An intelligent traffic data fusion analysis method and device based on cloud-edge collaboration and a medium

Abstract

The application provides a kind of intelligent traffic data fusion analysis method, equipment and medium based on cloud edge cooperation, belongs to intelligent traffic data fusion analysis technical field, method is collected in traffic edge side multi-source traffic data;Based on pre-processing data, local analysis is carried out to generate local traffic analysis results;Pre-processing data and local analysis results are uploaded to traffic cloud platform;Multi-level data fusion is carried out in cloud to obtain global traffic situation data;Global data is used to train and update traffic prediction and decision model;According to data processing complexity and real-time requirement, tasks are dynamically allocated to edge or cloud;Cloud issues control instructions, edge executes and feedback effect data, and cloud optimizes decision model accordingly.The application improves data processing efficiency and real-time performance, enhances data fusion accuracy, optimizes resource allocation and improves system response speed.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation data fusion and analysis technology, specifically relating to a cloud-edge collaborative intelligent transportation data fusion and analysis method, equipment, and medium. Background Technology

[0002] With the acceleration of urbanization and the continuous growth of motor vehicle ownership, problems such as traffic congestion, frequent accidents, and low traffic efficiency are becoming increasingly prominent. Against this backdrop, Intelligent Transportation Systems (ITS), as a comprehensive solution integrating sensing, communication, computing, and control technologies, have become a key means to alleviate traffic pressure and improve road network operational efficiency.

[0003] Traffic conditions are typically predicted and decisions are made using predictive models. These models are often trained based on historical data. However, since traffic flow is dynamic, training solely on historical data lacks a match with the actual situation, leading to large prediction errors. For example, the predicted traffic flow during the morning rush hour may differ significantly from the actual flow, which can result in traffic efficiency not improving after signal adjustments in decision-making schemes.

[0004] The raw traffic data that the prediction model aims to obtain is subject to noise and redundant information. Direct transmission and processing of this data would consume a large amount of bandwidth and computing resources, leading to increased analysis latency and making it difficult to meet the needs of real-time traffic management. Summary of the Invention

[0005] This invention provides a cloud-edge collaborative intelligent traffic data fusion and analysis method that improves traffic management efficiency, reduces congestion, and enhances safety.

[0006] The methods include: S101: At the traffic edge, collect multi-source traffic data from traffic cameras, radar sensors, and traffic signal controllers; S102: Clean and compress multi-source traffic data to obtain preprocessed traffic data; S103: Perform local traffic analysis based on preprocessed traffic data to generate local traffic analysis results; S104: Upload preprocessed traffic data and local traffic analysis results to the traffic cloud platform; S105: On the traffic cloud platform, the uploaded preprocessed traffic data and local traffic analysis results are fused to obtain global traffic situation data; S106: Train the traffic prediction and decision-making model based on global traffic situation data to obtain the updated traffic model; S107: Based on the complexity and real-time requirements of traffic data processing, allocate traffic processing tasks to the traffic edge side; S108: The traffic cloud platform sends traffic control commands to the traffic edge side, which executes the traffic control commands and provides feedback on the execution effect data.

[0007] Preferably, S103 specifically includes the following methods: Traffic flow parameters are extracted from preprocessed traffic data, and traffic flow, average speed, and density are calculated. Based on the extracted traffic flow parameters, a classification model is used to classify traffic conditions and identify traffic congestion and free flow. Perform moving target detection and tracking on video data in preprocessed traffic data to identify vehicle and pedestrian trajectories; Traffic incident detection is performed based on the results of moving target detection and tracking. By integrating traffic condition classification results and traffic incident detection results, local traffic analysis results are generated through data fusion methods.

[0008] Preferably, S105 specifically includes the following methods: The preprocessed traffic data and local traffic analysis results are aligned based on a unified timestamp and geographic location. For the aligned duplicate data, the Kalman filter algorithm is used to fuse them and output consistent traffic flow and average vehicle speed indicators. Extract the target trajectory features output by YOLOv5s and the traffic flow and occupancy rate from the radar data, calculate the feature weights, and weightedly fuse them to generate cross-modal traffic features; By combining the local traffic analysis results generated by S103 with the cloud-based historical traffic situation model, a preliminary global situation including anomaly impact assessment is obtained. Consistent indicators, cross-modal features, and preliminary global situation data are summarized and integrated into global traffic situation data.

[0009] Preferably, S106 specifically includes the following methods: S1061: Standardize and serialize the global traffic situation data to construct a model training dataset; S1062: Construct a spatiotemporal graph neural network model based on attention mechanism as a traffic prediction and decision-making model; S1063: Use a distributed computing framework to perform multiple rounds of iterative training on the model and optimize the model parameters; S1064: Use the validation dataset to evaluate the performance of the trained model and calculate the prediction accuracy metric; S1065: When the model performance meets the requirements, generate updated model parameters and prepare for distribution.

[0010] Preferably, S1062 specifically includes the following methods: Construct a traffic network graph structure, setting road segments or intersections as graph nodes. Node attributes include road segment length, number of lanes, and historical average traffic flow. The connection between adjacent nodes is set as graph edges, and the edge weight is calculated based on the actual distance between nodes and the historical traffic flow transfer rate. A bidirectional LSTM network layer is used, with the input being the temporal traffic data of the nodes. The short-term and long-term temporal dependencies of the traffic data are captured through forward and backward hidden states. A graph convolutional layer is used to perform neighborhood aggregation on node features based on the constructed graph structure. The feature information of neighboring nodes is fused into the current node through a weight matrix to capture the spatial correlation of the road network. Spatial attention layer and temporal attention layer are set up respectively. Spatial attention calculates the influence weight of different neighboring nodes on the current node, and temporal attention calculates the contribution weight of features at different time steps to the current prediction. After normalizing the weights through the Softmax function, the features are weighted and fused. The fused features are input into a fully connected layer, and the predicted traffic conditions and corresponding traffic control decision parameters for the future preset time period are output through linear transformation and activation function, respectively.

[0011] Preferably, S107 specifically includes the following methods: Configure the real-time performance and computational complexity metrics of the traffic tasks to be processed, and generate task feature vectors; Tasks are categorized into edge processing or cloud processing categories based on their feature vectors. Based on the task category identifier, the task data is routed to the corresponding traffic edge side; During task execution, the resource utilization status and task queue status at the traffic edge are continuously monitored.

[0012] Preferably, S108 specifically includes the following methods: The traffic cloud platform will transmit control commands to the designated traffic edge side; The traffic edge receives and parses control commands, converts them into operational commands that can be executed by the equipment, and drives the traffic signal controller or information dissemination equipment to execute them. Traffic status change data after the execution of instructions is collected at the traffic edge, and compared with the status before execution to generate execution effect evaluation indicators; The traffic edge will package the performance evaluation indicators with the original control commands and send them back to the cloud platform. The transportation cloud platform stores the execution effect data and corresponding decision records in the experience playback buffer.

[0013] According to another embodiment of this application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the cloud-edge collaborative intelligent transportation data fusion and analysis method.

[0014] According to another embodiment of this application, a storage medium is also provided, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the steps of the cloud-edge collaborative intelligent transportation data fusion and analysis method.

[0015] As can be seen from the above technical solutions, the present invention has the following advantages: The cloud-edge collaborative intelligent traffic data fusion and analysis method provided by this invention cleans and compresses multi-source traffic data, removing noise and filling in missing values ​​from sensor data. Targeted processing of video and text data improves data accuracy and completeness. Cloud-based data fusion integrates data from different sources and of different types, eliminating inconsistencies. Edge-based local traffic analysis quickly generates local traffic analysis results, enabling real-time monitoring and preliminary judgment of traffic conditions, such as local anomaly detection and simple trend prediction, providing a basis for timely countermeasures. Training traffic prediction and decision-making models based on global traffic situation data enables more accurate traffic predictions and more scientific decision-making. Tasks are allocated to the edge or cloud according to the complexity and real-time requirements of traffic data processing, allowing different types of tasks to be processed in appropriate environments, improving task processing efficiency.

[0016] Simple tasks with high real-time requirements are processed at the edge, ensuring rapid response; complex and non-real-time tasks are processed in the cloud, enabling deeper analysis and computation, thus balancing real-time response and complex processing. Through bidirectional command interaction, traffic prediction and decision-making models are optimized based on command execution effect data, allowing the models to continuously adapt to changes in traffic conditions and maintain good performance and accuracy. In this invention, the edge side is responsible for data acquisition, local analysis, and simple task processing, while the cloud side is responsible for data fusion, model training, and complex task processing. The two sides achieve information sharing and collaborative work through bidirectional command interaction, improving the performance of the entire traffic management system. Attached Figure Description

[0017] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart of a cloud-edge collaborative intelligent transportation data fusion and analysis method; Figure 2 This is a schematic diagram of an electronic device. Detailed Implementation

[0019] The following describes in detail the cloud-edge collaborative intelligent transportation data fusion and analysis method involved in this application. Specific details, such as particular system structures and technologies, are presented for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application can also be implemented in other embodiments without these specific details.

[0020] It should be understood that, when used in this specification, the term "comprising" indicates the presence of the described feature, integral, step, operation, element, and / or component, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0021] The terms "one embodiment" or "some embodiments" used in this application mean that one or more embodiments of this application include the specific features, structures, or characteristics described in that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this application do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized.

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figure 1 The diagram shows a flowchart of a cloud-edge collaborative intelligent transportation data fusion and analysis method in a specific embodiment. The method includes: S101: At the traffic edge, collect multi-source traffic data from traffic cameras, radar sensors, and traffic signal controllers.

[0024] In some embodiments, roadside edge devices are deployed in traffic scenarios such as road intersections, bridges, and tunnels. The roadside edge devices access video streams from traffic cameras, point cloud data from radar sensors, and operational data from traffic signal controllers to complete the collection of traffic data.

[0025] S102: Clean and compress the multi-source traffic data to obtain preprocessed traffic data.

[0026] In some embodiments, data quality is improved and data volume is reduced by cleaning to eliminate noise and missing data, compressing to remove redundant information.

[0027] S103: Perform local traffic analysis based on preprocessed traffic data to generate local traffic analysis results.

[0028] S103 specifically includes the following methods: S1031: Extract traffic flow parameters from preprocessed traffic data and calculate traffic flow, average speed, and density.

[0029] In some embodiments, traffic flow LB is calculated as LB=CN / T, where CN is the number of vehicles passing through in time T; average speed V is calculated as V=Σvi / CN, where vi is the speed of the i-th vehicle; and density ρ is calculated as ρ=LB / V.

[0030] This embodiment extracts vehicle passage events and speed values ​​from preprocessed sensor data, accumulates the number of vehicles within a preset time interval, and calculates the average speed and density.

[0031] S1032: Based on the extracted traffic flow parameters, a classification model is used to classify traffic conditions and identify traffic congestion and free flow.

[0032] The classification model in this embodiment is f(x)=sign(Σαi×yi×K(xi,x)+b), where x is the input feature vector based on flow, speed and density, yi is the class label, K is the kernel function, αi and b are model parameters, and xi represents the feature vector of the i-th support vector. In the support vector machine model, xi comes from the sample features in the training set that are determined to be support vectors and is used to construct the classification decision hyperplane.

[0033] `sign()` is the sign function, which takes the sign of the input value. Specifically, it outputs +1 when the input value is greater than 0, 0 when the input value is equal to 0, and -1 when the input value is less than 0. In classification models, `sign()` converts the calculation result of the Support Vector Machine (SVM) decision function into the final classification label, such as +1 representing congestion and -1 representing smooth flow.

[0034] In this embodiment, traffic flow parameters are used as features input to a pre-trained classification model. Features are segmented using a kernel function to map them, and the output state category is determined. The model parameters are trained based on historical data.

[0035] S1033: Perform moving target detection and tracking on video data in preprocessed traffic data to identify vehicle and pedestrian trajectories.

[0036] This embodiment extracts target data from video frames, distinguishes between background and moving objects using a Gaussian mixture model, and predicts and corrects target positions using a Kalman filter algorithm to form a continuous trajectory. In this way, by analyzing continuous frames to detect moving targets and maintain their trajectories, dynamic monitoring of traffic participants is achieved.

[0037] S1034: Based on the results of moving target detection and tracking, perform traffic incident detection and identify accidents and illegal parking incidents.

[0038] This embodiment uses a logistic regression model to extract features from trajectory data, such as sudden speed changes and positional stagnation.

[0039] S1035: Integrate traffic condition classification results and traffic incident detection results, and generate local traffic analysis results through data fusion methods.

[0040] This embodiment normalizes the outputs of state classification and event detection, and then fuses them according to predefined weights to generate a comprehensive analysis result. The weights are dynamically adjusted based on data quality. The fusion result includes text descriptions and numerical indicators.

[0041] Optionally, a weighted average fusion method can be used, with the final result JTR = w1 × S + w2 × E, where S is the state classification score, E is the event detection score, and w1 and w2 are the weights. JTR is an abbreviation for Local Traffic Analysis Result, representing the comprehensive output value obtained after weighted fusion. The comprehensive output value is a dimensionless scalar, ranging from 0 to 1. A larger value indicates a more complex traffic situation or a higher degree of urgency requiring intervention.

[0042] It should be noted that S is used to quantify the traffic state classification results in S1032, with a value ranging from 0 to 5. The specific execution method, combining the classification model output with traffic flow parameters, is as follows: Step 1: Obtain the output of the classification model, i.e., the output value of sign(); Step 2: Extract traffic flow parameters for the current area to be analyzed, calculate the normalized values ​​of the parameters, and eliminate the influence of dimensions. The normalized traffic flow value is processed as follows: LBnorm=(LB-LBmin) / (LBmax-LBmin), where LBmin is the historical minimum traffic flow in the area and LBmax is the historical maximum traffic flow in the area. The normalized velocity value is processed as follows: Vnorm=(V-Vmin) / (Vmax-Vmin), where Vmin is the historical minimum average velocity of the region and Vmax is the historical maximum average velocity of the region. The density normalization value is processed as follows: ρnorm=(ρmax-ρ) / (ρmax-ρmin), where ρmin is the historical minimum traffic density of the area and ρmax is the historical maximum traffic density of the area. The higher the density, the smaller the normalization value, which is negatively correlated with the degree of congestion. Step 3: Calculate the state classification score S, considering two cases: If the classification model outputs 1, it indicates a smooth flow condition: S = 5 × (0.3 × LBnorm + 0.5 × Vnorm + 0.2 × ρnorm). Speed ​​has the greatest impact on smooth flow, followed by flow rate and density. If the classification model outputs -1, it indicates a congestion state: S=5×(0.3×(1-LBnorm)+0.5×(1-Vnorm)+0.2×(1-ρnorm)), which inversely quantifies the degree of congestion. The more severe the congestion, the lower the S score.

[0043] It should be further explained that the event detection score E is a value between 0 and 1, used to quantify the probability of traffic events such as accidents and illegal parking. The calculation is based on a pre-trained logistic regression model. The specific steps for training and processing the logistic regression model are as follows: Construct a feature vector z, using event-related features extracted from the moving target tracking results as input.

[0044] Accident detection features may include: Define the instantaneous rate of decrease of vehicle speed z1, the angle of sudden change in trajectory direction z2, and the distance z3 at which multiple target trajectories intersect or are abnormally close.

[0045] For illegal parking detection, the features include: defining the time a vehicle remains in a non-parking area (z4); and defining the relative relationship between the vehicle's stopping position and the lane lines (z5).

[0046] Furthermore, the selected features are combined into a feature vector z, such as z=[z1,z2,z3], for accident detection.

[0047] Input the feature vector z into the trained logistic regression model. Logistic regression directly outputs a probability value: E = P(event|z) = 1 / (1 + exp(-(W T *z+c))) Where W is the model weight vector and c is the bias term, both learned from historical event data during the model training phase. T*z represents the dot product of the weight vector and the feature vector.

[0048] The closer E is to 1, the higher the probability of the corresponding event occurring. For example, a threshold of 0.7 can be set, and an event alarm will be triggered if E > 0.7.

[0049] Furthermore, the traffic analysis results can be integrated with JTR, adjusting the weights w1 and w2. The initial values ​​of w1 and w2 can be set as empirical values, such as w1=0.7, w2=0.3, defined as normal traffic conditions.

[0050] When the event detection score E exceeds a high threshold, such as 0.8, it indicates that a high-confidence event has occurred. The system automatically increases the weight of w2, making the JTR result more reflective of the sudden event. After the event ends, the weight returns to its initial value.

[0051] Furthermore, JTR = w1*S + w2*E, where relevant information is generated based on the JTR value, S, and E. For example, if the JTR is high and mainly due to high S, then the current area is experiencing traffic congestion and slow traffic.

[0052] If the JTR is high and mainly originates from E high: A suspected traffic accident has been detected in the current area. Please take precautions to avoid it.

[0053] Numerical indicators: JTR itself is a comprehensive index, which, together with S, E, w1, and w2, is uploaded to the transportation cloud platform as structured data.

[0054] This embodiment operates independently at the edge, enabling real-time analysis, reducing data transmission latency, and ensuring the accuracy and reliability of the analysis.

[0055] S104: Upload preprocessed traffic data and local traffic analysis results to the traffic cloud platform.

[0056] In some embodiments, the edge device uploads preprocessed traffic data and local traffic analysis results to the traffic cloud platform via the MQTT protocol, enabling data interoperability between the edge and the cloud and providing basic data for global analysis in the cloud.

[0057] S105: On the traffic cloud platform, the uploaded preprocessed traffic data and local traffic analysis results are fused to obtain global traffic situation data.

[0058] S105 specifically includes the following methods: S1051: Align the preprocessed traffic data and local traffic analysis results based on a unified timestamp and geographic location to eliminate time delays and spatial biases from different data sources.

[0059] S1052: For the aligned duplicate data, the Kalman filter algorithm is used to fuse them and output consistent traffic flow and average vehicle speed indicators.

[0060] S1053: Extract the target trajectory features from the YOLOv5s output and the traffic flow and occupancy rate from the radar data, calculate the feature weights, and perform weighted fusion to generate cross-modal traffic features.

[0061] This embodiment converts the target trajectory features of the video, such as vehicle driving direction, speed, and continuous trajectory, into multi-dimensional vectors. Traffic flow, occupancy rate, and queue length in a certain direction from the radar are also converted into multi-dimensional vectors. The cosine similarity between the two is calculated; the higher the similarity, the greater the weight of the corresponding feature. Weighted fusion generates cross-modal features that reflect both the behavior of traffic participants in the video and the traffic density information in the radar.

[0062] S1054: Combine the local traffic analysis results generated in S103 with the output of the cloud-based historical traffic situation model, and use Bayesian integration to generate a preliminary global situation that includes anomaly impact assessment.

[0063] This embodiment uses the local analysis results at the edge, such as a pedestrian running a red light at intersection A, as the prior probability. The historical situation model in the cloud, which shows that such anomalies at intersection A during the morning rush hour over the past week increased the probability of congestion in downstream road sections by 40%, is used as the likelihood. The posterior probability of the global situation is then calculated, corresponding to an anomaly at intersection A leading to a 55% increase in the probability of congestion in downstream road sections. By combining real-time local analysis at the edge with historical experience in the cloud, the impact of abnormal events on the global situation is quantified.

[0064] S1055: The consistent indicators of S1052, the cross-modal characteristics of S1053, and the preliminary global situation of S1054 are summarized and integrated into global traffic situation data, which specifically includes the congestion index of each road segment, the distribution of accident-prone areas, and the overall traffic efficiency.

[0065] This embodiment first calculates that when the traffic flow of each road segment reaches 80% of its capacity, the congestion index is 0.8. Then, the congestion indices of five adjacent road segments are aggregated into a regional situation. If the average congestion index in the region is 0.7, then the region is congested overall. Finally, the situation of all regions is integrated into the overall network congestion index and the distribution of accident-prone areas, thus realizing a general presentation of the global traffic status.

[0066] S106: Train the traffic prediction and decision-making model based on global traffic situation data to obtain the updated traffic model.

[0067] In some embodiments, a distributed framework is used to process large-scale data, train a deep model that can capture the spatiotemporal patterns of traffic, and generate prediction results.

[0068] S107: Based on the complexity and real-time requirements of traffic data processing, allocate traffic processing tasks to the traffic edge side.

[0069] In some embodiments, tasks are dynamically allocated based on task characteristics. Tasks with high real-time requirements and simple processing logic, such as real-time alarms for vehicles running red lights and immediate adjustments to individual traffic signals, are allocated to the edge side. Tasks with complex processing logic and low real-time requirements, such as regional traffic management plan planning and monthly traffic flow statistics analysis, are allocated to the cloud to achieve rational utilization of resources.

[0070] S108: The traffic cloud platform sends traffic control commands to the traffic edge side, which executes the traffic control commands and provides feedback on the execution effect data.

[0071] S108 specifically includes the following methods: S1081: The traffic cloud platform transmits control commands to the designated traffic edge side.

[0072] S1082: The traffic edge side receives and parses control commands, converts them into operational commands that can be executed by the device, and drives the traffic signal controller or information dissemination device to execute them.

[0073] S1083: Collect traffic status change data after the execution of the instruction at the traffic edge and compare it with the status before the execution to generate an evaluation index of the execution effect.

[0074] S1084: The traffic edge side will package the performance evaluation indicators with the original control commands and send them back to the cloud platform.

[0075] S1085: The traffic cloud platform stores the execution effect data and corresponding decision records in the experience playback buffer for the optimization and training of the decision model.

[0076] In S1081 of this embodiment, the control instruction is a quintuple C={t,n,a,p,v}, where t is a timestamp, n is the target node identifier, a is the operation type, p is the parameter type, and v is the set of parameter values. S1082 converts the signal timing scheme parameters into phase timing data of the traffic signal controller, or converts path guidance information into variable message sign display content. S1083 collects parameters such as traffic flow, average speed, and queue length at the same location within a fixed time interval before and after instruction execution. The changes in key performance indicators are calculated, normalized, and combined into a comprehensive effect evaluation index. The actual effect of the control instruction is quantitatively evaluated by comparing the states before and after. S1084 establishes a correlation between the effect evaluation index and the corresponding control instruction, adds timestamps and location information, and forms a complete feedback data packet. S1085 After receiving the feedback data, the cloud platform matches it with the original decision records, combines them into complete experience samples, and stores them in a fixed-capacity circular buffer. When the buffer accumulates enough samples, batch samples are extracted in chronological order for model parameter updates. This achieves continuous learning and optimization based on actual effects, improving the system's adaptive capability.

[0077] In one embodiment of the present invention, based on step S106, the following is a possible embodiment and its specific implementation will be described in a non-limiting manner. S106 specifically includes the following: S1061: Standardize and serialize the global traffic situation data to build a model training dataset.

[0078] S1062: Construct a spatiotemporal graph neural network model as a traffic prediction and decision-making model.

[0079] The spatiotemporal graph neural network model in this embodiment includes a graph structure construction module, which sets road segments as nodes. Node features include road segment length, number of lanes, traffic flow, and speed. Connections between adjacent road segments are set as edges, and edge weights are calculated based on historical traffic flow transfer rates. It also includes a spatial feature extraction module, employing a two-layer graph convolutional network. Each layer uses an adjacency matrix to weightedly fuse node features with the features of adjacent nodes, capturing the impact of main roads on branch roads. Furthermore, it includes a temporal feature extraction module, which connects to a bidirectional LSTM network after the GCN output, capturing short-term fluctuations and peak-hour patterns in traffic data through forward and backward propagation. Future traffic flow and speed are output through a fully connected layer, along with traffic control decision parameters, namely signal timing adjustments.

[0080] S1063: Use a distributed computing framework to perform multiple rounds of iterative training on the model to optimize the model parameters.

[0081] This embodiment utilizes a distributed computing framework to distribute large-scale global traffic situation data and model training tasks across multiple traffic edge sides, achieving parallel processing. Each traffic edge side independently calculates the model's forward output and local gradients based on its local fragmented data. The traffic cloud platform updates the global parameters and then synchronizes the updated parameters to all traffic edge sides. Through multiple iterations, the model parameters are continuously adjusted to minimize the joint loss, enabling the model to gradually learn the spatiotemporal patterns of traffic flow and the mapping relationship between decision parameters and traffic states, ultimately achieving parameter optimization.

[0082] S1064: Use the validation dataset to evaluate the performance of the trained model and calculate the prediction accuracy metric.

[0083] In this embodiment, samples are randomly selected from the validation set and input into the trained model to obtain predicted values ​​for traffic flow and speed, as well as decision parameters. Prediction accuracy metrics are calculated, including mean absolute error, root mean square error, and the percentage of samples showing improved traffic efficiency after applying the decision parameters. An evaluation threshold is set; if all metrics meet the threshold, the model is deemed to be performing satisfactorily; otherwise, it is marked as unsatisfactory.

[0084] S1065: When the model performance meets the requirements, generate updated model parameters and prepare for distribution.

[0085] In this embodiment, when the model performance meets the evaluation threshold, the model parameters are extracted and version-marked. Parameter distribution data is generated, including the adapted edge device type and update trigger conditions. This embodiment, under the premise that the model performance meets the standards, ensures that the parameters can be correctly received and applied by the edge side through version control and data configuration, preparing for the secure distribution and deployment of model parameters and improving the efficiency and reliability of model updates.

[0086] In one embodiment of the present invention, based on step S1062, the following is a possible embodiment and its specific implementation will be described in a non-limiting manner. S1062 specifically includes the following methods: S10621: Construct a traffic network graph structure, setting road segments or intersections as graph nodes. Node attributes include road segment length, number of lanes, and historical average traffic flow. The connection between adjacent nodes is set as graph edges, and the edge weight is calculated based on the actual distance between nodes and the historical traffic flow transfer rate.

[0087] S10622: Employs a bidirectional LSTM network layer, with the input being the temporal traffic data of the nodes. It captures the short-term and long-term temporal dependencies of the traffic data through forward and backward hidden states.

[0088] The bidirectional LSTM network in this embodiment contains two hidden layers, each with 128 hidden units. The input time-series traffic data is divided into fixed time periods. The forward LSTM calculates the hidden states from the beginning of the time series backward, while the backward LSTM calculates from the end of the time series backward. Finally, the hidden states from both directions are concatenated as the output time feature. In this way, it can simultaneously focus on past and future traffic state changes, capture periodic time patterns such as morning and evening peak hours, and improve the representation ability of time features.

[0089] S10623: Employs a graph convolutional layer. Based on the graph structure constructed in S10621, it performs neighborhood aggregation on node features and integrates the feature information of adjacent nodes into the current node through a weight matrix, thereby capturing the spatial correlation of the road network.

[0090] In this embodiment, the convolutional layer has two layers, and the output feature dimension of each layer is 64. The first layer aggregates the node's own features and the features of its directly adjacent nodes, and the second layer aggregates the features output by the first layer and the features of its second-order adjacent nodes.

[0091] Optionally, the aggregation method uses mean pooling, i.e., current node feature = (its own feature + Σ neighboring node features × edge weight) / (1 + number of neighboring nodes). This allows the node features to include spatial correlation information of the local road network and capture the mutual influence of traffic flow between road segments.

[0092] S10624: Set up spatial attention layer and temporal attention layer respectively. Calculate the influence weight of different neighboring nodes on the current node through spatial attention, and calculate the contribution weight of features at different time steps to the current prediction through temporal attention. After normalizing the weights through the Softmax function, the features are weighted and fused.

[0093] S10625: Input the fused features output from S10624 into the fully connected layer, and output the traffic state prediction value and the corresponding traffic control decision parameters for the future preset time period through linear transformation and activation function, respectively.

[0094] In this embodiment, feature integration employs a stitching method, stitching together the fused spatiotemporal features with road segment length and lane number. The fully connected layer consists of two layers. The first layer has an output dimension of 32 and uses the ReLU activation function. The output dimension of the second layer is set according to the task; for example, predicting traffic flow for the next three cycles results in a 3-dimensional output, while signal timing adjustment results in a 2-dimensional output, thus realizing the transformation from feature extraction to task output.

[0095] In one embodiment of the present invention, based on step S107, the following is a possible embodiment and its specific implementation will be described in a non-limiting manner. S107 specifically includes the following methods: S1071: Configure the real-time performance and computational complexity metrics of the traffic tasks to be processed, and generate task feature vectors.

[0096] S1072: Classify tasks into edge processing or cloud processing categories based on task feature vectors.

[0097] S1073: Based on the task category identifier, route the task data to the corresponding traffic edge side.

[0098] S1074: During task execution, continuously monitor the resource utilization status and task queue status at the traffic edge.

[0099] This embodiment transforms the real-time requirements and computational complexity of the tasks to be processed into quantifiable feature vectors. Tasks with high real-time requirements and low complexity are categorized as edge processing. Based on the feature vectors, tasks are classified into edge-side or cloud-side processing categories, clearly defining the task processing nodes. Task data is routed to the corresponding edge side according to the classification results, and task execution is initiated. During execution, S1074 continuously tracks the resource utilization status and task queue length on the edge side, capturing changes in system status. When S1075 detects that the status exceeds a preset threshold, a reallocation mechanism is triggered, transferring some tasks to the cloud or other edge nodes, dynamically adjusting the routing strategy. This invention ensures task-node capability matching through initial classification and responds to system status fluctuations through real-time monitoring and dynamic adjustment, achieving accuracy and adaptability in task allocation.

[0100] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0101] like Figure 2 As shown, this application also provides an electronic device, including a display module 103, a memory 102, a processor 101, a communication module 104, and a computer program stored in the memory and executable on the processor 101. When the processor 101 executes the program, it implements the steps of a cloud-edge collaborative intelligent transportation data fusion and analysis method.

[0102] In embodiments of the present invention, electronic devices include, but are not limited to, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments described and / or claimed herein.

[0103] In this embodiment, processor 101 may be implemented using at least one of an application-specific integrated circuit, a programmable logic device, a field-programmable gate array, a processor, a controller, a microcontroller, a microprocessor, or an electronic unit designed to perform the functions described herein. In some cases, such an implementation may be implemented within a controller. For software implementation, implementations such as processes or functions may be implemented with separate software modules that allow the performance of at least one function or operation. Software code may be implemented by a software application (or program) written in any suitable programming language, and the software code may be stored in memory and executed by the controller.

[0104] The display module 103 is used to display information input by the user or information provided to the user. The display module 103 may include a display panel, which may be configured in the form of a liquid crystal display, an organic light-emitting diode, or the like.

[0105] The memory 102 can be used to store software programs and various data. The memory 102 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0106] The communication module 104 transmits radio signals to and / or receives radio signals from at least one of a base station, an external terminal, and a server. Such radio signals may include voice call signals, video call signals, or various types of data sent and / or received according to text and / or multimedia messages.

[0107] The present invention also provides a storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of the cloud-edge collaborative intelligent transportation data fusion and analysis method.

[0108] The storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example,, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0109] The storage medium stores a program product capable of implementing the methods described above in this specification. In some possible implementations, various aspects of this disclosure may also be implemented as a program product comprising program code that, when run on a terminal device, causes the terminal device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure.

[0110] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A cloud-edge collaborative intelligent transportation data fusion and analysis method, characterized in that, The methods include: S101: At the traffic edge, collect multi-source traffic data from traffic cameras, radar sensors, and traffic signal controllers; S102: Clean and compress multi-source traffic data to obtain preprocessed traffic data; S103: Perform local traffic analysis based on preprocessed traffic data to generate local traffic analysis results; S104: Upload preprocessed traffic data and local traffic analysis results to the traffic cloud platform; S105: On the traffic cloud platform, the uploaded preprocessed traffic data and local traffic analysis results are fused to obtain global traffic situation data; S106: Train the traffic prediction and decision-making model based on global traffic situation data to obtain the updated traffic model; S107: Based on the complexity and real-time requirements of traffic data processing, allocate traffic processing tasks to the traffic edge side; S108: The traffic cloud platform sends traffic control commands to the traffic edge side, which executes the traffic control commands and provides feedback on the execution effect data.

2. The intelligent transportation data fusion and analysis method based on cloud-edge collaboration according to claim 1, characterized in that, S103 specifically includes the following methods: Traffic flow parameters are extracted from preprocessed traffic data, and traffic flow, average speed, and density are calculated. Based on the extracted traffic flow parameters, a classification model is used to classify traffic conditions and identify traffic congestion and free flow. Perform moving target detection and tracking on video data in preprocessed traffic data to identify vehicle and pedestrian trajectories; Traffic incident detection is performed based on the results of moving target detection and tracking. By integrating traffic condition classification results and traffic incident detection results, local traffic analysis results are generated through data fusion methods.

3. The intelligent transportation data fusion and analysis method based on cloud-edge collaboration according to claim 1, characterized in that, S105 specifically includes the following methods: The preprocessed traffic data and local traffic analysis results are aligned based on a unified timestamp and geographic location. For the aligned duplicate data, the Kalman filter algorithm is used to fuse them and output consistent traffic flow and average vehicle speed indicators. Extract the target trajectory features from the YOLOv5s output and the traffic flow and occupancy rate from the radar data, calculate the feature weights, and weightedly fuse them to generate cross-modal traffic features; By combining the local traffic analysis results generated by S103 with the cloud-based historical traffic situation model, a preliminary global situation including anomaly impact assessment is obtained. Consistent indicators, cross-modal features, and preliminary global situation data are summarized and integrated into global traffic situation data.

4. The intelligent transportation data fusion and analysis method based on cloud-edge collaboration according to claim 1, characterized in that, S106 specifically includes the following methods: S1061: Standardize and serialize the global traffic situation data to construct a model training dataset; S1062: Construct a spatiotemporal graph neural network model based on attention mechanism as a traffic prediction and decision-making model; S1063: Use a distributed computing framework to perform multiple rounds of iterative training on the model and optimize the model parameters; S1064: Use the validation dataset to evaluate the performance of the trained model and calculate the prediction accuracy metric; S1065: When the model performance meets the requirements, generate updated model parameters and prepare for distribution.

5. The intelligent transportation data fusion and analysis method based on cloud-edge collaboration according to claim 4, characterized in that, S1062 specifically includes the following methods: Construct a traffic network graph structure, setting road segments or intersections as graph nodes. Node attributes include road segment length, number of lanes, and historical average traffic flow. The connection between adjacent nodes is set as graph edges, and the edge weight is calculated based on the actual distance between nodes and the historical traffic flow transfer rate. A bidirectional LSTM network layer is used, with the input being the temporal traffic data of the nodes. The short-term and long-term temporal dependencies of the traffic data are captured through forward and backward hidden states. A graph convolutional layer is used to perform neighborhood aggregation on node features based on the constructed graph structure. The feature information of neighboring nodes is fused into the current node through a weight matrix to capture the spatial correlation of the road network. Spatial attention layer and temporal attention layer are set up respectively. Spatial attention calculates the influence weight of different neighboring nodes on the current node, and temporal attention calculates the contribution weight of features at different time steps to the current prediction. After normalizing the weights through the Softmax function, the features are weighted and fused. The fused features are input into a fully connected layer, and the predicted traffic conditions and corresponding traffic control decision parameters for the future preset time period are output through linear transformation and activation function, respectively.

6. The intelligent transportation data fusion and analysis method based on cloud-edge collaboration according to claim 4, characterized in that, S107 specifically includes the following methods: Configure the real-time performance and computational complexity metrics of the traffic tasks to be processed, and generate task feature vectors; Tasks are categorized into edge processing or cloud processing categories based on their feature vectors. Based on the task category identifier, the task data is routed to the corresponding traffic edge side; During task execution, the resource utilization status and task queue status at the traffic edge are continuously monitored.

7. The intelligent transportation data fusion and analysis method based on cloud-edge collaboration according to claim 1, characterized in that, S108 specifically includes the following methods: The traffic cloud platform will transmit control commands to the designated traffic edge side; The traffic edge receives and parses control commands, converts them into operational commands that can be executed by the equipment, and drives the traffic signal controller or information dissemination equipment to execute them. Traffic status change data after the execution of instructions is collected at the traffic edge, and compared with the status before execution to generate execution effect evaluation indicators; The traffic edge will package the performance evaluation indicators with the original control commands and send them back to the cloud platform. The transportation cloud platform stores the execution effect data and corresponding decision records in the experience playback buffer.

8. The intelligent transportation data fusion and analysis method based on cloud-edge collaboration according to claim 2, characterized in that, In S103, a classification model is used to classify traffic conditions and identify traffic congestion and smooth traffic conditions. The classification model is f(x)=sign(Σαi×yi×K(xi,x)+b), where x is the input feature vector based on flow, speed and density, yi is the class label, K is the kernel function, xi represents the feature vector of the i-th support vector, αi and b are the model parameters, and sign() is the sign function. In generating local traffic analysis results using data fusion methods, a weighted average fusion method is used. The final result JTR = w1×S + w2×E, where S is the state classification score, E is the event detection score, w1 and w2 are the weights, and JTR is the comprehensive output value obtained after weighted fusion.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the intelligent transportation data fusion and analysis method based on cloud-edge collaboration as described in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent transportation data fusion and analysis method based on cloud-edge collaboration as described in any one of claims 1 to 7.

Images