A model construction method and device, electronic equipment and storage medium

Abstract

This application discloses a model building method, apparatus, electronic device, and storage medium, relating to the field of smart home technology. The method includes: acquiring a first noise sample set, which is collected from different types of fluid power equipment under different operating conditions; expanding the first noise sample set based on preset sound quality feature constraints to obtain a second noise sample set; and constructing a sound quality prediction model for the fluid power equipment based on the second noise sample set. The technical solution provided by this application can effectively expand the sample size and reduce the cost of sample acquisition, solving the pain points of insufficient training data and low prediction reliability under small sample conditions, and providing support for the sound quality optimization of range hoods.

Description

Technical Field

[0001] This application relates to the field of smart home technology, and in particular to a model building method, device, electronic device and storage medium. Background Technology

[0002] As consumers increasingly demand greater comfort in their home environments, the noise performance of fluid-powered equipment (such as range hoods) has become a key indicator affecting user experience. Traditional range hood noise evaluations rely solely on sound pressure level, which fails to comprehensively reflect the impact of noise on subjective human perception.

[0003] Existing technologies for noise testing of range hoods are limited by factors such as the complexity of operating conditions (e.g., the coupling of airflow, air pressure, and oil fume concentration) and the difficulty in controlling the test environment. High sample acquisition costs and small dataset sizes (typically less than 1000 samples) make it difficult for data-driven sound quality modeling methods to achieve high-precision predictions due to insufficient training data. Existing data augmentation methods are not suitable for the one-to-one correspondence between sound quality features and labels, and modeling is prone to overfitting, reducing the reliability of sound quality predictions under small sample conditions and hindering the improvement of the overall performance of range hoods. Therefore, developing a sound quality modeling method for range hoods suitable for small sample scenarios has significant engineering value. Summary of the Invention

[0004] This application provides a model building method, apparatus, electronic device, and storage medium, which can effectively expand the sample size and reduce the cost of obtaining samples. It can solve the pain points of insufficient training data and low prediction reliability under small sample conditions, and can provide support for the sound quality optimization of fluid power equipment such as range hoods.

[0005] Firstly, this application provides a model construction method, which includes: A first noise sample set is obtained, which is collected from different types of fluid power equipment under different operating conditions; the first noise sample set is expanded based on preset sound quality feature constraints to obtain a second noise sample set; a sound quality prediction model is constructed for the fluid power equipment based on the second noise sample set.

[0006] Furthermore, the step of augmenting the first noise sample set based on preset sound quality feature constraints to obtain a second noise sample set includes: generating a corresponding target membership function for each psychoacoustic parameter, wherein the psychoacoustic parameters include at least loudness, sharpness, tone modulation, pure tone prominence ratio, sound-to-noise ratio, and speech intelligibility; determining multiple sample generation intervals based on the target membership functions; and performing perturbation operations within the sample generation intervals based on the sound quality feature constraints to obtain the second noise sample set.

[0007] Furthermore, generating a corresponding target membership function for each psychoacoustic parameter includes: constructing an initial membership function with the minimum and maximum values ​​of the subjective evaluation results of the first noise sample set as boundaries and the mean of the subjective evaluation results as the core point; constructing an attention network based on the subjective evaluation results and the psychoacoustic parameters; learning the degree of influence of the psychoacoustic parameters on sound quality through the attention network to obtain attention weights that conform to the sound quality feature constraints; and adjusting the initial membership function based on the attention weights to obtain the target membership function corresponding to each psychoacoustic parameter.

[0008] Furthermore, adjusting the initial membership function based on the attention weight to obtain the target membership function corresponding to each psychoacoustic parameter includes: sorting the psychoacoustic parameters based on the attention weight to obtain a sorting result; dividing the psychoacoustic parameters into at least one key parameter and at least one secondary parameter based on the sorting result; determining a first distribution interval corresponding to each key parameter and a second distribution interval corresponding to each secondary parameter in the subjective evaluation results of the first noise sample set; narrowing the perturbation interval and increasing the membership threshold of the first distribution interval to obtain the target membership function corresponding to the key parameter; and expanding the perturbation interval and decreasing the membership threshold of the second distribution interval to obtain the target membership function corresponding to the secondary parameter.

[0009] Furthermore, the step of performing a perturbation operation within the sample generation interval based on the sound quality feature constraints to obtain the second noise sample set includes: selecting a perturbation method for each sample generation interval based on the attention weight, wherein the perturbation method includes a fusion of linear perturbation and random perturbation; performing perturbation within the corresponding sample generation interval based on the perturbation method through a perturbation constraint factor to generate multiple candidate noise samples; and filtering the candidate noise samples to obtain the second noise sample set.

[0010] Furthermore, the step of filtering the candidate noise samples to obtain the second noise sample set includes: monitoring whether the candidate noise samples have an overfitting trend using a pre-constructed overfitting early warning model; if an overfitting trend exists, adjusting the membership threshold and perturbation interval of the key parameter corresponding to the first distribution interval, and the membership threshold and perturbation interval of the secondary parameter corresponding to the second distribution interval, according to a preset adjustment step size; re-executing the step of performing perturbation based on the perturbation method within the corresponding sample generation interval using perturbation constraint factors to generate multiple candidate noise samples, until the newly generated candidate noise samples do not have an overfitting trend; and using the candidate noise samples without an overfitting trend as the second noise sample set.

[0011] Furthermore, the overfitting early warning model is constructed based on the attention weights.

[0012] Secondly, this application provides a model building apparatus, which includes: The data acquisition module is used to acquire a first noise sample set, which is collected from different types of fluid power equipment under different operating conditions. The data generation module is used to augment the first noise sample set based on preset sound quality feature constraints to obtain a second noise sample set. The model building module is used to build a sound quality prediction model for the fluid power equipment based on the second noise sample set.

[0013] Thirdly, this application provides an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the model building method described in any embodiment of this application.

[0014] Fourthly, this application provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the model building method described in any embodiment of this application.

[0015] To address the shortcomings of existing technologies, this application provides a model construction method that offers the following advantages: First, by acquiring a first noise sample set from different types of fluid power equipment under various operating conditions, this method comprehensively covers the diverse variability of equipment and complex operating conditions, solving the problems of incomplete coverage and insufficient representativeness of traditional samples. This provides high-quality foundational data for subsequent modeling and reduces the cost of sample acquisition. Second, based on sound quality feature constraints, the first noise sample set is expanded to obtain a second noise sample set. This is adapted to the special data structure of one-to-one correspondence between sound quality features and labels, effectively expanding the sample size while avoiding overfitting. This addresses the pain points of insufficient training data and low prediction reliability under small sample conditions. Finally, a sound quality prediction model is constructed based on the second noise sample set. Relying on the expanded high-quality samples, the model's prediction accuracy and generalization ability are significantly improved. This overcomes the limitations of traditional data-driven methods in small sample scenarios, providing support for the sound quality optimization of range hoods and contributing to the improvement of overall equipment performance, thus possessing significant engineering application value.

[0016] It should be noted that the aforementioned computer instructions may be stored, in whole or in part, on a computer-readable storage medium. This computer-readable storage medium may be packaged together with the processor of the model-building apparatus, or it may be packaged separately from the processor of the model-building apparatus; this application does not impose any limitations on this.

[0017] The descriptions of the second, third, and fourth aspects in this application can be referenced to the detailed description of the first aspect; and the beneficial effects described in the second, third, and fourth aspects can be referenced to the analysis of the beneficial effects of the first aspect, which will not be repeated here.

[0018] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description.

[0019] It is understood that before using the technical solutions disclosed in the various embodiments of this application, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this application in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating a model building method provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a model building device provided in an embodiment of this application; Figure 3 This is a block diagram of an electronic device used to implement a model building method according to an embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0023] The terms "first," "second," "target," and "original," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in sequences other than those illustrated or described herein. Furthermore, the terms "comprising," "having," and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0024] It should be noted that the information collected in this application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, necessary confidentiality measures have been taken, and it does not violate public order and good morals. Corresponding operation portals are provided for users to choose to authorize or refuse.

[0025] Figure 1 This is a flowchart illustrating a model building method provided in an embodiment of this application. This embodiment is applicable to scenarios where sound quality prediction models are built in small sample situations. The model building method provided in this embodiment can be executed by the model building device provided in this embodiment. This device can be implemented through software and / or hardware and integrated into the electronic device executing the method. Preferably, the electronic device in this embodiment can be a hydrodynamic device.

[0026] See Figure 1 The method in this embodiment includes, but is not limited to, the following steps: S110. Obtain the first noise sample set, which is collected from different types of fluid power equipment under different operating conditions.

[0027] The first noise sample set in this embodiment may include a collection of original noise acquisition data, characteristic parameters and corresponding label information of fluid power equipment under multiple types and operating conditions. Each sample corresponds to a unique equipment type, operating condition and sound quality parameters.

[0028] The fluid power equipment in this embodiment can be a device that relies on fluid motion to transfer energy and achieve functions such as ventilation and smoke extraction, such as: range hoods (side-suction type, top-suction type), kitchen exhaust fans, industrial ventilators, household blowers and other similar noise-generating devices related to home and industrial scenarios. Among them, range hoods are a typical application scenario and are also the equipment type that this application focuses on. This embodiment uses range hoods as an example to explain the model construction method of this application.

[0029] In this embodiment, the operating condition can refer to the specific operating state of the fluid power equipment. It can be composed of the equipment's own operating parameters (such as motor speed and input power), load parameters (such as air volume, air pressure, and oil fume concentration) and environmental parameters (such as ambient temperature, atmospheric pressure, and ambient noise). Different operating conditions correspond to different operating loads of the equipment, and their noise emission characteristics and sound quality performance are significantly different. Comprehensive coverage of operating conditions is the key to ensuring the representativeness of the first noise sample set.

[0030] In this embodiment, firstly, different types of range hoods are screened to avoid modeling bias caused by a single sample; secondly, different operating conditions are set for each type of range hood to simulate various scenarios in actual use and ensure that the operating conditions cover the normal operating range of the equipment; finally, in a standardized test environment, noise data and corresponding operating condition parameters and characteristic parameters of the range hoods during operation are collected by professional acquisition equipment and then organized to form the first noise sample set.

[0031] For example, the specific steps for obtaining the first noise sample set are as follows: 1. Select three types of mainstream fluid power equipment, namely top-suction, side-suction, and ring-suction range hoods, to ensure sample diversity; 2. Adjust the air volume (e.g., 10~32m³ / h) 3 / min), wind pressure (e.g., 150~1500Pa), and oil fume concentration (e.g., 0~10mg / m³). 3 3. Using a noise acquisition instrument that meets the preset standards, place the instrument at a preset location (e.g., 1.5m directly in front of the range hood and 1.2m above the ground) to simulate the human hearing position. According to the preset sampling frequency and duration, each range hood will be sampled three times under each working condition, and the average of the three samples will be taken as the noise sample data for that working condition. 4. After obtaining the noise sample set of the range hood, preprocess the noise sample set to remove interfering noise. Then, further process the preprocessed noise samples. 5. Perform normalization processing, extract effective noise segments, and obtain standardized noise samples; 6. Conduct subjective evaluation experiments using the equal-interval direct one-dimensional evaluation method, and verify the rationality of the evaluation results by combining normality test, confidence interval analysis, and correlation analysis; At the same time, extract the psychoacoustic parameters of the noise samples and the corresponding subjective evaluation scores to construct a subjective and objective dataset of range hood sound quality, thereby obtaining the first noise sample set; Among them, the psychoacoustic parameters include, but are not limited to, loudness, sharpness, tone modulation, prominence ratio, sound-to-noise ratio, and speech intelligibility, and the subjective evaluation score is a quantitative indicator reflecting the quality of sound.

[0032] S120. Based on the preset sound quality characteristic constraints, the first noise sample set is augmented to obtain the second noise sample set.

[0033] In this embodiment, data augmentation can be a process of generating new noise samples that conform to the real distribution pattern based on the original small sample dataset through reasonable numerical perturbation, interval generation, etc., in order to expand the dataset size and improve the accuracy and generalization ability of subsequent model training.

[0034] In this embodiment, the acoustic quality feature constraints can be a series of pre-set restrictive conditions when augmenting the noise sample data to ensure that the augmented sample still conforms to the acoustic characteristics and subjective evaluation rules of the real device. The acoustic quality feature constraints include, but are not limited to, the range of psychoacoustic parameter values, the range of subjective evaluation scores, the absence of core acoustic quality features, and the constraint of overfitting of samples.

[0035] Optionally, the preset sound quality characteristic constraints can be that the values ​​of each psychoacoustic parameter after expansion do not exceed ±10% of the extreme values ​​of the corresponding parameters in the original sample (i.e., the first noise sample set), and the disturbances of operating parameters (such as air volume and speed) do not exceed the actual operating deviation range of the home, so as to ensure that the expanded sample fits the real acoustic characteristics of the equipment and does not cause distortion of the core sound quality characteristics.

[0036] In some embodiments, the first noise sample set is divided into a predetermined number of groups (e.g., 3 groups) according to equipment type. Within each group, samples with the same structure and similar operating conditions are selected as reference samples. The acoustic characteristics and operating parameters of the reference samples are extracted. The characteristics of different reference samples are then reasonably fused to generate intermediate samples that combine the characteristics of both types of samples, thus initially expanding the sample size. The operating parameters and acoustic parameters of the intermediate samples are then slightly perturbed, for example, the airflow is adjusted by ±0.6m. 3 The speed and rotational speed are perturbed by ±40 r / min, the sound pressure level by ±0.7 dB(A), and the loudness by ±0.25 sone, with random perturbation directions. All parameters meet preset constraints, thus perturbing the intermediate samples to obtain expanded samples and improve sample diversity. Invalid samples with parameters exceeding the constraint range or logical contradictions are removed. The consistency of sound quality patterns between the expanded samples and the baseline samples is verified. The effective expanded samples are merged with the baseline samples to form a second noise sample set, which can meet the requirements of sample size and diversity for subsequent sound quality modeling.

[0037] In some embodiments, the first noise sample set is augmented based on preset acoustic quality feature constraints to obtain a second noise sample set, including: generating a corresponding target membership function for each psychoacoustic parameter, wherein the psychoacoustic parameters include at least loudness, sharpness, tone modulation, pure tone prominence ratio, sound-to-noise ratio, and speech intelligibility; determining multiple sample generation intervals based on the target membership functions; and performing perturbation operations within the sample generation intervals based on acoustic quality feature constraints to obtain the second noise sample set.

[0038] The target membership function can be set for each psychoacoustic parameter, combined with a sound quality target (such as comfort), to define the membership range within which the psychoacoustic parameter meets the target requirements. The sample generation interval can be a multi-dimensional parameter space formed by combining the effective value ranges of multiple psychoacoustic parameters after determining the effective value range of each psychoacoustic parameter that meets the sound quality target based on the target membership function. The perturbation operation can be a small, controllable adjustment (i.e., perturbation) made to each psychoacoustic parameter of the initial sample (which can be obtained by random generation, correction based on the first noise sample, etc.) within the sample generation interval. The adjustment magnitude must meet the sound quality characteristic constraints. The second noise sample set can be the set of all noise samples that meet the target membership function and sound quality characteristic constraints after the perturbation operation.

[0039] Loudness quantifies the intensity of a sound, affecting the subjective perception of noise; excessive loudness can cause discomfort. Sharpness quantifies the sharpness of a sound, related to high-frequency components; higher sharpness makes the ear more easily annoyed. Tone modulation quantifies the pitch of a sound, describing its frequency characteristics and affecting its recognizability and comfort. Pure tone prominence ratio quantifies the prominence of pure tone components in noise; excessive pure tone components make noise sound harsher and affect sound quality. Sound-to-noise ratio (SNR) quantifies the ratio of useful signal (such as speech) to noise signal, affecting signal clarity; a higher SNR results in a clearer signal. Speech intelligibility quantifies the degree to which the human ear can recognize speech signals.

[0040] In one embodiment, generating a corresponding target membership function for each psychoacoustic parameter includes: using the collected first noise sample set as the original basic data, firstly, constructing an initial membership function in a preset form (such as a triangle) with the minimum and maximum values ​​of the subjective evaluation results of the first noise sample set as boundaries and the mean of the subjective evaluation results as the core point; then, constructing an attention network based on the subjective evaluation results and the psychoacoustic parameters; secondly, learning the degree of influence of multiple psychoacoustic parameters on sound quality through the attention network, distinguishing between key parameters and secondary parameters, and obtaining attention weights that conform to the sound quality feature constraints; finally, adjusting the core point and boundaries of the initial membership function based on the attention weights to obtain the target membership function corresponding to each psychoacoustic parameter.

[0041] Alternatively, the attention weights can be determined by: adaptively learning the correlation between each psychoacoustic parameter and the subjective evaluation result of sound quality through an attention network, assigning corresponding weights to each psychoacoustic parameter according to the correlation, giving high weights to key parameters that affect sound quality evaluation, and giving low weights to secondary parameters that affect sound quality evaluation, thereby achieving the enhancement of key features and the suppression of secondary features.

[0042] Furthermore, the initial membership function is adjusted based on attention weights to obtain the target membership function corresponding to each psychoacoustic parameter. This includes: sorting the psychoacoustic parameters based on attention weights to obtain a sorting result; dividing the psychoacoustic parameters into at least one key parameter and at least one secondary parameter based on the sorting result; determining the first distribution interval corresponding to each key parameter and the second distribution interval corresponding to each secondary parameter in the subjective evaluation results of the first noise sample set; narrowing the perturbation interval of the first distribution interval and increasing the membership threshold to obtain the target membership function corresponding to the corresponding key parameter; and expanding the perturbation interval of the second distribution interval and decreasing the membership threshold to obtain the target membership function corresponding to the corresponding secondary parameter.

[0043] Among them, the attention weight can be used to quantify the influence of each psychoacoustic parameter on the subjective evaluation result of sound quality. The higher the weight, the greater the influence of the parameter on the subjective perception of the range hood noise, and the more strictly it needs to be constrained; conversely, the constraints can be appropriately relaxed.

[0044] First, all psychoacoustic parameters are ranked according to their importance based on attention weight. Based on this ranking, the parameters are divided into critical and secondary parameters; high-weight parameters are classified as critical (with strict constraints), and low-weight parameters as secondary (with more relaxed constraints). Then, based on the subjective evaluation distribution of the first noise sample set, a first distribution interval is determined for the critical parameters, and a second distribution interval is determined for the secondary parameters. When constructing the membership function, the first distribution interval of the critical parameters is narrowed according to a preset adjustment method, while the membership threshold is increased to make their value range more concentrated and the judgment more stringent, thus obtaining a more constrained target membership function. For the secondary parameters, their distribution interval is expanded and the membership threshold is lowered to make their value range more relaxed and the fault tolerance higher, ultimately obtaining a target membership function that adapts to the importance of the parameters.

[0045] The initial membership function is shared by all psychoacoustic parameters, while the target membership function is tailored for each psychoacoustic parameter. For high-weight critical parameters, a target membership function with a strict value range is generated by narrowing the perturbation interval and increasing the membership threshold; for low-weight minor parameters, a target membership function with a loose value range is generated by widening the perturbation interval and lowering the membership threshold.

[0046] In one embodiment, a perturbation operation is performed within the sample generation interval based on acoustic quality feature constraints to obtain a second noise sample set, including: selecting a perturbation method for each sample generation interval based on attention weights, the perturbation method including fusion linear perturbation and random perturbation; performing perturbation within the corresponding sample generation interval based on the perturbation method through perturbation constraint factors to generate multiple candidate noise samples; and filtering the candidate noise samples to obtain the second noise sample set.

[0047] Optionally, the perturbation method for each sample interval can be dynamically selected based on the attention weight: the sample interval corresponding to the key parameter adopts a linear perturbation method to ensure the stability of the core features, while the sample interval corresponding to the minor parameter adopts a random perturbation method to improve sample diversity.

[0048] The perturbation constraint factor is related to the acoustic characteristics of the first noise sample. The higher the sound pressure level, the larger the value of the perturbation constraint factor and the smaller the perturbation amplitude. The perturbation constraint factor is used to limit the perturbation amplitude and avoid excessive perturbation that could lead to the loss of core acoustic quality characteristics.

[0049] Furthermore, the candidate noise samples are screened to obtain a second noise sample set, including: monitoring whether there is an overfitting trend in the candidate noise samples through a pre-built overfitting early warning model; if there is an overfitting trend, adjusting the membership threshold and perturbation interval of the key parameters corresponding to the first distribution interval, and the membership threshold and perturbation interval of the secondary parameters corresponding to the second distribution interval according to a preset adjustment step size; re-executing the step of perturbing in the corresponding sample generation interval based on the perturbation method through the perturbation constraint factor to generate multiple candidate noise samples until the newly generated candidate noise samples do not have an overfitting trend; and using the candidate noise samples without an overfitting trend as the second noise sample set.

[0050] Optionally, the overfit warning model is built based on attention weights and is used to monitor the impact of candidate noise samples on the sound quality prediction model in real time.

[0051] The overfitting early warning model monitors whether candidate noise samples are overfitted. If overfitting is detected, the membership threshold and perturbation interval are automatically adjusted, and new noise samples are regenerated until the new noise samples meet the constraints and are not overfitted. Finally, all valid samples are aggregated to form a second noise sample set. The entire process does not change the core sound quality characteristics, but only expands the amount and diversity of data within a reasonable range. The second noise sample set retains the core sound quality characteristics of the first noise sample set, and improves the sample size and diversity through reasonable expansion, effectively solving the problem of insufficient accuracy in traditional small-sample modeling. The constructed sound quality prediction model can accurately match the multivariate characteristics of fluid power equipment.

[0052] S130. Construct a sound quality prediction model for fluid power equipment based on the second noise sample set.

[0053] In this embodiment, the second noise sample set is preprocessed to remove invalid samples and correct abnormal data. The psychoacoustic parameters and operating parameters are standardized to eliminate the impact of differences in the magnitude of different parameters on model training. At the same time, the second noise sample set is divided into a training set (e.g., 70%), a validation set (e.g., 20%), and a test set (e.g., 10%) to provide a reasonable data division for model training, parameter adjustment, and performance verification.

[0054] Choose a model architecture that is suitable for small sample scenarios and has high prediction accuracy. Optionally, the model architecture can employ machine learning algorithms such as Least Absolute Shrinkage and Selection Operator (Lasso) regression, support vector regression, or generalized regression neural networks. Input the preprocessed training set into the model, using psychoacoustic parameters and operating condition parameters as input features, and subjective sound quality evaluation scores as output labels. Iteratively train the model to optimize its parameters, and use the validation set to adjust the model's hyperparameters (such as learning rate and number of iterations) in real time, ensuring that the model can fully learn the correlation between features and labels in the second noise sample set.

[0055] The trained model is validated using a test set. The model's prediction accuracy and generalization ability are verified by calculating the error and accuracy between the predicted score and the actual subjective evaluation score. If the model's performance does not meet the preset standard, the model can be retrained by adjusting the expanded parameters of the second noise sample set or the model's hyperparameters until the requirements are met. Finally, a sound quality prediction model adapted to fluid power equipment (range hood) is obtained, which can be directly used for rapid prediction of sound quality during equipment operation.

[0056] The technical solution provided in this embodiment involves acquiring a first noise sample set, which is collected from different types of fluid power equipment under different operating conditions; expanding the first noise sample set based on preset sound quality feature constraints to obtain a second noise sample set; and constructing a sound quality prediction model for the fluid power equipment based on the second noise sample set. Firstly, by acquiring a first noise sample set from different types of fluid power equipment under different operating conditions, this application can comprehensively cover the diverse variability characteristics and complex operating conditions of the equipment, solving the problems of incomplete coverage and insufficient representativeness of traditional samples, providing high-quality basic data for subsequent modeling, and reducing the cost of sample acquisition. Secondly, expanding the first noise sample set based on sound quality feature constraints to obtain the second noise sample set adapts to the special data structure of one-to-one correspondence between sound quality "features-labels," effectively expanding the sample size while avoiding overfitting, thus addressing the pain points of insufficient training data and low prediction reliability under small sample conditions. Finally, a sound quality prediction model was constructed based on the second noise sample set. Relying on the expanded uniform and high-quality samples, the model's prediction accuracy and generalization ability were greatly improved. This broke through the limitations of traditional data-driven methods in small sample scenarios, providing support for the sound quality optimization of fluid power equipment such as range hoods, and helping to improve the overall performance of the equipment, which has important engineering application value.

[0057] Figure 2 This is a schematic diagram of the structure of a model building device provided in an embodiment of this application, as shown below. Figure 2 As shown, the device 200 may include: The data acquisition module 210 is used to acquire a first noise sample set, which is collected from different types of fluid power equipment under different operating conditions. The data generation module 220 is used to augment the first noise sample set based on preset sound quality feature constraints to obtain a second noise sample set. The model building module 230 is used to build a sound quality prediction model for the fluid power equipment based on the second noise sample set.

[0058] In one embodiment, the data generation module 220 described above can also be specifically used to: generate a corresponding target membership function for each psychoacoustic parameter, wherein the psychoacoustic parameters include at least loudness, sharpness, tone modulation, pure tone prominence ratio, sound-to-noise ratio, and speech intelligibility; determine multiple sample generation intervals based on the target membership function; and perform a perturbation operation within the sample generation interval based on the sound quality feature constraints to obtain the second noise sample set.

[0059] In one embodiment, the data generation module 220 described above can also be specifically used to: construct an initial membership function with the minimum and maximum values ​​of the subjective evaluation results of the first noise sample set as boundaries and the mean of the subjective evaluation results as the core point; construct an attention network based on the subjective evaluation results and the psychoacoustic parameters; learn the degree of influence of the psychoacoustic parameters on sound quality through the attention network to obtain attention weights that conform to the sound quality feature constraints; and adjust the initial membership function based on the attention weights to obtain a target membership function corresponding to each psychoacoustic parameter.

[0060] In one embodiment, the data generation module 220 described above can also be specifically used for: sorting the psychoacoustic parameters based on the attention weights to obtain a sorting result; dividing the psychoacoustic parameters into at least one key parameter and at least one secondary parameter based on the sorting result; determining a first distribution interval corresponding to each key parameter and a second distribution interval corresponding to each secondary parameter in the subjective evaluation results of the first noise sample set; narrowing the perturbation interval of the first distribution interval and increasing the membership threshold to obtain the target membership function corresponding to the key parameter; expanding the perturbation interval of the second distribution interval and decreasing the membership threshold to obtain the target membership function corresponding to the secondary parameter.

[0061] In one embodiment, the data generation module 220 described above can also be specifically used for: selecting a perturbation method for each of the sample generation intervals based on the attention weight, wherein the perturbation method includes a fusion of linear perturbation and random perturbation; performing perturbation within the corresponding sample generation interval based on the perturbation method through a perturbation constraint factor to generate multiple candidate noise samples; and filtering the candidate noise samples to obtain the second noise sample set.

[0062] In one embodiment, the data generation module 220 can also be specifically used to: monitor whether the candidate noise samples have an overfitting trend through a pre-built overfitting early warning model; if an overfitting trend exists, adjust the membership threshold and perturbation interval of the key parameter corresponding to the first distribution interval and the membership threshold and perturbation interval of the secondary parameter corresponding to the second distribution interval according to a preset adjustment step size; re-execute the step of generating multiple candidate noise samples by performing perturbation based on the perturbation method through the perturbation constraint factor in the corresponding sample generation interval until the newly generated candidate noise samples do not have an overfitting trend; and use the candidate noise samples without an overfitting trend as the second noise sample set.

[0063] In one embodiment, the overfitting warning model is constructed based on the attention weights.

[0064] The model building apparatus provided in this embodiment can be applied to the model building methods provided in any of the above embodiments, and has corresponding functions and beneficial effects.

[0065] Figure 3 This is a block diagram of an electronic device used to implement a model-building method according to embodiments of this application. The electronic device 10 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the application described and / or claimed herein.

[0066] like Figure 3 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 may also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0067] Multiple components in electronic device 10 are connected to input / output (I / O) interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of monitors, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0068] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as model building methods.

[0069] In some embodiments, the model-building method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via read-only memory (ROM) 12 and / or communication unit 19. When the computer program is loaded into random access memory (RAM) 13 and executed by processor 11, one or more steps of the model-building method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to execute the model-building method by any other suitable means (e.g., by means of firmware).

[0070] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0071] Computer programs used to implement the methods of this application may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0072] In the context of this application, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0073] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a cathode ray tube (CRT) or liquid crystal display (LCD) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0074] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0075] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to address the shortcomings of traditional physical hosts and virtual private servers, such as high management difficulty and weak business scalability.

[0076] Note that the above are merely preferred embodiments and technical principles applied in this application. Those skilled in the art will understand that this application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of this application. For example, those skilled in the art can use the various forms of processes shown above to reorder, add, or delete steps; the steps described in this application can be executed in parallel, sequentially, or in different orders, as long as the desired results of the technical solution of this application can be achieved, and no limitations are imposed herein.

[0077] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A model construction method, characterized in that, The method includes: A first noise sample set is obtained, which is collected from different types of fluid power equipment under different operating conditions. The first noise sample set is augmented based on preset sound quality feature constraints to obtain the second noise sample set. A sound quality prediction model is constructed for the fluid power equipment based on the second noise sample set.

2. The model construction method according to claim 1, characterized in that, The second noise sample set is obtained by augmenting the first noise sample set based on preset sound quality feature constraints, including: Generate a corresponding target membership function for each psychoacoustic parameter, wherein the psychoacoustic parameters include at least loudness, sharpness, tone modulation, pure tone prominence ratio, sound-to-noise ratio, and speech intelligibility; Multiple sample generation intervals are determined based on the target membership function; Based on the acoustic quality characteristics constrained within the sample generation interval, a perturbation operation is performed to obtain the second noise sample set.

3. The model construction method according to claim 2, characterized in that, The step of generating a corresponding target membership function for each psychoacoustic parameter includes: Using the minimum and maximum values ​​of the subjective evaluation results of the first noise sample set as boundaries and the mean value of the subjective evaluation results as the core point, an initial membership function is constructed. An attention network is constructed based on the subjective evaluation results and the psychoacoustic parameters. The attention network is used to learn the degree of influence of the psychoacoustic parameters on sound quality, and attention weights that conform to the constraints of the sound quality characteristics are obtained. The initial membership function is adjusted based on the attention weights to obtain the target membership function corresponding to each psychoacoustic parameter.

4. The model construction method according to claim 3, characterized in that, The step of adjusting the initial membership function based on the attention weights to obtain the target membership function corresponding to each psychoacoustic parameter includes: The psychoacoustic parameters are sorted based on the attention weights to obtain the sorting results; Based on the sorting results, the psychoacoustic parameters are divided into at least one key parameter and at least one minor parameter; In the subjective evaluation results of the first noise sample set, a first distribution interval corresponding to each key parameter and a second distribution interval corresponding to each minor parameter are determined respectively. By narrowing the disturbance interval and increasing the membership threshold of the first distribution interval, the target membership function corresponding to the key parameter is obtained. By expanding the perturbation interval of the second distribution interval and lowering the membership threshold, the target membership function corresponding to the secondary parameter is obtained.

5. The model construction method according to claim 4, characterized in that, The perturbation operation based on the acoustic quality characteristics constraint within the sample generation interval is used to obtain the second noise sample set, including: The perturbation method for each sample generation interval is selected based on the attention weight, and the perturbation method includes a fusion of linear perturbation and random perturbation; Based on the perturbation method, perturbation is performed within the corresponding sample generation interval using a perturbation constraint factor to generate multiple candidate noise samples; The candidate noise samples are filtered to obtain the second noise sample set.

6. The model construction method according to claim 5, characterized in that, The step of filtering the candidate noise samples to obtain the second noise sample set includes: The candidate noise samples are monitored for overfitting trends using a pre-built overfitting early warning model. If an overfitting trend exists, the membership threshold and perturbation range of the key parameter corresponding to the first distribution interval, and the membership threshold and perturbation range of the secondary parameter corresponding to the second distribution interval are adjusted according to the preset adjustment step size. Repeat the step of performing perturbation based on the perturbation method and the perturbation constraint factor within the corresponding sample generation interval to generate multiple candidate noise samples until the newly generated candidate noise samples do not show an overfitting trend; Candidate noise samples that do not exhibit overfitting tendencies are used as the second noise sample set.

7. The model construction method according to claim 6, characterized in that, The overfitting early warning model is constructed based on the attention weights.

8. A model building apparatus, characterized in that, The device includes: The data acquisition module is used to acquire a first noise sample set, which is collected from different types of fluid power equipment under different operating conditions. The data generation module is used to augment the first noise sample set based on preset sound quality feature constraints to obtain a second noise sample set. The model building module is used to build a sound quality prediction model for the fluid power equipment based on the second noise sample set.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to said at least one processor; The memory stores a computer program that is executed by the at least one processor, such that the at least one processor is able to perform the model building method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that are used to cause a processor to execute the model building method of any one of claims 1 to 7.

Images