {0.368, 0.223, 0.166, 0.135, 0.108}.

[0134] In practical applications, for example, in an automatic driving vehicle communication network, the end-to-end delays of four available transmission paths are measured as 18 ms, 22 ms, 25 ms and 20 ms. By calculating the delay difference values, the delay difference value set of path 1 is {-4, -7, -2}, all of which are negative values, indicating that the delay performance of path 1 is better than that of all other paths; the delay difference value set of path 2 is {4, -3, 2}, the positive value subset is {4, 2}, and the negative value subset is {-3}; the delay difference value set of path 3 is {7, 3, 5}, all of which are positive values, indicating that the delay performance of path 3 is the worst; and the delay difference value set of path 4 is {2, -2, -5}, the positive value subset is {2}, and the negative value subset is {-2, -5}.

[0135] The delay disadvantage value and the advantage value of each path are calculated: PSum1 = 0, NSum1 = -13 for path 1; PSum2 = 6, NSum2 = -3 for path 2; PSum3 = 15, NSum3 = 0 for path 3; and PSum4 = 2, NSum4 = -7 for path 4. Accordingly, the weight coefficients are calculated: W1 = e (-(0 / 13)) = 1, W2 = e (-(6 / 3)) ≈ 0.135, W3 = 1 (since NSum3 = 0), and W4 = e (-(2 / 7)) ≈ 0.751. The sum of the weight coefficients Sum_W = 2.886, and after normalization, the delay weight distribution sequence {0.347, 0.047, 0.347, 0.260} is obtained.

[0136] The delay weight distribution sequence is applied to network resource allocation, such as bandwidth allocation or data packet scheduling priority setting. For example, in the case of a total bandwidth of 100 Mbps, path 1 is allocated 34.7 Mbps, path 2 is allocated 4.7 Mbps, path 3 is allocated 34.7 Mbps, and path 4 is allocated 26.0 Mbps. This allocation fully considers the delay characteristics of each path and prioritizes paths with better delay performance to obtain more resources.

[0137] In this embodiment, the time-sensitive network can dynamically adjust the resource allocation strategy according to the path delay performance, improve the real-time communication efficiency and reliability of the automatic driving system, meet the strict delay requirements of vehicle control, environmental perception and other functions, and provide a solid network foundation for safe driving.

[0138] In an alternative embodiment, the delay weight distribution sequence is iteratively updated based on hierarchical time scales, candidate transmission paths are scored based on node load rate data and node delay evaluation parameters, and the candidate transmission path with the highest comprehensive score is selected as the optimal transmission path, comprising:

[0139] The time delay weight distribution sequence of the candidate transmission path is obtained, and the time delay weight distribution sequence and the current path time delay state form a state parameter;

[0140] The weight learning is performed on each candidate transmission path, the current weight value is calculated according to the time delay weight distribution sequence of the candidate transmission path, the ratio of the historical transmission success rate of the candidate transmission path to the weight fluctuation rate is taken as a weight prediction coefficient, and the current weight value is corrected based on the weight prediction coefficient to determine an expected weight value;

[0141] A weight update mechanism of a hierarchical time scale is constructed, the current weight value is updated based on the path state change in a first time scale, the calculation parameter of the expected weight value is updated based on the path stability in a second time scale, and the weight value after iteration is obtained by weighted summation of the weight mean value of the first time scale and the weight prediction value of the second time scale;

[0142] The load rate data of each node on the candidate transmission path is collected, and the node time delay evaluation parameter is calculated based on the data transmission rate and the data packet round trip time between nodes;

[0143] The load balancing score is calculated according to the load rate data, the time delay score is calculated based on the node time delay evaluation parameter, and the weighted sum of the load balancing score and the time delay score is taken as the path basic score;

[0144] The path comprehensive score is obtained by multiplying the path basic score by the weight value after iteration, and the candidate transmission path with the highest path comprehensive score and greater than a preset score threshold is selected as the optimal transmission path.

[0145] In a specific embodiment, the time delay weight distribution sequence of the candidate transmission path is obtained, which reflects the weight distribution of the time delay of each node on the path. The time delay weight distribution sequence can be represented by a set of numerical values, for example, for a path including 5 nodes, its time delay weight distribution sequence can be [0.15, 0.22, 0.18, 0.25, 0.20]. These weight values and the time delay state of the current path form a state parameter, and the time delay state can include the average time delay, packet loss rate and other network quality indicators on the path.

[0146] When performing weight learning for each candidate transmission path, the current weight value is calculated according to the time delay weight distribution sequence of the candidate transmission path. For example, the weighted average of the weights of each node in the sequence is taken as the current weight value. If the weight distribution is [0.15, 0.22, 0.18, 0.25, 0.20] and the actual measured time delays of each node are [5ms, 8ms, 6ms, 9ms, 7ms] respectively, the current weight value can be calculated as 0.15x5+0.22x8+0.18x6+0.25x9+0.20x7=7.17ms. The historical transmission success rate and weight fluctuation rate of the candidate transmission path are obtained. Assuming that the historical transmission success rate is 0.95 and the weight fluctuation rate is 0.08, the weight prediction coefficient is 0.95 / 0.08=11.875. The current weight value is corrected using this prediction coefficient, for example, the expected weight value 6.85ms is obtained by a linear adjustment formula.

[0147] A weight update mechanism of hierarchical time scales is constructed, including two time scales. In the first time scale (such as 10 seconds), the path state change is monitored every second and the current weight value is updated. Assuming that in these 10 seconds, the measured weight value sequence is [7.17, 7.22, 7.15, 7.18, 7.25, 7.20, 7.16, 7.19, 7.21, 7.23], the average is 7.196ms. In the second time scale (such as 1 minute), the path stability is evaluated every 10 seconds, and the weight prediction parameter is updated. If the path is stable within 1 minute, the system may adjust the weight prediction coefficient from 11.875 to 12.5, and calculate the weight prediction value as 6.80ms based on the new coefficient. The weight average of the first time scale and the weight prediction value of the second time scale are weighted and summed, assuming that the weighting coefficients are 0.4 and 0.6, then the iterative weight value is 0.4x7.196+0.6x6.80=6.958ms.

[0148] The load rate data of each node on the candidate transmission path is collected, including CPU usage, memory occupancy, network bandwidth usage, etc. For example, the CPU usage of the 5 nodes on the path is [45%, 60%, 38%, 52%, 41%], the memory occupancy is [50%, 65%, 42%, 58%, 46%], and the network bandwidth usage is [35%, 48%, 30%, 42%, 37%]. At the same time, the node time delay evaluation parameter is calculated based on the data transmission rate and the data packet round trip time between nodes. If the data transmission rate between two nodes is 100Mbps, the data packet size is 1KB, and the round trip time is 10ms, the time delay evaluation parameter of the link can be calculated as 0.8.

[0149] The load balancing score is calculated according to the load rate data. The calculation method can be to take the inverse of the standard deviation of the load rate of each node, and the smaller the standard deviation, the more balanced the load, and the higher the score. For example, the standard deviation of the CPU usage rate described above is 8.44%, and after normalization, the load balancing score is 0.75. The latency score is calculated based on the node latency evaluation parameter, and the weighted average of each link latency evaluation parameter can be taken, for example, to obtain a latency score of 0.82. Assuming that the weights of the load balancing score and the latency score are 0.4 and 0.6 respectively, the path basic score is 0.4 x 0.75 + 0.6 x 0.82 = 0.792.

[0150] The path basic score is multiplied by the weight value after iteration to obtain the path comprehensive score. Assuming that the weight value after iteration is 6.958ms, for the convenience of calculation, it is normalized to 0.85, and the path comprehensive score is 0.792 x 0.85 = 0.6732. If the preset score threshold is 0.65, and the score of the path is 0.6732, which is higher than the threshold, and is the highest among all candidate paths, the path is selected as the optimal transmission path.

[0151] When changes in network state are detected, such as a sudden increase in the load of a certain node by more than 20% or an increase in link latency by more than 30%, the latency weight distribution sequence is reacquired and the path optimization process is performed. Through this dynamic adjustment mechanism, changes in the network environment can be adapted to, the optimal transmission path is always selected, and the efficiency and reliability of data transmission are improved.

[0152] The time-sensitive network latency control system for automatic driving according to the embodiment of the application comprises:

[0153] The first unit is configured to acquire data packet transmission latency information and network topology information of the automatic driving vehicle, and calculate node latency evaluation parameters according to the transmission latency information;

[0154] The second unit is configured to generate a network node connection relationship graph according to the network topology information, hierarchize the nodes based on the connection degree values of the nodes in the network node connection relationship graph, perform a state space search operation to generate a node state transition sequence, and obtain a candidate transmission path set;

[0155] The third unit is configured to collect historical latency data of each candidate transmission path, perform mean calculation based on a preset time window, and obtain predicted latency values of each candidate transmission path through weighted calculation;

[0156] The fourth unit is configured to determine latency weight coefficients of each candidate transmission path through subset division based on a comparison result of latency difference values between the candidate transmission paths, and generate a latency weight distribution sequence;

[0157] The fifth unit is configured to iteratively update the time delay weight distribution sequence based on a hierarchical time scale, score the candidate transmission paths according to the node load rate data and the node time delay evaluation parameters, and select the candidate transmission path with the highest comprehensive score as the optimal transmission path.

[0158] The sixth unit is configured to forward the data packet through the optimal transmission path, and when detecting that the real-time time delay change value exceeds the preset time delay change range, recompute the time delay weight coefficients of the paths by using the latest collected historical time delay data and the node time delay evaluation parameters.

[0159] In a third aspect, an electronic device is provided, comprising:

[0160] a processor;

[0161] a memory for storing processor-executable instructions;

[0162] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0163] In a fourth aspect, a computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0164] The present application can be a method, apparatus, system and / or computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions stored therein, which are used to perform various aspects of the present application.

[0165] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A time-sensitive network delay control method for autonomous driving, characterized in that, include: Obtain data packet transmission latency information and network topology information of autonomous vehicles, and calculate node latency evaluation parameters based on the transmission latency information; A network node connection graph is generated based on the network topology information. The nodes are then layered based on their connectivity values ​​in the graph. A state space search operation is performed to generate a node state transition sequence, resulting in a set of candidate transmission paths. Historical latency data for each candidate transmission path is collected, the mean is calculated based on a preset time window, and the predicted latency value for each candidate transmission path is obtained through weighted calculation. Based on the comparison results of delay differences between candidate transmission paths, the delay weight coefficient of each candidate transmission path is determined by subset partitioning, and a delay weight distribution sequence is generated. The delay weight distribution sequence is iteratively updated based on a hierarchical time scale. Candidate transmission paths are scored based on node load rate data and node delay evaluation parameters. The candidate transmission path with the highest comprehensive score is selected as the optimal transmission path. Data packets are forwarded via the optimal transmission path. When the real-time latency change value is detected to exceed the preset latency change range, the latency weight coefficient of each path is recalculated using the latest collected historical latency data and node latency evaluation parameters.

2. The method according to claim 1, characterized in that, Based on the connectivity values ​​of nodes in the network node connection graph, nodes are hierarchically classified, and a state space search operation is performed to generate node state transition sequences, resulting in a set of candidate transmission paths, including: The node connectivity value is obtained by counting the number of connecting edges of each node in the network node connection graph. The nodes in the network node connection graph are divided into multiple connection layers according to a preset first connectivity threshold and a preset second connectivity threshold. Obtain the geographic coordinate information of each node, perform regional clustering operations on the nodes in each connection layer, calculate the Euclidean distance between nodes, and generate a node regional distribution sequence; Calculate the Euclidean distance between each node, obtain the connectivity ratio of each node, record the historical successful transmission count and total transmission count between each node, and calculate the transmission state transition value using weighted average. The inter-node transition probability is determined based on the transmission state transition value. A state space search operation is performed based on the dynamic search depth value to generate a node state transition sequence. The cost value of each transition is calculated. The node state transition sequence is optimized based on the cost value. The state space is traversed and searched multiple times using a random sampling method. State transition paths whose replacement value meets a preset cost threshold are selected. The state transition paths are then mapped to the actual node access order to generate a set of candidate transmission paths.

3. The method according to claim 2, characterized in that, The process involves determining the inter-node transition probability based on the transmission state transition value, performing a state space search operation based on the dynamic search depth value to generate a node state transition sequence, calculating the cost value of each transition, and optimizing the node state transition sequence based on the cost value, including: Determine the inter-node transition probability based on the transmission state transition value; Using the starting node as the search starting point, perform state space search operation, set the starting node as the current search node, and initialize the dynamic search depth value; Based on the inter-node transition probability, the next hop node of the current search node is selected, and the next hop node is recorded in the set of visited nodes; the historical access frequency of the next hop node is counted, and when the historical access frequency is greater than a preset access threshold, the inter-node transition probability of the next hop node is multiplied by a preset attenuation coefficient. When the next hop node is the target node, the current search path is recorded as a node state transition sequence; when the next hop node is not the target node and is not in the set of visited nodes, the next hop node is set as the current search node, the dynamic search depth value is updated based on the preset depth adjustment coefficient, and the state space search operation continues; when the next hop node is in the set of visited nodes, the process returns to the previous level node to reselect the next hop node; when the search depth reaches the dynamic search depth value, the path cost is calculated based on the cost value of each node in the set of visited nodes and the distance to the target node; when the path cost is greater than the preset path cost threshold, the process returns to the previous level node to reselect the next hop node. Calculate the cost of each transition in the node state transition sequence; The node state transition sequence is optimized based on the cost value, and the node state transition sequence with a cost value less than a preset cost threshold and a node hop count less than a preset hop count threshold is selected as the final node state transition sequence.

4. The method according to claim 1, characterized in that, Historical latency data for each candidate transmission path is collected, and the mean is calculated based on a preset time window. The predicted latency values ​​for each candidate transmission path are obtained through weighted calculation, including: Historical latency data of each candidate transmission path is collected, and the historical latency data is periodically sampled according to a preset sampling time interval to generate a latency sampling sequence; Based on a preset first time window and a preset second time window, the time-delayed sampling sequence is divided into data segments, wherein the second time window is larger than the first time window; The average values ​​of the sampled data within the first time window and the sampled data within the second time window are calculated to determine the short-term delay average and the long-term delay average, respectively. Calculate the deviation rate between the short-term delay mean and the long-term delay mean, and determine the delay prediction weighting coefficient based on the deviation rate; The predicted delay weighting coefficient is weighted and calculated with the current delay sample value to obtain the predicted delay value of each candidate transmission path; The prediction accuracy of the predicted delay value is statistically analyzed. When the prediction accuracy is lower than a preset accuracy threshold, the delay prediction weight coefficient is updated based on a preset adjustment step size, and the predicted delay value is recalculated.

5. The method according to claim 1, characterized in that, Based on the comparison results of delay differences among candidate transmission paths, the delay weight coefficient of each candidate transmission path is determined through subset partitioning, and a delay weight distribution sequence is generated, including: Calculate the delay difference between each candidate transmission path and all other candidate transmission paths, and establish a delay difference set for each transmission path; Compare the positive and negative attributes of each delay difference in the delay difference set of each candidate transmission path, and classify delay differences greater than zero into the positive value subset and delay differences less than zero into the negative value subset; The latency disadvantage value is obtained by summing the latency differences in the positive subset of each candidate transmission path, and the latency advantage value is obtained by summing the latency differences in the negative subset. Based on the ratio of delay disadvantage value to delay advantage value, the weight coefficient of the corresponding candidate transmission path is determined by exponential operation; the weight coefficients of all candidate transmission paths are normalized to determine the delay weight distribution sequence.

6. The method according to claim 1, characterized in that, The latency weight distribution sequence is iteratively updated based on a hierarchical time scale. Candidate transmission paths are scored based on node load rate data and node latency evaluation parameters. The candidate transmission path with the highest comprehensive score is selected as the optimal transmission path, including: Obtain the delay weight distribution sequence of the candidate transmission path, and combine the delay weight distribution sequence with the current path delay state to form a state parameter; For each candidate transmission path, weight learning is performed. The current weight value is calculated based on the delay weight distribution sequence of the candidate transmission path. The ratio of the historical transmission success rate of the candidate transmission path to the weight volatility is used as the weight prediction coefficient. The current weight value is corrected based on the weight prediction coefficient to determine the expected weight value. A weight update mechanism with hierarchical time scales is constructed. The current weight value is updated based on the path state change in the first time scale, and the calculation parameters of the expected weight value are updated based on the path stability in the second time scale. The weight mean of the first time scale and the weight prediction value of the second time scale are weighted and summed to obtain the iterative weight value. Collect load rate data of each node on the candidate transmission path, and calculate node latency evaluation parameters based on the data transmission rate between nodes and the round-trip time of data packets; The load balancing score is calculated based on the load rate data, and the latency score is calculated based on the node latency evaluation parameters. The weighted sum of the load balancing score and the latency score is used as the path base score. The path base score is multiplied by the iterative weight value to obtain the path comprehensive score. The candidate transmission path with the highest path comprehensive score that is greater than the preset score threshold is selected as the optimal transmission path.

7. A time-sensitive network delay control system for autonomous driving, used to implement the method of any one of claims 1-6, characterized in that, include: The first unit is used to obtain data packet transmission delay information and network topology information of autonomous vehicles, and to calculate node delay evaluation parameters based on the transmission delay information. The second unit is used to generate a network node connection diagram based on network topology information, to classify nodes based on the connectivity values ​​of nodes in the network node connection diagram, to perform state space search operation to generate node state transition sequences, and to obtain a set of candidate transmission paths. The third unit is used to collect historical latency data of each candidate transmission path, calculate the mean based on a preset time window, and obtain the predicted latency value of each candidate transmission path through weighted calculation. The fourth unit is used to determine the delay weight coefficient of each candidate transmission path by subset partitioning based on the comparison results of delay difference between candidate transmission paths, and to generate a delay weight distribution sequence. The fifth unit is used to iteratively update the delay weight distribution sequence based on a hierarchical time scale, score candidate transmission paths based on node load rate data and node delay evaluation parameters, and select the candidate transmission path with the highest comprehensive score as the optimal transmission path. The sixth unit is used to forward data packets through the optimal transmission path. When the real-time delay change value is detected to exceed the preset delay change range, the delay weight coefficient of each path is recalculated using the latest collected historical delay data and node delay evaluation parameters.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images