Intelligent modeling and sound elimination method and system for sound field of silence shelter

Abstract

This invention discloses an intelligent sound field modeling and anechoic treatment method and system for silent modular cabins, belonging to the field of computer simulation and acoustic modeling technology. The method acquires multi-point transient sound pressure sequence data and frequency distribution characteristics through an acoustic sensor array deployed in three-dimensional space, and simultaneously collects operating status parameters of noise source equipment. Using spatial coordinates, temporal coordinates, and operating status parameters as inputs, a sound field reconstruction model based on a physically constrained neural network is constructed to predict the sound pressure level and phase distribution within the three-dimensional space of the silent modular cabin. The loss function of the sound field reconstruction model is a weighted aggregation of a data-driven loss term, a scalar sound wave equation residual loss term, and a boundary condition constraint loss term. This invention solves the technical problems of low sound field modeling accuracy, static and singular anechoic strategies, inability to cope with complex dynamic noise environments, and insufficient low-frequency anechoic performance in existing silent modular cabin noise reduction processes.

Description

Technical Field

[0001] This invention belongs to the field of computer simulation and acoustic modeling technology, specifically relating to intelligent modeling and noise reduction methods and systems for the sound field of silent cabins. Background Technology

[0002] As an integrated, multi-functional special equipment, the silent modular housing incorporates high-density generator sets, precision instruments, and high-power computing devices. The noise generated by these devices threatens the operational stability and lifespan of electronic components and poses physical and mental health risks to operators who spend extended periods in a confined environment.

[0003] Existing noise reduction strategies for silent modular cabins primarily rely on physical isolation and passive structural optimization. Conventional solutions involve installing soundproof panels within the cabin to physically decouple the noise source area from the sensitive area, or embedding bend-shaped channels in the air intake and exhaust system to utilize sound wave reflection within the geometric structure to attenuate sound energy. These traditional solutions, based on materials mechanics and classical acoustics, face inherent physical limitations. Traditional technologies follow a static, passive defense logic, and their noise reduction characteristics become fixed once the hardware structure is finalized. When fluctuations in the rotational speed of equipment within the cabin cause a drift in the dominant noise frequency, the physical sound-absorbing structure is highly susceptible to a sharp drop in sound-absorbing efficiency due to frequency mismatch.

[0004] The sound field inside the shelter is affected by complex boundary conditions and multi-source interference effects, exhibiting a significant spatial non-uniform distribution. Current technologies lack systematic dynamic monitoring and precise intervention methods for structural secondary radiation noise excited by mechanical vibration and standing waves and sound focusing phenomena formed in the corners of the shelter, resulting in widespread noise reduction blind spots. When passively reducing low-frequency noise, standing wave phenomena are severe in low-frequency regions where the sound wave wavelength is comparable to the characteristic dimensions of the shelter, making it difficult for porous sound-absorbing materials to achieve effective absorption.

[0005] Existing technologies lack an active control mechanism that can perceive the evolution of the cabin sound field in real time and provide closed-loop feedback. When faced with a sound field with complex spatiotemporal evolution, they cannot perform adaptive phase compensation and sound field energy reconstruction. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for intelligent sound field modeling and noise reduction of silent cabins, so as to solve the technical problems of low sound field modeling accuracy, static and single noise reduction strategy, inability to cope with complex dynamic noise environment and insufficient low frequency noise reduction efficiency in the noise reduction process of silent cabins in the prior art.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0008] The method for intelligent sound field modeling and noise reduction in silent modular housing includes the following steps:

[0009] Step 1: Acoustic sensor arrays deployed in the three-dimensional space inside the silent cabin are used to acquire transient sound pressure sequence data and frequency distribution characteristics of multiple measurement points inside the silent cabin in real time. Simultaneously, the operating status parameters of noise source equipment inside the silent cabin are collected. The operating status parameters include rotational speed, load rate, power consumption, vibration acceleration and operating temperature.

[0010] Step 2: Using the transient sound pressure sequence data as direct input features, the operating status parameters of the noise source device are parsed into dynamic adjustment parameters of the sound source excitation boundary conditions and frequency domain focusing regularization term parameters;

[0011] The rotational speed parameter of the noise source device is used to calculate the device's fundamental frequency and harmonic components to update the frequency domain focusing regularization term in real time. The load rate and power consumption parameters of the noise source device are mapped to the sound source volume velocity amplitude to update the boundary conditions of the sound field reconstruction model. The processed data is input into a pre-trained sound field reconstruction model based on a physical constraint neural network. The boundary conditions updated by the operating state parameters are used as the physical driver, and the transient sound pressure sequence data is used as the data driver to predict the sound pressure level distribution and phase distribution inside the silent cabin.

[0012] Step 3: Calculate the residual between the sound pressure level distribution and the target silence threshold, generate an active noise cancellation compensation control law based on the residual, and simultaneously generate a tuning command for the acoustic impedance parameters of the semi-active noise cancellation structure.

[0013] Step 4: Drive the secondary sound source array to emit canceling sound waves according to the active noise cancellation compensation control law, and drive the semi-active noise cancellation structure to adjust the acoustic tuning characteristic parameters according to the acoustic impedance parameter tuning command.

[0014] Furthermore, the sound field reconstruction model based on physically constrained neural networks embeds acoustic physics equations as regularization constraints into the loss function. These acoustic physics equations include scalar sound wave equations:

[0015]

[0016] In the formula, Sound pressure level, unit: ; Speed ​​of sound, unit: ; For the Laplace operator; For time, the unit is .

[0017] Furthermore, the acoustic tuning characteristic parameters include the tuning frequency of the semi-active noise cancellation structure, which is calculated according to the following formula:

[0018]

[0019] In the formula, The tuning frequency is expressed in units of 1000 ppm. ; Speed ​​of sound, unit: ; The neck cross-sectional area of ​​the semi-active noise reduction structure is given in units of... ; The volume of the cavity in the semi-active noise reduction structure is expressed in units of... ; The length of the neck of the semi-active noise reduction structure, in units of ; This is the terminal correction term, in units of .

[0020] Furthermore, in the step of generating the active noise cancellation compensation control law based on the residual, the transient sound pressure field output by the sound field reconstruction model based on the physical constraint neural network is... The complex sound pressure distribution is obtained by mapping the data to the frequency domain using a fast Fourier transform. The particle velocity distribution in each spatial region within the silent container is derived and calculated using the complex sound pressure distribution. The time-averaged acoustic energy flow vector field distribution is then calculated based on the following formula:

[0021]

[0022]

[0023] In the formula, The frequency domain particle velocity vector, in units of ; The imaginary unit; Angular frequency, unit: ; air density, unit: ; It is a complex sound pressure gradient vector in the frequency domain; This is the time-averaged acoustic energy flow vector, in units of ; Indicates taking the real part; It represents complex conjugate; calculates the time-averaged acoustic energy flow divergence of each spatial region in the time-averaged acoustic energy flow vector field distribution, and increases the secondary cancellation signal energy allocation weight of the secondary sound source array to the region for regions with time-averaged acoustic energy flow divergence greater than the preset divergence threshold.

[0024] Furthermore, the sound field reconstruction model based on physical constraint neural network predicts the sound pressure at spatial coordinates of un-deployed acoustic sensors within the silent cabin, establishing virtual error sensing points. The predicted sound pressure values ​​output by the virtual error sensing points are filtered by a virtual secondary channel transfer function model obtained through pre-identification or calculation using the sound field reconstruction model and the free-field Green's function to generate virtual error signals. The virtual error signals and the physical error signals collected by the physical acoustic sensors jointly participate in the iterative update of the weight coefficients of the active noise cancellation compensation control law.

[0025] Furthermore, the prediction variance of the sound field reconstruction model based on a physically constrained neural network at the virtual error sensing point is extracted as a prediction uncertainty index. When the prediction uncertainty index exceeds a preset uncertainty threshold, the weighting coefficient of the virtual error signal in the total error cost function of the active noise cancellation compensation control law is reduced according to a preset linear attenuation function. The preset linear attenuation function is... ,in To predict uncertainty, To preset an uncertainty threshold, The function is determined by a preset attenuation coefficient, which is exited by gradually decreasing the amount of prediction uncertainty as it increases.

[0026] Furthermore, the active noise cancellation compensation control law includes a Volterra series compensation model for pre-compensating the nonlinear distortion of the secondary sound source array; the Volterra series compensation model uses the control signal sequence output by the adaptive filtering algorithm as the input signal sequence. The expression for the output signal of the Volterra series compensation model is:

[0027]

[0028] In the formula, Output signal for Volterra series compensation model; The sequence of control signals output by the adaptive filtering algorithm; For the first Volterra kernel function of order; The highest order of the Volterra series expansion; This represents the memory length of each kernel function. For discrete-time indexing; For the first One delayed variable.

[0029] Furthermore, the predicted sound pressure level difference before and after anechoic chambering and the total system power consumption generated by the silent cabin sound field intelligent modeling and anechoic chamber system are continuously calculated through environmental adaptive optimization closed loop; a joint reward function is constructed using the predicted sound pressure level difference and the total system power consumption using reinforcement learning algorithm, and the network weight parameters of the sound field reconstruction model based on physical constraint neural network are updated according to the joint reward function.

[0030] In practice, the predicted sound pressure level difference before and after anechoic chambering, as well as the total system power consumption generated by the intelligent sound field modeling and anechoic chamber system, are continuously calculated through an environmental adaptive optimization closed loop. A joint reward function is constructed using a policy gradient-based reinforcement learning algorithm to combine the predicted sound pressure level difference with the total system power consumption. ,in This represents the predicted sound pressure level difference before and after noise reduction. The total power consumption of the system is and The network weight parameters of the sound field reconstruction model based on the physical constraint neural network are updated according to the joint reward function, with preset weight coefficients.

[0031] In addition, this invention also discloses a sound field intelligent modeling and anechoic system for a silent modular cabin, used to implement the sound field intelligent modeling and anechoic method for a silent modular cabin as described above, including:

[0032] The acoustic feature acquisition module is used to acquire transient sound pressure sequence data and frequency distribution characteristics of multiple measurement points in the silent cabin;

[0033] The equipment status monitoring module is used to collect the operating status parameters of noise source equipment in the silent cabin. The operating status parameters include rotational speed, load rate, power consumption, vibration acceleration, and operating temperature.

[0034] The computational processing module is pre-loaded with a sound field reconstruction model and adaptive control algorithm based on a physical constraint neural network. It is used to generate active noise cancellation compensation control law and acoustic impedance parameter tuning instructions based on transient sound pressure sequence data, frequency distribution characteristics and operating status parameters.

[0035] An active control execution unit includes a secondary sound source array for emitting canceling sound waves according to an active noise cancellation compensation control law;

[0036] The physical impedance adjustment unit includes a semi-active noise cancellation structure, which is used to adjust the acoustic tuning characteristic parameters of the semi-active noise cancellation structure according to the acoustic impedance parameter tuning command.

[0037] Furthermore, the acoustic feature acquisition module includes a three-dimensional spatially distributed acoustic sensor array; the acoustic sensor array adopts a non-uniform deployment topology, and the distribution density of the acoustic sensor array in the noise source equipment installation area inside the silent cabin is greater than the distribution density of the acoustic sensor array in the central area of ​​the cabin ceiling inside the silent cabin.

[0038] Furthermore, the semi-active noise reduction structure includes a resonant cavity, a telescopic neck, and a drive mechanism; the drive mechanism includes a stepper motor and a lead screw pair, the lead screw pair being connected to the cavity base plate inside the resonant cavity; the stepper motor receives tuning commands for acoustic impedance parameters, and drives the lead screw pair to push the cavity base plate to slide along the axial direction of the resonant cavity, thereby changing the cavity volume of the resonant cavity.

[0039] Furthermore, the drive mechanism is equipped with a current closed-loop feedback controller; the current closed-loop feedback controller monitors the load torque of the stepper motor in real time. When the load torque of the stepper motor increases, the current closed-loop feedback controller increases the drive current of the stepper motor through a proportional-integral-derivative control algorithm to maintain the position adjustment step accuracy of the cavity base plate within the preset accuracy range.

[0040] Furthermore, the inner wall of the silent cabin is covered with a basic sound-absorbing layer made of melamine sponge and damping rubber; the basic sound-absorbing layer absorbs high-frequency sound energy, and the active control execution unit and the physical impedance adjustment unit jointly cancel out mid- and low-frequency noise.

[0041] Furthermore, the sound velocity in the scalar sound wave equation embedded in the sound field reconstruction model based on the physical constraint neural network is configured as a scalar sound velocity field distribution with spatial non-uniform distribution characteristics; the calculation and processing module extracts the operating temperature from the operating status parameters of the noise source equipment, maps the operating temperature to a three-dimensional temperature distribution field inside the silent cabin using a multiple Gaussian heat source diffusion model, and updates the scalar sound velocity field distribution in the scalar sound wave equation according to the following formula:

[0042]

[0043] In the formula, The spatial sound velocity at each coordinate point in a three-dimensional coordinate system, in units of . ; The reference speed of sound at zero degrees Celsius under standard atmospheric pressure, in units of 1000 m / s. ; This represents the Celsius temperature value of the three-dimensional temperature distribution field at the corresponding coordinate point, in units of 1. .

[0044] Furthermore, in the step of generating the active noise cancellation compensation control law based on the residual, the calculation and processing module extracts the phase difference between the driving current and the terminal voltage of the power amplifier module of the secondary sound source array and converts the phase difference into voice coil heat dissipation temperature rise data by combining the resistance temperature coefficient of the voice coil material. A fractional-order lead-lag phase compensation operator is then added to the output stage of the active noise cancellation compensation control law. The transfer function of the fractional-order lead-lag phase compensation operator is approximated by a rational proper fraction and discretized using a bilinear transform method. The discretized digital filter is then directly applied to the digital discrete signal sequence output by the multi-channel adaptive filtering algorithm. The transfer function expression of the fractional-order lead-lag phase compensation operator is:

[0045]

[0046] In the formula, For the transfer function of the fractional-order lead-lag phase compensation operator; For complex frequency domain variables; This is the phase angle compensation gain coefficient, which is linearly positively correlated with the currently estimated voice coil heat dissipation temperature rise data; The time constant is expressed in units of 1. ; Let be the order of a fractional calculus, and satisfy . ; A fractional core term structure to eliminate dimensional conflicts.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] This invention directly embeds the scalar acoustic wave equation as a regularization constraint into the neural network loss function. Through automatic differentiation technology, it achieves a deep integration of physical laws and data-driven approaches. This overcomes the shortcomings of traditional pure data-driven models, which suffer from a sharp drop in prediction accuracy in sparse sensor regions and cannot guarantee physical consistency. It achieves high-precision sound field reconstruction with a sound pressure level error of less than 1.0 dB within the target noise reduction frequency band.

[0049] This invention uses a sound field reconstruction model as a unified feedback source to drive the phase cancellation of the secondary sound source array of the active control execution unit and the impedance tuning of the physical impedance adjustment unit to work in tandem. The secondary sound source array of the active control execution unit performs rapid phase compensation for broadband non-stationary noise, while the semi-active noise cancellation structure physically absorbs narrowband peak noise. The combination of these two technologies overcomes the limitations of single noise cancellation methods in terms of frequency band coverage and dynamic adaptability, constructing a deeply silent environment covering the entire audible frequency range.

[0050] This invention further improves the spatial coverage and energy suppression efficiency of noise reduction by employing a virtual error sensing point mechanism and a time-averaged acoustic energy flow vector field driven control strategy. The virtual error sensing point mechanism utilizes the remote prediction capability of the sound field reconstruction model to expand the spatial noise reduction point coverage without adding physical sensors, and ensures engineering robustness through prediction uncertainty quantification and dynamic confidence weight adjustment. The time-averaged acoustic energy flow vector field driven control strategy derives the sound pressure gradient and particle velocity distribution, calculates the time-averaged acoustic energy flow vector field, and allocates secondary cancellation signal energy based on the time-averaged acoustic energy flow divergence difference. This suppresses sound field energy accumulation from the essence of energy propagation, achieving a uniform silencing effect across the entire domain. Attached Figure Description

[0051] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0052] Figure 1 This is a flowchart of the intelligent sound field modeling and noise reduction method for the silent modular cabin of the present invention.

[0053] Figure 2 This is a flowchart of the virtual error sensing point mechanism of the present invention.

[0054] Figure 3 This is a flowchart of the time-averaged acoustic energy flow vector field energy distribution strategy of the present invention.

[0055] Figure 4 This is a flowchart of the pre-compensation process for the Volterra series compensation model of the present invention.

[0056] Figure 5 A flowchart for adaptive optimization of the closed loop in the invention environment.

[0057] Figure 6 This is one of the system operation interface diagrams when using this invention.

[0058] Figure 7 This is the second diagram of the system operation interface when using this invention. Detailed Implementation

[0059] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the embodiments of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0060] The following is in conjunction with the appendix Figures 1-7The embodiments of the present invention will be described in detail below.

[0061] Example 1: This example discloses the basic architecture and core algorithm logic of the intelligent sound field modeling and anechoic system for a silent modular cabin, as detailed below:

[0062] The intelligent sound field modeling and anechoic system of the silent cabin is integrated into the interior space of the silent cabin. The inner wall of the silent cabin is covered with a basic anechoic layer composed of multiple layers of composite sound-absorbing materials. The basic anechoic layer is made of melamine sponge with a gradient density structure and damping rubber, which absorbs high-frequency sound energy by utilizing the viscous friction mechanism of porous media.

[0063] The silent modular cabin sound field intelligent modeling and anechoic system includes an acoustic feature acquisition module, an equipment status monitoring module, a calculation and processing module, an active control execution unit, and a physical impedance adjustment unit.

[0064] The acoustic feature acquisition module consists of a three-dimensional spatially distributed acoustic sensor array, signal conditioning circuitry, and a multi-channel synchronous sampling unit. The acoustic sensor array employs a non-uniform three-dimensional spatial topology layout. Within the quiet cabin, the deployment density of the acoustic sensor array in the noise source equipment base area, the corners of the air inlet and outlet, and the area surrounding the operator's head movement is increased to more than three times the standard deployment density compared to the central area of ​​the cabin ceiling. Each acoustic sensor unit in the array is connected to the computing module via an industrial Ethernet bus, and a time-sensitive networking protocol ensures data synchronization accuracy between sensor nodes is better than 10 microseconds.

[0065] The equipment status monitoring module directly collects the underlying operating status parameters of the noise source equipment through the data communication bus, including speed, load rate, power consumption, vibration acceleration, and operating temperature.

[0066] The computational processing module adopts an integrated heterogeneous computing platform, pre-loaded with a sound field reconstruction model based on a physically constrained neural network and a multi-channel adaptive filtering algorithm. The network architecture of the sound field reconstruction model based on the physically constrained neural network includes a high-dimensional coordinate encoding layer, multiple nonlinear fully connected hidden layers, and a linear output layer. The high-dimensional coordinate encoding layer maps three-dimensional spatial coordinates and time coordinates to a high-frequency Fourier feature space to enhance the network's ability to capture details of high-frequency spatial fluctuations.

[0067] The loss function of the sound field reconstruction model is a weighted aggregation of the data-driven loss term, the physical equation residual loss term, and the boundary condition constraint loss term. The complete expression of the loss function of the sound field reconstruction model is:

[0068]

[0069] In the formula, To measure sound pressure With predicted sound pressure The mean square error between; the second term is the mean square error across all The mean of the squared residuals of the scalar acoustic wave equation calculated at each computational grid point; the third term is the mean of the squared residuals calculated at all grid points. At each boundary node, the predicted boundary conditions With true boundary conditions The mean square error. , , These are preset weighting coefficients used to balance various losses; their typical value range is... , .

[0070] The specific process by which the computational processing module constructs the residual loss term of the physical equation is as follows: a regular three-dimensional computational grid is constructed inside the silent cabin with a preset spatial resolution;

[0071] After each forward propagation of the network, the partial derivatives of the sound pressure output by the sound field reconstruction model with respect to three-dimensional spatial coordinates and time are calculated using an automatic differentiation mechanism. The residuals are then calculated at each grid point using the scalar acoustic wave equation. Finally, the sum of the squares of the residuals from all grid points is averaged. This mechanism forces the network to output a sound pressure level distribution that conforms to the laws of acoustic wave propagation even in spatial regions without physical measurement support.

[0072] The active control execution unit consists of a multi-channel signal generator, a power amplifier module, and a secondary sound source array. The secondary sound source array is composed of multiple long-stroke electroacoustic transducers, which are evenly arranged in the internal frame structure of the silent cabin.

[0073] The physical impedance adjustment unit includes a semi-active anechoic structure and a drive mechanism distributed on the walls of the silent cabin. The semi-active anechoic structure is an acoustic resonance absorption device with a tunable resonant frequency, and its acoustic impedance parameters can be adjusted in response to control commands output by the computational processing module.

[0074] In practical implementation:

[0075] The acoustic feature acquisition module acquires transient sound pressure sequence data and frequency distribution characteristics of multiple measurement points in the silent cabin in real time through an acoustic sensor array, while the equipment status monitoring module collects the operating status parameters of the noise source equipment simultaneously.

[0076] The computational processing module uses the transient sound pressure sequence data as the direct input feature of the sound field reconstruction model based on the physical constraint neural network; and uses the operating state parameters as physical constraint adjustment factors to dynamically adjust the physical constraint boundary conditions in the sound field reconstruction model according to the operating state parameters before model inference.

[0077] In this model, the rotational speed parameter of the noise source device is used to calculate the device's fundamental frequency and harmonic components, thereby updating the frequency-domain focusing regularization term embedded in the loss function of the sound field reconstruction model in real time. The load rate and power consumption parameters of the noise source device are mapped to the volume velocity amplitude of the sound source, used to update the non-homogeneous Neumann boundary conditions of the sound field reconstruction model on the mesh nodes of the noise source device surface. This sound field reconstruction model is driven by the boundary conditions updated by the operating state parameters as the physics driver and by transient sound pressure sequence data as the data driver, outputting a continuous three-dimensional sound pressure level distribution throughout the entire interior of the silent cabin.

[0078] The calculation and processing module compares the sound pressure level distribution with the target silence threshold point by point to calculate the residual matrix. Based on the residual matrix, it dynamically generates an active noise cancellation compensation control law and simultaneously generates tuning instructions for the acoustic impedance parameters of the semi-active noise cancellation structure.

[0079] The active noise cancellation compensation control law employs a multi-channel adaptive filtering algorithm. The computational processing module identifies the physical secondary channel transfer function between each electroacoustic transducer in the secondary sound source array and the acoustic sensor unit, which serves as an error sensor. The adaptive filtering algorithm uses a variable convergence factor. When the gradient of the error signal energy change exceeds a preset threshold, the variable convergence factor increases accordingly to improve the dynamic tracking speed; when the error signal energy tends to stabilize, the variable convergence factor decreases accordingly to reduce steady-state misalignment.

[0080] The active control execution unit drives the secondary sound source array to emit canceling sound waves that are out of phase with the current noise signal, according to the active noise cancellation compensation control law. The physical impedance adjustment unit drives the semi-active noise cancellation structure to change its acoustic tuning characteristic parameters according to the acoustic impedance parameter tuning command. The tuning frequency of the semi-active noise cancellation structure is determined by the following formula:

[0081]

[0082] In the formula, The tuning frequency is expressed in units of 1000 ppm. ; Speed ​​of sound, unit: ; The cross-sectional area of ​​the neck is given in units of... ; The volume of the cavity is expressed in units of 1000 liters. ; Neck length, in units of ; This is the terminal correction term, in units of .

[0083] The phase cancellation of the secondary sound source array in the active control actuator works in conjunction with the impedance tuning of the physical impedance adjustment unit. The active control actuator mainly performs real-time phase compensation for broadband non-stationary noise, while the physical impedance adjustment unit mainly performs physical absorption for narrowband peak noise. Together, they achieve a deep silencing effect across the entire frequency band. A closed-loop feedback mechanism continuously monitors residual fluctuations, and the acoustic sensor array continuously collects sound field data after noise cancellation and feeds it back to the sound field reconstruction model to update the prediction results, thus achieving full-frequency adaptive closed-loop control.

[0084] The silent cabin sound field intelligent modeling and anechoic system in this embodiment also features an environment-adaptive optimization closed loop. This closed loop continuously compares the sound field reconstruction model output before and after anechoic ...

[0085] The computational processing module utilizes a policy gradient-based reinforcement learning algorithm, with the joint reward function being the maximization of noise reduction efficiency and the minimization of system power consumption. ,in This represents the predicted sound pressure level difference before and after noise reduction. The total power consumption of the system is and The weighting coefficients are preset, and the weight parameters of the sound field reconstruction model and the control parameters of the adaptive control algorithm are updated periodically. The environmental adaptive optimization closed loop ensures that the system maintains the optimal noise reduction state throughout its entire life cycle.

[0086] Example 2: Based on Example 1, Example 2 specifies the electromechanical actuation details of the semi-active noise reduction structure and optimizes the nonlinear distortion compensation mechanism of the secondary sound source array under high-intensity output.

[0087] In this embodiment, the semi-active anechoic structure consists of a resonant cavity, a telescopic neck, and a drive mechanism. The resonant cavity is a sealed cylindrical structure with an adjustable volume, and the telescopic neck connects the interior of the resonant cavity with the interior space of the soundproof cabin. The drive mechanism includes a stepper motor, a lead screw pair, and a cavity base plate. The cavity base plate is disposed inside the resonant cavity and can slide along the inner wall of the resonant cavity. According to the tuning command of the acoustic impedance parameters output by the calculation and processing module, the stepper motor drives the cavity base plate to move axially along the resonant cavity through the lead screw pair, thereby changing the effective cavity volume of the resonant cavity.

[0088] The drive mechanism incorporates a current closed-loop feedback controller. This controller employs a proportional-integral-derivative (PI-DE) control algorithm to monitor changes in the stepper motor's load torque in real time and adjust the drive current accordingly. When changes in air humidity within the silent cabin increase the internal frictional damping of the resonant cavity, leading to a rise in load torque, the controller ensures that the position adjustment step accuracy of the cavity's base plate remains within 0.02mm, corresponding to a tuning frequency deviation controlled within ±0.5Hz.

[0089] In alternative implementations of the semi-active noise reduction structure, the drive mechanism may also employ a linear motor drive, a piezoelectric ceramic drive, or a magnetostrictive drive. The object adjusted by the drive mechanism is not limited to the cavity volume of the resonant cavity, but may also include the neck cross-sectional area or neck length of the telescopic neck.

[0090] To address the nonlinear intermodulation distortion problem caused by electroacoustic transducers in secondary sound source arrays under high sound pressure level and large displacement conditions, Example 2 embeds a Volterra series compensation model before the output stage of the active noise cancellation compensation control law. The Volterra series compensation model utilizes a high-order polynomial convolutional network to inversely model and pre-compensate the distortion of the dynamic displacement-electromagnetic force nonlinear response curve of the electroacoustic transducer. The output signal of the Volterra series compensation model... Represented as:

[0091]

[0092] In the formula, The sequence of control signals output by the adaptive filtering algorithm; For the first Volterra kernel function of order; The highest order of the Volterra series expansion; This represents the memory length of each kernel function. For discrete-time indexing; For the first One delay variable. The Volterra series compensation model is connected in series between the output of the adaptive filtering algorithm and the input of the power amplifier module, which effectively suppresses the influence of nonlinear distortion of the electroacoustic transducer on the compensation accuracy in high sound pressure level noise environments.

[0093] The Volterra kernel function The following steps are obtained through an offline identification process: the electroacoustic transducer in the secondary sound source array is excited by a white noise signal, the input voltage signal and output sound pressure signal of the electroacoustic transducer are collected simultaneously, the kernel function coefficients of each order are estimated by an adaptive identification algorithm based on the minimum mean square error criterion, and the identified kernel function coefficients are stored in the calculation and processing module.

[0094] The other technical solutions in this embodiment are the same as those in Embodiment 1, and will not be repeated here.

[0095] Example 3: Based on Example 2, Example 3 introduces a Bayesian uncertainty management logic based on a virtual error sensing point mechanism and establishes a global energy distribution control strategy based on the divergence of the time-averaged acoustic energy flow vector field.

[0096] In the actual operation of a silent modular shelter, it is inconvenient to deploy a large number of physical acoustic sensors in the area where the operator's head moves, but this area is precisely where noise reduction is most urgently needed. Example 3 utilizes the interpolation prediction capability of a sound field reconstruction model based on a physically constrained neural network to project and generate a dense matrix of virtual error sensing points within the envelope area of ​​the operator's head movement.

[0097] In practical implementation:

[0098] For each virtual error sensing point, the calculation and processing module obtains the predicted sound pressure value at that virtual error sensing point. A virtual error signal is generated by comparing the predicted sound pressure value with the expected sound pressure value. Virtual error signal The signal needs to be filtered using a virtual secondary channel transfer function model. This virtual secondary channel transfer function is calculated using the sound field reconstruction model based on a physically constrained neural network, under the condition that a preset excitation signal is applied by the active control execution unit, to obtain the acoustic transfer function from the secondary sound source to the virtual error sensing point. The filtered virtual error signal and the physical error signal collected and filtered by the physical acoustic sensor are jointly input into the weight coefficient update equation of the adaptive filtering algorithm, jointly driving the optimization direction of the filter.

[0099] To enhance system robustness, Example 3 includes a continuous monitoring mechanism for prediction uncertainty based on Bayesian inference. The computational processing module approximates the prediction variance of the sound field reconstruction model's output distribution at each virtual error sensing point using the Monte Carlo random deactivation method. The prediction variance serves as an indicator of prediction uncertainty.

[0100] When the prediction uncertainty index of a certain virtual error sensing point exceeds a preset uncertainty threshold, the calculation and processing module reduces the weight coefficient of the virtual error signal in the total error cost function according to a preset linear attenuation function. The preset linear attenuation function is in the form of:

[0101] ;

[0102] in To predict uncertainty, To preset an uncertainty threshold, The preset attenuation coefficient is used. At this point, the update control of the adaptive filtering algorithm smoothly transfers to the physical error signal, preventing virtual node prediction divergence from causing the control system to malfunction. When the prediction uncertainty index falls below the preset uncertainty threshold, the weight coefficients of the virtual error signal slowly recover according to an inverse linear function.

[0103] In the system cascaded control logic, Example 3 establishes a three-dimensional spatial energy diversion strategy based on the divergence of the time-averaged acoustic energy flow vector field. The computational processing module maps the transient sound pressure field output by the sound field reconstruction model to the frequency domain via a fast Fourier transform to obtain the complex sound pressure distribution. Based on the complex sound pressure distribution, the particle velocity distribution of the three-dimensional mesh nodes inside the silent cabin is derived and solved.

[0104]

[0105] In the formula, The frequency domain particle velocity vector, in units of ; The imaginary unit; Angular frequency, unit: ; air density, unit: ; It is the frequency domain complex sound pressure gradient vector.

[0106] The computational processing module further processes the complex sound pressure at each spatial coordinate point in the reconstructed sound field. With particle velocity vector Calculations were performed to obtain the time-averaged acoustic energy flow vector field. ,in This is the time-averaged acoustic energy flow vector, in units of , Indicates taking the real part, Indicates complex conjugation.

[0107] The calculation and processing module performs divergence calculations on the vector field. Regions with significantly positive divergence values ​​correspond to the sound source region or standing wave antinode region with net acoustic energy outflow. The calculation and processing module then calculates the time-averaged acoustic energy flow divergence for each spatial region. Calculate the energy distribution weights of the secondary cancellation signal The formula for calculating energy allocation weights is as follows: Normalization is performed on all regions with positive divergence or that meet the threshold condition.

[0108] Accordingly, the output power distribution of each channel of the secondary sound source array in the active control execution unit is adjusted, prioritizing the concentration of canceling energy onto the sound energy accumulation area. This strategy suppresses sound field energy accumulation from the perspective of energy propagation, achieving a uniform and quiet noise reduction effect across the entire domain.

[0109] In specific implementation, the complete technical solution of this invention is adopted, namely, a sound field reconstruction model based on a physically constrained neural network, coordinated control of active and semi-active noise cancellation, a time-averaged acoustic energy flow vector field energy distribution strategy, and fusion of equipment operating status data. In a volume of... Inside the silent cabin, a three-dimensional acoustic sensor array composed of 64 acoustic sensor units is deployed, along with an electroacoustic transducer with 8 secondary sound source arrays and 12 semi-active noise reduction structures. The noise source equipment uses rated power... Centrifugal fans generate complex low-to-mid-frequency noise, including fundamental and harmonic components, under rated operating conditions. The system sampling frequency is set to... The sound field reconstruction model based on physically constrained neural networks adopts a 4-layer fully connected network structure, and the weights of the loss function of the sound field reconstruction model are set to... The drive mechanism of the semi-active noise reduction structure adopts current closed-loop feedback control, and the tuning frequency resolution reaches [value missing]. .

[0110] The experimental ambient temperature was The initial cabin noise level was Before the experiment, in a silent cabin without any noise reduction measures, only the noise source equipment was turned on and operated stably at its rated speed. The original noise level at each frequency was measured using 64 acoustic sensors. Subsequently, the systems of the comparative and example embodiments were deployed and started, and measurements were taken after noise reduction once the systems were running stably. All sensors were used for error sensing and sound field reconstruction.

[0111] Comparative Example 1: Comparative Example 1 uses the same number and layout of 64-channel acoustic sensors and 8-channel secondary sound source array as the embodiment of the present invention, but uses a purely data-driven, unconstrained neural network for sound field reconstruction, and adopts a standard multi-channel filtering-x least mean square algorithm without energy distribution strategy for active control, and does not integrate a semi-active noise cancellation structure.

[0112] Table 1. Comparison of noise reduction effects between Comparative Example 1 and Example 3;

[0113]

[0114] As shown in Table 1, in the key low-mid frequency band from 50Hz to 500Hz, Example 3 achieved a higher noise reduction gain compared to Comparative Example 1 with the same hardware configuration. The noise reduction was increased by 9.7dB at the 50Hz fundamental frequency, by 9.1dB at the 100Hz harmonic frequency, and by 9.7dB at the 250Hz structural resonant frequency. The overall sound pressure level was reduced by 10.3dB compared to the comparative example. These data demonstrate that the present invention, through a sound field reconstruction model based on a physically constrained neural network, coordinated control of active and semi-active noise cancellation, and time-averaged acoustic energy flow vector field energy distribution, brings incremental noise reduction contributions to complex systems.

[0115] Furthermore, in the test of sudden change in operating conditions, when the rotation speed of the noise source equipment changes suddenly within 2 seconds, causing the main frequency of the noise to drift, the comparative noise reduction system takes more than 5 seconds to reconverge and the noise level fluctuates significantly during the convergence process; the embodiment relies on the sound field reconstruction model based on physical constraint neural network to quickly predict the evolution of the sound field and dynamically adjust the variable convergence factor, and completes convergence within 0.3 seconds, and the noise level fluctuation amplitude during the convergence process does not exceed 2dB.

[0116] In the long-term operational stability test, after 168 hours of continuous operation, the noise reduction of Comparative Example 1 decreased by about 8dB from the initial value due to the aging of the sound-absorbing material due to moisture and the temperature drift of the electronic components; while the environmental adaptive optimization closed loop of Example 3 continuously compensated for the changes in acoustic characteristics caused by environmental factors, and the fluctuation range of the noise reduction index was maintained within ±1dB of the initial value.

[0117] Example 4: Based on Example 3 above, Example 4 introduces a thermoacoustic coupling physical constraint mechanism and a fractional-order phase compensation algorithm to address the drastic gradient changes in the spatial temperature field caused by the start-up and shutdown of high-density power equipment in the silent cabin, as well as the problem of voice coil thermal compression phase shift caused by the continuous output of high sound pressure level of the secondary sound source array.

[0118] In practical implementation:

[0119] The enclosed space inside the silent modular cabin is affected by the large amount of irregular heat emitted by the equipment, resulting in temperature stratification and uneven diffusion of the air medium. The equipment status monitoring module reads the operating temperature and power consumption parameters of each noise source device via a data communication bus at a sampling rate of no less than 50Hz. After receiving the underlying data, the calculation and processing module substitutes it into a multi-Gaussian heat source diffusion model. The multi-Gaussian heat source diffusion model constructs independent Gaussian thermodynamic diffusion cores based on the known three-dimensional geometric installation coordinates of each noise source device. These cores are then spatially linearly superimposed with the convective heat dissipation boundary conditions of the cabin's ventilation system to output a high spatial resolution three-dimensional temperature distribution field.

[0120] The computational processing module extracts the gridded Celsius temperature values ​​of each spatial node in the three-dimensional temperature distribution field. And combined with the zero-degree Celsius standard reference speed of sound Calculate the scalar field distribution of sound velocity with spatially non-uniform characteristics. The distribution of the sound velocity scalar field with spatial analytical properties is determined by the following equation:

[0121]

[0122] In the formula, The spatial sound velocity at each coordinate point in a three-dimensional coordinate system, in units of . ; The reference speed of sound at zero degrees Celsius under standard atmospheric pressure, in units of 1000 m / s. ; This represents the Celsius temperature value of the three-dimensional temperature distribution field at the corresponding coordinate point, in units of 1. .

[0123] The computational processing module dynamically injects the scalar field distribution of sound velocity into the sound field reconstruction model based on a physically constrained neural network, directly replacing the global constant sound velocity term in the original scalar sound wave equation.

[0124] During the forward propagation of network training and inference, the automatic differential engine calculates the output sound pressure. Second-order Laplace operator for spatial variables and the second derivative with respect to time When calculating the residual loss term of the physical equations, the partial derivatives at the grid nodes are strictly related to the local spatial sound velocity at that coordinate point. Matching multiplication. This mechanism forces the reconstructed sound field to still fit the true wavefront refraction and phase distortion phenomena in spatial regions experiencing severe temperature gradients.

[0125] When the secondary sound source array is subjected to high-intensity steady-state or narrowband noise for a long time, the resistive heat dissipation of the voice coil inside the electroacoustic transducer leads to an increase in effective resistance and a change in quality factor, which in turn induces dynamic phase hysteresis and disrupts the destructive interference conditions of the active control execution unit.

[0126] The calculation and processing module extracts the drive voltage and feedback current of the power amplifier module in real time, calculates the phase difference between the drive voltage and feedback current, and converts the phase difference into voice coil heat dissipation temperature rise data by combining the resistance temperature coefficient of the voice coil material. The output stage of the active noise cancellation compensation control law is pre-configured with a fractional-order lead-lag phase compensation operator. The transfer function of the fractional-order lead-lag phase compensation operator in the complex frequency domain is:

[0127]

[0128] In the formula, For the transfer function of the fractional-order lead-lag phase compensation operator; For complex frequency domain variables; The phase angle compensation gain coefficient is linearly positively correlated with the currently estimated voice coil heat dissipation temperature rise data. The time constant is expressed in units of 1. The setting is based on the thermal time constant calibration value of the electroacoustic transducer in the secondary sound source array; Let be the order of a fractional calculus, and satisfy . ; A fractional core term structure to eliminate dimensional conflicts.

[0129] To enable the physical deployment of continuous operators on a heterogeneous digital platform for computational processing modules, the system employs a modified Oustaloup approximation algorithm for the transfer function. Perform rational proper fraction approximation.

[0130] In the set noise reduction target frequency band Inside, the fractional calculus operator is decomposed into The cascade of integer-order zero-pole models is expressed as:

[0131]

[0132] In the formula, For the first Approximate zero-point frequency, unit: ; For the first Approximate pole frequency of order 1, in units of ; As approximate order parameters, based on the target frequency band Select the required fitting accuracy for the frequency response. The typical value range is 2 to 5; zero frequency With pole frequency Based on the parameters in the original transfer function , , The result is obtained through a modified Oustaloup recursive formula. The system employs a bilinear transformation method to map the approximate transfer function in the Laplace domain to the discrete domain. The domain generates the coefficients of the difference equations for the digital filter and directly applies them to the digital discrete signal sequence output by the multi-channel adaptive filtering algorithm.

[0133] The drive signal, fed forward by a digital fractional-order lead-lag operator, provides a wider bandwidth and smoother lead phase injection compared to conventional integer-order filters, effectively neutralizing the thermal hysteresis generated at the physical device level.

[0134] Under extreme test conditions where the ambient temperature rises at a rate of 20°C / h and exhibits a non-uniform thermal distribution due to continuous operation of the fully loaded equipment, the phase tracking error of the conventional physical constraint model changes significantly with the temperature gradient. In this embodiment, the system maintains high control accuracy across the entire frequency band by relying on a spatiotemporal heterogeneous sound velocity field and digital fractional-order phase compensation. Specific test data comparisons are shown in Table 2.

[0135] Table 2. Comparison of noise reduction performance under extreme thermoacoustic coupling conditions;

[0136]

[0137] Test results show that the phase tracking error of the system is stably compressed to within 5.5° at 500Hz, and the system's steady-state high-load noise reduction jumps from 11.4dB and remains at 18.2dB. These data verify the engineering robustness and feasibility of the thermoacoustic coupling mechanism and fractional-order phase compensation algorithm under harsh temperature-changing conditions.

[0138] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0139] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent sound field modeling and noise reduction in a silent modular cabin, characterized in that, Includes the following steps: Step 1: Acoustic sensor arrays deployed in the three-dimensional space inside the silent cabin are used to acquire transient sound pressure sequence data and frequency distribution characteristics of multiple measurement points inside the silent cabin in real time. Simultaneously, the operating status parameters of noise source equipment inside the silent cabin are collected. The operating status parameters include rotational speed, load rate, power consumption, vibration acceleration and operating temperature. Step 2: Using the transient sound pressure sequence data as direct input features, the operating status parameters of the noise source device are parsed into dynamic adjustment parameters of the sound source excitation boundary conditions and frequency domain focusing regularization term parameters; The rotational speed parameter of the noise source device is used to calculate the device's fundamental frequency and harmonic components to update the frequency domain focusing regularization term in real time. The load rate and power consumption parameters of the noise source device are mapped to the sound source volume velocity amplitude to update the boundary conditions of the sound field reconstruction model. The processed data is input into a pre-trained sound field reconstruction model based on a physical constraint neural network. The boundary conditions updated by the operating state parameters are used as the physical driver, and the transient sound pressure sequence data is used as the data driver to predict the sound pressure level distribution and phase distribution inside the silent cabin. Step 3: Calculate the residual between the sound pressure level distribution and the target silence threshold, generate an active noise cancellation compensation control law based on the residual, and simultaneously generate a tuning command for the acoustic impedance parameters of the semi-active noise cancellation structure. Step 4: Drive the secondary sound source array to emit canceling sound waves according to the active noise cancellation compensation control law, and drive the semi-active noise cancellation structure to adjust the acoustic tuning characteristic parameters according to the acoustic impedance parameter tuning command.

2. The method for intelligent sound field modeling and noise reduction of a silent modular cabin according to claim 1, characterized in that, The sound field reconstruction model based on physically constrained neural networks embeds acoustic physics equations as regularization constraints into the loss function. These acoustic physics equations include scalar sound wave equations. In the formula, Sound pressure level, unit: ; Speed ​​of sound, unit: ; For the Laplace operator; For time, the unit is .

3. The method for intelligent sound field modeling and noise reduction of a silent modular cabin according to claim 1, characterized in that, The acoustic tuning characteristic parameters include the tuning frequency of the semi-active noise cancellation structure, which is calculated using the following formula: In the formula, The tuning frequency is expressed in units of 1000 ppm. ; Speed ​​of sound, unit: ; The neck cross-sectional area of ​​the semi-active noise reduction structure is given in units of... ; The volume of the cavity in the semi-active noise reduction structure is expressed in units of... ; The length of the neck of the semi-active noise reduction structure, in units of ; This is the terminal correction term, in units of .

4. The method for intelligent sound field modeling and noise reduction of a silent modular cabin according to claim 1, characterized in that, In the step of generating the active noise cancellation compensation control law based on the residual, the transient sound pressure field output by the sound field reconstruction model based on the physical constraint neural network is... The complex sound pressure distribution is obtained by mapping the data to the frequency domain using a fast Fourier transform. The particle velocity distribution in each spatial region within the silent container is derived and calculated using the complex sound pressure distribution. The time-averaged acoustic energy flow vector field distribution is then calculated based on the following formula: In the formula, The frequency domain particle velocity vector, in units of ; The imaginary unit; Angular frequency, unit: ; air density, unit: ; It is a complex sound pressure gradient vector in the frequency domain; This is the time-averaged acoustic energy flow vector, in units of ; Indicates taking the real part; Indicates complex conjugation; Calculate the time-averaged acoustic energy flow divergence of each spatial region in the time-averaged acoustic energy flow vector field distribution. For regions where the time-averaged acoustic energy flow divergence is greater than a preset divergence threshold, increase the secondary cancellation signal energy allocation weight of the secondary sound source array to that region.

5. The method for intelligent sound field modeling and noise reduction of a silent modular cabin according to claim 1, characterized in that, A sound field reconstruction model based on physical constraints neural networks predicts sound pressure at spatial coordinates of un-deployed acoustic sensors within a silent cabin, establishing virtual error sensing points. The predicted sound pressure values ​​output from these virtual error sensing points are then filtered using a virtual secondary channel transfer function model obtained through pre-identification or calculation using the sound field reconstruction model and the free-field Green's function to generate virtual error signals. These virtual error signals, along with the physical error signals collected by the physical acoustic sensors, participate in the iterative update of the weight coefficients of the active noise cancellation compensation control law.

6. The method for intelligent sound field modeling and noise reduction of a silent modular cabin according to claim 5, characterized in that, The prediction variance of the sound field reconstruction model based on physical constraint neural network at the virtual error sensing point is extracted as the prediction uncertainty index. When the prediction uncertainty index exceeds the preset uncertainty threshold, the weight coefficient of the virtual error signal in the total error cost function of the active noise compensation control law is reduced according to the preset linear decay function. The preset linear decay function is: , in To predict uncertainty, To preset an uncertainty threshold, The function is determined by a preset attenuation coefficient, which is exited by gradually decreasing the amount of prediction uncertainty as it increases.

7. The method for intelligent sound field modeling and noise reduction of a silent modular cabin according to claim 1, characterized in that, The active noise cancellation compensation control law includes a Volterra series compensation model for pre-compensating the nonlinear distortion of the secondary sound source array; the Volterra series compensation model uses the control signal sequence output by the adaptive filtering algorithm as the input signal sequence. The expression for the output signal of the Volterra series compensation model is: In the formula, Output signal for Volterra series compensation model; The sequence of control signals output by the adaptive filtering algorithm; For the first Volterra kernel function of order; The highest order of the Volterra series expansion; This represents the memory length of each kernel function. For discrete-time indexing; For the first One delayed variable.

8. The method for intelligent sound field modeling and noise reduction of a silent modular cabin according to claim 1, characterized in that, The predicted sound pressure level difference before and after anechoic chambering and the total power consumption of the sound field intelligent modeling and anechoic chamber system are continuously calculated through environmental adaptive optimization closed loop. A joint reward function is constructed using the predicted sound pressure level difference and the total power consumption of the system through reinforcement learning algorithm. The network weight parameters of the sound field reconstruction model based on physical constraint neural network are updated according to the joint reward function.

9. A sound field intelligent modeling and anechoic system for a silent modular cabin, used to implement the sound field intelligent modeling and anechoic method for a silent modular cabin as described in any one of claims 1 to 8, characterized in that, include: The acoustic feature acquisition module is used to acquire transient sound pressure sequence data and frequency distribution characteristics of multiple measurement points in the silent cabin; The equipment status monitoring module is used to collect operating status parameters of noise source equipment in the silent cabin, including rotational speed, load rate, power consumption, vibration acceleration, and operating temperature. The calculation and processing module is pre-loaded with a sound field reconstruction model and adaptive control algorithm based on a physically constrained neural network, used to generate active noise cancellation compensation control laws and acoustic impedance parameter tuning commands based on transient sound pressure sequence data, frequency distribution characteristics, and operating status parameters. The active control execution unit includes a secondary sound source array, used to emit canceling sound waves according to the active noise cancellation compensation control law. The physical impedance adjustment unit includes a semi-active noise cancellation structure, used to adjust the acoustic tuning characteristic parameters of the semi-active noise cancellation structure according to the acoustic impedance parameter tuning commands.

10. The intelligent sound field modeling and anechoic system for a silent modular cabin according to claim 9, characterized in that, The acoustic feature acquisition module includes a three-dimensional spatially distributed acoustic sensor array. The acoustic sensor array adopts a non-uniform deployment topology, and the distribution density of the acoustic sensor array in the noise source equipment installation area inside the silent cabin is greater than the distribution density of the acoustic sensor array in the central area of ​​the cabin ceiling inside the silent cabin.

Images