A communication load detection system and method for vehicle networking environment

Abstract

The present invention provides a communication load detection system and method for an Internet of Vehicles environment, comprising obtaining communication link status information of a node including a current connection protocol type and a signal strength; extracting corresponding communication load data including a data packet size and a transmission frequency from the node according to the protocol type in the communication link status information; establishing a new communication link through an optimal communication protocol, and recording load distribution characteristics including a mean value of the data packet size and a transmission frequency variance before and after the switching; updating adaptive detection model parameters using a gradient descent method according to the load distribution characteristics, with an objective function of minimizing a load change rate prediction error; deploying the updated adaptive detection model to the Internet of Vehicles, detecting the communication load in real time, and continuously monitoring the model performance after deployment for optimization; in this way, the communication protocol can be dynamically adjusted according to the network environment, thereby improving the communication efficiency and stability of the Internet of Vehicles.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a communication load detection system and method for an Internet of Vehicles environment. Background Art

[0002] In the Internet of Vehicles environment, devices are widely distributed and highly mobile, resulting in frequent switching of communication links. This dynamicity makes the distribution of communication load highly uneven and bursty. When vehicles move between different areas, the communication demand will change significantly due to factors such as geographical location, traffic density, and user behavior. For example, in the central area of ​​the city, the vehicle density is high and the communication demand is concentrated, while in the suburbs or on highways, the vehicle distribution is sparse and the communication demand is relatively dispersed. This uneven load distribution makes it difficult for traditional static load detection methods to adapt. In addition, multiple communication protocols coexist in the Internet of Vehicles, including cellular networks, dedicated short-range communications, and low-power wide area networks. These protocols have significant differences in coverage, transmission rate, and power consumption, resulting in different load characteristics. For example, cellular networks are suitable for large-scale coverage, but may face high latency problems; dedicated short-range communications have high transmission efficiency within a short distance, but limited coverage; low-power wide area networks are suitable for low-rate, long-distance communication needs. This coexistence of multiple protocols further increases the complexity of load detection.

[0003] Therefore, how to design a detection method that can adapt to different communication protocols and reflect the changes in communication load in real time in a dynamically changing network environment has become the core technical problem currently faced. Summary of the invention

[0004] The purpose of the present invention is to overcome the above-mentioned shortcomings and provide a communication load detection system and method for a vehicle networking environment, which can dynamically adjust the communication protocol according to the network environment to improve the efficiency and stability of vehicle networking communication.

[0005] A first aspect of the present invention provides a communication load detection method for a vehicle networking environment, which mainly includes the following steps:

[0006] Obtain the node's communication link status information including the current connection protocol type and signal strength;

[0007] According to the protocol type, extract the corresponding communication load data including the data packet size and transmission frequency;

[0008] For the communication load data, a sliding window mechanism is used to calculate the load change rate within each time window. The load change rate is defined as the weighted change value of the packet size and transmission frequency within the window.

[0009] If the load change rate exceeds the preset threshold, the protocol switching detection mode is triggered, and the optimal communication protocol is selected according to the network environment through the pre-established protocol switching model;

[0010] Establish a new communication link through the optimal communication protocol, and record the load distribution characteristics before and after the switch, including the mean value of the packet size and the variance of the transmission frequency;

[0011] According to the load distribution characteristics, the gradient descent method is used to update the parameters of the adaptive detection model, and the objective function is to minimize the load change rate prediction error;

[0012] Deploy the updated adaptive detection model to the Internet of Vehicles to detect the communication load in real time, and continuously monitor the performance of the adaptive detection model after deployment for optimization.

[0013] Optionally, for the communication load data, a sliding window mechanism is used to calculate the load change rate in each time window, where the load change rate is defined as a weighted change value of the data packet size and the transmission frequency in the window, including:

[0014] For communication load data, a sliding window mechanism is used to divide the time window;

[0015] The network traffic is monitored using a preset time window to obtain the data packet size value and the transmission frequency value, and the size value and the frequency value are weighted according to a pre-established weight model to obtain a weighted value;

[0016] The weighted value is input into the load change rate calculation model, and the load change rate in the current time window is calculated by a linear regression algorithm.

[0017] Optionally, it also includes building an adaptive detection model based on the load distribution characteristics using a support vector machine algorithm, and training the adaptive detection model, the training data includes the load distribution characteristics before and after the switching, and the support vector machine uses a radial basis kernel function.

[0018] Optionally, if the load change rate exceeds a preset threshold, a protocol switching detection mode is triggered, and an optimal communication protocol is selected according to the network environment through a pre-established protocol switching model, including:

[0019] The input features of the pre-established protocol switching model include signal strength, data packet size and transmission frequency, and the feature selection criterion is to maximize information gain;

[0020] Obtain the load change rate in the network environment of the Internet of Vehicles. If the load change rate exceeds a preset threshold, the protocol is triggered to switch to a detection mode.

[0021] In the protocol switching detection mode, the signal strength, data packet size and transmission frequency in the network environment are collected, and these feature data are processed using the minimum and maximum normalization method;

[0022] The normalized data is input into a pre-established protocol switching model to determine the information gain value of each feature for protocol selection;

[0023] Based on the information gain maximization standard, the communication protocol corresponding to the feature with the largest information gain is selected as the optimal protocol, and a protocol switching decision is generated;

[0024] Convert the protocol switching decision into a network configuration instruction, use the network configuration protocol to send the instruction to the network device, and the device completes the dynamic switching of the communication protocol according to the instruction;

[0025] After the switching is completed, the load change rate in the network environment is re-obtained to determine whether it meets the preset threshold;

[0026] If not, repeat the above process until the load change rate stabilizes within the preset threshold.

[0027] Optionally, establishing a new communication link through an optimal communication protocol and recording load distribution characteristics including a mean value of a data packet size and a transmission frequency variance before and after the switching include:

[0028] Establish a new communication link for the Internet of Vehicles under the optimal communication protocol, and obtain the mean value of the packet size and the variance of the transmission frequency before switching;

[0029] If the new communication link is established successfully, the mean value of the packet size and the variance of the transmission frequency after the switch are collected;

[0030] According to the mean values ​​of the data packet sizes before and after the switching, the mean value change is calculated to determine the first characteristic value of the load distribution characteristic;

[0031] According to the transmission frequency variance before and after the switching, the variance change is calculated to determine the second characteristic value of the load distribution characteristic;

[0032] Inputting the first eigenvalue and the second eigenvalue into a pre-established load distribution characteristic model to determine whether the load distribution characteristic is stable;

[0033] If the load distribution characteristics are stable, a load distribution characteristic record is generated and stored in the network environment database;

[0034] According to the load distribution feature records, the pre-established protocol switching model is updated and the information gain calculation standard is optimized.

[0035] Optionally, based on the load distribution characteristics, a support vector machine algorithm is used to build an adaptive detection model, and the adaptive detection model is trained, the training data includes the load distribution characteristics before and after the switching, and the support vector machine uses a radial basis kernel function, including:

[0036] Obtain load distribution characteristic data before and after the switching, extract the load amount, perform normalization processing on the load amount, and obtain standardized load data;

[0037] Extract histograms and statistical characteristic values ​​from the standardized load data to determine training set samples;

[0038] Use the support vector machine algorithm, set the radial basis kernel function parameters, and build a model training framework;

[0039] According to the training set samples and the model training framework, the model training process is executed to obtain a preliminary training model;

[0040] If the accuracy of the initial training model is lower than the preset threshold, the kernel function parameters are adjusted and the training process is re-executed;

[0041] Obtain the optimized training model and determine whether the model performance indicators meet the requirements;

[0042] If the requirements are met, the optimized training model is determined as the adaptive detection model.

[0043] Optionally, the method of updating the adaptive detection model parameters by using a gradient descent method according to the load distribution characteristics, wherein the objective function is minimization of the load change rate prediction error, includes:

[0044] Obtain load distribution characteristic data, and extract load change rate based on the characteristic data;

[0045] Adopting an adaptive detection model to predict the load change rate and obtain an initial prediction value;

[0046] Calculate the prediction error based on the initial prediction value and the actual load change rate;

[0047] If the prediction error exceeds the preset threshold, the gradient descent method is used to update the adaptive detection model parameters;

[0048] The load change rate is re-predicted through the updated adaptive detection model parameters to obtain the optimized prediction value;

[0049] According to the optimized prediction value, determine whether the error meets the minimization condition. If not, update the adaptive detection model parameters cyclically.

[0050] The optimized adaptive detection model is used to output the final load change rate prediction result.

[0051] Optionally, deploying the updated adaptive detection model to the Internet of Vehicles to detect the communication load in real time includes:

[0052] Use the TensorRT model conversion tool to convert the updated adaptive detection model to ONNX format;

[0053] During the conversion process, the adaptive nature of the adaptive detection model is preserved using TensorRT’s quantization capabilities.

[0054] Use CUDA to configure the inference parameters of the converted model according to the hardware architecture and computing resources of the connected car system;

[0055] Deploy the converted model to the Internet of Vehicles system, configure the data interface between the converted model and the vehicle communication module, and realize the real-time collection and preprocessing of communication data;

[0056] Obtain preprocessed data from the vehicle communication module, perform inference calculation of the model through the vehicle computing unit, and obtain the detection result of the communication load;

[0057] The test results are transmitted to the vehicle display module for real-time presentation.

[0058] Optionally, the continuously monitoring the model performance after deployment for optimization includes:

[0059] Use Prometheus to establish a model performance monitoring mechanism, collect performance indicators such as accuracy, latency, and resource usage of the converted model reasoning, and store performance data for subsequent analysis;

[0060] If the performance monitoring data is lower than the preset threshold, the model optimization process is started, and the parameters of the adaptive detection model are incrementally updated using the stochastic gradient descent algorithm to improve the adaptability of the adaptive detection model in the vehicle environment;

[0061] During the model optimization process, the optimized performance indicators are monitored. If the performance meets the preset requirements, the optimization process is completed.

[0062] A second aspect of the present invention provides a communication load detection system for a vehicle networking environment, comprising:

[0063] A status acquisition module is used to acquire the communication link status information of the node including the current connection protocol type and signal strength;

[0064] A data extraction module, used to extract corresponding communication load data including data packet size and transmission frequency according to the protocol type;

[0065] The change rate calculation module is used to calculate the load change rate in each time window using a sliding window mechanism for communication load data. The load change rate is defined as a weighted change value of the data packet size and the transmission frequency in the window.

[0066] A switching detection module is used to trigger a protocol switching detection mode when the load change rate exceeds a preset threshold, and select the optimal communication protocol according to the network environment through a pre-established protocol switching model;

[0067] A link establishment module, used to establish a new communication link through an optimal communication protocol and record load distribution characteristics including the mean value of data packet size and transmission frequency variance before and after switching;

[0068] The parameter updating module is used to update the parameters of the adaptive detection model using the gradient descent method according to the load distribution characteristics of the IoV nodes, and the objective function is to minimize the load change rate prediction error;

[0069] The model deployment and monitoring module is used to deploy the updated adaptive detection model to the Internet of Vehicles, detect the communication load in real time, and continuously monitor the model performance after deployment for optimization.

[0070] Beneficial effects of the present invention: The present invention obtains the communication link status information of each node, extracts the communication load data, and adopts the sliding window mechanism to calculate the load change rate. When the load change rate exceeds the threshold, the protocol switching detection mode is triggered, and the optimal communication protocol is selected using the pre-established protocol switching model, and the load distribution characteristics before and after the switching are recorded. The support vector machine algorithm is used to build an adaptive detection model, and the gradient descent method is used to update the model parameters. Finally, the updated model is deployed to the Internet of Vehicles for real-time detection and continuous optimization. In this way, the communication protocol can be dynamically adjusted according to the network environment, the communication efficiency and stability of the Internet of Vehicles can be improved, and an intelligent communication protocol management solution is provided for the complex and changeable Internet of Vehicles environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 The present invention is a flow chart of a communication load detection method for a vehicle networking environment. DETAILED DESCRIPTION

[0072] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments. The present invention is further described in detail in conjunction with the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are only used to explain the relevant inventions, rather than to limit the invention. It is also necessary to explain that, for the convenience of description, only the parts related to the invention are shown in the accompanying drawings.

[0073] like Figure 1As shown, the communication load detection method for a vehicle networking environment described in this embodiment specifically includes the following steps:

[0074] Step S101, obtaining the communication link status information of the node including the current connection protocol type and signal strength.

[0075] Specifically, the communication link status information of the target node is obtained from the Internet of Vehicles according to the node number. The status information includes the protocol type and signal strength value. For the obtained protocol type, if the protocol type is TCP protocol, the transport layer packet analysis algorithm is used to extract the delay value and bit error rate. For the obtained signal strength value, if the signal strength value is lower than the preset threshold, the network layer routing optimization algorithm is used to update the routing table and adjust the bandwidth value. According to the delay value and bit error rate, the link status evaluation model is used to determine whether the link status is normal. If the link status is abnormal, the device code matching algorithm is used to find the adjacent nodes of the target node as a backup link. Through the preset network layer protocol, the routing table of the backup link is updated to the transport layer of the target node. According to the updated routing table and bandwidth value, the signal strength value is re-evaluated to obtain the final communication link status information.

[0076] For example, in the Internet of Vehicles, the management of the communication link status involves technical points at multiple levels. In the process of node acquisition, taking the vehicle node as an example, each node has a unique identification code, and the communication status of the target vehicle can be obtained through node scanning. The protocol type and signal strength value of the communication link are key indicators. The protocol type determines the way of data transmission, and the signal strength value reflects the communication quality. In terms of protocol analysis, the transport layer packet analysis focuses on delay and bit error. For example, when a vehicle node adopts traditional Internet of Vehicles communication, the delay value is usually maintained at the millisecond level. For example, the measured delay value is 20 milliseconds and the bit error rate is one thousandth. These indicators directly affect the communication quality. When the signal strength value is detected to be lower than the preset threshold, such as the signal strength is lower than -85 decibels, it is necessary to start routing optimization. The routing optimization algorithm updates the routing table by analyzing the network topology and adjusts the bandwidth allocation accordingly. The link status evaluation model adopts a multi-dimensional evaluation system. For example, the delay threshold is set to 50 milliseconds and the bit error rate threshold is set to five thousandths. Any indicator exceeding the threshold will trigger an abnormality judgment. When an abnormality is found, it is necessary to quickly locate the backup link. The device code matching algorithm finds the best candidate node by traversing the device feature codes of adjacent nodes. For example, if a vehicle node detects that the current link is abnormal, the algorithm will automatically search for surrounding vehicle nodes and select the node with the strongest signal and lightest load as the backup link. Updating the routing table is a key step to ensure communication reliability. The network layer protocol is responsible for the transmission of routing information and needs to ensure that the new routing table can be accurately delivered to the target node. In specific implementation, the updated routing table can be encapsulated in a dedicated data packet for transmission through the control signaling channel. For example, if a vehicle node needs to switch the communication link, the Internet of Vehicles system will send the new routing information to the target node through the control channel to ensure smooth routing switching. The final evaluation process must ensure that the communication quality is substantially improved. By comparing the bandwidth utilization and signal strength before and after the update, the optimization effect can be quantified. For example, before optimization, the bandwidth utilization was 60% and the signal strength was -85 decibels. After link switching and bandwidth adjustment, the bandwidth utilization increased to 80% and the signal strength increased to -70 decibels, indicating that the link status has been effectively improved. This dynamic optimization mechanism can improve the overall communication reliability of the Internet of Vehicles and ensure driving safety.

[0077] Step S102: extract corresponding communication load data including data packet size and transmission frequency from the node according to the protocol type in the communication link status information.

[0078] Specifically, the packet size and transmission frequency are extracted from the target node according to the protocol type. If the protocol type is TCP, the transport layer packet analysis algorithm is used to extract the delay value and bit error rate. According to the delay value and bit error rate, the link state evaluation model is used to determine whether the link state is normal. If the link state is abnormal, the device code matching algorithm is used to find the adjacent points of the target node. Through the preset network layer protocol, the routing table of the adjacent points is updated to the transport layer of the target node. According to the updated routing table and bandwidth value, the signal strength value is re-evaluated. According to the re-evaluated signal strength value and packet size, the transmission frequency is adjusted to obtain the optimized communication load data.

[0079] Exemplarily, the communication protocol analysis first needs to accurately identify the properties of the node transmission data packet. For example, through the transport layer protocol analysis, the current vehicle node uses fifty packets of one thousand bytes per second. The end-to-end delay is measured by the detection packet. Through the difference between the sending and receiving timestamps, the current link delay of the node is obtained to be sixty milliseconds, and the bit error rate is 2 / 10,000 through the receiver's checksum statistics. The link status evaluation model uses the delay value and the bit error rate as key indicators, and combines the preset threshold to judge the link status. In the Internet of Vehicles communication, the link status indicator score adopts a ten-point system. The current node link delay value exceeds the preset fifty millisecond threshold, and the bit error rate is also higher than the standard value of one in ten thousand. According to the scoring rules, it gets five points and is in a critical state. To ensure the quality of communication, it is necessary to find a backup link. The node device code is DEV001, and the adjacent nodes are queried in the topology database through the device code matching algorithm. A backup node numbered DEV002 is found. The current link load of this node is low and can be used as an alternative communication path.

[0080] The open shortest path first protocol is used to send the routing information of the backup node to the target node. The routing table update includes the new next hop address, outbound interface identifier, and link metric. The original link bandwidth is adjusted from 10 megabits per second to 5 megabits per second, and the bit error rate is improved by reducing the bandwidth. When re-evaluating the signal strength, the impact of route changes and bandwidth adjustments needs to be considered. The signal strength value is measured by the received signal strength indicator value. The original signal strength value is negative 75 decibel milliwatts, which is increased to negative 68 decibel milliwatts after route optimization, meeting the communication quality requirements. Based on the optimized network environment, load balancing is performed on data packet transmission. The original transmission frequency of 50 data packets per second is adjusted to 40 per second, and the size of a single data packet remains unchanged at 1,000 bytes. By reducing the transmission frequency, the link load is further reduced, the delay value is reduced to 45 milliseconds, the bit error rate is reduced to 0.5 per 10,000, and the link status score is increased to 9 points, achieving the expected goal of link optimization.

[0081] This optimization process reflects the adaptive adjustment capability of the Internet of Vehicles communication. Through multi-level state perception and dynamic adjustment, it ensures the communication quality while ensuring data transmission efficiency. This link optimization mechanism based on real-time monitoring provides strong support for the stable operation of the Internet of Vehicles system.

[0082] Step S103, for the communication load data, a sliding window mechanism is used to calculate the load change rate in each time window, where the load change rate is defined as a weighted change value of the data packet size and the transmission frequency in the window.

[0083] Specifically, for communication load data, a sliding window mechanism is used to divide the time window. The sliding window size can be set to 1 second and the sliding step size can be set to 0.5 seconds. The weighted values ​​of the packet size and transmission frequency are obtained within the time window, and the load change rate is calculated. The network traffic is monitored using a preset time window to obtain the packet size value and transmission frequency value. The size value and frequency value are weighted according to the pre-established weight model to obtain the weighted value. The weighted value is input into the load change rate calculation model, and the load change rate within the current time window is calculated using a linear regression algorithm.

[0084] The weighted change value is calculated using a pre-established weight model. The process of establishing the weight model is as follows: historical Internet of Vehicles communication data, including packet size and transmission frequency information at different times and nodes, is collected, the impact of the data on load changes is analyzed, and the weight values ​​of packet size and transmission frequency are determined using a linear regression method.

[0085] Exemplarily, the sliding window mechanism monitors the communication data of vehicle nodes in the Internet of Vehicles in real time, collects data by setting a one-second time window, and slides back every 0.5 seconds to achieve continuous sampling of network traffic. During the window movement, communication load information is continuously collected, including data packet size and transmission frequency. For example, in the communication monitoring of vehicle nodes, load data of fifty data packets per second and each data packet size of one thousand bytes is collected during a single sliding window. The establishment of the weight model is based on historical data analysis, and the degree of influence of each indicator is determined by collecting communication data of vehicle nodes in different scenarios. In the test of vehicle nodes, by analyzing one month of historical data, it is found that the weight of the impact of data packet size on load change is 0.6, and the weight of the impact of transmission frequency is 0.4. This weight distribution reflects that in the Internet of Vehicles environment, the impact of data packet size on network load is more significant. The load change rate is calculated using a linear regression method, with the weighted data packet size and transmission frequency as input variables. Taking the vehicle node as an example, in a sliding window, the original data packet size of one thousand bytes is weighted by 0.6 and the transmission frequency of fifty packets per second is weighted by 0.4 and the load change rate is 15%, indicating that the network load is on the rise in this time window. The protocol switching model is constructed using a decision tree algorithm and trained by analyzing historical network environment data. After collecting network data of vehicle nodes for three months, including signal strength, data packet size and transmission frequency at different times, the data shows that when the signal strength is lower than minus 70 dbmW, switching the communication protocol from dedicated short-range communication of the Internet of Vehicles to cellular network can improve the communication quality. By learning these rules, the decision tree model establishes a protocol selection strategy based on multi-dimensional indicators. By continuously optimizing the weight model and protocol switching strategy, the system achieves precise control of the network load. In the vehicle node application, the original communication load fluctuation range is 20%. The optimized model reduces the load fluctuation to less than 5%, and reduces the average delay from 60 milliseconds to 40 milliseconds, significantly improving the stability and reliability of the Internet of Vehicles communication.

[0086] Step S104, if the load change rate exceeds a preset threshold, the protocol switching detection mode is triggered, and the optimal communication protocol is selected according to the network environment through a pre-established protocol switching model; the input features of the pre-established protocol switching model include signal strength, data packet size and transmission frequency, and the feature selection criterion is to maximize information gain.

[0087] Specifically, the load change rate in the network environment of the Internet of Vehicles is obtained. If the load change rate exceeds the preset threshold, the protocol switching detection mode is triggered. In the protocol switching detection mode, the signal strength, data packet size and transmission frequency in the network environment are collected, and these feature data are processed using the minimum and maximum normalization method. The normalized data is input into the pre-established protocol switching model to determine the information gain value of each feature for protocol selection. The construction process of the protocol switching model is as follows: collect historical network environment data, including signal strength, data packet size, transmission frequency and corresponding optimal protocol selection results under different protocols, divide them into training sets and test sets, use the decision tree algorithm to build the initial model and optimize the performance through parameter adjustment, train the initial model through the training set, and test the trained initial model through the test set. Based on the information gain maximization standard, the communication protocol corresponding to the feature with the largest information gain is selected as the optimal protocol, and a protocol switching decision is generated. The protocol switching decision is converted into a network configuration instruction, and the instruction is sent to the network device using the network configuration protocol. The device completes the dynamic switching of the communication protocol according to the instruction. After the switching is completed, the load change rate in the network environment is re-obtained to determine whether it meets the preset threshold. If not, repeat the above process until the load change rate stabilizes within the preset threshold.

[0088] For example, monitoring of load change rate is crucial in the Internet of Vehicles. By setting the threshold reasonably, network anomalies can be discovered in time. For example, during the morning rush hour, the load change rate is monitored to rise rapidly from 40% to 120%, exceeding the preset threshold of 80%, and the protocol switching detection mode is automatically triggered. In the protocol switching detection mode, it is necessary to comprehensively collect network environment characteristics. The signal strength collected by the system is negative 65 decibels, the data packet size is 400 bytes, and the transmission frequency is 15 times per second. In order to make the features of different dimensions comparable, the minimum and maximum normalization processing is adopted. The normalized signal strength is 0.8, the data packet size is 0.6, and the transmission frequency is 0.7. The protocol switching model is constructed based on the decision tree principle. By analyzing historical data, it is found that the information gain value of signal strength for protocol selection is 0.6, the information gain value of data packet size is 0.3, and the information gain value of transmission frequency is 0.1. Since the signal strength has the maximum information gain value, the system selects the low-power wide area network protocol based on signal strength as the optimal solution. The protocol switching decision needs to be converted into specific network configuration instructions. When the LPWAN protocol is selected, a configuration instruction containing parameters such as channel, power, and modulation mode is generated. It is sent to the control unit through the network configuration protocol, and the control unit immediately switches to the new communication mode. The load change rate needs to be continuously monitored after the protocol is switched. After the switch is completed, the load change rate drops to 60%, which is still higher than the preset threshold. The system continues to collect environmental characteristics and finds that the information gain value of the packet size for the protocol selection rises to 0.5. Based on this, it switches to a low-bandwidth, high-reliability protocol, and finally stabilizes the load change rate at 40%. In this way, dynamic protocol switching can effectively respond to changes in the network environment and improve the reliability and stability of communication load detection.

[0089] Step S105: establish a new communication link through the optimal communication protocol, and record the load distribution characteristics before and after the switch, including the mean value of the data packet size and the transmission frequency variance.

[0090] Specifically, a new communication link of the Internet of Vehicles is established under the optimal communication protocol, and the mean value of the data packet size and the variance of the transmission frequency before the switch are obtained. If the new communication link is successfully established, the mean value of the data packet size and the variance of the transmission frequency after the switch are collected. According to the mean value of the data packet size before and after the switch, the change in the mean is calculated to determine the first eigenvalue of the load distribution characteristic. According to the variance of the transmission frequency before and after the switch, the change in the variance is calculated to determine the second eigenvalue of the load distribution characteristic. The first eigenvalue and the second eigenvalue are input into the pre-established load distribution characteristic model to determine whether the load distribution characteristic is stable. If the load distribution characteristic is stable, a load distribution characteristic record is generated and stored in the network environment database. According to the load distribution characteristic record, the pre-established protocol switching model is updated to optimize the information gain calculation standard.

[0091] The process of establishing the load distribution feature model includes: obtaining historical load data, extracting the time series characteristics and spatial distribution characteristics of the load, and constructing a feature vector set. For the feature vector set, the random forest algorithm is used for feature selection to determine the key feature subset. According to the key feature subset, the load distribution feature model is trained by the support vector machine algorithm. If the prediction error of the load distribution feature model exceeds the preset threshold, the gradient boosting algorithm is used to retrain the model. Real-time load data is obtained, and the trained load distribution feature model is input to obtain the load prediction result. It is judged whether the load prediction result meets the preset conditions. If not, the feature weights in the feature vector set are adjusted. According to the adjusted feature weights, the load prediction result is recalculated and the final load distribution feature is output.

[0092] For example, the switching of communication links in the Internet of Vehicles involves the monitoring and analysis of changes in various parameters. Communication quality is usually evaluated from two dimensions: the mean size of the data packet and the variance of the transmission frequency. These indicators need to be measured continuously during the establishment of a new link. For example, before the switch, the mean size of the data packet was 500 bytes and the variance of the transmission frequency was 0.3. After switching to the new link, the mean size of the data packet dropped to 300 bytes and the variance of the transmission frequency dropped to 0.1. The change in the mean was calculated to be negative 200 bytes and the change in the variance was negative 0.2. These two values ​​constitute the key indicators of the load distribution characteristics. The training process of the load distribution characteristic model requires a large amount of historical data support. The load data of vehicle nodes in different periods and different regions within a year are collected. The load fluctuation law during the morning and evening peak hours is found through time series analysis, and the spatial distribution analysis reveals the load difference between the central area of ​​the city and the suburbs or highways. When the random forest algorithm is used to screen features, it is found that the importance of indicators such as data transmission delay and packet loss rate ranks high. When training the support vector machine algorithm, the radial basis kernel function is selected to build the model, and the prediction error is controlled within 5%. For example, when the real-time load data shows that the average size of the data packet is 400 bytes and the transmission frequency variance is 0.2, after inputting these data into the trained model, the prediction results show that the load will show a stable trend in the next hour. Based on this, the system generates a load distribution feature record, including timestamp, location information, load indicators, etc., and stores it in the network environment database. These records are used to continuously optimize the protocol switching model and improve the accuracy of information gain calculation. By analyzing the load distribution feature records, it is found that the original protocol switching model is not sensitive enough to respond to emergencies. The updated model adds an emergency weight factor when calculating information gain. When a traffic accident occurs, the system can identify load changes more quickly and switch to the appropriate communication protocol in time. In this way, the system maintains communication stability while improving its ability to respond to emergencies. When the prediction results show that the load distribution does not meet the preset conditions, the system automatically adjusts the weights of features such as congestion level and user density. The recalculated prediction results are more in line with the actual situation and provide a reliable basis for protocol switching decisions. This dynamic adjustment mechanism ensures the continuous and stable operation of vehicle network communication.

[0093] Step S106, based on the load distribution characteristics, the adaptive detection model parameters are updated using the gradient descent method, and the objective function is to minimize the load change rate prediction error.

[0094] Specifically, load distribution characteristic data is obtained, and the load change rate is extracted based on the characteristic data. The load change rate is predicted using an adaptive detection model to obtain an initial prediction value. The prediction error is calculated based on the initial prediction value and the actual load change rate. If the prediction error exceeds the preset threshold, the adaptive detection model parameters are updated using the gradient descent method. The load change rate is re-predicted using the updated adaptive detection model parameters to obtain an optimized prediction value. Based on the optimized prediction value, it is determined whether the error meets the minimization condition. If not, the adaptive detection model parameters are updated cyclically. The optimized adaptive detection model is used to output the final load change rate prediction result.

[0095] Exemplarily, the load distribution characteristic data refers to the load change collected during the operation of the system, including indicators such as processor usage, memory usage, and network bandwidth. In practical applications, the load change rate can be obtained by sampling the load change every five minutes. The adaptive detection model processes these characteristic data through the support vector machine algorithm to predict the load change trend in the future. The prediction error is calculated by comparing the predicted value with the actual observed value. For example, the model predicts that the network bandwidth usage rate in a certain period of time is 60%, while the actual usage rate is 75%, and the prediction error is 15%. When this error exceeds the pre-set threshold of 10%, the model parameter update mechanism needs to be started. The gradient descent method is used to optimize the model parameters and reduce the prediction error by continuously adjusting the weights. In practical applications, if the model has a large deviation from the prediction, it may require multiple iterations of optimization. Each iteration will calculate the gradient of the error function and adjust the parameters in the opposite direction of the gradient until the error falls within an acceptable range. The model after the parameter update will re-predict and obtain a new prediction result. If the prediction result shows that the load in a certain period of time will exceed the warning line of 85%, the system can start the backup resources in advance. The minimization condition means that the prediction error must be less than a certain threshold. Assuming the threshold is set to 5%, if the error between the predicted database access volume and the actual value is still greater than this value, the model will continue to optimize the parameters. This process will be repeated until the prediction error falls within an acceptable range. The final output load change rate prediction result can be used to guide the actual communication link switching decision. Such a prediction result can effectively avoid the degradation of communication quality caused by insufficient resources.

[0096] Furthermore, before the step of updating the parameters of the adaptive detection model using the gradient descent method, step S1061 is also included: based on the load distribution characteristics, an adaptive detection model is constructed using a support vector machine algorithm, and the adaptive detection model is trained. The training data includes the load distribution characteristics before and after the switching, and the support vector machine uses a radial basis kernel function.

[0097] Specifically, obtain the load distribution characteristic data before and after the switch, extract the load amount, perform normalization processing on the load amount, and obtain standardized load data. Extract the histogram and statistical eigenvalues ​​from the standardized load data, and determine the training set samples. Use the support vector machine algorithm, set the radial basis kernel function parameters, and build a model training framework. According to the training set samples and the model training framework, execute the model training process to obtain a preliminary training model. If the accuracy of the preliminary training model is lower than the preset threshold, adjust the kernel function parameters and re-execute the training process. Obtain the optimized training model and determine whether the model performance indicators meet the requirements. If the requirements are met, the optimized training model is determined as the adaptive detection model.

[0098] For example, in the detection of communication load distribution characteristics, obtaining load distribution data before and after switching is a core task. When the protocol is switched, it is recorded that the traffic load changes from one gigabyte per second to eight hundred megabytes per second. This load needs to be normalized. Normalization can use the maximum and minimum method to map the load data to the interval from zero to one, so that load data of different sizes are comparable. When extracting features from the standardized load data, a load distribution histogram can be constructed. For example, the load interval is divided into ten levels, and the frequency of occurrence of data at each level is counted to obtain the load distribution law. At the same time, statistical characteristic values ​​such as mean and variance are calculated, and these data together constitute the training sample. The normalized load mean is 0.6 and the variance is 0.04, which can be used for subsequent modeling. When the support vector machine algorithm uses the radial basis kernel function, the parameter value of the kernel function is set. The optimal parameters are selected by the cross-validation method. For example, when the kernel function parameter value is 0.1, the model performs best. When building a model training framework, data preprocessing, feature selection, and model evaluation should be considered. For example, the first 100 sets of load data are selected as training sets, and the model performance is evaluated by 5-fold cross validation. During the model training process, if the accuracy is lower than the preset threshold of 95%, the kernel function parameters need to be adjusted and retrained. The parameters are gradually adjusted by the grid search method, and the model accuracy eventually reaches 97%. The optimized training model also needs to be evaluated for its generalization ability, and the test set data can be used to verify the model performance. The performance indicators of the adaptive detection model include accuracy, recall rate, and running time. The time for the model to process each piece of data should be controlled within one millisecond, and the accuracy should reach more than 95%. The adaptive detection model constructed in this way can quickly identify load anomalies and adjust the communication strategy in time.

[0099] Step S107, deploying the updated adaptive detection model to the Internet of Vehicles, detecting the communication load in real time, and continuously monitoring the model performance after deployment for optimization.

[0100] Specifically, the TensorRT model conversion tool is used to convert the updated adaptive detection model into the ONNX format. During the conversion process, the quantization function of TensorRT is used to retain the adaptive characteristics of the adaptive detection model. According to the hardware architecture and computing resources of the Internet of Vehicles system, CUDA is used to configure the inference parameters of the converted model, and the frequency of real-time communication load detection is set to 10 times per second with an accuracy requirement of 99%. The converted model is deployed to the Internet of Vehicles system, and the data interface between the converted model and the on-board communication module is configured through the CAN bus to realize real-time collection and preprocessing of communication data. The preprocessed data is obtained from the communication module, and the inference calculation of the model is performed through the inference calculation unit to obtain the detection results of the communication load. The detection results are transmitted to the display module for real-time presentation.

[0101] Use Prometheus to establish a model performance monitoring mechanism, collect performance indicators such as accuracy, latency, and resource utilization of the converted model reasoning, and store performance data for subsequent analysis. If the performance monitoring data is lower than the preset threshold, the model optimization process is started, and the stochastic gradient descent algorithm is used to incrementally update the adaptive detection model parameters to improve the adaptability of the adaptive detection model in the Internet of Vehicles environment. During the model optimization process, monitor the optimized performance indicators. If the performance meets the preset requirements, the optimization process is completed.

[0102] For example, the conversion of the communication load detection model into the vehicle deployment format is a key link to ensure the real-time performance of the system. In practical applications, taking self-driving cars as an example, the vehicle needs to process massive data from multiple sensors every second. For example, the lidar generates about one million data points per second, and the camera generates tens of megabytes of image data per second. When these data are transmitted through the vehicle network, they will generate huge communication load. The use of quantization technology during model conversion can effectively reduce the computing overhead. For example, converting the floating-point parameters of the model to eight-bit fixed-point numbers can compress the model size to one-fourth of the original size, while still maintaining more than 99% of the detection accuracy. The inference delay of the model on the vehicle chip after quantization optimization is reduced from 100 milliseconds to 10 milliseconds. When the communication rate can reach one megabit per second, millisecond-level data sampling can be achieved by configuring the standard data frame format. The performance monitoring mechanism is crucial to the continuous optimization of the model. By storing performance indicators in a time series database, the performance of the model under different working conditions can be tracked. For example, in actual measurements, it was found that the detection accuracy of the model in a high temperature environment would drop by 5%. The model was optimized in a targeted manner through an incremental update algorithm to improve the environmental adaptability of the model. Monitoring indicators include the amount of data processed per second, the accuracy of detection results, and the system resource occupancy rate. The stochastic gradient descent algorithm plays a key role in model optimization. Taking communication load prediction as an example, the algorithm dynamically adjusts model parameters based on real-time collected data. When a vehicle is detected in a congested section, the prediction model will pay more attention to short-term load changes, and the weight parameters will be adjusted in the direction of rapid response. When driving at high speeds, the model will pay more attention to the grasp of long-term load trends. This adaptive optimization mechanism enables the model to adapt to different driving scenarios. Real-time display during deployment can provide more intuitive load status feedback. The current network load level, future load forecast trend and other information are displayed through a graphical interface. For example, different colors are used to mark the load level on the on-board display screen, green indicates normal load, yellow indicates high load, and red indicates that the load is close to the threshold, helping the driver to understand the vehicle communication status in a timely manner.

[0103] This embodiment also provides a communication load detection system for a vehicle networking environment, including:

[0104] A status acquisition module is used to acquire the communication link status information of the node including the current connection protocol type and signal strength;

[0105] A data extraction module, used to extract corresponding communication load data including data packet size and transmission frequency according to the protocol type;

[0106] The change rate calculation module is used to calculate the load change rate in each time window using a sliding window mechanism for communication load data. The load change rate is defined as a weighted change value of the data packet size and the transmission frequency in the window.

[0107] A switching detection module is used to trigger a protocol switching detection mode when the load change rate exceeds a preset threshold, and select the optimal communication protocol according to the network environment through a pre-established protocol switching model;

[0108] A link establishment module, used to establish a new communication link through an optimal communication protocol and record load distribution characteristics including the mean value of data packet size and transmission frequency variance before and after switching;

[0109] The parameter updating module is used to update the parameters of the adaptive detection model using the gradient descent method according to the load distribution characteristics of the IoV nodes, and the objective function is to minimize the load change rate prediction error;

[0110] The model deployment and monitoring module is used to deploy the updated adaptive detection model to the Internet of Vehicles, detect the communication load in real time, and continuously monitor the model performance after deployment for optimization.

[0111] It also includes a detection model building module, which is used to build an adaptive detection model based on load distribution characteristics by using a support vector machine algorithm, and train the adaptive detection model. The training data includes load distribution characteristics before and after switching, and the support vector machine uses a radial basis kernel function.

[0112] The input features of the pre-established protocol switching model include signal strength, packet size, and transmission frequency, and the feature selection criterion is to maximize information gain.

[0113] The above description is only a preferred embodiment of the present invention. Therefore, any equivalent changes or modifications made according to the structure, characteristics and principles described in the scope of the patent application of the present invention are included in the protection scope of the patent application of the present invention.

Claims

1. A communication load detection method for a vehicle networking environment, characterized in that: The steps include: Obtain the node's communication link status information including the current connection protocol type and signal strength; According to the protocol type, extract the corresponding communication load data including the data packet size and transmission frequency; For the communication load data, a sliding window mechanism is used to calculate the load change rate within each time window. The load change rate is defined as the weighted change value of the packet size and transmission frequency within the window. If the load change rate exceeds the preset threshold, the protocol switching detection mode is triggered, and the optimal communication protocol is selected according to the network environment through the pre-established protocol switching model; Establish a new communication link through the optimal communication protocol, and record the load distribution characteristics before and after the switch, including the mean value of the packet size and the variance of the transmission frequency; According to the load distribution characteristics, the gradient descent method is used to update the parameters of the adaptive detection model, and the objective function is to minimize the load change rate prediction error; Deploy the updated adaptive detection model to the Internet of Vehicles to detect the communication load in real time, and continuously monitor the performance of the adaptive detection model after deployment for optimization.

2. The communication load detection method according to claim 1, characterized in that: The communication load data is calculated by using a sliding window mechanism to calculate the load change rate in each time window. The load change rate is defined as a weighted change value of the data packet size and the transmission frequency in the window, including: For communication load data, a sliding window mechanism is used to divide the time window; The network traffic is monitored using a preset time window to obtain the data packet size value and the transmission frequency value, and the size value and the frequency value are weighted according to a pre-established weight model to obtain a weighted value; The weighted value is input into the load change rate calculation model, and the load change rate in the current time window is calculated by a linear regression algorithm.

3. The communication load detection method according to claim 1, characterized in that: Before the step of updating the adaptive detection model parameters by using the gradient descent method, it also includes: Based on the load distribution characteristics, the support vector machine algorithm is used to build an adaptive detection model, and the adaptive detection model is trained. The training data includes the load distribution characteristics before and after switching. The support vector machine uses the radial basis kernel function.

4. The communication load detection method according to claim 1, characterized in that: If the load change rate exceeds the preset threshold, the protocol switching detection mode is triggered, and the optimal communication protocol is selected according to the network environment through the pre-established protocol switching model, including: The input features of the pre-established protocol switching model include signal strength, data packet size and transmission frequency, and the feature selection criterion is to maximize information gain; Obtain the load change rate in the network environment of the Internet of Vehicles. If the load change rate exceeds a preset threshold, the protocol is triggered to switch to a detection mode. In the protocol switching detection mode, the signal strength, data packet size and transmission frequency in the network environment are collected, and these feature data are processed using the minimum and maximum normalization method; The normalized data is input into a pre-established protocol switching model to determine the information gain value of each feature for protocol selection; Based on the information gain maximization standard, the communication protocol corresponding to the feature with the largest information gain is selected as the optimal protocol, and a protocol switching decision is generated; Convert the protocol switching decision into a network configuration instruction, use the network configuration protocol to send the instruction to the network device, and the device completes the dynamic switching of the communication protocol according to the instruction; After the switch is completed, the load change rate in the network environment is re-obtained to determine whether it meets the preset threshold; If not satisfied, repeat the above process until the load change rate stabilizes within the preset threshold.

5. The communication load detection method according to claim 1, characterized in that: The establishing of a new communication link through the optimal communication protocol and recording of load distribution characteristics including the mean value of the data packet size and the transmission frequency variance before and after the switching include: Establish a new communication link for the Internet of Vehicles under the optimal communication protocol, and obtain the mean value of the packet size and the variance of the transmission frequency before switching; If the new communication link is established successfully, the mean value of the packet size and the variance of the transmission frequency after the switch are collected; According to the mean values ​​of the data packet sizes before and after the switching, the mean value change is calculated to determine the first characteristic value of the load distribution characteristic; According to the transmission frequency variance before and after the switching, the variance change is calculated to determine the second characteristic value of the load distribution characteristic; Inputting the first eigenvalue and the second eigenvalue into a pre-established load distribution characteristic model to determine whether the load distribution characteristic is stable; If the load distribution characteristics are stable, a load distribution characteristic record is generated and stored in the network environment database; According to the load distribution feature records, the pre-established protocol switching model is updated and the information gain calculation standard is optimized.

6. The communication load detection method according to claim 3, characterized in that: Based on the load distribution characteristics, the support vector machine algorithm is used to build an adaptive detection model, and the adaptive detection model is trained. The training data includes the load distribution characteristics before and after the switching. The support vector machine uses a radial basis kernel function, including: Obtain load distribution characteristic data before and after the switching, extract the load amount, perform normalization processing on the load amount, and obtain standardized load data; Extract histograms and statistical characteristic values ​​from the standardized load data to determine training set samples; Use the support vector machine algorithm, set the radial basis kernel function parameters, and build a model training framework; According to the training set samples and the model training framework, the model training process is executed to obtain a preliminary training model; If the accuracy of the initial training model is lower than the preset threshold, the base kernel function parameters are adjusted and the training process is re-executed; Obtain the optimized training model and determine whether the model performance indicators meet the requirements; If the requirements are met, the optimized training model is determined as the adaptive detection model.

7. The communication load detection method according to claim 1, characterized in that: The method of updating the adaptive detection model parameters by using the gradient descent method according to the load distribution characteristics, wherein the objective function is to minimize the load change rate prediction error, includes: Obtain load distribution characteristic data, and extract load change rate based on the characteristic data; Adopting an adaptive detection model to predict the load change rate and obtain an initial prediction value; Calculate the prediction error based on the initial prediction value and the actual load change rate; If the prediction error exceeds the preset threshold, the gradient descent method is used to update the adaptive detection model parameters; The load change rate is re-predicted through the updated adaptive detection model parameters to obtain the optimized prediction value; According to the optimized prediction value, determine whether the error meets the minimization condition. If not, update the adaptive detection model parameters cyclically. The optimized adaptive detection model is used to output the final load change rate prediction result.

8. The communication load detection method according to claim 1, characterized in that: The updated adaptive detection model is deployed to the Internet of Vehicles to detect the communication load in real time, including: Use the TensorRT model conversion tool to convert the updated adaptive detection model to ONNX format; During the conversion process, the adaptive nature of the adaptive detection model is preserved using TensorRT’s quantization capabilities. Use CUDA to configure the inference parameters of the converted model according to the hardware architecture and computing resources of the connected car system; Deploy the converted model to the Internet of Vehicles system, configure the data interface between the converted model and the vehicle communication module, and realize the real-time collection and preprocessing of communication data; Obtain preprocessed data from the vehicle communication module, perform inference calculation of the model through the vehicle computing unit, and obtain the detection result of the communication load; The test results are transmitted to the vehicle display module for real-time presentation.

9. The communication load detection method according to claim 1, characterized in that: The model performance is continuously monitored after deployment for optimization, including: Use Prometheus to establish a model performance monitoring mechanism, collect performance indicators such as accuracy, latency, and resource usage of the converted model reasoning, and store performance data for subsequent analysis; If the performance monitoring data is lower than the preset threshold, the model optimization process is started, and the parameters of the adaptive detection model are incrementally updated using the stochastic gradient descent algorithm to improve the adaptability of the adaptive detection model in the vehicle environment; During the model optimization process, the optimized performance indicators are monitored. If the performance meets the preset requirements, the optimization process is completed.

10. A communication load detection system for a vehicle networking environment, characterized in that: include: A status acquisition module is used to acquire the communication link status information of the node including the current connection protocol type and signal strength; A data extraction module, used to extract corresponding communication load data including data packet size and transmission frequency according to the protocol type; The change rate calculation module is used to calculate the load change rate in each time window using a sliding window mechanism for communication load data. The load change rate is defined as a weighted change value of the data packet size and the transmission frequency in the window. A switching detection module is used to trigger a protocol switching detection mode when the load change rate exceeds a preset threshold, and select the optimal communication protocol according to the network environment through a pre-established protocol switching model; A link establishment module, used to establish a new communication link through an optimal communication protocol and record load distribution characteristics including the mean value of data packet size and transmission frequency variance before and after switching; The parameter updating module is used to update the parameters of the adaptive detection model using the gradient descent method according to the load distribution characteristics of the IoV nodes, and the objective function is to minimize the load change rate prediction error; The model deployment and monitoring module is used to deploy the updated adaptive detection model to the Internet of Vehicles, detect the communication load in real time, and continuously monitor the model performance after deployment for optimization.

Images