Step calibration method for active noise reduction system and active noise reduction system

Abstract

The application discloses a step calibration method of an active noise reduction system and the active noise reduction system, relates to the technical field of active noise control, and the method comprises the following steps: acquiring an error signal and determining a reference noise energy value and a residual noise energy value under different steps; calculating a physical comprehensive evaluation index corresponding to each step according to the reference noise energy value and the residual noise energy value; acquiring simulation error signals of the active noise reduction system under multiple steps, and calculating multiple simulation comprehensive evaluation indexes according to the simulation error signals; based on each step and the corresponding physical comprehensive evaluation index and simulation comprehensive evaluation index, a mapping relationship among the step, the simulation comprehensive evaluation index and the physical comprehensive evaluation index is constructed, the mapping relationship is input into a deep learning model for model training, and a target noise reduction performance prediction model is obtained; based on the target noise reduction performance prediction model, step optimization is performed in a preset step value interval, a target step is obtained, and the secondary sound source is driven to complete step calibration.

Description

Technical Field

[0001] This application relates to the field of active noise control technology, and in particular to a step size calibration method and an active noise reduction system. Background Technology

[0002] Active noise control technology, based on the principle of destructive interference of sound waves, has been widely applied in automotive cabins, indoor areas, and open spaces. The step size, as a core parameter for updating filter weights, directly determines the controller's response amplitude to error signals, and its value affects the system's overall performance, including noise reduction, convergence speed, and stability.

[0003] Currently, the step size calibration of active noise cancellation systems mainly relies on manual trial and error and empirical iteration, which has drawbacks such as large testing workload and long calibration cycle. In traditional methods, there is a performance deviation between the simulation domain and the experimental domain, the prediction accuracy is limited, and it is difficult to directly obtain the optimal value of the step size during actual calibration to ensure the overall optimal control performance.

[0004] Therefore, improving the prediction accuracy and control performance of active noise cancellation systems is a problem that urgently needs to be solved. Summary of the Invention

[0005] The main purpose of this application is to provide a step size calibration method and an active noise cancellation system, aiming to solve the technical problem of how to improve the prediction accuracy and control performance of the active noise cancellation system.

[0006] To achieve the above objectives, this application proposes a step size calibration method for an active noise cancellation system. The step size calibration method for the active noise cancellation system is applied to an active noise cancellation system and includes: Acquire a first error signal and a second error signal, wherein the first error signal is the error signal when the active noise cancellation system is in the off state, and the second error signal is the error signal after the active noise cancellation system has run at multiple step sizes and converged. The reference noise energy value is determined based on the first error signal, the corresponding residual noise energy value is determined based on the second error signal corresponding to each step length, and the physical comprehensive evaluation index corresponding to each step length is calculated based on the reference noise energy value and the residual noise energy value. The physical comprehensive evaluation index is used to quantify the noise reduction performance of the active noise reduction system at the corresponding step length. Obtain the simulation error signal output by the simulation model of the active noise reduction system under the multiple step sizes, and calculate the simulation comprehensive evaluation index corresponding to each of the multiple step sizes based on the simulation error signal. Based on each step size and the corresponding physical comprehensive evaluation index and simulation comprehensive evaluation index, a mapping relationship is constructed between the step size, the simulation comprehensive evaluation index and the physical comprehensive evaluation index. The mapping relationship is then input into a deep learning model for model training to obtain a target noise reduction performance prediction model. Based on the target noise reduction performance prediction model, the step size is optimized within the preset step size range to obtain the target step size. The target step size corresponds to the optimal physical comprehensive evaluation index and is used to drive the secondary sound source to complete the step size calibration.

[0007] In one embodiment, the step of calculating the physical comprehensive evaluation index corresponding to each step size based on the reference noise energy value and the residual noise energy value includes: For each step size, take the logarithm of the ratio of the corresponding reference noise energy value to the residual noise energy value to obtain the steady-state noise reduction component at the current step size; Obtain the real-time noise reduction curve after the active noise reduction system is turned on, and determine the convergence time component at the current step size based on the time required for the instantaneous noise reduction on the real-time noise reduction curve to reach the preset ratio of the steady-state noise reduction. The steady-state noise reduction component and the convergence time component are normalized respectively, and the two normalized components are weighted and summed to obtain the physical comprehensive evaluation index corresponding to the current step size.

[0008] In one embodiment, the step of obtaining the real-time noise reduction curve after the active noise reduction system is turned on, and determining the convergence time component at the current step size based on the time required for the instantaneous noise reduction on the real-time noise reduction curve to reach a preset proportion of the steady-state noise reduction, includes: During the operation of the active noise reduction system at any step length, the instantaneous error signal of the error signal acquisition device at each moment is acquired, and the instantaneous error signal includes the instantaneous energy value; The instantaneous noise reduction amount at each moment is calculated based on the reference noise energy value and the instantaneous energy value at each moment. A real-time noise reduction curve is constructed by plotting time on the horizontal axis and the instantaneous noise reduction amount at each moment on the vertical axis. The steady-state noise reduction amount of the active noise reduction system is determined based on the real-time noise reduction curve. The moment when the instantaneous noise reduction on the real-time noise reduction curve first reaches a preset proportion of the steady-state noise reduction and continuously meets the preset holding time is determined as the convergence moment. The convergence time component is determined based on the time difference between the convergence time and the start time of the active noise cancellation system.

[0009] In one embodiment, the step of obtaining the simulation error signal output by the simulation model of the active noise reduction system at the plurality of step sizes, and calculating the simulation comprehensive evaluation index corresponding to each of the plurality of step sizes based on the simulation error signal, includes: Input the operating parameters of the active noise cancellation system into the simulation model of the active noise cancellation system; The simulation is run in the simulation model at each of the stated step sizes to obtain the simulation error signal output by the simulation model at each stated step size; Based on the simulation error signal at each step size, and using the calculation method of the physical comprehensive evaluation index, the simulation comprehensive evaluation index corresponding to each step size is determined.

[0010] In one embodiment, the deep learning model includes a physical information neural network model. The step of constructing a mapping relationship between step size, simulation comprehensive evaluation index, and physical comprehensive evaluation index based on each step size and the corresponding physical comprehensive evaluation index and simulation comprehensive evaluation index, and inputting the mapping relationship into the deep learning model for model training to obtain a target noise reduction performance prediction model, includes: A portion of step sizes are selected from multiple step sizes as training samples. Each training sample includes the step size and the corresponding simulation comprehensive evaluation index and physical comprehensive evaluation index. Within the preset step size range, physical points are selected, and based on the existing correspondence between step size and simulation comprehensive evaluation index, the simulation comprehensive evaluation index corresponding to each physical point is determined. The step size and simulation comprehensive evaluation index of each training sample are used as inputs, and the physical comprehensive evaluation index is used as the output label. Using the step size of each physical point and the comprehensive simulation evaluation index as the points for applying physical constraints, the physical information neural network model is jointly trained to obtain the target noise reduction performance prediction model.

[0011] In one embodiment, the step of jointly training the physical information neural network model using the step size of each physical point and the comprehensive simulation evaluation index as physical constraint application points to obtain the target noise reduction performance prediction model includes: Based on the predicted values ​​of the physical comprehensive evaluation index output by the physical information neural network model at each physical coordinate point, a smoothing constraint loss term is calculated. The smoothing constraint loss term is used to constrain the output values ​​at adjacent physical coordinate points to meet the preset differential smoothing conditions. Based on the gradient of the predicted value of the physical comprehensive evaluation index output by the physical information neural network model at each physical coordinate point with respect to the input features, a gradient constraint loss term is calculated. The gradient constraint loss term is used to constrain the sensitivity of the predicted value of the physical comprehensive evaluation index to changes in the input features. The smoothing constraint loss term, the gradient constraint loss term, and the data fitting loss term are weighted and summed with preset weights to obtain the total loss value. The network parameters of the physical information neural network model are updated based on the total loss value. When the training termination condition is met, the target noise reduction performance prediction model is obtained.

[0012] In one embodiment, the step of optimizing the step size within a preset step size range based on the target noise reduction performance prediction model to obtain the target step size includes: Multiple candidate step sizes are generated within the preset step size range, and the simulation comprehensive evaluation index corresponding to each candidate step size is determined. Input each candidate step size and its corresponding simulation comprehensive evaluation index into the target noise reduction performance prediction model to obtain the predicted value of the physical comprehensive evaluation index corresponding to each candidate step size. Determine the maximum value of the predicted physical comprehensive evaluation index within the preset step size range; The target step size is determined based on the maximum value of the predicted value of the current operating conditions and the physical comprehensive evaluation index.

[0013] In addition, to achieve the above objectives, this application also proposes an active noise reduction system, which includes a processor, a controller, a data acquisition device, an error signal acquisition device, and a secondary sound source; The data acquisition device is connected to the error signal acquisition device and is used to acquire and record the error signal output by the error signal acquisition device. The error signal acquisition device is connected to the data acquisition device and the processor respectively, and is used to acquire the error signal of the target position and transmit it to the processor; The processor is configured to execute the step size calibration method of the active noise cancellation system as described above, obtain the target step size, and send the target step size to the controller; The controller is connected to the processor and the secondary sound source respectively, and is used to receive the target step size and output a drive signal to the secondary sound source according to the target step size; The secondary sound source is used to receive the driving signal and emit secondary sound waves.

[0014] In addition, to achieve the above objectives, this application also proposes an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the step size calibration method for the active noise cancellation system as described above.

[0015] In addition, to achieve the above objectives, this application also proposes a non-transitory computer storage medium storing a computer program, which, when executed by a processor, implements the steps of the step size calibration method for the active noise reduction system described above.

[0016] One or more technical solutions proposed in this application have at least the following technical effects: (1) Obtain the first error signal and the second error signal; determine the reference noise energy value based on the first error signal, determine the corresponding residual noise energy value based on the second error signal corresponding to each step length, and calculate the physical comprehensive evaluation index corresponding to each step length based on the reference noise energy value and the residual noise energy value. The physical comprehensive evaluation index is used to quantify the noise reduction performance of the active noise reduction system at the corresponding step length; obtain the simulation error signal output by the simulation model of the active noise reduction system at multiple step lengths, and calculate the simulation comprehensive evaluation index corresponding to each step length based on the simulation error signal; construct the mapping relationship between step length, simulation comprehensive evaluation index and physical comprehensive evaluation index based on each step length and the corresponding physical comprehensive evaluation index and simulation comprehensive evaluation index, and input the mapping relationship into the deep learning model for model training to obtain the target noise reduction performance prediction model; optimize the step length within the preset step length value range based on the target noise reduction performance prediction model to obtain the target step length. The physical comprehensive evaluation index corresponding to the target step length is optimal and is used to drive the secondary sound source to complete the step length calibration. This study employs a technique of simultaneously acquiring simulation error signals output by the simulation model at each step size and calculating the comprehensive simulation evaluation index. This addresses the inter-domain bias inherent in relying solely on simulation prediction, achieving a unified representation and complementary advantages of noise reduction performance in both the simulation and experimental domains. A mapping relationship is constructed based on each step size and its corresponding physical and simulation comprehensive evaluation indices, which is then input into a deep learning model for training, resulting in a target noise reduction performance prediction model. This solves the problem of insufficient prediction accuracy caused by performance deviations between simulation and actual measurements, enabling high-precision prediction of the system's measured performance across the entire range and effectively correcting simulation biases. Based on the target noise reduction performance prediction model, step size optimization is performed within a preset step size range to obtain the target step size with the optimal physical comprehensive evaluation index. This drives the secondary sound source to complete calibration, solving the problems of large workload, long cycle, reliance on experience, and difficulty in obtaining the globally optimal step size in traditional calibration testing. This achieves automatic output and efficient calibration of the globally optimal single step size for the active noise reduction system, ensuring that the system's comprehensive control performance, including noise reduction, convergence speed, and stability, reaches its best state.

[0017] (2) Generate multiple candidate step lengths within a preset step length range and determine the simulation comprehensive evaluation index corresponding to each candidate step length; input each candidate step length and its corresponding simulation comprehensive evaluation index into the target noise reduction performance prediction model to obtain the predicted value of the physical comprehensive evaluation index corresponding to each candidate step length; determine the maximum value point of the predicted value of the physical comprehensive evaluation index within the preset step length range; determine the target step length based on the current working condition information and the maximum value point of the predicted value of the physical comprehensive evaluation index. Among them, the target step length includes the absolutely optimal step length and the robust optimal step length. The candidate step length corresponding to the maximum value point of the predicted value of the physical comprehensive evaluation index is the absolutely optimal step length. Taking the maximum value point as the center, determine the step length range within a preset neighborhood where the predicted value of the comprehensive evaluation index satisfies the preset stability condition, and determine the specified candidate step length within the step length range as the robust optimal step length. The target step size is determined based on the maximum value of the predicted value of the current operating condition information and the physical comprehensive evaluation index, which solves the problem that the fixed step size cannot adapt to the stability requirements of multiple scenarios, and enables the system to balance the optimal noise reduction and operational adaptability. The candidate step size corresponding to the maximum value point is defined as the absolute optimal step size, and it is clarified that the target step size includes both absolute and robust optimal forms, providing a basis for flexible application in the future. By adopting the step size interval that satisfies the preset stability conditions of the predicted value of the comprehensive evaluation index within a preset neighborhood with the maximum value point as the center, and determining the specified candidate step size within the interval as the robust optimal step size, the problem of noise amplification and divergence of the system is solved, and a reliable step size selection that balances peak performance and neighborhood robustness is achieved, thereby improving the stability of real vehicle engineering. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, 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 illustrating the first embodiment of the step size calibration method for the active noise reduction system of this application; Figure 2 This is a complete flowchart illustrating the embodiments and implementation methods of this application; Figure 3 This is a schematic diagram of the intelligent optimization prediction results for the bench step length in this application; Figure 4 This is a schematic diagram illustrating the noise reduction effect of the robust optimal step size in a bench environment as described in this application; Figure 5This is a schematic diagram of the intelligent optimization prediction results for the actual vehicle step length in this application; Figure 6 This is a schematic diagram illustrating the noise reduction effect of the driver's seat in a real vehicle under the robust optimal step length at 3500r / min operating conditions, as described in this application.

[0021] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0023] This application provides a step size calibration method for an active noise cancellation system. Specifically, refer to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the step size calibration method for the active noise cancellation system of this application. In this embodiment, the step size calibration method for the active noise cancellation system includes steps S10 to S50:

[0024] Step S10: Obtain the first error signal and the second error signal.

[0025] It should be noted that the first error signal is the error signal when the active noise cancellation system is in the off state, and the second error signal is the error signal after the active noise cancellation system runs at multiple steps and converges.

[0026] For example, the sound pressure signal at the target location can be picked up by an error microphone and a sound pressure sensor, and recorded by the Simcenter data acquisition system and LMS Test.Lab. Specifically, a narrowband notch filter FXLMS algorithm can be used, with a sampling rate set to 5340Hz. A piecewise non-uniform sampling strategy is adopted, selecting 50 step points within a step size range of 0 to 1, acquiring the error microphone signal at each step size, and calculating the corresponding measured noise reduction performance index value.

[0027] Step S20: Determine the reference noise energy value based on the first error signal, determine the corresponding residual noise energy value based on the second error signal corresponding to each step length, and calculate the physical comprehensive evaluation index corresponding to each step length based on the reference noise energy value and the residual noise energy value.

[0028] It should be noted that the reference noise energy value is the total noise energy calculated based on the first error signal, representing the noise intensity benchmark without active control, and is used as a reference standard for evaluating the noise reduction effect. The residual noise energy value is the noise energy calculated based on the second error signal corresponding to each step length, representing the intensity level of residual noise at that specific step length. The physical comprehensive evaluation index is calculated from the reference noise energy value and the residual noise energy value, and is used to quantify the actual noise reduction performance of the active noise cancellation system at the corresponding step length.

[0029] Step S30: Obtain the simulation error signals output by the simulation model of the active noise reduction system under multiple step sizes, and calculate the simulation comprehensive evaluation index corresponding to each of the multiple step sizes based on the simulation error signals.

[0030] It should be noted that the simulation model can be understood as a computer digital model established based on the physical mechanism and transfer characteristics of the active noise cancellation system, used to simulate the sound field response and control results of the system under different step size parameters. The simulation error signal refers to the error signal output by the above simulation model when running at multiple given step sizes, corresponding to the residual noise signal in the virtual simulation environment, used to characterize the system performance under the model prediction. The simulation comprehensive evaluation index is calculated based on the simulation error signal, used to quantify the noise reduction performance of the active noise cancellation system in the simulation environment, and forms a corresponding reference between the simulation domain and the experimental domain with the physical comprehensive evaluation index.

[0031] Step S40: Based on each step length and the corresponding physical comprehensive evaluation index and simulation comprehensive evaluation index, construct the mapping relationship between the step length, simulation comprehensive evaluation index and physical comprehensive evaluation index, and input the mapping relationship into the deep learning model for model training to obtain the target noise reduction performance prediction model.

[0032] It should be noted that the mapping relationship can be understood as a data set or functional relationship describing the one-to-one correspondence between each step size value and its corresponding simulation comprehensive evaluation index and physical comprehensive evaluation index; it serves as a bridge connecting the simulation domain and the experimental domain. The deep learning model can be an artificial neural network model composed of multiple layers of nonlinear processing units. The constructed mapping relationship data can be used as input, and by adjusting the internal parameters of the deep learning model, it can learn the mapping relationship between "step size, simulation evaluation index, and physical evaluation index." The target noise reduction performance prediction model is used to predict the noise reduction performance of an active noise reduction system in a real physical environment at any step size.

[0033] Step S50: Based on the target noise reduction performance prediction model, optimize the step size within the preset step size range to obtain the target step size.

[0034] It should be noted that the preset step size range is a pre-defined range of step size values ​​based on actual engineering constraints and hardware capabilities, serving as the boundary condition for subsequent optimization searches. Within the preset step size range, different step size values ​​can be traversed or searched according to the target noise reduction performance prediction model, their corresponding prediction performance can be compared, and the comprehensive optimal value can be found. The target step size corresponds to the optimal physical comprehensive evaluation index, which is used to drive the secondary sound source to complete the step size calibration.

[0035] In this embodiment, by acquiring the first error signal while the system is off and acquiring the second error signal after convergence at multiple step sizes, the baseline noise energy value, residual noise energy value, and physical comprehensive evaluation index are calculated. This solves the problem of lacking objective quantitative evaluation basis in traditional manual trial-and-error calibration, and achieves objective and accurate quantification of the noise reduction performance of the active noise reduction system in a real physical environment. Because the technique of simultaneously acquiring the simulation error signals output by the simulation model at each step size and calculating the simulation comprehensive evaluation index is adopted, the inter-domain bias caused by relying solely on simulation prediction is solved, achieving a unified representation and complementary advantages of noise reduction performance in the simulation domain and the experimental domain. By employing a technique that constructs a mapping relationship based on each step size and its corresponding physical and simulation comprehensive evaluation indicators, and then inputs this relationship into a deep learning model for training to obtain a target noise reduction performance prediction model, the problem of insufficient prediction accuracy caused by performance deviation between simulation and actual measurement is solved. This achieves high-precision prediction of the system's measured performance across the entire range and effectively corrects simulation deviations. Based on the target noise reduction performance prediction model, step size optimization is performed within a preset step size range to obtain the target step size with the optimal physical comprehensive evaluation indicators. This drives the secondary sound source to complete calibration, solving the problems of large workload, long cycle, reliance on experience, and difficulty in obtaining the globally optimal step size in traditional calibration testing. This achieves automatic output and efficient calibration of the globally optimal single step size for the active noise reduction system, enabling the system to achieve optimal comprehensive control performance in terms of noise reduction, convergence speed, and stability.

[0036] In one implementation, step S20 includes: Step S201: Determine the reference noise energy value based on the first error signal, and determine the corresponding residual noise energy value based on the second error signal corresponding to each step size.

[0037] It should be noted that the reference noise energy value is the noise energy value calculated based on the first error signal, which can be expressed as: This represents the noise intensity baseline before active control is initiated. The residual noise energy value is the noise energy value calculated based on the second error signal corresponding to each step size, and can be expressed as: , representing the intensity level of residual noise at a specific step size.

[0038] Step S202: For each step size, take the logarithm of the ratio of the corresponding reference noise energy value to the residual noise energy value to obtain the steady-state noise reduction component at the current step size.

[0039] For example, the formula for calculating the steady-state noise reduction component is: in, This represents the mean square value of the target position error signal during the off phase of the active noise cancellation system, serving as a benchmark reference for the noise level. This represents the mean square value of the error signal within the last 20% of the time window after the active noise cancellation system is turned on and the system converges. A higher value indicates a better steady-state noise reduction effect.

[0040] Step S203: Obtain the real-time noise reduction curve after the active noise reduction system is turned on, and determine the convergence time component under the current step size based on the time required for the instantaneous noise reduction amount on the real-time noise reduction curve to reach the preset ratio of the steady-state noise reduction amount.

[0041] Specifically, step S203 includes: during the operation of the active noise reduction system at any step length, acquiring the instantaneous error signal of the error signal acquisition device at each moment, the instantaneous error signal including the instantaneous energy value; calculating the instantaneous noise reduction amount at each moment based on the reference noise energy value and the instantaneous energy value at each moment; constructing a real-time noise reduction curve with time as the horizontal axis and the instantaneous noise reduction amount at each moment as the vertical axis; determining the steady-state noise reduction amount of the active noise reduction system based on the real-time noise reduction curve; determining the moment when the instantaneous noise reduction amount on the real-time noise reduction curve first reaches a preset proportion of the steady-state noise reduction amount and continuously meets the preset holding time as the convergence moment; and determining the convergence time component based on the time difference between the convergence moment and the start time of the active noise reduction system.

[0042] For example, the formula for calculating the convergence time component is as follows: in, The real-time noise reduction curve is given, where MSE(t) is the instantaneous error energy calculated using a sliding window with a length of 100 sampling points; the final steady-state noise reduction is also given. , This refers to the last 20% of the system's runtime. To ensure the criteria are consistently met, a time interval of 0.5 seconds is used to filter out spurious convergence caused by transient noise disturbances. The smaller the index value, the faster the system converges.

[0043] In step S204, the steady-state noise reduction component and the convergence time component are normalized respectively, and the two normalized components are weighted and summed to obtain the physical comprehensive evaluation index corresponding to the current step size.

[0044] For example, the formula for calculating the comprehensive physical evaluation index is as follows: Among them, steady-state noise reduction scale The magnitude of the optimal steady-state noise reduction level that the system can achieve under this operating condition; convergence time scale. This represents the order of magnitude of the fastest convergence speed that the system can achieve under this operating condition. , These are the weighting coefficients for the steady-state noise reduction and the convergence time, respectively, satisfying... It can be flexibly adjusted according to the performance priority of different application scenarios.

[0045] For example, for active road noise control under constant speed driving conditions of pure electric vehicles, since the vehicle's operating state is stable, the requirement for system convergence speed is relatively low, but there is an extremely high demand for steady-state noise reduction to improve cabin quietness. Therefore, the following can be selected: =0.8、 =0.2; For hybrid electric vehicle engine start-stop and speed transient conditions, due to the drastic changes in noise characteristics, the system is required to have fast tracking capability. In this case, the weight of convergence time should be increased, and a value of 0.2 can be selected. =0.4、 =0.6. For a very small positive number, take This is used to avoid situations where the denominator is zero. Physics Comprehensive Evaluation Index The larger the positive value, the better the overall performance of the system at that step size; a negative value indicates that the system is noisy and divergent.

[0046] In this embodiment, a reference noise energy value is determined based on a first error signal, and a residual noise energy value is determined based on a second error signal for each step. This solves the problem of traditional manual trial-and-error calibration lacking a unified noise reference and making it difficult to objectively quantify the true residual noise level at each step. It achieves accurate energy level quantification of the measured noise reduction effect at each step using the system's off-state noise as a reference. By taking the logarithm of the ratio of the reference to the residual noise energy to obtain the steady-state noise reduction component, the problem of large absolute energy value ranges and difficulty in intuitively comparing the noise reduction capabilities at different step lengths is solved, achieving standardized characterization of the system's steady-state noise reduction at the decibel scale. By acquiring the real-time noise reduction curve and determining the convergence time component based on the time required for the instantaneous noise reduction to reach a preset proportion of the steady-state noise reduction, the active noise reduction system is subjected to standardized characterization. The method quantitatively extracts the dynamic convergence process; it acquires instantaneous error signals at various times through an error signal acquisition device, calculates instantaneous noise reduction, constructs real-time noise reduction curves, and determines the convergence time component based on the preset holding time and time difference, thus solving the problem that the determination of the convergence time is easily affected by instantaneous noise disturbances; finally, it normalizes and weights the steady-state noise reduction component and the convergence time component to obtain a physical comprehensive evaluation index, solving the problem that the steady-state noise reduction and convergence time have different dimensions, cannot be uniformly evaluated, and rely on empirical judgment. It realizes the integration of steady-state noise reduction capability and transient convergence speed into a unified, calculable, and rankable comprehensive evaluation framework, transforming step size evaluation from a single empirical observation to an objective quantitative ranking, and providing a reliable physical performance measurement basis for subsequent global optimal step size optimization.

[0047] In one implementation, step S30 includes: Step S301: Input the operating parameters of the active noise cancellation system into the simulation model of the active noise cancellation system.

[0048] It should be noted that the simulation model refers to the simulation model of the active noise reduction algorithm based on the active noise reduction algorithm. For example, a virtual system can be constructed in the Matlab R2021a environment using measured primary noise, secondary paths, and the same algorithm, such as narrowband notch filter FXLMS, and with the same sampling rate.

[0049] Step S302: Run the simulation in the simulation model at each step size to obtain the simulation error signal output by the simulation model at each step size.

[0050] It should be noted that the simulation error signal is the error signal output by the simulation model after it converges at a given step size. It corresponds to the residual noise signal in the virtual simulation environment and is used to calculate the comprehensive evaluation index of the simulation in subsequent calculations.

[0051] Step S303: Based on the simulation error signal at each step size, determine the simulation comprehensive evaluation index corresponding to each step size using the calculation method of the physical comprehensive evaluation index.

[0052] It should be noted that the simulation comprehensive evaluation index is a comprehensive metric used to quantify the system's noise reduction performance under simulation conditions, obtained by using the same calculation method as the physical comprehensive evaluation index based on the simulation error signals at each step size. It is denoted as J_sim(μ) and forms a cross-domain correspondence with the physical comprehensive evaluation index J_rig(μ), for subsequent mapping and bias correction. In the Matlab environment, the primary noise, secondary path, and... (The sentence is incomplete and requires further context to translate accurately.) and The values ​​were calculated, and the same active noise reduction algorithm was used for simulation under the same sampling rate conditions.

[0053] For example, the formula for calculating the comprehensive evaluation index of simulation is as follows: in, The formula represents the comprehensive evaluation index of simulation. The other meanings in the formula and the selection of coefficients correspond to the formula of the comprehensive evaluation index of physics, which will not be elaborated here.

[0054] In this embodiment, the technical means of inputting the operating parameters of the active noise cancellation system into the simulation model of the active noise cancellation system solves the problem of simulation prediction deviating from reality. It realizes the reproduction of primary noise, secondary paths and algorithm parameters of the actual test environment in the virtual model, laying the foundation for high-precision alignment between the simulation domain and the test domain. Since the simulation is run at each step size in the simulation model to obtain the simulation error signal at each step size, it solves the problem of large testing workload and long calibration cycle caused by relying on repeated trial and error to collect multi-step signals on the bench or real vehicle. It realizes the rapid batch generation of residual noise signals in the entire step size range in a virtual environment without physical loss, significantly reducing the data acquisition cost. Based on the simulation error signal and the same calculation method as the physical comprehensive evaluation index, the simulation comprehensive evaluation index corresponding to each step size is determined. This solves the problem of the fragmented definition of evaluation indexes between the simulation domain and the test domain and the difficulty in establishing cross-domain mapping correction relationships. It realizes the quantification of simulation performance with a unified comprehensive evaluation framework, so that the simulation index and the subsequent measured physical index can be directly compared and integrated, supporting deviation correction and global optimal step size optimization under small sample conditions.

[0055] In one implementation, the deep learning model includes, but is not limited to, the Physics-Informed Neural Network (PINN), the Domain Adversarial Neural Network (DANN), and the Bayesian Neural Network (BNN). Taking PINN as an example, PINN can integrate data-driven fitting capabilities with physical constraints. By employing a PINN to learn and correct the performance deviation between the simulation domain and the experimental domain, in practical applications, only a small number of measured samples at a few step lengths are needed to achieve high-precision prediction of the measured performance of the active noise reduction system and output the globally optimal step length.

[0056] Based on this, step S40 includes: Step S401: Select a portion of the step lengths from multiple step lengths as training samples. Each training sample includes the step length and the corresponding simulation comprehensive evaluation index and physical comprehensive evaluation index.

[0057] For example, an equally spaced partitioning strategy for 50 sets of data can be adopted, with a fixed random seed of 2026 to ensure the reproducibility of the data processing. First, all valid samples are sorted in ascending order of step size. Then, points taken every 5 samples are used as the test set, i.e., the 5th, 10th, 15th, 20th, 25th, 30th, 35th, 40th, 45th, and 50th samples are used as test samples, totaling 10 sets, to cover the entire range of step size variations. The remaining 40 sets of samples are used as candidate samples for training and validation sets. Then, validation set samples are selected at equally spaced intervals, totaling 8 sets, and the remaining 32 sets are used as the training set.

[0058] Step S402: Select physical points within the preset step size range, and determine the simulation comprehensive evaluation index corresponding to each physical point based on the existing correspondence between step size and simulation comprehensive evaluation index.

[0059] Step S403: Take the step size and simulation comprehensive evaluation index of each training sample as input, and take the physical comprehensive evaluation index as output label.

[0060] For example, in this embodiment, the number of samples in the training set, validation set, and test set are 32 sets, 8 sets, and 10 sets, respectively. Where X=[μ, J_sim(μ)] is the input feature matrix; y=J_rig(μ) is the output label; , The mean and standard deviation are inputs to the training set; , The mean and standard deviation of the training set output are set. If the standard deviation of a feature is 0, it is set to 1 to avoid division by zero error. 200 equally spaced physical points are generated within the step size coverage interval of 50 sets of samples. The simulation performance at the corresponding step size was obtained using the PCHIP interpolation method. Construct the physical collocation feature matrix .

[0061] Step S404: Using the step size of each physical point and the comprehensive simulation evaluation index as the points for applying physical constraints, the physical information neural network model is jointly trained to obtain the target noise reduction performance prediction model.

[0062] It should be noted that each physical collocation point is only used to apply smoothness and gradient constraints to ensure the physical rationality of the prediction curve throughout the entire step size range.

[0063] Specifically, step S404 includes: calculating a smoothing constraint loss term based on the predicted values ​​of the physical comprehensive evaluation index output by the physical information neural network model at each physical coordinate point; the smoothing constraint loss term is used to constrain the output values ​​at adjacent physical coordinate points to meet a preset differential smoothing condition; calculating a gradient constraint loss term based on the gradient of the predicted values ​​of the physical comprehensive evaluation index output by the physical information neural network model at each physical coordinate point with respect to the input features; the gradient constraint loss term is used to constrain the sensitivity of the predicted values ​​of the physical comprehensive evaluation index to changes in the input features; the smoothing constraint loss term and the gradient constraint loss term are weighted and summed with the data fitting loss term with preset weights to obtain a total loss value; the network parameters of the physical information neural network model are updated based on the total loss value; and when the training termination condition is met, a target noise reduction performance prediction model is obtained.

[0064] For example, a few-sample PINN fully connected feedforward neural network can be used as the basic structure, with a network structure of 2→32→32→16→1. The input layer contains two features, corresponding to the stride μ and the simulation performance index J_sim(μ); the three hidden fully connected layers have 32, 32, and 16 neurons respectively, all using the hyperbolic tangent function (tanh) as the activation function, with the corresponding expression as follows:

[0065] The output layer consists of one neuron, corresponding to the predicted experimental performance metric J_rig,pred(μ). The loss function is expressed as:

[0066] in, To smooth the loss weights corresponding to the constraint loss, for example, ; For example, the loss weights corresponding to the gradient constraint loss. This allows the model to achieve a balance between data fitting accuracy, prediction curve smoothness, and robustness to input perturbations.

[0067] Data loss This represents the mean squared error between the predicted values ​​and the true values ​​in the training set, ensuring that the model fits the measured data. The corresponding formula is: in, This indicates the number of samples in the training set. This represents the actual measured comprehensive performance index of the i-th sample. This represents the neural network's prediction for the i-th training sample.

[0068] Smoothing Constraint Loss The sum of squared second-order differences of the predicted physical collocation points, ensuring curve smoothness, is represented by the following formula: in, Indicates the total number of physical points. This represents the prediction performance at the j-th physical coordinate point.

[0069] Gradient-constrained loss This represents the sum of squared gradients of the predicted output with respect to the input features, used to improve robustness. The corresponding formula is: in, This indicates taking the partial derivative with respect to the input vector x. This represents the predicted output at the j-th physical coordinate point.

[0070] For example, the Adam optimizer can be used with a learning rate set to 1^ The maximum number of iterations is 5000 epochs. The validation set loss is calculated every 20 epochs after the first epoch. If the current validation set mean squared error is less than the historical best validation set loss, the current network is saved as the optimal model. If the validation set loss does not improve further for 600 consecutive epochs, an early stopping mechanism is triggered, training is stopped, and the parameters of the network that performed best on the validation set are restored. During training, the changes in total loss, data fitting loss, smoothing physical constraint loss, gradient sensitivity constraint loss, and validation set loss are recorded. After training, the model that performed best on the validation set is saved, along with key parameters such as input / output normalization parameters, dataset partitioning information, and loss weights, for later retrieval.

[0071] In this embodiment, a subset of step lengths is selected from multiple step lengths as training samples, and each training sample includes the step length and the corresponding simulation and physical comprehensive evaluation index. This solves the problem of large workload and long calibration cycle in traditional full-step-length field testing, and achieves efficient small-sample calibration that can start supervised model training with only a small number of measured step length points. By employing the technique of selecting physical collocation points within a preset step length value range and determining the simulation comprehensive evaluation index of each collocation point based on the existing correspondence between step lengths and simulation comprehensive evaluation indexes, the problem of missing continuous mapping in the simulation domain and the inability to constrain the full-range morphology of the model by only discrete interpolation is solved. By using the training sample step length and simulation comprehensive evaluation index as input and the physical comprehensive evaluation index as the output label, the problem of difficulty in establishing the simulation-to-measurement deviation correction relationship is solved. By using each physical collocation point as a physical constraint application point, a physical information neural network model is jointly trained to obtain the target noise reduction performance prediction model, solving the problem of purely data-driven small-sample... This paper addresses the overfitting problem by implementing a high-precision deviation correction model that integrates physical constraints from the simulation domain to the experimental domain. A smoothing constraint loss term is calculated based on the predicted values ​​of physical collocations to constrain the outputs of adjacent physical collocations to meet the differential smoothing condition. This solves the problems of discontinuous prediction curve jitter and distortion across step sizes, achieving physical smoothness and readability of the prediction curve across the entire step size range. By employing a technique of calculating a gradient constraint loss term based on the gradient of the input features using the predicted values ​​of physical collocations to constrain the sensitivity of the predicted values ​​to changes in input features, the paper solves the problem of the model being overly sensitive to small perturbations in the simulation input and exhibiting poor robustness, thus improving the anti-interference capability of the predicted output against feature perturbations. The smoothing constraint loss term, gradient constraint loss term, and data fitting loss term are weighted and summed with preset weights to obtain the total loss value. The network parameters are then updated based on this total loss value until the training termination condition is met to obtain the target model. This solves the problem of conflicting multi-objective optimizations and provides a reliable prediction foundation for subsequent optimization across the entire step size range.

[0072] In one implementation, step S50 includes: generating multiple candidate step lengths within a preset step length range and determining the simulation comprehensive evaluation index corresponding to each candidate step length; inputting each candidate step length and its corresponding simulation comprehensive evaluation index into the target noise reduction performance prediction model to obtain the predicted value of the physical comprehensive evaluation index corresponding to each candidate step length; determining the maximum value point of the predicted value of the physical comprehensive evaluation index within the preset step length range; and determining the target step length based on the current operating condition information and the maximum value point of the predicted value of the physical comprehensive evaluation index.

[0073] It should be noted that the target step size can include both the absolutely optimal step size and the robust optimal step size. The candidate step size corresponding to the maximum value of the predicted value of the physical comprehensive evaluation index is the absolutely optimal step size. Furthermore, with the maximum value as the center, a step size interval is determined within a preset neighborhood where the predicted value of the comprehensive evaluation index satisfies the preset stability condition, and the specified candidate step size within the step size interval is determined as the robust optimal step size.

[0074] Furthermore, the step of writing the target step size into the controller so that the controller drives the secondary sound source according to the target step size includes: acquiring the operating condition information of the active noise cancellation system; determining whether a preset stability priority condition is met based on the operating condition information; if so, writing the robust optimal step size as the target step size into the controller; otherwise, writing the absolutely optimal step size as the target step size into the controller. The complete flowcharts corresponding to the embodiments and implementation methods of the step size calibration method of the active noise cancellation system of this application are as follows: Figure 2 As shown.

[0075] For example, high-density simulations are performed across the entire step size range from 0 to 1, with a step size interval of 0.00001, as a prediction reference. The target noise reduction performance prediction model is loaded to predict the overall system performance across the entire step size range and automatically output the globally optimal step size. The system performance results predicted by the PINN model are as follows: Figure 3 As shown, the absolute optimal step size for intelligent optimization is μ=0.00182, and the robust optimal step size is μ=0.00177.

[0076] Table 1. Noise reduction in dB(A) for the optimal robust step size under bench conditions. The noise reduction of the test bench corresponding to the optimal step size is as follows: Figure 4 As shown in Table 1, the predicted robust optimal step size demonstrates good noise reduction performance and system stability. When other step sizes are selected, such as μ=0.001, the noise reduction at each frequency point is unsatisfactory. While μ=0.0026 achieves good noise reduction at three frequencies, popping noises occur in the initial stage of active noise cancellation, and significant noise amplification occurs in other frequency bands, indicating the system is close to instability. In summary, the robust optimal step size obtained through intelligent optimization using the PINN model has significant advantages in terms of noise reduction, convergence speed, and stability.

[0077] In this embodiment, the target step size is determined based on the current operating condition information and the maximum value of the predicted value of the physical comprehensive evaluation index. This solves the problem that a fixed step size cannot adapt to the stability requirements of multiple scenarios, enabling the system to balance the optimal noise reduction and operational adaptability. The candidate step size corresponding to the maximum value point is defined as the absolute optimal step size, and it is clarified that the target step size includes both absolute and robust optimal forms, providing a basis for flexible application in the future. Since the step size interval that satisfies the preset stability conditions of the predicted value of the comprehensive evaluation index is determined within a preset neighborhood centered on the maximum value point, and the designated candidate step size within the interval is determined as the robust optimal step size, the problem of noise amplification and divergence in the system is solved. This achieves reliable step size selection that balances peak performance and neighborhood robustness, improving the stability of real vehicle engineering.

[0078] To verify the effectiveness of the step size calibration method for the active noise cancellation system in this application, a real vehicle active noise cancellation system step size calibration optimization test was conducted.

[0079] Specifically, a plug-in hybrid electric vehicle was used as the test vehicle. The test condition was a stationary engine speed of 3500 r / min. The hardware and software used were the same as those used in the active noise cancellation system of this application, including the step size calibration method and the hardware and software involved in the active noise cancellation system. The control target was the second and fourth order engine noise at the four seats in the front and middle rows, with frequencies of 116.66 Hz and 233.33 Hz, respectively. A narrowband notch filter FXLMS algorithm was used for noise reduction control, with a sampling frequency of 2790 Hz.

[0080] The step size selection, simulation performance index calculation, measured performance index calculation, and PINN model training methods are as described above and will not be repeated here. The final system performance results predicted by the PINN model are as follows: Figure 5 As shown, the absolute optimal step size for intelligent optimization is μ=0.000244, and the robust optimal step size is μ=0.00024.

[0081] Table 2. Real-world vehicle noise reduction (dB(A)) corresponding to the robust optimal step size under 3500 r / min operating condition. The actual vehicle noise reduction effect corresponding to the optimal step length is as follows: Figure 6 As shown in Table 2, the results demonstrate that the in-vehicle multi-channel active noise reduction system achieved good noise reduction effects under the robust optimal step size for different operating conditions. The noise reduction of the second-order and fourth-order engine noise reached 18.22 dB(A) and 10.73 dB(A), respectively, which fully proves the effectiveness of the proposed intelligent step size optimization method in real vehicle active noise reduction applications.

[0082] Based on the same inventive concept, this application also provides an active noise reduction system, which includes a processor, a controller, a data acquisition device, an error signal acquisition device, and a secondary sound source. The data acquisition device is connected to the error signal acquisition device and is used to acquire and record the error signal output by the error signal acquisition device. The error signal acquisition device is connected to both the data acquisition device and the processor and is used to acquire the error signal at the target position and transmit it to the processor. The processor is configured to execute the step size calibration method of the active noise reduction system as described above to obtain a target step size and send the target step size to the controller. The controller is connected to both the processor and the secondary sound source and is used to receive the target step size and output a drive signal to the secondary sound source according to the target step size. The secondary sound source is used to receive the drive signal and emit secondary sound waves.

[0083] It should be noted that the active noise cancellation system also includes a primary loudspeaker and a monitoring microphone. The primary loudspeaker emits the original primary noise and is the controlled object of the active noise cancellation. For example, the processor can be equipped with CCES, LMS Test.Lab, and MATLAB R2021a software environments to execute the step size calibration method of the active noise cancellation system described above, obtain the target step size, and send this target step size to the controller. The controller is an ADSP21489 controller, which is connected to both the processor and the secondary sound source. It receives the target step size from the processor and runs the narrowband notch filter FXLMS algorithm in real time accordingly, outputting a drive signal to the secondary sound source. The error signal acquisition device includes a MAX9814 error microphone and a GRAS46AE sound pressure sensor, which can be arranged at the noise cancellation target location, coaxially with the primary and secondary loudspeakers, to convert the sound pressure at the target location into an error electrical signal, which is transmitted to the data acquisition device and the processor, respectively. The data acquisition device includes a Simcenter data acquisition system, which is connected to the error signal acquisition device to acquire and record the error signal output by the latter. A separate monitoring microphone can be placed adjacent to the error microphone and connected to the LMS Test.Lab system for data acquisition. This monitoring microphone is an auxiliary front-end of the data acquisition device, used for redundant recording and monitoring. The secondary sound source is the secondary speaker in the Hivi X3 loudspeaker, which is coaxially arranged with the primary speaker and the error microphone. It receives the drive signal output by the controller ADSP21489 and emits equal-amplitude, anti-phase secondary sound waves.

[0084] It should be noted that the step size calibration device for the active noise cancellation system provided in this application embodiment and the step size calibration method for the active noise cancellation system provided in this application embodiment are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned step size calibration method for the active noise cancellation system, and the repeated parts will not be described again.

[0085] In some embodiments, an electronic device provided in this application includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the step size calibration method of the above-described active noise cancellation system.

[0086] Specifically, the processor may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor may also include onboard memory for caching purposes. The processor may be a single processing unit or multiple processing units for performing different actions of the method flow according to embodiments of this application.

[0087] Memory can be any medium capable of containing, storing, transmitting, propagating, or transmitting instructions. For example, memory can include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, instruments, or propagation media. Specific examples of memory include: magnetic storage devices such as magnetic tape or hard disk drives (HDDs); optical storage devices such as optical discs (CD-ROMs); and also random access memory (RAM) or flash memory; and / or wired / wireless communication links.

[0088] This application also provides a non-transitory computer storage medium storing a computer program that, when executed by a processor, implements the step size calibration method of the active noise reduction system described above. This storage medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not be assembled into the device / apparatus / system. The aforementioned non-transitory computer storage medium carries one or more programs, which, when executed, implement the method as described in the embodiments or implementations of this application.

[0089] According to embodiments of this application, a non-transitory computer storage medium can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. A computer-readable signal medium can also be any storage medium other than a computer-readable storage medium that can transmit, propagate, or transfer a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the storage medium can be transmitted using any suitable medium, including but not limited to: wireless, wired, optical fiber, radio frequency signals, etc., or any suitable combination thereof.

[0090] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0091] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application. Therefore, the scope of this application should not be limited to the above embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A step size calibration method for an active noise reduction system, characterized in that, The step size calibration method for the active noise cancellation system is applied to the active noise cancellation system, and the step size calibration method for the active noise cancellation system includes: Acquire a first error signal and a second error signal, wherein the first error signal is the error signal when the active noise cancellation system is in the off state, and the second error signal is the error signal after the active noise cancellation system has run at multiple step sizes and converged. The reference noise energy value is determined based on the first error signal, the corresponding residual noise energy value is determined based on the second error signal corresponding to each step length, and the physical comprehensive evaluation index corresponding to each step length is calculated based on the reference noise energy value and the residual noise energy value. The physical comprehensive evaluation index is used to quantify the noise reduction performance of the active noise reduction system at the corresponding step length. Obtain the simulation error signal output by the simulation model of the active noise reduction system under the multiple step sizes, and calculate the simulation comprehensive evaluation index corresponding to each of the multiple step sizes based on the simulation error signal. Based on each step size and the corresponding physical comprehensive evaluation index and simulation comprehensive evaluation index, a mapping relationship is constructed between the step size, the simulation comprehensive evaluation index and the physical comprehensive evaluation index. The mapping relationship is then input into a deep learning model for model training to obtain a target noise reduction performance prediction model. The target noise reduction performance prediction model is used to predict the noise reduction performance of the active noise reduction system in a real physical environment under any step size. Based on the target noise reduction performance prediction model, the step size is optimized within the preset step size range to obtain the target step size. The target step size corresponds to the optimal physical comprehensive evaluation index and is used to drive the secondary sound source to complete the step size calibration.

2. The method as described in claim 1, characterized in that, The step of calculating the physical comprehensive evaluation index corresponding to each step size based on the reference noise energy value and the residual noise energy value includes: For each step size, take the logarithm of the ratio of the corresponding reference noise energy value to the residual noise energy value to obtain the steady-state noise reduction component at the current step size; Obtain the real-time noise reduction curve after the active noise reduction system is turned on, and determine the convergence time component at the current step size based on the time required for the instantaneous noise reduction on the real-time noise reduction curve to reach the preset ratio of the steady-state noise reduction. The steady-state noise reduction component and the convergence time component are normalized respectively, and the two normalized components are weighted and summed to obtain the physical comprehensive evaluation index corresponding to the current step size.

3. The method as described in claim 2, characterized in that, The step of obtaining the real-time noise reduction curve after the active noise reduction system is turned on, and determining the convergence time component at the current step size based on the time required for the instantaneous noise reduction on the real-time noise reduction curve to reach a preset proportion of the steady-state noise reduction, includes: During the operation of the active noise reduction system at any step length, the instantaneous error signal of the error signal acquisition device at each moment is acquired, and the instantaneous error signal includes the instantaneous energy value; The instantaneous noise reduction amount at each moment is calculated based on the reference noise energy value and the instantaneous energy value at each moment. A real-time noise reduction curve is constructed by plotting time on the horizontal axis and the instantaneous noise reduction amount at each moment on the vertical axis. The steady-state noise reduction amount of the active noise reduction system is determined based on the real-time noise reduction curve. The moment when the instantaneous noise reduction on the real-time noise reduction curve first reaches a preset proportion of the steady-state noise reduction and continuously meets the preset holding time is determined as the convergence moment. The convergence time component is determined based on the time difference between the convergence time and the start time of the active noise cancellation system.

4. The method as described in claim 1, characterized in that, The steps of obtaining the simulation error signals output by the simulation model of the active noise reduction system at the multiple step sizes, and calculating the simulation comprehensive evaluation index corresponding to each of the multiple step sizes based on the simulation error signals, include: Input the operating parameters of the active noise cancellation system into the simulation model of the active noise cancellation system; The simulation is run in the simulation model at each of the stated step sizes to obtain the simulation error signal output by the simulation model at each stated step size; Based on the simulation error signal at each step size, and using the calculation method of the physical comprehensive evaluation index, the simulation comprehensive evaluation index corresponding to each step size is determined.

5. The method as described in claim 1, characterized in that, The deep learning model includes a physical information neural network model. The step of constructing a mapping relationship between step size, simulation comprehensive evaluation index, and physical comprehensive evaluation index based on each step size and the corresponding physical comprehensive evaluation index, and inputting the mapping relationship into the deep learning model for model training to obtain the target noise reduction performance prediction model includes: A portion of step sizes are selected from multiple step sizes as training samples. Each training sample includes the step size and the corresponding simulation comprehensive evaluation index and physical comprehensive evaluation index. Within the preset step size range, physical points are selected, and based on the existing correspondence between step size and simulation comprehensive evaluation index, the simulation comprehensive evaluation index corresponding to each physical point is determined. The step size and simulation comprehensive evaluation index of each training sample are used as inputs, and the physical comprehensive evaluation index is used as the output label. Using the step size of each physical point and the comprehensive simulation evaluation index as the points for applying physical constraints, the physical information neural network model is jointly trained to obtain the target noise reduction performance prediction model.

6. The method as described in claim 5, characterized in that, The step of jointly training the physical information neural network model using the step size of each physical point and the comprehensive simulation evaluation index as physical constraint application points to obtain the target noise reduction performance prediction model includes: Based on the predicted values ​​of the physical comprehensive evaluation index output by the physical information neural network model at each physical coordinate point, a smoothing constraint loss term is calculated. The smoothing constraint loss term is used to constrain the output values ​​at adjacent physical coordinate points to meet the preset differential smoothing conditions. Based on the gradient of the predicted value of the physical comprehensive evaluation index output by the physical information neural network model at each physical coordinate point with respect to the input features, a gradient constraint loss term is calculated. The gradient constraint loss term is used to constrain the sensitivity of the predicted value of the physical comprehensive evaluation index to changes in the input features. The smoothing constraint loss term, the gradient constraint loss term, and the data fitting loss term are weighted and summed with preset weights to obtain the total loss value. The network parameters of the physical information neural network model are updated based on the total loss value. When the training termination condition is met, the target noise reduction performance prediction model is obtained.

7. The method as described in claim 1, characterized in that, The step of optimizing the step size within a preset step size range based on the target noise reduction performance prediction model to obtain the target step size includes: Multiple candidate step sizes are generated within the preset step size range, and the simulation comprehensive evaluation index corresponding to each candidate step size is determined. Input each candidate step size and its corresponding simulation comprehensive evaluation index into the target noise reduction performance prediction model to obtain the predicted value of the physical comprehensive evaluation index corresponding to each candidate step size. Determine the maximum value of the predicted physical comprehensive evaluation index within the preset step size range; The target step size is determined based on the maximum value of the predicted value of the current operating conditions and the physical comprehensive evaluation index.

8. An active noise cancellation system, characterized in that, The active noise cancellation system includes a processor, a controller, a data acquisition device, an error signal acquisition device, and a secondary sound source; The data acquisition device is connected to the error signal acquisition device and is used to acquire and record the error signal output by the error signal acquisition device. The error signal acquisition device is connected to the data acquisition device and the processor respectively, and is used to acquire the error signal of the target position and transmit it to the processor; The processor is configured to execute the step size calibration method of the active noise cancellation system as described in any one of claims 1 to 7, obtain a target step size, and send the target step size to the controller; The controller is connected to the processor and the secondary sound source respectively, and is used to receive the target step size and output a drive signal to the secondary sound source according to the target step size; The secondary sound source is used to receive the driving signal and emit secondary sound waves.

9. An electronic device, characterized in that, The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the step size calibration method for the active noise cancellation system as described in any one of claims 1 to 7.

10. A non-transitory computer storage medium, characterized in that, The non-transitory computer storage medium stores a computer program, which, when executed by a processor, implements the steps of the step size calibration method for the active noise reduction system as described in any one of claims 1 to 7.

Images