Brake master cylinder pressure prediction method based on MAML-AUKF model
Through the internal and external loop training strategy of the MAML-AUKF model, the problems of reduced robustness and generalization ability caused by frequent changes in sEMG signal noise were solved, and the accuracy and stability of brake master cylinder pressure prediction in a human-machine co-driving environment were improved.
Patent Information
- Application Number
- CN202510765387.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-10
- Publication Date
- 2025-09-05
AI Technical Summary
In the existing technology, surface electromyography (sEMG) noise changes frequently, which leads to the reduction of robustness and generalization ability of the unscented Kalman filter algorithm in brake master cylinder pressure prediction.
The MAML-AUKF model is adopted, and the AUKF model parameters are updated through the inner and outer loop training strategy to adapt it to the frequently changing noise conditions of sEMG signals and improve its robustness and generalization ability.
The AUKF model parameters are quickly updated with a small number of samples, which improves the accuracy and stability of the brake master cylinder pressure prediction and reduces abnormal fluctuations caused by noise interference.
Smart Images
Figure CN120596802A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of human-machine co-driving technology, and in particular to a brake master cylinder pressure prediction method based on a MAML-AUKF model. Background Art
[0002] Surface electromyography (sEMG) is a weak physiological signal that can effectively characterize muscle activity. It can reflect the state and function of nerves and muscles. However, its acquisition process is easily interfered by various noises. In the "human-machine co-driving" mode, the surface electromyography (sEMG) signal in the human-machine co-driving interaction information is often used to characterize the driver's continuous braking intention and predict the master cylinder pressure through unscented Kalman filtering (UKF). However, due to the frequent changes in sEMG signal noise, the use of ordinary unscented Kalman filtering algorithm will greatly reduce the robustness and generalization ability of the model in predicting the master cylinder pressure filter. Summary of the Invention
[0003] The purpose of this paper is to quickly update the design parameters of an adaptive unscented Kalman filter (AUKF) with a small number of samples to adapt it to the frequently changing noise conditions of sEMG signals, thereby improving the robustness and generalization ability of the AUKF model in predicting master cylinder pressure filtering.
[0004] Specifically, the present invention provides a brake master cylinder pressure prediction method based on the MAML-AUKF model, comprising: A MAML-AUKF model was constructed and the AUKF model was trained using the MAML training strategy to adapt it to the noisy conditions of frequently changing sEMG signals. The brake master cylinder pressure prediction task is divided into multiple meta-learning tasks according to the driver's braking deceleration peak value; The driver's leg sEMG signals and brake master cylinder pressure are acquired when the driver presses the brake pedal, and a dataset is established. The dataset is then divided into support sets and query sets for different tasks according to a preset ratio. Based on the support sets of different tasks, the parameters of the AUKF model are updated through the inner loop training of the MAML-AUKF model; based on the query sets of different tasks, the parameters of the AUKF model are updated through the outer loop training of the MAML-AUKF model; The driver's real-time sEMG signal is collected and input into the AUKF model trained and updated by the MAML method to predict the brake master cylinder pressure.
[0005] Furthermore, the training process of MAML uses tasks as basic units, and the number of tasks is the same as the number of data subsets.
[0006] Furthermore, during the MAML training process, the dataset is divided into a support set and a query set. The support set consists of training samples in each task, which is used to simulate the situation of learning tasks under small sample conditions and update the AUKF model parameters through samples of different tasks; the query set is a set of samples for evaluating the performance of the model on each task, which contains verification samples in each task. The loss function is calculated through the query set samples, and the initial parameters of the AUKF model are optimized in the outer loop.
[0007] Furthermore, the inner loop training process is specifically as follows: Assume the number of tasks in a single training batch is , then the j-th task is expressed as Task The corresponding support set is ,but ,in, To support the set The number of samples in the medium; the design parameters of the AUKF model are , then the MAML-AUKF model uses the parameter equation express; Use the gradient descent algorithm to extract specific tasks In the support set The inner loop training and AUKF model parameter update are implemented above. The specific process is as follows: (1) Where, , , for the The updated AUKF model parameters are: is the number of inner loop training iterations, represents the inner loop optimization learning rate, Optimize the objective function for the inner loop gradient.
[0008] Furthermore, the objective function It is expressed as follows: (2).
[0009] Furthermore, the outer loop training process is specifically as follows: Assume the number of tasks in a single training batch is , then the j-th task is expressed as Task The corresponding query set is ,but ,in, For query set The number of samples in the medium; the design parameters of the AUKF model are , then the MAML-AUKF model uses the parameter equation express; Task-based Query set , the AUKF model parameters are updated through the outer loop training of the MAML-AUKF model. The process is expressed as: (3) Where, are the updated AUKF model parameters, is the learning rate of the outer loop training, is the weight of the j-th task, Training objective function for the outer loop gradient.
[0010] Further, It is expressed as follows: (4).
[0011] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: The present invention constructs a new brake master cylinder pressure prediction model (MAML-AUKF model). The AUKF model is trained using the MAML (Model-Agnostic Meta-Learning, MAML) training strategy. This allows the adaptive unscented Kalman filter (AUKF) design parameters to be quickly updated with a small number of samples, making it adaptable to the frequently changing noise conditions of sEMG signals. Ultimately, this improves the robustness and generalization ability of the AUKF model in predicting brake master cylinder pressure filtering. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] The drawings described herein are used to provide a further understanding of the embodiments of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings: Figure 1 This is the automated search and optimization process for the AUKF model design parameters of the MAML-AUKF framework proposed in this invention.
[0013] Figure 2 The training and testing losses of the MAML-AUKF model proposed in this invention.
[0014] Figure 3 This is the filtering result of the AUKF model under mixed noise proposed in this invention. DETAILED DESCRIPTION
[0015] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0016] The present invention provides a brake master cylinder pressure prediction method based on the MAML-AUKF model, comprising the following steps: Step 1: Construct a MAML-AUKF model and use the MAML training strategy to train the AUKF model to adapt it to the noisy conditions of frequently changing sEMG signals.
[0017] In order to quickly update the design parameters of the adaptive unscented Kalman filter (AUKF) with a small number of samples, adapt it to the noise conditions of frequently changing sEMG signals, and improve the robustness and generalization ability of the AUKF model in predicting the brake master cylinder pressure filtering, the present invention provides a new model MAML-AUKF model.
[0018] like Figure 1 As shown in the figure, the process flow of the automated search and optimization of the design parameters of the AUKF model based on model-agnostic meta-learning (MAML) provided by the present invention is shown. The training process of MAML takes tasks as the basic unit, and the number of tasks is the same as the number of data subsets. In addition, during the training process of MAML, the data set is divided into a support set and a query set. The support set consists of training samples in each task, which is used to simulate the situation of learning tasks under small sample conditions. The AUKF model parameters are updated through samples of different tasks, thereby ensuring that the AUKF model has stronger universality; the query set is a sample set for evaluating the performance of the model on each task, including verification samples in each task. The loss function is calculated through the query set samples, and the initial parameters of the AUKF model are optimized in the outer loop.
[0019] Step 2: Divide the brake master cylinder pressure prediction task into multiple meta-learning tasks according to the driver's braking deceleration peak.
[0020] In order to design the influence of the diversity and uncertainty of sEMG signal noise levels on the prediction of brake master cylinder pressure under different braking levels, the present invention divides the brake master cylinder pressure prediction task into multiple meta-learning tasks according to the braking deceleration peak value, for example, it can be divided into three meta-learning tasks of light braking, moderate braking and heavy braking.
[0021] Step 3: Obtain the driver's leg sEMG signals and brake master cylinder pressure when pressing the brake pedal and establish a dataset. The dataset is then divided into support sets and query sets for different tasks according to a preset ratio.
[0022] The selected driver's sEMG signal and brake master cylinder pressure dataset are divided into a support set and a query set for inner-loop training and outer-loop verification of the MAML-AUKF model.
[0023] Step 4: Based on the support sets of different tasks, the parameters of the AUKF model are updated through the inner loop training of the MAML-AUKF model; based on the query sets of different tasks, the parameters of the AUKF model are updated through the outer loop training of the MAML-AUKF model.
[0024] (1) The above inner loop training process is specifically as follows: Assume the number of tasks in a single training batch is , then the j-th task is expressed as Task The corresponding support set is ,but ,in, To support the set The number of samples in the medium.
[0025] The design parameters of the AUKF model are , then the MAML-AUKF model uses the parameter equation According to the MAML training strategy, in a training batch, there are The tasks are respectively in the support set and queryset Complete the inner loop optimization. Use the gradient descent algorithm to extract specific tasks In the support set The inner loop training and AUKF model parameter update are implemented above. The specific process is as follows: (1) Where, , , for the The updated AUKF model parameters are: is the number of inner loop training iterations, represents the inner loop optimization learning rate, Optimize the objective function for the inner loop gradient.
[0026] Task In the support set The inner loop training is performed on It is expressed as follows: (2).
[0027] (2) The above outer loop training process is specifically as follows: Assume the number of tasks in a single training batch is , then the j-th task is expressed as Task The corresponding query set is ,but ,in, For query set The number of samples in the medium; the design parameters of the AUKF model are , then the MAML-AUKF model uses the parameter equation express.
[0028] Outer loop training realizes the optimization of AUKF model design parameters for multiple tasks, so that it can quickly adapt to new tasks. Query set , the AUKF model parameters are updated through the outer loop training of the MAML-AUKF model. The process is expressed as: (3) Where, are the updated AUKF model parameters, is the learning rate of the outer loop training, is the weight of the j-th task, Training objective function for the outer loop gradient.
[0029] It is expressed as follows: (4).
[0030] (3) Metamodel performance testing After MAML-AUKF training is completed, the updated AUKF design parameters can be used Evaluate its generalization ability for new tasks on the test set. Let the task corresponding to the test set be , and its corresponding support set and query set are and ,but and Support set and queryset In the test phase, if the trained MAML-AUKF model is only trained on a support set containing a small amount of data, After a quick adjustment, you can The higher the accuracy, the stronger the generalization ability of the trained model. The MAML-AUKF model updates the parameters in the test phase through the gradient descent algorithm. : (5) Where, , represents the AUKF design parameters after the lth update, Represents a test task Objective function The gradient, As shown below: (6) Different from the training phase, the query set in the test phase is only used to evaluate the accuracy of the MAML-AUKF model and does not participate in parameter update. The calculation method of the model accuracy is: (7) Step 5: Collect the driver's real-time sEMG signal dataset and input it into the AUKF model trained and updated by the MAML method to predict the brake master cylinder pressure.
[0031] Using the electromyographic (sEMG) signal in the human-machine co-driving interaction information and the AUKF model with updated parameters, the predicted master cylinder pressure of the driver's continuous braking intention can be accurately characterized, the myoelectric (sEMG) signal noise in the human-machine co-driving interaction information can be reduced, and the abnormal fluctuations of the predicted master cylinder pressure that characterizes the driver's continuous braking intention can be reduced.
[0032] The following examples verify the robustness and generalization ability of the MAML-AUKF model proposed in the present invention in predicting brake master cylinder pressure filtering: The effects of the diversity and uncertainty of sEMG signal noise levels on the prediction of master cylinder pressure under different braking levels are designed, and three meta-learning tasks are divided according to the peak value of braking deceleration, including light braking , moderate braking , and heavy braking .
[0033] The driver's sEMG and brake master cylinder pressure datasets were selected and divided into support set and query set in a ratio of 3:1, which were used for inner loop training and outer loop verification of the MAML-AUKF model respectively. The initial values of the AUKF model design parameters are as follows: As shown in the figure, the detailed settings of the design parameter optimization of the AUKF model based on MAML are as follows shown.
[0034] surface Initial values of AUKF model design parameters
[0035] surface Detailed settings for optimizing design parameters of the AUKF model based on MAML
[0036] like As shown in the figure, during the training of the MAML-AUKF model, both the training loss and the test loss tended to be stable at the 250th round of training, and no significant changes occurred. This shows that the model performance has reached a bottleneck at this stage, and further increasing the number of training rounds will not bring significant performance improvement. At the same time, it also shows that the MAML-AUKF model has well fitted the training data and has shown stable generalization ability on the test data. The optimal design parameters of the AUKF model obtained at this time are as follows: shown.
[0037] surface Optimal design parameters of AUKF model
[0038] To verify the filtering effect of the AUKF model on sEMG signals in a complex noise environment, the following three types of uncertainty noise are superimposed to form a mixed noise. The superimposed noise signal can be expressed as: (8) in, 、 and They correspond to three types of uncertainty noise respectively.
[0039] The predicted value of the brake master cylinder pressure under mixed noise is: (9) The signal-to-noise ratio (SNR), mean square error (MSE) and correlation coefficient (CC) are used as indicators to evaluate the filtering performance of the AUKF model. The performance comparison results of EKF, UKF and MAML-AUKF denoising algorithms under mixed noise are shown in Figure 2. The three denoising algorithms are used to denoise the master cylinder pressure signal under mixed noise interference. The results are shown in Figure 2. Compared to the EKF algorithm, the UKF algorithm's SNR and CC improved by 8.13% and 0.84, respectively, while the MSE decreased by 20.52%, demonstrating that the UKF algorithm outperforms the EKF algorithm in complex noise environments. The AUKF model, trained using the MAML method, further improved its noise reduction performance, with SNR and CC ratios increasing by 12.12% and 1.57%, respectively, and the MSE decreasing by 31.04%, demonstrating that the AUKF algorithm achieves the best noise reduction results.
[0040] Table 4 Performance comparison of EKF, UKF and AUKF
[0041] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0042] The present invention is described with reference to flowcharts and / or block diagrams of systems, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0043] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.
[0044] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.
[0045] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.
[0046] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.
Claims
1. A brake master cylinder pressure prediction method based on the MAML-AUKF model, characterized in that: include: A MAML-AUKF model was constructed and the AUKF model was trained using the MAML training strategy to adapt it to the noisy conditions of frequently changing sEMG signals. The brake master cylinder pressure prediction task is divided into multiple meta-learning tasks according to the driver's braking deceleration peak value; The driver's leg sEMG signals and brake master cylinder pressure are acquired when the driver presses the brake pedal, and a dataset is established. The dataset is then divided into support sets and query sets for different tasks according to a preset ratio. Based on the support sets of different tasks, the parameters of the AUKF model are updated through the inner loop training of the MAML-AUKF model; based on the query sets of different tasks, the parameters of the AUKF model are updated through the outer loop training of the MAML-AUKF model; The driver's real-time sEMG signal is collected and input into the AUKF model trained and updated by the MAML method to predict the brake master cylinder pressure.
2. The method for predicting brake master cylinder pressure based on the MAML-AUKF model according to claim 1, characterized in that: The training process of MAML is based on tasks, and the number of tasks is the same as the number of data subsets.
3. The method for predicting brake master cylinder pressure based on the MAML-AUKF model according to claim 1, characterized in that: During the training process of MAML, the dataset is divided into a support set and a query set. The support set consists of training samples in each task, which is used to simulate the situation of learning tasks under small sample conditions and update the AUKF model parameters through samples of different tasks; the query set is a set of samples for evaluating the performance of the model on each task, which contains verification samples in each task. The loss function is calculated through the query set samples, and the initial parameters of the AUKF model are optimized in the outer loop.
4. The method for predicting brake master cylinder pressure based on the MAML-AUKF model according to claim 1, characterized in that: The inner loop training process is specifically as follows: Assume the number of tasks in a single training batch is , then the j-th task is expressed as Task The corresponding support set is ,but ,in, To support the set The number of samples in the medium; the design parameters of the AUKF model are , then the MAML-AUKF model uses the parameter equation express; Use the gradient descent algorithm to extract specific tasks In the support set The inner loop training and AUKF model parameter update are implemented above. The specific process is as follows: (1) Where, , , for the The updated AUKF model parameters are: is the number of inner loop training iterations, represents the inner loop optimization learning rate, Optimize the objective function for the inner loop gradient.
5. The method for predicting brake master cylinder pressure based on the MAML-AUKF model according to claim 4, characterized in that: Objective function It is expressed as follows: (2)。 6. The method for predicting brake master cylinder pressure based on the MAML-AUKF model according to claim 1, characterized in that: The outer loop training process is specifically as follows: Assume the number of tasks in a single training batch is , then the j-th task is expressed as Task The corresponding query set is ,but ,in, For query set The number of samples in the medium; the design parameters of the AUKF model are , then the MAML-AUKF model uses the parameter equation express; Task-based Query set , the AUKF model parameters are updated through the outer loop training of the MAML-AUKF model. The process is expressed as: (3) Where, are the updated AUKF model parameters, is the learning rate of the outer loop training, is the weight of the j-th task, Training objective function for the outer loop gradient.
7. The method for predicting brake master cylinder pressure based on the MAML-AUKF model according to claim 6, characterized in that: It is expressed as follows: (4)。