Methods and apparatus for acoustic measurement and correction of airflow interference in pipelines

Abstract

This application proposes a method and apparatus for correcting acoustic measurements of pipelines that address airflow interference, relating to the field of acoustic measurement correction technology. The method includes: acquiring acoustic measurement values ​​of the pipeline under different parameter combinations to construct an initial dataset; preprocessing the initial dataset, extracting airflow parameters as input features and corresponding acoustic measurement deviations as output labels to construct a structured dataset; optimizing the key parameters of the RBF-BP hybrid model using a sparrow search algorithm to obtain an initial deviation prediction model, training the initial deviation prediction model using the structured dataset to obtain a target deviation prediction model; inputting the current measured airflow parameters into the target deviation prediction model to obtain airflow interference prediction values; and inputting the current acoustic measurement values ​​and the predicted airflow interference values ​​into a preset correction model to obtain corrected acoustic performance parameters. This application can achieve the correction of pipeline acoustic measurement data using the above method.

Description

Technical Field

[0001] This application relates to the field of acoustic measurement and correction technology, and in particular to a method and apparatus for acoustic measurement and correction of airflow interference ducts. Background Technology

[0002] Pipeline acoustic measurement is widely used in fields such as automobiles and aerospace. However, the airflow inside the pipe interacts with the sound waves, producing convection and other effects, which leads to significant deviations in the measurement of acoustic parameters.

[0003] Existing correction methods mainly rely on theoretical formulas, empirical fitting, or static table lookup. These methods are usually based on the ideal flow assumption, have limited generalization ability, and are difficult to achieve real-time and continuous correction, thus failing to meet the actual needs of high-precision online measurement under complex working conditions. Summary of the Invention

[0004] In view of this, this application proposes a method and apparatus for acoustic measurement and correction of airflow interference in pipelines.

[0005] In a first aspect, this application provides a method for correcting acoustic measurements of airflow interference in a pipeline, the method comprising: Acoustic measurement values ​​of the measuring pipe under different parameter combinations are obtained to construct an initial dataset, wherein the parameter combinations include airflow parameters and sound source parameters; The initial dataset is preprocessed, and airflow parameters are extracted from the preprocessed dataset as input features, and the corresponding acoustic measurement deviations are used as output labels to construct a structured dataset. The key parameters of the RBF-BP hybrid model are optimized using the Sparrow Search algorithm to obtain an initial deviation prediction model. The initial deviation prediction model is then trained using the structured dataset to obtain the target deviation prediction model. Input the current measured airflow parameters into the target deviation prediction model to obtain the predicted airflow disturbance value; The current acoustic measurement value and the predicted airflow interference value are input into a preset correction model to obtain the corrected acoustic performance parameters. The preset correction model controls the degree of correction for different frequency bands in the predicted airflow interference value through frequency adaptive weight control.

[0006] In one embodiment, the step of inputting the current acoustic measurement value and the predicted airflow interference value into a preset correction model to obtain the corrected acoustic performance parameters includes: The frequency adaptive weight is multiplied by the predicted airflow disturbance value and added to the airflow stability correction term to obtain the airflow disturbance correction value, wherein the airflow stability correction term is used to compensate for additional disturbances caused by temperature and flow fluctuations. The corrected acoustic performance parameters are obtained by subtracting the airflow interference correction value from the current acoustic measurement value.

[0007] In one embodiment, the preprocessing of the initial dataset includes: The raw data in the initial dataset is filtered to obtain a coarsely filtered dataset. The outliers in the coarsely screened dataset are identified and removed using preset anomaly identification rules to obtain a valid dataset. When a continuous parameter missing value is detected in the valid dataset, the missing value in the valid dataset is filled in using linear interpolation to obtain a corrected dataset.

[0008] In one embodiment, the step of extracting airflow parameters as input features from the preprocessed dataset and using the corresponding acoustic measurement deviations as output labels to construct a structured dataset includes: Airflow parameters are extracted from the preprocessed dataset as input features to obtain the initial input feature set, where each sample in the preprocessed dataset corresponds to a set of feature vectors; Obtain the reference value corresponding to each input feature in the initial input feature set, calculate the difference between the corresponding acoustic measurement value and the reference value to obtain the acoustic measurement deviation, wherein the reference value is the acoustic performance parameter measured under normal temperature environment with the same sound source parameters and no airflow interference; New input features are generated by randomly selecting reference samples from the initial input feature set and superimposing controllable Gaussian noise, until the number of new input features reaches a predetermined number, thus obtaining an extended input feature set. The new input features generated based on the reference samples have the same acoustic measurement deviation as the reference samples. The acoustic measurement deviation corresponding to each input feature in the extended input feature set is used as the output label to construct the structured dataset.

[0009] In one embodiment, the RBF-BP hybrid model includes: an RBF sub-model, a BP sub-model, and a fusion module; The output features of the RBF sub-model are used as the input features of the BP sub-model. The fusion module is used to perform weighted fusion of the output features of the RBF sub-model and the BP sub-model, and the resulting weighted fused features are used as the output features of the RBF-BP hybrid model.

[0010] In one embodiment, the sparrow search algorithm uses a preset multi-objective adaptive optimization function as the fitness calculation function; The multi-objective adaptive optimization function includes a frequency adaptive error term, a physical constraint term, a model complexity term, and a computational efficiency term. The multi-objective adaptive optimization function dynamically adjusts the weight ratios of the four sub-items according to different optimization stages of the sparrow search algorithm through a weight adaptive allocation mechanism. The frequency adaptive error term controls the optimization focus on key parameters in key frequency bands. The physical constraint term constrains the output of the RBF-BP hybrid model to conform to the laws of fluid dynamics. The model complexity term dynamically controls the model complexity of the RBF-BP hybrid model. The computational efficiency term enforces constraints on the model inference time of the RBF-BP hybrid model.

[0011] In one embodiment, training the initial bias prediction model using the structured dataset to obtain the target bias prediction model includes: The initial deviation prediction model is trained using a preset mean squared error function as the target loss function and the structured dataset until the calculated value of the target loss function converges, thus obtaining the target deviation prediction model.

[0012] Secondly, this application also provides an acoustic measurement and correction device for airflow interference ducts, comprising: The acquisition module is used to acquire acoustic measurement values ​​of the measuring pipe under different parameter combinations and construct an initial dataset, wherein the parameter combinations include airflow parameters and sound source parameters; The construction module is used to preprocess the initial dataset, extract airflow parameters as input features from the preprocessed dataset, and use the corresponding acoustic measurement deviation as output labels to construct a structured dataset. The optimization module is used to optimize the key parameters of the RBF-BP hybrid model using the sparrow search algorithm to obtain an initial deviation prediction model. The initial deviation prediction model is then trained using the structured dataset to obtain the target deviation prediction model. The prediction module is used to input the currently measured airflow parameters into the target deviation prediction model to obtain the predicted airflow disturbance value; The calibration module is used to input the current acoustic measurement value and the predicted airflow interference value into a preset calibration model to obtain the corrected acoustic performance parameters. The preset calibration model controls the degree of correction for different frequency bands in the predicted airflow interference value through frequency adaptive weighting.

[0013] Thirdly, this application also provides an electronic device, including a processor and a memory; the memory has a stored computer program, wherein the computer program, when executed by the processor, implements the airflow interference duct acoustic measurement correction method as described in the first aspect.

[0014] Fourthly, this application also provides a non-transitory computer storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the airflow interference duct acoustic measurement correction method as described in the first aspect.

[0015] The airflow interference duct acoustic measurement and correction method of this application has the following advantages over related technologies: 1. The airflow interference duct acoustic measurement correction method of this application first obtains the acoustic measurement values ​​of the duct under different combinations of airflow and sound source parameters to construct an initial dataset, and then extracts the airflow parameters as input features and the acoustic measurement deviation as output labels to construct a structured dataset. It abandons the modeling method based on the ideal flow assumption of the existing method, and instead relies on the real multi-parameter combination data under actual working conditions to lay the foundation for model training, thus breaking through the application scenario limitations brought about by the ideal assumption.

[0016] 2. By employing the Sparrow Search algorithm to optimize the key parameters of the RBF-BP hybrid model, an initial deviation prediction model is obtained. Then, a target deviation prediction model is obtained through training using a structured dataset. The efficient optimization capability of the Sparrow Search algorithm improves the parameter adaptability and prediction accuracy of the RBF-BP hybrid model. Furthermore, training based on diverse real-world operating data enhances the model's generalization ability, enabling it to adapt to complex and variable pipeline airflow conditions. By inputting the currently measured airflow parameters into the target deviation prediction model in real time, the predicted airflow disturbance value can be quickly obtained. This overcomes the drawbacks of static table lookup and empirical fitting methods, which cannot continuously obtain disturbance values, thus achieving real-time and continuous prediction of airflow disturbance values.

[0017] 3. Finally, the current acoustic measurement value and the predicted airflow interference value are input into the preset correction model. Since the preset correction model can control the degree of correction of different frequency bands in the predicted airflow interference value through frequency adaptive weighting, it can achieve accurate acoustic data correction based on the frequency band characteristics of airflow interference. Therefore, the use of the preset correction model for correction ultimately enables the entire correction method to achieve real-time and continuous correction of pipeline acoustic measurement data, which greatly improves the accuracy of acoustic measurement under complex working conditions and effectively meets the actual needs of high-precision online measurement of pipeline acoustics in industrial scenarios. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic flowchart of an acoustic measurement and correction method for airflow interference in a pipeline according to an embodiment of this application. Figure 2 This is a schematic diagram illustrating the process of constructing a structured dataset in one embodiment of this application; Figure 3 This is a schematic diagram of the framework flow of the RBF-BP hybrid model in one embodiment of this application; Figure 4 This is a schematic diagram of the structure of an acoustic measurement and correction device for airflow interference in a pipeline according to one embodiment of this application. Detailed Implementation

[0020] 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.

[0021] As described in the background section, pipe acoustic measurement is widely used in fields such as automotive and aerospace. However, the airflow inside the pipe interacts with sound waves, producing convection and other effects, which leads to significant deviations in the measurement of acoustic parameters. Existing correction methods mainly rely on theoretical formulas, empirical fitting, or static table lookup. These methods are usually based on the assumption of ideal flow, have limited generalization ability, and are difficult to achieve real-time and continuous correction, thus failing to meet the actual needs of high-precision, online measurement under complex operating conditions.

[0022] In recent years, machine learning technology has provided a new approach to handling such complex nonlinear mapping problems. Neural network models can automatically build predictive models from airflow parameters to acoustic deviations by learning from large amounts of experimental data, without relying on strong physical assumptions, and have good adaptability and generalization potential. However, a complete methodological framework and implementation process are still lacking for their systematic application in real-time correction of airflow interference in duct acoustic measurements.

[0023] In some embodiments, such as Figure 1 As shown, this application provides a method for correcting the acoustic measurement of a duct that interferes with airflow, which includes the following steps S101 to S105.

[0024] S101: Obtain acoustic measurements of the measuring pipe under different parameter combinations to construct an initial dataset. The parameter combinations include airflow parameters and sound source parameters. Airflow parameters may include temperature, flow rate, and pressure; sound source parameters may include frequency, sound pressure level, and location. Acoustic measurements can be acoustic performance parameter values, including transmission loss and sound absorption coefficient.

[0025] In applications, the measuring pipe can be a straight, round tube with a smooth inner wall, and the middle section of the pipe can be set as a standard test section for the acoustic element under test. The airflow parameter control unit can include a variable frequency fan, a gas temperature control device, a flow regulating valve, and a pressure regulating valve, used to accurately generate and control the temperature T, volumetric flow rate Q, and pressure P of the airflow within the pipe. Each parameter can be monitored in real time by corresponding temperature sensors, flow meters, and pressure transmitters. The sound source parameter control unit can include a function signal generator, a power amplifier, and a high-quality loudspeaker, used to generate a stable broadband or single-frequency acoustic signal with a specified frequency and sound pressure level (SPL). The sound source installation position can be adjusted along the pipe axis. High-precision microphones can be placed upstream and downstream of the pipe test section. These high-precision microphones can be connected to a multi-channel data acquisition instrument to synchronously acquire sound pressure signals for subsequent calculations of acoustic performance parameters such as transmission loss TL and absorption coefficient α.

[0026] Based on this, under normal temperature conditions, airflow and sound source parameters can be systematically changed, and the original acoustic signals under each parameter combination can be measured simultaneously to construct an initial dataset. Airflow parameters include: temperature T (°C), flow rate Q (m³ / h), and pressure P (MPa), which are adjusted by the airflow parameter control unit and recorded in real time by sensors. Sound source parameters include: frequency (Hz), sound pressure level (dB), and location, which are adjusted and fixed by the sound source parameter control unit. Acoustic performance data, such as transmission loss and absorption coefficient, corresponding to each airflow and sound source parameter combination are synchronously collected using acoustic sensors.

[0027] For example, the ambient temperature range can refer to the industry's standard ambient temperature definition (20-25℃), selecting three levels to cover minor fluctuations within the ambient temperature range, ensuring that the experimental conditions closely resemble actual application scenarios. The frequency range can be strictly limited to 200-1600Hz, with levels evenly divided at 200Hz intervals to cover the main frequency bands affected by mid-to-low frequency pipeline acoustic interference, avoiding parameter redundancy. Other parameters such as flow rate, pressure, sound pressure level, and sound source location are all selected from key levels within the conventional test range, balancing experimental operability and interference pattern coverage. Specific divisions are shown in Table 1 below.

[0028] Table 1 is a table of parameter combinations.

[0029] S102: Preprocess the initial dataset, extract airflow parameters as input features from the preprocessed dataset, and use the corresponding acoustic measurement deviations as output labels to construct a structured dataset.

[0030] It is understandable that preprocessing the initial dataset containing acoustic measurements under different combinations of airflow and sound source parameters (e.g., data cleaning and interpolation) can ensure the validity, integrity, and accuracy of the data, avoiding interference from messy raw data in subsequent data applications. Then, airflow parameters are precisely extracted from the preprocessed high-quality data as input features, focusing on core influencing factors directly related to airflow interference, discarding irrelevant parameters, allowing data applications to accurately target the analysis of the impact of airflow interference on acoustic measurements. Simultaneously, acoustic measurement deviation is used as the output label, clarifying the core correspondence between changes in airflow parameters and the interference-induced deviation in acoustic measurements, transforming the abstract impact of airflow interference into concrete, quantifiable labeled data. Finally, the one-to-one correspondence of input features and output labels is integrated to construct a structured dataset, transforming scattered raw experimental data into a standardized dataset with a unified format and clear logical connections.

[0031] S103: The key parameters of the RBF-BP hybrid model are optimized using the sparrow search algorithm to obtain the initial deviation prediction model. The initial deviation prediction model is then trained using a structured dataset to obtain the target deviation prediction model.

[0032] In applications, the RBF model possesses the core characteristic of local approximation, exhibiting a strong ability to capture and extract local nonlinear features of the input features. Using a Gaussian kernel function as the activation function allows for rapid responses to local input features, resulting in fast modeling convergence and outstanding performance in nonlinear feature extraction. However, when used alone, its ability to fit global features and perform regression prediction is relatively weak. The BP model, on the other hand, is a global approximation neural network model. It iteratively updates weights and biases through the backpropagation algorithm, excelling at fitting and regressing global nonlinear correlations of input features, demonstrating strong global prediction capabilities. However, its ability to capture fine local features of the original input features is insufficient, and it is prone to getting trapped in local optima due to initial parameter settings, resulting in a relatively slow convergence speed. The RBF-BP hybrid model fully integrates the advantages of both models while avoiding their respective shortcomings. Using the RBF model as the pre-feature extraction unit, it can fully leverage its local feature capture and fast convergence characteristics to perform refined local nonlinear feature extraction on input features such as airflow parameters, providing high-dimensional feature vectors that fit the actual laws for subsequent predictions. Then, the BP model is used as the subsequent regression prediction unit, relying on its powerful global fitting and regression capabilities to perform global nonlinear mapping and deviation prediction on the features extracted by the RBF model. The two form an efficient feature extraction-regression prediction complementary structure, which not only improves the model's learning efficiency of the complex nonlinear mapping relationship between airflow parameters and acoustic measurement deviations, but also significantly enhances the overall fitting accuracy and prediction reliability of the model. At the same time, it effectively alleviates the problem that the single BP model is prone to getting trapped in local optima, allowing the model to have better feature learning capabilities and stable deviation prediction capabilities when facing complex and ever-changing pipeline airflow conditions.

[0033] It is understandable that the Sparrow Search algorithm, with its high optimization accuracy, fast convergence speed, and strong resistance to local optima, performs global and precise optimization of key parameters in the RBF-BP hybrid model. This allows the initial parameters of the model to overcome the limitations of random setting, adapt to the nonlinear mapping relationship between airflow parameters and acoustic measurement deviations, and effectively avoid problems such as slow model convergence and low prediction accuracy caused by unreasonable initial parameters. This results in an initial deviation prediction model with better parameter configuration and basic prediction capabilities. Based on this, using a structured dataset with standardized format, high data quality, and clear association between features and labels as the training basis, the initial deviation prediction model fully learns the inherent rules between different airflow parameter input features and corresponding acoustic measurement deviation output labels during iterative training. Through continuous error feedback and parameter adjustment, the model's prediction deviation is continuously corrected, allowing the model's prediction ability to gradually optimize and stabilize, ultimately converging to form a target deviation prediction model with high-precision prediction capabilities.

[0034] S104: Input the current measured airflow parameters into the target deviation prediction model to obtain the predicted airflow disturbance value.

[0035] In actual ambient temperature pipeline acoustic testing, sensors are used to collect airflow parameters of the current test environment in real time, forming a real-time feature vector, i.e., the currently measured airflow parameters:

[0036] in, , and These represent real-time temperature, real-time flow rate, and real-time pressure, respectively.

[0037] Will Input the trained target bias prediction model, and then use the nonlinear mapping relationship of the target bias prediction model. Real-time output of predicted airflow disturbance values:

[0038] S105: Input the current acoustic measurement value and the predicted airflow interference value into the preset correction model to obtain the corrected acoustic performance parameters. The preset correction model controls the degree of correction of different frequency bands in the predicted airflow interference value through frequency adaptive weight control.

[0039] It is understandable that by co-processing the current acoustic measurement values ​​collected in actual testing with the airflow interference prediction values ​​obtained from the target deviation prediction model, interference subtraction of the original acoustic measurement data can be achieved with the help of a preset correction model. Since this correction model does not uniformly correct all frequency bands of the airflow interference prediction values, but is based on a frequency adaptive weight design, it matches appropriate correction weights to the airflow interference prediction values ​​of different frequency bands, taking into account the frequency selectivity of airflow interference on pipeline acoustic measurements. Higher correction weights are given to key frequency bands where airflow interference is more significant, and relatively lower correction weights are set for frequency bands with weaker interference. Therefore, it is possible to achieve differentiated and accurate subtraction of interference values ​​in different frequency bands, and finally obtain the corrected acoustic performance parameters after eliminating airflow interference.

[0040] The aforementioned method for correcting acoustic measurements of airflow interference in pipelines first constructs an initial dataset by acquiring acoustic measurements of the pipeline under different combinations of airflow and sound source parameters. Then, through preprocessing, airflow parameters are extracted as input features, and acoustic measurement deviations are used as output labels to construct a structured dataset. This method abandons the modeling approach based on ideal flow assumptions found in existing methods, instead relying on real multi-parameter combination data under actual working conditions to lay the foundation for model training, thus overcoming the application scenario limitations imposed by ideal assumptions. The method uses a sparrow search algorithm to optimize the key parameters of the RBF-BP hybrid model to obtain an initial deviation prediction model, and then uses the structured dataset to complete model training to obtain a target deviation prediction model. The efficient optimization capability of the sparrow search algorithm improves the parameter adaptability and prediction accuracy of the RBF-BP hybrid model, and training based on diverse real-world working condition data gives the model stronger generalization ability, adapting to complex and varied pipeline airflow conditions. Furthermore, by inputting the current measured airflow parameters into the target deviation prediction model in real time, the predicted airflow interference value can be quickly obtained, overcoming the drawbacks of static table lookup and empirical fitting methods that cannot continuously obtain interference values, thus achieving real-time and continuous prediction of airflow interference values. Finally, the current acoustic measurement value and the predicted airflow interference value are input into the preset correction model. Since the preset correction model can control the degree of correction of different frequency bands in the predicted airflow interference value through frequency adaptive weighting, it can achieve accurate acoustic data correction based on the frequency band characteristics of airflow interference. Therefore, the use of the preset correction model for correction ultimately enables the entire correction method to achieve real-time and continuous correction of pipeline acoustic measurement data, which greatly improves the accuracy of acoustic measurement under complex working conditions and effectively meets the actual needs of high-precision online measurement of pipeline acoustics in industrial scenarios.

[0041] In some embodiments, step S105, inputting the current acoustic measurement value and the predicted airflow interference value into a preset correction model to obtain the corrected acoustic performance parameters, includes: multiplying the frequency adaptive weight with the predicted airflow interference value and adding it to the airflow stability correction term to obtain the airflow interference correction value; and subtracting the airflow interference correction value from the current acoustic measurement value to obtain the corrected acoustic performance parameters. The airflow stability correction term is used to compensate for additional interference caused by temperature and flow fluctuations.

[0042] In application, current acoustic measurements and The time dimensions correspond one-to-one, meaning the two correspond to the acoustic data and airflow interference value at the same moment. Based on the following correction model, the airflow interference value is subtracted from the original acoustic data to obtain the corrected acoustic performance parameters:

[0043] in, This represents the final acoustic performance parameters after correction; This indicates the current acoustic measurement value, which is consistent with the output label parameter type during model training; This represents the predicted airflow disturbance value of the target deviation prediction model for the current measured airflow parameters; The frequency-adaptive weight is expressed as:

[0044] The frequency adaptive weighting is designed based on the characteristic that interference is more significant in the mid-to-low frequency band (200-800Hz), making the correction of key frequency bands more accurate. The airflow stability correction term is expressed as:

[0045]

[0046] The airflow stability correction term is used to compensate for additional disturbances caused by temperature and flow fluctuations, which is in line with the basic laws of fluid mechanics.

[0047] In this embodiment, by multiplying the frequency adaptive weight with the predicted airflow interference value, differentiated correction of airflow interference in different frequency bands is achieved, accurately matching the frequency selectivity of airflow interference and avoiding the accuracy deviation caused by indiscriminate correction. The addition of the airflow stability correction term allows the interference correction to cover the additional interference caused by airflow parameter fluctuations, making the dimensions of interference correction more comprehensive and the results more in line with actual working conditions. Finally, the simple and direct subtraction operation can efficiently eliminate the influence of various airflow interferences from the original acoustic measurement values, effectively improving the overall accuracy of acoustic measurement correction. Moreover, the entire calculation process is clear and has low computational load, enabling rapid real-time correction and adapting to the correction requirements of online and continuous pipeline acoustic measurement under complex working conditions, ensuring that the corrected acoustic performance parameters can accurately reflect the true acoustic characteristics of the test specimen.

[0048] In some embodiments, step S102 involves preprocessing the initial dataset, including: filtering the original data in the initial dataset to obtain a coarse-screened dataset; identifying and removing outliers in the coarse-screened dataset using a preset anomaly identification rule to obtain a valid dataset; and when a continuous parameter missing value is detected in the valid dataset, using linear interpolation to fill in the missing value in the valid dataset to obtain a corrected dataset.

[0049] This can be achieved by deleting samples corresponding to sensor malfunctions or parameter control failures during the experiment, such as temperature samples, to effectively filter the raw data in the initial dataset and obtain a coarse-screened dataset. Then, a 3D model can be used... The principle is to identify and remove outliers in the coarse screening dataset, that is, to calculate the mean of the acoustic measurements of each parameter: temperature T, flow rate Q, and pressure P. and standard deviation If the parameter value in the sample is outside the range If any values ​​are found to be outliers, they are removed, resulting in a valid dataset. Finally, it is determined whether any continuous parameters are missing. If a small number of continuous parameters are missing, linear interpolation is used to fill in the missing values, ensuring data integrity and ultimately yielding a valid and complete corrected dataset.

[0050] In some embodiments, such as Figure 2 As shown, in step S102, airflow parameters are extracted from the preprocessed dataset as input features, and the corresponding acoustic measurement deviations are used as output labels to construct a structured dataset, including the following steps S201 to S204.

[0051] S201: Extract airflow parameters from the preprocessed dataset as input features to obtain the initial input feature set, where each sample in the preprocessed dataset corresponds to a set of feature vectors.

[0052] The airflow parameters temperature T, flow rate Q, and pressure P in the correction dataset are used as input features. Each sample corresponds to a feature vector X=[T,Q,P], and the corresponding features can be obtained by directly reading the parameter values ​​recorded by the sensor.

[0053] S202: Obtain the reference value corresponding to each input feature in the initial input feature set, calculate the difference between the corresponding acoustic measurement value and the reference value to obtain the acoustic measurement deviation, wherein the reference value is the acoustic performance parameter measured under normal temperature environment with the same sound source parameters and no airflow interference.

[0054] In application, acoustic performance baseline values, such as transmission loss TL0 and absorption coefficient α0, are first measured under normal temperature conditions with identical sound source parameters and no airflow interference. For each parameter combination, the corresponding acoustic measurement values ​​are then calculated. Its acoustic measurement deviation If the acoustic parameters are of other types, the deviation formula is adjusted according to the parameter characteristics, and each sample corresponds to an output label y.

[0055] S203: Randomly select reference samples from the initial input feature set and superimpose controllable Gaussian noise to generate new input features until the number of new input features reaches a predetermined number, thereby obtaining an extended input feature set. The new input features generated based on the reference samples have the same acoustic measurement deviation as the reference samples.

[0056] The initial input feature set is denoted as:

[0057] In the above formula, Represents a single set of input features; It is the total number of sets of original input features.

[0058] Assume the number of new input feature groups to be generated: New samples are generated by superimposing controllable Gaussian noise onto the original input features. The specific formula is as follows:

[0059]

[0060] in, Based on the original sample The kth new input feature sample is generated; This is the noise intensity (select a value between 0.01 and 0.05 based on the actual physical range of the input features). Is and Dimensionally consistent standard Gaussian noise, satisfying Generate M sets of new input features. Then, it is merged with the original input features to obtain the expanded dataset:

[0061]

[0062] At this point, the total number of groups in the expanded dataset is: This completes the expansion of the input features.

[0063] S204: Use the acoustic measurement deviation corresponding to each input feature in the extended input feature set as the output label to construct a structured dataset.

[0064] Understandable. Based on the original sample The generated k-th new input feature sample is based on the original sample. Generate all new samples and The corresponding acoustic measurement deviations are the same. Based on this principle, the output labels corresponding to all extended input features can be determined, and a structured dataset can be constructed based on the extended input feature set.

[0065] In this embodiment, core airflow parameters are precisely extracted as input features, eliminating interference from irrelevant parameters and making the input dimensions for model learning more targeted. By calculating acoustic measurement deviations based on a unified benchmark reference value, the physical relationship between features and labels is clarified, enabling the model to accurately learn the intrinsic mapping relationship between changes in airflow parameters and acoustic measurement deviations. The input feature set is expanded by superimposing controllable Gaussian noise, eliminating the need for additional physical experiments and effectively increasing the data volume while controlling experimental costs. Furthermore, the reuse of benchmark sample labels for new features avoids repetitive experiments and calculations, significantly improving the efficiency of data expansion while ensuring the rationality of the correspondence between new samples and labels. The final structured dataset achieves accurate matching between input features and output labels, possesses a standardized format, and has the data volume required for model training, providing a high-quality, highly adaptable data source for subsequent model parameter optimization and model training.

[0066] In some embodiments, the RBF-BP hybrid model includes: an RBF sub-model, a BP sub-model, and a fusion module.

[0067] In this model, the output features of the RBF sub-model are used as the input features of the BP sub-model. The fusion module is used to perform weighted fusion of the output features of the RBF sub-model and the BP sub-model, and the resulting weighted fused features are used as the output features of the RBF-BP hybrid model.

[0068] In applications, such as Figure 3 As shown, the RBF sub-model may include a first input layer, a first hidden layer, and a first output layer.

[0069] First input layer: Set up 3 nodes, the input is the preprocessed airflow parameter feature vector. .

[0070] First hidden layer: 20 nodes are set, and a Gaussian kernel function is used as the activation function to extract local nonlinear features of airflow parameters. The activation function expression is:

[0071] in, , is the center vector of the i-th node in the hidden layer, corresponding to a typical airflow condition; The width parameter controls the local activation range of the Gaussian kernel function; This is the Euclidean distance between the input vector and the center vector.

[0072] First output layer: Outputs a 20-dimensional feature vector. This feature vector is directly used as the input data for the subsequent BP sub-model, realizing the effective transmission of local features.

[0073] The BP sub-model can include a second input layer, a second hidden layer, and a second output layer.

[0074] Second input layer: Set with 20 nodes, corresponding to the output feature vector of the RBF network layer. The dimensions are in one-to-one correspondence, receiving the local feature vectors output by the RBF network layer.

[0075] The second hidden layer has N nodes, where N ranges from 5 to 20 (integers). The Sigmoid function is used as the activation function to achieve a non-linear mapping. The activation function expression is:

[0076] Second output layer: Sets 1 node, uses a linear activation function, and outputs the predicted value of acoustic measurement deviation. To ensure that the predicted value is not limited by the output range, the output expression is:

[0077] in, The weight matrix and bias vector from the second input layer to the second hidden layer are given. The weight matrix and bias vector from the second hidden layer to the second output layer (updated via backpropagation).

[0078] Setting fusion weights in the fusion module This is used to balance the output contributions of the RBF sub-model and the BP sub-model, satisfying the constraints. ,and The final fusion expression is:

[0079] In the formula, The mean of the output feature vectors of the RBF network layer is used to normalize the overall contribution of local features.

[0080] In this embodiment, the RBF sub-model plays a core role in pre-processing local nonlinear feature extraction. For input features such as airflow parameters, it accurately captures the local characteristic patterns of these parameters by leveraging its local approximation properties. The output high-dimensional feature vector is directly used as the input feature of the BP sub-model, providing a feature foundation that fits the actual operating conditions for subsequent predictions. The BP sub-model, using the output of the RBF sub-model as input, leverages its global approximation advantage to perform global nonlinear fitting and regression prediction on the high-dimensional local features, outputting the corresponding acoustic measurement deviation-related prediction features. The fusion module, as the model's output integration unit, integrates the original local features output by the RBF sub-model with the BP sub-model using preset fusion weights. The predicted features of the model output are weighted and fused to balance the contribution of both to the model output, allowing the detailed information of local features to be fully combined with the regular information of global prediction. Finally, the fused features obtained by weighted fusion are used as the output features of the entire RBF-BP hybrid model, realizing a complete nonlinear mapping from the input of airflow parameters to the output of acoustic measurement deviation related features. This structure allows the advantages of the RBF sub-model and the BP sub-model to be fully utilized. At the same time, the feature information output by both is integrated through the fusion module, which makes up for the shortcomings of the single model in feature capture or prediction fitting. This gives the hybrid model a stronger ability to learn and express the complex nonlinear relationship between airflow parameters and acoustic measurement deviation.

[0081] In some embodiments, the sparrow search algorithm uses a preset multi-objective adaptive optimization function as the fitness calculation function.

[0082] The multi-objective adaptive optimization function includes a frequency adaptive error term, a physical constraint term, a model complexity term, and a computational efficiency term. The multi-objective adaptive optimization function dynamically adjusts the weight ratio of the four sub-items according to different optimization stages of the sparrow search algorithm through a weight adaptive allocation mechanism. The frequency adaptive error term is used to control the key parameters that focus the optimization on the key frequency band. The physical constraint term is used to constrain the output of the RBF-BP hybrid model to conform to the laws of fluid dynamics. The model complexity term is used to dynamically control the model complexity of the RBF-BP hybrid model. The computational efficiency term is used to forcibly constrain the model inference time of the RBF-BP hybrid model.

[0083] In the application, the Sparrow Search Algorithm (SSA) is used to globally optimize six key parameters of the RBF-BP hybrid model. These six key parameters include the center vector of the RBF network layer. Width parameters The number of hidden layer nodes in a BP network layer Learning rate and the fusion weight of the fusion module , The specific optimization process is as follows: Addressing the frequency selectivity effect of airflow interference on duct acoustic measurements (i.e., the interference is most significant in the 200–800Hz frequency band), a multi-objective adaptive optimization function is proposed to optimize the key parameters of the RBF-BP hybrid model using the Sparrow Search Algorithm (SSA). This optimization function is:

[0084]

[0085] in, For frequency points (Hz) , As the dominant frequency center of airflow noise, it can be selected ; To control frequency adaptive weight decay, you can select... , so that the weight is When it is 1, in or It decays to 0.22 over time; For frequency The actual measured sound pressure value at that location. The bias value predicted by the RBF-BP mixture model; Let these represent the temperature, flow rate, and pressure at the i-th operating point, respectively. This is the actual measured sound pressure value at this operating point.

[0086] It should be noted that, As a frequency adaptive error term, the optimization focus can be automatically shifted to the mid-to-low frequency band (200–800Hz) where airflow interference is most severe. It is a frequency-adaptive weighting function to avoid over-optimization of irrelevant high-frequency bands.

[0087] These are physical constraint terms used to force the model output to conform to the physical laws of fluid mechanics, thereby improving interpretability and generalization ability. To ensure the physical correlation between sound pressure and operating parameters, a physical constraint function based on Bernoulli's equation is used. a, b, c You can choose according to your needs, for example, it can be 0.001.

[0088] This is a model complexity term. C This is a model complexity metric (the ratio of the number of effective RBF centers to the maximum allowed number of centers). This represents the simplification level of the model. Setting this option can balance accuracy and the number of parameters, ensuring that the model runs efficiently on embedded devices.

[0089] For calculating the efficiency term, T The inference time of the model on the target hardware platform. The weights are dynamic (increased in the later stages of optimization), allowing the focus of later optimization to shift to reducing computation time.

[0090] , , and All are dynamic weights, among which, , , , which can make , For example, the maximum number of iterations for the SSA algorithm is... It can be set to 500.

[0091] The encoding optimization parameters are individual sparrows and population size. Each individual is a 63-dimensional vector. , The initial population is generated randomly:

[0092] in, For the k-th sparrow in the t-th generation, This represents the boundary of the parameter's value range. The random numbers are uniformly distributed in the range [0,1].

[0093] For each initial individual, the value of the optimization function F is calculated by substituting it into the RBF-BP hybrid model and used as the individual fitness value. The top 20% of individuals are selected as explorers for global search, the remaining 80% are only followers (local development), and 10% are randomly selected as vigilants.

[0094] Then, the population positions are updated according to a preset formula, retaining the individual with the best fitness in each generation. The explorer position update formula is:

[0095] in, , , These are random numbers that follow a standard normal distribution.

[0096] The follower position update process is as follows: The top 50% of followers (with better fitness) follow the globally optimal individual. The corresponding formula is:

[0097] The bottom 50% of followers (those with poor fitness) randomly follow other explorers, according to the following formula:

[0098] in, Let the vector be a random vector whose elements take values ​​{1, -1}. These are randomly selected explorer individuals.

[0099] The formula for updating the vigilant's position is:

[0100] The iteration termination condition is as follows: iteration stops when any of the following conditions are met, and the number of iterations reaches [a certain threshold]. Or the change in fitness of the best individual over 10 consecutive generations After the iteration terminates, the parameter combination corresponding to the individual with the lowest fitness is output as the initial parameters of the RBF-BP hybrid model.

[0101] In some embodiments, step S103, training the initial deviation prediction model using a structured dataset to obtain the target deviation prediction model, includes: using a preset mean squared error function as the target loss function, training the initial deviation prediction model using the target loss function and the structured dataset until the calculated value of the target loss function converges, thereby obtaining the target deviation prediction model.

[0102] In the application, the optimal parameters output by SSA are loaded to initialize the parameters of the RBF-BP hybrid model. The training set is divided from the structured dataset.

[0103] Forward propagation: Input the feature vectors from the training set into the model, and calculate the fused prediction value according to the aforementioned formula. .

[0104] Error calculation: The training set is calculated using the mean squared error function and used as the target loss function.

[0105] Backpropagation: Updating the weights and biases of the BP sub-model using gradient descent.

[0106] Iterative training: Repeat the forward propagation, error calculation, and backpropagation steps until the loss function converges (loss value < 0.001 or the number of iterations reaches 500), thereby completing the training of the initial bias prediction model and obtaining the target bias prediction model.

[0107] In some embodiments, please refer to Figure 4 This application also provides an acoustic measurement and correction device 40 for airflow interference ducts, which includes: an acquisition module 41, a construction module 42, an optimization module 43, a prediction module 44, and a correction module 45.

[0108] The acquisition module 41 is used to acquire acoustic measurement values ​​of the measuring pipe under different parameter combinations and construct an initial dataset, wherein the parameter combinations include airflow parameters and sound source parameters.

[0109] The construction module 42 is used to preprocess the initial dataset, extract airflow parameters as input features from the preprocessed dataset, and use the corresponding acoustic measurement deviation as output labels to construct a structured dataset.

[0110] The optimization module 43 is used to optimize the key parameters of the RBF-BP hybrid model using the sparrow search algorithm to obtain the initial deviation prediction model. The initial deviation prediction model is then trained using a structured dataset to obtain the target deviation prediction model.

[0111] The prediction module 44 is used to input the current measured airflow parameters into the target deviation prediction model to obtain the predicted airflow disturbance value.

[0112] The calibration module 45 is used to input the current acoustic measurement value and the airflow interference prediction value into the preset calibration model to obtain the corrected acoustic performance parameters. The preset calibration model controls the degree of correction of different frequency bands in the airflow interference prediction value through frequency adaptive weight control.

[0113] It should be noted that the airflow interference duct acoustic measurement and correction device 40 provided in this application embodiment and the airflow interference duct acoustic measurement and correction method provided in this application embodiment are based on the same inventive concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned airflow interference duct acoustic measurement and correction method, and the repeated parts will not be described again.

[0114] In some embodiments, an electronic device provided in this application includes a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the above-described method for correcting acoustic measurements of airflow interference ducts.

[0115] 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.

[0116] 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.

[0117] This application also provides a non-transitory computer storage medium storing a computer program thereon, which, when executed by a processor, implements the aforementioned method for correcting acoustic measurements of airflow interference ducts. This computer-readable medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into that device / apparatus / system. The aforementioned computer-readable medium carries one or more programs, which, when executed, implement the method as described in the embodiments of this application.

[0118] According to embodiments of this application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples 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 may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable 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.

[0119] 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.

[0120] 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. An air flow disturbance duct acoustic measurement correction method, characterized by, The method for correcting airflow interference duct acoustic measurement includes: Acoustic measurement values ​​of the measuring pipe under different parameter combinations are obtained to construct an initial dataset. The parameter combinations include airflow parameters and sound source parameters. The airflow parameters include temperature, flow rate and pressure. The acoustic measurement values ​​include transmission loss and sound absorption coefficient. The initial dataset is preprocessed, and airflow parameters are extracted from the preprocessed dataset as input features, and the corresponding acoustic measurement deviations are used as output labels to construct a structured dataset. A sparrow search algorithm is used to optimize the key parameters of the RBF-BP hybrid model to obtain an initial bias prediction model. This initial bias prediction model is then trained using the structured dataset to obtain a target bias prediction model. The sparrow search algorithm employs a pre-defined multi-objective adaptive optimization function as its fitness calculation function. This multi-objective adaptive optimization function includes a frequency adaptive error term, a physical constraint term, a model complexity term, and a computational efficiency term. The multi-objective adaptive optimization function dynamically adjusts the weight ratios of these four sub-items according to different optimization stages of the sparrow search algorithm through a weight adaptive allocation mechanism. The frequency adaptive error term controls the optimization focus on key parameters in key frequency bands. The physical constraint term ensures that the output of the RBF-BP hybrid model conforms to the laws of fluid dynamics. The model complexity term dynamically controls the model complexity of the RBF-BP hybrid model. The computational efficiency term enforces constraints on the model inference time of the RBF-BP hybrid model. Input the current measured airflow parameters into the target deviation prediction model to obtain the predicted airflow disturbance value; The current acoustic measurement value and the predicted airflow interference value are input into a preset correction model to obtain the corrected acoustic performance parameters. The preset correction model controls the degree of correction for different frequency bands in the predicted airflow interference value through frequency adaptive weight control.

2. The method for acoustic measurement and correction of airflow interference in pipelines as described in claim 1, characterized in that, The step of inputting the current acoustic measurement value and the predicted airflow interference value into a preset correction model to obtain the corrected acoustic performance parameters includes: The frequency adaptive weight is multiplied by the predicted airflow disturbance value and added to the airflow stability correction term to obtain the airflow disturbance correction value, wherein the airflow stability correction term is used to compensate for additional disturbances caused by temperature and flow fluctuations. The corrected acoustic performance parameters are obtained by subtracting the airflow interference correction value from the current acoustic measurement value.

3. The method for acoustic measurement and correction of airflow interference in pipelines as described in claim 1, characterized in that, The preprocessing of the initial dataset includes: The raw data in the initial dataset is filtered to obtain a coarsely filtered dataset. The outliers in the coarsely screened dataset are identified and removed using preset anomaly identification rules to obtain a valid dataset. When a continuous parameter missing value is detected in the valid dataset, the missing value in the valid dataset is filled in using linear interpolation to obtain a corrected dataset.

4. The method for acoustic measurement and correction of airflow interference in pipelines as described in claim 1, characterized in that, The process of extracting airflow parameters as input features from the preprocessed dataset and using the corresponding acoustic measurement deviations as output labels to construct a structured dataset includes: Airflow parameters are extracted from the preprocessed dataset as input features to obtain the initial input feature set, where each sample in the preprocessed dataset corresponds to a set of feature vectors; Obtain the reference value corresponding to each input feature in the initial input feature set, calculate the difference between the corresponding acoustic measurement value and the reference value to obtain the acoustic measurement deviation, wherein the reference value is the acoustic performance parameter measured under normal temperature environment with the same sound source parameters and no airflow interference; New input features are generated by randomly selecting reference samples from the initial input feature set and superimposing controllable Gaussian noise, until the number of new input features reaches a predetermined number, thus obtaining an extended input feature set. The new input features generated based on the reference samples have the same acoustic measurement deviation as the reference samples. The acoustic measurement deviation corresponding to each input feature in the extended input feature set is used as the output label to construct the structured dataset.

5. The method for acoustic measurement and correction of airflow interference in pipelines as described in claim 1, characterized in that, The RBF-BP hybrid model includes: an RBF sub-model, a BP sub-model, and a fusion module; The output features of the RBF sub-model are used as the input features of the BP sub-model. The fusion module is used to perform weighted fusion of the output features of the RBF sub-model and the BP sub-model, and the resulting weighted fused features are used as the output features of the RBF-BP hybrid model.

6. The method for acoustic measurement and correction of airflow interference in pipelines as described in claim 1, characterized in that, The step of training the initial deviation prediction model using the structured dataset to obtain the target deviation prediction model includes: The initial deviation prediction model is trained using a preset mean squared error function as the target loss function and the structured dataset until the calculated value of the target loss function converges, thus obtaining the target deviation prediction model.

7. A device for measuring and correcting the acoustic properties of a duct that allows for airflow interference, characterized in that... include: The acquisition module is used to acquire acoustic measurement values ​​of the measuring pipe under different parameter combinations and construct an initial dataset. The parameter combinations include airflow parameters and sound source parameters. The airflow parameters include temperature, flow rate and pressure. The acoustic measurement values ​​include transmission loss and sound absorption coefficient. The construction module is used to preprocess the initial dataset, extract airflow parameters as input features from the preprocessed dataset, and use the corresponding acoustic measurement deviation as output labels to construct a structured dataset. An optimization module is used to optimize the key parameters of the RBF-BP hybrid model using a sparrow search algorithm to obtain an initial bias prediction model. The initial bias prediction model is then trained using the structured dataset to obtain a target bias prediction model. The sparrow search algorithm uses a preset multi-objective adaptive optimization function as the fitness calculation function. This multi-objective adaptive optimization function includes a frequency adaptive error term, a physical constraint term, a model complexity term, and a computational efficiency term. The multi-objective adaptive optimization function dynamically adjusts the weight ratio of the four sub-items according to different optimization stages of the sparrow search algorithm through a weight adaptive allocation mechanism. The frequency adaptive error term controls the optimization focus to key parameters in key frequency bands. The physical constraint term constrains the output of the RBF-BP hybrid model to conform to fluid dynamics physical laws. The model complexity term dynamically controls the model complexity of the RBF-BP hybrid model. The computational efficiency term enforces constraints on the model inference time of the RBF-BP hybrid model. The prediction module is used to input the currently measured airflow parameters into the target deviation prediction model to obtain the predicted airflow disturbance value; The calibration module is used to input the current acoustic measurement value and the predicted airflow interference value into a preset calibration model to obtain the corrected acoustic performance parameters. The preset calibration model controls the degree of correction for different frequency bands in the predicted airflow interference value through frequency adaptive weighting.

8. An electronic device, characterized in that, It includes a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the airflow interference duct acoustic measurement correction method as described in any one of claims 1 to 6.

9. A non-transitory computer storage medium, characterized in that, It stores a computer program, wherein the computer program, when executed by a processor, implements the airflow interference duct acoustic measurement correction method as described in any one of claims 1 to 6.

Images