Computer-aided design method for noise barrier based on noise simulation prediction model

Abstract

The present application belongs to the technical field of computer aided design and noise simulation, in particular to a sound barrier computer aided design method based on a noise simulation prediction model. The method obtains a sound source, a sound receiving point, a terrain boundary, an acoustic medium and a sound barrier design parameter, establishes a parameterized scene, generates a sound propagation path graph according to a visible relationship, a shielding boundary and a propagation interface, obtains a path band contribution and a sound receiving point combined sound level by using a prediction model, calculates a computer residual error based on a path attenuation, a diffraction transition and a multi-path synthesis constraint, positions the propagation path, calls a high-precision noise simulation to generate a correction sample and update a residual error compensation function, maps the correction result to the design parameter for iteration, and when a path topological change is detected, reconstructs the path and the compensation relationship, and outputs a scheme meeting the noise constraint. The present application can suppress the prediction distortion in the topological switching area, reduce the irrelevant path repeated simulation, and keep the scheme evaluation consistent with the sound propagation state.

Description

Technical Field

[0001] This invention belongs to the field of computer-aided design and noise simulation technology, specifically a computer-aided design method for sound barriers based on a noise simulation prediction model. Background Technology

[0002] Existing computer-aided design methods for sound barriers typically begin by establishing a parameterized scenario of roads, sound sources, terrain, receiving points, and the sound barrier itself. Then, the frequency band sound level at each receiving point is calculated according to rules such as sound propagation distance, obstruction relationships, edge diffraction, interface reflection, and ground attenuation. To reduce the computational burden of high-precision simulations for each scheme, a common practice is to extract several sound barrier design states within a preset parameter range, using numerical acoustic calculation results to construct training samples, and establishing a mapping model from sound source parameters, environmental parameters, sound barrier parameters, and receiving point parameters to the predicted sound level. During the design phase, candidate parameters are generated through grid search, heuristic search, or gradient iteration. These candidate parameters are input into the prediction model to obtain noise evaluation results, and then the barrier position, height, or other designable parameters are adjusted based on the receiving point sound level limits. Some schemes also supplement the entire simulation sample when the prediction error is large, and retrain the prediction model, enabling the computer-aided design process to complete multiple rounds of scheme selection under limited computing resources.

[0003] Noise prediction models typically take the entire design scenario as a whole as input and directly output the synthesized sound level at the receiving point, without decomposing the contributions of different paths such as direct, diffracted, reflected, and ground propagation. When changes in sound barrier parameters cause a propagation path to change from visible to blocked, replace diffraction boundaries, or add or remove reflection paths, the sound propagation relationship undergoes a topological shift. The original continuous mapping is prone to local distortion near the switching boundary. Since the model cannot determine the specific propagation path and mechanism corresponding to the error, supplementing samples can only expand the entire simulation range, and parameter optimization may still be based on the distorted sound level change relationship to update the design parameters. Even with increased training samples, if the samples do not cover the specific topological switching neighborhood, the overall mapping still cannot establish the correspondence between the propagation path and the sound level change before and after the switch. The prediction error will accumulate along the design iteration process, making the evaluation order and parameter update direction of candidate schemes in the path topological change region lack a stable basis, and making it difficult to ensure that the computer-aided design results are consistent with the actual propagation state. Summary of the Invention

[0004] The purpose of this invention is to provide a computer-aided design method for sound barriers based on a noise simulation and prediction model, which can solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] The computer-aided design method for sound barriers based on noise simulation and prediction models includes: obtaining sound source parameters, sound receiving point parameters, terrain boundary, acoustic medium parameters and sound barrier designable parameters, and establishing a parametric design scenario;

[0007] A sound propagation path diagram is generated based on the visibility relationship between the sound source and the receiving point, the occlusion boundary, and the propagation interface.

[0008] Input the sound propagation path diagram into the noise simulation prediction model to obtain the frequency band contribution of each propagation path and the synthesized sound level at the receiving point;

[0009] Based on path attenuation constraints, diffraction transition constraints, and multipath synthesis constraints, the target propagation path in the residual set is determined by computer.

[0010] For the target propagation path, a high-precision noise simulation is invoked to generate local correction samples and update the residual compensation function;

[0011] The corrected frequency band contribution can be mapped to the sound barrier design parameters and the parameters can be updated.

[0012] When a change in the propagation path topology is detected, the sound propagation path diagram and residual compensation function are reconstructed, and a sound barrier design scheme that meets noise constraints is output.

[0013] Preferably, generating the sound propagation path map includes: using the sound source, the receiving point, the diffraction boundary of the sound barrier, and the propagation interface as path nodes, and establishing a candidate node sequence according to the spatial connectivity order between the sound source and the receiving point;

[0014] Perform visibility and interface intersection checks on adjacent path nodes, and delete candidate edges that traverse non-propagable regions or do not meet the propagation order.

[0015] Based on the mechanisms of direct propagation, occlusion diffraction, interface reflection, and ground propagation, propagation type identifiers are configured for the retained candidate edges, and path length, incident direction, outgoing direction, occlusion margin, and frequency band medium properties are recorded.

[0016] Candidate paths with the same node sequence and propagation type sequence are merged into a propagation path. A directed graph is constructed according to the correspondence between the propagation path and the receiving point to form a sound propagation path graph for the noise simulation and prediction model to read.

[0017] Preferably, the computer-generated residual includes: for each propagation path, comparing the frequency band contribution output by the noise simulation prediction model with an attenuation benchmark determined based on the path length and frequency band medium properties to obtain the path attenuation residual;

[0018] For the sound barrier, the propagation paths with the same propagation type sequence under adjacent disturbance states of the design parameters are paired, and the diffraction transition residual is obtained based on the continuous relationship of the frequency band contribution of the paired paths during the process of changing the blocking margin.

[0019] The consistency of the synthesized frequency band contribution of each propagation path with the synthesized sound level at the receiving point is verified to obtain the multi-path synthesis residual.

[0020] The path attenuation residual, the diffraction transition residual, and the multipath synthesis residual are combined according to the propagation path, frequency band, and receiver point index, and the target propagation path is determined based on the correlation between the residuals and adjacent path nodes.

[0021] Preferably, generating the local correction sample and updating the residual compensation function includes: locking the propagation path state other than the target propagation path, applying a local perturbation to the sound barrier design parameters associated with the target propagation path to keep the path topology unchanged, and forming a local design state;

[0022] Noise simulation prediction model calculation and high-precision noise simulation calculation are performed on the local design state respectively, and the prediction results are paired according to the same propagation path, frequency band and receiving point index.

[0023] Local correction samples are constructed based on the difference in pairing results, changes in path length, changes in occlusion margin, and propagation interface relationships.

[0024] The residual compensation function, which takes the propagation path state as input and the frequency band contribution compensation value as output, is updated using the local correction samples, and the updated compensation value is superimposed on the frequency band contribution of the corresponding propagation path.

[0025] Preferably, detecting the topological change of the propagation path includes: extracting the node sequence, propagation type sequence, interface intersection order, and occlusion margin symbol of each propagation path before and after the update of the designable parameters of the sound barrier, and generating the path signature before the update and the path signature after the update.

[0026] The path signature before and after the update are matched to identify the addition, disappearance, node replacement, change of propagation type, and change of interface order in the propagation path.

[0027] Based on the matching results, the sound propagation path graph is divided into a common subgraph that maintains topological consistency and a difference subgraph that undergoes topological changes;

[0028] The residual compensation function and the local correction sample corresponding to the propagation path in the common subgraph are retained, the path nodes, candidate edges and propagation type identifiers of the difference subgraph are regenerated, and an independent residual index is established for the regenerated propagation path.

[0029] Preferably, invoking high-precision noise simulation for the target propagation path includes: jointly ranking the target propagation path based on the mechanism residual amplitude, the contribution ratio of the target propagation path to the synthesized sound level at the receiving point, and the local sensitivity associated with the designable parameters of the sound barrier;

[0030] For the target propagation path selected based on the sorting results, generate mutually independent combinations of local perturbations within the parameter range while keeping the common subgraph unchanged;

[0031] For each combination of local disturbances, a path signature is pre-calculated, and only the combination whose path signature matches the target propagation path is submitted to the high-precision noise simulation.

[0032] When the path signature corresponding to the local disturbance combination changes, the topology change interval is determined according to the parameter state before and after the path signature change. The topology change interval is divided into sampling intervals for different path states, and local correction samples are generated respectively.

[0033] Preferably, updating the residual compensation function further includes: according to the path signature corresponding to the local correction sample, classifying samples with the same propagation type sequence into the path common sample set, and classifying samples with the same node adjacency relationship and interface intersection order into the path local sample set.

[0034] Common residual components of the frequency band residuals that vary with path length and propagation type are extracted from the common sample set of the path, and local residual components of the frequency band residuals that vary with occlusion margin and interface relationship are extracted from the local sample set of the path.

[0035] According to the sample coverage density of the target propagation path in the path state space, the common residual component and the local residual component are weighted and combined.

[0036] By applying adjacent frequency band continuity constraints and path energy synthesis constraints at the same receiving point to the combined compensation value, the updated residual compensation function is obtained.

[0037] Preferably, mapping the corrected frequency band contribution to the designable parameters of the sound barrier and performing parameter updates includes: calculating the local sensitivity of the designable parameters of the sound barrier relative to the sound level constraints of each sound receiving point and each frequency band based on the frequency band contribution of each propagation path after residual compensation, and constructing a sensitivity matrix between the design parameters and the sound level constraints;

[0038] The topological stability region is determined based on the parameter range in which the path signature remains unchanged, and the parameter update amount, which is simultaneously limited by the design parameter value boundary and sound level constraint, is solved within the topological stability region.

[0039] The parameter update amount is used to generate candidate design states and recalculate path signatures. When the path signatures of the candidate design states are consistent, they are written into the parameterized design scenario. When the path signatures change, the parameter update amount is revoked and the difference subgraph reconstruction is performed.

[0040] Preferably, when performing parameter updates within the boundary neighborhood of the topologically stable domain, the method further includes: obtaining path signatures and acoustic propagation path diagrams on both sides of the topological change interval, and establishing a transitional correspondence between newly added propagation paths and disappearing propagation paths based on common path nodes and propagation interfaces.

[0041] The frequency band contribution in the boundary neighborhood is calculated by using the residual compensation function on both sides of the topology change interval, and the combined weight of the acoustic propagation path map on both sides is generated based on the occlusion margin, interface intersection distance and path node visibility status.

[0042] The segmented frequency band contribution mapping of the boundary neighborhood is formed according to the combined weights;

[0043] The candidate parameter update is calculated based on the segmented frequency band contribution mapping, and the candidate parameter update across the topology change interval is split into continuous update amounts located in the topology stable domains on both sides.

[0044] Preferably, the output sound barrier design scheme that satisfies the noise constraints includes: recording the sound propagation path diagram, path signature, mechanism residual, local correction sample and residual compensation function corresponding to each parameter update, forming an iterative data chain associated with the candidate design state;

[0045] After each parameter update, the propagation path to be verified is determined according to the sound level constraint deviation at the sound receiving point, the frequency band contribution ratio of the propagation path, and the mechanism residual. High-precision noise simulation is performed on the propagation path to be verified, and the verification results are written into the iterative data chain.

[0046] When the frequency band difference between the verification result and the frequency band contribution after residual compensation exceeds the residual envelope of the local correction sample of the corresponding path signature, return to the corresponding topological stability domain to supplement the local correction sample and recalculate the parameter update amount.

[0047] When the verified sound level at each sound receiving point meets the sound level constraint and the path signature no longer changes, output the designable parameters of the sound barrier in the current parametric design scenario.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] 1. The sound source, receiver, diffraction boundary, and propagation interface are organized into a sound propagation path diagram. The frequency band contribution of each propagation path is output by the noise simulation prediction model, ensuring that the propagation source corresponding to the synthesized sound level remains distinguishable. After forming mechanistic residuals based on path attenuation constraints, diffraction transition constraints, and multipath synthesis constraints, the target propagation path in the residual set can be determined according to the propagation path, frequency band, and receiver index. Then, high-precision noise simulation is performed only on the local design state associated with the target propagation path, and the residual compensation function is updated accordingly. The corrected path contribution is then mapped to the design parameters of the sound barrier, so that the parameter update is based on the verified local acoustic response. When parameter changes cause the addition, disappearance, or switching of propagation type of a path, the topology change is identified by the path signature and the difference subgraph is reconstructed. The original parameter update relationship across the topology boundary is frozen to avoid using residual compensation results that are inconsistent with the current propagation state. Path decomposition, mechanistic residual location, local correction, and topology reconstruction form a continuous closed loop, thereby suppressing the transmission of local prediction distortion to subsequent design iterations and maintaining the consistency between the evaluation order of candidate schemes and the actual changes in propagation paths.

[0050] 2. Based on the path signatures before and after the update, the acoustic propagation path graph is divided into a topologically consistent common subgraph and a changed difference subgraph. The residual compensation function and local correction samples corresponding to the common subgraph are retained, while only the relevant path nodes, candidate edges, and residual indices are regenerated in the difference subgraph, reducing redundant calculations unrelated to topological changes. Local correction samples are grouped according to propagation type sequence, node adjacency relationship, and interface intersection order, and common residual components and local residual components are extracted separately, enabling existing samples to participate in subsequent corrections under the same propagation mechanism. Design parameter updates are restricted to the topologically stable domain where the path signature remains unchanged. Candidate update quantities that cross the topological change interval are split into continuous update quantities under different path states to avoid mixing two types of propagation relationships in a single update. During the design iteration process, the acoustic propagation path graph, path signature, mechanism residual, correction samples, and compensation function are recorded simultaneously, and high-precision noise simulation is performed on the propagation path to be verified. When the verification difference exceeds the local sample residual envelope, the corresponding topologically stable domain is returned to supplement samples, so that the output design parameters have a calculation basis corresponding to the noise prediction results. Attached Figure Description

[0051] Figure 1 A flowchart illustrating the overall process of computer-aided design of a sound barrier based on a noise simulation and prediction model;

[0052] Figure 2 A flowchart for constructing the sound propagation path diagram, locating the mechanism residual, and performing local correction.

[0053] Figure 3 Map showing the contribution of path frequency bands;

[0054] Figure 4 Flowchart for identifying propagation path topology changes, updating parameters, and verifying design results;

[0055] Figure 5 Match the path signature to the corresponding graph;

[0056] Figure 6 This is a graph showing the weights of the combined topological boundaries.

[0057] Figure 7 This is a comparison diagram of residual envelopes. Detailed Implementation

[0058] In one embodiment, please refer to Figure 1 This paper provides a computer-aided design method for sound barriers based on a noise simulation prediction model. The method includes acquiring sound source parameters, receiver parameters, terrain boundaries, acoustic medium parameters, and sound barrier design parameters; establishing a parametric design scenario; generating a sound propagation path map based on the visibility relationship between the sound source and receiver, obstruction boundaries, and propagation interfaces; inputting the sound propagation path map into the noise simulation prediction model to obtain the frequency band contribution of each propagation path and the synthesized sound level at the receiver; calculating the theoretical residuals based on path attenuation constraints, diffraction transition constraints, and multi-path synthesis constraints to determine the target propagation path in the residual set; calling high-precision noise simulation to generate local correction samples for the target propagation path and updating the residual compensation function; mapping the corrected frequency band contribution to the sound barrier design parameters and performing parameter updates; reconstructing the sound propagation path map and residual compensation function when a propagation path topology change is detected; and outputting a sound barrier design scheme that meets noise constraints. The entire calculation process uses the propagation path as the minimum prediction and correction object, enabling the synthesized sound level at the receiver to be traced back to the sound barrier design parameters along the specific propagation path, and forming a closed data processing chain of parameter updates, path topology detection, and local correction.

[0059] In this embodiment, the parametric design scenario uses a unified coordinate reference to represent the sound source, receiving point, terrain boundary, propagation interface, and sound barrier design object. The sound source parameters include at least the spatial location of the sound source, the direction of the sound source, and the reference sound level of each analysis frequency band. The receiving point parameters include at least the spatial location of the receiving point and the corresponding frequency band sound level constraints. The terrain boundary is represented by boundary points or boundary surfaces arranged in spatial connection order. The acoustic medium parameters are indexed and associated with the corresponding terrain region or propagation interface. The designable parameters of the sound barrier are stored in a design variable vector. Each element in the design variable vector is associated with a geometric attribute in the parametric model that can change the sound propagation relationship. The parametric design scenario stores the design variable vector, geometric object identifier, boundary connection relationship, and version identifier. Each parameter update generates a new scenario version without directly overwriting the previous version, so as to return to the corresponding design state when the path topology changes or the verification result does not meet the correction conditions.

[0060] Specifically, the training samples of the noise simulation prediction model consist of multiple parametric design scenarios and their high-precision noise simulation results. Each training sample does not use the synthesized sound level of a single receiver point as the sole supervisory data. Instead, it saves the path frequency band contribution according to the index relationship between the sound source, propagation path, analysis frequency band, and receiver point, and saves the receiver point synthesized sound level formed by energy synthesis of each propagation path. Before training, the path length, occlusion margin, incident direction, outgoing direction, interface intersection order, propagation type, and frequency band medium properties are uniformly encoded. Propagation paths with different numbers of nodes are represented by a combination of node sequence and edge attribute sequence. The prediction model reads the internal features of the path, the path set features under the same sound source, and the path set features under the same receiver point, respectively, and outputs the frequency band contribution corresponding to each input path. It also performs energy domain synthesis through receiver point index, so that the training process is constrained by both path-level results and receiver-level results.

[0061] In one embodiment, please refer to Figure 2 The generation of the sound propagation path map includes using the sound source, the receiving point, the diffraction boundary of the sound barrier, and the propagation interface as path nodes. A candidate node sequence is established according to the spatial connectivity order between the sound source and the receiving point. Visibility and interface intersection are determined for adjacent path nodes. Candidate edges that cross non-propagation areas or do not meet the propagation order are deleted. Propagation type identifiers are configured for the retained candidate edges based on direct reach, obstruction diffraction, interface reflection, and ground propagation mechanisms. The path length, incident direction, outgoing direction, obstruction margin, and frequency band medium properties are recorded. Candidate paths with the same node sequence and propagation type sequence are merged into a propagation path. A directed graph is constructed according to the correspondence between the propagation path and the receiving point to form a sound propagation path map for the noise simulation and prediction model to read. The sound source node is only used as the starting point of the path, the receiving point node is only used as the ending point of the path, and the diffraction boundary node and the propagation interface node are located between the starting point and the ending point according to the geometric relationship, thereby avoiding the formation of cyclic paths with contradictory propagation directions.

[0062] In this embodiment, visibility determination uses the connecting line segments between adjacent path nodes as the detection object. The connecting line segments are intersected with the terrain boundary, the sound barrier boundary, and the propagation interface item by item. When the connecting line segment has an internal intersection with the inaccessible boundary, the corresponding candidate edge is marked as invisible. When the connecting line segment only intersects with the boundary that is the starting point or the ending point, the candidate edge is retained. The interface intersection determination records the geometric relationship between the intersection position, the intersection direction, and the interface normal. For the reflection propagation path, the mirror position of the sound source or the previous node with respect to the propagation interface is used to verify whether the incident segment and the emitting segment meet the same interface constraint. For the diffraction propagation path, the position of the line connecting the sound source and the receiving point relative to the sound barrier boundary is used to determine the occlusion state, and the edge points that can connect the two sides of the occlusion area are selected as the diffraction boundary nodes. For the ground propagation path, continuous terrain areas located within the projection range of the propagation line segment and with consistent acoustic medium properties are merged into one path segment.

[0063] Furthermore, the candidate node sequence is generated by expanding layer by layer from the sound source to the receiving point. Each expansion only allows the addition of diffraction boundary nodes or propagation interface nodes that are reachable in the current propagation direction, and the intersection order of the same interface in the same propagation path is limited. After the expansion is completed, the candidate path is normalized. Duplicate nodes with the same geometric position but different object identifiers are uniformly mapped to the main node. Adjacent diffraction points that are generated due to boundary discretization and have the same propagation relationship are merged into the same diffraction boundary node. Candidate paths with path length differences within the model's resolution range and consistent node sequence and propagation type sequence are merged into a single propagation path. The merged path preserves the spatial coverage of the original candidate path so that when the design parameters of the sound barrier change locally, it can be determined whether the original propagation path is still within the corresponding coverage range.

[0064] In a preferred embodiment, the noise simulation prediction model predicts path residuals based on acoustic baseline calculations to reduce the possibility of unconstrained outputs when training data is insufficient. The predicted frequency band contribution of each propagation path is calculated according to the following formula:

[0065] ;

[0066] in, This represents the predicted sound level of the r-th propagation path in the f-th analysis frequency band. This represents the reference sound level of the sound source in the f-th analysis frequency band. This represents the normalized path length of the r-th propagation path. This indicates the diffraction attenuation. Indicates the effect of the propagation interface. Indicates the attenuation of propagation over the ground. This represents the correction amount output by the noise simulation prediction model based on the path map characteristics. When the reference sound level of a certain frequency band is 90, the normalized path length is 100, the diffraction attenuation is 8, the propagation interface effect is 2, the ground propagation attenuation is 1, and the correction amount is 1, the corresponding predicted contribution sound level is 40. The correction amount only corrects the local path relationships that the reference calculation fails to represent, and does not replace the path length and the constraint terms corresponding to each propagation mechanism.

[0067] Please refer to Figure 3 , Figure 3 The horizontal axis represents the analysis frequency band, covering a range from 63Hz to 8000Hz, and the vertical axis represents the contribution sound level of a single propagation path to the receiving point, in dB. Figure 3 Each curve in the figure corresponds to a representative propagation path in the same engineering scenario. The legend distinguishes different paths by path number and path code. The points on the curves represent the results of each analysis frequency band after being calculated by the source frequency band reference, normalized path length, diffraction attenuation, interface effect, ground propagation attenuation and correction.

[0068] from Figure 3 As can be seen, the contribution sound levels of different propagation paths in the low, mid, and high frequency bands do not change completely synchronously. Some paths show a significant decrease or plateau in the mid and high frequency bands, indicating that path length, diffraction boundary, and propagation interface jointly determine the distribution pattern of frequency band contribution. The spacing between the curves reflects the differences in the contribution of different path geometric relationships and propagation types to the receiving point. If a certain path maintains a high level in multiple frequency bands, this path should be given priority in subsequent residual localization and barrier parameter adjustment.

[0069] In one embodiment, the computational residual includes, for each propagation path, comparing the frequency band contribution output by the noise simulation prediction model with an attenuation benchmark determined based on path length and frequency band medium properties to obtain a path attenuation residual; pairing propagation paths with the same propagation type sequence under adjacent disturbance states of the sound barrier's design parameters; obtaining a diffraction transition residual based on the continuity of the frequency band contribution of the paired paths during the change of the blocking margin; verifying the consistency between the synthesized frequency band contribution of each propagation path and the synthesized sound level at the receiving point to obtain a multi-path synthesis residual; combining the path attenuation residual, diffraction transition residual, and multi-path synthesis residual according to the propagation path, frequency band, and receiving point index; and determining the target propagation path based on the correlation between the residuals and adjacent path nodes. The mechanistic residual is not used as a simple training error but retains the correspondence between the propagation path, frequency band, receiving point, and design state that generated the residual.

[0070] Specifically, the path attenuation residual is obtained by comparing the difference in contributed sound level before and after the path length change under the same propagation type with the reference attenuation difference. The diffraction transition residual is obtained by pairing continuous design states where the blocking margin changes from a positive value to a near-zero value and then to a negative value. The multipath synthesis residual is obtained by converting the contributed sound levels of each path to the energy domain, summing them, and comparing them with the synthesized sound level of the receiving point output by the model. If a residual is concentrated only in a single path and adjacent frequency bands, then that path is considered a local anomaly candidate. If multiple residuals are concentrated in a path set that shares the same diffraction boundary node or propagation interface node, then the shared node and its adjacent paths are considered as anomaly candidates. If the residual only appears in the receiving point synthesis stage and the attenuation relationship of each path remains consistent, then the synthesis relationship of the path set under the corresponding receiving point is considered as an anomaly candidate.

[0071] In this embodiment, the mechanistic residuals are used to form a unified evaluation metric according to the following formula:

[0072] ;

[0073] in, This represents the mechanism residual evaluation quantity corresponding to the r-th propagation path, the f-th analysis frequency band, and the j-th receiving point. Represents the path decay residual. Indicates the diffraction transition residual. This represents the composite residual from multiple paths. This represents the residual correlation between the current path and its adjacent paths at shared nodes. , , and These represent the non-negative combination coefficients of the four types of residuals, with a sum of 1. When the four types of combination coefficients are 0.35, 0.30, 0.20, and 0.15 respectively, and the four types of residuals are 4, 3, 2, and 1 respectively, the mechanistic residual evaluation value is 2.85. The values ​​mentioned above ensure that path-level constraints maintain a high proportion in the target propagation path screening, while retaining the correction of abnormal locations by the composition relationship and the adjacent path association relationship.

[0074] Preferably, the target propagation path is determined using a residual space aggregation method. Propagation paths with the same sound source, the same receiving point, and shared path nodes are formed into local path clusters. The mechanistic residual evaluation quantities within the local path clusters are aggregated according to the frequency band continuity relationship. If high residual frequency bands appear continuously between adjacent frequency bands, the corresponding path cluster is retained. If high residuals appear only in isolation in a single frequency band and there is no identical change in the high-precision simulation samples, they are marked as data to be reviewed without directly triggering correction. After the target propagation path is determined, the design variables, path signatures, and parameter perturbation directions that cause the residuals are recorded, so that subsequent local corrections only change the design parameters of the sound barrier that are geometrically related to the target propagation path, while keeping the design variables corresponding to unrelated propagation paths unchanged.

[0075] In one embodiment, generating local correction samples and updating the residual compensation function includes locking the propagation path state outside the target propagation path, applying local perturbations to the design parameters of the sound barrier associated with the target propagation path while maintaining the path topology, forming a local design state, performing the noise simulation prediction model calculation and high-precision noise simulation calculation on the local design state respectively, establishing prediction result pairings according to the same propagation path, frequency band, and receiver point index, constructing local correction samples based on the difference of the pairing results, path length changes, occlusion margin changes, and propagation interface relationships, updating the residual compensation function with the propagation path state as input and the frequency band contribution compensation value as output using the local correction samples, and superimposing the updated compensation value onto the frequency band contribution of the corresponding propagation path. The generation range of the local perturbation is limited by the topological stability interval of the target propagation path to avoid writing the results of different propagation path states into the same compensation relationship.

[0076] In this embodiment, the design parameters of the sound barrier associated with the target propagation path are determined through geometric dependencies. If the target propagation path includes a diffraction boundary node of the sound barrier, design variables that can change the position of the diffraction boundary node, the blocking margin, or the length of the adjacent path are selected. If the target propagation path includes a propagation interface node, design variables that can change the incident segment, the exit segment, or the intersection position of the interface are selected. The local disturbance is expanded from the current design state to the direction of allowable change of the design variables. Each time a local design state is generated, the path signature is recalculated and compared with the path signature of the target propagation path. Only states with consistent node sequence, propagation type sequence, and interface intersection order are retained. The high-precision noise simulation results are redistributed to the corresponding path index according to the path geometry. Results that cannot establish a correspondence with the original path are not included in the current residual compensation function.

[0077] Furthermore, the residual compensation function is calculated using a locally weighted method based on path state similarity:

[0078] ;

[0079] in, This represents the compensation value of the r-th propagation path in the f-th analysis frequency band under the path state vector u. This represents the path state vector of the m-th local correction sample. This represents the difference between the corresponding high-precision simulation result and the prediction result, where M represents the number of local correction samples under the current path signature. This represents the state distance decay coefficient. Let represent the Euclidean distance, where the squared distances between the path state to be corrected and the two samples are 0.1 and 0.4, respectively. When the difference between the two samples is 3 and 1 respectively, the sample that is closer to the other receives a larger combined weight. The calculated compensation value is between 1 and 3 and is close to the difference between the two samples.

[0080] In a preferred embodiment, the path state vector consists of normalized path length, occlusion margin, incident direction encoding, outgoing direction encoding, propagation interface relationship encoding, and frequency band medium attribute encoding. Continuous variables are processed using an interval normalization method based on the current topological stable domain boundary, while discrete variables are processed using a category encoding method that maintains the distinction of propagation type. Sample distance calculation is performed only between samples with the same path signature. When the number of local correction samples is insufficient to cover the neighborhood of the path state to be corrected, the compensation value is not directly extrapolated. Instead, the corresponding path state is added to the high-precision noise simulation queue, and the residual compensation function is recalculated after supplementing the samples. After the compensation values ​​are superimposed, the path attenuation constraint and multi-path synthesis constraint verification are performed again. Compensation results with inconsistent verification are restricted to the interval formed by the difference between adjacent samples.

[0081] In one embodiment, please refer to Figure 4 The detection of propagation path topology changes includes extracting the node sequence, propagation type sequence, interface intersection order, and occlusion margin symbol for each propagation path before and after the update of the designable parameters of the sound barrier, generating path signatures before and after the update, matching the path signatures before and after the update, identifying the addition, disappearance, node replacement, propagation type switching, and interface order change of the propagation path, and dividing the sound propagation path graph into a common subgraph that maintains topological consistency and a difference subgraph that has undergone topological changes based on the matching results, retaining the residual compensation function and local correction samples corresponding to the propagation path in the common subgraph, regenerating the path nodes, candidate edges, and propagation type identifiers of the difference subgraph, and establishing an independent residual index for the regenerated propagation path. The path topology change detection is performed after each candidate parameter update and before the parameter status is written to the official scene version.

[0082] Specifically, the path signature is represented by an ordered combination of fields. The ordered fields sequentially record the source node identifier, intermediate node type and its relative order, receiver node identifier, propagation type of each path edge, intersection order of each propagation interface, and occlusion margin symbol corresponding to each diffraction boundary. Small-range continuous changes in the geometric position of a node do not change the node identifier. The node identifier changes when a node changes from a reachable state to an inaccessible state or migrates from one boundary segment to another discontinuous boundary segment. Different status codes are used when the occlusion margin is positive, zero, or negative. When comparing path signatures, the node sequence is compared first, then the propagation type sequence and the intersection order of the interfaces are compared. After completing the above discrete field comparison, geometric quantities are used to determine whether they belong to the same topological stable domain.

[0083] In this embodiment, the matching degree between the path signature before and after the update is calculated according to the following formula:

[0084] ;

[0085] in, This represents the matching value between the signature of the a-th path before the update and the signature of the b-th path after the update. Indicates the consistency value of the node sequence. Indicates the consistency value of the propagation type sequence. This indicates the consistency value of the intersection order of interfaces. Indicates the sign consistency value of the occlusion margin. , , and These represent the non-negative weights of the four consistency values, with a sum of 1. When the four consistency values ​​are 1, 1, 1, and 0 respectively, and the four weights are 0.35, 0.30, 0.20, and 0.15 respectively, the matching value is 0.85. The corresponding path retains the common nodes and propagation type but enters the occlusion state change check. When the node sequence consistency value is 0, regardless of the other consistency values, the two paths are not directly regarded as the same path.

[0086] Please refer to Figure 5 , Figure 5 The horizontal axis represents the pairing sequence number of the propagation paths before and after the update, and the vertical axis represents the path signature matching value. A higher value indicates a greater degree of consistency between the two paths in terms of node sequence, propagation type sequence, interface intersection order, and occlusion margin sign. The bars represent the comprehensive matching result of each path pairing. Figure 5 The reference lines in the diagram are used to help distinguish between paths that can inherit common subgraphs and paths that need to enter the differential subgraph for reconstruction.

[0087] from Figure 5As can be seen, the matching values ​​of some path pairings are significantly higher than the reference line, indicating that these paths maintained the same main propagation structure before and after the parameter update, and can continue to inherit the original local correction samples and residual compensation functions. Several bars are also below the reference line, indicating that their node sequences, propagation types, or interface intersection order have changed, and they cannot be directly considered the same path; especially when the node sequence consistency is zero, even if other indicators are high, they should be included in the difference subgraph processing.

[0088] Furthermore, the common subgraph consists of propagation paths whose path signatures remain consistent and whose geometric state is still within the coverage of the original local correction samples. The difference subgraph includes newly added paths, disappearing paths, paths where path nodes have been replaced, paths where propagation types have been switched, and paths where the path signatures remain consistent but the path state exceeds the coverage of the original samples. The residual compensation function and sample index in the common subgraph are directly inherited to the updated scenario version. The original compensation function in the difference subgraph is only saved as a historical record and does not participate in the calculation of new paths. When regenerating the difference subgraph, the local search range is the boundary nodes that have changed and their adjacent propagation interfaces, and connections are established with the entry and exit nodes of the common subgraph to avoid re-performing candidate node expansion on all paths that are unrelated to topology changes.

[0089] In one embodiment, the high-precision noise simulation for the target propagation path includes jointly ranking the target propagation paths based on the mechanistic residual amplitude, the contribution ratio of the target propagation path to the synthesized sound level at the receiving point, and the local sensitivity of the designable parameters of the associated sound barrier. For the target propagation paths selected by the ranking result, mutually independent local perturbation combinations are generated within the parameter range while keeping the common subgraph unchanged. Path signatures are pre-calculated for each local perturbation combination. Only combinations whose path signatures match the target propagation path are submitted to the high-precision noise simulation. When the path signature corresponding to a local perturbation combination changes, the topology change range is determined based on the parameter state before and after the path signature change. The topology change range is divided into sampling ranges for different path states, and local correction samples are generated for each. The joint ranking only determines the processing order of the high-precision noise simulation and does not delete target propagation paths that have not yet met the correction conditions.

[0090] In this embodiment, the contribution ratio of the target propagation path to the synthesized sound level at the receiving point is determined by the energy domain proportion. The local sensitivity of the associated design parameters is calculated by the perturbation results of maintaining the same path signature on both sides of the current design state. The mechanism residual amplitude, contribution ratio, and local sensitivity are normalized to a unified numerical range to form a sorted value. The local perturbation combination is generated by variable separation. Each combination only changes the design variables that have a geometric dependence on the current target propagation path, while keeping the other design variables fixed. When multiple design variables jointly determine the position of the same diffraction boundary node, a joint perturbation combination containing the multiple design variables can be generated. However, it is necessary to confirm through path signature pre-calculation that it does not change the node sequence and propagation type sequence of the common subgraph.

[0091] Preferably, when the path signatures corresponding to two adjacent local disturbance states are different, the design variable intervals of the two states are taken as the initial topology change intervals. Within the initial topology change intervals, intermediate design states are generated and path signatures are calculated using an interval decreasing method until the continuous parameter intervals on both sides of the path signature change can be distinguished. Samples located in the left path state are only written into the left residual compensation function, and samples located in the right path state are only written into the right residual compensation function. Boundary states where the path signature cannot be stably determined are saved as topology boundary samples. The topology boundary samples are used to determine the intervals that parameter updates cannot directly cross and do not participate in the conventional weighted calculation of the compensation function on either side.

[0092] In one embodiment, the updated residual compensation function further includes, based on the path signature corresponding to the local correction samples, classifying samples with the same propagation type sequence into the path common sample set, and classifying samples with the same node adjacency relationship and interface intersection order into the path local sample set; extracting common residual components from the path common sample set that vary with path length and propagation type, and extracting local residual components from the path local sample set that vary with occlusion margin and interface relationship; weighting and combining the common residual components and local residual components according to the sample coverage density of the target propagation path in the path state space; and applying adjacent frequency band continuity constraints and path energy synthesis constraints at the same receiving point to the combined compensation value to obtain the updated residual compensation function.

[0093] Specifically, the path common sample set can come from different parametric design scenarios, but requires consistent propagation type sequence, corresponding frequency band medium attribute encoding, and path length within a comparable range. The common residual component fits the relationship between path length change and frequency band residual change after eliminating the local occlusion margin and interface position differences of each sample. The path local sample set is limited to the same node adjacency relationship and interface intersection order. The local residual component fits the residual difference generated by the occlusion margin, incident direction, and outgoing direction changes after eliminating the common residual component. When the local sample distribution in the target path neighborhood is dense, the combined weight of the local residual component is increased. When there are only a few local samples in the target path neighborhood, the combined weight of the common residual component is increased, and the output compensation value is limited to the residual envelope of the same path signature sample.

[0094] Furthermore, the continuity constraint of adjacent frequency bands is determined by comparing the change direction of the current frequency band compensation value with that of the adjacent frequency band compensation values ​​on both sides. When the current frequency band compensation value forms an isolated abrupt change that cannot be explained by the characteristics of the sound source frequency band or the properties of the acoustic medium, the current value is re-estimated by combining the compensation values ​​of adjacent frequency bands and the original correction sample of the current frequency band. The path energy synthesis constraint of the same receiving point is achieved by converting the compensated path contribution sound level to the energy domain and summing it, and then comparing it with the synthesized sound level of the receiving point in the high-precision noise simulation. When the synthesis difference exceeds the local sample residual envelope, the synthesis difference is redistributed according to the energy contribution ratio of each path at the current receiving point, and the path attenuation residual and diffraction transition residual verification are re-executed.

[0095] In one embodiment, mapping the corrected frequency band contribution to the designable parameters of the sound barrier and performing parameter updates includes: calculating the local sensitivity of the designable parameters of the sound barrier relative to each receiving point and each frequency band sound level constraint based on the frequency band contribution of each propagation path after residual compensation; constructing a sensitivity matrix between the design parameters and the sound level constraint; determining the topological stability domain based on the parameter range in which the path signature remains unchanged; solving the parameter update amount within the topological stability domain that is simultaneously limited by the design parameter value boundary and the sound level constraint; generating candidate design states using the parameter update amount and recalculating the path signature; writing the parameterized design scenario when the path signature of the candidate design states remains consistent; and revoking the parameter update amount and performing difference subgraph reconstruction when the path signature changes.

[0096] In this embodiment, local sensitivity is obtained by applying positive and negative perturbations to a single design variable within the topological stability domain. The positive perturbation state, the current state, and the negative perturbation state have the same path signature. The path frequency band contribution and the synthesized sound level at the receiving point are calculated in the three states after residual compensation. The corresponding elements of the sensitivity matrix are formed based on the local difference of the synthesized sound level as the design variable changes. If a design variable changes multiple propagation paths at the same time, the sensitivity element corresponding to the design variable is determined by the changes in the contribution of each path after synthesis in the energy domain. If a receiving point has multiple frequency band constraints, a constraint row is established for each frequency band, and the corresponding index of the receiving point and the propagation path is retained.

[0097] In a preferred embodiment, the parameter update amount is calculated using the following formula:

[0098]

[0099] ;

[0100] in, Let S represent the obtained design parameter update vector, S represent the sensitivity matrix between the design parameters and the frequency band constraints of the receiving point, and e represent the deviation vector of the current frequency band sound level of the receiving point relative to the corresponding constraint condition. This represents a non-negative coefficient that limits the magnitude of a single parameter change. This represents the current design parameter vector, where l and h represent the lower and upper bound vectors of the allowable range of the design parameters, respectively. This represents the topological stability domain corresponding to the current path signature. When the current value of a certain design variable is 4, the allowable range is 3 to 6, and the topological stability domain is 3.5 to 5.2, even if the updated value obtained by the unconstrained solution is 5.6, the candidate value will be restricted to the topological stability domain and the solution will be re-solved instead of directly using 5.6.

[0101] Specifically, the topological stability region is jointly determined by local perturbation samples that maintain the current path signature. For a single design variable, the continuous value interval before the path signature changes is taken as the stability interval of the design variable. For the case where multiple design variables jointly affect the path topology, the local design states with verified consistent path signatures constitute the stability region sample set, and the local region for executable parameter updates is determined by the connectivity between adjacent samples. After the parameter update vector is generated, several intermediate states are inserted according to the connection from the current design state to the candidate design state, and the path signature is calculated. When all intermediate states maintain the current path signature, the candidate state is written into the parameterized design scenario. When any intermediate state experiences a change in path signature, the candidate state is revoked.

[0102] In one embodiment, when performing parameter updates within the boundary neighborhood of the topologically stable domain, the method further includes obtaining path signatures and acoustic propagation path maps on both sides of the topological change interval, establishing a transitional correspondence between newly added propagation paths and disappearing propagation paths based on common path nodes and propagation interfaces, calculating the frequency band contribution within the boundary neighborhood using residual compensation functions on both sides of the topological change interval, generating combined weights for the acoustic propagation path maps on both sides based on occlusion margin, interface intersection distance, and path node visibility status, forming a segmented frequency band contribution mapping of the boundary neighborhood according to the combined weights, calculating candidate parameter update amounts based on the segmented frequency band contribution mapping, and splitting the candidate parameter update amounts that cross the topological change interval into continuous update amounts located within the topologically stable domains on both sides.

[0103] In this embodiment, the transition correspondence uses the common sound source node, common sound receiver node, and unchanged intermediate path node in the path diagram on both sides of the topology change interval as anchor points. The path segment of the disappearing path after the last common node and the path segment of the new path after the same common node are established as transition correspondences. When the two paths share the same propagation interface but the diffraction boundary node is replaced, the correspondence is established according to the intersection position of the interface and the propagation direction. When one side is a direct path and the other side is a diffraction path, the correspondence is established according to the boundary state where the occlusion margin is close to zero. The segmented frequency band contribution mapping is only used to generate candidate update directions close to the topology boundary and does not merge the compensation functions on both sides into a unified compensation function.

[0104] Furthermore, the combined weights of the two path graphs are calculated according to the following formula:

[0105]

[0106] ;

[0107] in, The combined weights represent the path states on the left side of the topology change interval. The combined weights represent the path states to the right of the topology change interval. This represents the combined distance between the candidate design state and the left-side stable state, formed by the difference in occlusion margin, the difference in interface intersection distance, and the difference in the visible state of path nodes. This represents the combined distance between the candidate design state and the right-hand steady state. Represents the distance attenuation coefficient, when Take 0.2, Take 0.8 and When the value is 2, the left-side path states receive a larger combined weight. When the candidate design state moves to the right-side stability region, The weights decrease and the right-side combination weights increase accordingly; these weights are used only within the corresponding topological change interval.

[0108] Please refer to Figure 6 , Figure 6 The horizontal axis represents the sampling location within the topological boundary neighborhood, and the vertical axis represents the combined weight of the path states on both sides of the boundary, ranging from 0 to 1. The solid line represents the weight of the stable state on the left, the dashed line represents the weight of the stable state on the right, and the horizontal reference line indicates the position where the weights on both sides are equal. The curve changes with the boundary neighborhood position, reflecting the piecewise mapping relationship when the candidate design state transitions from the stable domain on the left to the stable domain on the right.

[0109] from Figure 6 As can be seen, in the region near the stable state on the left, the weight of the left side is significantly higher than that of the right side, indicating that the frequency band contribution should be mainly determined by the left path map and the left residual compensation function. As the sampling position moves to the right, the weight of the left side gradually decreases, while the weight of the right side gradually increases, forming a dominant relationship transition near the middle of the boundary. This change indicates that the boundary neighborhood cannot simply use the results from either side, but should be smoothly combined based on distance relationships.

[0110] In a preferred embodiment, the update amount of candidate parameters spanning the topology change interval is split in the order from the current design state to the state to the left of the topology boundary, and from the state to the right of the topology boundary to the target design state. The current design state is first updated to the boundary neighborhood within the original topology stable domain and a high-precision noise simulation verification is completed. Then, a residual compensation function for the right path state is established based on the reconstruction result of the difference subgraph. Parameter updates are then performed from the starting point of the right stable domain. Parameter intervals located between two stable domains and with unstable path signatures are not considered as the dwell state of the design iteration. If a continuous update direction that satisfies the sound level constraint cannot be obtained in the right stable domain, the update combination of other design variables is recalculated in the left stable domain.

[0111] In one embodiment, the output of a sound barrier design scheme that satisfies noise constraints includes recording the sound propagation path diagram, path signature, mechanism residual, local correction samples, and residual compensation function corresponding to each parameter update, forming an iterative data chain associated with the candidate design state. After each parameter update, the propagation path to be reviewed is determined according to the sound level constraint deviation of the receiving point, the frequency band contribution ratio of the propagation path, and the mechanism residual. High-precision noise simulation is performed on the propagation path to be reviewed, and the review result is written into the iterative data chain. When the frequency band difference between the review result and the frequency band contribution after residual compensation exceeds the residual envelope of the local correction sample of the corresponding path signature, the corresponding topological stability domain is returned to supplement the local correction sample and the parameter update amount is recalculated. When the reviewed sound level of each receiving point satisfies the sound level constraint and the path signature no longer changes, the designable parameters of the sound barrier in the current parameterized design scenario are output.

[0112] Specifically, the iterative data chain uses the scenario version as the main index. Each scenario version stores the parent version identifier, design parameter vector, sound propagation path graph version, path signatures, mechanism residual records, local design state used when calling high-precision noise simulation, local correction sample index, residual compensation function version, parameter update vector, and verification state. When returning to the corresponding topological stability domain, the design parameters and path graph are restored according to the parent version identifier. Candidate scenario versions that have been generated but not adopted are not deleted. The unadopted versions are marked as backtracking branches and the path signature changes or residual envelope exceedance records that caused the backtracking are retained to avoid subsequent iterations from repeatedly using the same invalid parameter update direction.

[0113] In this embodiment, the envelope of the local corrected sample residuals is determined according to the following formula:

[0114] ;

[0115] in, This represents the upper bound of the residual envelope of the r-th propagation path in the f-th analysis frequency band. This represents the median of the absolute values ​​of the differences between locally corrected samples under the same path signature. This represents the median deviation of the absolute value of the locally corrected sample difference relative to the median. Let represent the envelope expansion coefficient. When the absolute values ​​of the locally corrected sample differences are 1, 1.5, 2, 2.5, and 6, respectively, the median is 2 and the median deviation is 0.5. When the value is 3, the upper bound of the residual envelope is 3.5. Samples with a verification difference of 6 do not directly change the current parameter update result, but instead trigger supplementary sampling of the corresponding path state neighborhood and update of the residual compensation function.

[0116] Please refer to Figure 7 , Figure 7 The horizontal axis represents the sequence number of the path signature and analysis frequency band combination, and the vertical axis represents the upper bound of the verification difference and residual envelope, in dB. The bars represent the latest verification difference, and the line represents the upper bound of the envelope calculated from samples of the same path signature. When the bar is below the line, it indicates that the verification difference of the path frequency band combination is still within the coverage of the current sample; when the bar is close to or exceeds the line, it suggests that additional sampling or a re-update of the compensation function is needed.

[0117] from Figure 7 As can be seen, most of the verification differences are below the upper bound of the envelope, indicating that the current residual compensation function has acceptable local stability for these path frequency band combinations. If the bar of a certain combination is significantly higher than that of the adjacent combination, even if it has not yet crossed the upper bound of the envelope, it should be checked for isolated abrupt changes, changes in interface order, or insufficient sample coverage by combining the mechanistic residual plot and the path signature matching plot.

[0118] Preferably, the propagation path to be verified is determined by a joint index of the sound level constraint deviation of the sound-receiving point, the energy contribution of the path frequency band, and the mechanism residual. When the sound level of the sound-receiving point is close to the constraint condition and is dominated by a small number of paths, the dominant path is included in the verification range. When the path frequency band contribution is low but the mechanism residual is continuously concentrated in adjacent frequency bands, it is also included in the verification range. During the verification, the current parametric design scenario and path signature are fixed, and the high-precision noise simulation results are saved using the same propagation path indexing method as the local correction samples. Verification results that cannot establish a correspondence with the current path signature trigger the reconstruction of the difference subgraph without being written into the original residual compensation function.

[0119] In a specific application, the parametric design scenario includes a set of sound sources, multiple receiving points, continuous terrain boundaries, several propagation interfaces, and a parametric object of a sound barrier. In the initial design state, direct paths, barrier boundary diffraction paths, ground propagation paths, and interface reflection paths are generated. After the noise simulation prediction model outputs the frequency band contribution of each path, the mechanism residuals identify that some diffraction paths have continuity deviations when the occlusion margin is close to zero. The computer-aided design program locks the states of irrelevant paths and generates local disturbance states only for the sound barrier design parameters associated with the corresponding diffraction boundary. For states that maintain the same path signature, high-precision noise simulation is performed and the residual compensation function is updated. Then, based on the compensated frequency band contribution, a sensitivity matrix is ​​constructed and the parameter update is calculated in the current topological stability domain.

[0120] When parameter updates cause the original direct path to become a diffracted path, the computer-aided design program compares the node sequence, propagation type sequence, interface intersection order, and occlusion margin sign before and after the update. The ground propagation path and reflection path that remain unchanged are assigned to the common subgraph, while the path that changes from direct to diffracted is assigned to the difference subgraph. The existing local correction samples and residual compensation functions in the common subgraph are retained. Diffracted boundary nodes and path indices are regenerated for the difference subgraph. Local correction sample sets are established on both sides of the topology change interval. The parameter update that crosses the boundary is split into two continuous update processes through segmented frequency band contribution mapping. The sensitivity matrix is ​​recalculated after entering the new topological stability domain.

[0121] When the parameter iteration is nearing completion, the computer-aided design program reads the frequency band sound level constraint deviation, propagation path contribution ratio, and mechanism residual of each receiving point. It selects the propagation path that is still close to the constraint conditions or residual envelope boundary to perform high-precision noise simulation verification. When the verification difference is within the residual envelope of the local correction sample of the corresponding path signature, the verification result is written to the current scene version and the sample coverage record is updated. When the verification difference exceeds the residual envelope, it returns to the topological stability domain, supplements the local correction sample in the neighborhood of the current path state, and recalculates the residual compensation function and parameter update amount until the verification sound level of each receiving point meets the corresponding constraint, the frequency band difference of the path to be verified is within the residual envelope, and the path signature remains stable during continuous parameter updates. Then, it outputs the design parameters of the sound barrier, the corresponding sound propagation path diagram, and the calculation record associated with the design state from the current scene version.

Claims

1. A computer-aided design method for sound barriers based on a noise simulation and prediction model, characterized in that, include: Obtain sound source parameters, receiver parameters, terrain boundary parameters, acoustic medium parameters, and designable parameters of the sound barrier to establish a parametric design scenario; A sound propagation path diagram is generated based on the visibility relationship between the sound source and the receiving point, the occlusion boundary, and the propagation interface. Input the sound propagation path diagram into the noise simulation prediction model to obtain the frequency band contribution of each propagation path and the synthesized sound level at the receiving point; Based on path attenuation constraints, diffraction transition constraints, and multipath synthesis constraints, the target propagation path in the residual set is determined by computer. For the target propagation path, a high-precision noise simulation is invoked to generate local correction samples and update the residual compensation function; The corrected frequency band contribution can be mapped to the sound barrier design parameters and the parameters can be updated. When a change in the propagation path topology is detected, the sound propagation path diagram and residual compensation function are reconstructed, and a sound barrier design scheme that meets noise constraints is output.

2. The computer-aided design method for sound barriers based on a noise simulation and prediction model according to claim 1, characterized in that, Generating the sound propagation path map includes: using the sound source, the receiving point, the diffraction boundary of the sound barrier, and the propagation interface as path nodes, and establishing a candidate node sequence according to the spatial connectivity order between the sound source and the receiving point; Perform visibility and interface intersection checks on adjacent path nodes, and delete candidate edges that traverse non-propagable regions or do not meet the propagation order. Based on the mechanisms of direct propagation, occlusion diffraction, interface reflection, and ground propagation, propagation type identifiers are configured for the retained candidate edges, and path length, incident direction, outgoing direction, occlusion margin, and frequency band medium properties are recorded. Candidate paths with the same node sequence and propagation type sequence are merged into a propagation path. A directed graph is constructed according to the correspondence between the propagation path and the receiving point to form a sound propagation path graph for the noise simulation and prediction model to read.

3. The computer-aided design method for sound barriers based on a noise simulation and prediction model according to claim 1, characterized in that, The computer-generated residual includes: for each propagation path, comparing the frequency band contribution output by the noise simulation prediction model with the attenuation benchmark determined based on the path length and frequency band medium properties to obtain the path attenuation residual; For the sound barrier, the propagation paths with the same propagation type sequence under adjacent disturbance states of the design parameters are paired, and the diffraction transition residual is obtained based on the continuous relationship of the frequency band contribution of the paired paths during the process of changing the blocking margin. The consistency of the synthesized frequency band contribution of each propagation path with the synthesized sound level at the receiving point is verified to obtain the multi-path synthesis residual. The path attenuation residual, the diffraction transition residual, and the multipath synthesis residual are combined according to the propagation path, frequency band, and receiver point index, and the target propagation path is determined based on the correlation between the residuals and adjacent path nodes.

4. The computer-aided design method for sound barriers based on a noise simulation and prediction model according to claim 1, characterized in that, Generating the local correction sample and updating the residual compensation function includes: locking the propagation path state outside the target propagation path, applying a local perturbation to the sound barrier design parameters associated with the target propagation path to keep the path topology unchanged, and forming a local design state; Noise simulation prediction model calculation and high-precision noise simulation calculation are performed on the local design state respectively, and the prediction results are paired according to the same propagation path, frequency band and receiving point index. Local correction samples are constructed based on the difference in pairing results, changes in path length, changes in occlusion margin, and propagation interface relationships. The residual compensation function, which takes the propagation path state as input and the frequency band contribution compensation value as output, is updated using the local correction samples, and the updated compensation value is superimposed on the frequency band contribution of the corresponding propagation path.

5. The computer-aided design method for sound barriers based on a noise simulation prediction model according to claim 2, characterized in that, Detecting the topological changes of the propagation path includes: extracting the node sequence, propagation type sequence, interface intersection order, and occlusion margin symbol of each propagation path before and after the update of the design parameters of the sound barrier, and generating the path signature before the update and the path signature after the update. The path signature before and after the update are matched to identify the addition, disappearance, node replacement, change of propagation type, and change of interface order in the propagation path. Based on the matching results, the sound propagation path graph is divided into a common subgraph that maintains topological consistency and a difference subgraph that undergoes topological changes; The residual compensation function and the local correction sample corresponding to the propagation path in the common subgraph are retained, the path nodes, candidate edges and propagation type identifiers of the difference subgraph are regenerated, and an independent residual index is established for the regenerated propagation path.

6. The computer-aided design method for sound barriers based on a noise simulation prediction model according to claim 5, characterized in that, The high-precision noise simulation for the target propagation path includes: jointly ranking the target propagation path based on the mechanism residual amplitude, the contribution ratio of the target propagation path to the synthesized sound level at the receiving point, and the local sensitivity associated with the designable parameters of the sound barrier; For the target propagation path selected based on the sorting results, generate mutually independent combinations of local perturbations within the parameter range while keeping the common subgraph unchanged; For each combination of local disturbances, a path signature is pre-calculated, and only the combination whose path signature matches the target propagation path is submitted to the high-precision noise simulation. When the path signature corresponding to the local disturbance combination changes, the topology change interval is determined according to the parameter state before and after the path signature change. The topology change interval is divided into sampling intervals for different path states, and local correction samples are generated respectively.

7. The computer-aided design method for sound barriers based on a noise simulation prediction model according to claim 4, characterized in that, Updating the residual compensation function further includes: according to the path signature corresponding to the local correction sample, classifying samples with the same propagation type sequence into the path common sample set, and classifying samples with the same node adjacency relationship and interface intersection order into the path local sample set; Common residual components of the frequency band residuals that vary with path length and propagation type are extracted from the common sample set of the path, and local residual components of the frequency band residuals that vary with occlusion margin and interface relationship are extracted from the local sample set of the path. According to the sample coverage density of the target propagation path in the path state space, the common residual component and the local residual component are weighted and combined. By applying adjacent frequency band continuity constraints and path energy synthesis constraints at the same receiving point to the combined compensation value, the updated residual compensation function is obtained.

8. The computer-aided design method for sound barriers based on a noise simulation and prediction model according to claim 5, characterized in that, Mapping the corrected frequency band contribution to the designable parameters of the sound barrier and performing parameter updates includes: calculating the local sensitivity of the designable parameters of the sound barrier relative to the sound level constraints of each sound receiving point and each frequency band based on the frequency band contribution of each propagation path after residual compensation, and constructing a sensitivity matrix between the design parameters and the sound level constraints; The topological stability region is determined based on the parameter range in which the path signature remains unchanged, and the parameter update amount, which is simultaneously limited by the design parameter value boundary and sound level constraint, is solved within the topological stability region. The parameter update amount is used to generate candidate design states and recalculate path signatures. When the path signatures of the candidate design states are consistent, they are written into the parameterized design scenario. When the path signatures change, the parameter update amount is revoked and the difference subgraph reconstruction is performed.

9. The computer-aided design method for sound barriers based on a noise simulation prediction model according to claim 8, characterized in that, When performing parameter updates within the boundary neighborhood of the topologically stable domain, the method further includes: obtaining path signatures and acoustic propagation path graphs on both sides of the topological change interval, and establishing a transitional correspondence between newly added propagation paths and disappearing propagation paths based on common path nodes and propagation interfaces. The frequency band contribution in the boundary neighborhood is calculated by using the residual compensation function on both sides of the topology change interval, and the combined weight of the acoustic propagation path map on both sides is generated based on the occlusion margin, interface intersection distance and path node visibility status. The segmented frequency band contribution mapping of the boundary neighborhood is formed according to the combined weights; The candidate parameter update is calculated based on the segmented frequency band contribution mapping, and the candidate parameter update across the topology change interval is split into continuous update amounts located in the topology stable domains on both sides.

10. The computer-aided design method for sound barriers based on a noise simulation prediction model according to claim 8, characterized in that, The output of a sound barrier design scheme that satisfies the noise constraints includes: recording the sound propagation path diagram, path signature, mechanism residual, local correction sample and residual compensation function corresponding to each parameter update, forming an iterative data chain associated with the candidate design state; After each parameter update, the propagation path to be verified is determined according to the sound level constraint deviation at the sound receiving point, the frequency band contribution ratio of the propagation path, and the mechanism residual. High-precision noise simulation is performed on the propagation path to be verified, and the verification results are written into the iterative data chain. When the frequency band difference between the verification result and the frequency band contribution after residual compensation exceeds the residual envelope of the local correction sample of the corresponding path signature, return to the corresponding topological stability domain to supplement the local correction sample and recalculate the parameter update amount. When the verified sound level at each sound receiving point meets the sound level constraint and the path signature no longer changes, output the designable parameters of the sound barrier in the current parametric design scenario.

Images