Multi-city noise automatic monitoring data analysis method and system based on time period classification

Abstract

The present invention discloses a multi-city noise automatic monitoring data analysis method and system based on time period classification, which specifically relates to the field of urban noise automatic monitoring, including collecting sound pressure data and performing time slot slicing, thereby separating urban common and specific features through feature weighted classification and feature decoupling migration, and realizing dynamic analysis and evolution of multi-city noise monitoring data. The multi-city noise automatic monitoring data analysis method and system based on time period classification accurately captures the dynamic evolution law of noise in the time domain and frequency domain by constructing an association framework between time period features and noise patterns, effectively reducing the regional misjudgment rate of cross-time period monitoring; through a dual adaptive mechanism, combined with feature decoupling migration and incremental learning, cross-time period and cross-city feature migration and dynamic evolution are realized, significantly reducing the occurrence of period-specific false alarms; by forming a self-optimizing closed loop, passive correction is converted into active learning, solving the pain point of frequent manual intervention in the existing technology.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic urban noise monitoring, and more particularly to a method and system for analyzing multi-city noise automatic monitoring data based on time period classification. Background Art

[0002] With the acceleration of the intelligentization of urban environmental monitoring, the real-time identification and dynamic control of large-scale noise pollution sources in multiple cities are particularly important. Traditional noise monitoring consists of distributed sound pressure sensors, cellular network transmission modules and cloud-based data processing platforms. It collects voiceprint data through a periodic polling mechanism, extracts spectral features through Fourier transform, and uploads them to the central server for noise classification. However, in actual applications, there are data transmission delays of up to minutes, and node energy consumption is concentrated in the spectrum calculation link, which makes it difficult to meet the real-time response needs of large-scale data from multiple cities and multiple measurement points.

[0003] In order to solve the bottlenecks of data transmission delay and energy consumption in traditional architecture, the existing improved technology proposes to combine big data training models with multi-layer wavelet transforms to achieve hierarchical compression expression of voiceprint features, and ultimately realize automatic analysis, intelligent recognition and manual supervision of noise.

[0004] However, it still has some shortcomings in actual use. For example, the current monitoring system faces a bottleneck in adaptability to dynamic environments. Although the general model based on static big data training improves the efficiency of single-point recognition, it does not take into account the dynamic evolution of urban noise sources in the time dimension under multi-city heterogeneous scenarios. Since the association framework between time period characteristics and noise patterns has not been constructed, the noise classification threshold and feature weight cannot be dynamically adjusted during typical time periods such as day and night transitions and commuting peaks, resulting in a surge in regional misjudgment rates during cross-time period monitoring, and an increase in the frequency of manual supervision to correct time period false alarms. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a multi-city noise automatic monitoring data analysis method and system based on time period classification, and solves the problems raised in the above-mentioned background technology through the following scheme.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The multi-city noise automatic monitoring data analysis method based on time period classification includes:

[0008] S1: Acquire noise monitoring data through a multi-city sensor node network. The noise monitoring data includes a sound pressure signal acquired through dual-mode positioning and high-precision clock synchronization, a millisecond-level timestamp synchronized with the sound pressure signal, and geographic location information.

[0009] S2: performing a spatiotemporal feature extraction operation based on the noise monitoring data, wherein the spatiotemporal feature extraction operation slices the sound pressure signal into time slots and uses an attention mechanism to generate a dynamic time period fingerprint map containing time-sensitive weights and spatial correlation weights;

[0010] S3: performing noise iterative analysis on the dynamic period fingerprint including feature weighted classification, feature decoupling migration, incremental learning and abnormal feedback;

[0011] S4: Continuously perform clock alignment between nodes across cities and use clock signals as basic timing constraints for feedback adjustment.

[0012] Preferably, the S1, in the multi-city sensor node network, automatically triggers a scene switching event by determining the spatial relationship between the coordinate point corresponding to the sensor node and the preset geographic fence, specifically including:

[0013] Load the baseline noise signature library associated with the target geo-fence and send an update signal of the dynamic period fingerprint map to the monitoring data central processing module;

[0014] Reconfigure the noise threshold based on the target geofence scenario type and target time period;

[0015] Dynamically adjust the data upload frequency based on the scene sensitivity of the target geofence.

[0016] Preferably, the step S2, generating a dynamic time period fingerprint map, specifically includes:

[0017] The power spectrum values ​​corresponding to the continuous PCM waveform data in the noise monitoring data are formed into a power spectrum matrix in frame order, and input into a time slicing convolutional network to obtain time domain features, frequency domain features and compressed feature vectors;

[0018] Using time slot indexing , generating time slot embedding vectors through a trainable lookup table;

[0019] Integrated attention mechanism, based on time slot embedding vector , construct the query vector , key vector and value vector ;

[0020] Based on query vector and key vector , calculate the attention weight , specifically expressed as:

[0021] ,

[0022] in, Represented as a key vector Dimensions, Represented as query vector and key vector The matrix product of the transpose of ;

[0023] Through attention weight Pair value vector Perform weighted summation to obtain the time period sensitive feature vector , specifically expressed as:

[0024] ;

[0025] The compressed feature vector, the time slot index, the time period sensitive feature vector and the power spectrum matrix corresponding to each city are encapsulated to form a dynamic time period fingerprint map.

[0026] Preferably, the S2 is to calculate the power spectrum value corresponding to the continuous PCM waveform data in the noise monitoring data. , specifically expressed as:

[0027] ,

[0028] in, Expressed as The power spectrum value of the frame, Expressed as the number of sampling points corresponding to the Hamming window, Represents the index of the sampling point corresponding to the Hamming window, Expressed as The signal value of the sampling point, Expressed as the Hamming window function value, Expressed as the basis functions of the discrete Fourier transform.

[0029] Preferably, the step S3, performing feature weighted classification, specifically includes:

[0030] By constructing an LSTM time period prediction unit and a noise pattern discriminator driven by a gradient reversal network, a dual adaptive mechanism is formed;

[0031] Adversarial training is performed on the dynamic period fingerprint to generate the probability distribution of each noise type.

[0032] Preferably, the step S3, performing feature decoupling migration, specifically includes:

[0033] By constructing a feature decoupling distillation framework, the feature space of the multi-city noise analysis model is decomposed into a city-independent common subspace and a city-specific private subspace, thus realizing the cross-domain transfer of common features.

[0034] Preferably, the step S3, performing incremental learning, specifically includes:

[0035] Adopting an elastic weight solidification algorithm to realize the dynamic evolution of the multi-city noise analysis model;

[0036] If a periodic false alarm is detected, the common feature parameters are frozen and the specific feature parameters are dynamically adjusted, and incremental updates are achieved through a dual-channel back-propagation mechanism.

[0037] Preferably, the step S3, executing abnormal feedback construction, specifically includes:

[0038] A spatiotemporal joint anomaly detection mechanism based on Mahalanobis distance;

[0039] If the classification result deviates from the preset threshold of the historical period pattern, the manual review interface is automatically triggered and a training sample stream with corrected labels is generated.

[0040] To achieve the above-mentioned object, the present invention provides the following technical solution: a multi-city noise automatic monitoring data analysis system based on time period classification, comprising a multi-city noise monitoring database and a monitoring data central processing module, and implementing the multi-city noise automatic monitoring data analysis method based on time period classification, comprising:

[0041] Multi-source spatiotemporal data acquisition module: This module acquires noise monitoring data through a multi-city sensor node network. The noise monitoring data includes sound pressure signals acquired through dual-mode positioning and high-precision clock synchronization, millisecond-level timestamps synchronized with the sound pressure signals, and geographic location information.

[0042] A time segment feature encoding module performs a spatiotemporal feature extraction operation based on the noise monitoring data. The spatiotemporal feature extraction operation slices the sound pressure signal into time slots and uses an attention mechanism to generate a dynamic time segment fingerprint with time-sensitive weights and spatial correlation weights.

[0043] Dynamic evolution collaboration module: performs noise iterative analysis on the dynamic period fingerprint map, including feature weighted classification, feature decoupling migration, incremental learning and abnormal feedback;

[0044] Clock synchronization module: continuously performs clock alignment between nodes across cities and uses clock signals as basic timing constraints for feedback adjustment;

[0045] The multi-city noise monitoring database includes all data texts of the multi-city noise automatic monitoring data analysis system based on time period classification, and collects information data output by each module in real time; the monitoring data central processing module is used for information data instructions output by each module in the central control system.

[0046] Preferably, the clock synchronization module continuously performs clock alignment between nodes across cities, specifically including:

[0047] The master clock node sends synchronization messages every 100ms;

[0048] Use linear regression compensation algorithm to eliminate transmission delay jitter;

[0049] Maintain network-wide clock deviation less than 1μs.

[0050] The technical effects and advantages of the present invention are as follows:

[0051] 1. Through time slot slicing and an attention mechanism, this paper converts sound pressure signals into dynamic time-period fingerprints containing time-sensitive weights and spatial correlation weights. This framework establishes an association framework between time-period characteristics and noise patterns, accurately capturing the dynamic evolution of noise in both the time and frequency domains during periods such as daytime and peak commuting times. This enables the model to adaptively adjust the classification strategy based on the characteristic weights of different time periods, effectively reducing the regional misjudgment rate in cross-period monitoring.

[0052] 2. This invention uses a dual adaptive mechanism of LSTM time period prediction units and gradient reversal networks, combined with feature decoupling migration and incremental learning, to achieve cross-time period and cross-city feature migration of noise patterns and dynamic model evolution. While freezing common features, it dynamically optimizes city-specific parameters. This not only retains the classification experience of historical time periods, but also can quickly adapt to noise changes in new time periods, significantly reducing the occurrence of time-specific false alarms.

[0053] 3. The present invention uses a spatiotemporal joint detection mechanism based on Mahalanobis distance to automatically trigger manual review and generate a correction sample stream when classification is abnormal, forming a self-optimizing closed loop of the monitoring system. By reducing the frequency of manual active supervision, passive correction is transformed into active learning, effectively reducing the cost of manual supervision and solving the pain point of frequent manual intervention in the existing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a diagram of the implementation steps of the multi-city noise automatic monitoring data analysis method based on time period classification provided in an embodiment of the present application.

[0055] Figure 2 This is a diagram of the implementation steps of noise iterative analysis in the multi-city noise automatic monitoring data analysis method based on time period classification provided in an embodiment of the present application.

[0056] Figure 3 This is a block diagram of the architecture of a multi-city noise automatic monitoring data analysis system based on time period classification provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0058] The terms used in the following embodiments of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification of the present application, the singular expressions "one", "a kind of", "said", "above", "the" and "this" are intended to also include plural expressions, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in the present application refers to and includes any or all possible combinations of one or more listed items. In the description of the embodiments of the present application, unless otherwise stated, the meaning of "multiple" is two or more.

[0059] As attached Figure 1 The time-slot classification-based multi-city noise automatic monitoring data analysis method shown here collects time-stamped sound pressure data and slices it into time slots. It then separates common and specific city features through feature weighted classification and feature decoupling and migration, achieving dynamic analysis and evolution of multi-city noise monitoring data. Specifically, it includes the following steps:

[0060] S1: Acquire noise monitoring data through a multi-city sensor node network. The noise monitoring data includes a sound pressure signal acquired through dual-mode positioning and high-precision clock synchronization, a millisecond-level timestamp synchronized with the sound pressure signal, and geographic location information.

[0061] S2: performing a spatiotemporal feature extraction operation based on the noise monitoring data, wherein the spatiotemporal feature extraction operation slices the sound pressure signal into time slots and uses an attention mechanism to generate a dynamic time period fingerprint map containing time-sensitive weights and spatial correlation weights;

[0062] S3: performing noise iterative analysis on the dynamic period fingerprint including feature weighted classification, feature decoupling migration, incremental learning and abnormal feedback;

[0063] S4: Continuously perform clock alignment between nodes across cities and use clock signals as basic timing constraints for feedback adjustment.

[0064] Specifically, in S1, by deploying multiple sensor nodes in multi-city areas, the sensor nodes integrate positioning functions with GPS and Beidou dual-mode positioning capabilities, high-precision clock synchronization functions and acoustic signal acquisition functions, and complete the necessary network configuration and initialization; the data packets containing the microsecond timestamp provided by the synchronized clock, the longitude and latitude obtained by the fusion positioning, the instantaneous sound pressure level, the PCM waveform data of the preset sampling window and the dynamically associated geographic fence identifier are encapsulated as noise monitoring data.

[0065] In this embodiment, the acoustic signal acquisition function captures raw PCM waveform data at a 48 kHz sampling rate through the ALSA audio driver layer, and applies a real-time noise dose calculation method to calculate the instantaneous sound pressure level.

[0066] Furthermore, the high-precision clock synchronization function is based on the improved IEEE 1588 PTPv2.1 protocol stack, which periodically sends Sync messages containing sending timestamps through the master clock node. , record the arrival time after receiving the message from the node , obtain the slave node response time through message interaction and master node reception time , to calculate the delay compensation , specifically expressed as:

[0067] ,

[0068] This embodiment uses a Kalman filter to eliminate network jitter from the synchronization result of delay compensation, so that the clock deviation between nodes is less than 1 μs.

[0069] Furthermore, the underlying API of the corresponding device of each sensor node is called, and the GPS and Beidou dual-mode positioning engine is integrated to obtain GPS positioning data in the WGS-84 coordinate system in real time. and Beidou positioning data and its corresponding horizontal dilution of precision ; Then execute the multi-source positioning data fusion algorithm, and obtain the coordinate points corresponding to each sensor node according to the dynamic weighting formula , specifically expressed as:

[0070] ,

[0071] in, and Respectively expressed as and The corresponding weights in the fusion process; this embodiment The value is set to , Expressed as GPS horizontal dilution of precision; The value of , Expressed as the horizontal precision factor of the Beidou system, it ensures that the positioning error of the weighted fusion is less than 3 meters, and the fused latitude and longitude data are obtained at an interval of 50ms.

[0072] It should be noted that by pre-loading the multi-city GIS geo-fence data stored in the multi-city noise monitoring database, the ray method point-surface inclusion judgment algorithm is executed in real time to judge the spatial relationship between the coordinate point corresponding to the sensor node and the preset geo-fence, and dynamically associate the acoustic scene tags. The acoustic scene tags include but are not limited to residential quiet areas, industrial operation areas, traffic artery areas, commercial gathering areas, cultural, educational and medical areas, parks and green spaces, etc.; if the mobile node crosses the fence boundary, the triple scene switching event is automatically triggered.

[0073] In one possible embodiment, the scene switching event includes: loading a baseline noise feature library associated with the target geofence and sending an update signal of the dynamic time period fingerprint map to the monitoring data central processing module; reconfiguring the noise judgment threshold according to the scene type and target time period of the target geofence; and dynamically adjusting the data upload frequency according to the scene sensitivity of the target geofence.

[0074] In this embodiment, the residential quiet zone is the night time period from 22:00 to 6:00, with an equivalent sound level limit of ≤45dB(A); the industrial operation area is dominated by continuous mechanical noise, with the energy proportion of the 63Hz-4kHz frequency band being greater than 70%; the traffic artery area is a typical motor vehicle noise spectrum with a peak frequency of 500Hz±50Hz; the commercial gathering area is a mixed noise of human voices, with a sound pressure fluctuation variance of >15dB in the 1-2kHz frequency band; the cultural, educational and medical area is a sudden high-noise event threshold of ≤65dB(C) and a duration of <200ms; the park green space is a natural soundscape with intermittent pulses of bird songs, wind sounds, etc. in the 2-8kHz frequency band.

[0075] Specifically, in S2, the noise monitoring data is received, and time slot-based slicing and preprocessing are performed to divide the 24-hour period into minute-level time slots. The time-slicing convolutional network is combined with the attention mechanism to extract the time-period sensitive feature vector to generate a dynamic time-period fingerprint map including time-sensitive weights and spatial correlation weights. The dynamic time-period fingerprint map provides a spatiotemporal coupling feature expression for noise classification.

[0076] It should be noted that the slicing and preprocessing of the time slots include: using a sliding window mechanism to divide the continuous PCM waveform data in the noise monitoring data into 1440 independent 1-minute time slots according to a 24-hour period; based on the microsecond timestamps corresponding to the continuous PCM waveform data in the noise monitoring data, the time slot index corresponding to the noise data is determined by a periodic slicing algorithm. , specifically expressed as:

[0077] ,

[0078] in, , The noise monitoring data is represented by the microsecond timestamp corresponding to the continuous PCM waveform data; the sound pressure signal data in each time slot is preprocessed, and the preprocessing includes but is not limited to applying an A-weighted filter to filter the sound pressure signal, using a frame-by-frame windowing technique to frame the filtered signal using a Hamming window, and performing Fourier transform on each frame to calculate the corresponding power spectrum. , specifically expressed as:

[0079] ,

[0080] in, Expressed as The power spectrum value of the frame, this embodiment , Expressed as the number of sampling points corresponding to the Hamming window, Represents the index of the sampling point corresponding to the Hamming window, Expressed as The signal value of the sampling point, Expressed as the Hamming window function value, Represented as a basis function for the discrete Fourier transform, which is used to convert a signal from the time domain to the frequency domain.

[0081] In a possible implementation, generating a dynamic time period fingerprint includes: forming a power spectrum matrix from power spectrum values ​​corresponding to continuous PCM waveform data in the noise monitoring data in a frame order of a Hamming window, and inputting the matrix into a time-slicing convolutional network.

[0082] It should be noted that the time-slicing convolutional network includes multiple layers; the input layer receives the power spectrum matrix; the time domain convolution layer uses multiple 1D convolution kernels and the ReLU activation function to extract time domain features; the frequency domain convolution layer contains multiple 2D convolution kernels and uses the LeakyReLU activation function to extract frequency domain features; the attention pooling layer performs multi-scale feature compression through multiple scaling factors to obtain a compressed feature vector.

[0083] Furthermore, the time domain features are preliminary features extracted from the power spectrum matrix, reflecting the energy distribution pattern or change trend within a specific frequency range, including but not limited to the time domain envelope features of short-time energy mutations of transient events, the time domain periodic features, zero-crossing rate, autocorrelation function of short-time energy integral value and peak interval of continuous noise, etc. The transient events include but are not limited to impact sounds, explosion sounds, etc., and the continuous noise includes but is not limited to engine roars, fireworks explosions, etc.; the frequency domain features are more complex and discriminative spectral patterns further extracted on the basis of time domain features, including but not limited to frequency range energy distribution, the proportion of fundamental frequency and harmonic components, spectral centroid, special frequency identification, etc. The special frequencies include but are not limited to traffic noise, animal calls, etc.; the compressed feature vector is a compact representation that integrates and compresses the features extracted by time domain and frequency domain convolution.

[0084] In a possible implementation, generating a dynamic time period fingerprint map further includes: using a time slot index , generating time slot embedding vectors through a trainable lookup table ;Integrated attention mechanism, based on time slot embedding vector , construct the query vector , key vector and value vector , specifically expressed as;

[0085] ,

[0086] ,

[0087] ,

[0088] in, 、 and Represented as query vectors , key vector and value vector The corresponding weight matrix, It is represented as the power spectrum matrix corresponding to the time slot embedding vector, 、 and Represented as the adjusted query vector , key vector and value vector The offset of; based on the query vector and key vector , calculate the attention weight , specifically expressed as:

[0089] ,

[0090] in, Represented as a key vector Dimensions, Represented as query vector and key vector The matrix product of the transpose of ; through the attention weight Pair value vector Perform weighted summation to obtain the time period sensitive feature vector , specifically expressed as:

[0091] ;

[0092] In this embodiment, during the morning rush hour, i.e., 7:00-9:00, the noise feature weight is enhanced by applying a gradient amplification factor to the convolution kernel corresponding to the 500Hz frequency band; during the late night period, i.e., 22:00-6:00, the activation threshold of the feature channel in the 2-4kHz frequency band is increased by setting the feature channel activation threshold.

[0093] In a possible implementation, generating a dynamic time period fingerprint map further includes: encapsulating the compressed feature vector, the time slot index, the time period sensitive feature vector, and the power spectrum matrix corresponding to each city to form a dynamic time period fingerprint map.

[0094] Specifically, in S3, the feature weighted classification, feature decoupling migration, incremental learning and abnormal feedback form a technical closed loop through the shared feature space and parameter update chain. The result of the feature decoupling determines the parameter freezing range of the incremental learning. The model update amount generated by the incremental learning reversely constrains the adversarial training process of the feature decoupling. The threshold dynamic adjustment algorithm of the abnormal feedback depends on the distribution of common urban features output by the feature decoupling.

[0095] In one possible implementation, the noise iterative analysis includes: the feature weighted classification forms a dual adaptive mechanism by constructing an LSTM time period prediction unit and a noise pattern discriminator driven by a gradient reversal network to perform adversarial training on dynamic time period fingerprint maps.

[0096] It should be noted that the dynamic time period fingerprint map is used as input, and the probability distribution of each noise type is generated through the final classification layer of the noise pattern discriminator, and the time period weight and spatial correlation information are applied to the feature vector, and the final classification result is encapsulated. The encapsulation content includes: noise type code, probability array of each category, time dimension weight of the current time slot, spatial dimension weight of the current position and time slot index.

[0097] Furthermore, the network architecture of the LSTM time period prediction unit includes an input layer, an LSTM hidden layer, an attention fusion layer and an output layer, wherein the input layer is used to receive the dynamic time period fingerprint map and perform time series modeling based on the feature vector sequence in the dynamic time period fingerprint map, the LSTM hidden layer contains multiple memory units, the attention fusion layer is used to calculate the attention weight of the historical time period features, and the output layer is used to generate the time period feature weight matrix; the noise pattern discriminator constructed by the gradient reversal network includes a feature input layer, a gradient reversal layer, a fully connected layer and a classification output layer, wherein the feature input layer is used to receive fingerprint map features, the gradient reversal layer is used to reverse the gradient during back propagation, the fully connected layer contains multiple fully connected nodes, and the classification output layer is used to output the probability of each noise type.

[0098] In this example, the time step is set to 60, and the attention mechanism is applied to the output hidden state of the past 60 time steps to calculate the attention score of each hidden state , and calculate the attention weight through the Softmax function , specifically expressed as:

[0099] ,

[0100] in, Expressed as The attention weight of the hidden state corresponding to each time step, Expressed as The attention score of the hidden state corresponding to each time step, Represented as a preset weight vector, used to perform weighted summation on the transformation results of the hidden state, Represented as a preset weight matrix, used to map the hidden state, 、 Respectively expressed as and The hidden state of the LSTM hidden layer at each time step is obtained; the weighted summed hidden state is passed through a fully connected layer and a Sigmoid activation function to generate a predicted time period feature weight matrix.

[0101] Furthermore, an adversarial training mechanism is adopted to optimize the LSTM period prediction unit and the noise pattern discriminator driven by the gradient reversal network. In each training cycle: the LSTM unit generates prediction weights based on the input information, performs feature weighting on the original features and the prediction weights, obtains weighted features and inputs them into the noise pattern discriminator to obtain the classification probability of each noise type; the classification loss of the noise pattern discriminator and the loss of the LSTM prediction unit are calculated, and the total loss is defined, which is to maximize the loss of the noise pattern discriminator while minimizing the prediction error of the prediction unit; the total loss is backpropagated, the noise pattern discriminator parameters are normally updated according to the classification loss of the noise pattern discriminator, and the parameter update of the LSTM prediction unit is affected by the gradient reversal layer.

[0102] In a possible implementation, the noise iterative analysis also includes: the feature decoupling migration decomposes the feature space of the multi-city noise analysis model into a city-independent common subspace and a city-specific private subspace by constructing a feature decoupling distillation framework, thereby realizing cross-domain transmission of common features.

[0103] In this embodiment, the dynamic time period fingerprint map is taken as input and decomposed into common features and private features through two independent feature extraction channels. The common features are extracted by the ReLU function to extract universal features shared across cities, and the private features are extracted by the tanh function to extract city-specific features. An orthogonal regularization loss function is introduced to ensure that the common features and the private features are decoupled from each other in the feature space. The element-wise product of the common features and the private features is calculated and summed to obtain a correlation measure. The mean square value of the correlation measure is used as the orthogonal regularization loss term, and forced learning is used to generate independent common features and private features.

[0104] Furthermore, by using city-independent common features as the knowledge source, the common feature extraction capability is transferred to the model of the target city through knowledge distillation technology, and the learned common features are applied to the model of the target city. The common features representing specific city types are loaded from the pre-built common feature library, and then the common features are frozen so that they remain unchanged during the training process of the target city model. Only the layer responsible for extracting private features in the target model is trained to adapt it to the specific noise pattern of the target city while retaining the transferred common feature extraction capability.

[0105] In one possible implementation, the noise iterative analysis further includes: the incremental learning uses an elastic weight solidification algorithm to achieve dynamic evolution of the multi-city noise analysis model; if a periodic false alarm is detected, the common feature parameters are frozen and the specific feature parameters are dynamically adjusted; and the model is incrementally updated through a dual-channel back-propagation mechanism.

[0106] It should be noted that by calculating the diagonal elements of the Fisher information matrix corresponding to the parameters of each multi-city noise analysis model, its sensitivity to the change of the loss function on the marked time period fingerprint sample is evaluated, which is used as a measure of parameter importance. , specifically expressed as;

[0107] ,

[0108] in, Expressed as a parameter The importance measure, Expressed as the number of samples in the historical dataset, Represented as the sample index in the historical dataset, Represented as a model in the sample The loss function on Expressed as a loss function with respect to parameters gradient;

[0109] Furthermore, the model parameters are sorted in descending order of importance, and the cumulative importance ratio of the parameters when arranged in this order is calculated. The parameters whose cumulative importance ratio reaches a preset threshold are selected as the core parameter set. The preset threshold in this embodiment is 85%. For the parameters determined to be core parameters, their gradient calculation function is disabled in the subsequent incremental learning process, and the gradient calculation of the auxiliary parameters includes the standard loss gradient term and the EWC-based regularization term to balance the learning of new tasks and the retention of old task knowledge.

[0110] In one possible implementation, the noise iterative analysis also includes: the anomaly feedback constructs a spatiotemporal joint anomaly detection mechanism based on the Mahalanobis distance. If the classification result deviates from the preset threshold of the historical period pattern, the manual review interface is automatically triggered and a training sample stream with corrected labels is generated, thereby realizing a self-optimizing closed loop of the noise monitoring system.

[0111] It should be noted that the calculation of the current dynamic period fingerprint and the corresponding time slot Historical Mode The Mahalanobis distance is compared with the dynamically set threshold; the Mahalanobis distance The calculation is specifically expressed as:

[0112] ,

[0113] in, It is expressed as the inverse of the historical covariance matrix; the anomaly detection threshold is dynamically set based on the statistical distribution of the historical Mahalanobis distance, using the 3 standard deviation principle;

[0114] Furthermore, when it is detected that the Mahalanobis distance exceeds the dynamic threshold, a manual review process is triggered, and a sample stream for model self-optimization is generated based on the review results. In this embodiment, different strategies are adopted to generate and transmit correction samples according to the type and persistence of the anomaly to optimize feedback efficiency and model training effect. When a single point of sudden anomaly occurs, an independent sample is generated immediately, and the transmission priority is the highest. When the threshold is adaptively updated, the dynamic threshold is adjusted according to the new historical data, and additional metadata is generated to explain the reason and scope of the threshold adjustment. No specific sample is generated, and the transmission priority is the lowest.

[0115] Specifically, in S4, the improved PTPv2.1 protocol is used to achieve microsecond-level time synchronization between nodes across cities, build a unified spatiotemporal reference system, and provide a benchmark guarantee for the spatiotemporal correlation analysis of noise monitoring data.

[0116] As attached Figure 2 The multi-city noise automatic monitoring data analysis system based on time period classification shown includes a multi-city noise monitoring database and a monitoring data central processing module, and also includes:

[0117] Multi-source spatiotemporal data acquisition module: This module acquires noise monitoring data through a multi-city sensor node network. The noise monitoring data includes sound pressure signals acquired through dual-mode positioning and high-precision clock synchronization, millisecond-level timestamps synchronized with the sound pressure signals, and geographic location information.

[0118] A time segment feature encoding module performs a spatiotemporal feature extraction operation based on the noise monitoring data. The spatiotemporal feature extraction operation slices the sound pressure signal into time slots and uses an attention mechanism to generate a dynamic time segment fingerprint with time-sensitive weights and spatial correlation weights.

[0119] Dynamic evolution collaboration module: performs noise iterative analysis on the dynamic period fingerprint map, including feature weighted classification, feature decoupling migration, incremental learning and abnormal feedback;

[0120] Clock synchronization module: continuously performs clock alignment between nodes across cities and uses clock signals as basic timing constraints for feedback adjustment;

[0121] In this embodiment, the clock synchronization module continuously performs clock alignment between nodes across cities, including: generating and broadcasting Sync synchronization messages every 100ms through the master clock node pre-deployed in the multi-city noise monitoring database. The message structure adopts a simplified frame header design and uses a multicast address to achieve one-to-many efficient transmission. The frequency is automatically reduced to 200ms when the network is congested. If no response is received from the slave node for two consecutive times, retransmission is triggered;

[0122] Furthermore, the slave node records multiple sets of continuous time deviation data and uses a linear regression compensation algorithm to solve the current moment prediction deviation, and then adjusts the local clock to eliminate transmission delay jitter;

[0123] Furthermore, each node generates a time slot index based on a unified clock, maintaining a network-wide clock deviation of less than 1μs, providing a unified benchmark for time slot division;

[0124] The multi-city noise monitoring database includes all data texts of the multi-city noise automatic monitoring data analysis system based on time period classification, and collects information data output by each module in real time. The monitoring data central processing module is used for information data instructions output by each module in the central control system.

[0125] Among them, the multi-city noise monitoring database adopts a spatiotemporal database storage architecture to store a large amount of data related to the analysis of multi-city noise automatic monitoring data based on time period classification, including a cross-city feature decoupling model established based on adversarial discriminators and orthogonal constraints, historical versions of the LSTM time period prediction weight matrix, and dynamic time period fingerprint maps stored in the city-time period dual dimensions; when performing dynamic feature weighted classification, the monitoring data central processing module calls the common feature template data in the model library in real time, and combines the spatiotemporal weight matrix in the feature library to perform adversarial training gradient calculation, parameter importance evaluation and dual-channel back propagation operations.

[0126] Among them, the monitoring data central processing module is the core operation and control unit of the entire multi-city noise automatic monitoring data analysis system based on time period classification; it includes a multi-core heterogeneous computing architecture, which is connected to the distributed clock synchronization module, multi-source spatiotemporal calibration acquisition module and dynamic evolution collaboration module through a high-speed bus and a dedicated interface; by running or executing the convolutional neural network instruction set, incremental learning algorithm code set and spatiotemporal attention calculation instructions stored in the system operation database, and being able to call the cross-city feature template data stored therein, it performs spatiotemporal joint feature encoding, dynamic weight matrix generation, feature decoupling distillation calculation and other operations to ensure that the minute-level feature slicing of the time slot is strictly synchronized with the incremental learning process.

[0127] Secondly: The drawings of the embodiments disclosed in the present invention only involve structures related to the embodiments disclosed in the present invention. Other structures may refer to conventional designs. The same embodiment and different embodiments of the present invention may be combined with each other without conflict.

[0128] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A multi-city noise automatic monitoring data analysis method based on time period classification is characterized by: include: S1: Acquire noise monitoring data through a multi-city sensor node network. The noise monitoring data includes a sound pressure signal acquired through dual-mode positioning and high-precision clock synchronization, a millisecond-level timestamp synchronized with the sound pressure signal, and geographic location information. S2: performing a spatiotemporal feature extraction operation based on the noise monitoring data, wherein the spatiotemporal feature extraction operation slices the sound pressure signal into time slots and uses an attention mechanism to generate a dynamic time period fingerprint map containing time-sensitive weights and spatial correlation weights; Generate dynamic time period fingerprint, including: The power spectrum values ​​corresponding to the continuous PCM waveform data in the noise monitoring data are formed into a power spectrum matrix in frame order, and input into a time slicing convolutional network to obtain time domain features, frequency domain features and compressed feature vectors; Calculate the power spectrum value corresponding to the continuous PCM waveform data in the noise monitoring data , specifically expressed as: in, Expressed as The power spectrum value of the frame, Expressed as the number of sampling points corresponding to the Hamming window, Represents the index of the sampling point corresponding to the Hamming window, Expressed as The signal value of the sampling point, Expressed as the Hamming window function value, Expressed as a basis function of the discrete Fourier transform; Using time slot indexing , generating time slot embedding vectors through a trainable lookup table; Integrated attention mechanism, based on time slot embedding vector , construct the query vector , key vector and value vector ; Based on query vector and key vector , calculate the attention weight , specifically expressed as: in, Represented as a key vector Dimensions, Represented as query vector and key vector The matrix product of the transpose of ; Through attention weight Pair value vector Perform weighted summation to obtain the time period sensitive feature vector , specifically expressed as: Encapsulating the compressed feature vector, the time slot index, the time period sensitive feature vector, and the power spectrum matrix corresponding to each city to form a dynamic time period fingerprint map; S3: performing noise iterative analysis on the dynamic period fingerprint including feature weighted classification, feature decoupling migration, incremental learning and abnormal feedback; S4: Continuously perform clock alignment between nodes across cities and use clock signals as basic timing constraints for feedback adjustment.

2. The method for analyzing multi-city noise automatic monitoring data based on time period classification according to claim 1 is characterized by: The S1, in the multi-city sensor node network, automatically triggers a scene switching event by determining the spatial relationship between the coordinate point corresponding to the sensor node and the preset geographic fence, specifically including: Load the baseline noise signature library associated with the target geo-fence and send an update signal of the dynamic period fingerprint map to the monitoring data central processing module; Reconfigure the noise threshold based on the target geofence scenario type and target time period; Dynamically adjust the data upload frequency based on the scene sensitivity of the target geofence.

3. The multi-city noise automatic monitoring data analysis method based on time period classification according to claim 1 is characterized by: S3, performing feature weighted classification, specifically includes: By constructing an LSTM time period prediction unit and a noise pattern discriminator driven by a gradient reversal network, a dual adaptive mechanism is formed; Adversarial training is performed on the dynamic period fingerprint to generate the probability distribution of each noise type.

4. The method for analyzing multi-city noise automatic monitoring data based on time period classification according to claim 1 is characterized by: S3, performing feature decoupling migration, specifically includes: By constructing a feature decoupling distillation framework, the feature space of the multi-city noise analysis model is decomposed into a city-independent common subspace and a city-specific private subspace, realizing the cross-domain transfer of common features.

5. The method for analyzing multi-city noise automatic monitoring data based on time period classification according to claim 1 is characterized in that: S3, performing incremental learning, specifically includes: Adopting an elastic weight solidification algorithm to realize the dynamic evolution of the multi-city noise analysis model; If a periodic false alarm is detected, the common feature parameters are frozen and the specific feature parameters are dynamically adjusted, and incremental updates are achieved through a dual-channel back-propagation mechanism.

6. The method for analyzing multi-city noise automatic monitoring data based on time period classification according to claim 1 is characterized by: S3 executes abnormal feedback construction, specifically including: A spatiotemporal joint anomaly detection mechanism based on Mahalanobis distance; If the classification result deviates from the preset threshold of the historical period pattern, the manual review interface is automatically triggered and a training sample stream with corrected labels is generated.

7. A multi-city noise automatic monitoring data analysis system based on time period classification, comprising a multi-city noise monitoring database and a monitoring data central processing module, wherein the multi-city noise automatic monitoring data analysis method based on time period classification according to any one of claims 1 to 6 is used, characterized in that: Also includes: Multi-source spatiotemporal data acquisition module: This module acquires noise monitoring data through a multi-city sensor node network. The noise monitoring data includes sound pressure signals acquired through dual-mode positioning and high-precision clock synchronization, millisecond-level timestamps synchronized with the sound pressure signals, and geographic location information. A time segment feature encoding module performs a spatiotemporal feature extraction operation based on the noise monitoring data. The spatiotemporal feature extraction operation slices the sound pressure signal into time slots and uses an attention mechanism to generate a dynamic time segment fingerprint with time-sensitive weights and spatial correlation weights. Dynamic evolution collaboration module: performs noise iterative analysis on the dynamic period fingerprint map, including feature weighted classification, feature decoupling migration, incremental learning and abnormal feedback; Clock synchronization module: continuously performs clock alignment between nodes across cities and uses clock signals as basic timing constraints for feedback adjustment; The multi-city noise monitoring database includes all data texts of the multi-city noise automatic monitoring data analysis system based on time period classification, and collects information data output by each module in real time; the monitoring data central processing module is used for information data instructions output by each module in the central control system.

8. The multi-city noise automatic monitoring data analysis system based on time period classification according to claim 7 is characterized by: The clock synchronization module continuously performs clock alignment between nodes across cities, specifically including: The master clock node sends synchronization messages every 100ms; Use linear regression compensation algorithm to eliminate transmission delay jitter; Maintain network-wide clock deviation less than 1μs.

Images