A method for identification and evaluation of noise pollution sources

Abstract

The application discloses a kind of identification and evaluation method for noise pollution source, specifically related to environmental acoustic monitoring and intelligent signal processing technical field;Multi-channel sound pressure signal and structure vibration response signal in target area are obtained, multi-dimensional acoustic feature matrix is constructed, and sound-structure coupling energy density field is generated;Energy aggregation coefficient is calculated, directed weighted propagation diagram is constructed, coupling path of sound energy in air propagation channel and structure conduction channel is deduced, and propagation attenuation factor and chain transition probability are obtained;Candidate noise source evolution trajectory is generated by space-time folding mapping, stability index, mutation frequency index and structure resonance matching coefficient are calculated, finally nonlinear coupling operation is carried out to propagation attenuation factor, chain transition probability and structure resonance matching coefficient, noise risk coupling strength distribution is formed and risk grade evaluation result is output;The present application can realize the accurate identification and risk quantization evaluation of micro-scale, hidden type noise pollution source.

Description

Technical Field

[0001] This invention relates to the field of environmental acoustic monitoring and intelligent signal processing technology, specifically to a method for identifying and evaluating noise pollution sources. Background Technology

[0002] In high-density urban built-up areas, underground utility tunnels, precision manufacturing cleanrooms, and high-rise building curtain wall structures, there exists a type of concealed noise pollution source characterized by low sound power, narrow frequency band, intermittency, microscale, and structural coupling propagation. Examples include pressure relief pulses from microcracks in pipelines, abnormal cavitation sounds from vacuum pumps, high-frequency howling caused by wind from loose components, and high-frequency structurally conducted noise induced by localized failures in vibration damping systems. While this type of noise typically does not reach environmental standard limits at the sound level, its energy diffuses non-uniformly through structural paths and generates localized accumulation effects. Long-term exposure may lead to equipment fatigue failure, structural loosening and expansion, or hidden health risks. Existing noise identification methods... The current method mainly relies on sound pressure level statistics or single spectral feature matching, lacking the ability to couple and model the relationship between sound field characteristics and structural topology. It is difficult to distinguish the contribution ratio of airborne sound and structure-conducted sound in environments with strong reverberation, small scale, and dual-channel propagation. It also cannot dynamically quantify the formation mechanism, evolution path, and risk coupling degree of abnormal sound events, resulting in insufficient sensitivity in identifying early weak abnormal noise sources and lagging risk assessment. Therefore, it is urgent to construct a noise pollution source identification and evaluation method that integrates multi-domain acoustic characteristics, structural propagation topology, and energy coupling deduction mechanism to achieve accurate location and quantitative risk assessment of concealed microscale noise sources. Summary of the Invention

[0003] The purpose of this invention is to provide a method for identifying and evaluating noise pollution sources, in order to address the shortcomings of the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for identifying and evaluating noise pollution sources, comprising: Acquire multi-channel sound pressure signals and corresponding structural vibration response signals within the target area, perform time-frequency decomposition on the sound pressure signals, extract instantaneous spectral density, phase difference gradient and higher-order spectral entropy parameters, and construct a multi-dimensional acoustic feature matrix; The multidimensional acoustic feature matrix and the structural vibration response signal are fused across modes to generate an acoustic-structure coupled energy density field, and the energy concentration coefficient of each spatial sampling unit is calculated. A sound energy propagation topology network is constructed based on the energy concentration coefficient, and the network edges are weighted according to the energy gradient direction to form a directed weighted propagation graph; Based on the directed weighted propagation graph, the coupling path of sound energy in the air propagation channel and the structural conduction channel is deduced, and the propagation attenuation factor and the chain transition probability between nodes are calculated. Based on the chain transition probability, the multidimensional acoustic feature matrix is ​​spatiotemporally folded and mapped to establish the coupling relationship between the duration of the abnormal sound event and the spatial offset, and the evolution trajectory of the candidate noise source is generated. Based on the evolution trajectory of the candidate noise sources, the stability index, mutation frequency index, and structural resonance matching coefficient are calculated to identify the location and type of noise pollution sources. The propagation attenuation factor, chain transition probability, and structural resonance matching coefficient are subjected to nonlinear coupling calculation to generate a noise risk coupling intensity distribution and output the corresponding risk level evaluation results.

[0005] Preferably, the process of constructing the multidimensional acoustic feature matrix includes: The multi-channel sound pressure signal is segmented using an adaptive window function to obtain equal-duration signal subsequences; Continuous wavelet transform and bispectral joint analysis are performed on each of the signal subsequences to calculate the corresponding instantaneous spectral density and higher-order spectral entropy parameters; A phase difference gradient distribution map is constructed based on the time delay difference results between different channels, and the phase difference gradient distribution map is reconstructed by spatial interpolation to generate a phase difference gradient matrix in a unified spatial coordinate system. The instantaneous spectral density, higher-order spectral entropy parameters, and phase difference gradient matrix are aligned according to timestamps and then tensor-concatenated to form a multidimensional acoustic feature matrix with spatiotemporal correspondence.

[0006] Preferably, the process of generating the acoustic-structure coupled energy density field includes: The vibration response signal of the structure is subjected to frequency decomposition processing, and synchronous segmentation is performed using the same timestamp as the multidimensional acoustic feature matrix. The vibration energy spectral density vector corresponding to each time slice is calculated. The vibration energy spectral density vector is matched with the instantaneous spectral density in the multidimensional acoustic feature matrix in terms of frequency dimension to construct a sound-vibration frequency correspondence mapping matrix, and the initial coupling energy value is calculated based on the frequency amplitude product. The spatial coupling weight coefficient is calculated based on the initial coupling energy value and the phase difference gradient matrix. The spatial coupling weight coefficient is obtained by weighted superposition of the normalized result of the absolute value of the phase difference gradient and the structural vibration amplitude. The initial coupling energy value is multiplied point by point with the spatial coupling weight coefficient, and reconstructed into a three-dimensional energy distribution tensor according to the spatial coordinates to generate an acoustic-structure coupling energy density field with spatiotemporal correspondence.

[0007] Preferably, the process of calculating the energy concentration coefficient of each spatial sampling unit includes: The acoustic-structure coupled energy density field is divided into several spatial sampling units according to a unified spatial coordinate system, and the total coupled energy value of each spatial sampling unit in the same time slice is extracted to form a spatial energy distribution sequence. Taking the current spatial sampling unit as the center, select its neighboring spatial sampling units in three-dimensional space that are less than a preset radius r, calculate the average value of the total coupling energy of the neighboring spatial sampling units, and obtain the local background energy reference value. An energy deviation is constructed based on the total coupled energy value and the local background energy reference value. The energy deviation is defined as the ratio of the difference between the total coupled energy value and the local background energy reference value to the local background energy reference value. The energy deviation is weighted and superimposed with the rate of change of coupling energy of the current spatial sampling unit in a continuous time slice to obtain the energy concentration coefficient of each spatial sampling unit, which is used to represent the degree of abnormal concentration of acoustic-structural coupling energy in space.

[0008] Preferably, the process of forming the directed weighted propagation graph includes: Using a unified spatial coordinate system, each spatial sampling unit is taken as a network node, and the corresponding energy concentration coefficient is used as the node attribute value to construct an initial node set. An adjacency determination matrix is ​​constructed based on the Euclidean distance between any two spatial sampling units. When the distance is less than the preset connection threshold d0, an undirected connection relationship is established to form an initial acoustic energy propagation topology network. The energy gradient value is calculated based on the difference in energy concentration coefficients between adjacent spatial sampling units, and the direction of the edge is determined by pointing from the spatial sampling unit with a high energy concentration coefficient to the spatial sampling unit with a low energy concentration coefficient. The energy gradient value is multiplied by the reciprocal of the corresponding spatial distance to obtain the edge weight. This weight is then assigned to each connected edge to generate a directed weighted propagation graph with directional and weight attributes.

[0009] Preferably, the process of calculating the propagation attenuation factor and the chain transition probability between nodes includes: Based on the edge weights of each edge in the directed weighted propagation graph, a path cost function is constructed, and a minimum cost path search method is used to extract candidate propagation paths for air propagation channels and structural conduction channels respectively. The air propagation channels are selected based on the principle of spatial distance weight dominance, and the structural conduction channels are selected based on the principle of energy concentration coefficient continuity dominance. For each candidate propagation path, a propagation attenuation model is constructed based on the product of the weights of each edge on the path, and a path length correction factor is introduced to perform exponential attenuation calculation on the edge weights to obtain the propagation attenuation factor between nodes. Based on the propagation attenuation factor, a node transition probability matrix is ​​constructed, and the outgoing edge propagation attenuation factor of each node is normalized to form the chain transition probability between nodes. The chain transition probabilities corresponding to the air propagation channel and the structural conduction channel are weighted and fused to obtain the acoustic energy coupling propagation path and its chain transition probability results that reflect the dual-channel coupling characteristics.

[0010] Preferably, the process of generating the evolution trajectory of the candidate noise source includes: Based on the chain transition probabilities, a time recursive weight matrix is ​​constructed, and the feature vectors of each time slice in the multidimensional acoustic feature matrix are weighted and accumulated in time order to form a folded time feature sequence. In the folded time feature sequence, time segments that continuously exceed a preset energy threshold are identified, and the length of the time segment is defined as the duration of the abnormal sound event. The spatial expected coordinates corresponding to each time slice are calculated based on the chain transition probability matrix. The spatial offset of the abnormal sound event is obtained by weighted summation of the spatial sampling unit coordinates and the corresponding chain transition probabilities. The duration of the abnormal sound event is fitted with the spatial offset to construct a time-space coupling curve, and the desired spatial coordinate points are connected in chronological order to generate the evolution trajectory of the candidate noise source.

[0011] Preferably, the process of identifying the location and type of noise pollution sources by calculating the stability index, abrupt change frequency index, and structural resonance matching coefficient based on the evolution trajectory of candidate noise sources includes: Discrete difference operation is performed on the expected spatial coordinates of the evolution trajectory of the candidate noise source within a continuous time slice to calculate the variance of the spatial offset between adjacent time slices, and the reciprocal of the variance is used as the stability index. Based on the time-space coupling relationship curve corresponding to the evolution trajectory of the candidate noise source, the number of times the sign of the second derivative of the curve changes is counted, and the number of changes per unit time is defined as the mutation frequency index. Extract the main frequency component of the time slice corresponding to the evolution trajectory of the candidate noise source, and match it one by one with the set of natural frequencies of the structure in the target area. Calculate the minimum absolute value of the difference between the main frequency component and each natural frequency of the structure, and use the reciprocal of the minimum value as the structural resonance matching coefficient. The stability index, mutation frequency index, and structural resonance matching coefficient are weighted and fused according to preset weights to construct a noise source discrimination function, and the spatial location and type label of the noise pollution source are determined based on the output of the discrimination function.

[0012] Preferably, the process of generating the noise risk coupling strength distribution by performing nonlinear coupling calculations on the propagation attenuation factor, the chain transition probability, and the structural resonance matching coefficient includes: The propagation attenuation factor, chain transition probability, and structural resonance matching coefficient are respectively normalized by interval normalization to obtain normalized propagation attenuation factor, normalized chain transition probability, and normalized structural resonance matching coefficient. A nonlinear coupling function is constructed, and the normalized propagation attenuation factor and the normalized chain transition probability are exponentially multiplied. The normalized structural resonance matching coefficient is introduced as a power-law modulation factor to calculate the noise risk coupling strength value corresponding to each spatial sampling unit. The noise risk coupling strength values ​​of each spatial sampling unit are spatially mapped according to a unified spatial coordinate system to form a noise risk coupling strength distribution. Based on the statistical mean and standard deviation of noise risk coupling intensity under historical normal operating conditions, a graded threshold interval is constructed to classify the distribution of noise risk coupling intensity and output the corresponding risk level evaluation results.

[0013] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention achieves unified modeling and coupled deduction of air propagation channels and structural conduction channels by constructing a multidimensional acoustic feature matrix, a sound-structure coupled energy density field, and a directed weighted propagation diagram. This overcomes the technical bottleneck of traditional methods that rely solely on sound pressure level or single spectrum analysis, which cannot identify microscale, low-power, concealed noise sources. By introducing a chain transition probability and spatiotemporal folding mapping method, this invention can characterize the dynamic evolution of anomalous sound events in both time and space dimensions, significantly improving the early identification capability and positioning accuracy of intermittent, structurally coupled noise pollution sources, and reducing the interference of strong reverberation environments and background noise on the identification results.

[0014] 2. This invention constructs a discriminant function using a stability index, a mutation frequency index, and a structural resonance matching coefficient, and generates a noise risk coupling intensity distribution using nonlinear coupling operations, thus expanding the technology from "sound source identification" to "risk quantification assessment." This technical solution can identify potential structural resonance risks and abnormal energy accumulation risks before noise reaches environmental limits, providing a quantitative basis for equipment fatigue early warning, structural safety assessment, and control of hidden health risks. It has the technical advantages of high identification accuracy, strong anti-interference ability, and highly intuitive and visual risk expression. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0016] Figure 1 This is a flowchart of a method for identifying and evaluating noise pollution sources according to the present invention.

[0017] Figure 2 This is a flowchart of the method for identifying the location and type of noise pollution sources according to the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] For examples, please refer to Figure 1 , 2 As shown in this embodiment, a method for identifying and evaluating noise pollution sources includes: Acquire multi-channel sound pressure signals and corresponding structural vibration response signals within the target area, perform time-frequency decomposition on the sound pressure signals, extract instantaneous spectral density, phase difference gradient and higher-order spectral entropy parameters, and construct a multi-dimensional acoustic feature matrix; In this embodiment, firstly, multi-channel sound pressure signals and corresponding structural vibration response signals within the target area are acquired. The multi-channel sound pressure signals are simultaneously acquired by no fewer than three sound pressure sensors, with a sampling frequency set to 48,000 Hz, and the sampling time error is corrected to within 0.001 Hz using a unified time reference signal. The structural vibration response signals are acquired by accelerometers, with a sampling frequency consistent with that of the multi-channel sound pressure signals.

[0020] When performing adaptive window function segmentation on the multi-channel sound pressure signal, the short-time energy within the sliding time window is first calculated. The short-time energy is calculated according to the following expression: Where E(k) represents the short-time energy value corresponding to the sliding time window starting from the k-th sampling point; N represents the window length, initially set to 1024; x(k+i) represents the amplitude of the (k+i)-th sampling point; i is the sampling point number within the window; k increases according to the step size L, which is set to 512. Then, the energy change rate is calculated: Where M(k) is the average short-time energy of the current window and the two preceding windows. When R(k) is greater than 0.15, the window length N is adjusted to 0.5 times the original length; when R(k) is less than 0.05, the window length N is adjusted to 1.5 times the original length; when R(k) is between 0.05 and 0.15, it remains unchanged. Through the above adjustments, equal-duration signal subsequences are obtained, and a unique timestamp is assigned to each signal subsequence.

[0021] When performing continuous wavelet transform on each of the aforementioned signal subsequences, the complex Morlet function is selected as the mother wavelet function, and the scale parameter ranges from 1 to 128. For each scale a and time position b, the wavelet coefficients W(a,b) are calculated, and the squared modulus of the wavelet coefficients is used as the instantaneous spectral density.

[0022] Subsequently, bispectral analysis was performed on the same signal subsequence. First, a fast Fourier transform was performed on the signal to obtain the frequency domain representation X(f). The bispectral B(f1,f2) is defined as the product of X(f1) and X(f2), multiplied by the conjugate of X(f1+f2), and then averaged multiple times.

[0023] Normalize the bispectral amplitude into a probability distribution form by dividing the bispectral amplitude corresponding to each frequency pair by the sum of the bispectral amplitudes of all frequency pairs.

[0024] Higher-order spectral entropy parameters are calculated using the information entropy formula: ; where p(i) is the normalized bispectral amplitude corresponding to the i-th frequency pair.

[0025] When constructing the phase difference gradient distribution map based on the time delay difference results between different channels, the generalized cross-correlation algorithm is first used to calculate the time delay difference between any two sound pressure channels. The time delay difference is obtained by finding the time position corresponding to the maximum peak value of the cross-correlation function.

[0026] The time delay difference is converted into a phase difference, and the calculation method is as follows: Where φ is the phase difference; π is pi; f0 is the frequency corresponding to the maximum instantaneous spectral density; and τ is the time delay difference.

[0027] Based on the actual installation coordinates of each sound pressure sensor in three-dimensional space, the phase difference is mapped to the spatial coordinate system. Spatial interpolation reconstruction is performed using an inverse distance-weighted interpolation method. The interpolation weight wi is defined as: Where di is the Euclidean distance between the point to be interpolated and the i-th sensor. The phase difference gradient matrix in a unified spatial coordinate system is obtained by weighted averaging.

[0028] The instantaneous spectral density, higher-order spectral entropy parameters, and phase difference gradient matrix are aligned according to timestamps. The minimum time resolution is selected as the unified time base, and linear interpolation is used to numerically compensate for data with inconsistent times.

[0029] Then, the data corresponding to each timestamp is processed by vector expansion: the instantaneous spectral density is expanded into a frequency-dimensional vector, the higher-order spectral entropy parameter is used as a scalar additional feature, and the phase difference gradient matrix is ​​expanded into a spatial-dimensional vector.

[0030] Tensors are concatenated in the order of frequency, space, and time dimensions, and all time-slice feature vectors are stacked in chronological order to form a multidimensional acoustic feature matrix with spatiotemporal correspondence.

[0031] Each row in the multidimensional acoustic feature matrix corresponds to a time slice, and each column corresponds to a frequency or spatial location feature, which is used for subsequent calculation of the acoustic-structure coupling energy density field.

[0032] The multidimensional acoustic feature matrix and the structural vibration response signal are fused across modes to generate an acoustic-structure coupled energy density field, and the energy concentration coefficient of each spatial sampling unit is calculated.

[0033] When performing frequency decomposition processing on the structural vibration response signal, the signal is first synchronously segmented according to the same timestamps as the multidimensional acoustic feature matrix, ensuring that the length of each time slice is consistent with the length of the sound pressure signal time slice. Within each time slice, a fast Fourier transform is performed on the structural vibration response signal to obtain the frequency domain representation V(f). The vibration energy spectral density vector Sv(f) is defined as: Where |V(f)| represents the vibration amplitude at frequency f; T represents the time length of the time slice. Arranging Sv(f) corresponding to all frequency points in frequency order constitutes the vibration energy spectral density vector corresponding to that time slice.

[0034] The vibration energy spectral density vector is matched with the instantaneous spectral density in the multidimensional acoustic feature matrix in terms of frequency dimension. The frequency matching method is as follows: using a unified frequency resolution Δf as a reference, the frequency axes of both are resampled to ensure that the number of frequency points is consistent. A sound-vibration frequency correspondence mapping matrix M(fi,fj) is constructed, where fi represents the acoustic frequency component and fj represents the vibration frequency component. M(fi,fj) is set to 1 if the condition is met, and 0 otherwise. The initial coupling energy value Ec(fi) is defined as: Where Sa(fi) represents the instantaneous spectral density; Sv(fj) represents the vibrational energy spectral density.

[0035] The spatial coupling weight coefficients are calculated based on the initial coupling energy value and the phase difference gradient matrix. First, the absolute value of the gradient at each spatial sampling point in the phase difference gradient matrix is ​​taken and normalized. The normalization method is as follows: Where Gmax is the maximum gradient value in the phase difference gradient matrix. Simultaneously, the structural vibration amplitude Av(x,y,z) at the corresponding spatial location is calculated and normalized to its maximum value. The spatial coupling weighting coefficient W(x,y,z) is defined as: Wherein, α and β are weighting coefficients, satisfying α+β=1. In this embodiment, α is set to 0.6 and β is set to 0.4.

[0036] The initial coupling energy value is multiplied point-by-point with the spatial coupling weight coefficient to obtain the spatial coupling energy value Es(x,y,z). All Es(x,y,z) are arranged in three-dimensional space according to the spatial sampling point coordinates to form a three-dimensional energy distribution tensor. This three-dimensional energy distribution tensor is stacked sequentially over consecutive time slices to form an acoustic-structure coupling energy density field with spatiotemporal correspondence.

[0037] The acoustic-structure coupled energy density field is divided into cubic spatial sampling units with side length d according to a unified spatial coordinate system. In this embodiment, d is set to 0.5. The spatial coupled energy values ​​of all spatial sampling points within each spatial sampling unit are summed to obtain the total coupled energy value of that spatial sampling unit in the current time slice, forming a spatial energy distribution sequence.

[0038] Taking the current spatial sampling unit as the center, calculate the Euclidean distance between it and other spatial sampling units. Select spatial sampling units with a distance less than a preset radius r as adjacent spatial sampling units, where r is set to 1.5 in this embodiment. Calculate the arithmetic mean of the total coupling energy of the adjacent spatial sampling units to obtain a local background energy reference value.

[0039] Energy deviation D is defined as: Where Etotal is the total coupled energy of the current spatial sampling cell; Ebg is the local background energy reference value. The rate of change Rt of the total coupled energy between two consecutive time slices is also calculated: The energy deviation D and the coupling energy change rate Rt are weighted and superimposed: Wherein, γ is the time stability weighting coefficient, which is set to 0.7 in this embodiment. The resulting C is the energy concentration coefficient of each spatial sampling unit, used to represent the degree of abnormal concentration of acoustic-structural coupling energy in space.

[0040] A sound energy propagation topology network is constructed based on the energy concentration coefficient, and the network edges are weighted according to the energy gradient direction to form a directed weighted propagation graph.

[0041] Using a unified spatial coordinate system, the geometric center coordinates of each spatial sampling unit are denoted as (xi, yi, zi), where i represents the spatial sampling unit number. Each spatial sampling unit is treated as a network node, and its corresponding energy concentration coefficient Ci is stored as a node attribute value, along with its spatial coordinates, forming a node data structure Ni={xi,yi,zi,Ci}. All spatial sampling units are traversed, and all node data structures are added to the node set in numerical order to construct the initial node set.

[0042] For any two spatial sampling units i and j, calculate their Euclidean distance Dij using the following formula: Set a connection threshold d0. The connection threshold d0 is determined by multiplying the average Euclidean distance between all spatial sampling units by a scaling factor of 0.5. Construct an adjacency decision matrix Aij. When Dij ≤ d0, set Aij = 1; when Dij > d0, set Aij = 0. For all node pairs that satisfy Aij = 1, establish an undirected connection between node i and node j to form the initial acoustic propagation topology network.

[0043] For node pairs i and j that have established a connection, calculate the energy concentration coefficient difference ΔCij: When ΔCij > 0, the edge direction is changed from node i to node j; when ΔCij < 0, the edge direction is changed from node j to node i; when ΔCij = 0, no direction is set. Through these rules, the undirected connections in the initial acoustic energy propagation topology are transformed into connections with directional attributes.

[0044] For each connection edge with a defined direction, the energy gradient value Gij is calculated, defined as: Where |ΔCij| represents the absolute value of the difference in energy concentration coefficients. Simultaneously, the reciprocal of the spatial distance, Rij = 1 / Dij, is calculated. The edge weight Wij is defined as: Wij = Gij × Rij; the edge weight Wij is assigned to the connecting edges in the corresponding directions. After traversing all connecting edges, a directed weighted propagation graph containing the node set, directional attributes, and edge weights is formed. This directed weighted propagation graph is used for subsequent acoustic energy propagation path deduction and chain transition probability calculation.

[0045] Based on the directed weighted propagation diagram, the coupling path of acoustic energy in the air propagation channel and the structural conduction channel is deduced, and the propagation attenuation factor and the chain transition probability between nodes are calculated.

[0046] A path cost function F(P) is constructed based on the edge weights Wij of each edge in the directed weighted propagation graph, where P represents a propagation path from the starting node to the target node. The path cost function is defined as follows: Where Wij represents the edge weight of each directed edge on path P. A minimum-cost path search method is used to extract candidate propagation paths. The specific implementation steps are as follows: First, select the node with the largest energy aggregation coefficient as the starting node; second, traverse the nodes directly connected to the starting node and calculate the cumulative path cost; then, expand the nodes layer by layer in order of increasing path cost until all reachable nodes have been traversed; finally, select several paths with the minimum path cost as candidate propagation paths.

[0047] In the candidate propagation paths, if the sum of the spatial distances on the path is less than 0.8 times the average spatial distance of all candidate paths, it is determined to be an air propagation channel; if the average absolute value of the difference in energy concentration coefficients between adjacent nodes on the path is less than a preset threshold of 0.1, it is determined to be a structural conduction channel.

[0048] For each candidate propagation path, a propagation attenuation model is constructed. The propagation attenuation factor AP is defined as: Where Π represents the product operation of all edge weights on the path; LP represents the total spatial distance of path P; λ is the path length correction coefficient, which is set to 0.05 in this embodiment; and exp represents an exponential function with the natural constant as the base. The above expression is used to characterize the attenuation effect of sound energy as the path length increases during propagation.

[0049] Construct a node transition probability matrix based on the propagation attenuation factors. For any node i, the propagation attenuation factors corresponding to all its outgoing edges are denoted as Ai1, Ai2, ..., Ain. Normalize the above propagation attenuation factors to obtain the chain transition probability Pij from node i to node j: Where Σ Aik represents the sum of the propagation attenuation factors of all outgoing edges of node i. The complete node transition probability matrix is ​​constructed in this way.

[0050] Calculate the chain transition probability matrices for the air propagation channel and the structural conduction channel, respectively. Let the weighting coefficient for the air propagation channel be μ, and the weighting coefficient for the structural conduction channel be... In this embodiment, μ is set to 0.6. The two types of chain transition probability matrices are fused using a weighted superposition method to obtain the final acoustic energy coupling propagation path and its chain transition probability results. This result is used for subsequent abnormal acoustic event evolution trajectory deduction.

[0051] Based on the chain transition probability, the multidimensional acoustic feature matrix is ​​spatiotemporally folded and mapped to establish the coupling relationship between the duration of the abnormal sound event and the spatial offset, thereby generating the evolution trajectory of the candidate noise source.

[0052] A time-recursive weight matrix is ​​constructed based on the chain transition probabilities. Let P(t) be the node transition probability matrix corresponding to the t-th time slice. The time-recursive weight matrix W(t) is defined as the product of the weight matrix of the previous time slice and the current node transition probability matrix, i.e.: The initial time-recursive weight matrix W(0) is defined as the identity matrix. The time-recursive weight matrix W(t) is multiplied by the eigenvector corresponding to the t-th time slice in the multidimensional acoustic feature matrix to obtain the folded time feature vector F(t). F(t) is accumulated and superimposed in chronological order to form a folded time feature sequence. This folded time feature sequence is used to reinforce the propagation trend characteristics dominated by the probability of chain transitions.

[0053] In the folded time feature sequence, the energy amplitude of the feature vector for each time slice is first calculated. The energy amplitude is defined as the square root of the sum of the squares of the components of that feature vector. The mean Emean and standard deviation σ of the energy amplitudes for all time slices are then calculated. An abnormal energy threshold Te is set as follows: When the energy amplitude of multiple consecutive time slices is greater than the abnormal energy threshold Te, the consecutive time segment is defined as an abnormal sound event. The duration of this consecutive time segment is recorded as the duration of the abnormal sound event.

[0054] Within each time slice of an abnormal sound event, the expected spatial coordinates are calculated using the chain transition probability matrix of the corresponding time slice. Let the spatial coordinates of spatial sampling unit i be (xi, yi, zi), and its chain transition probability in the current time slice be Pi. The expected spatial coordinates (xt, yt, zt) are defined as: xt = Σ(Pi × xi); yt = Σ(Pi × yi); zt = Σ(Pi × zi); where Σ represents the summation over all spatial sampling units. The difference in expected spatial coordinates between adjacent time slices is calculated as follows: The spatial offset of the abnormal sound event is obtained by accumulating ΔS(t) for each time slice.

[0055] Plotting the duration of the abnormal sound event on the horizontal axis and the spatial offset of the corresponding time slice on the vertical axis, a quadratic polynomial fitting using the least squares method is employed to obtain the time-space coupling relationship curve. The expression for the quadratic polynomial is: Where a1, b1, and c1 are the fitting coefficients obtained by the least squares method. Connecting the spatial desired coordinates of each time slice in chronological order forms a continuous spatial trajectory curve. This continuous spatial trajectory curve is the evolution trajectory of the candidate noise source, used for subsequent identification of the location and type of noise pollution sources.

[0056] Based on the evolution trajectory of the candidate noise sources, the stability index, mutation frequency index, and structural resonance matching coefficient are calculated to identify the location and type of noise pollution sources.

[0057] Based on the expected spatial coordinates of the candidate noise source evolution trajectory within consecutive time slices, let the expected spatial coordinates of the t-th time slice be (xt, yt, zt). First, calculate the spatial offset ΔS(t) between adjacent time slices. Then, within the duration of the anomalous sound event, calculate the variance VarS for all ΔS(t). The stability index Is is defined as: Where ε is a correction coefficient to prevent the denominator from being zero, which is set to 0.001 in this embodiment. The stability index is used to characterize the spatial fluctuation of the evolution trajectory of candidate noise sources; the smaller the variance, the larger the stability index.

[0058] Based on the time-space coupling curve S(t), the second derivative of the curve is calculated. The expression for the second derivative is: Under discrete-time conditions, the discrete second-order difference value D2(t) is obtained by performing a second-order difference operation on the discrete spatial offset sequence. The number of times Nc changes the symbol of D2(t) during the duration of the anomalous sound event is counted. Let the duration of the anomalous sound event be Td, then the mutation frequency index If is defined as: If = Nc / Td; this index is used to characterize the suddenness of the evolution process of the candidate noise source.

[0059] Extract the dominant frequency component fp corresponding to the time slice of the candidate noise source evolution trajectory. The dominant frequency component is defined as the frequency corresponding to the maximum instantaneous spectral density. The set of natural frequencies {f1, f2, ..., fn} of the target region structure is obtained in advance through structural modal testing. Calculate the absolute value Δfi of the difference between the dominant frequency component and each structure natural frequency. The minimum difference Δfmin is selected. The structural resonance matching coefficient Ir is defined as: Where δ is the frequency resolution correction coefficient, which is set to 0.1 in this embodiment. The structural resonance matching coefficient is used to characterize the proximity between the candidate noise source and the natural frequency of the structure.

[0060] A noise source discrimination function F is constructed by weighting and fusing the stability index, mutation frequency index, and structural resonance matching coefficient: F = 0.4 × Is + 0.3 × If + 0.3 × Ir. The discrimination threshold Tf is set as the mean of the discrimination function under historical normal operating conditions plus twice the standard deviation. When F ≥ Tf, the area corresponding to the expected spatial coordinates is determined to be the location of the noise pollution source. Based on the combination of Is, If, and Ir, the source is classified as follows: a high stability index and a high structural resonance matching coefficient indicate a structural resonance noise source; a high mutation frequency index indicates an intermittent burst noise source; and a low stability index and a low mutation frequency index indicate a spatially drifting noise source. Through these steps, the location and type of noise pollution sources are identified.

[0061] The propagation attenuation factor, chain transition probability, and structural resonance matching coefficient are subjected to nonlinear coupling calculation to generate a noise risk coupling intensity distribution and output the corresponding risk level evaluation results.

[0062] The propagation attenuation factor, the chain transition probability, and the structural resonance matching coefficient are all subjected to interval normalization.

[0063] Let Ai be the propagation attenuation factor corresponding to a certain spatial sampling cell. Among all spatial sampling cells, the maximum value of the propagation attenuation factor is Amax, and the minimum value is Amin. Then, the normalized propagation attenuation factor Ai' is defined as: The chain transition probability Pi and the structural resonance matching coefficient Iri were normalized in the same way to obtain the normalized chain transition probability Pi' and the normalized structural resonance matching coefficient Iri', respectively.

[0064] A nonlinear coupling function Ri is constructed to calculate the coupling strength value of the noise risk. The nonlinear coupling function is defined as follows: The exponential function is used to amplify the effect of the structural resonance matching coefficient on the risk intensity, thereby reflecting the risk superposition characteristics under structural resonance conditions. Ri is calculated for all spatial sampling units to obtain the noise risk coupling strength value of each spatial sampling unit.

[0065] Using a unified spatial coordinate system, the noise risk coupling intensity value Ri of each spatial sampling unit is mapped to its corresponding spatial coordinate position and arranged in the order of spatial sampling unit numbering to form a three-dimensional risk numerical matrix. This three-dimensional risk numerical matrix is ​​the noise risk coupling intensity distribution, used to represent the spatial distribution of risk.

[0066] Based on the noise risk coupling intensity value sequence under historical normal operating conditions, its mean Rmean and standard deviation σR are calculated. The risk classification threshold intervals are constructed as follows: Level 1 Risk: Ri <Rmean+1×σR; Level 2 risk: Rmean+1×σR≤Ri <Rmean+2×σR; Level 3 risk: Ri≥Rmean+2×σR.

[0067] The noise risk coupling strength values ​​of all spatial sampling units are classified into levels, and the corresponding risk level evaluation results are output.

[0068] Through the above steps, nonlinear coupling calculations of propagation attenuation factor, chain transition probability, and structural resonance matching coefficient are achieved, as well as noise risk level determination.

[0069] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for identifying and evaluating noise pollution sources, characterized in that: include: Acquire multi-channel sound pressure signals and corresponding structural vibration response signals within the target area, perform time-frequency decomposition on the sound pressure signals, extract instantaneous spectral density, phase difference gradient and higher-order spectral entropy parameters, and construct a multi-dimensional acoustic feature matrix; The multidimensional acoustic feature matrix and the structural vibration response signal are fused across modes to generate an acoustic-structure coupled energy density field, and the energy concentration coefficient of each spatial sampling unit is calculated. A sound energy propagation topology network is constructed based on the energy concentration coefficient, and the network edges are weighted according to the energy gradient direction to form a directed weighted propagation graph; Based on the directed weighted propagation graph, the coupling path of sound energy in the air propagation channel and the structural conduction channel is deduced, and the propagation attenuation factor and the chain transition probability between nodes are calculated. Based on the chain transition probability, the multidimensional acoustic feature matrix is ​​spatiotemporally folded and mapped to establish the coupling relationship between the duration of the abnormal sound event and the spatial offset, and the evolution trajectory of the candidate noise source is generated. Based on the evolution trajectory of the candidate noise sources, the stability index, mutation frequency index, and structural resonance matching coefficient are calculated to identify the location and type of noise pollution sources. The propagation attenuation factor, chain transition probability, and structural resonance matching coefficient are subjected to nonlinear coupling calculation to generate a noise risk coupling intensity distribution and output the corresponding risk level evaluation results.

2. The method for identifying and evaluating noise pollution sources according to claim 1, characterized in that: The process of constructing the multidimensional acoustic feature matrix includes: The multi-channel sound pressure signal is segmented using an adaptive window function to obtain equal-duration signal subsequences; Continuous wavelet transform and bispectral joint analysis are performed on each of the signal subsequences to calculate the corresponding instantaneous spectral density and higher-order spectral entropy parameters; A phase difference gradient distribution map is constructed based on the time delay difference results between different channels, and the phase difference gradient distribution map is reconstructed by spatial interpolation to generate a phase difference gradient matrix in a unified spatial coordinate system. The instantaneous spectral density, higher-order spectral entropy parameters, and phase difference gradient matrix are aligned according to timestamps and then tensor-concatenated to form a multidimensional acoustic feature matrix with spatiotemporal correspondence.

3. The method for identifying and evaluating noise pollution sources according to claim 1, characterized in that: The process of generating the acoustic-structure coupled energy density field includes: The vibration response signal of the structure is subjected to frequency decomposition processing, and synchronous segmentation is performed using the same timestamp as the multidimensional acoustic feature matrix. The vibration energy spectral density vector corresponding to each time slice is calculated. The vibration energy spectral density vector is matched with the instantaneous spectral density in the multidimensional acoustic feature matrix in terms of frequency dimension to construct a sound-vibration frequency correspondence mapping matrix, and the initial coupling energy value is calculated based on the frequency amplitude product. The spatial coupling weight coefficient is calculated based on the initial coupling energy value and the phase difference gradient matrix. The spatial coupling weight coefficient is obtained by weighted superposition of the normalized result of the absolute value of the phase difference gradient and the structural vibration amplitude. The initial coupling energy value is multiplied point by point with the spatial coupling weight coefficient, and reconstructed into a three-dimensional energy distribution tensor according to the spatial coordinates to generate an acoustic-structure coupling energy density field with spatiotemporal correspondence.

4. The method for identifying and evaluating noise pollution sources according to claim 3, characterized in that: The process of calculating the energy concentration coefficient of each spatial sampling unit includes: The acoustic-structure coupled energy density field is divided into several spatial sampling units according to a unified spatial coordinate system, and the total coupled energy value of each spatial sampling unit in the same time slice is extracted to form a spatial energy distribution sequence. Taking the current spatial sampling unit as the center, select its neighboring spatial sampling units in three-dimensional space that are less than a preset radius r, calculate the average value of the total coupling energy of the neighboring spatial sampling units, and obtain the local background energy reference value. An energy deviation is constructed based on the total coupled energy value and the local background energy reference value. The energy deviation is defined as the ratio of the difference between the total coupled energy value and the local background energy reference value to the local background energy reference value. The energy deviation is weighted and superimposed with the rate of change of coupling energy of the current spatial sampling unit in a continuous time slice to obtain the energy concentration coefficient of each spatial sampling unit, which is used to represent the degree of abnormal concentration of acoustic-structural coupling energy in space.

5. The method for identifying and evaluating noise pollution sources according to claim 1, characterized in that: The process of forming the directed weighted propagation graph includes: Using a unified spatial coordinate system, each spatial sampling unit is taken as a network node, and the corresponding energy concentration coefficient is used as the node attribute value to construct an initial node set. An adjacency determination matrix is ​​constructed based on the Euclidean distance between any two spatial sampling units. When the distance is less than the preset connection threshold d0, an undirected connection relationship is established to form an initial acoustic energy propagation topology network. The energy gradient value is calculated based on the difference in energy concentration coefficients between adjacent spatial sampling units, and the direction of the edge is determined by pointing from the spatial sampling unit with a high energy concentration coefficient to the spatial sampling unit with a low energy concentration coefficient. The energy gradient value is multiplied by the reciprocal of the corresponding spatial distance to obtain the edge weight. This weight is then assigned to each connected edge to generate a directed weighted propagation graph with directional and weight attributes.

6. The method for identifying and evaluating noise pollution sources according to claim 1, characterized in that: The process of calculating the propagation attenuation factor and the chain transition probability between nodes includes: Based on the edge weights of each edge in the directed weighted propagation graph, a path cost function is constructed, and a minimum cost path search method is used to extract candidate propagation paths for air propagation channels and structural conduction channels respectively. The air propagation channels are selected based on the principle of spatial distance weight dominance, and the structural conduction channels are selected based on the principle of energy concentration coefficient continuity dominance. For each candidate propagation path, a propagation attenuation model is constructed based on the product of the weights of each edge on the path, and a path length correction factor is introduced to perform exponential attenuation calculation on the edge weights to obtain the propagation attenuation factor between nodes. Based on the propagation attenuation factor, a node transition probability matrix is ​​constructed, and the outgoing edge propagation attenuation factor of each node is normalized to form the chain transition probability between nodes. The chain transition probabilities corresponding to the air propagation channel and the structural conduction channel are weighted and fused to obtain the acoustic energy coupling propagation path and its chain transition probability results that reflect the dual-channel coupling characteristics.

7. The method for identifying and evaluating noise pollution sources according to claim 1, characterized in that: The process of generating candidate noise source evolution trajectories includes: Based on the chain transition probabilities, a time recursive weight matrix is ​​constructed, and the feature vectors of each time slice in the multidimensional acoustic feature matrix are weighted and accumulated in time order to form a folded time feature sequence. In the folded time feature sequence, time segments that continuously exceed a preset energy threshold are identified, and the length of the time segment is defined as the duration of the abnormal sound event. The spatial expected coordinates corresponding to each time slice are calculated based on the chain transition probability matrix. The spatial offset of the abnormal sound event is obtained by weighted summation of the spatial sampling unit coordinates and the corresponding chain transition probabilities. The duration of the abnormal sound event is fitted with the spatial offset to construct a time-space coupling curve, and the desired spatial coordinate points are connected in chronological order to generate the evolution trajectory of the candidate noise source.

8. The method for identifying and evaluating noise pollution sources according to claim 1, characterized in that: The process of identifying the location and type of noise pollution sources by calculating the stability index, abrupt change frequency index, and structural resonance matching coefficient based on the evolution trajectory of candidate noise sources includes: Discrete difference operation is performed on the expected spatial coordinates of the evolution trajectory of the candidate noise source within a continuous time slice to calculate the variance of the spatial offset between adjacent time slices, and the reciprocal of the variance is used as the stability index. Based on the time-space coupling relationship curve corresponding to the evolution trajectory of the candidate noise source, the number of times the sign of the second derivative of the curve changes is counted, and the number of changes per unit time is defined as the mutation frequency index. Extract the main frequency component of the time slice corresponding to the evolution trajectory of the candidate noise source, and match it one by one with the set of natural frequencies of the structure in the target area. Calculate the minimum absolute value of the difference between the main frequency component and each natural frequency of the structure, and use the reciprocal of the minimum value as the structural resonance matching coefficient. The stability index, mutation frequency index, and structural resonance matching coefficient are weighted and fused according to preset weights to construct a noise source discrimination function, and the spatial location and type label of the noise pollution source are determined based on the output of the discrimination function.

9. The method for identifying and evaluating noise pollution sources according to claim 1, characterized in that: The process of generating the noise risk coupling strength distribution by performing nonlinear coupling calculations on the propagation attenuation factor, chain transition probability, and structural resonance matching coefficient includes: The propagation attenuation factor, chain transition probability, and structural resonance matching coefficient are respectively normalized by interval normalization to obtain normalized propagation attenuation factor, normalized chain transition probability, and normalized structural resonance matching coefficient. A nonlinear coupling function is constructed, and the normalized propagation attenuation factor and the normalized chain transition probability are exponentially multiplied. The normalized structural resonance matching coefficient is introduced as a power-law modulation factor to calculate the noise risk coupling strength value corresponding to each spatial sampling unit. The noise risk coupling strength values ​​of each spatial sampling unit are spatially mapped according to a unified spatial coordinate system to form a noise risk coupling strength distribution. Based on the statistical mean and standard deviation of noise risk coupling intensity under historical normal operating conditions, a graded threshold interval is constructed to classify the distribution of noise risk coupling intensity and output the corresponding risk level evaluation results.

Images