Road adaptation method, device and medium applicable to icv driving model

Abstract

This invention discloses a road adaptation method, device, and medium suitable for ICV driving models. The method includes: acquiring a pre-training sample dataset and preprocessing it; training an intelligent driving model using the pre-training sample dataset; determining whether the vehicle is on an unfamiliar road segment; if the vehicle is on an unfamiliar road segment, acquiring vehicle trajectory data from the unfamiliar road as an adaptive sample set and preprocessing it; fine-tuning the intelligent driving model using an adaptive fine-tuning strategy based on the preprocessed adaptive sample set to obtain a new intelligent driving model; and adding the obtained adaptive sample set to the pre-training sample dataset to update it. This invention uses an adaptive fine-tuning strategy to fine-tune the model parameters, adapting it to the road environment of unfamiliar road segments, improving the safety and adaptability of ICVs on unfamiliar road segments, and can be widely applied in intelligent transportation, intelligent driving models, and other fields.

Description

Technical Field

[0001] This invention relates to the fields of intelligent transportation and intelligent driving models, and in particular to a road adaptation method, device and medium suitable for ICV driving models. Background Technology

[0002] With the rapid development of intelligent transportation systems, intelligent and connected vehicles (ICVs), as an important component, are gradually demonstrating their enormous potential in improving traffic efficiency and enhancing driving safety. Therefore, research on intelligent driving models for ICVs has attracted the attention of relevant researchers and has become a current hot research topic.

[0003] To achieve widespread application and industrialization of ICVs, it is essential to ensure their safety in various road environments, meaning that intelligent driving models must be adaptable to different road conditions. To address this challenge, current solutions for data-driven intelligent driving models primarily involve optimizing the network structure and expanding the training dataset. Optimizing the network structure often relies on the complexity of the existing pre-trained intelligent driving model network and requires significant computational resources, resulting in relatively poor improvements in training efficiency and adaptability, failing to adequately meet the adaptability requirements of ICVs. Dataset expansion methods mainly include joint updates and incremental updates. Joint updates combine historical and new data to update the model, thoroughly resolving the model adaptability issue across different scenarios; however, as the dataset increases, the computational and storage costs become extremely high, limiting its practical application capabilities. Incremental updates update the model with small-scale new data, offering higher training efficiency. Summary of the Invention

[0004] In order to at least partially solve one of the technical problems existing in the prior art, the purpose of this invention is to provide a road adaptation method, device and medium suitable for ICV driving models.

[0005] The first technical solution adopted in this invention is:

[0006] A road adaptation method suitable for ICV driving models includes the following steps:

[0007] Obtain a large number of pre-trained sample datasets and preprocess the obtained pre-trained sample datasets;

[0008] The intelligent driving model was trained using a pre-processed pre-training sample dataset.

[0009] Determine if the vehicle is on an unfamiliar road; if the vehicle is on an unfamiliar road, obtain a small amount of vehicle trajectory data from unfamiliar roads as an adaptive sample set, and preprocess the obtained adaptive sample set.

[0010] Based on the preprocessed adaptive sample set, an adaptive fine-tuning strategy is used to fine-tune the intelligent driving model to obtain a new intelligent driving model.

[0011] The obtained adaptive sample set is added to the pre-trained sample dataset to update the pre-trained sample dataset.

[0012] Further, the step of obtaining the pre-trained sample dataset and preprocessing the obtained pre-trained sample dataset includes:

[0013] Obtain a predetermined number of vehicle trajectory data as a pre-training sample dataset;

[0014] A threshold method is used to process outlier data in the pre-training sample dataset;

[0015] The dataset after anomaly handling is expanded using a preset data expansion method.

[0016] Furthermore, the intelligent driving model is a data-driven intelligent driving model;

[0017] The process of training the intelligent driving model using a pre-processed pre-training sample dataset includes:

[0018] The intelligent driving model is trained using a pre-processed pre-trained sample dataset, and the training effect is measured by a preset evaluation standard to select the best-performing model as the benchmark intelligent driving model.

[0019] Furthermore, determining whether the vehicle is in an unfamiliar road section includes:

[0020] The vehicle's location data is collected according to a preset cycle, and the location data is used to determine whether the vehicle is currently on an unfamiliar road segment.

[0021] Furthermore, the adaptive sample set data can be acquired in two ways: vehicle-to-infrastructure communication and vehicle-to-vehicle communication.

[0022] Furthermore, the step of fine-tuning the intelligent driving model using an adaptive fine-tuning strategy to obtain a new intelligent driving model includes:

[0023] An adaptive fine-tuning strategy is employed to fine-tune the parameters of the intelligent driving model. The weight update method uses root mean square propagation (RMSprop), introducing a variable called adaptive performance α(n). This variable represents the percentage improvement in performance of the baseline intelligent driving model after n (n>0) rounds of fine-tuning, expressed by the formula:

[0024]

[0025] In the formula, MSE0 represents the transfer performance of the model, and MSE n This represents the model performance after n iterations;

[0026] The performance improvement of the intelligent driving model after parameter fine-tuning is measured by mean square error (MSE) and adaptive performance α(n). When the difference in adaptive performance between two adjacent iterations is less than a preset threshold, the model is determined to have reached the best performance, the model parameter adjustment is terminated, and the best-performing intelligent driving model is output.

[0027] Furthermore, for each training sample, the root mean square propagation method is used to update the weights, as shown in the following expression:

[0028]

[0029] In the formula, Δθ is the parameter change, η is the fine-tuning learning rate; ε μ It is to prevent numbers with a denominator of 0; g μ Let be the gradient of the μth training iteration. β is the cumulative squared gradient of the first μ-1 training iterations, and β is the decay factor.

[0030] The second technical solution adopted in this invention is:

[0031] A road adaptation device for ICV driving models, comprising:

[0032] The dataset building module is used to acquire a large number of pre-trained sample datasets and preprocess the acquired pre-trained sample datasets.

[0033] The model pre-training module is used to train the intelligent driving model using a pre-processed pre-training sample dataset.

[0034] The adaptive data acquisition module is used to determine whether the vehicle is on an unfamiliar road segment; if the vehicle is on an unfamiliar road segment, it acquires a small amount of vehicle trajectory data on the unfamiliar road as an adaptive sample set, and preprocesses the obtained adaptive sample set.

[0035] The model fine-tuning module is used to fine-tune the intelligent driving model based on the preprocessed adaptive sample set and an adaptive fine-tuning strategy to obtain a new intelligent driving model.

[0036] The dataset update module is used to add the obtained adaptive sample set to the pre-training sample dataset to update the pre-training sample dataset.

[0037] The third technical solution adopted in this invention is:

[0038] A road adaptation device for ICV driving models, comprising:

[0039] At least one processor;

[0040] At least one memory for storing at least one program;

[0041] When the at least one program is executed by the at least one processor, the at least one processor implements the method described above.

[0042] The fourth technical solution adopted in this invention is:

[0043] A computer-readable storage medium storing a processor-executable program, which, when executed by a processor, performs the method described above.

[0044] The beneficial effects of this invention are as follows: Based on a "pre-training-fine-tuning" approach, the pre-trained intelligent driving model, through learning features and patterns from large-scale data, can provide rich prior knowledge for the model. Using the best-performing intelligent driving model from the pre-training process as the benchmark, an adaptive sample set—namely, vehicle trajectory data on unfamiliar road sections—is obtained through both vehicle-to-infrastructure (V2I) and vehicle-to-vehicle (V2V) communication. An adaptive fine-tuning strategy is then employed to fine-tune the parameters of the intelligent driving model. Through this fine-tuning, the model can adjust its parameters specifically according to road conditions and environmental characteristics, thereby rapidly adapting and learning on unfamiliar roads, improving the model's generalization ability and adaptability to new scenarios, and enhancing the practical applicability and reliability of ICV (Intelligent Driving Vehicle). Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following description is provided with accompanying drawings of the relevant technical solutions in the embodiments of the present invention or the prior art. It should be understood that the accompanying drawings described below are only for the purpose of clearly illustrating some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart illustrating a road adaptation method applicable to an ICV driving model in an embodiment of the present invention;

[0047] Figure 2 This is a schematic diagram illustrating the steps of the adaptive fine-tuning strategy in an embodiment of the present invention;

[0048] Figure 3 This is a flowchart illustrating the steps of a road adaptation method applicable to an ICV driving model in an embodiment of the present invention. Detailed Implementation

[0049] The embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0050] In the description of this invention, it should be understood that the orientation descriptions, such as up, down, front, back, left, right, etc., are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0051] In the description of this invention, "several" means one or more, "multiple" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. If "first" or "second" is used, it is only for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features. Furthermore, "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0052] In the description of this invention, unless otherwise explicitly defined, terms such as "set up," "install," and "connect" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0053] The pre-training-fine-tuning strategy, as an effective machine learning method, first pre-trains the model on a large-scale dataset, enabling it to possess basic recognition and processing capabilities. Subsequently, the model parameters are fine-tuned for specific tasks or scenarios to further improve its performance. This strategy has achieved significant results in multiple fields, including natural language processing and image recognition. Furthermore, current traffic detection and communication technologies make it possible for ICVs to acquire small-scale vehicle trajectory data on unfamiliar roads in real time. Therefore, an incremental update approach is adopted to apply the pre-training-fine-tuning strategy to address the adaptability of ICVs to unfamiliar road environments. In the pre-training phase, the ICV utilizes a large amount of labeled road data to learn basic road environment understanding capabilities. In the fine-tuning phase, the ICV acquires a small-scale dataset and fine-tunes the intelligent driving model parameters based on the environmental characteristics of specific unfamiliar road sections, allowing it to better adapt to the actual conditions of unfamiliar road sections. Thus, this method demonstrates strong feasibility and high practical application value.

[0054] like Figure 3 As shown, this embodiment provides a road adaptation method suitable for ICV driving models, including the following steps:

[0055] S1. Obtain a large number of pre-trained sample datasets and preprocess the obtained pre-trained sample datasets.

[0056] Acquiring and preprocessing the pre-training sample dataset requires collecting a large amount of vehicle trajectory data (more than 1,000,000 vehicle trajectory data points) and performing preprocessing operations on the collected trajectory data. The main focus is on handling outlier data in the dataset, using a threshold method. Thresholds are set based on actual conditions and professional knowledge; for example, accelerations greater than 5 m / s² are considered outlier. 2 and less than -8m / s 2 Vehicle trajectories are considered abnormal data. Abnormal data in vehicle trajectory data are filtered and deleted, unreasonable data is corrected, and appropriate data expansion methods such as data augmentation are adopted to increase the amount and richness of data.

[0057] In addition, the data format needs to be adjusted according to different intelligent driving models to make it conform to the model data input format.

[0058] S2. Train the intelligent driving model using the pre-processed pre-training sample dataset.

[0059] Specifically, the intelligent driving model refers to a data-driven intelligent driving model, which is an autonomous driving system designed and optimized by utilizing a large amount of collected data and combining machine learning and artificial intelligence technologies, such as the LSTM-ID model and the FNN-ID model.

[0060] The intelligent driving model is trained using the pre-trained sample dataset from step S1, and an appropriate evaluation criterion is selected to measure the training effect. The model with the best performance is selected as the benchmark intelligent driving model.

[0061] S3. Determine if the vehicle is on an unfamiliar road segment; if the vehicle is on an unfamiliar road segment, obtain a small amount of vehicle trajectory data from unfamiliar roads as an adaptive sample set, and preprocess the obtained adaptive sample set.

[0062] The vehicle's location data is collected at regular intervals (e.g., every 10 seconds) to determine whether the ICV is currently on an unfamiliar road segment.

[0063] As an optional implementation method, there are two ways to obtain the adaptive sample set: vehicle-to-infrastructure communication and vehicle-to-vehicle communication.

[0064] Vehicle-to-everything (V2X) communication primarily involves communication between the Intelligent Vehicle Vehicle (ICV) and intelligent roadside equipment (ILVs) installed along the road. This communication utilizes vehicle-to-everything (V2X) or short-range communication (SMR) technologies to acquire, process, and store data collected by the ICVs. Related research indicates that in the 5G era, upon receiving a data request signal from an ICV, the ICV will transmit 50,000-250,000 data samples to the ICV within 0.1-0.5 seconds. These samples can meet the data requirements for fine-tuning intelligent driving models.

[0065] Vehicle-to-vehicle (V2V) communication primarily involves direct communication with other ICVs on unfamiliar roads via wireless network technology, obtaining data collected by these ICVs using onboard detection equipment such as cameras and radar. The number of vehicle trajectories collected in this way depends on the number of other ICVs within the ICV's communication range; typically, communication between one ICV and another will yield 1,000 to 10,000 sample data points.

[0066] The processing method for vehicle trajectory data obtained through these two methods is the same as that for the pre-training sample dataset. The processed data format is also consistent with that described in step S1, and the data is randomly divided into training set and test set.

[0067] S4. Based on the preprocessed adaptive sample set, the intelligent driving model is fine-tuned using an adaptive fine-tuning strategy to obtain a new intelligent driving model.

[0068] An adaptive fine-tuning strategy is employed to fine-tune the model parameters. The inputs are an adaptive sample set and a baseline intelligent driving model. The weight update method uses the root mean square propagation (RMSprop) method. A variable called adaptive performance α(n) is introduced, which represents the percentage improvement in performance of the baseline intelligent driving model after n (n>0) rounds of fine-tuning. This variable is expressed by the formula:

[0069]

[0070] Where MSE0 represents the model's transfer performance, MSE n This represents the model performance after n iterations.

[0071] The performance improvement of the intelligent driving model after parameter fine-tuning is measured by mean squared error (MSE) and adaptive performance α(n). When the difference in adaptive performance between two adjacent iterations is less than a threshold, such as 0.1%, the model is considered to have reached the best performance. Based on this, a termination rule is set. After the termination rule is met, the model parameter adjustment ends and the intelligent driving model with the best performance is output.

[0072] The new benchmark intelligent driving model is the intelligent driving model with the best performance output by the adaptive fine-tuning strategy.

[0073] S5. Add the obtained adaptive sample set to the pre-training sample dataset to update the pre-training sample dataset.

[0074] In summary, the method in this embodiment utilizes a pre-trained sample dataset to train an intelligent driving model as a baseline intelligent driving model for ICVs. When the ICV enters an unfamiliar road segment, it acquires a small number of adaptive samples through vehicle-to-infrastructure communication or vehicle-to-vehicle communication, and employs an adaptive fine-tuning strategy to fine-tune the model parameters to adapt it to the road environment of the unfamiliar road segment, thereby improving the safety and adaptability of the ICV in unfamiliar road segments.

[0075] The above method will be explained in detail below with reference to the accompanying drawings and specific embodiments.

[0076] This embodiment provides a road adaptation method suitable for ICV driving models, which is applicable to improving the adaptability of ICVs to unfamiliar roads and achieving safe driving on unfamiliar roads. The step framework diagram of this method is attached. Figure 1 As shown, taking the example of an intelligent driving model using the LSTM-ID model, a pre-training sample dataset using the US NGSIM dataset, and an adaptive sample set acquisition method using vehicle-to-infrastructure communication, further explanation will be provided.

[0077] The pre-training sample dataset comes from the NGSIM dataset I-101 road segment. Data preprocessing mainly addresses two issues in the dataset: unreasonable vehicle motion states and positional relationships between vehicles. A four-step correction method is used to correct the vehicle trajectory data. The first step uses a threshold method to filter vehicles with acceleration greater than 5 m / s². 2 and less than -8m / s 2The first step involves treating vehicle trajectories as abnormal data, deleting abnormal data from the vehicle trajectory data, and replacing the deleted data with trajectory data generated by cubic spline curve interpolation. The second step involves smoothing and denoising the data using a first-order Butterworth filter with a cutoff frequency of 1Hz. The third step involves reconstructing the local vehicle data trajectory using a cubic polynomial trajectory planning method, mainly modifying data on unreasonable positional relationships between vehicles. The fourth step involves using the same first-order Butterworth filter as the second step to eliminate the noise introduced in the third step.

[0078] After correcting the data issues, the data format was changed to the format of the LSTM-ID model input. The modified data format includes target vehicle state features (including the target vehicle's horizontal and vertical coordinates and speed), the target vehicle's relative state with surrounding vehicles (including relative horizontal and vertical coordinates and relative speed), the target vehicle's spatiotemporal distance to its short-term destination (including lateral, longitudinal, and temporal relative distances), target vehicle size information (including vehicle width and length), surrounding vehicle size information (including vehicle width and length), and basic road information (including the horizontal coordinate of the centerline of the lane where the target vehicle is located and the horizontal coordinates of the centerlines of the lanes adjacent to the target vehicle on the left and right).

[0079] The final pre-training sample dataset with 2,000,000 samples was obtained and randomly divided into training set (50%), validation set (20%), and test set (30%).

[0080] The LSTM-ID model was chosen for the intelligent driving model. The LSTM-ID model combines the ability of Long Short-Term Memory networks to capture long-term dependencies in time series data with the ability of intent detection to deeply understand driver intent, providing a powerful tool for data-driven intelligent driving models.

[0081] The learning rate for the pre-training of the intelligent driving model is 0.001, and the performance evaluation metric is the mean squared error (MSE), calculated as follows:

[0082]

[0083] Where MSE represents the mean square error, the unit is m. 2 ; x represents the predicted value of the k-th trajectory sample in the horizontal and vertical directions; k y k represents the actual values ​​of the k-th trajectory sample in the horizontal and vertical directions; K is the number of samples used for testing.

[0084] After pre-training, the optimal MSE of the LSTM-ID model was 0.29m. 2 This model serves as the benchmark for intelligent driving models.

[0085] Using the NGSIM dataset I-80 and the High-D dataset, we simulated vehicle trajectory data obtained by ICV communicating with intelligent roadside devices on unfamiliar road sections. The trajectory data was processed in the same way as the pre-training sample dataset, resulting in a training dataset of 250,000 data points and a test set of 400,000 data points.

[0086] The baseline intelligent driving model was directly transferred to the test set of the adaptive sample set for transfer testing, and the MSE was 0.55m. 2 and 0.24m 2 .

[0087] An adaptive fine-tuning strategy is used to update the parameters of the baseline intelligent driving model. The step-by-step framework diagram of the adaptive fine-tuning strategy is attached. Figure 2 As shown. The freezing strategy is set to fine-tune all parameters, the fine-tuning learning rate is 0.0001, the maximum number of iterations is 10, and the evaluation criteria are mean squared error (MSE) and adaptive performance α(n). Adaptive performance is defined as the percentage increase in adaptability of the trained driving model after n (n>0) rounds of fine-tuning, expressed by the formula:

[0088]

[0089] Where MSE0 represents the model's transfer performance, MSE n This represents the model performance after n iterations.

[0090] The fine-tuning termination rule is set to α(n)-α(n-1)<0.1%, which means that the driving model is considered to have adapted to the new road environment at this time, and the number of training iterations is n-1; if the fine-tuning termination rule is not met when n=10 during the fine-tuning process, the number of training iterations is the preset maximum number of training iterations.

[0091] For each training sample, the RMSprop weight update method is used. The update principle is as follows:

[0092]

[0093] Where Δθ is the parameter change, η is the fine-tuning learning rate; ε μ To prevent the denominator from being 0, we take 10. -6 g μ Let be the gradient of the μth training iteration. β is the cumulative squared gradient of the first μ-1 training iterations, and β is the decay factor.

[0094] After each iteration, the model is validated using a test set to obtain the mean squared error (MSE) and adaptive performance α(n), and it is determined whether the termination rule is met. If the termination rule is not met, the next iteration of training continues; if the termination rule is met, the best model is output as the new benchmark intelligent driving model, and the adaptive sample set is merged into the pre-training sample dataset.

[0095] To simulate specific road environments, adaptive sample sets with training sample sizes of 50,000, 100,000, 150,000, 200,000, and 250,000 were used to fine-tune the baseline intelligent driving model, obtaining the optimal MSE and α(n). The results are shown in Table 1.

[0096] Table 1. Model performance and adaptive performance in V2I environment

[0097]

[0098] As shown in Table 1, the adaptability of the intelligent driving model was significantly improved after parameter fine-tuning, reaching a maximum of 81%. In addition, the adaptive performance of the intelligent driving model improved as the number of samples in the adaptive sample set increased. Finally, the adaptability of the intelligent driving model improved even more on the German High-D dataset, which is more different from the pre-training sample dataset.

[0099] In summary, experiments have shown that this method significantly improves the adaptability of ICVs. The greater the difference in road environment, the greater the improvement in adaptive performance, which can improve the safety of ICVs driving on unfamiliar roads and has high practical application value.

[0100] This embodiment also provides a road adaptation device suitable for ICV driving models, including:

[0101] The dataset building module is used to acquire a large number of pre-trained sample datasets and preprocess the acquired pre-trained sample datasets.

[0102] The model pre-training module is used to train the intelligent driving model using a pre-processed pre-training sample dataset.

[0103] The adaptive data acquisition module is used to determine whether the vehicle is on an unfamiliar road segment; if the vehicle is on an unfamiliar road segment, it acquires a small amount of vehicle trajectory data on the unfamiliar road as an adaptive sample set, and preprocesses the obtained adaptive sample set.

[0104] The model fine-tuning module is used to fine-tune the intelligent driving model based on the preprocessed adaptive sample set and an adaptive fine-tuning strategy to obtain a new intelligent driving model.

[0105] The dataset update module is used to add the obtained adaptive sample set to the pre-training sample dataset to update the pre-training sample dataset.

[0106] This embodiment provides a road adaptation device suitable for ICV driving models, which can execute a road adaptation method suitable for ICV driving models provided in the method embodiment of the present invention. It can execute any combination of implementation steps of the method embodiment and has the corresponding functions and beneficial effects of the method.

[0107] This embodiment also provides a road adaptation device suitable for ICV driving models, including:

[0108] At least one processor;

[0109] At least one memory for storing at least one program;

[0110] When the at least one program is executed by the at least one processor, the at least one processor implements Figure 3 The method shown.

[0111] This embodiment provides a road adaptation device suitable for ICV driving models, which can execute a road adaptation method suitable for ICV driving models provided in the method embodiment of the present invention. It can execute any combination of implementation steps of the method embodiment and has the corresponding functions and beneficial effects of the method.

[0112] This application also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform... Figure 3 The method shown.

[0113] This embodiment also provides a storage medium storing instructions or programs that can execute the road adaptation method for ICV driving models provided in the method embodiment of the present invention. When the instructions or programs are run, any combination of implementation steps of the method embodiment can be executed, and the method has the corresponding functions and beneficial effects.

[0114] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is altered and sub-operations described as part of a larger operation are executed independently.

[0115] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the described functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0116] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0117] 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 processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0118] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0119] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0120] In the foregoing description of this specification, references to terms such as "one embodiment," "another embodiment," or "some embodiments" indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0121] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0122] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A road adaptation method suitable for ICV driving models, characterized in that, Includes the following steps: Obtain the pre-training sample dataset and preprocess the obtained pre-training sample dataset; The intelligent driving model was trained using a pre-processed pre-training sample dataset. Determine if the vehicle is in an unfamiliar area; If the vehicle is on an unfamiliar road, obtain the vehicle trajectory data of the unfamiliar road as an adaptive sample set, and preprocess the obtained adaptive sample set. Based on the preprocessed adaptive sample set, an adaptive fine-tuning strategy is used to fine-tune the intelligent driving model to obtain a new intelligent driving model. The obtained adaptive sample set is added to the pre-training sample dataset to update the pre-training sample dataset; The process of fine-tuning the intelligent driving model using an adaptive fine-tuning strategy to obtain a new intelligent driving model includes: An adaptive fine-tuning strategy is used to fine-tune the parameters of the intelligent driving model; The weight update method employs root mean square propagation, and introduces a feature called adaptive performance. The variable represents the baseline intelligent driving model after... The percentage performance improvement after wheel fine-tuning is expressed by the formula: In the formula, Indicates the model's transfer performance. Indicates the process Model performance after rounds of iterations; Mean Square Error (MSE) and Adaptive Performance The performance improvement of the intelligent driving model after parameter fine-tuning is measured. When the difference in adaptive performance between two adjacent iterations is less than a preset threshold, the model is determined to have reached the best performance, the model parameter adjustment is terminated, and the best-performing intelligent driving model is output. For each training sample, the weights are updated using the root mean square propagation method, as shown in the following expression: In the formula, For parameter changes, To fine-tune the learning rate; It is to prevent numbers with a denominator of 0; For the first The gradient of the training iterations. For the front The cumulative squared gradient of training iterations. This is the attenuation factor.

2. The road adaptation method for ICV driving models according to claim 1, characterized in that, The process of obtaining the pre-training sample dataset and preprocessing the obtained pre-training sample dataset includes: Obtain a predetermined number of vehicle trajectory data as a pre-training sample dataset; A threshold method is used to process outlier data in the pre-training sample dataset; The dataset after anomaly handling is expanded using a preset data expansion method.

3. The road adaptation method for ICV driving models according to claim 1, characterized in that, The intelligent driving model is a data-driven intelligent driving model; The process of training the intelligent driving model using a pre-processed pre-training sample dataset includes: The intelligent driving model is trained using a pre-processed pre-trained sample dataset, and the training effect is measured by a preset evaluation standard to select the best-performing model as the benchmark intelligent driving model.

4. The road adaptation method for ICV driving models according to claim 1, characterized in that, The determination of whether a vehicle is on an unfamiliar road section includes: The vehicle's location data is collected according to a preset cycle, and the location data is used to determine whether the vehicle is currently on an unfamiliar road segment.

5. The road adaptation method for ICV driving models according to claim 1, characterized in that, There are two ways to acquire the adaptive sample set data: vehicle-to-infrastructure communication and vehicle-to-vehicle communication.

6. A road adaptation device suitable for ICV driving models, for implementing the method according to any one of claims 1-5, characterized in that, include: The dataset building module is used to obtain the pre-training sample dataset and preprocess the obtained pre-training sample dataset; The model pre-training module is used to train the intelligent driving model using a pre-processed pre-training sample dataset. An adaptive data acquisition module is used to determine whether the vehicle is in an unfamiliar road section; If the vehicle is on an unfamiliar road, obtain the vehicle trajectory data of the unfamiliar road as an adaptive sample set, and preprocess the obtained adaptive sample set. The model fine-tuning module is used to fine-tune the intelligent driving model based on the preprocessed adaptive sample set and an adaptive fine-tuning strategy to obtain a new intelligent driving model. The dataset update module is used to add the obtained adaptive sample set to the pre-training sample dataset to update the pre-training sample dataset.

7. A road adaptive device suitable for ICV driving models, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method of any one of claims 1-5.

8. A computer-readable storage medium storing a processor-executable program, characterized in that, The processor-executable program, when executed by the processor, is used to perform the method as described in any one of claims 1-5.

Images