Flying dust noise monitoring data intelligent analysis method based on deep learning

Abstract

The invention discloses a flying dust noise monitoring data intelligent analysis method based on deep learning, and the method comprises the steps: obtaining initial multi-source monitoring data, carrying out the noise reduction of flying dust data in the data through employing a wavelet threshold value, carrying out the noise reduction of non-environmental interference in the data through adaptive frequency band filtering, and carrying out the noise reduction of the non-environmental interference in the data; the method comprises the following steps: extracting multi-scale time sequence features by using a 1D-CNN (Convolutional Neural Network) to establish a flying dust branch, extracting long-range frequency spectrum dependence through a Transform encoder to establish a noise branch, and performing cross-modal feature interaction through an attention fusion mechanism to obtain a double-flow deep neural network model; inputting the noise reduction monitoring data into a model for identification, and outputting an event classification probability and a decision factor; and performing intelligent early warning according to an evaluation result. The recognition accuracy of complex environment events is effectively improved, and the recognition accuracy of construction dust raising events is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental monitoring, and particularly to an intelligent analysis method for dust and noise monitoring data based on deep learning. Background Art

[0002] Traditional dust and noise monitoring technologies mainly rely on single sensors and simple threshold judgments, which have significant limitations. A single sensor can only obtain single-type data and cannot comprehensively consider the influence of multiple factors such as meteorological conditions and environmental interference on the monitoring results, resulting in insufficient accuracy and comprehensiveness of the data. Traditional dust monitoring equipment is prone to misjudging non-pollution source particles lifted by wind as dust pollution, causing distortion of the monitoring data. Simple threshold judgments lack the ability to deeply analyze complex environmental characteristics and have low analysis accuracy. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and design an intelligent analysis method for dust and noise monitoring data based on deep learning.

[0004] Furthermore, in the above intelligent analysis method for dust and noise monitoring data based on deep learning, the intelligent analysis method for dust and noise monitoring data includes the following steps: Obtain initial multi-source monitoring data, use wavelet threshold to denoise the dust data in the data, and use adaptive band filtering to denoise the non-environmental interference in the data to obtain denoised monitoring data; Use a 1D-CNN dilated convolutional neural network to extract multi-scale temporal features to establish a dust branch, use a Transformer encoder to extract long-range spectral dependencies to establish a noise branch, and perform cross-modal feature interaction through an attention fusion mechanism to obtain a CNN-Transformer dual-stream deep neural network model; Input the denoised monitoring data into the CNN-Transformer dual-stream deep neural network model for recognition, and output the event classification probability and decision factor; perform environmental status evaluation according to the event classification probability and decision factor, and perform intelligent early warning according to the evaluation result.

[0005] Furthermore, in the above intelligent analysis method for dust and noise monitoring data based on deep learning, the step of obtaining initial multi-source monitoring data, using wavelet threshold to denoise the dust data in the data, and using adaptive band filtering to denoise the non-environmental interference in the data to obtain denoised monitoring data includes: Synchronously collect dust concentration data, noise spectrum data, and meteorological parameters through multi-source sensors to obtain multi-source monitoring data, Perform outlier detection on the original data of each sensor. For the moment data of a single sensor, if its value exceeds 3 times the standard deviation of the mean of the historical data of that sensor, it is judged as an outlier and the outlier is marked; For the marked outliers, use the adjacent time point data interpolation method to repair them to obtain processed multi-source monitoring data.

[0006] Furthermore, in the above intelligent analysis method for dust and noise monitoring data based on deep learning, when obtaining the initial multi-source monitoring data, use wavelet threshold to reduce the noise of the dust data in the data, and perform noise reduction on the non-environmental interference in the data through adaptive band filtering to obtain noise-reduced monitoring data, including: Obtain the processed multi-source monitoring data, perform 3-layer wavelet decomposition on the preprocessed dust concentration data through the db4 wavelet basis function to obtain detail coefficients and approximation coefficients at different frequency levels; Use the threshold to calculate the threshold of each layer of detail coefficients, process the detail coefficients through the soft threshold function, and perform wavelet reconstruction on the processed coefficients to obtain the noise-reduced dust concentration data.

[0007] Furthermore, in the above intelligent analysis method for dust and noise monitoring data based on deep learning, when obtaining the initial multi-source monitoring data, use wavelet threshold to reduce the noise of the dust data in the data, and perform noise reduction on the non-environmental interference in the data through adaptive band filtering to obtain noise-reduced monitoring data, including: Obtain the processed multi-source monitoring data, use FFT (Fast Fourier Transform) to convert the noise spectrum data in the initial multi-source monitoring data and the time-domain signal collected by the noise sensor into frequency-domain signals, and perform real-time spectrum analysis to obtain the frequency components of the noise to obtain the spectrum analysis result; Judge the frequency band range of environmental interference according to the spectrum analysis result. If the periodic noise generated by industrial equipment is judged to be in the 50 - 200 Hz frequency band, then set this frequency band as the filtered frequency band; Use an IIR (Infinite Impulse Response) filter to perform adaptive band filtering on the filtered frequency band, use a band-stop filter to set the passband and stopband parameters of the filter according to the determined frequency band range, and perform noise reduction processing on the non-environmental interference to obtain noise-reduced monitoring data.

[0008] Furthermore, in the above intelligent analysis method for dust and noise monitoring data based on deep learning, when using a 1D-CNN (1D Convolutional Neural Network) dilated convolutional neural network to extract multi-scale temporal features to establish a dust branch, using a Transformer encoder to extract long-range spectral dependencies to establish a noise branch, and performing cross-modal feature interaction through an attention fusion mechanism to obtain a CNN-Transformer dual-stream deep neural network model, including: The 1D-CNN dilated convolutional neural network at least includes an input layer, a dilated convolutional layer, a batch normalization layer, and a pooling layer; The input layer is used to receive the dust data and meteorological parameters in the noise-reduced monitoring data, and splice the dust data and meteorological parameters into one-dimensional time series data; Three dilated convolutional layers are set. The dilation rate of the first layer is 1, the convolutional kernel size is 3, and the number of output channels is 64; the dilation rate of the second layer is 2, the convolutional kernel size is 3, and the number of output channels is 128; the dilation rate of the third layer is 4, the convolutional kernel size is 3, and the number of output channels is 256. Multi-scale time series features are obtained through different dilation rates; A batch normalization layer is added after each dilated convolutional layer, and an average pooling layer is added after the last dilated convolutional layer, and the pooling window size is 2.

[0009] Furthermore, in the above intelligent analysis method for dust and noise monitoring data based on deep learning, the multi-scale time series features are extracted by using the 1D-CNN dilated convolutional neural network to establish a dust branch, the long-range spectral dependence is extracted by using a Transformer encoder to establish a noise branch, and cross-modal feature interaction is performed through an attention fusion mechanism to obtain a CNN-Transformer dual-stream deep neural network model, including: The Transformer encoder at least includes an input layer, a position encoding layer, a Transformer encoder layer, and a normalization layer; The input layer is used to take the noise-reduced noise spectrum data as the input, and add position information to the noise spectrum data based on sine-cosine position encoding; Six Transformer encoders are set. Each encoder contains a multi-head self-attention mechanism and a feed-forward neural network. The multi-head self-attention mechanism enables the model to focus on different parts in different representation subspaces, capture long-range spectral dependence relationships, and layer normalization layers are added before and after the multi-head self-attention mechanism and the feed-forward neural network.

[0010] Furthermore, in the above intelligent analysis method for dust and noise monitoring data based on deep learning, the noise-reduced monitoring data is input into the CNN-Transformer dual-stream deep neural network model for recognition, and the event classification probability and decision factor are output; the environmental state is evaluated according to the event classification probability and decision factor, and intelligent early warning is performed according to the evaluation result, including: The event classification probability at least includes construction dust, traffic noise, industrial noise, and normal environment, and the decision factor is used to measure the credibility of the model prediction; The classification probability threshold for each category of events is set to 0.7. When the classification probability of a category of events is greater than or equal to 0.7, it is considered that the possibility of the event occurring is relatively high; The decision factor ranges from [0, 1]. Set the decision factor threshold to 0.6. When the decision factor is greater than or equal to 0.6, the model prediction result is considered credible.

[0011] Furthermore, in a system for implementing an intelligent analysis method for dust and noise monitoring data based on deep learning, the system includes the following sub-modules: A monitoring data acquisition module, which is used to acquire initial multi-source monitoring data, denoise the dust data in the data using wavelet threshold, and denoise the non-environmental interference in the data through adaptive band filtering to obtain denoised monitoring data; A network model establishment module, which is used to extract multi-scale time series features using a 1D-CNN dilated convolutional neural network to establish a dust branch, extract long-range spectral dependencies using a Transformer encoder to establish a noise branch, and perform cross-modal feature interaction through an attention fusion mechanism to obtain a CNN-Transformer dual-stream deep neural network model; A monitoring intelligent analysis module, which is used to input the denoised monitoring data into the CNN-Transformer dual-stream deep neural network model for identification, and output the event classification probability and the decision factor; perform environmental status evaluation according to the event classification probability and the decision factor, and perform intelligent warning according to the evaluation result.

[0012] Furthermore, in a system for implementing an intelligent analysis method for dust and noise monitoring data based on deep learning, the system includes the following sub-modules: An acquisition sub-module, which is used to synchronously collect dust concentration data, noise spectrum data, and meteorological parameters through multi-source sensors to obtain multi-source monitoring data, A judgment sub-module, which is used to detect outliers in the original data of each sensor, judge the moment data of a single sensor, and if its value exceeds 3 times the standard deviation of the mean of the historical data of the sensor, it is judged as an outlier and the outlier is marked; A repair sub-module, which is used to repair the marked outliers using the adjacent time point data interpolation method to obtain processed multi-source monitoring data.

[0013] Furthermore, in a system for implementing an intelligent analysis method for dust and noise monitoring data based on deep learning, the system includes the following sub-modules: A decomposition sub-module, which is used to obtain processed multi-source monitoring data, perform 3-layer wavelet decomposition on the preprocessed dust concentration data using the db4 wavelet basis function to obtain detail coefficients and approximation coefficients at different frequency levels; A reconstruction sub-module, which is used to calculate the threshold of each layer of detail coefficients using a threshold, process the detail coefficients through a soft threshold function, and perform wavelet reconstruction on the processed coefficients to obtain denoised dust concentration data.

[0014] The beneficial effects are as follows: by obtaining the initial multi-source monitoring data, using wavelet threshold to denoise the dust data in the data, and using adaptive band filtering to denoise the non-environmental interference in the data, the denoised monitoring data is obtained; using the 1D-CNN dilated convolutional neural network to extract multi-scale time series features to establish a dust branch, using the Transformer encoder to extract long-range spectral dependencies to establish a noise branch, and performing cross-modal feature interaction through the attention fusion mechanism to obtain the CNN-Transformer dual-stream deep neural network model; inputting the denoised monitoring data into the CNN-Transformer dual-stream deep neural network model for recognition, and outputting the event classification probability and decision factor; evaluating the environmental state according to the event classification probability and decision factor, and performing intelligent early warning according to the evaluation result. 1. Compared with the traditional single-sensor scheme, the data dimension is increased, the pollution source characteristics and environmental impact factors are completely covered, providing full-dimensional data support for subsequent analysis. 2. The recognition accuracy of complex environmental events is effectively improved, the recognition accuracy of construction dust events is improved, and the comprehensive performance is improved compared with the traditional scheme in complex scenarios such as construction sites and industrial parks. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention.

[0016] Figure 1 Schematic diagram of the first embodiment of an intelligent analysis method for dust and noise monitoring data based on deep learning in an embodiment of the present invention; Figure 2 Schematic diagram of the second embodiment of an intelligent analysis method for dust and noise monitoring data based on deep learning in an embodiment of the present invention; Figure 3 Schematic diagram of the third embodiment of an intelligent analysis method for dust and noise monitoring data based on deep learning in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0018] Those skilled in the art can understand that, unless specifically stated, the singular forms "a", "an", "the" and "said" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present invention means the presence of features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or their groups.

[0019] The present invention will be specifically described below with reference to the accompanying drawings, as Figure 1 shown, an intelligent analysis method for dust and noise monitoring data based on deep learning, the intelligent analysis method for dust and noise monitoring data includes the following steps: Step 101, obtain initial multi-source monitoring data, use wavelet threshold to denoise the dust data in the data, and perform denoising on non-environmental interference in the data through adaptive band filtering to obtain denoised monitoring data; Specifically, in this embodiment, dust concentration data, noise spectrum data and meteorological parameters are synchronously collected by multi-source sensors to obtain multi-source monitoring data. Perform outlier detection on the original data of each sensor, judge the moment data of a single sensor, if its value exceeds 3 times the standard deviation of the mean of the historical data of the sensor, it is judged as an outlier, and the outlier is marked; For the marked outliers, use the adjacent time point data interpolation method to repair them to obtain processed multi-source monitoring data.

[0020] Obtain the processed multi-source monitoring data, perform 3-layer wavelet decomposition on the preprocessed dust concentration data through the db4 wavelet basis function to obtain detail coefficients and approximation coefficients at different frequency levels; Use the threshold to calculate the threshold of each layer of detail coefficients, process the detail coefficients through the soft threshold function, and perform wavelet reconstruction on the processed coefficients to obtain the denoised dust concentration data.

[0021] Obtain the processed multi-source monitoring data, use FFT fast Fourier transform to convert the noise spectrum data and the time-domain signal collected by the noise sensor in the initial multi-source monitoring data into frequency-domain signals, and perform real-time spectrum analysis to obtain the frequency components of the noise to obtain the spectrum analysis result; Judge the frequency band range of environmental interference according to the spectrum analysis result. If the periodic noise generated by industrial equipment is judged to be in the 50-200 Hz frequency band, then set this frequency band as the filtered band; Use an IIR infinite impulse response filter to perform adaptive band filtering on the filtered band, use a band-stop filter to set the passband and stopband parameters of the filter according to the determined frequency band range, and perform denoising processing on non-environmental interference to obtain denoised monitoring data.

[0022] Specifically, I. Multi-source Sensor Deployment Sensor Types: High-precision dust sensors (laser scattering dust sensors, measurement range 0 - 10 mg / m³, accuracy ±10%), noise sensors (capacitive microphone noise sensors, measurement range 30 - 130 dB, accuracy ±1.5 dB), and meteorological sensors (including temperature and humidity sensors, wind speed and direction sensors, and barometric pressure sensors) are adopted. The temperature and humidity sensor is of the SHT30 type, with a temperature measurement range of -40 - 125 °C, accuracy ±0.3 °C, humidity range 0 - 100%RH, accuracy ±2%RH; the wind speed and direction sensor is ultrasonic, with a wind speed measurement range of 0 - 60 m / s, accuracy ±0.3 m / s, wind direction measurement range of 0 - 360°, accuracy ±3°; the barometric pressure sensor is of the MS5611 type, with a measurement range of 10 - 1200 hPa, accuracy ±0.12 hPa.

[0023] Sensor Layout: Sensor nodes are evenly deployed within the monitoring area. Each node contains the above-mentioned various sensors, and the spacing is set to 50 - 100 meters according to the monitoring accuracy requirements. For key monitoring areas (construction sites, beside main traffic roads), the sensor deployment is appropriately densified.

[0024] II. Synchronous Acquisition Implementation A high-precision GPS timing module (u-blox NEO-M8N) is used to provide a unified time reference for each sensor, and the time synchronization accuracy can reach the nanosecond level. Each sensor performs data acquisition with a sampling period of 1 second, and the acquisition moment is synchronously triggered by the GPS timing module to ensure that multi-source data is strictly aligned on the time axis.

[0025] III. Preprocessing Process Data Format Conversion: Analog Signal Processing: The analog signals output by the dust sensor and some meteorological sensors are converted into digital signals through a 24-bit analog-to-digital converter (ADC), and the conversion accuracy meets the resolution requirements of the monitoring data.

[0026] Unified Data Format: The digital signals output by all sensors are converted into the JSON format specified by the system. The data structure includes fields such as timestamp, sensor type, monitoring value, and unit. For example: Outlier Detection and Processing: The Z-score method is used to detect outliers. For the monitoring data sequence of each sensor, the mean and standard deviation are calculated.

[0027] Outlier Processing: If the outlier is an isolated point, interpolation processing is performed using the average value of the two adjacent normal data points; if multiple consecutive data points are outliers, they are marked as data missing, and a sliding window mean is used for filling during subsequent model input.

[0028] Noise reduction processing I. Wavelet threshold denoising of dust emission data Wavelet basis selection and decomposition level: The db4 wavelet basis is selected. This wavelet basis has good regularity and vanishing moments and is suitable for the noise reduction processing of dust emission data. According to the sampling frequency (1 Hz) and noise characteristics of the dust emission data, the decomposition level is determined to be 4 layers.

[0029] Wavelet decomposition and threshold processing: Perform 4-layer wavelet decomposition on the preprocessed dust emission data to obtain the low-frequency coefficients and high-frequency coefficients of each layer.

[0030] Use the soft threshold function to process the high-frequency coefficients. The threshold calculation adopts VisuShrink. Wavelet reconstruction: Perform wavelet reconstruction on the processed coefficients of each layer to obtain the dust emission data after noise reduction.

[0031] II. Adaptive band filtering denoising for non-environmental interference Noise spectrum analysis: Perform real-time spectrum analysis on the noise spectrum data in the initial multi-source monitoring data (the time-domain signal collected by the noise sensor is converted into a frequency-domain signal using the fast Fourier transform (FFT), with the number of transformation points being 1024 and the frequency resolution being 0.9766 Hz) to obtain the main frequency components of the noise.

[0032] Determination of the adaptive band: According to the spectrum analysis results, dynamically determine the main frequency band range of environmental interference. For example, if it is detected that the periodic noise generated by industrial equipment is mainly concentrated in the 50 - 200 Hz frequency band, then this frequency band is set as the band to be filtered.

[0033] Filtering processing: Use an infinite impulse response (IIR) filter for adaptive band filtering. The filter type is selected as a band-stop filter, and the passband and stopband parameters of the filter are set according to the determined frequency band range to achieve noise reduction processing for non-environmental interference.

[0034] Step 102: Use a 1D-CNN dilated convolutional neural network to extract multi-scale temporal features to establish a dust emission branch, use a Transformer encoder to extract long-range spectral dependencies to establish a noise branch, and perform cross-modal feature interaction through an attention fusion mechanism to obtain a CNN-Transformer dual-stream deep neural network model; Specifically, in this embodiment, the 1D-CNN dilated convolutional neural network at least includes an input layer, a dilated convolutional layer, a batch normalization layer, and a pooling layer; Use the input layer to receive the dust emission data and meteorological parameters in the noise-reduced monitoring data, and concatenate the dust emission data and meteorological parameters into one-dimensional time series data; Set 3 dilated convolutional layers. The dilation rate of the first layer is 1, the convolutional kernel size is 3, and the number of output channels is 64; the dilation rate of the second layer is 2, the convolutional kernel size is 3, and the number of output channels is 128; the dilation rate of the third layer is 4, the convolutional kernel size is 3, and the number of output channels is 256. Obtain multi-scale temporal features through different dilation rates; Add a batch normalization layer after each dilated convolutional layer, and add an average pooling layer after the last dilated convolutional layer. The pooling window size is 2.

[0035] The Transformer encoder includes at least an input layer, a positional encoding layer, a Transformer encoder layer, and a normalization layer; Use the input layer to take the denoised noise spectrum data as input, and add positional information to the noise spectrum data based on sine-cosine positional encoding; Set 6 Transformer encoders. Each encoder contains a multi-head self-attention mechanism and a feed-forward neural network. The multi-head self-attention mechanism enables the model to focus on different parts in different representation subspaces, capture long-range spectral dependencies, and add layer normalization layers before and after the multi-head self-attention mechanism and the feed-forward neural network.

[0036] Specifically, Dust branch: 1D-CNN dilated convolutional neural network Network structure: Input layer: Receive denoised dust data and meteorological parameters (temperature, humidity, wind speed, wind direction, air pressure), and concatenate them into one-dimensional time series data.

[0037] Dilated convolutional layer: Set 3 dilated convolutional layers. The dilation rate of the first layer is 1, the convolutional kernel size is 3, and the number of output channels is 64; the dilation rate of the second layer is 2, the convolutional kernel size is 3, and the number of output channels is 128; the dilation rate of the third layer is 4, the convolutional kernel size is 3, and the number of output channels is 256. Obtain multi-scale temporal features through different dilation rates.

[0038] Batch normalization layer (BatchNormalization): Add a batch normalization layer after each dilated convolutional layer to accelerate the network convergence speed and reduce overfitting.

[0039] Activation function: Adopt the ReLU activation function to increase the non-linear expression ability of the network.

[0040] Pooling layer: Add an average pooling layer after the last dilated convolutional layer. The pooling window size is 2 to reduce the feature dimension.

[0041] Noise branch: Transformer encoder Input layer: The denoised noise spectrum data (frequency-domain signal after FFT conversion, with the number of frequency points set to 512) is used as the input, and the input dimension is \(T\times512\).

[0042] Position encoding: Since the Transformer model does not have temporal sequence information, sine-cosine position encoding is used to add position information to the noise spectrum data. The position encoding formula is as follows: Transformer encoder layer: Six Transformer encoders are set, and each encoder contains a multi-head self-attention mechanism (the number of heads is set to 8) and a feed-forward neural network. The multi-head self-attention mechanism allows the model to focus on different parts in different representation subspaces and capture long-range spectral dependencies.

[0043] Layer Normalization: Layer normalization layers are added before and after the multi-head self-attention mechanism and the feed-forward neural network to stabilize network training.

[0044] Attention fusion mechanism Feature extraction: The feature vector output by the dust branch is, with the dimension of ; the feature vector output by the noise branch is, with the dimension of. The dimensions of the two are unified to (set to 1024) through a fully connected layer.

[0045] Cross-modal attention calculation: A bidirectional attention mechanism is used to calculate the attention weights of the dust features to the noise features and the attention weights of the noise features to the dust features Feature fusion: The features of the two branches are weighted and fused through the attention weights to obtain the fused features.

[0046] Model recognition and output I. Data input processing The denoised monitoring data (including dust data, noise spectrum data, and meteorological parameters) is normalized. The normalization method uses min-max normalization to scale the data to the [0,1] interval. The data is divided according to a time window (length is 100, sliding step is 50) to form input samples.

[0047] II. Model recognition process The normalized input samples are input into the CNN-Transformer dual-stream deep neural network model. The dust emission branch extracts multi-scale temporal features through a dilated convolutional neural network, and the noise branch extracts long-range spectral dependence features through a Transformer encoder. The two perform cross-modal feature interaction through an attention fusion mechanism. After being processed by a fully connected layer and a softmax function, the event classification probability is output. The classification categories include construction dust, traffic noise, industrial noise, normal environment, etc., a total of 4 categories. At the same time, a decision factor calculation module (implemented by a fully connected layer) outputs a decision factor, which is used to measure the credibility of the model prediction.

[0048] Step 103: Input the noise-reduced monitoring data into the CNN-Transformer dual-stream deep neural network model for recognition, and output the event classification probability and the decision factor; perform environmental status evaluation according to the event classification probability and the decision factor, and conduct intelligent early warning according to the evaluation result.

[0049] Specifically, in this embodiment, the event classification probability at least includes construction dust, traffic noise, industrial noise, and normal environment, and the decision factor is used to measure the credibility of the model prediction; Set the classification probability threshold for each category of event to 0.7. When the classification probability of the category event is greater than or equal to 0.7, it is considered that the possibility of the event occurring is relatively high; The range of the decision factor is [0,1]. Set the decision factor threshold to 0.6. When the decision factor is greater than or equal to 0.6, the model prediction result is considered credible.

[0050] Specifically, Processing of the dust emission branch: The dust emission concentration sequence is concatenated with 5 meteorological parameters such as temperature, humidity, wind speed and direction, and air pressure to form a 100×6 two-dimensional time series matrix, and features are extracted layer by layer through 3 dilated convolutional layers. The first layer of convolution captures short-term fluctuations of 1-3 seconds (the impact of instantaneous wind speed changes on dust emissions), the second layer extends the perception range to 5-7 seconds through a convolutional kernel with a dilation rate of 2, and the third layer captures medium- and long-term trends of 30-40 seconds with a dilation rate of 4 (the cumulative effect of humidity changes on particulate matter sedimentation), and finally outputs a time series feature vector containing multi-scale environmental associations.

[0051] Noise Branch Processing: The noise time-domain signal per second is converted into 512-point frequency-domain features, forming a 100×512 spectral matrix. After injecting temporal sequence information through positional encoding, global dependency modeling is performed through 6 layers of Transformer encoders. Each layer of the encoder simultaneously calculates self-attention in 8 different subspaces to capture the energy distribution correlation between high-frequency noise (mechanical vibration) and low-frequency noise (traffic roar) in the long term, and outputs a spectral feature vector containing cross-band and cross-time-point dependency relationships. The feature vectors output by the two branches are fused through a bidirectional attention mechanism: first, calculate the attention degree of the dust feature to the noise feature (whether there is an increase in specific frequency band noise during high-concentration dust periods), then calculate the contribution degree of the noise feature to the dust feature in the reverse direction, and dynamically adjust the weights of the two and splice them into a comprehensive feature vector. After being mapped to a 4-classification space through a fully connected layer, the occurrence probabilities of 4 types of events, namely construction dust, traffic noise, industrial noise, and normal environment, are output through the Softmax function (the construction dust probability of 0.82 indicates that the current scene is 82% likely to be a construction dust event). At the same time, a fully connected layer is separately set to evaluate the credibility of the comprehensive feature, and a decision factor in the range of 0-1 is output (the higher the value, the more certain the model prediction), which is used to filter out fuzzy judgments with low confidence.

[0052] I. Environmental Status Evaluation Criteria Double criteria are adopted to ensure the reliability of the evaluation: Event Classification Probability: Trigger thresholds of 0.7 are set for each of the 4 types of events. When the probability of a certain type of event ≥0.7, it indicates that the model matches based on historical data and judges that the likelihood of this event occurring is significantly higher than other categories (the traffic noise probability of 0.75 means that the matching degree of the current noise feature with historical traffic noise samples exceeds 75%).

[0053] Decision Factor: The value range is 0-1, which is obtained by optimizing the prediction entropy value during model training. When the decision factor ≥0.6, it indicates that the model clearly recognizes the features of the current input sample and the prediction result is credible (the decision factor of 0.8 means that the uncertainty of the model's judgment for this time is low). Only events that simultaneously meet the probability threshold and the decision factor threshold will be included in the final evaluation result to avoid false alarms caused by data anomalies or fuzzy model judgments.

[0054] II. Intelligent Early Warning Mechanism Single-event instant warning: When the output result of a certain time window meets the double-threshold condition, the system immediately triggers the corresponding event warning. For example, if the probability of construction dust in a certain window is 0.78, the decision factor is 0.65, and the monitoring location is around a construction site, it is determined as a construction dust event, and warning information is generated immediately. The warning information includes the event type ("abnormal industrial noise"), the occurrence time (accurate to the second level), the monitoring node location (through the built-in GPS of the sensor or the deployed coordinate positioning), and the preliminary severity level (classified as "mild" and "moderate" based on the probability value. For example, the probability of 0.7 - 0.8 is mild, and 0.8 - 1 is moderate).

[0055] Multi-window trend analysis and warning upgrade: Establish a dynamic verification mechanism for 3 consecutive time windows. If the same monitoring node triggers the same type of event warning (traffic noise) within 150 seconds continuously (3 sliding windows of 50 seconds each), and the decision factor of each window is ≥ 0.6, it is determined as a persistent pollution event, and the warning level is upgraded to "severe". The first window triggers a mild warning for traffic noise, and the subsequent two windows continue to trigger. The system automatically upgrades the warning to severe and marks that the event has lasted for 5 minutes, indicating that there may be long-term pollution sources such as traffic congestion or equipment abnormalities.

[0056] Diversified warning methods: Adopt a hierarchical notification strategy according to the warning level: Mild warning: Pop-up reminder through the monitoring platform and send a text message (including the event type, time, and location) to the on-site management personnel for timely investigation.

[0057] Severe warning: In addition to the above methods, activate the on-site sound and light alarm device (flashing red light + buzzer reminder at the construction site), and at the same time send an email warning to the environmental protection department (attached with the event duration and characteristic trend chart) to trigger remote supervision intervention. All warning information is synchronously stored in the cloud database, supporting historical event query and spatio-temporal distribution analysis (viewing the concentrated occurrence period of construction dust warnings in a certain area within a week through the GIS map), providing data support for environmental management decisions.

[0058] Its beneficial effects are as follows: 1. Compared with the traditional single-sensor solution, the data dimension is enhanced, comprehensively covering the characteristics of pollution sources and environmental impact factors, providing full-dimensional data support for subsequent analysis; 2. Effectively improving the recognition accuracy of complex environmental events, the recognition accuracy of construction dust events is improved, and the comprehensive performance is improved compared with the traditional solution in complex scenarios such as construction sites and industrial parks.

[0059] Please refer to Figure 2, in an intelligent analysis method for dust and noise monitoring data based on deep learning, obtaining initial multi-source monitoring data, using wavelet threshold to reduce noise in the dust data, and using adaptive band filtering to reduce non-environmental interference in the data, the steps for obtaining the noise-reduced monitoring data are as follows: Step 201: Synchronously collect dust concentration data, noise spectrum data, and meteorological parameters through multi-source sensors to obtain multi-source monitoring data. Step 202: Perform outlier detection on the original data of each sensor. For the moment data of a single sensor, if its value exceeds 3 times the standard deviation of the historical data mean of the sensor, it is determined as an outlier and marked. Step 203: For the marked outliers, use the adjacent time point data interpolation method for repair to obtain the processed multi-source monitoring data.

[0060] Please refer to Figure 3 , in an intelligent analysis method for dust and noise monitoring data based on deep learning, obtaining initial multi-source monitoring data, using wavelet threshold to reduce noise in the dust data, and using adaptive band filtering to reduce non-environmental interference in the data, the steps for obtaining the noise-reduced monitoring data are as follows: Step 301: Obtain the processed multi-source monitoring data, use the FFT (Fast Fourier Transform) to convert the noise spectrum data and the time-domain signal collected by the noise sensor in the initial multi-source monitoring data into frequency-domain signals, and perform real-time spectrum analysis to obtain the frequency components of the noise, resulting in a spectrum analysis result. Step 302: Determine the frequency band range of environmental interference according to the spectrum analysis result. If the periodic noise generated by industrial equipment is determined to be in the 50 - 200 Hz frequency band, set this frequency band as the filter-out band. Step 303: Use an IIR (Infinite Impulse Response) filter to perform adaptive band filtering on the filter-out band, use a band-stop filter to set the passband and stopband parameters of the filter according to the determined frequency band range, and perform noise reduction processing on non-environmental interference to obtain the noise-reduced monitoring data.

[0061] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent analysis method for dust and noise monitoring data based on deep learning, characterized in that, The intelligent analysis method for dust and noise monitoring data includes the following steps: Obtain initial multi-source monitoring data, use wavelet threshold to reduce noise in the dust data, and perform noise reduction on non-environmental interference in the data through adaptive band filtering to obtain noise-reduced monitoring data; Use a 1D-CNN dilated convolutional neural network to extract multi-scale temporal features to establish a dust branch, use a Transformer encoder to extract long-range spectral dependencies to establish a noise branch, and perform cross-modal feature interaction through an attention fusion mechanism to obtain a CNN-Transformer dual-stream deep neural network model; Input the noise-reduced monitoring data into the CNN-Transformer dual-stream deep neural network model for identification, and output the event classification probability and decision factor; perform environmental status evaluation based on the event classification probability and decision factor, and perform intelligent early warning based on the evaluation result.

2. The intelligent analysis method for dust and noise monitoring data based on deep learning according to claim 1, characterized in that The step of obtaining initial multi-source monitoring data, using wavelet threshold to reduce noise in the dust data, and performing noise reduction on non-environmental interference in the data through adaptive band filtering to obtain noise-reduced monitoring data includes: Synchronously collect dust concentration data, noise spectrum data, and meteorological parameters through multi-source sensors to obtain multi-source monitoring data, Perform outlier detection on the original data of each sensor, judge the moment data of a single sensor, if its value exceeds 3 times the standard deviation of the mean of the historical data of this sensor, then judge it as an outlier and mark the outlier; For the marked outliers, use the adjacent time point data interpolation method for repair to obtain processed multi-source monitoring data.

3. The intelligent analysis method for dust and noise monitoring data based on deep learning according to claim 2, wherein The step of obtaining initial multi-source monitoring data, using wavelet threshold to reduce noise in the dust data, and performing noise reduction on non-environmental interference in the data through adaptive band filtering to obtain noise-reduced monitoring data includes: Obtain processed multi-source monitoring data, perform 3-layer wavelet decomposition on the preprocessed dust concentration data through the db4 wavelet basis function to obtain detail coefficients and approximation coefficients at different frequency levels; Use the threshold to calculate the threshold of each layer of detail coefficients, process the detail coefficients through the soft threshold function, and perform wavelet reconstruction on the processed coefficients to obtain noise-reduced dust concentration data.

4. The intelligent analysis method for dust and noise monitoring data based on deep learning according to claim 2, wherein, The step of obtaining initial multi-source monitoring data, using wavelet threshold to reduce noise in the dust data, and performing noise reduction on non-environmental interference in the data through adaptive band filtering to obtain noise-reduced monitoring data includes: Obtain processed multi-source monitoring data, use FFT (Fast Fourier Transform) to convert the noise spectrum data and the time-domain signal collected by the noise sensor in the initial multi-source monitoring data into frequency-domain signals, and perform real-time spectrum analysis to obtain the frequency components of the noise to obtain the spectrum analysis result; Judge the frequency band range of environmental interference according to the spectrum analysis result. If the periodic noise generated by industrial equipment is judged to be in the 50-200Hz frequency band, then set this frequency band as the filtered frequency band; Adaptive band filtering is performed on the filtered band using an IIR (Infinite Impulse Response) filter. A band-stop filter is used to set the passband and stopband parameters of the filter according to the determined frequency band range, and noise reduction processing is performed on non-environmental interference to obtain noise-reduced monitoring data.

5. The intelligent analysis method for dust and noise monitoring data based on deep learning according to claim 1, characterized in that, Using a 1D-CNN (One-Dimensional Convolutional Neural Network) dilated convolutional neural network to extract multi-scale temporal features to establish a dust branch, using a Transformer encoder to extract long-range spectral dependencies to establish a noise branch, and performing cross-modal feature interaction through an attention fusion mechanism to obtain a CNN-Transformer two-stream deep neural network model, including: The 1D-CNN dilated convolutional neural network at least includes an input layer, a dilated convolutional layer, a batch normalization layer, and a pooling layer; The input layer is used to receive the dust data and meteorological parameters in the noise-reduced monitoring data, and the dust data and meteorological parameters are concatenated into one-dimensional time series data; Three dilated convolutional layers are set. The dilation rate of the first layer is 1, the convolutional kernel size is 3, and the number of output channels is 64; the dilation rate of the second layer is 2, the convolutional kernel size is 3, and the number of output channels is 128; the dilation rate of the third layer is 4, the convolutional kernel size is 3, and the number of output channels is 256. Multi-scale temporal features are obtained through different dilation rates; A batch normalization layer is added after each dilated convolutional layer, and an average pooling layer with a pooling window size of 2 is added after the last dilated convolutional layer.

6. The intelligent analysis method for dust and noise monitoring data based on deep learning according to claim 1, characterized in that, Using a 1D-CNN dilated convolutional neural network to extract multi-scale temporal features to establish a dust branch, using a Transformer encoder to extract long-range spectral dependencies to establish a noise branch, and performing cross-modal feature interaction through an attention fusion mechanism to obtain a CNN-Transformer two-stream deep neural network model, including: The Transformer encoder at least includes an input layer, a positional encoding layer, a Transformer encoder layer, and a normalization layer; The input layer is used to take the noise-reduced noise spectral data as input, and positional information is added to the noise spectral data based on sine-cosine positional encoding; Six Transformer encoders are set. Each encoder contains a multi-head self-attention mechanism and a feed-forward neural network. The multi-head self-attention mechanism enables the model to focus on different parts in different representation subspaces to capture long-range spectral dependency relationships, and layer normalization layers are added before and after the multi-head self-attention mechanism and the feed-forward neural network.

7. The intelligent analysis method for dust and noise monitoring data based on deep learning according to claim 1, characterized in that Inputting the noise-reduced monitoring data into the CNN-Transformer two-stream deep neural network model for identification, and outputting event classification probabilities and decision factors; performing environmental state evaluation according to the event classification probabilities and decision factors, and performing intelligent early warning according to the evaluation results, including: The event classification probabilities at least include construction dust, traffic noise, industrial noise, and normal environment, and the decision factor is used to measure the credibility of the model prediction; The classification probability threshold for each type of event is set to 0.

7. When the classification probability of a type of event is greater than or equal to 0.7, it is considered that the possibility of the event occurring is relatively high; The decision factor ranges from [0, 1]. Set the decision factor threshold to 0.

6. When the decision factor is greater than or equal to 0.6, the model prediction result is considered credible.

8. A system for implementing the intelligent analysis method of dust and noise monitoring data based on deep learning as described in claim 1, characterized in that, The system includes the following modules: The monitoring data acquisition module is used to acquire the initial multi-source monitoring data, denoise the dust data in the data using wavelet threshold, and denoise the non-environmental interference in the data through adaptive band filtering to obtain the denoised monitoring data; The network model establishment module is used to extract multi-scale time-series features using a 1D-CNN dilated convolutional neural network to establish a dust branch, extract long-range spectral dependencies using a Transformer encoder to establish a noise branch, and perform cross-modal feature interaction through an attention fusion mechanism to obtain a CNN-Transformer dual-stream deep neural network model; The monitoring intelligent analysis module is used to input the denoised monitoring data into the CNN-Transformer dual-stream deep neural network model for recognition, and output the event classification probability and decision factor; evaluate the environmental state according to the event classification probability and decision factor, and perform intelligent warning according to the evaluation result.

9. A system for implementing the intelligent analysis method of dust and noise monitoring data based on deep learning as described in claim 1, characterized in that, The system includes the following sub-modules: The acquisition sub-module is used to synchronously collect dust concentration data, noise spectrum data, and meteorological parameters through multi-source sensors to obtain multi-source monitoring data. The judgment sub-module is used to detect outliers in the original data of each sensor, judge the moment data of a single sensor. If its value exceeds 3 times the standard deviation of the mean of the historical data of this sensor, it is judged as an outlier and the outlier is marked; The repair sub-module is used to repair the marked outliers using the adjacent time point data interpolation method to obtain the processed multi-source monitoring data.

10. A system for implementing the intelligent analysis method of dust and noise monitoring data based on deep learning as described in claim 1, characterized in that, The system includes the following sub-modules: The decomposition sub-module is used to obtain the processed multi-source monitoring data, perform 3-layer wavelet decomposition on the preprocessed dust concentration data using the db4 wavelet basis function to obtain the detail coefficients and approximation coefficients at different frequency levels; The reconstruction sub-module is used to calculate the threshold of each layer of detail coefficients using the threshold, process the detail coefficients through the soft threshold function, and perform wavelet reconstruction on the processed coefficients to obtain the denoised dust concentration data.

Images