Automobile collaborative driving method and system and medium

Abstract

The invention relates to an automobile cooperative driving method and system and a medium, and relates to the technical field of intelligent network connection automobiles, and the method comprises the steps: constructing a block chain network comprising a plurality of intelligent automobile nodes; deploying a federated learning framework on each intelligent automobile node; each intelligent automobile node trains a collaborative driving model based on local data; the trained model parameters are uploaded to the block chain network; aggregating the model parameters through the block chain network and updating the collaborative driving model; and executing a collaborative driving decision based on the updated collaborative driving model. According to the multi-vehicle cooperative driving method, efficient and safe multi-vehicle cooperative driving can be realized while vehicle data privacy is protected, and the credibility, the real-time performance and the model generalization ability of multi-vehicle cooperative driving are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent connected vehicle technology, and in particular to a method, system and medium for cooperative driving of automobiles. Background Technology

[0002] With the rapid development of intelligent connected vehicle technology, multi-vehicle cooperative driving has become an important direction for improving road utilization and driving safety. Currently, multi-vehicle cooperative driving mainly relies on real-time data sharing between vehicles and a centralized decision-making system. Vehicles collect information about their surrounding environment through onboard sensors, upload the data to a cloud server for processing and decision-making, and then issue instructions to each vehicle to execute cooperative actions. This method can achieve information sharing and cooperative control between vehicles, but it also faces some challenges.

[0003] However, existing multi-vehicle cooperative driving systems suffer from problems such as data privacy risks, lack of trust between vehicles, low data sharing efficiency, and high model training costs. The need to upload large amounts of raw data to a central server easily leads to the leakage of user privacy information. Simultaneously, the lack of effective trust mechanisms between vehicles makes it difficult to guarantee the authenticity and integrity of shared data. Furthermore, centralized data processing methods easily create data transmission bottlenecks, affecting the real-time performance of collaborative decision-making. Moreover, centralized model training methods are ill-suited to complex and ever-changing road environments, resulting in limited model generalization capabilities. Summary of the Invention

[0004] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a method, system and medium for cooperative driving of automobiles, which can achieve efficient and safe multi-vehicle cooperative driving while protecting vehicle data privacy, and improve the system's credibility, real-time performance and model generalization ability.

[0005] To achieve the above objectives, the present invention adopts the following technical solution.

[0006] Firstly, the present invention provides a vehicle cooperative driving method, which adopts the following technical solution: Construct a blockchain network containing multiple smart car nodes; Deploy a federated learning framework on each of the aforementioned intelligent vehicle nodes; Each of the aforementioned intelligent vehicle nodes trains a cooperative driving model based on local data; The trained model parameters are uploaded to the blockchain network; The model parameters are aggregated and the cooperative driving model is updated through the blockchain network; and Cooperative driving decisions are made based on the updated cooperative driving model.

[0007] Furthermore, in the aforementioned vehicle cooperative driving method, the construction of a blockchain network comprising multiple intelligent vehicle nodes includes: Establish communication connections between intelligent vehicle nodes, roadside equipment nodes, and traffic management center nodes; and Deploy smart contracts on the blockchain network, which define cooperative driving rules, data sharing protocols, and reward mechanisms.

[0008] Furthermore, in the above-mentioned vehicle cooperative driving method, before each intelligent vehicle node trains the cooperative driving model based on local data, the method further includes: Vehicle status data and surrounding environment data are collected through onboard sensors; and The collected data is preprocessed, including data cleaning and feature extraction.

[0009] Furthermore, in the above-mentioned cooperative driving method, the training of the cooperative driving model by each intelligent vehicle node based on local data includes: Download the initial cooperative driving model from the blockchain network; The initial cooperative driving model is trained using preprocessed local data; and The performance of the cooperative driving model was evaluated using a validation set.

[0010] Furthermore, in the above-mentioned vehicle cooperative driving method, the initial cooperative driving model includes an environmental perception sub-model, a decision-making and planning sub-model, and a communication and coordination sub-model. Training the initial cooperative driving model using preprocessed local data includes: The preprocessed local data is then input into the environmental perception sub-model, the decision-making and planning sub-model, and the communication and collaboration sub-model, respectively. A multi-task learning approach is adopted to simultaneously optimize the loss functions of the three sub-tasks: environmental perception, decision planning, and communication and coordination. Based on a preset joint loss function, the outputs of the three sub-models are weighted and fused. The joint loss function is optimized using the gradient descent algorithm, while the parameters of the three sub-models are updated simultaneously. During training, the weights between the sub-models are dynamically adjusted to balance the importance of different tasks; The backpropagation algorithm is used to pass the fused gradient information to each sub-model, achieving end-to-end joint optimization.

[0011] Furthermore, in the above-mentioned vehicle cooperative driving method, uploading the trained model parameters to the blockchain network includes: Encrypt the trained model parameters; and The encrypted model parameters are uploaded to the blockchain network.

[0012] Furthermore, in the above-mentioned vehicle cooperative driving method, the step of aggregating the model parameters and updating the cooperative driving model through the blockchain network includes: Use smart contracts to automatically aggregate model parameters uploaded by each smart car node; Update the global cooperative driving model based on the aggregated model parameters; and The updated global collaborative driving model is distributed to each intelligent vehicle node.

[0013] Furthermore, in the above-mentioned vehicle cooperative driving method, the step of performing cooperative driving decisions based on the updated cooperative driving model includes: Collect real-time vehicle status data and surrounding environment data; Input the real-time data into the updated cooperative driving model; and Based on the output of the cooperative driving model, perform cooperative lane changing or cooperative deceleration operations.

[0014] Furthermore, the aforementioned cooperative driving method for automobiles also includes: Evaluate the effectiveness of the cooperative driving system and iteratively optimize the cooperative driving model based on the evaluation results; The evaluation of the cooperative driving effect includes: Calculate safety and efficiency indicators during cooperative driving; and The performance of the cooperative driving model in different driving scenarios was analyzed.

[0015] Furthermore, the aforementioned cooperative driving method for automobiles also includes: Based on a pre-designed blockchain-based incentive mechanism, smart car nodes are encouraged to actively participate in data sharing and collaborative driving.

[0016] Secondly, the present invention provides a vehicle cooperative driving system, which adopts the following technical solution: A blockchain network building module is used to build a blockchain network containing multiple smart car nodes; A federated learning framework deployment module is used to deploy the federated learning framework on each of the intelligent vehicle nodes; The model training module is used to train a cooperative driving model based on local data on each of the intelligent vehicle nodes; The parameter upload module is used to upload the trained model parameters to the blockchain network; A model update module is used to aggregate the model parameters and update the cooperative driving model through the blockchain network; and The decision execution module is used to execute cooperative driving decisions based on the updated cooperative driving model.

[0017] Thirdly, the present invention provides a readable storage medium, which adopts the following technical solution: A readable storage medium storing computer instructions that, when executed by a processor, implement the vehicle cooperative driving method as described in any one of the first aspects above.

[0018] In summary, compared with existing technologies, this invention can achieve efficient and safe multi-vehicle cooperative driving while protecting vehicle data privacy. This method can reduce the risk of data leakage, increase trust between vehicles, and improve the model's generalization ability and real-time performance through distributed learning. Furthermore, the decentralized nature of blockchain networks helps reduce data transmission bottlenecks, improving reliability and scalability. This innovative combination of technologies can significantly improve the safety, efficiency, and adaptability of cooperative driving, providing a new solution for the development of intelligent connected vehicle technology. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of an embodiment of a vehicle cooperative driving method according to the present invention.

[0021] Figure 2 This is a schematic diagram illustrating an implementation scenario of the vehicle cooperative driving method of the present invention.

[0022] Figure 3 This is a flowchart of an embodiment of the present invention for establishing a network communication connection for intelligent vehicles.

[0023] Figure 4 This is a flowchart of an embodiment of the present invention for downloading and training a cooperative driving model.

[0024] Figure 5 This is a flowchart of another embodiment of the present invention for training a cooperative driving model.

[0025] Figure 6 This is a flowchart of an embodiment of the present invention for processing cooperative driving model parameters.

[0026] Figure 7 This is a flowchart of an embodiment of the present invention for updating the parameters of the cooperative driving model.

[0027] Figure 8 This is a flowchart of an embodiment of the present invention performing cooperative driving.

[0028] Figure 9 This is a flowchart of another embodiment of a vehicle cooperative driving method of the present invention.

[0029] Figure 10 This is a flowchart of yet another embodiment of a vehicle cooperative driving method according to the present invention.

[0030] Figure 11 This is a schematic diagram of an embodiment of a vehicle cooperative driving system according to the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Furthermore, it should be understood that the specific embodiments described herein are only for illustration and explanation of this application and are not intended to limit this application.

[0032] It should be noted that the order of description of the following embodiments is not intended to limit the preferred order of the embodiments of this application. Furthermore, the descriptions of each embodiment in the following embodiments have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0033] The method steps described in this embodiment of the invention can be executed in the order described in the specific implementation, or the execution order of each step can be adjusted according to actual needs, provided that the technical problem can be solved. These are not listed one by one here.

[0034] To address the problems existing in current multi-vehicle cooperative driving systems, this invention proposes a smart car cooperative driving method based on blockchain and federated learning. This method constructs a decentralized blockchain network to achieve secure and reliable data sharing among vehicles, while utilizing federated learning technology to perform distributed model training while protecting data privacy, thereby improving the efficiency and safety of cooperative driving.

[0035] The specific embodiments of the present invention will be described in detail below so that those skilled in the art can better understand and implement the present invention.

[0036] Reference Figure 1 and Figure 2 This invention discloses a method for cooperative driving of automobiles, which includes the following steps.

[0037] S1. Construct a blockchain network comprising multiple intelligent vehicle nodes. Specifically, construct a blockchain network composed of multiple intelligent vehicle nodes, which may also include participating nodes such as Roadside Units (RSUs) and traffic management centers. Each node establishes a secure connection through an encrypted communication protocol and collaboratively constructs a distributed ledger based on a pre-defined consensus mechanism (such as PoS, PBFT, etc.) to store model parameters and collaborative driving-related data. Simultaneously, deploy smart contracts within this blockchain network to define rules for data sharing, model updates, and incentive mechanisms, supporting the decentralized operation and automated management of the entire collaborative driving system.

[0038] S2. Deploy a federated learning framework on each intelligent vehicle node. Specifically, deploy a federated learning framework on each intelligent vehicle node, enabling it to perform local model training, parameter updates, and result uploads. This federated learning framework may include a task coordination module, a local training module, and a parameter communication interface. The task coordination module receives the model structure and training tasks from the blockchain network. The local training module is responsible for training the cooperative driving model using sensor data and status information collected locally by the vehicle. The parameter communication interface is used to upload encrypted model parameters to the blockchain network and receive global model updates, ensuring that the collaborative modeling process is completed without leaking the original data.

[0039] S3: Each intelligent vehicle node trains a cooperative driving model based on local data. Each intelligent vehicle node utilizes locally collected sensor data (such as LiDAR, cameras, millimeter-wave radar, etc.) and vehicle state information (such as speed, acceleration, steering angle, etc.) to train the cooperative driving model locally within the deployed federated learning framework. The specific process includes: downloading the initial model as the training starting point; dividing the preprocessed data (training set, validation set, test set); designing an appropriate loss function (such as path error or decision accuracy); iteratively training using optimization algorithms (such as stochastic gradient descent (SGD); and continuously adjusting the model structure and parameters based on model evaluation results, thereby improving the model's adaptability and generalization ability in local scenarios.

[0040] S4. The trained model parameters are uploaded to the blockchain network. Specifically, after completing local model training, each smart car node encrypts the trained model parameters to ensure data security and privacy protection during transmission. The encrypted model parameters include, but are not limited to, the vehicle's driving trajectory, speed, acceleration, steering angle, and other relevant feature weights. Subsequently, using the communication interface connected to the blockchain network, and following the upload rules defined by the smart contract, the encrypted model parameters are packaged into transaction data and uploaded to the blockchain network. Each node jointly verifies its validity and integrity and records it in the distributed ledger for subsequent aggregation.

[0041] S5, the model parameters are aggregated and the cooperative driving model is updated through the blockchain network. Specifically, the blockchain network automatically triggers the aggregation process of model parameters through a preset smart contract. First, each node verifies the uploaded model parameters to ensure their integrity and reliable source. Then, an aggregation algorithm in federated learning (such as FedAvg) is used to perform a weighted average or other form of fusion on the model parameters uploaded by multiple smart car nodes to generate a unified global cooperative driving model. This aggregation process can be executed by a specific node in the blockchain network (such as RSU or management center) or can be completed collaboratively through a distributed consensus mechanism. The updated model parameters are synchronously distributed to each smart car node through the blockchain network to achieve consistent model updates across the entire network.

[0042] S6. Based on the updated cooperative driving model, cooperative driving decisions are executed. Specifically, each intelligent vehicle node receives and loads the latest cooperative driving model aggregated by the blockchain network, and combines it with currently collected real-time vehicle status information (such as speed, acceleration, steering angle, etc.) and environmental perception data (such as road conditions, obstacle positions, surrounding vehicle behavior, etc.) to perform model inference and decision calculation locally. Based on the updated model output, the intelligent vehicle executes corresponding cooperative driving behaviors, such as cooperating with other vehicles to change lanes, decelerate, accelerate, or avoid obstacles in scenarios such as lane changing, merging guidance, or platooning, thereby improving overall traffic efficiency and driving safety.

[0043] The vehicle cooperative driving method described in this invention, by constructing a blockchain network and deploying a federated learning framework, achieves distributed training and updating of the cooperative driving model without disclosing the original data. This not only effectively protects the privacy data of vehicle owners but also establishes trust relationships among multiple vehicles using the immutability of blockchain and smart contract mechanisms. Simultaneously, the method improves the efficiency of data sharing and model iteration through encrypted parameter uploading and network-wide aggregated updates, and significantly enhances the system's response speed and driving safety through local collaborative decision-making based on real-time data. Overall, it enhances the intelligence, reliability, and scalability of the multi-vehicle cooperative driving system.

[0044] Furthermore, as one embodiment of the present invention, refer to Figure 3 Step S1, constructing a blockchain network containing multiple smart car nodes includes: S11 establishes communication connections between intelligent vehicle nodes, roadside equipment nodes, and traffic management center nodes; and S12, Deploy a smart contract on the blockchain network, the smart contract defining cooperative driving rules, data sharing protocols and reward mechanisms.

[0045] Specifically, firstly, step S11 establishes a secure and reliable communication connection between multiple smart car nodes, roadside unit (RSU) nodes, and traffic management center nodes using encrypted communication protocols (such as TLS or SSL). Each node accesses the blockchain network through an onboard communication module (supporting technologies such as 5G and C-V2X) to ensure stable transmission of real-time data and model parameters between nodes. Simultaneously, to ensure network scalability and low latency, a layered communication architecture can be adopted. The smart cars communicate with the RSUs over short distances at high speed, and the RSUs are then connected to the traffic management center via fiber optic or wireless links, forming a multi-layered, interconnected decentralized network structure that provides the communication foundation for subsequent smart contract execution and model synchronization.

[0046] Next, in step S12, smart contracts are developed and deployed on the established blockchain network using a smart contract programming language (such as Solidity). These smart contracts include cooperative driving rules (such as the interaction logic for lane changing, following, and obstacle avoidance between vehicles), a data sharing protocol (specifying the upload format, verification method, and update frequency of model parameters), and a reward mechanism (allocating digital tokens or points as incentives based on node participation). During deployment, the smart contracts are registered on the blockchain platform (such as Ethereum or Hyperledger Fabric) and verified through consensus among nodes in the network, ensuring that they can be automatically triggered, tamper-proof, and executed synchronously across the entire network during cooperative driving, thereby guaranteeing the standardization of collaboration and the fairness of system operation.

[0047] Finally, step S2 deploys a federated learning framework within each smart car node. This framework consists of a task scheduling module, a local training module, and a parameter communication interface. The smart contract automatically triggers the distribution of training tasks according to on-chain rules. The task scheduling module downloads the initial model and training configuration accordingly; the local training module uses real-time data collected by vehicle sensors to train the model; and the parameter communication interface encrypts the trained model parameters according to the upload rules defined by the smart contract and prepares them for uploading to the blockchain network. Throughout this process, the federated learning framework and the on-chain smart contract work together to achieve automated initiation of training tasks, privacy protection, and unified control.

[0048] Furthermore, as an embodiment of the present invention, step S3, before each intelligent vehicle node trains the cooperative driving model based on local data, further includes: collecting vehicle status data and surrounding environment data through onboard sensors; and preprocessing the collected data, including data cleaning and feature extraction.

[0049] Specifically, each intelligent vehicle node collects real-time vehicle status data (such as speed, acceleration, steering angle, and braking status) and surrounding environment data (such as road structure, obstacle location, and neighboring vehicle behavior) through onboard sensor systems (such as LiDAR, millimeter-wave radar, cameras, and ultrasonic sensors). After collection, the raw data undergoes preprocessing. First, data cleaning is performed to remove noise, outliers, and duplicate data, improving data quality. Then, feature extraction is performed, using image processing algorithms, point cloud analysis techniques, and signal processing methods to extract key driving features, such as lane edges, target trajectories, object categories, and relative speeds, providing high-quality input data for the subsequent local training of the cooperative driving model.

[0050] Furthermore, as one embodiment of the present invention, refer to Figure 4 Step S3, each intelligent vehicle node trains a cooperative driving model based on local data, including the following steps.

[0051] S31, Download the initial cooperative driving model from the blockchain network. Specifically, after receiving the training task instruction issued by the smart contract in the blockchain network, each smart car node accesses the initial cooperative driving model stored on the chain through the blockchain interface module. This model is generated from the parameter aggregation results of the previous round and stored in encrypted form in the distributed ledger. The node uses on-chain index information or locates the corresponding model file through the Web3 interface, and uses the lightweight blockchain client embedded in the vehicle to verify, decrypt, and download it, ensuring the integrity and trustworthiness of the model data. After downloading, the initial model will serve as the starting point for local training, used to further optimize it in conjunction with the local data currently collected by the vehicle.

[0052] S32, the initial cooperative driving model is trained using preprocessed local data. Specifically, each intelligent vehicle node inputs preprocessed local data (including vehicle status information, environmental perception data, and communication interaction data) into the downloaded initial cooperative driving model for local training. The training process includes dividing the data into training and validation sets, using appropriate loss functions (such as mean squared error of path error, cross-entropy loss for behavior classification) to measure the difference between the model output and the expected result, and using optimization algorithms (such as stochastic gradient descent SGD or Adam) to iteratively update the model parameters. During this process, the model continuously adjusts to better adapt to the local driving environment, realizing functions such as cooperative perception, cooperative decision-making, and path planning, thereby enhancing the model's adaptability and generalization ability to specific scenarios.

[0053] S33. The performance of the cooperative driving model is evaluated using a validation set. After completing model training, each intelligent vehicle node uses a locally partitioned validation set to evaluate the performance of the trained cooperative driving model. During the evaluation process, the input data from the validation set is fed into the model, and the difference between the model's output and the true labels is compared. Multiple metrics such as accuracy, recall, mean squared error, and path deviation are calculated to measure the model's predictive ability and decision-making reliability in different driving scenarios. Simultaneously, methods such as visual path reconstruction or decision behavior reproduction can be used to assist in analyzing the model's generalization ability and potential weaknesses, providing a quality reference for subsequent model uploading and aggregation.

[0054] Furthermore, as one embodiment of the present invention, the initial cooperative driving model includes an environmental perception sub-model, a decision-making and planning sub-model, and a communication and coordination sub-model. Correspondingly, refer to... Figure 5 Step S32, training the initial cooperative driving model using preprocessed local data, includes the following sub-steps.

[0055] S321, the preprocessed local data is input into the environmental perception sub-model, the decision-making and planning sub-model, and the communication and coordination sub-model, respectively. Specifically, the preprocessed local data is input into the three sub-models of the initial cooperative driving model according to different task requirements. The environmental perception sub-model receives data from sensors such as LiDAR and cameras to extract features such as road structure, obstacles, and lane edges; the decision-making and planning sub-model receives vehicle state information (such as speed, acceleration, and steering angle) and environmental perception results to formulate driving strategies and path planning; and the communication and coordination sub-model receives other vehicle behavior information and collaborative intentions from the V2X communication module to analyze the interaction relationships and coordination commands between vehicles. By inputting different types of data into each sub-model in a targeted manner, it is ensured that each sub-task obtains the key feature information required, providing an accurate input foundation for subsequent multi-task joint training.

[0056] S322 employs a multi-task learning approach, simultaneously optimizing the loss functions for three sub-tasks: environmental perception, decision planning, and communication collaboration. Specifically, during training, a multi-task learning method is used, constructing independent loss functions for each of the three sub-tasks. For example, the cross-entropy loss function is used to evaluate the accuracy of the environmental perception sub-model in target recognition and classification, the mean squared error is used to evaluate the deviation of the decision planning sub-model in path prediction, and a graph structure loss function is used to evaluate the performance of the communication collaboration sub-model in modeling vehicle-to-vehicle interaction relationships. These three loss functions are then calculated in parallel and incorporated into a unified training process, achieving synchronous optimization of the parameters of each sub-model and improving the overall performance and collaboration of the model in multiple cooperative driving tasks.

[0057] S323, based on a preset joint loss function, weightedly fuses the outputs of the three sub-models. Specifically, after obtaining the outputs and corresponding task losses of the environment perception sub-model, decision-making and planning sub-model, and communication and coordination sub-model, the outputs of the three sub-models are weighted and fused according to the preset joint loss function. This joint loss function adopts a weighted summation form, that is, multiplying the loss of each sub-task by a corresponding task weight coefficient, and then adding them together to obtain the overall loss value. The weight coefficients can be preset according to the importance of the tasks or dynamically adjusted during training to take into account the contribution of each sub-task in model training. In this way, the optimization objectives of multiple tasks are integrated into a global objective, providing a unified loss basis for subsequent gradient calculation and parameter updates, and realizing the collaborative training of multiple sub-models.

[0058] S324 uses gradient descent to optimize the joint loss function while updating the parameters of the three sub-models. Specifically, after obtaining the weighted and fused joint loss function, gradient descent is used to optimize it. This involves calculating the gradient of the joint loss function relative to the parameters of each sub-model and updating the weight parameters of the environmental perception, decision-making, and communication / coordination sub-models based on the gradient values. Standard stochastic gradient descent (SGD) or adaptive learning rate algorithms (such as Adam or RMSProp) can be used during optimization to improve convergence efficiency and training stability. This optimization process iteratively minimizes the joint loss, enabling the three sub-models to achieve parameter co-evolution while maintaining task independence, thereby improving the overall performance and robustness of the cooperative driving model.

[0059] S325 dynamically adjusts the weights of each sub-model during training to balance the importance of different tasks. Specifically, during joint training, the weight ratios of the three sub-models—environmental perception, decision-making and planning, and communication and collaboration—in the joint loss function are dynamically adjusted based on the training convergence of each sub-task and the changing trend of the loss function. For example, when the loss of a certain sub-task decreases slowly or fluctuates significantly, its corresponding weight can be appropriately increased to enhance the optimization effect; conversely, its weight can be decreased to avoid overfitting or wasting resources. The weight adjustment strategy can be implemented based on methods such as uncertainty weighting, adaptive weighting, or gradient normalization, thereby achieving a reasonable allocation of training resources among multiple tasks and improving the overall model's learning efficiency and task balance.

[0060] S326 utilizes the backpropagation algorithm to transmit the fused gradient information to each sub-model, achieving end-to-end joint optimization. Specifically, after calculating the total gradient of the joint loss function, the backpropagation algorithm transmits this gradient information layer by layer from the output layer back to each network layer of the environmental perception sub-model, decision-making and planning sub-model, and communication and coordination sub-model, ensuring that each sub-model receives the corresponding gradient signal according to its weight in the joint loss. The backpropagation process calculates the partial derivatives of the parameters of each layer and updates their weights accordingly, achieving end-to-end collaborative optimization among the sub-models. This process is repeated in each training iteration, ensuring that multiple tasks share information while maintaining task specificity, thereby improving the overall performance and consistency of the entire cooperative driving model in perception, planning, and collaborative decision-making.

[0061] Furthermore, as one embodiment of the present invention, refer to Figure 6 Step S4, uploading the trained model parameters to the blockchain network includes: S41, encrypt the parameters of the trained model; and S42, upload the encrypted model parameters to the blockchain network.

[0062] Specifically, after completing local cooperative driving model training, the intelligent vehicle nodes encrypt the model parameters to ensure data security and privacy protection during the upload process. This is achieved by using symmetric or asymmetric encryption algorithms, such as AES (Advanced Encryption Standard) or RSA, to encapsulate key weight matrices, bias parameters, and other learnable variables in the model. Simultaneously, a hash algorithm (such as SHA-256) can be used to generate a model parameter digest for subsequent integrity verification on the blockchain. Differential privacy mechanisms can also be introduced during encryption to perturb the parameters, further preventing potential reverse engineering attacks and thus achieving parameter-level data protection.

[0063] After encrypting the model parameters, the smart car nodes encapsulate the encrypted model parameters into transaction data via the blockchain communication interface and broadcast it to the blockchain network according to a pre-defined blockchain protocol (such as Ethereum or Hyperledger Fabric). During the upload process, the smart contract verifies the format and permissions of the transaction data to ensure that the data source is legitimate and meets the conditions for uploading to the blockchain. Subsequently, the transaction is confirmed by multiple nodes in the blockchain network through a consensus mechanism (such as PoS or PBFT), and the model parameter data is recorded in the distributed ledger. After the upload is complete, other nodes can access the parameter data through the blockchain, providing a trusted foundation for subsequent model aggregation and global updates.

[0064] Furthermore, as one embodiment of the present invention, refer to Figure 7Step S5, aggregating the model parameters and updating the cooperative driving model through the blockchain network includes: S51 uses smart contracts to automatically aggregate model parameters uploaded by each smart car node; S52 updates the global cooperative driving model based on the aggregated model parameters; and S53 distributes the updated global cooperative driving model to each intelligent vehicle node.

[0065] Specifically, after detecting that multiple smart car nodes have successfully uploaded encrypted model parameters, the smart contract in the blockchain network automatically triggers a model aggregation task. The smart contract first verifies the integrity and signature of all uploaded parameters to ensure their legitimate origin and tamper-proof nature. Then, it calls the on-chain aggregation logic, using a commonly used aggregation algorithm in federated learning (such as the FedAvg algorithm) to perform a weighted average or weighted fusion of the model parameters from each node. The weights can be dynamically adjusted based on the number of samples, training epochs, or model convergence degree of each node. The entire aggregation process is executed transparently on-chain, without relying on any central node, thereby achieving reliable fusion and collaborative evolution of multi-vehicle model parameters.

[0066] After aggregating the model parameters of each intelligent vehicle node, the smart contract in the blockchain network uses the aggregation result as the global model parameters. This is then combined with the structure of the original cooperative driving model to update it, generating a new global cooperative driving model. The update process can employ model parameter replacement or incremental updates. The aggregated parameters can directly overwrite the old model parameters, or the model weights can be smoothly adjusted based on strategies such as moving averages to improve model stability and generalization ability. The updated model is version-identified and stored on the blockchain, and its data integrity and consistency are ensured through an on-chain hash verification mechanism, providing a reliable initial model foundation for the next round of cooperative training.

[0067] After generating and storing the updated global cooperative driving model, the blockchain network synchronizes the model version information and storage address to all smart car nodes via an on-chain broadcast mechanism. Each node queries the latest model's hash checksum and metadata through a blockchain client or interface module, and downloads the model parameter file using the Web3 interface or light node function. During the download process, nodes verify the integrity of the model data to ensure it has not been tampered with. Subsequently, each smart car node loads the global model into its local federated learning framework as the starting point for a new round of local training, thus achieving network-wide synchronization and iterative closed-loop model updates.

[0068] Furthermore, as one embodiment of the present invention, refer to Figure 8 Step S6, performing cooperative driving decisions based on the updated cooperative driving model includes: S61 collects real-time vehicle status data and surrounding environment data; S62, input the real-time data into the updated cooperative driving model; and S63, perform cooperative lane changing or cooperative deceleration operations based on the output of the cooperative driving model.

[0069] Specifically, each intelligent vehicle node collects real-time vehicle status data and surrounding environmental information through its onboard sensor system. Vehicle status data includes vehicle speed, acceleration, steering wheel angle, and braking status, while environmental data is acquired through sensors such as LiDAR, cameras, millimeter-wave radar, and ultrasonic radar, covering information such as road boundaries, obstacle locations, the behavior of vehicles in front and behind, and traffic sign recognition. Furthermore, it can receive location information and driving intentions from nearby vehicles via a V2X communication module, achieving comprehensive perception of the surrounding traffic scene and providing accurate and real-time data support for subsequent model inference and collaborative decision-making.

[0070] After completing real-time data acquisition, the intelligent vehicle nodes normalize, encode, or perform tensor transformations on the vehicle status data and environmental perception data according to the input format required by the cooperative driving model. The processed data is then fed into the updated cooperative driving model. This model consists of an environmental perception sub-model, a decision-making and planning sub-model, and a communication and coordination sub-model. These sub-models collaboratively process the input data, extract key features, identify the current traffic situation, and predict multi-vehicle interaction intentions and potential risks. The model can run in real-time on a local computing platform (such as an onboard GPU or a dedicated AI chip), ensuring low latency and high reliability in decision response.

[0071] Based on the updated collaborative driving model's decision-making results, combined with the current driving environment and vehicle status, intelligent vehicle nodes execute specific collaborative operations, such as collaborative lane changing or collaborative deceleration. For collaborative lane changing, the model outputs the timing of the lane change, the target lane, and the required action parameters. The vehicle control system adjusts the steering wheel angle and vehicle speed accordingly to achieve a safe lane change. For collaborative deceleration, the model judges the traffic flow or the status of the vehicle ahead, outputs the deceleration intention and deceleration rate, and the control system then executes braking commands and notifies following vehicles through V2X communication to cooperate in deceleration. This achieves coordinated dynamic behavior among multiple vehicles, improving overall traffic safety and smoothness.

[0072] Furthermore, as one embodiment of the present invention, refer to Figure 9 The vehicle cooperative driving method further includes: S7. Evaluate the effectiveness of cooperative driving and iteratively optimize the cooperative driving model based on the evaluation results.

[0073] Specifically, after completing steps S1 to S6, each intelligent vehicle node continuously records relevant operational data and results feedback after performing cooperative driving operations, such as lane change success rate, deceleration response time, traffic flow impact, ride comfort, and safety incident occurrence. These data are quantitatively analyzed using a pre-defined evaluation algorithm to generate a cooperative driving performance evaluation result, which is then compared with the model's expected behavior to identify performance deficiencies in specific scenarios. Based on the evaluation results, the intelligent vehicle node can selectively retrain its local sub-model or adjust hyperparameters, and then re-encrypt and upload the optimized model parameters to the blockchain network to participate in the next round of parameter aggregation and model updates. This achieves closed-loop self-iteration and continuous optimization of the model, thereby continuously improving the adaptability and robustness of the cooperative driving system in real-world complex road conditions.

[0074] Furthermore, the evaluation of the cooperative driving effect includes: calculating safety and efficiency indicators during the cooperative driving process; and analyzing the performance of the cooperative driving model in different driving scenarios. Specifically, during cooperative driving, operational data is collected in real time through onboard sensors and communication modules, and safety indicators such as collision risk coefficient, minimum vehicle distance, emergency braking frequency, and lane change success rate are calculated; simultaneously, efficiency indicators such as vehicle speed stability, traffic flow rate, average driving time, and fuel or energy consumption levels are evaluated. To analyze the model's performance in different driving scenarios, the driving process is categorized according to typical scenarios, such as highway lane merging, urban intersection driving, following other vehicles, and congestion queuing. Then, the model's decision accuracy, response time, and operational smoothness are statistically analyzed and compared for each scenario to identify the model's strengths and weaknesses in specific scenarios, thereby providing data support for subsequent model adjustments and optimizations.

[0075] Furthermore, as one embodiment of the present invention, refer to Figure 10 The vehicle cooperative driving method further includes: S8, based on a pre-designed blockchain-based incentive mechanism, encourages smart car nodes to actively participate in data sharing and collaborative driving.

[0076] Specifically, pre-designed incentive mechanism smart contracts are deployed in the blockchain network to clarify the reward rules for data sharing and model training participation. For example, when a smart car node completes effective actions such as uploading model parameters, sharing collaborative driving data, and responding to collaborative operations, the smart contract automatically records its contribution and distributes incentive resources, such as digital tokens, points, or usage priority, based on the contribution level. The contribution level can be comprehensively evaluated through dimensions such as upload frequency, data quality, and model improvement effect, and is recorded publicly and transparently on the blockchain to ensure the fairness and traceability of incentive distribution, thereby stimulating each node to actively participate in system collaboration and enhancing the continuous operation capability and performance optimization efficiency of the entire collaborative driving network.

[0077] This invention also discloses a vehicle cooperative driving system.

[0078] Reference Figure 11 A collaborative driving system for automobiles includes a blockchain network construction module 1, a federated learning framework deployment module 2, a model training module 3, a parameter uploading module 4, a model update module 5, and a decision execution module 6.

[0079] Blockchain network construction module 1 is used to build a blockchain network containing multiple smart car nodes. Specifically, blockchain network construction module 1 establishes secure communication connections between smart car nodes, roadside equipment nodes, and traffic management center nodes through encrypted communication protocols, forming a decentralized and distributed blockchain network. Simultaneously, blockchain network construction module 1 deploys a blockchain underlying architecture supporting consensus mechanisms (such as PoS or PBFT) within the network and embeds smart contract functionality to define collaborative driving rules, data sharing protocols, and incentive mechanisms. All nodes access the network through this module, possessing data recording, verification, and synchronization capabilities, providing fundamental support for the secure storage, trusted transmission, and transparent management of model parameters during collaborative driving.

[0080] The Federated Learning Framework Deployment Module 2 is used to deploy the Federated Learning Framework on each of the aforementioned intelligent vehicle nodes. Specifically, the Federated Learning Framework Deployment Module 2 deploys the Federated Learning Framework on the local computing platform of each intelligent vehicle node. The framework includes a task scheduling unit, a local training module, and a model communication interface. The task scheduling unit coordinates the training process according to instructions issued by the blockchain smart contract. The local training module uses local data collected by onboard sensors to perform customized training on the cooperative driving model. The model communication interface is used to encrypt the training results and interact with the blockchain network to achieve the uploading and synchronization of model parameters. Through the deployment of this module, each intelligent vehicle can independently participate in model training while ensuring data privacy, thereby supporting distributed collaborative modeling and iterative optimization.

[0081] Model training module 3 is used to train the cooperative driving model on each of the intelligent vehicle nodes based on local data. Specifically, model training module 3 receives the initial cooperative driving model issued by the federated learning framework and uses the pre-processed local data for sub-model level training. This includes inputting data into the environmental perception sub-model, decision planning sub-model, and communication and coordination sub-model respectively, calculating the loss function of each sub-task through multi-task learning, and constructing a joint loss function for weighted fusion. Subsequently, model training module 3 uses the gradient descent algorithm to optimize the fusion loss and achieves end-to-end parameter updates through backpropagation, while dynamically adjusting the sub-task weights to improve training balance. Model training module 3 also has a built-in verification mechanism to evaluate model performance based on the local validation set, ensuring that the training results reach the expected accuracy before proceeding to the upload stage, laying the foundation for subsequent parameter aggregation.

[0082] The parameter upload module 4 is used to upload the trained model parameters to the blockchain network. Specifically, after completing the local cooperative driving model training, the parameter upload module 4 encrypts the model parameters, using encryption algorithms such as AES or RSA to ensure transmission security, and combining differential privacy technology to enhance reverse engineering capabilities. Subsequently, the parameter upload module 4 packages the encrypted model parameters into standardized transaction data and uploads the parameters to the blockchain network through a secure connection established with the blockchain communication interface. During the upload process, the parameter upload module 4 also performs integrity verification of the model parameters and calls the smart contract to verify and record the legality of the upload behavior, thereby ensuring the credibility, traceability, and privacy protection of the uploaded data, providing reliable input for subsequent aggregation updates.

[0083] The model update module 5 is used to aggregate the model parameters and update the cooperative driving model through the blockchain network. Specifically, the model update module 5 automatically triggers the aggregation processing of encrypted model parameters uploaded by each intelligent vehicle node by calling a smart contract deployed in the blockchain network. The aggregation method can adopt federated learning algorithms such as FedAvg, which allocate weights according to the amount of data or contribution of each node and perform weighted fusion. The fused parameters are used to generate an updated global cooperative driving model, and version identification, hash verification and storage are performed through the blockchain network. Subsequently, the model update module 5 synchronously distributes the updated model to all intelligent vehicle nodes, and ensures the consistency and integrity of the model through on-chain broadcast and verification mechanisms, thereby realizing iterative optimization and unified upgrade of the entire network model.

[0084] The decision execution module 6 is used to execute cooperative driving decisions based on the updated cooperative driving model. Specifically, the decision execution module 6 receives and loads the updated cooperative driving model from the blockchain network in real time, and inputs vehicle status data and environmental perception data collected from onboard sensors and V2X communication modules into the model for inference analysis. The decision results output by the model include lane change timing, target trajectory, deceleration commands, etc. The decision execution module 6 automatically controls the vehicle's steering, acceleration and deceleration, and cooperative behavior with surrounding vehicles based on the output results, realizing operations such as cooperative lane changing, cooperative deceleration, and automatic platooning. At the same time, the decision execution module 6 also monitors the safety and response efficiency during the execution process and feeds back the relevant execution data to the system for subsequent model evaluation and optimization closed loop.

[0085] The vehicle cooperative driving system, composed of a blockchain network construction module 1, a federated learning framework deployment module 2, a model training module 3, a parameter upload module 4, a model update module 5, and a decision execution module 6, integrates the decentralized, traceable, and tamper-proof characteristics of blockchain with the privacy protection and distributed modeling capabilities of federated learning. This enables distributed model training, secure parameter transmission, reliable model updates, and intelligent decision execution in multi-vehicle cooperative driving. It not only effectively protects the privacy of vehicle owner data and establishes a trust mechanism between vehicles, but also improves the training efficiency, update speed, and adaptability of the cooperative driving model, thereby significantly enhancing the system's safety, real-time performance, and scalability in complex and dynamic traffic environments.

[0086] This invention also discloses a readable storage medium.

[0087] A computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the vehicle cooperative driving method described in any of the above embodiments. The computer-readable storage medium may include any entity or device capable of carrying a computer program, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), and a software distribution medium, etc. The computer program includes computer program code. The computer program code may be in the form of source code, object code, an executable file, or some intermediate form, etc. The computer-readable storage medium may include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), and a software distribution medium, etc.

[0088] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0089] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a system including a processing module or other system that can fetch and execute instructions from, an instruction execution system, apparatus or device).

[0090] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for cooperative driving of automobiles, characterized in that, include: Construct a blockchain network containing multiple smart car nodes; Deploy a federated learning framework on each of the aforementioned intelligent vehicle nodes; Each of the aforementioned intelligent vehicle nodes trains a cooperative driving model based on local data; The trained model parameters are uploaded to the blockchain network; The model parameters are aggregated and the cooperative driving model is updated through the blockchain network; as well as Cooperative driving decisions are made based on the updated cooperative driving model.

2. The vehicle cooperative driving method according to claim 1, characterized in that, The construction of the blockchain network containing multiple smart car nodes includes: Establish communication connections between intelligent vehicle nodes, roadside equipment nodes, and traffic management center nodes; and Deploy smart contracts on the blockchain network, which define cooperative driving rules, data sharing protocols, and reward mechanisms.

3. The vehicle cooperative driving method according to claim 1, characterized in that, Before each intelligent vehicle node trains the cooperative driving model based on local data, the following steps are also included: Vehicle status data and surrounding environment data are collected through onboard sensors; and The collected data is preprocessed, including data cleaning and feature extraction.

4. The vehicle cooperative driving method according to claim 1, characterized in that, Each intelligent vehicle node trains a cooperative driving model based on local data, including: Download the initial cooperative driving model from the blockchain network; The initial cooperative driving model is trained using preprocessed local data; and The performance of the cooperative driving model was evaluated using a validation set.

5. The vehicle cooperative driving method according to claim 4, characterized in that, The initial cooperative driving model includes an environmental perception sub-model, a decision-making and planning sub-model, and a communication and coordination sub-model. Training the initial cooperative driving model using preprocessed local data includes: The preprocessed local data is then input into the environmental perception sub-model, the decision-making and planning sub-model, and the communication and collaboration sub-model, respectively. A multi-task learning approach is adopted to simultaneously optimize the loss functions of the three sub-tasks: environmental perception, decision planning, and communication and coordination. Based on a preset joint loss function, the outputs of the three sub-models are weighted and fused. The joint loss function is optimized using the gradient descent algorithm, while the parameters of the three sub-models are updated simultaneously. During training, the weights between the sub-models are dynamically adjusted to balance the importance of different tasks; The backpropagation algorithm is used to pass the fused gradient information to each sub-model, achieving end-to-end joint optimization.

6. The vehicle cooperative driving method according to claim 1, characterized in that, Uploading the trained model parameters to the blockchain network includes: Encrypt the trained model parameters; and The encrypted model parameters are uploaded to the blockchain network.

7. The vehicle cooperative driving method according to claim 1, characterized in that, The step of aggregating the model parameters and updating the cooperative driving model through the blockchain network includes: Use smart contracts to automatically aggregate model parameters uploaded by each smart car node; Update the global cooperative driving model based on the aggregated model parameters; and The updated global collaborative driving model is distributed to each intelligent vehicle node.

8. The vehicle cooperative driving method according to claim 1, characterized in that, The process of making cooperative driving decisions based on the updated cooperative driving model includes: Collect real-time vehicle status data and surrounding environment data; Input the real-time data into the updated cooperative driving model; and Based on the output of the cooperative driving model, perform cooperative lane changing or cooperative deceleration operations.

9. The vehicle cooperative driving method according to any one of claims 1 to 8, characterized in that, Also includes: Evaluate the effectiveness of the cooperative driving system and iteratively optimize the cooperative driving model based on the evaluation results; The evaluation of the cooperative driving effect includes: Calculate safety and efficiency indicators during cooperative driving; and The performance of the cooperative driving model in different driving scenarios was analyzed.

10. The vehicle cooperative driving method according to any one of claims 1 to 8, characterized in that, Also includes: Based on a pre-designed blockchain-based incentive mechanism, smart car nodes are encouraged to actively participate in data sharing and collaborative driving.

11. A vehicle cooperative driving system, characterized in that, The system includes: A blockchain network building module is used to build a blockchain network containing multiple smart car nodes; A federated learning framework deployment module is used to deploy the federated learning framework on each of the intelligent vehicle nodes; The model training module is used to train a cooperative driving model based on local data on each of the intelligent vehicle nodes; The parameter upload module is used to upload the trained model parameters to the blockchain network; A model update module is used to aggregate the model parameters and update the cooperative driving model through the blockchain network; and The decision execution module is used to execute cooperative driving decisions based on the updated cooperative driving model.

12. A readable storage medium, characterized in that, The readable storage medium stores computer instructions that, when executed by a processor, implement the vehicle cooperative driving method as described in any one of claims 1-10.

Images