A noise map dynamic deduction method and system based on multi-source heterogeneous data fusion

Abstract

The present invention relates to the field of noise map technology, and is a method and system for dynamic deduction of noise maps based on multi-source heterogeneous data fusion. The method and system comprise: in an edge computing node, performing spatiotemporal alignment processing on multiple clusters of low-dimensional feature vectors to generate three-dimensional sound field reconstruction parameters of a first target data sequence and spectral energy distribution parameters of a second target data sequence; dynamically controlling the dormancy of a sensor cluster and generating spatiotemporal differential parameters based on the three-dimensional sound field reconstruction parameters and the spectral energy distribution parameters; performing layered encryption on the spatiotemporal differential parameters and two types of data sequences during the dormancy period to generate a basic ciphertext sequence and a feature ciphertext sequence for verification; the present invention generates a noise deduction map with spatiotemporal continuity, realizes high-precision dynamic deduction and energy efficiency optimization of the noise map, effectively reflects the evolution law of noise over time and space, improves the real-time and accuracy of noise data, and can reflect the actual situation of noise in a more realistic, efficient and real-time manner.

Description

Technical Field

[0001] The present invention relates to the technical field of noise maps, and in particular to a noise map dynamic deduction method and system based on multi-source heterogeneous data fusion. Background Art

[0002] Traditional noise mapping methods rely heavily on single-type sensor data (such as sound pressure level data) and lack the ability to efficiently integrate heterogeneous data from multiple sources, such as the spatial distribution of the sound field and spectral characteristics. This is particularly true in complex scenarios like industrial parks, where the spatiotemporally heterogeneous data streams generated by numerous noise sensor nodes are difficult to precisely align, resulting in low sound field reconstruction accuracy and poor real-time performance.

[0003] For example, in an industrial park, noise sources may include both mechanical vibrations (dominated by low frequencies) and gas emissions (bursts of high frequencies). Traditional methods rely solely on the calculation of the average sound pressure level and are unable to distinguish differences in spectral characteristics, leading to misjudgment of noise sources. When data delays occur at nodes within a sensor cluster due to network congestion, linear interpolation alignment is directly used when fusing data with adjacent clusters, resulting in phase errors in the sound wave propagation path during sound field reconstruction (typical errors >10%).

[0004] For example, the cluster head node of a high-density sensor cluster (such as 500 nodes / cluster) needs to undertake the task of pre-fusion of raw data, but its computing resources are limited (such as a single-core ARM processor) and cannot complete multi-channel data compression within a 10ms time window, resulting in edge node queue overflow and data loss. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to overcome the shortcomings of the existing technology and provide a noise map dynamic deduction method and system based on multi-source heterogeneous data fusion, which realizes high-precision dynamic deduction and energy efficiency optimization of noise maps.

[0006] In order to solve the above technical problems, the basic concept of the technical solution adopted by the present invention is:

[0007] In the first aspect, a noise map dynamic deduction method based on multi-source heterogeneous data fusion includes:

[0008] In the edge computing node, multiple clusters of low-dimensional feature vectors are subjected to spatiotemporal alignment to generate three-dimensional sound field reconstruction parameters of the first target data sequence and spectrum energy distribution parameters of the second target data sequence;

[0009] Based on the three-dimensional sound field reconstruction parameters and spectrum energy distribution parameters, the sensor cluster is dynamically controlled to sleep and the spatiotemporal differential parameters are generated.

[0010] Layered encryption is performed on the two types of data sequences, namely, the time-space differential parameters and the dormant period, to generate a basic ciphertext sequence and a characteristic ciphertext sequence for verification;

[0011] Data correction and rendering control are performed based on basic ciphertext sequences and characteristic ciphertext sequences to generate a noise deduction map with spatiotemporal continuity.

[0012] The second aspect is a noise map dynamic deduction system based on multi-source heterogeneous data fusion, including:

[0013] A generation module is used to perform spatiotemporal alignment processing on multiple clusters of low-dimensional feature vectors in an edge computing node to generate three-dimensional sound field reconstruction parameters of a first target data sequence and spectral energy distribution parameters of a second target data sequence;

[0014] A fusion module is used to dynamically control the dormancy of the sensor cluster and generate spatiotemporal differential parameters based on the three-dimensional sound field reconstruction parameters and spectrum energy distribution parameters;

[0015] An encryption module is used to implement layered encryption on the spatiotemporal differential parameters and two types of data sequences during the dormant period to generate a basic ciphertext sequence and a characteristic ciphertext sequence for verification;

[0016] The correction module is used to perform data correction and rendering control based on the basic ciphertext sequence and the characteristic ciphertext sequence to generate a noise deduction map with spatiotemporal continuity.

[0017] After adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art:

[0018] By aligning time and space to process multi-source heterogeneous data, cross-domain fusion of low-dimensional features is achieved, the accuracy of three-dimensional sound field reconstruction and spectral energy analysis is improved, more reliable basic parameters are provided for noise modeling, and sensor cluster dormancy is dynamically controlled. While ensuring the integrity of data's time and space characteristics, equipment energy consumption and data redundancy are reduced, and the resource utilization efficiency of edge computing nodes is optimized. The layered encryption mechanism differentiates between basic data and feature data, implements lightweight encryption at the edge, generates a spatiotemporal continuous noise map, and supports real-time / near real-time dynamic rendering of noise distribution, effectively reflecting the evolution of noise over time and space, improving the real-time and accuracy of noise data, and being able to reflect the actual situation of noise in a more realistic, efficient, and real-time manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of the noise map dynamic deduction system based on multi-source heterogeneous data fusion of the present invention.

[0020] Figure 2 It is a flow chart of the noise map dynamic deduction method based on multi-source heterogeneous data fusion of the present invention. DETAILED DESCRIPTION

[0021] The following embodiment of the present application takes a noise map dynamic deduction system based on multi-source heterogeneous data fusion as an example to explain the solution of the present application in detail, but this embodiment does not limit the scope of protection of the present application.

[0022] like Figure 1 As shown, the present invention provides a noise map dynamic deduction method based on multi-source heterogeneous data fusion, including:

[0023] Step S1: In the edge computing node, a spatiotemporal alignment process is performed on multiple clusters of low-dimensional feature vectors to generate three-dimensional sound field reconstruction parameters of a first target data sequence and spectrum energy distribution parameters of a second target data sequence;

[0024] Step S2: Based on the three-dimensional sound field reconstruction parameters and spectrum energy distribution parameters generated in step S1, dynamically control the dormancy of the sensor cluster and generate spatiotemporal differential parameters;

[0025] Step S3: performing layered encryption on the spatiotemporal differential parameters generated in step S2 and the two types of data sequences during the dormant period to generate a basic ciphertext sequence and a characteristic ciphertext sequence for verification;

[0026] Step S4: Perform data correction and rendering control based on the encrypted ciphertext sequence in step S3 to generate a noise deduction map with spatiotemporal continuity.

[0027] In an embodiment of the present invention, multi-source heterogeneous data is processed through spatiotemporal alignment (step S1) to achieve cross-domain fusion of low-dimensional features, improve the accuracy of three-dimensional sound field reconstruction and spectral energy analysis, provide more reliable basic parameters for noise modeling, and dynamically control the dormancy of sensor clusters (step S2). While ensuring the integrity of the spatiotemporal characteristics of the data, the energy consumption of the equipment and data redundancy are reduced, and the resource utilization efficiency of the edge computing nodes is optimized. The layered encryption mechanism (step S3) performs differentiated processing on basic data and feature data, and realizes lightweight encryption at the edge, which not only meets the secure transmission and storage requirements of noise data, but also avoids the high computing overhead of traditional full encryption, and generates a spatiotemporal continuous noise deduction map (step S4). It supports real-time / near real-time dynamic rendering of noise distribution, and effectively reflects the evolution of noise over time and space.

[0028] In a preferred embodiment of the present invention, before step S1, the method further includes:

[0029] Implement cluster topology management for no less than 500 noise sensor nodes in the industrial park, dynamically divide sensor clusters based on node signal strength, and generate intra-cluster communication link topology;

[0030] A cluster head node is set based on the intra-cluster communication link topology, and the cluster head node performs pre-fusion calculations on the original noise data stream to generate de-redundant compressed data; the compressed data is converted into a low-dimensional feature vector by the cluster head node, and the low-dimensional feature vector is transmitted to the edge computing node through the edge gateway as input of the multi-cluster low-dimensional feature vector.

[0031] In an embodiment of the present invention, each noise sensor node periodically broadcasts signal strength detection packets (e.g., RSSI signals) to collect real-time signal interaction data from at least 50 nodes within an industrial park. A density-based clustering algorithm (e.g., DBSCAN) or an adaptive signal strength threshold method is then used to dynamically divide sensor clusters, using the node's physical location (latitude and longitude coordinates) as the spatial dimension and signal strength as the connection weight. The division principle is: the mean signal strength between nodes within the same cluster must be ≥ a preset threshold (e.g., -70dBm), and the number of nodes within a cluster must be limited to 50-100 to balance the computational load. A star or mesh communication link topology is generated for each cluster, and the communication paths between the cluster nodes and the cluster head node are clearly defined (e.g., establishing point-to-point connections via the ZigBee or LoRa protocols).

[0032] A distributed election algorithm (such as an improved version of the LEACH protocol) dynamically elects a cluster head node based on node residual energy, computing power (CPU frequency, memory capacity), and location centrality (near the cluster's geometric center). Reelection occurs every five minutes to prevent rapid energy depletion at a single node. After initialization, the cluster head establishes a two-way authentication mechanism (such as symmetric key encryption) with cluster member nodes to ensure data security. Cluster member nodes collect noise time-domain signals (voltage amplitude sequences) at a fixed frequency (such as 10 Hz) and transmit them to the cluster head via a communication link. The cluster head uses a sliding window de-redundancy algorithm (e.g., removing duplicate data with a fluctuation of less than 5% over three consecutive cycles) combined with a discrete Fourier transform (DFT) to extract frequency-domain features (such as octave-band sound pressure level). Principal component analysis (PCA) then compresses these multi-dimensional features to three to five dimensions, generating de-redundant compressed data (data compression ratios of 80%-90%).

[0033] The cluster head node maps the compressed data into a low-dimensional feature vector (such as a five-dimensional vector containing the time domain mean, frequency domain peak, and spatial coordinates). The vector elements use floating-point data format (single-precision floating point, each vector is approximately 20 bytes). The feature vector is normalized (such as standardized to the [-1,1] interval) to avoid dimensionality effects during subsequent processing by the edge computing node. Each cluster head node is connected to the edge gateway via wired (RJ45) or wireless (Wi-Fi) means. The edge gateway uses a publish-subscribe model (such as the MQTT protocol) to collect the low-dimensional feature vectors of each cluster.

[0034] The edge gateway synchronizes the timestamps of the multi-cluster feature vectors (accurate to milliseconds), encapsulates them into a unified data format (such as Protocol Buffers), and transmits them to the edge computing node via Ethernet or 5G network to form the "multi-cluster low-dimensional feature vector" input required for step S1.

[0035] In an embodiment of the present invention, dynamic clustering is used to reduce ineffective communication between nodes. The star topology within the cluster shortens the communication distance of a single node by 30%-50%, reduces the energy consumption of sensor nodes by more than 40%, and extends the life of the equipment. Nodes within the same cluster are spatially distributed closely, and signal transmission is highly consistent despite the influence of industrial electromagnetic interference, reducing the interference of heterogeneous noise on data synchronization. Redundancy compression reduces the amount of original data by more than 85%, significantly reducing the input data scale of edge computing nodes and improving the subsequent spatiotemporal alignment processing speed (processing time is reduced by 60%). Through frequency domain conversion and PCA dimensionality reduction, the key features of the noise signal (such as sound pressure level and main frequency distribution) are retained, and high-frequency noise interference is filtered out, providing a purer feature input for the three-dimensional sound field reconstruction in step S1. The data volume transmitted by a single cluster feature vector is reduced from the KB level of the original time domain signal to the byte level, the edge gateway throughput is increased by more than 5 times, and the network delay is reduced to less than 50ms. The timestamp synchronization mechanism of the edge gateway ensures that the time consistency error of multi-cluster data is less than 1ms, laying a high-precision time benchmark for the spatiotemporal alignment processing in step S1 and avoiding reconstruction parameter errors caused by clock deviation.

[0036] In a preferred embodiment of the present invention, step S1: in an edge computing node, performing spatiotemporal alignment processing on multiple clusters of low-dimensional feature vectors to generate three-dimensional sound field reconstruction parameters of a first target data sequence and spectral energy distribution parameters of a second target data sequence, including:

[0037] Based on the multiple clusters of low-dimensional feature vectors, the multiple clusters of feature vectors are spatially aligned through the convolution layer of the lightweight hybrid neural network to generate spatially aligned feature vectors; and the spatially aligned feature vectors are temporally aligned using a long short-term memory network layer to generate a continuous data stream;

[0038] The continuous data stream is divided into data blocks with a time window length of 10 milliseconds, and a three-dimensional sound field grid model is constructed based on the spatial coordinates of each node in the sensor cluster; the sound field intensity values ​​are filled in for discrete data points in the three-dimensional sound field grid model using an inverse distance weighted interpolation algorithm to generate three-dimensional sound field reconstruction parameters for the first target data sequence;

[0039] The 10 millisecond data blocks are subjected to fast Fourier transform to extract 1 / 3 octave spectrum energy, and multi-sensor data are fused through an energy superposition algorithm to generate spectrum energy distribution parameters of the second target data sequence.

[0040] In this embodiment of the present invention, spatial alignment and feature extraction first involves inputting multiple clusters of low-dimensional feature vectors (including information such as the spatial coordinates, time-domain mean, and frequency-domain peak values ​​of each sensor node) into the convolutional layer of a lightweight hybrid neural network. This network uses a simplified two-dimensional convolution kernel (e.g., 3×3) and the plane coordinates (X and Y axes) of the sensor nodes as input dimensions. This network scans the spatial distribution characteristics of different clusters using a sliding window, automatically identifying spatial relationships between adjacent sensor clusters (e.g., mapping the locations of noise sources and monitoring nodes within a factory floor). The spatially aligned feature vectors output by the convolutional layer contain the relative position weights of each cluster node in three-dimensional space and signal strength correlation features.

[0041] Time alignment and sequence generation: The spatially aligned feature vectors are input into the long short-term memory (LSTM) layer. The LSTM's gating mechanism is used to capture long-term dependencies in the time series (such as the periodic pattern of noise changes with production shifts). The network input is a sequence of feature vectors sorted by timestamps (with a time resolution accurate to the millisecond level). Through the hidden layer state transfer, the output is a continuous data stream with timestamp bias removed, so that data at different times and in different clusters form a seamless sequence in the time dimension.

[0042] Data block segmentation and grid modeling divide the continuous data stream into independent data blocks according to 10-millisecond time windows (for example, each data block contains multi-cluster features of 100 time points). At the same time, based on the three-dimensional spatial coordinates of the nodes in each sensor cluster (X, Y, Z axes, where the Z axis represents the height, such as different floors of a factory building), the target area is divided into a regular three-dimensional grid (such as a 0.5-meter × 0.5-meter × 0.5-meter cube unit). Each grid point corresponds to a location to be filled in for sound field intensity calculation.

[0043] The sound field reconstruction parameters are generated by collecting data from several sensor nodes closest to each discrete data point in the three-dimensional grid (such as the eight nodes with the closest Euclidean distance). The inverse distance weighting method is used: the closer the node is, the higher the contribution weight to the sound field intensity of the grid point is. The sound field intensity value of each grid point is filled by weighted average calculation, and finally the three-dimensional sound field grid data covering the target area is generated, that is, the three-dimensional sound field reconstruction parameters of the first target data sequence.

[0044] To generate spectral energy parameters, a Fast Fourier Transform (FFT) is performed on each 10-millisecond data block, converting the time-domain noise signal into a frequency-domain energy distribution. Energy values ​​in the 1 / 3 octave band (such as 63 Hz, 80 Hz, and 100 Hz, commonly used noise analysis bands) are specifically extracted. The energy values ​​in the same frequency band across multiple sensor clusters within the same time window are then averaged (e.g., using the arithmetic mean or weighted average) to eliminate random errors from individual sensors. This generates spectral energy distribution parameters for the second target data sequence, reflecting the spatial distribution characteristics of noise energy in different frequency bands.

[0045] In an embodiment of the present invention, a lightweight hybrid neural network combines the spatial feature extraction of the convolutional layer with the time series modeling of the LSTM to achieve low-computational-complexity spatiotemporal alignment at the edge computing node. Compared with the traditional interpolation method, the accuracy of spatial position association recognition is improved by 30%, and the time series continuity error is reduced to less than 5%. A three-dimensional grid model is constructed based on the actual spatial coordinates of the sensor, and the sound field intensity is filled by inverse distance weighted interpolation, effectively solving the problem of sound field gaps between discrete sensor nodes. The reconstructed sound field parameters can accurately reflect the noise gradient changes on different floors and near equipment in industrial plants (such as the sound field intensity error within 5 meters from the noise source is less than 3%). The 1 / 3 octave energy extraction and multi-sensor fusion algorithm can not only retain the key frequency domain characteristics of the noise signal (such as the energy anomaly in a specific frequency band of mechanical vibration), but also reduce the single-point measurement noise through data superposition, thereby improving the signal-to-noise ratio of the spectral energy distribution parameters by 40%.

[0046] In a preferred embodiment of the present invention, based on the multiple clusters of low-dimensional feature vectors, the multiple clusters of feature vectors are spatially aligned by a convolutional layer of a lightweight hybrid neural network to generate a spatially aligned feature vector; and the spatially aligned feature vectors are temporally aligned by a long short-term memory network layer to generate a continuous data stream, including:

[0047] The spatial features of multiple clusters of low-dimensional feature vectors are extracted using the multi-layer convolution kernel of the convolution layer, and the coordinate data of different sensor clusters are normalized to the same spatial reference system through the feature mapping matrix;

[0048] A long short-term memory network layer is used to perform sliding window analysis on the timing characteristics of continuous data streams, and a gating mechanism is used to dynamically compensate for the time offset of adjacent data blocks to achieve millisecond-level time synchronization.

[0049] In this embodiment of the present invention, multi-layer convolutional kernel spatial feature extraction first inputs multiple clusters of low-dimensional feature vectors (containing information such as the three-dimensional X / Y / Z coordinates of each sensor node, signal strength, and device type) into the convolutional layer of a lightweight neural network. The convolutional layer uses two or three layers of two-dimensional convolutional kernels of varying sizes (e.g., a first layer of 5×5 kernels captures large-scale spatial correlations, while a second layer of 3×3 kernels focuses on local node relationships). Using the plane coordinates (X, Y) of the sensor nodes as input, the convolutional layer scans the spatial distribution of different sensor clusters layer by layer.

[0050] The first-layer convolution kernel calculates the spatial distance weights of adjacent nodes through a sliding window to identify the spatial dependence of signal propagation between clusters (for example, nodes within 10 meters are considered strongly correlated). The second-layer convolution kernel extracts the spatial density characteristics of nodes within the cluster (such as the number of nodes per 20 square meters), generating a preliminary feature map that includes spatial position correlation and signal attenuation patterns. Coordinate normalization and spatial reference system - to address possible local coordinate system differences between different sensor clusters (for example, different factories use independent coordinate origins), coordinate transformation is performed using a preset feature mapping matrix:

[0051] First, the physical coordinates of all sensor nodes are collected (absolute coordinates are obtained through GPS or UWB positioning systems) to determine the global coordinate system of the industrial park (with the center of the park as the origin, the X axis is due east, the Y axis is due north, and the Z axis is vertical height);

[0052] The coordinate data of each sensor cluster (such as the local coordinates of a workshop with the corner as the origin) is normalized to the global coordinate system through translation (X'=X+offset), rotation (correction of coordinate system angle deviation), and scaling (unified to meters) to ensure that the spatial data of different clusters are comparable at the same scale.

[0053] Sliding window time series feature analysis sorts the spatially aligned feature vectors by timestamp and feeds them into a long short-term memory (LSTM) layer. A fixed-length sliding window (e.g., a 500ms window containing 50 10ms data points) is used to process the data stream segment by segment. The window slides 10ms at a time, ensuring a 490ms overlap between adjacent data blocks, capturing the continuity of the time series.

[0054] The data within the window contains the changes in signal strength of each sensor cluster at consecutive time points (such as the sudden noise changes caused by the start and stop of night shift machines). LSTM transfers the hidden layer state and memorizes the feature dependencies at different time points (such as the impact of the noise peak in the previous 100ms on the current moment).

[0055] The gating mechanism dynamically compensates for time offsets. To address potential clock synchronization errors among different sensor clusters (e.g., timestamp deviations of ±5ms due to communication delays on some nodes), the LSTM gating mechanism (input gate, forget gate, and output gate) is used for time calibration.

[0056] Input gate: Calculates the offset between the current time point data and the theoretical timestamp (such as the difference between the actual reception time and the sensor reporting time) and generates an offset weight factor;

[0057] Forget Gate: Reduces the weight of data points with a time offset of more than 3ms to suppress the influence of outdated information (for example, ignores abnormal data with a delay of more than 5ms);

[0058] Output gate: Dynamically adjusts the output value at the current moment based on the temporal continuity of adjacent data blocks (for example, compensating for missing intermediate values ​​by interpolating data from previous and subsequent moments), ultimately generating a continuous data stream with a timestamp error of less than 1ms.

[0059] In an embodiment of the present invention, cross-cluster coordinate unification converts the scattered local coordinate system into the global coordinate system through the feature mapping matrix, completely solving the problem of inconsistent spatial references of multi-source sensors (such as the conversion of monitoring data from different factory buildings from "local coordinates of the workshop" to "global coordinates of the park"), so that the spatial positioning accuracy of subsequent three-dimensional sound field modeling is improved to within 0.5 meters, avoiding sound field reconstruction distortion caused by coordinate deviation (such as traditional unnormalized methods may cause spatial positioning errors of more than 10 meters). Spatial correlation intelligent recognition multi-layer convolution kernels automatically capture the spatial distribution patterns of sensor nodes (such as densely deployed node clusters along the production line and sparsely distributed nodes at the edge of the factory area), generate spatial alignment features including distance weights and signal attenuation characteristics, and provide more accurate spatial correlation basis for subsequent inverse distance weighted interpolation (for example, correctly identifying the high-weight contribution of nodes near the noise source to the grid points). Compared with the method of manually setting the distance threshold, the efficiency of spatial feature extraction is improved by 40%. The millisecond-level time synchronization capability uses a sliding window combined with a gating mechanism to effectively address sensor clock asynchrony, compressing the timestamp error of multi-cluster data from the ±10ms of traditional NTP synchronization to within ±1ms. This ensures that the same noise event (such as the moment a machine starts or stops) is displayed synchronously in data from different clusters, avoiding misjudgment of noise evolution patterns due to time misalignment (for example, accurately capturing the millisecond-level time correspondence between noise peaks and device actions). The continuous enhancement of time series features enhances the memory characteristics of LSTM, which can retain long-term noise change trends (such as periodic fluctuations in noise within an 8-hour shift). At the same time, it maintains feature continuity between data blocks through overlapping sliding windows (avoiding signal breaks at window boundaries), so that the continuous data stream can fully reflect the gradual change of noise over time (such as the smooth decline curve of noise during equipment maintenance at noon). The lightweight network design (such as limiting the depth of the convolution layer to 3 layers and the dimension of the LSTM hidden layer to 64 dimensions) reduces the power consumption of single-node processing by 60%, and controls the inference delay on edge computing nodes (such as NVIDIA Jetson AGX Orin) to within 15ms, meeting the stringent requirements of real-time industrial noise monitoring for low latency and low energy consumption. Compared with traditional cloud-based full-data processing solutions, the response speed is more than 3 times faster and the network transmission load is reduced by 70%.

[0060] In a preferred embodiment of the present invention, a fast Fourier transform is performed on the 10 millisecond data block to extract the 1 / 3 octave spectrum energy, and multi-sensor data is fused through an energy superposition algorithm to generate spectrum energy distribution parameters of the second target data sequence, including:

[0061] Performing Hanning windowing processing on the time domain signal of each of the 10 millisecond data blocks to eliminate spectrum leakage, and then obtaining frequency domain energy distribution through fast Fourier transform;

[0062] Based on the standard 1 / 3 octave frequency band division rule, the frequency domain energy distribution is segmented according to the center frequency, the energy value in each frequency band is integrated and calculated, and the initial energy distribution parameters are generated with the frequency band as the index;

[0063] Based on the spatial distribution density of each node in the sensor cluster, spatial weighted fusion is performed on the initial energy distribution parameters to generate spectrum energy distribution parameters of the second target data sequence.

[0064] In this embodiment of the present invention, a Hanning window function is applied to the time-domain noise signal (voltage amplitude sequence) of each 10-millisecond data block. The window width matches the data block length, smoothing the signal edges to reduce spectral leakage (preventing spectral energy from spreading to adjacent frequencies). A fast Fourier transform (FFT) is performed on the windowed signal to convert the time-domain signal into a frequency-domain energy distribution. The output frequency range covers 20 Hz to 20 kHz (the audible range of the human ear), with a frequency resolution of approximately 100 Hz (determined by the number of FFT points).

[0065] The continuous spectrum output by the FFT is divided into 30 frequency bands (each with a bandwidth of approximately 23% of the current center frequency) based on the 1 / 3 octave center frequencies defined in international standards such as ISO 266 (e.g., 31.5 Hz, 40 Hz, 50 Hz, ..., 16 kHz). The discrete spectrum energy values ​​within each band are integrated (that is, the energy of all frequency points within the band is accumulated) to generate initial energy distribution parameters indexed by the 1 / 3 octave center frequency. Each parameter represents the noise energy intensity in a specific frequency band.

[0066] Calculate the spatial distribution density of each node in the sensor cluster: with each node as the center, delineate a circular area with a radius of 2 meters, and count the number of other nodes in the area as a density indicator; assign weights to each node based on the density indicator: nodes in high-density areas (such as those near equipment in the center of the workshop) have lower weights (such as 0.8), and nodes in low-density areas (such as the edge of the factory) have higher weights (such as 1.2). Balance the monitoring contributions of different areas, perform weighted averaging fusion on the initial energy parameters of multiple nodes in the same frequency band, and generate the final spectrum energy distribution parameters, which reflect the spatial distribution of the noise frequency characteristics of the entire monitoring area.

[0067] In the embodiment of the present invention, the Hanning windowing process effectively suppresses spectrum leakage, making the energy boundaries of each frequency band clearer (for example, the error of energy leakage from the 500Hz band to the 400Hz band is reduced from 15% of the traditional rectangular window to 3%), improving the frequency resolution, and the 1 / 3 octave analysis is more refined than the traditional 1 octave (for example, the three frequency bands of 800Hz, 1kHz, and 1.25kHz can be distinguished near 1kHz), which can more accurately identify the characteristic frequencies of the equipment (such as the 1000Hz electromagnetic noise of the motor and the 1250Hz mechanical noise of the gearbox). The spatial weighted fusion algorithm avoids the problem of overfitting of data in high-density areas, so that the noise characteristics of the edge areas (such as factory boundary noise) account for a reasonable proportion in the final parameters, and fully presents the spatial distribution of the noise source. The energy superposition processing of multi-sensor data eliminates the accidental errors of single-point measurement (such as random environmental noise interference), making the spectral energy distribution parameters closer to the actual sound field characteristics (such as the energy fluctuation range in a certain frequency band is reduced from ±5dB to ±1.5dB). The 10-millisecond processing window balances the time resolution (capturing sudden noise events) and frequency resolution (meeting the 1 / 3 octave analysis requirements), which is suitable for the dual scenarios of industrial equipment start-up and shutdown (time scale at the 100ms level) and steady-state noise monitoring. FFT and energy calculation are performed on the edge, and the data processing delay is less than 5ms, meeting the real-time noise warning requirements (such as abnormal frequency energy mutation alarm). Compared with cloud-based processing solutions, the response speed is increased by 90%.

[0068] In a preferred embodiment of the present invention, step S2: dynamically controlling the dormancy of the sensor cluster and generating spatiotemporal differential parameters based on the three-dimensional sound field reconstruction parameters and spectral energy distribution parameters generated in step S1, includes:

[0069] Calculate the activity score of each sensor cluster based on the sound field intensity value in the three-dimensional sound field reconstruction parameter and the frequency band energy ratio in the spectrum energy distribution parameter; if the activity score is lower than a preset threshold, send a sleep instruction to the corresponding sensor cluster and record the sleep start timestamp and the spatiotemporal coordinates of the missing data during the sleep period;

[0070] Based on the recorded sleep start timestamp and spatiotemporal coordinates, obtaining three-dimensional sound field reconstruction parameters before sleep, and obtaining three-dimensional sound field reconstruction parameters after sleep after sleep ends, and calculating the sound field intensity difference matrix between the two;

[0071] The sound field intensity difference matrix is ​​input into the spatiotemporal Kalman filter algorithm. By fusing the historical sound field intensity change rate and the spatiotemporal distribution characteristics of the current sensor cluster, the dynamic sound field intensity change in the dormant area is predicted, and the spatiotemporal difference parameters including timestamp compensation and spatial interpolation weights are generated.

[0072] In an embodiment of the present invention, based on the sound field intensity value in the three-dimensional sound field reconstruction parameter, the average sound pressure level (such as the A sound level) of the monitoring area of ​​each sensor cluster is calculated. Combined with the frequency band energy ratio in the spectral energy distribution parameter, the characteristic frequency band that contributes most to the total sound pressure is identified (for example, industrial equipment noise is usually concentrated in the range of 500Hz-4kHz). The sound field intensity and the stability of the characteristic frequency band energy (for example, a fluctuation coefficient of <5% is considered stable) are comprehensively considered to generate an activity score (0-100 points) for each sensor cluster.

[0073] Hibernation Decision:

[0074] If the score of a sensor cluster is lower than the preset threshold (such as 40 points) for three consecutive periods (such as 30 seconds), the sound field in the area is determined to be stable, and a sleep command is sent to it, recording the sleep start timestamp (accurate to milliseconds) and the spatiotemporal coordinates (three-dimensional grid position) of the missing data during the sleep period.

[0075] Sound field intensity difference matrix calculation, data collection:

[0076] Before the sensor cluster goes to sleep (e.g., within 100ms before sleep), the sound field intensity value in the three-dimensional sound field reconstruction parameters of the area is extracted to form a pre-sleep intensity matrix. After the sensor cluster wakes up (e.g., within 50ms after wakeup), the sound field intensity value of the same area is immediately collected to form a post-sleep intensity matrix.

[0077] Difference calculation:

[0078] The difference between the intensity matrices before and after sleep is calculated point by point to generate a sound field intensity difference matrix. The matrix elements represent the change in sound pressure level at each grid point during the sleep period (e.g., ΔdB = dB after sleep - dB before sleep).

[0079] Spatiotemporal Kalman filter prediction and differential parameter generation, historical feature extraction:

[0080] Analyze the rate of change of sound field intensity in the area from historical data (such as the hourly sound pressure level change trend), identify diurnal periodic patterns (such as noise reduction caused by equipment shutdown at night), and combine the spatial distribution density of sensor clusters to determine the sensitivity coefficient of sound field changes at different locations (such as the change rate is higher in areas close to the noise source).

[0081] Spatiotemporal fusion prediction:

[0082] The sound field intensity difference matrix is ​​input into the spatiotemporal Kalman filter algorithm, and the historical change rate and spatial sensitivity coefficient are integrated. The algorithm dynamically adjusts the prediction weight according to the sleep time (for example, the longer the sleep time, the higher the historical trend weight), and outputs the predicted sound field intensity change.

[0083] Differential parameter generation:

[0084] Generate spatiotemporal difference parameters including timestamp compensation values ​​(such as aligning the prediction results to the current time) and spatial interpolation weights (such as the grid points closer to the dormant area have higher weights) for subsequent data correction.

[0085] In an embodiment of the present invention, through sleep control, the power consumption of sensors in non-critical areas is reduced by more than 60% (for example, sensors that originally worked 24 hours a day are now intermittently operated), frequent data transmission is reduced, network bandwidth usage is reduced by 50%, and the service life of edge gateways and communication equipment is extended. The spatiotemporal Kalman filter algorithm effectively fills the data gaps during sleep, and the sound field intensity prediction error is controlled within ±2dB (the industrial noise standard allows an error of ±3dB). The timestamp compensation mechanism ensures the temporal continuity of dynamic deduction, avoids noise map jumps caused by data interruptions, and reduces the real-time processing load of edge computing nodes. The system reduces load (for example, the number of active sensor clusters processed simultaneously is reduced by 30%), freeing up computing resources for more complex noise propagation modeling. The sleep strategy enables the system to concentrate computing power on high-activity areas (such as near production lines where equipment starts and stops frequently), improving the monitoring accuracy of key areas. The spatiotemporal differential parameters reflect the dynamic change trend of the noise field (such as the rate of increase of sound pressure level when equipment starts), increasing the noise map update frequency to 10Hz (traditional methods are usually 1Hz), and shortening the response speed to sudden noise events (such as high-frequency noise caused by pipeline leakage) from seconds to milliseconds, meeting the timeliness requirements of industrial safety warnings.

[0086] In a preferred embodiment of the present invention, step S3: performing layered encryption on the spatiotemporal differential parameters generated in step S2 and the two types of data sequences during the dormant period to generate a basic ciphertext sequence and a characteristic ciphertext sequence for verification, including:

[0087] Based on the timestamp compensation and spatial interpolation weight in the spatiotemporal differential parameters, the sound pressure value and three-dimensional coordinates of the first target data sequence during the dormant period are checked for data integrity. After the check passes, the data is encrypted using the national secret SM4 algorithm to generate a basic ciphertext sequence;

[0088] For the spectrum energy values ​​and frequency band labels of the second target data sequence during the dormant period, adjusting the energy distribution ratio according to the spatial interpolation weight in the spatiotemporal difference parameter, and encrypting the adjusted energy values ​​and frequency band labels using a homomorphic encryption algorithm to generate a characteristic ciphertext sequence;

[0089] The basic ciphertext sequence and the characteristic ciphertext sequence are input into the watermark embedding module, a digital watermark identifier is generated based on the identifier of the sensor cluster and the edge node number, and the digital watermark identifier is hashed and bound with the synchronization timestamp in the spatiotemporal differential parameter to generate an encrypted data block for verification.

[0090] In an embodiment of the present invention, the timestamp compensation value in the spatiotemporal differential parameter is extracted and compared with the timestamp of the first target data sequence (sound pressure value and three-dimensional coordinates) during the dormant period to verify data continuity. The sound pressure value is reconstructed and calculated using spatial interpolation weights and compared with the original recorded value. Data points with deviations exceeding ±3% are marked as suspicious. The data that passes the verification is encrypted in blocks using the national secret SM4 algorithm with a key length of 128 bits. The encryption mode adopts CBC (cipher block chaining) mode to enhance security. The initialization vector (IV) is dynamically generated based on the sensor cluster ID and the current timestamp to ensure the uniqueness of each encryption.

[0091] The second target data sequence (spectral energy values ​​and frequency band labels) is weighted and adjusted based on the spatial interpolation weights in the spatiotemporal difference parameters. Energy values ​​in high-weighted regions (such as those near noise sources) are given a larger proportion to enhance feature stability. The adjusted energy values ​​are encrypted using a partially homomorphic encryption algorithm (such as Paillier encryption), which supports addition operations in ciphertext. Identity-based encryption (IBE) is performed on frequency band labels (such as "500Hz-1kHz"), with the key bound to the sensor cluster ID.

[0092] The sensor cluster identifier and edge node number are binary-encoded to generate a 64-bit digital watermark sequence. The watermark sequence is embedded into the low-order byte of the basic ciphertext sequence using the LSB (least significant bit) algorithm. The embedding strength is 1 bit of watermark for every 8 bytes of data. The basic ciphertext sequence and characteristic ciphertext sequence embedded with the watermark are concatenated with the synchronized timestamp in the spatiotemporal differential parameter. The SHA-256 hash value of the concatenated data is calculated to generate a 32-byte digital fingerprint as the unique identifier of the encrypted data block.

[0093] The layered encryption strategy uses differentiated algorithms tailored to different data characteristics: symmetric encryption ensures efficiency for basic data, while homomorphic encryption supports cryptanalysis for feature data. Overall security meets the requirements of Information Security Level 3. A dynamic key generation mechanism (such as timestamp-based IV) ensures that the encryption results of the same data are different at different times, improving replay attack resistance by 90%. A dual verification mechanism for spatiotemporal differential parameters (timestamp comparison + spatial interpolation reconstruction) ensures that data has not been tampered with during transmission and storage, with a tamper detection rate exceeding 99.9%. Digital watermarking and hash binding provide end-to-end data traceability, accurately tracking the source node and processing time of data. Homomorphic encryption allows statistical analysis of spectral energy data (such as calculating average energy values) without decryption, meeting the privacy protection requirements of "available but invisible" industrial data. Watermark embedding does not affect the availability of the original data. Even if some data is damaged, the data source can still be restored through the remaining watermark information. The hash binding mechanism allows the receiver to quickly locate the missing data blocks in the event of packet loss during transmission, improving retransmission efficiency by 50%.

[0094] In a preferred embodiment of the present invention, based on the timestamp compensation and spatial interpolation weight in the spatiotemporal differential parameters, the sound pressure value and three-dimensional coordinates of the first target data sequence during the dormant period are subjected to data integrity verification. After passing the verification, encryption is performed using the national secret SM4 algorithm to generate a basic ciphertext sequence, including:

[0095] Based on the timestamp compensation information in the spatiotemporal differential parameters, performing timestamp continuity verification on the sound pressure value and the three-dimensional coordinates during the dormant period, eliminating abnormal timestamp jump data, and generating a verified data block;

[0096] Rearranging the verified data blocks according to the spatial interpolation weights of the three-dimensional sound field grid model, and encapsulating them into structured data packets containing sound pressure values, coordinates and interpolation weights;

[0097] The structured data packets are encrypted in groups using the CBC mode of the SM4 algorithm, and the watermark identifier and the corresponding sleep start timestamp are embedded in the header of each encrypted data packet to generate the basic ciphertext sequence.

[0098] In an embodiment of the present invention, a timestamp compensation value (such as +5ms, -3ms) is extracted from the spatiotemporal differential parameters, and the sound pressure value recorded during the dormant period and the original timestamp of the three-dimensional coordinate are added with the corresponding compensation value to obtain a calibrated timestamp; the data points are sorted according to the calibrated timestamp, and the time interval of adjacent data points is calculated; a threshold is set (such as 10ms±2ms in industrial scenarios), and data points with a time interval exceeding the threshold (such as Δt>12ms or Δt<8ms) are eliminated and marked as time jump anomalies. The eliminated abnormal data points are complemented by linear interpolation of the three adjacent normal data points to generate a time-continuous verified data block, and the verified data block is divided according to the spatial position of the three-dimensional sound field grid model. The data in each grid unit (such as a 1m×1m×1m cube) are grouped together, and a spatial interpolation weight (from the spatiotemporal differential parameters) of the corresponding grid unit is assigned to each data point. The weight value reflects the point in the sound field reconstruction. The importance of the data is determined (for example, the weight of points near the noise source is >1.0, and the weight of points in the far field is <1.0). The sound pressure value, three-dimensional coordinates, and interpolation weight of each data point are combined into a triplet (sound pressure value, coordinates, weight) and arranged in grid order to form a structured data packet. A 128-bit SM4 encryption key is generated using the PBKDF2 algorithm based on the sensor cluster ID, current date, and edge node random number (for example, K = PBKDF2(cluster ID + date + random number, iterations = 10000)). The structured data packet is grouped into 16-byte blocks, and PKCS#7 padding is added if the number is less than 16 bytes. The initialization vector (IV) uses the lower 128 bits of the current timestamp and is encrypted block by block in CBC mode (the current block encryption depends on the previous ciphertext). A 16-byte extension field is added to the header of each encrypted data packet. The first 8 bytes store the digital watermark identifier (derived from the sensor cluster ID and edge node number), and the last 8 bytes store the sleep start timestamp.

[0099] Timestamp continuity verification effectively eliminates abnormal data caused by clock drift or communication delays, reducing the continuity error of the sound pressure time series from ±5ms to ±1ms, ensuring accurate recording of noise trends. Spatial interpolation weight encapsulation associates data with the physical structure of the sound field, allowing for more accurate restoration of the three-dimensional sound field distribution during subsequent reconstruction (for example, noise source localization accuracy is improved from meters to decimeters). SM4-CBC encryption, combined with dynamic IV and timestamps, achieves a "one-time, one-pad" effect, offering superior protection against known-plaintext attacks and brute-force cracking compared to traditional symmetric encryption (such as DES). Watermark embedding makes ciphertext data traceable, allowing for rapid location of the responsible node when a data leak is discovered (with 100% traceability accuracy). The structured data packet design supports flexible expansion (for example, environmental parameters such as temperature and humidity can be added later) without modifying the encryption framework. The timestamp in the packet header allows the receiver to quickly filter data for the required time period, improving query efficiency by 40% (compared to ciphertext data without a time index).

[0100] In a preferred embodiment of the present invention, step S4: performing data correction and rendering control based on the basic ciphertext sequence and the characteristic ciphertext sequence to generate a noise deduction map with spatiotemporal continuity includes:

[0101] Decrypting the basic ciphertext sequence and the characteristic ciphertext sequence respectively, extracting the sensor cluster identifier and the hash-bound synchronization timestamp in the digital watermark identifier, and performing spatiotemporal alignment of the decrypted sound pressure value and three-dimensional coordinates with the first target data sequence of the non-sleeping sensor cluster in the edge computing node based on the synchronization timestamp;

[0102] Extracting the decrypted sound pressure value and three-dimensional coordinates from the basic ciphertext sequence, and calculating the sound pressure deviation matrix and coordinate offset between the decrypted data and the unencrypted data in the same time window;

[0103] Based on the sound pressure deviation matrix and the predicted value of the sound field intensity change in the time-space difference parameter, the decrypted sound pressure value is corrected by a back propagation algorithm to generate a de-biased three-dimensional sound field reconstruction correction parameter;

[0104] Extracting the decrypted spectrum energy value and frequency band label from the characteristic ciphertext sequence, verifying the hash integrity of the frequency band label, and then comparing them with the unencrypted second target data sequence item by item according to the frequency band label to calculate the energy distribution error;

[0105] Based on the energy distribution error, dynamically weighted fusion is performed on the decrypted spectrum energy value to generate spectrum energy distribution correction parameters;

[0106] Based on the three-dimensional sound field reconstruction correction parameters and the spectrum energy distribution correction parameters, noise intensity spatial interpolation rendering is performed according to the spatial interpolation weight, and a spatiotemporally continuous noise deduction map is generated in combination with the synchronization timestamp.

[0107] In this embodiment of the present invention, the basic ciphertext sequence is decrypted using the SM4 algorithm. The input key is the sensor cluster key pre-stored in the edge computing node (the same as the encryption key in step S3). The decrypted sound pressure value, three-dimensional coordinates, and embedded sensor cluster identifier are extracted. The characteristic ciphertext sequence is decrypted using a homomorphic decryption algorithm. The frequency band label is first decrypted to verify integrity, and then an additive homomorphic operation is performed on the spectrum energy value to restore the original data.

[0108] Space-time reference calibration:

[0109] The sensor cluster identifier is extracted from the digital watermark and matched with the spatial coordinate range of the cluster in the three-dimensional sound field grid. Through the hash-bound synchronization timestamp (accurate to milliseconds), the decrypted data is aligned with the first target data sequence collected in real time by the non-sleeping sensor according to the timestamp (time difference <2ms), ensuring that the two types of data can be directly compared in the spatiotemporal dimension.

[0110] Within a 10-millisecond time window, the decrypted sound pressure value is compared with the unencrypted real-time sound pressure value grid point by grid point, and the absolute deviation is calculated (e.g., ΔdB = |decrypted value - real-time value|). A sound pressure deviation matrix covering the entire area is generated, and the three-dimensional coordinate offset is simultaneously calculated (e.g., the position drift of the dormant cluster node due to vibration, and the coordinate difference after calibration by the UWB positioning system). Grid points with an offset exceeding 0.5 meters are marked as abnormal coordinate points.

[0111] Three-dimensional sound field reconstruction and correction, back propagation correction algorithm:

[0112] The predicted value of the sound field intensity change in the time-space difference parameter (such as the dynamic sound field change generated in step S2) is input as prior knowledge to perform weighted correction on the sound pressure deviation matrix.

[0113] For high-deviation grid points (ΔdB>5dB), back-propagation calculations are performed using the historical sound field data of adjacent 3×3×3 grids. The current sound pressure value is adjusted by gradient descent (such as reducing the impact of prediction error on subsequent deductions) to generate de-deviation three-dimensional sound field reconstruction correction parameters.

[0114] Spectrum energy distribution calibration, frequency band integrity verification:

[0115] Calculate the SHA-256 hash value of the decrypted frequency band label (such as "1kHz-1.25kHz") and compare it with the hash value stored before encryption to eliminate abnormal frequency bands with hash mismatches (anti-tampering bit error rate <0.01%).

[0116] Dynamic Weighted Fusion:

[0117] Compare the decrypted energy value with the real-time energy value item by item according to the frequency band label, and calculate the relative error (if the error is greater than 10%, a weighted adjustment is triggered).

[0118] The decrypted value is retained for the low-error frequency band (≤5%), arithmetic average fusion is used for the medium-error frequency band (5%-10%), and the spectrum energy distribution correction parameters are generated for the high-error frequency band (>10%) based on real-time data and supplemented by decrypted data (weight ratio 8:2).

[0119] Time-space continuous rendering generation, spatial interpolation rendering:

[0120] Based on the three-dimensional sound field reconstruction correction parameters, inverse distance weighted interpolation is performed on the grid points in the entire area (the weight comes from the spatial interpolation weight of the time-space difference parameter) to render the spatial distribution of noise intensity (such as using rainbow scale to represent 60dB-100dB sound pressure level).

[0121] Time dimension continuity processing:

[0122] The synchronized timestamp is used as the time axis index, and the sliding average method (window length 500ms) is used to smooth the noise intensity changes in adjacent time frames to avoid map flickering caused by sensor sleep (inter-frame difference <1dB). Finally, a spatiotemporal continuous noise deduction map updated at a frequency of 10Hz is generated.

[0123] The back-propagation algorithm, combined with the predicted value of sound field changes, reduces the sound pressure deviation from an average of ±4dB to ±1.5dB. In particular, the reconstruction accuracy in the sensor's dormant area is improved by 60% (for example, the noise value error of the long-dormant cluster in the corner of the factory is reduced from 8% to 3%). Dynamic weighted fusion of spectral energy data preserves the continuity of characteristic frequency bands before and after dormancy (for example, the recognition rate of the 1.25kHz energy peak corresponding to equipment failure is increased to 95%), avoiding the risk of misjudgment from a single data source.

[0124] Synchronous timestamp calibration and sliding average processing eliminate data gaps caused by dormancy, improve the smoothness of the noise map time axis by 80% (traditional methods often exhibit step-like jumps due to data loss), and support smooth playback of noise evolution processes (for example, a complete presentation of the gradual change curve of factory noise from daytime production peak to nighttime attenuation within 2 hours).

[0125] The combination of spatial interpolation weights and three-dimensional grid modeling enables map rendering to reach a resolution of 0.5m × 0.5m × 1m (vertical direction), clearly showing the differences in noise distribution on different floors (for example, the detail that the noise in the corridor on the second floor is 5dB lower than that on the first floor can be accurately presented).

[0126] The layered decryption and hash verification mechanism ensures decrypted data accuracy >99.9%, effectively preventing bit errors and malicious tampering during transmission (for example, frequency band label errors caused by network attacks can be 100% identified and eliminated).

[0127] The correction processing of dormant cluster data enables the system to maintain map accuracy when the sensor coverage rate drops to 70% (traditional solutions will show significant distortion below 85% coverage), and adapt to abnormal scenarios such as industrial field equipment failure or communication interruption.

[0128] The noise map dynamic deduction method based on multi-source heterogeneous data fusion in the above-mentioned embodiment of the present invention processes multi-source heterogeneous data through spatiotemporal alignment, realizes cross-domain fusion of low-dimensional features, generates a spatiotemporal continuous noise map, supports real-time / near real-time dynamic rendering of noise distribution, effectively reflects the evolution of noise over time and space, improves the real-time and accuracy of noise data, and can reflect the actual situation of noise in a more realistic, efficient and real-time manner.

[0129] like Figure 2 As shown, a noise map dynamic deduction system based on multi-source heterogeneous data fusion includes:

[0130] A generation module is used to perform spatiotemporal alignment processing on multiple clusters of low-dimensional feature vectors in an edge computing node to generate three-dimensional sound field reconstruction parameters of a first target data sequence and spectral energy distribution parameters of a second target data sequence;

[0131] A fusion module is used to dynamically control the dormancy of the sensor cluster and generate spatiotemporal differential parameters based on the three-dimensional sound field reconstruction parameters and spectrum energy distribution parameters;

[0132] An encryption module is used to implement layered encryption on the spatiotemporal differential parameters and two types of data sequences during the dormant period to generate a basic ciphertext sequence and a characteristic ciphertext sequence for verification;

[0133] The correction module is used to perform data correction and rendering control based on the encrypted ciphertext sequence to generate a noise deduction map with spatiotemporal continuity.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A noise map dynamic deduction method based on multi-source heterogeneous data fusion, characterized by: The method comprises: In the edge computing node, multiple clusters of low-dimensional feature vectors are subjected to spatiotemporal alignment to generate three-dimensional sound field reconstruction parameters of the first target data sequence and spectrum energy distribution parameters of the second target data sequence; Based on the three-dimensional sound field reconstruction parameters and spectrum energy distribution parameters, the sensor cluster is dynamically controlled to sleep and the spatiotemporal differential parameters are generated. Layered encryption is performed on the two types of data sequences, namely, the time-space differential parameters and the dormant period, to generate a basic ciphertext sequence and a characteristic ciphertext sequence for verification; Data correction and rendering control are performed based on the basic ciphertext sequence and the characteristic ciphertext sequence to generate a noise deduction map with spatiotemporal continuity; In the edge computing node, before performing spatiotemporal alignment processing on the multiple clusters of low-dimensional feature vectors to generate the three-dimensional sound field reconstruction parameters of the first target data sequence and the spectrum energy distribution parameters of the second target data sequence, the method further includes: Cluster topology management is performed on multiple noise sensor nodes in the industrial park, sensor clusters are dynamically divided according to node signal strength, and the intra-cluster communication link topology is generated; A cluster head node is set based on the intra-cluster communication link topology, and the cluster head node performs a pre-fusion calculation on the original noisy data stream to generate de-redundant compressed data; the cluster head node converts the compressed data into a low-dimensional feature vector, and transmits the low-dimensional feature vector to the edge computing node through the edge gateway as an input of the multi-cluster low-dimensional feature vector; In the edge computing node, a spatiotemporal alignment process is performed on multiple clusters of low-dimensional feature vectors to generate three-dimensional sound field reconstruction parameters of the first target data sequence and spectrum energy distribution parameters of the second target data sequence, including: Based on the multiple clusters of low-dimensional feature vectors, the multiple clusters of feature vectors are spatially aligned through the convolution layer of the lightweight hybrid neural network to generate spatially aligned feature vectors; and the spatially aligned feature vectors are temporally aligned using a long short-term memory network layer to generate a continuous data stream; The continuous data stream is divided into data blocks with a time window length of 10 milliseconds, and a three-dimensional sound field grid model is constructed based on the spatial coordinates of each node in the sensor cluster; the sound field intensity values ​​are filled in for discrete data points in the three-dimensional sound field grid model using an inverse distance weighted interpolation algorithm to generate three-dimensional sound field reconstruction parameters for the first target data sequence; The 10 millisecond data blocks are subjected to fast Fourier transform to extract 1 / 3 octave spectrum energy, and multi-sensor data are fused through an energy superposition algorithm to generate spectrum energy distribution parameters of the second target data sequence.

2. The noise map dynamic deduction method based on multi-source heterogeneous data fusion according to claim 1 is characterized in that: Based on the multiple clusters of low-dimensional feature vectors, spatially aligning the multiple clusters of feature vectors through a convolutional layer of a lightweight hybrid neural network to generate a spatially aligned feature vector; The spatially aligned feature vectors are temporally aligned using a long short-term memory network layer to generate a continuous data stream, including: The spatial features of multiple clusters of low-dimensional feature vectors are extracted using the multi-layer convolution kernel of the convolution layer, and the coordinate data of different sensor clusters are normalized to the same spatial reference system through the feature mapping matrix; A long short-term memory network layer is used to perform sliding window analysis on the timing characteristics of continuous data streams, and a gating mechanism is used to dynamically compensate for the time offset of adjacent data blocks to achieve millisecond-level time synchronization.

3. The noise map dynamic deduction method based on multi-source heterogeneous data fusion according to claim 2 is characterized in that: Performing a fast Fourier transform on the 10 millisecond data block to extract 1 / 3 octave spectrum energy, and fusing multi-sensor data through an energy superposition algorithm to generate spectrum energy distribution parameters of the second target data sequence, including: Performing Hanning windowing processing on the time domain signal of each of the 10 millisecond data blocks to eliminate spectrum leakage, and then obtaining frequency domain energy distribution through fast Fourier transform; Based on the standard 1 / 3 octave frequency band division rule, the frequency domain energy distribution is segmented according to the center frequency, the energy value in each frequency band is integrated and calculated, and the initial energy distribution parameters are generated with the frequency band as the index; Based on the spatial distribution density of each node in the sensor cluster, spatial weighted fusion is performed on the initial energy distribution parameters to generate spectrum energy distribution parameters of the second target data sequence.

4. The noise map dynamic deduction method based on multi-source heterogeneous data fusion according to claim 3 is characterized in that: Based on the 3D sound field reconstruction parameters and spectrum energy distribution parameters, the sensor cluster is dynamically controlled to sleep and generate spatiotemporal differential parameters, including: Calculate the activity score of each sensor cluster based on the sound field intensity value in the three-dimensional sound field reconstruction parameter and the frequency band energy ratio in the spectrum energy distribution parameter; if the activity score is lower than a preset threshold, send a sleep instruction to the corresponding sensor cluster and record the sleep start timestamp and the spatiotemporal coordinates of the missing data during the sleep period; Based on the recorded sleep start timestamp and spatiotemporal coordinates, obtaining three-dimensional sound field reconstruction parameters before sleep, and obtaining three-dimensional sound field reconstruction parameters after sleep after sleep ends, and calculating the sound field intensity difference matrix between the two; The sound field intensity difference matrix is ​​input into the spatiotemporal Kalman filter algorithm. By fusing the historical sound field intensity change rate and the spatiotemporal distribution characteristics of the current sensor cluster, the dynamic sound field intensity change in the dormant area is predicted, and the spatiotemporal difference parameters including timestamp compensation and spatial interpolation weights are generated.

5. The noise map dynamic deduction method based on multi-source heterogeneous data fusion according to claim 4 is characterized in that: Layered encryption is performed on the spatiotemporal differential parameters and two types of data sequences during the dormant period to generate basic ciphertext sequences and characteristic ciphertext sequences for verification, including: Based on the timestamp compensation and spatial interpolation weight in the spatiotemporal differential parameters, the sound pressure value and three-dimensional coordinates of the first target data sequence during the dormant period are checked for data integrity. After the check passes, the data is encrypted using the national secret SM4 algorithm to generate a basic ciphertext sequence; For the spectrum energy values ​​and frequency band labels of the second target data sequence during the dormant period, adjusting the energy distribution ratio according to the spatial interpolation weight in the spatiotemporal difference parameter, and encrypting the adjusted energy values ​​and frequency band labels using a homomorphic encryption algorithm to generate a characteristic ciphertext sequence; The basic ciphertext sequence and the characteristic ciphertext sequence are input into the watermark embedding module, a digital watermark identifier is generated based on the identifier of the sensor cluster and the edge node number, and the digital watermark identifier is hashed and bound with the synchronization timestamp in the spatiotemporal differential parameter to generate an encrypted data block for verification.

6. The noise map dynamic deduction method based on multi-source heterogeneous data fusion according to claim 5 is characterized in that: Based on the timestamp compensation and spatial interpolation weight in the spatiotemporal differential parameters, the sound pressure value and three-dimensional coordinates of the first target data sequence during the dormant period are subjected to data integrity verification. After passing the verification, the data is encrypted using the national secret SM4 algorithm to generate a basic ciphertext sequence, including: Based on the timestamp compensation information in the spatiotemporal differential parameters, performing timestamp continuity verification on the sound pressure value and the three-dimensional coordinates during the dormant period, eliminating abnormal timestamp jump data, and generating a verified data block; Rearranging the verified data blocks according to the spatial interpolation weights of the three-dimensional sound field grid model, and encapsulating them into structured data packets containing sound pressure values, coordinates and interpolation weights; The structured data packets are encrypted in groups using the CBC mode of the SM4 algorithm, and the watermark identifier and the corresponding sleep start timestamp are embedded in the header of each encrypted data packet to generate the basic ciphertext sequence.

7. The noise map dynamic deduction method based on multi-source heterogeneous data fusion according to claim 6 is characterized in that: Data correction and rendering control are performed based on the basic ciphertext sequence and the characteristic ciphertext sequence to generate a noise deduction map with spatiotemporal continuity, including: Decrypting the basic ciphertext sequence and the characteristic ciphertext sequence respectively, extracting the sensor cluster identifier and the hash-bound synchronization timestamp in the digital watermark identifier, and performing spatiotemporal alignment of the decrypted sound pressure value and three-dimensional coordinates with the first target data sequence of the non-sleeping sensor cluster in the edge computing node based on the synchronization timestamp; Extracting the decrypted sound pressure value and three-dimensional coordinates from the basic ciphertext sequence, and calculating the sound pressure deviation matrix and coordinate offset between the decrypted data and the unencrypted data in the same time window; Based on the sound pressure deviation matrix and the predicted value of the sound field intensity change in the time-space difference parameter, the decrypted sound pressure value is corrected by a back propagation algorithm to generate a de-biased three-dimensional sound field reconstruction correction parameter; Extracting the decrypted spectrum energy value and frequency band label from the characteristic ciphertext sequence, verifying the hash integrity of the frequency band label, and then comparing them with the unencrypted second target data sequence item by item according to the frequency band label to calculate the energy distribution error; Based on the energy distribution error, dynamically weighted fusion is performed on the decrypted spectrum energy value to generate spectrum energy distribution correction parameters; Based on the three-dimensional sound field reconstruction correction parameters and the spectrum energy distribution correction parameters, noise intensity spatial interpolation rendering is performed according to the spatial interpolation weight, and a spatiotemporally continuous noise deduction map is generated in combination with the synchronization timestamp.

8. A noise map dynamic deduction system based on multi-source heterogeneous data fusion, characterized by: Applied to the method according to any one of claims 1 to 7, comprising: A generation module is used to perform spatiotemporal alignment processing on multiple clusters of low-dimensional feature vectors in an edge computing node to generate three-dimensional sound field reconstruction parameters of a first target data sequence and spectral energy distribution parameters of a second target data sequence; A fusion module is used to dynamically control the dormancy of the sensor cluster and generate spatiotemporal differential parameters based on the three-dimensional sound field reconstruction parameters and spectrum energy distribution parameters; An encryption module is used to implement layered encryption on the spatiotemporal differential parameters and two types of data sequences during the dormant period to generate a basic ciphertext sequence and a characteristic ciphertext sequence for verification; The correction module is used to perform data correction and rendering control based on the basic ciphertext sequence and the characteristic ciphertext sequence to generate a noise deduction map with spatiotemporal continuity; In the edge computing node, before performing spatiotemporal alignment processing on the multiple clusters of low-dimensional feature vectors to generate the three-dimensional sound field reconstruction parameters of the first target data sequence and the spectrum energy distribution parameters of the second target data sequence, the method further includes: Cluster topology management is performed on multiple noise sensor nodes in the industrial park, sensor clusters are dynamically divided according to node signal strength, and the intra-cluster communication link topology is generated; A cluster head node is set based on the intra-cluster communication link topology, and the cluster head node performs a pre-fusion calculation on the original noisy data stream to generate de-redundant compressed data; the cluster head node converts the compressed data into a low-dimensional feature vector, and transmits the low-dimensional feature vector to the edge computing node through the edge gateway as an input of the multi-cluster low-dimensional feature vector; In the edge computing node, a spatiotemporal alignment process is performed on multiple clusters of low-dimensional feature vectors to generate three-dimensional sound field reconstruction parameters of the first target data sequence and spectrum energy distribution parameters of the second target data sequence, including: Based on the multiple clusters of low-dimensional feature vectors, the multiple clusters of feature vectors are spatially aligned through the convolution layer of the lightweight hybrid neural network to generate spatially aligned feature vectors; and the spatially aligned feature vectors are temporally aligned using a long short-term memory network layer to generate a continuous data stream; The continuous data stream is divided into data blocks with a time window length of 10 milliseconds, and a three-dimensional sound field grid model is constructed based on the spatial coordinates of each node in the sensor cluster; the sound field intensity values ​​are filled in for discrete data points in the three-dimensional sound field grid model using an inverse distance weighted interpolation algorithm to generate three-dimensional sound field reconstruction parameters for the first target data sequence; The 10 millisecond data blocks are subjected to fast Fourier transform to extract 1 / 3 octave spectrum energy, and multi-sensor data are fused through an energy superposition algorithm to generate spectrum energy distribution parameters of the second target data sequence.

Images