Building sound wave reflection path analysis and noise source positioning method and system

Abstract

The invention discloses a building sound wave reflection path analysis and noise source positioning method and system, and the method comprises the steps: S1, arranging a plurality of sound sensors according to the geometric structure of a target building, collecting sound signals, carrying out the signal sampling of the sound signals, and storing the sound signals in the form of a time sequence; s2, preprocessing the sound signal, extracting a time difference matrix, a frequency spectrum feature and an energy feature, merging into a three-dimensional time sequence feature, and normalizing to form a three-dimensional time sequence feature matrix; s3, establishing an LSTM network, and predicting a three-dimensional position and a sound wave reflection path of an output noise source; s4, optimizing and calculating a sound wave reflection path by using the geometric structure data of the target building, and generating a three-dimensional reflection path diagram; and S5, dynamically displaying a noise source position and a sound wave reflection path according to the three-dimensional reflection path diagram. According to the invention, high-precision three-dimensional positioning of a noise source can be realized in a complex building environment, accurate modeling is realized, and propagation and reflection paths of sound waves in a building are analyzed.

Description

Technical Field

[0001] The present invention relates to the technical fields of architectural acoustics and artificial intelligence, and particularly to a method and system for analyzing the reflection path of building sound waves and locating noise sources. Background Art

[0002] With the continuous development of architectural acoustics technology, noise control and acoustic optimization inside buildings have become important contents for improving the quality of the building environment. In indoor acoustic analysis, the accurate positioning of noise sources and the modeling of sound wave reflection paths are key technical links. However, multi-path reflections and dynamic noise sources in complex building environments pose great challenges to traditional acoustic analysis methods.

[0003] The physical modeling method, sensor array method, and shallow machine learning methods widely used in the prior art have problems of insufficient accuracy and real-time performance when dealing with multi-reflection paths and complex sound fields. Traditional physical modeling methods rely on building geometric structures and sound wave propagation theories, with complex calculations and poor adaptability to dynamic changes in the building environment, making it difficult to meet real-time requirements; the traditional method of the sensor array method for calculating the time difference of arrival (TDOA) through multi-point sensors performs well in simple environments, but in multi-reflection path or strong background noise scenarios, the positioning accuracy significantly decreases; although some shallow machine learning models (such as support vector machines and simple neural networks) improve the automation degree of noise analysis, their modeling ability for time series data is insufficient, making it difficult to capture key features in multi-path reflections.

[0004] The prior art has significant deficiencies in terms of accuracy, real-time performance, and adaptability to complex scenarios, and it is difficult to meet the actual needs of modern architectural acoustic optimization and noise management. Therefore, there is an urgent need for an efficient, accurate, and highly adaptable solution to meet the needs of modern architectural acoustic optimization and noise management. Summary of the Invention

[0005] To solve the problems of insufficient accuracy in locating complex building noise sources and poor adaptability of models in multi-reflection path environments in the prior art, the present invention proposes a method and system for analyzing the reflection path of building sound waves and locating noise sources.

[0006] The specific technical solution is as follows: A method for analyzing the reflection path of building sound waves and locating noise sources, the steps including:

[0007] S1: According to the geometric structure of the target building, arrange a number of sound sensors, collect sound signals, perform signal sampling on the sound signals, and store them in the form of a time series.

[0008] S2: Preprocess the sound signals, extract the time difference matrix, spectral features, and energy features, merge them into three-dimensional time series features, and normalize them to form a three-dimensional time series feature matrix.

[0009] S3: Establish an LSTM network. Input the three-dimensional time series feature matrix in the form of a time series window. The LSTM layer models the propagation pattern and reflection characteristics of the sound signal in the building space, conducts time series analysis, uses the attention mechanism to weight and generate a context vector for noise source localization, and predicts and outputs the three-dimensional position of the noise source and the sound wave reflection path;

[0010] S4: Optimize and analyze the sound wave reflection path using the geometric structure data of the target building. The LSTM network learns historical acoustic data and dynamically simulates the three-dimensional position of the noise source and the sound wave reflection path to generate a three-dimensional reflection path map;

[0011] S5: Dynamically display the position of the noise source and the sound wave reflection path according to the three-dimensional reflection path map.

[0012] Furthermore, the arrangement method of the sound sensors includes: Based on the acoustic propagation characteristics of the building space, configure the sound sensor array as at least one topological structure among a regular grid type, a hierarchical coupling type, a partition gradient type, and a linear waveguide type, where:

[0013] Regular grid type: Uniformly arrange in a three-dimensional grid within a symmetric open space, and capture the reflection symmetry of the sound field through multi-node synchronous sampling;

[0014] Hierarchical type: Deploy vertically along multiple floors of a building, and analyze the vertical sound field superposition effect by combining the time delay correlation between adjacent layer sensor groups;

[0015] Partition type: Divide the sound field mutation area in a complex space, and dynamically encrypt the sensors based on the reflection path density to form an acoustic energy attenuation gradient tracking network;

[0016] Linear type: Arrange a sensor chain along the axis of a long and narrow channel, and analyze the acoustic wave guiding mode through the axial time of arrival time series.

[0017] Furthermore, the sound signal features stored in the time series form include:

[0018] Time difference: Used to calculate the distance between the noise source and the sensor;

[0019] Spectrum feature: Extract the frequency distribution of the signal through fast Fourier transform;

[0020] Energy feature: Record the change of the signal intensity over time.

[0021] Furthermore, the LSTM network includes:

[0022] Input layer: Receive the sound signals collected by the sound sensors and input them in the form of a time series window;

[0023] Convolutional layer: Extract time-frequency features to reduce noise interference and improve the network's adaptability to different noise types. 1D or 2D convolutional neural networks are used for feature extraction;

[0024] LSTM layer: Model the propagation pattern and reflection characteristics of sound signals in the building space for time series analysis;

[0025] Attention mechanism: Automatically focus on key time steps to improve the model's sensitivity to important noise signals;

[0026] Fully connected layer: Convert the output of the LSTM layer into the final localization result, output three-dimensional coordinates, and optimize the predicted noise source position;

[0027] Output layer: Output the three-dimensional position of the predicted noise source and the probability distribution of the reflection path generated by combining the LSTM layer and the attention mechanism.

[0028] Furthermore, the LSTM layer includes a first LSTM, a second LSTM, and a Dropout mechanism:

[0029] First LSTM: Capture local time dependencies and learn the acoustic wave change patterns within a short time range;

[0030] Second LSTM: The second LSTM is a bidirectional LSTM, which enhances bidirectional dependency modeling and improves the time series modeling ability;

[0031] Dropout mechanism: The dropout probability is 0.2 - 0.5, which is used to prevent overfitting.

[0032] Furthermore, the attention mechanism includes:

[0033] Weight calculation: For each time step and hidden state calculate the main force weights :

[0034] , ,

[0035] where , , are learnable parameter matrices;

[0036] Weighted summation: Weight the time series features according to the weights to generate a context vector:

[0037] ,

[0038] The context vector contains the most critical information for noise source localization and serves as the output of the attention mechanism.

[0039] Further, the acoustic wave reflection path analysis includes:

[0040] Optimizing the calculation of the acoustic wave reflection path using building geometric structure data, and simulating the acoustic wave reflection path by the mirror sound source method or ray tracing;

[0041] Dynamically simulating the reflection path: The LSTM network learns historical acoustic data, inputs the time series data received by the sound sensor and the building reflection surface data, predicts the possible acoustic wave reflection path and adjusts it in real time, and outputs the reflection path of the acoustic wave and the corresponding energy distribution;

[0042] Generating a reflection path map: Matching the prediction result with the building reflection surface, generating a three-dimensional acoustic wave reflection path map, calculating the energy distribution, visualizing the finally calculated acoustic wave reflection path in three dimensions, and dynamically rendering it on the building three-dimensional model.

[0043] Further, the building geometric structure data includes the material of the wall surface, the position of the wall surface, the reflection coefficient, sound absorption coefficient and light transmittance of the reflection surface.

[0044] Further, the three-dimensional visualization of the acoustic wave reflection path is performed by real-time simulation using OpenGL or Unity3D, which is used to display the distribution of key reflection surfaces, the acoustic wave propagation path and the acoustic information of different areas of the building, and mark the acoustic wave reflection paths with different energy distributions respectively.

[0045] A building acoustic wave reflection path analysis and noise source localization system, applicable to a building acoustic wave reflection path analysis and noise source localization method, includes:

[0046] Multi-point sensor acquisition module: Highly sensitively acquiring the time difference, frequency spectrum and energy characteristics of the acoustic wave;

[0047] Data preprocessing module: Denoising the acquired acoustic wave signal, removing background noise and non-target frequency band signals, and normalizing the feature data;

[0048] LSTM network module: Constructing time series features and a deep model to realize the prediction of the noise source position and the acoustic wave reflection path, capturing the long-term dependencies in the time series through hierarchical design, and introducing an attention mechanism to dynamically focus on the key time step features to filter out redundant information;

[0049] Acoustic wave reflection path analysis module: Combining the LSTM network prediction result and the building geometric model to simulate the propagation and reflection process of the acoustic wave and generate an accurate path map;

[0050] Real-time Visualization and Alarm Module: Provides a user-friendly 3D modeling interface to display noise sources and reflection paths, and offers alarms and optimization suggestions when the noise exceeds the standard, quickly locating and solving noise problems.

[0051] The above technical solutions have the following advantages or technical effects:

[0052] 1. By optimizing the noise source localization accuracy and enhancing the modeling adaptability in complex scenarios, the present invention achieves high real-time performance, significantly superior to traditional physical modeling and shallow learning methods. It can dynamically simulate the sound wave reflection path, and trigger alarms and optimization suggestions when the noise exceeds the standard, providing efficient and reliable technical support for architectural acoustics design, noise monitoring, and environmental optimization, and being applicable to various complex building scenarios.

[0053] 2. Through the reasonable arrangement of multi-point sensors and high-precision data acquisition, combined with the LSTM network and attention mechanism in deep learning technology, the present invention realizes the three-dimensional precise localization of noise sources and the dynamic modeling of sound wave reflection paths in complex building acoustic environments. Based on time series data, key features such as time difference, spectral features, and energy distribution are extracted, and the attention mechanism dynamically focuses on key path information, enhancing the modeling ability and adaptability of the model in multi-reflection path scenarios. Description of the Drawings

[0054] Figure 1 is a schematic flowchart of the method of the present invention;

[0055] Figure 2 is a schematic diagram of the connection of system modules of the present invention. Detailed Embodiments

[0056] To make the technical solutions of the present invention clearer, the following further describes the present invention in detail with reference to the accompanying drawings and specific embodiments.

[0057] Embodiment 1

[0058] As Figure 1 shown, a method for analyzing the sound wave reflection path and locating noise sources in a building, the steps include:

[0059] S1: According to the geometric structure of the target building, arrange a number of sound sensors, collect sound signals, perform signal sampling on the sound signals and store them in the form of time series;

[0060] S2: Preprocess the sound signals, extract the time difference matrix, spectral features, and energy features, merge them into three-dimensional time series features, and normalize them to form a three-dimensional time series feature matrix;

[0061] S3: Establish an LSTM network, input the three-dimensional time series feature matrix in the form of a time series window. The LSTM layer models the propagation pattern and reflection characteristics of sound signals in the building space, conducts time series analysis, uses the attention mechanism to weight and generate a context vector for noise source localization, and predicts and outputs the three-dimensional position of the noise source and the sound wave reflection path;

[0062] S4: Optimize and analyze the sound wave reflection path using the geometric structure data of the target building. The LSTM network learns historical acoustic data and dynamically simulates the three-dimensional position of the noise source and the sound wave reflection path to generate a three-dimensional reflection path map;

[0063] S5: Dynamically display the position of the noise source and the sound wave reflection path according to the three-dimensional reflection path map.

[0064] The method of the present invention adopts an acoustic positioning algorithm and multi-dimensional signal processing technology, which can achieve high-precision three-dimensional positioning of noise sources, especially for noise source positioning in complex building environments, such as buildings with large open spaces, complex geometric structures, and buildings with special functions. It accurately models the complex building environment and analyzes the propagation and reflection paths of sound waves in the building.

[0065] S1: According to the geometric structure of the target building, arrange a number of sound sensors to collect sound signals, sample the sound signals, and store them in the form of a time series.

[0066] S11: Arrange sound sensors: According to the geometric structure of the target building, arrange a number of sound sensors. The sound sensors can be high-sensitivity microphone sensors. Ensure time synchronization through PrecisionTimeProtocol (PTP) with an error less than 1 ms. The number of sound sensors can be adjusted by the user according to the building area and noise positioning accuracy requirements, and the preferred range is 30 - 50, with the spacing set between 0.5 meters and 5 meters, evenly distributed around the building and on the ceiling to ensure a wide sampling range without blind spots.

[0067] Among them, the arrangement method of the sound sensors is designed differently according to different building structures as follows:

[0068] Regular grid arrangement: Suitable for open and symmetrical buildings, specifically applicable to regular spaces such as office buildings, shopping malls, and halls. Uniformly arrange in a three-dimensional grid in the symmetrical open space. Capture the symmetry of the sound field reflection through multi-node synchronous sampling. The sound sensors are arranged on the ceiling or wall to form a uniform grid to ensure uniform sampling of sound waves. The preferred spacing is: 5 m - 10 m, and the specific spacing is adjusted according to the noise attenuation situation;

[0069] Stratified layout: Applicable to multi-storey buildings or vertical structures, deployed vertically along the multi-storey building, analyzing the vertical sound field superposition effect by associating the time delay between adjacent layer sensor groups. Multiple groups of sound sensors are arranged according to the storey height on each floor, with at least 3 - 4 sensors arranged on each floor for acoustic wave propagation analysis at different heights. The sensor spacing is adjusted according to the floor height, usually at least 1 group per floor;

[0070] Zonal layout: Applicable to complex spaces or local monitoring, specifically used for high-noise areas such as machine rooms, elevator shafts, ventilation ducts, etc. In complex spaces, sound field mutation areas are divided, and sensors are dynamically encrypted based on the reflection path density to form an acoustic energy attenuation gradient tracking network. Sound sensors are locally densely arranged in high-noise areas, and additional sensors are added near corners, wall reflection points, doors, and windows to capture reflected and diffracted sound waves;

[0071] Linear layout: Applicable to narrow and long spaces, specifically used for corridors, passageways, etc. Sound sensors are arranged at intervals along the main axis direction, and a sensor chain is laid along the axis of the narrow passage. The acoustic wave guiding mode is analyzed through the axial time-of-arrival time series. One is placed every about 5 m and installed on one or more of the walls, ceilings, or floors to obtain the maximum amount of information.

[0072] Based on the real-time acoustic wave reflection path reconstruction results, each topological structure correlates the probability distribution of the noise source location, adaptively adjusts the sensor space density and signal fusion weight. In actual use, it is necessary to select a suitable layout method according to different usage scenarios.

[0073] S12: Collect sound signals, sample and synchronize the sound signals: Set the sampling signal to 96 kHz, covering the acoustic wave frequency range (20 Hz to 20 kHz) and its high-frequency harmonic components. Use high-precision timestamps to synchronize the sampling signals of all sensors, with a calibration error less than 1 ms to ensure the accuracy of time difference calculation.

[0074] S13: Data formatting: The data collected and processed through the above steps is stored in the form of a time series, including the following features:

[0075] Time difference (TDOA): Used to calculate the distance between the noise source and the sensor;

[0076] Spectrum characteristics: Extract the frequency distribution of the signal through the fast Fourier transform (FFT);

[0077] Energy characteristics: Record the change of signal intensity over time;

[0078] Data is transmitted to the backend via the MQTT or WebSocket protocol. Among them, the data is stored in a time-series database, such as InfluDB or TimescaleDB. The time format is a time series, and the data is written into the database table SensorData at fixed time intervals. The table structure is designed as follows:

[0079] Table 1: Database table SensorData

[0080] Field Name Type Description SensorID INT Sensor number Timestamp DATETIME Data acquisition time TDOA FLOATARRAY Time difference matrix Spectrum FLOATARRAY Spectrum characteristics Energy FLOAT Energy distribution value

[0081] S2: Perform data preprocessing on the sound signal, extract the time difference matrix, spectral features, and energy features, merge them into three-dimensional time series features, and normalize them to form a three-dimensional time series feature matrix.

[0082] S21: Noise reduction processing: Use an adaptive filter to remove background noise and filter out low-frequency interference (<20 Hz) and high-frequency interference (>20 kHz).

[0083] S22: Feature extraction:

[0084] Extract the time difference (TDOA): Calculate the time delay matrix between sensors, ΔT = {Δt ij I | i,j = 1,...,32};

[0085] Extract spectral features: Obtain the frequency distribution through FFT;

[0086] Extract energy features: Calculate the sound wave intensity within each time window;

[0087] Merge into three-dimensional time series features.

[0088] S23: Normalization and serialization: Normalize all features, standardize the data range to [0,1], and improve the efficiency of model training. Generate a time series feature matrix with an input format of N×T×F:

[0089] N = X (N represents the number of sensors);

[0090] T = 1000 (T represents the time step);

[0091] F = 3 (F represents the number of features: time difference, spectral features, energy).

[0092] After data preprocessing, it is immediately written into the database, providing input for the LSTM network. Indexes are established according to timestamps and sensor numbers to support fast query and batch loading. The database table ProcessedData involved is as follows:

[0093] Table 2: Database table ProcessedData

[0094] Field Name Type Description SensorID INT Sensor number Timestamp DATETIME Data processing time NormalizedTDOA FLOAT ARRAY Normalized time difference matrix Spectrum FLOAT ARRAY Spectrum characteristics Energy FLOAT Energy distribution value

[0095] S3: Establish an LSTM network, input the three-dimensional time series feature matrix in the form of a time series window. The LSTM layer models the propagation pattern and reflection characteristics of the sound signal in the building space, conducts time series analysis, uses the attention mechanism to generate a context vector for noise source localization by weighting, and predicts the three-dimensional position of the noise source and the sound wave reflection path.

[0096] The LSTM network includes:

[0097] Input layer: Receive the sound signal collected by the sound sensor and input it in the form of a time series window. The format is as follows: Input dimension: (N, T, F).

[0098] Convolution layer: Extract time-frequency features to reduce noise interference and improve the adaptability of the network to different noise types. Use a 1D or 2D convolutional neural network (CNN) for feature extraction, where:

[0099] 1D convolution: Suitable for raw time-domain waveform data;

[0100] 2D convolution: Suitable for spectral data (such as Mel spectrum, STFT transform);

[0101] Convolution kernel size: 3×3 3\times3 3×3 or 5×5 5\times5 5×5;

[0102] Number of convolution channels: 64 - 128.

[0103] LSTM layer: Model the propagation pattern and reflection characteristics of the sound signal in the building space, conduct time series analysis. The LSTM layer includes a first LSTM, a second LSTM, and a Dropout mechanism. Adopt a bidirectional LSTM (Bi-LSTM) structure to capture the forward and backward dependencies of the time series, so as to more accurately model the complex building sound wave propagation path:

[0104] First LSTM: Capture local time dependencies and learn the sound wave change pattern within a short time range. Number of hidden neuron units: 128 - 256;

[0105] Second LSTM: The second LSTM is a bidirectional LSTM (Bi-LSTM) to enhance bidirectional dependency modeling and improve time series modeling ability. The optional number of neurons is 128. The second LSTM can optionally use a GRU (gated recurrent unit) as a lightweight alternative;

[0106] Dropout mechanism: The dropout probability is 0.2 - 0.5, used to prevent overfitting.

[0107] Attention mechanism: used to process a large amount of redundant information contained in noisy data, can automatically focus on key time steps, improve the model's sensitivity to important noise signals. The attention mechanism includes:

[0108] Weight calculation: for each time step and hidden state calculate the main force weights :

[0109] , ,

[0110] where , , are learnable parameter matrices;

[0111] Weighted summation: weight the time series features according to the weights to generate a context vector:

[0112] ,

[0113] The context vector contains the most crucial information for noise source localization and serves as the output of the attention mechanism.

[0114] Fully connected layer: convert the output of the LSTM layer into the final localization result, output three-dimensional coordinates (x, y, z), optimize the predicted noise source position, and use the mean squared error (MSE) loss function for optimization;

[0115] Output layer: output the three-dimensional position (x, y, z) of the predicted noise source generated by the combination of the LSTM layer and the attention mechanism and the reflection path probability distribution (including the positions and reflection coefficients of the main reflecting surfaces).

[0116] S4: Optimize the analysis of the acoustic wave reflection path using the geometric structure data of the target building. The LSTM network learns historical acoustic data and dynamically simulates the three-dimensional position of the noise source and the acoustic wave reflection path to generate a three-dimensional reflection path map.

[0117] S41: Optimize the calculation of the acoustic wave reflection path using the building geometric structure data (materials and positions of walls, ceilings, etc.), introduce BIM data or CAD floor plans to accurately obtain the positions and material information of the building's walls, ceilings, floors, and other structures. In the analysis of the acoustic wave reflection path, the building geometric structure plays a key role, and these structures have different effects on the propagation of acoustic waves. For example, concrete walls usually have a high reflectivity, while carpets have a high absorption capacity. Using this data, the reflection coefficients, sound absorption coefficients, and transmission coefficients of each reflecting surface can be calculated, thereby optimizing the modeling of the acoustic wave propagation path.

[0118] To simulate the reflection path of sound waves, the mirror source method or ray tracing can be used. The mirror source method creates virtual mirror sources behind each reflecting surface to calculate the first-order, second-order, and higher-order reflection paths; the ray tracing method emits multiple sound wave rays, traces their propagation trajectories in the building, and calculates the reflection, attenuation, and arrival time at the sensor. By selecting either of the above two methods and combining with the building geometry data, the sound wave propagation pattern in different areas can be accurately predicted.

[0119] S42: Dynamically simulate the sound wave reflection path: The LSTM network predicts the possible sound wave reflection path in the current environment by learning historical acoustic data (historical sensor data is stored in the SensorData and ProcessedData tables), and provides the ability for real-time adjustment to achieve dynamic simulation. The input of this prediction process includes the time series data received by the sensor and the building reflecting surface data (such as the material properties, positions, etc. of each reflecting surface), and the output includes the reflection path of the sound wave and the corresponding energy distribution.

[0120] Among them, if the network output result is stored in the AnalysisResults table, record the noise source location and the reflection path:

[0121] Table 3: AnalysisResults table

[0122] Field Name Type Description AnalysisID INT Analysis number SourceLocation JSON Three-dimensional position of the noise source (x, y, zx, y, zx, y, z) PathData JSON Acoustic wave reflection path data Timestamp DATETIME Analysis time

[0123] If the network model is stored in the LSTMModel table:

[0124] Table 4: LSTMModel table

[0125] Field Name Type Description ModelID INT Model number ModelWeights BLOB Model weight file TrainingDataID INT Training data number LastUpdated DATETIME Time of the last model update

[0126] S43: Generate a sound wave reflection path diagram: After obtaining the prediction results, match them with the reflecting surfaces in the building geometry model to generate a three-dimensional sound wave reflection path diagram, marking the reflection point coordinates, energy attenuation, and number of reflections of the path nodes. This path diagram can be used to analyze the main reflection paths inside the building, calculate and display the energy distribution of the main reflection paths, and evaluate the attenuation of different paths. For example, the total energy loss of each path can be calculated, and different reflection paths can be distinguished by color coding. High-energy reflection paths can be marked in red, low-energy paths in blue, and diffraction paths in green. This dynamic simulation process enables the system to continuously optimize the accuracy of noise source localization according to real-time data.

[0127] Perform 3D visualization on the finally calculated reflection paths, use a visualization engine (such as Three.js or Unity3D) to draw the path diagram, and perform dynamic rendering on the 3D building model. This process can be implemented using an OpenGL / Unity3D rendering engine for real-time simulation. In this way, users can clearly see the distribution of key reflection surfaces, the propagation paths of sound waves, and the acoustic characteristics of different regions.

[0128] The building geometric model data is stored in GeometryData, mainly including 3D coordinates and material properties. The path data is stored in the table PathSimulation in JSON format, including path node coordinates, reflection times, and energy distribution.

[0129] Table 5: GeometryData Table

[0130] Field Name Type Description ModelID INT Model number WallCoordinates JSON Wall coordinate data Material VARCHAR Material characteristics ReflectionCoeff FLOAT Reflection coefficient

[0131] Table 6: PathSimulation Table

[0132] Field Name Type Description SimulationID INT Path simulation number PathData JSON Simulated path data EnergyLoss FLOAT Energy attenuation value Timestamp DATETIME Simulation time

[0133] In the visualized path diagram, each reflection surface will be marked to show its impact on the propagation of sound waves. For example, a concrete wall may be the main high-energy reflection surface, while the sound-absorbing material on the ceiling may significantly reduce the reflection intensity of sound waves. In addition, the path diagram can also mark the number of reflections of sound waves, that is, distinguish single reflection, double reflection, and multiple reflections, to help optimize the acoustic environment of the building. If the reflected sound waves in a certain area are too strong (such as echoes in a meeting room or office), sound-absorbing materials can be added in this area, or the building layout and sensor positions can be adjusted to reduce noise interference.

[0134] S5: Dynamically display the noise source location and the sound wave reflection path according to the 3D reflection path diagram, and give an alarm when the noise intensity exceeds the threshold.

[0135] S51: Real-time visualization: Dynamically display the noise source location (red marked points) and the sound wave reflection path (blue and green curves), and provide model interaction functions, including zooming and rotating the view.

[0136] S52: Alarm mechanism: When the noise intensity exceeds the threshold (such as 80 dB), trigger visual alarm (red flashing) and sound warning, and record the exceeding time, area, and noise source location.

[0137] The visualization configuration is stored in the table VisualizationConfig, and the alarm records are stored in the table Alarms, supporting historical queries. The modeling files are stored in STL or OBJ format.

[0138] Table 7: VisualizationConfig Table

[0139] Field Name Type Description ConfigID INT Visualization configuration number PathData JSON Visualization path data GeometryModel JSON Building geometric model Timestamp DATETIME Configuration update time

[0140] Table 8: Alarms Table

[0141] Field Name Type Description AlarmID INT Alarm number SourceLocation JSON Three-dimensional position of the noise source NoiseLevel FLOAT Noise intensity Suggestions TEXT Optimization suggestions Timestamp DATETIME Alarm time

[0142] The present invention optimizes the LSTM network by introducing an attention mechanism, achieving precise modeling of noise source localization and sound wave reflection paths in complex building acoustic environments. Compared with traditional methods, the attention mechanism dynamically focuses on key path information, significantly improving the localization accuracy, enhancing the system's adaptability to multi-path reflections and dynamic noise scenarios, and controlling the real-time delay within 50 ms, providing an efficient and reliable solution for building acoustic design, noise monitoring, and environmental optimization.

[0143] Example 2

[0144] As Figure 2 shown, a building sound wave reflection path analysis and noise source localization system includes:

[0145] Multi-point sensor acquisition module: Highly sensitively acquire the time difference, spectrum, and energy characteristics of sound waves to ensure the coverage and synchronization of data in key areas;

[0146] Data preprocessing module: Denoise the acquired sound wave signals, remove background noise and non-target frequency band signals, and normalize the feature data;

[0147] LSTM network module: Construct time series features and a deep model to predict the position of the noise source and the sound wave reflection path, capture long-term dependencies in the time series through hierarchical design, and introduce an attention mechanism to dynamically focus on key time step features to filter out redundant information;

[0148] Sound wave reflection path analysis module: Combine the prediction results of the LSTM network and the building geometric model to simulate the propagation and reflection process of sound waves and generate an accurate path map;

[0149] Real-time visualization and alarm module: Provide a user-friendly 3D modeling interface to display the noise source and reflection path, and provide alarms and optimization suggestions when the noise exceeds the standard to quickly locate and solve noise problems.

[0150] The system of the present invention is used to implement the building sound wave reflection path analysis and noise source localization method. Based on the above functional modules, it can improve the system's adaptability in multi-reflection path and dynamic noise source scenarios, ensure the real-time performance of the system, and meet the requirements of building acoustic design, noise management, and public place environmental optimization.

[0151] The above-described embodiments merely represent several implementation manners of the present invention. The description thereof is relatively specific and detailed, but it should not be construed as a limitation to the scope of the patent for the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent for the present invention shall be subject to the appended claims.

Claims

1. A method for analyzing the acoustic wave reflection path of a building and locating the noise source, characterized in that the steps Including: S1: According to the geometric structure of the target building, arrange a number of sound sensors to collect sound signals, sample the sound signals and store them in the form of a time series; S2: Preprocess the sound signals, extract the time difference matrix, spectral features and energy features, merge them into three-dimensional time series features, and normalize them to form a three-dimensional time series feature matrix; S3: Establish an LSTM network, input the three-dimensional time series feature matrix in the form of a time series window, the LSTM layer models the propagation mode and reflection characteristics of sound signals in the building space, conducts time series analysis, uses the attention mechanism to weight and generate a context vector for noise source localization, and predicts the three-dimensional position of the noise source and the sound wave reflection path; S4: Optimize and analyze the sound wave reflection path using the geometric structure data of the target building, the LSTM network learns historical acoustic data and dynamically simulates the three-dimensional position of the noise source and the sound wave reflection path, and generates a three-dimensional reflection path map; S5: Dynamically display the noise source position and the sound wave reflection path according to the three-dimensional reflection path map.

2. The method for analyzing the acoustic wave reflection path and locating the noise source of a building according to claim 1, wherein The arrangement method of the sound sensors includes: based on the acoustic propagation characteristics of the building space, configuring the sound sensor array into at least one topological structure of a regular grid type, a hierarchical coupling type, a partition gradient type, and a linear waveguide type, where: Regular grid type: Uniformly arranged in a three-dimensional grid in a symmetric open space, and capture the symmetry of the sound field reflection through multi-node synchronous sampling; Hierarchical type: Deployed vertically along multiple floors of the building, and analyze the vertical sound field superposition effect by combining the time delay correlation between adjacent floor sensor groups; Partition type: Divide the sound field mutation area in a complex space, dynamically encrypt the sensors based on the reflection path density, and form an acoustic energy attenuation gradient tracking network; Linear type: Arrange a sensor chain along the axis of a long and narrow channel, and analyze the acoustic wave guiding mode through the axial time of arrival time series.

3. The method for analyzing the building sound wave reflection path and locating the noise source according to claim 1, wherein The characteristics of the sound signals stored in the form of a time series include: Time difference: Used to calculate the distance between the noise source and the sensor; Spectral features: Extract the frequency distribution of the signal through fast Fourier transform; Energy features: Record the change of the signal intensity over time.

4. A method for analyzing the acoustic wave reflection path of a building and locating the noise source according to claim 1, characterized in that, The LSTM network includes: Input layer: Receive the sound signals collected by the sound sensors and input them in the form of a time series window; Convolution layer: Extract time-frequency features to reduce noise interference and improve the adaptability of the network to different noise types, and use 1D or 2D convolutional neural networks for feature extraction; LSTM layer: Model the propagation mode and reflection characteristics of sound signals in the building space, and conduct time series analysis; Attention mechanism: Automatically focus on key time steps to improve the sensitivity of the model to important noise signals; Fully connected layer: Convert the output of the LSTM layer into the final positioning result, output three-dimensional coordinates, and optimize the predicted noise source position; Output layer: Output the three-dimensional position of the predicted noise source and the probability distribution of the reflection path generated by the combination of the LSTM layer and the attention mechanism.

5. A method for analyzing the acoustic wave reflection path and locating the noise source in a building according to claim 4, characterized in that The LSTM layer includes a first LSTM, a second LSTM and a Dropout mechanism: First LSTM: Capture local time dependencies and learn the acoustic wave change pattern within a short time range; Second LSTM: The second LSTM is a bidirectional LSTM, which enhances bidirectional dependency modeling and improves the time series modeling ability; Dropout mechanism: The dropout probability is 0.2 - 0.5, which is used to prevent overfitting.

6. A method for analyzing the acoustic wave reflection path and locating the noise source in a building according to claim 4, characterized in that, The attention mechanism includes: Weight calculation: For each time step and the hidden state calculate the main force weight : ， ， Among them, , , are learnable parameter matrices; Weighted summation: According to the weights weight the time series features to generate a context vector: ， Context vector Contains the most crucial information for noise source localization and serves as the output of the attention mechanism.

7. A method for analyzing the acoustic wave reflection path and locating the noise source in a building according to claim 1, characterized in that The acoustic wave reflection path analysis includes: Optimizing the calculation of the acoustic wave reflection path using building geometric structure data, and adopting the mirror sound source method or ray tracing to simulate the acoustic wave reflection path; Dynamically simulating the reflection path: The LSTM network learns historical acoustic data, inputs the time series data received by the sound sensors and the building reflection surface data, predicts the possible acoustic wave reflection paths and adjusts them in real time, and outputs the reflection paths of the acoustic waves and the corresponding energy distributions; Generating a reflection path map: Matching the prediction results with the building reflection surface to generate a three-dimensional acoustic wave reflection path map, calculating the energy distribution, and performing three-dimensional visualization on the finally calculated acoustic wave reflection paths and dynamically rendering them on the building three-dimensional model.

8. A method for analyzing the acoustic wave reflection path and locating the noise source in a building according to claim 7, characterized in that The building geometric structure data includes the material of the wall, the position of the wall, the reflection coefficient, sound absorption coefficient, and light transmission coefficient of the reflection surface.

9. The method for analyzing the building sound wave reflection path and locating the noise source according to claim 7, characterized in that, The three-dimensional visualization of the acoustic wave reflection path is performed in real time using OpenGL or Unity3D, which is used to display the distribution of key reflection surfaces, the acoustic wave propagation path, and the acoustic information of different areas of the building, and mark the acoustic wave reflection paths with different energy distributions respectively.

10. A building sound wave reflection path analysis and noise source localization system, applicable to the building sound wave reflection path analysis and noise source localization method according to any one of claims 1 to 9, characterized in that, Includes: Multi-point sensor acquisition module: Highly sensitively acquire the time difference, frequency spectrum, and energy characteristics of acoustic waves; Data preprocessing module: Denoise the acquired acoustic wave signals, remove background noise and non-target frequency band signals, and normalize the feature data; LSTM network module: Construct time series features and a deep model to realize the prediction of the noise source position and the acoustic wave reflection path, capture the long-term dependencies in the time series through hierarchical design, and introduce an attention mechanism to dynamically focus on the key time step features to filter out redundant information; Acoustic wave reflection path analysis module: Combine the LSTM network prediction results and the building geometric model to simulate the propagation and reflection process of acoustic waves and generate an accurate path map; Real-time visualization and alarm module: Provide a user-friendly 3D modeling interface, display the noise source and the reflection path, and provide alarms and optimization suggestions when the noise exceeds the standard, quickly locate and solve the noise problem.

Images