A real-time processing and intelligent analysis system for ground-air time-frequency electromagnetic data

Abstract

The present invention discloses a real-time processing and intelligent analysis system for ground-to-air time-frequency electromagnetic data, which relates to the field of information processing technology. It solves the technical problems of difficulty in decomposing non-stationary broadband signals into multiple sub-bands, inconvenience in performing more detailed time-frequency analysis on electromagnetic signals, and difficulty in effectively identifying data points that are significantly different from normal patterns. By analyzing signals at different time and frequency resolutions through wavelet packet transform, it is convenient to process the decomposition of non-stationary broadband signals into multiple sub-bands. Combining wavelet packet transform with STFT, the characteristics of each frequency band can be effectively extracted, providing a high-quality information basis for subsequent data analysis. The time-frequency analysis results can be displayed in a visual manner such as a time-frequency graph, making the complex signal characteristics more intuitive and facilitating subsequent data interpretation and decision support. Similar data points are grouped together through a clustering algorithm, thereby achieving unsupervised classification, reducing manual intervention, and improving classification efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of information processing technology, and in particular is a real-time processing and intelligent analysis system for ground-space time-frequency electromagnetic data. Background Art

[0002] With the rapid development of technologies such as wireless communications, radar detection, the Internet of Things (IoT), and intelligent transportation, the generation and application of ground-space time-frequency electromagnetic data is increasing. These electromagnetic signals cover a wide frequency range and are involved in multiple application areas. Traditional data processing methods often fail to meet the requirements of real-time performance and accuracy, necessitating an efficient system to process and analyze this complex electromagnetic data. The real-time processing and intelligent analysis system for ground-space time-frequency electromagnetic data, combined with cloud computing technology, enables efficient processing and in-depth analysis of complex electromagnetic signals, providing strong data support and decision-making basis for various industries.

[0003] Most current real-time electromagnetic data processing and analysis systems perform unified signal analysis, making it difficult to decompose non-stationary wideband signals into multiple sub-bands. This makes it difficult to perform more detailed time-frequency analysis of electromagnetic signals, effectively identify data points that are significantly different from normal patterns, and promptly discover potential problems. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a real-time processing and intelligent analysis system for ground-to-space time-frequency electromagnetic data, which is used to solve the technical problems that it is difficult to decompose non-stationary wide-band signals into multiple sub-bands, it is not convenient to perform more detailed time-frequency analysis of electromagnetic signals, and it is difficult to effectively identify data points that are significantly different from normal patterns.

[0005] To solve the above problems, the first aspect of the present invention provides a system for real-time processing and intelligent analysis of ground-space time-frequency electromagnetic data, comprising:

[0006] A multi-sensor module is used to set up sensor nodes in the ground and air environments of the target area, and deploy electromagnetic signal sensors for ground environment, aviation environment and semi-aeronautical environment at the sensor nodes;

[0007] The front-end data processing module is used to divide the ground and air environments of the target area into different detection blocks, set up edge computing nodes in each detection block, and perform preliminary electromagnetic signal processing and feature extraction through the edge computing nodes;

[0008] The data planning module extracts the characteristics of electromagnetic signals from edge computing nodes, evaluates the level of each detection block, determines the criticality of the detection signal based on the level of the detection block, and determines whether to send the data to the cloud data center or retain it at the edge node based on the criticality of the detection signal;

[0009] The time-frequency analysis module is used in the cloud data center to decompose the broadband signal into multiple sub-bands using the wavelet packet transform method, perform time-frequency analysis on the electromagnetic signal of each sub-band using the STFT short-time Fourier transform method, and extract the characteristics of each frequency sub-band;

[0010] The predictive analysis module is used to receive key data from edge nodes through the established cloud data center, and classify, predict and detect anomalies of the key data through K-means clustering algorithm and deep learning algorithm.

[0011] As a further solution of the present invention: the front-end data processing module performs preliminary electromagnetic signal processing and feature extraction through the edge computing node, including the following steps:

[0012] Pre-process the electromagnetic signals in the ground environment, aviation environment and semi-aviation environment, and remove background noise and interference signals through band-pass filters;

[0013] Extract features of electromagnetic signals, including time domain feature extraction and frequency domain feature extraction;

[0014] Among them, various electromagnetic signals are extracted for time domain feature extraction, including instantaneous amplitude, mean and variance data;

[0015] Frequency domain features are extracted for various electromagnetic signals respectively. The time domain signals are converted into frequency domain through FFT fast Fourier transform. The amplitude square of the frequency components of various electromagnetic signals is calculated through spectrum analysis to obtain the power spectrum. The power spectrum is divided by the total duration of the sampling time interval to obtain the energy distribution of various electromagnetic signals.

[0016] Specifically, in this embodiment, the basic distribution characteristics of the signal can be described by features such as instantaneous amplitude, mean, and variance, which facilitates the identification of different types of signals; by calculating the power spectral density, the energy distribution of the signal in different frequency ranges can be understood, which helps to discover potential frequency characteristics.

[0017] As a further solution of the present invention, the data planning module evaluates the level of each detection block based on the feature extraction of the electromagnetic signal by the edge computing node, including the following steps:

[0018] The electromagnetic signal data of the detected ground environment, aviation environment and semi-aviation environment are divided into several detection time intervals, and the time domain characteristics of the electromagnetic signal of each environment in each detection time interval, including instantaneous amplitude, mean and variance data, as well as the frequency domain characteristics of the electromagnetic signal, including the energy distribution of the electromagnetic signal, are obtained;

[0019] The level of each detection block is evaluated using the following formula:

[0020]

[0021] Among them, K is the hierarchical evaluation value of each detection block, μei is the mean value of the energy distribution of the i-th electromagnetic signal in the detection block in each detection time interval, Sei is the variance of the energy distribution of the i-th electromagnetic signal in the detection block in each detection time interval, μai is the mean value of the peak value of the i-th electromagnetic signal in the detection block in each detection time interval, Sai is the variance of the peak value of the i-th electromagnetic signal in the detection block in each detection time interval, μpi is the mean data of the time domain characteristics of the i-th electromagnetic signal in the detection block, and Spi is the variance data in the time domain characteristics of the i-th electromagnetic signal in the detection block;

[0022] The hierarchical evaluation values of each detection block are sorted from high to low, and the detection blocks corresponding to the first 20% of the hierarchical evaluation values are set as high-level detection blocks, the detection blocks corresponding to the last 20% of the hierarchical evaluation values are set as low-level detection blocks, and the remaining detection blocks are set as medium-level detection blocks.

[0023] As a further solution of the present invention, the data planning module determines the criticality of the detection signal according to the level of the detection block, and determines whether to send the data to the cloud data center or retain it at the edge node according to the criticality of the detection signal, including the following steps:

[0024] The criticality coefficient of the detection signal is calculated using the following formula:

[0025]

[0026] Wherein, α is the critical coefficient of the signal, μmax is the maximum value of the energy distribution mean of the electromagnetic signal of the detection block, and Smax is the maximum value of the variance data in the time domain characteristics of the electromagnetic signal of the detection block;

[0027] The criticality of the detection signal is determined based on the level of the detection block and the criticality coefficient of the electromagnetic signal within the detection block:

[0028] For high-level detection blocks, the detection signals of the first 20% of the critical coefficients of the electromagnetic signals in the detection block are set as critical detection signals, and the detection signals of the last 5% are set as non-critical detection signals;

[0029] For the middle-level detection block, the detection signals with the top 15% of the critical coefficients of the electromagnetic signals in the detection block are set as critical detection signals, and the detection signals with the bottom 10% are set as non-critical detection signals;

[0030] For low-level detection blocks, the detection signals of the first 10% of the critical coefficients of the electromagnetic signals in the detection block are set as critical detection signals, and the detection signals of the last 15% are set as non-critical detection signals;

[0031] The non-critical detection signal data in the detection block is retained in the edge node, the critical detection signal data in the detection block is sent to the cloud data center in real time, and other signal data in the detection block is sent to the cloud data center every preset time period.

[0032] As a further solution of the present invention: the time-frequency analysis module decomposes the broadband signal into multiple sub-bands by wavelet packet transform, performs time-frequency analysis on the electromagnetic signal of each sub-band by STFT short-time Fourier transform, and extracts the characteristics of each frequency sub-band, including the following steps:

[0033] The time-frequency analysis module constructs a wavelet packet tree based on the Haar wavelet basis using the wavelet packet transform method;

[0034] Extract decomposed subbands from the wavelet packet tree;

[0035] The window function and window length are selected by STFT short-time Fourier transform. In this embodiment, the window function Hamming window is selected, and the window length is 350 sample data.

[0036] Slide the selected window on each sub-band electromagnetic signal and perform Fourier transform at each window position;

[0037] By performing Fourier transform on the signal in each window, the time-frequency diagram of each sub-band electromagnetic signal is obtained;

[0038] According to the time-frequency diagram of each sub-band electromagnetic signal, the characteristics of each frequency sub-band are extracted, including amplitude spectrum, phase spectrum and energy spectrum;

[0039] Among them, the amplitude spectrum: calculates the amplitude of each frequency component; the phase spectrum: calculates the phase of each frequency component; the energy spectrum: calculates the energy of each frequency band.

[0040] According to the energy spectrum of each frequency sub-band extracted, the PSD power spectrum density graph is drawn with the frequency component of the electromagnetic signal as the X-axis and the energy distribution on the frequency component of the electromagnetic signal as the Y-axis;

[0041] The main frequency components and their harmonics are identified based on the power spectrum density graph, and the main frequency, bandwidth and harmonic ratio data of the sub-band are extracted as the characteristics of each frequency sub-band.

[0042] Harmonics are usually integer multiples of the fundamental frequency and can be identified by observing the PSD graph. Feature extraction includes: extracting the main frequency, bandwidth, and harmonic ratio.

[0043] As a further solution of the present invention: constructing a wavelet packet tree based on the Haar wavelet basis by wavelet packet transform method, including the following steps:

[0044] Perform a wavelet transform on the original signal to obtain the low-frequency part and the high-frequency part. The low-frequency part is passed through a low-pass filter to generate the approximate coefficient, and the high-frequency part is passed through a high-pass filter to generate the detail coefficient.

[0045] DWT discrete wavelet transform is performed using Haar wavelet basis;

[0046] Suppose the original signal is x(t). After a DWT discrete wavelet transform, the initial low-frequency coefficient and high-frequency coefficient are obtained. After a DWT discrete wavelet transform, we get: c0=DWT(x(t)); where c0 is the initial low-frequency coefficient and d0 is the initial high-frequency coefficient.

[0047] The low-frequency portion is further subjected to a wavelet transform. After a second DWT discrete wavelet transform of the low-frequency portion, the secondary low-frequency and high-frequency coefficients are obtained. Further DWT is performed on the low-frequency portion c0 to obtain finer low-frequency and high-frequency components. After the second DWT, we obtain: c1, d1 = DWT(c0), where c1 and d1 are the secondary low-frequency and high-frequency coefficients.

[0048] This process is repeated to decompose each low-frequency subband until the predetermined number of layers is reached.

[0049] As a further solution of the present invention: the prediction and analysis module classifies, predicts and detects anomalies of key data using a K-means clustering algorithm and a deep learning algorithm, including the following steps:

[0050] Obtain the main frequency, bandwidth and harmonic ratio data of the sub-band extracted by the time-frequency analysis module;

[0051] All features are integrated into a data table, where each row represents a sample and each column represents a feature;

[0052] The key data is classified by similarity using the K-means clustering algorithm, and the feature data of the sub-band is divided into different similarity categories;

[0053] Obtain the features of each frequency subband extracted by the time-frequency analysis module, including the amplitude spectrum, phase spectrum, and energy spectrum of the subband. Using deep learning algorithms, build anomaly detection and prediction models for different similarity classifications.

[0054] Through the anomaly detection and prediction models of different similarity classifications, the abnormal conditions of the sub-band are detected and predicted.

[0055] As a further solution of the present invention, similarity classification is performed on key data using a K-means clustering algorithm, and the feature data of the sub-bands are divided into different similarity classifications, including the following steps:

[0056] S1: Standardize the feature data in the data table;

[0057] S2: Calculate the silhouette coefficients under different K values by using the silhouette coefficient method, and select the K value with the highest silhouette coefficient as the final choice;

[0058] S3: Randomly select K initial cluster centers;

[0059] S4: For each sub-band feature data in the data table, calculate the distance between it and all cluster centers, and assign it to the cluster where the nearest cluster center is located;

[0060] S5: Calculate the mean of each sub-band feature data in all data tables in the current cluster according to the position of each sub-band feature data in all data tables in the current cluster, and update the position of the mean of each sub-band feature data in all data tables in the current cluster as the cluster center;

[0061] S6: Repeat steps S4 and S5 until the change in the latest updated cluster center position is less than the preset distance, and use the feature data of each sub-band in all data tables in each cluster as data in the same similarity classification.

[0062] As a further solution of the present invention: using a deep learning algorithm, constructing anomaly detection and prediction models for different similarity classifications, including the following steps:

[0063] Acquire historical data of sub-band feature data of different similarity classifications, and image data of amplitude spectrum, phase spectrum, and energy spectrum of corresponding sub-bands in the sub-band feature data of different similarity classifications;

[0064] Divide historical data into data at the time of anomaly, data in a preset time period before the anomaly, and normal data, and add data labels to the corresponding data according to the classification;

[0065] The neural network model is trained by adding labeled amplitude spectrum, phase spectrum and energy spectrum image data of different similarity classifications to build anomaly detection and prediction models of different similarity classifications.

[0066] Compared with the prior art, the present invention has the following beneficial effects:

[0067] The present invention analyzes signals at different time and frequency resolutions through wavelet packet transform, making it easy to process non-stationary broadband signals and decompose them into multiple sub-bands; through the STFT short-time Fourier transform method, it is easy to obtain the time and frequency information of the signal at the same time, making the understanding of the signal characteristics more comprehensive, especially when processing dynamically changing signals. Combining wavelet packet transform with STFT, the characteristics of each frequency band can be effectively extracted, providing a high-quality information basis for subsequent data analysis, classification or identification. The time-frequency analysis results can be displayed in a visual way such as a time-frequency graph, making the complex signal characteristics more intuitive and convenient for subsequent data interpretation and decision support.

[0068] This invention uses the K-means clustering algorithm to group similar data points, enabling unsupervised classification, reducing manual intervention and improving classification efficiency. Deep learning models can effectively identify data points that deviate significantly from normal patterns, enabling real-time anomaly detection and helping to promptly identify potential problems. Deep learning algorithms can capture complex nonlinear relationships in data, improving the accuracy and robustness of prediction models. By performing real-time analysis of key data, system status can be monitored promptly, and feedback based on the analysis results can be provided to optimize system operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0070] Figure 1 Schematic diagram of the system framework of the present invention. DETAILED DESCRIPTION

[0071] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all 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.

[0072] See also Figure 1 The first embodiment of the present invention provides a system for real-time processing and intelligent analysis of ground-space time-frequency electromagnetic data, comprising:

[0073] A multi-sensor module is used to set up sensor nodes in the ground and air environments of the target area, and deploy electromagnetic signal sensors for ground environment, aviation environment and semi-aeronautical environment at the sensor nodes;

[0074] The front-end data processing module is used to divide the ground and air environments of the target area into different detection blocks, set up edge computing nodes in each detection block, and perform preliminary electromagnetic signal processing and feature extraction through the edge computing nodes;

[0075] The data planning module extracts the characteristics of electromagnetic signals from edge computing nodes, evaluates the level of each detection block, determines the criticality of the detection signal based on the level of the detection block, and determines whether to send the data to the cloud data center or retain it at the edge node based on the criticality of the detection signal;

[0076] The time-frequency analysis module is used in the cloud data center to decompose the broadband signal into multiple sub-bands using the wavelet packet transform method, perform time-frequency analysis on the electromagnetic signal of each sub-band using the STFT short-time Fourier transform method, and extract the characteristics of each frequency sub-band;

[0077] The predictive analysis module is used to receive key data from edge nodes through the established cloud data center, and classify, predict and detect anomalies of key data through K-means clustering algorithm and deep learning algorithm.

[0078] Specifically, in this embodiment, the ground and air environments of the target area are divided into different detection blocks through the front-end data processing module, and edge computing nodes are set in each detection block to perform preliminary electromagnetic signal processing and feature extraction. The edge computing nodes are close to the data source, which facilitates low-latency data processing and timely response to detection data from different detection blocks. By processing data locally, only necessary information is transmitted to the central server, which reduces network bandwidth consumption and data transmission costs. By distributing computing tasks and storage tasks to multiple edge nodes, the computing load can be effectively shared and the overall performance of the system can be improved.

[0079] The data planning module extracts electromagnetic signal features from edge computing nodes, assesses the hierarchical level of each detection block, determines the criticality of the detection signal based on the detection block's hierarchy, and determines whether to send the data to the cloud data center or retain it at the edge node based on the criticality of the detection signal. This criticality assessment enables more intelligent data processing and decision-making, ensuring that important information is transmitted promptly while less important information can be processed locally. Sending only critical data to the cloud reduces unnecessary data transmission, conserving network bandwidth, and improving transmission efficiency. This data screening and hierarchical management effectively utilizes computing resources, avoids cloud overload, and reduces storage costs.

[0080] Dynamically adjust data transmission strategies based on real-time assessment results, enabling the system to better adapt to environmental changes and demand fluctuations. By focusing on key signals, the quality of data uploaded to the cloud can be improved, facilitating subsequent data analysis and decision support.

[0081] The time-frequency analysis module uses the wavelet packet transform (STFT) in the cloud data center to decompose broadband signals into multiple subbands. The STFT (Short-Time Fourier Transform) method then performs time-frequency analysis on the electromagnetic signals in each subband, extracting the characteristics of each frequency subband. The wavelet packet transform (STFT) can analyze signals at varying time and frequency resolutions, facilitating the decomposition of non-stationary broadband signals into multiple subbands. The STFT (Short-Time Fourier Transform) method simultaneously obtains both time and frequency information, enabling a more comprehensive understanding of signal characteristics, particularly when processing dynamically changing signals. Combining the wavelet packet transform with the STFT effectively extracts the characteristics of each frequency band, providing a high-quality information foundation for subsequent data analysis, classification, or recognition.

[0082] The results of time-frequency analysis can be displayed in visual ways such as time-frequency graphs, making the characteristics of complex signals more intuitive and facilitating subsequent data interpretation and decision support.

[0083] The predictive analysis module receives key data from edge nodes through the established cloud data center, and classifies, predicts and detects anomalies of the key data through K-means clustering algorithm and deep learning algorithm.

[0084] Cloud data centers have powerful computing capabilities that can quickly process and analyze large amounts of critical data, improving data processing efficiency. By combining K-means clustering and deep learning, they can perform more complex data analysis tasks, identify potential patterns and trends, and enhance intelligent decision-making capabilities.

[0085] The K-means clustering algorithm groups similar data points into a single group, enabling unsupervised classification, reducing manual intervention and improving classification efficiency. Deep learning models can effectively identify data points that differ significantly from normal patterns, enabling real-time anomaly detection and helping to identify potential problems promptly.

[0086] Deep learning algorithms can capture complex nonlinear relationships in data, improving the accuracy and robustness of predictive models. By analyzing key data in real time, system status can be monitored promptly, and feedback based on the analysis results can be provided to optimize system operation.

[0087] In one embodiment of the present invention, the front-end data processing module performs preliminary electromagnetic signal processing and feature extraction through the edge computing node, including the following steps:

[0088] Pre-process the electromagnetic signals in the ground environment, aviation environment and semi-aviation environment, and remove background noise and interference signals through band-pass filters;

[0089] Extract features of electromagnetic signals, including time domain feature extraction and frequency domain feature extraction;

[0090] Among them, various electromagnetic signals are extracted for time domain feature extraction, including instantaneous amplitude, mean and variance data;

[0091] Frequency domain features are extracted for various electromagnetic signals respectively. The time domain signals are converted into frequency domain through FFT fast Fourier transform. The amplitude square of the frequency components of various electromagnetic signals is calculated through spectrum analysis to obtain the power spectrum. The power spectrum is divided by the total duration of the sampling time interval to obtain the energy distribution of various electromagnetic signals.

[0092] Specifically, in this embodiment, the basic distribution characteristics of the signal can be described by features such as instantaneous amplitude, mean, and variance, which facilitates the identification of different types of signals; by calculating the power spectral density, the energy distribution of the signal in different frequency ranges can be understood, which helps to discover potential frequency characteristics.

[0093] In one embodiment of the present invention, the data planning module evaluates the level of each detection block based on the feature extraction of the electromagnetic signal by the edge computing node, including the following steps:

[0094] The electromagnetic signal data of the detected ground environment, aviation environment and semi-aviation environment are divided into several detection time intervals, and the time domain characteristics of the electromagnetic signal of each environment in each detection time interval, including instantaneous amplitude, mean and variance data, as well as the frequency domain characteristics of the electromagnetic signal, including the energy distribution of the electromagnetic signal, are obtained;

[0095] The level of each detection block is evaluated using the following formula:

[0096]

[0097] Among them, K is the hierarchical evaluation value of each detection block, μei is the mean value of the energy distribution of the i-th electromagnetic signal in the detection block in each detection time interval, Sei is the variance of the energy distribution of the i-th electromagnetic signal in the detection block in each detection time interval, μai is the mean value of the peak value of the i-th electromagnetic signal in the detection block in each detection time interval, Sai is the variance of the peak value of the i-th electromagnetic signal in the detection block in each detection time interval, μpi is the mean data of the time domain characteristics of the i-th electromagnetic signal in the detection block, and Spi is the variance data in the time domain characteristics of the i-th electromagnetic signal in the detection block;

[0098] The hierarchical evaluation values of each detection block are sorted from high to low, and the detection blocks corresponding to the first 20% of the hierarchical evaluation values are set as high-level detection blocks, the detection blocks corresponding to the last 20% of the hierarchical evaluation values are set as low-level detection blocks, and the remaining detection blocks are set as medium-level detection blocks.

[0099] Specifically, in this embodiment, the entire monitoring area is divided into multiple detection blocks, and a significance score is calculated for each block. The blocks are then sorted based on the scores. Based on actual needs, the monitoring area is divided into several smaller blocks, each corresponding to a set of electromagnetic signal data. The significance scoring model from the previous step is applied to each block to obtain a significance score. All blocks are then sorted based on the scores to determine which areas are most critical and require priority attention and processing.

[0100] In one embodiment of the present invention, the data planning module determines the criticality of the detection signal according to the level of the detection block, and determines whether to send the data to the cloud data center or retain it at the edge node according to the criticality of the detection signal, including the following steps:

[0101] The criticality coefficient of the detection signal is calculated using the following formula:

[0102]

[0103] Wherein, α is the critical coefficient of the signal, μmax is the maximum value of the energy distribution mean of the electromagnetic signal of the detection block, and Smax is the maximum value of the variance data in the time domain characteristics of the electromagnetic signal of the detection block;

[0104] The criticality of the detection signal is determined based on the level of the detection block and the criticality coefficient of the electromagnetic signal within the detection block:

[0105] For high-level detection blocks, the detection signals of the first 20% of the critical coefficients of the electromagnetic signals in the detection block are set as critical detection signals, and the detection signals of the last 5% are set as non-critical detection signals;

[0106] For the middle-level detection block, the detection signals with the top 15% of the critical coefficients of the electromagnetic signals in the detection block are set as critical detection signals, and the detection signals with the bottom 10% are set as non-critical detection signals;

[0107] For low-level detection blocks, the detection signals of the first 10% of the critical coefficients of the electromagnetic signals in the detection block are set as critical detection signals, and the detection signals of the last 15% are set as non-critical detection signals;

[0108] The non-critical detection signal data in the detection block is retained in the edge node, the critical detection signal data in the detection block is sent to the cloud data center in real time, and other signal data in the detection block is sent to the cloud data center every preset time period.

[0109] In one embodiment of the present invention, the time-frequency analysis module decomposes the broadband signal into multiple sub-bands by wavelet packet transform, performs time-frequency analysis on the electromagnetic signal of each sub-band by STFT short-time Fourier transform, and extracts the characteristics of each frequency sub-band, including the following steps:

[0110] The time-frequency analysis module constructs a wavelet packet tree based on the Haar wavelet basis using the wavelet packet transform method;

[0111] Extract decomposed subbands from the wavelet packet tree;

[0112] The window function and window length are selected by STFT short-time Fourier transform. In this embodiment, the window function Hamming window is selected, and the window length is 350 sample data.

[0113] Slide the selected window on each sub-band electromagnetic signal and perform Fourier transform at each window position;

[0114] By performing Fourier transform on the signal in each window, the time-frequency diagram of each sub-band electromagnetic signal is obtained;

[0115] According to the time-frequency diagram of each sub-band electromagnetic signal, the characteristics of each frequency sub-band are extracted, including amplitude spectrum, phase spectrum and energy spectrum;

[0116] Among them, the amplitude spectrum: calculates the amplitude of each frequency component; the phase spectrum: calculates the phase of each frequency component; the energy spectrum: calculates the energy of each frequency band.

[0117] According to the energy spectrum of each frequency sub-band extracted, the PSD power spectrum density graph is drawn with the frequency component of the electromagnetic signal as the X-axis and the energy distribution on the frequency component of the electromagnetic signal as the Y-axis;

[0118] The main frequency components and their harmonics are identified based on the power spectrum density graph, and the main frequency, bandwidth and harmonic ratio data of the sub-band are extracted as the characteristics of each frequency sub-band.

[0119] Harmonics are usually integer multiples of the fundamental frequency and can be identified by observing the PSD graph. Feature extraction includes extracting the main frequency, bandwidth, and harmonic ratio.

[0120] In one embodiment of the present invention, a wavelet packet tree is constructed based on a Haar wavelet basis using a wavelet packet transform method, including the following steps:

[0121] Perform a wavelet transform on the original signal to obtain the low-frequency part and the high-frequency part. The low-frequency part is passed through a low-pass filter to generate the approximate coefficient, and the high-frequency part is passed through a high-pass filter to generate the detail coefficient.

[0122] DWT discrete wavelet transform is performed using Haar wavelet basis;

[0123] Suppose the original signal is x(t). After a DWT discrete wavelet transform, the initial low-frequency coefficient and high-frequency coefficient are obtained. After a DWT discrete wavelet transform, we get: c0=DWT(x(t)); where c0 is the initial low-frequency coefficient and d0 is the initial high-frequency coefficient.

[0124] The low-frequency portion is further subjected to a wavelet transform. After a second DWT discrete wavelet transform of the low-frequency portion, the secondary low-frequency and high-frequency coefficients are obtained. Further DWT is performed on the low-frequency portion c0 to obtain finer low-frequency and high-frequency components. After the second DWT, we obtain: c1, d1 = DWT(c0), where c1 and d1 are the secondary low-frequency and high-frequency coefficients.

[0125] This process is repeated to decompose each low-frequency subband until the predetermined number of layers is reached.

[0126] Specifically, in this embodiment, each node of the wavelet packet tree constructed using the Haar wavelet basis represents a frequency band, and the depth of the tree determines the number of decomposition levels.

[0127] For each layer, the signal is passed through a low-pass filter to generate approximate coefficients, and through a high-pass filter to generate detail coefficients;

[0128] After each layer, the same operation can be continued on the approximation coefficients and detail coefficients to obtain more sub-bands.

[0129] Through multi-level decomposition, several sub-bands are obtained, each of which corresponds to a signal component in a different frequency range.

[0130] Normally, after decomposing to the Nth level, we get A sub-band.

[0131] In one embodiment of the present invention, the prediction and analysis module classifies, predicts, and detects anomalies of key data using a K-means clustering algorithm and a deep learning algorithm, including the following steps:

[0132] Obtain the main frequency, bandwidth and harmonic ratio data of the sub-band extracted by the time-frequency analysis module;

[0133] All features are integrated into a data table, where each row represents a sample and each column represents a feature;

[0134] The key data is classified by similarity using the K-means clustering algorithm, and the feature data of the sub-band is divided into different similarity categories;

[0135] Obtain the features of each frequency subband extracted by the time-frequency analysis module, including the amplitude spectrum, phase spectrum, and energy spectrum of the subband. Using deep learning algorithms, build anomaly detection and prediction models for different similarity classifications.

[0136] The abnormal conditions of the sub-band are detected and predicted through the anomaly detection and prediction models of different similarity classifications.

[0137] In one embodiment of the present invention, similarity classification is performed on key data using a K-means clustering algorithm to classify feature data of sub-bands into different similarity categories, including the following steps:

[0138] S1: Standardize the feature data in the data table;

[0139] S2: Calculate the silhouette coefficients under different K values by using the silhouette coefficient method, and select the K value with the highest silhouette coefficient as the final choice;

[0140] S3: Randomly select K initial cluster centers;

[0141] S4: For each sub-band feature data in the data table, calculate the distance between it and all cluster centers, and assign it to the cluster where the nearest cluster center is located;

[0142] S5: Calculate the mean of each sub-band feature data in all data tables in the current cluster according to the position of each sub-band feature data in all data tables in the current cluster, and update the position of the mean of each sub-band feature data in all data tables in the current cluster as the cluster center;

[0143] S6: Repeat steps S4 and S5 until the change in the latest updated cluster center position is less than the preset distance, and use the feature data of each sub-band in all data tables in each cluster as data in the same similarity classification.

[0144] In one embodiment of the present invention, a deep learning algorithm is used to construct anomaly detection and prediction models for different similarity classifications, including the following steps:

[0145] Acquire historical data of sub-band feature data of different similarity classifications, and image data of amplitude spectrum, phase spectrum, and energy spectrum of corresponding sub-bands in the sub-band feature data of different similarity classifications;

[0146] Divide historical data into data at the time of anomaly, data in a preset time period before the anomaly, and normal data, and add data labels to the corresponding data according to the classification;

[0147] The neural network model is trained by adding labeled amplitude spectrum, phase spectrum and energy spectrum image data of different similarity classifications to build anomaly detection and prediction models of different similarity classifications.

[0148] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A real-time processing and intelligent analysis system for ground-to-air time-frequency electromagnetic data, characterized in that: include: A multi-sensor module is used to set up sensor nodes in the ground and air environments of the target area, and deploy electromagnetic signal sensors for ground environment, aviation environment and semi-aeronautical environment at the sensor nodes; The front-end data processing module is used to divide the ground and air environments of the target area into different detection blocks, set up edge computing nodes in each detection block, and perform preliminary electromagnetic signal processing and feature extraction through the edge computing nodes; The data planning module extracts the characteristics of electromagnetic signals from edge computing nodes, evaluates the level of each detection block, determines the criticality of the detection signal based on the level of the detection block, and determines whether to send the data to the cloud data center or retain it at the edge node based on the criticality of the detection signal; The time-frequency analysis module is used in the cloud data center to decompose the broadband signal into multiple sub-bands using the wavelet packet transform method, perform time-frequency analysis on the electromagnetic signal of each sub-band using the STFT short-time Fourier transform method, and extract the characteristics of each frequency sub-band; The predictive analysis module is used to receive key data from edge nodes through the established cloud data center, and classify, predict and detect anomalies of key data through K-means clustering algorithm and deep learning algorithm; The data planning module evaluates the level of each detection block based on the feature extraction of electromagnetic signals by the edge computing node, including the following steps: The electromagnetic signal data of the detected ground environment, aviation environment and semi-aviation environment are divided into several detection time intervals, and the time domain characteristics of the electromagnetic signal of each environment in each detection time interval, including instantaneous amplitude, mean and variance data, as well as the frequency domain characteristics of the electromagnetic signal, including the energy distribution of the electromagnetic signal, are obtained; The level of each detection block is evaluated using the following formula: Among them, K is the hierarchical evaluation value of each detection block, μei is the mean value of the energy distribution of the i-th electromagnetic signal in the detection block in each detection time interval, Sei is the variance of the energy distribution of the i-th electromagnetic signal in the detection block in each detection time interval, μai is the mean value of the peak value of the i-th electromagnetic signal in the detection block in each detection time interval, Sai is the variance of the peak value of the i-th electromagnetic signal in the detection block in each detection time interval, μpi is the mean data of the time domain characteristics of the i-th electromagnetic signal in the detection block, and Spi is the variance data in the time domain characteristics of the i-th electromagnetic signal in the detection block; Sort the hierarchical evaluation values of each detection block from high to low, set the detection blocks corresponding to the first 20% of the hierarchical evaluation values as high-level detection blocks, set the detection blocks corresponding to the last 20% of the hierarchical evaluation values as low-level detection blocks, and set the remaining detection blocks as medium-level detection blocks; The data planning module determines the criticality of the detection signal according to the level of the detection block, and determines whether to send the data to the cloud data center or retain it at the edge node according to the criticality of the detection signal, including the following steps: The criticality coefficient of the detection signal is calculated using the following formula: Wherein, α is the critical coefficient of the signal, μmax is the maximum value of the energy distribution mean of the electromagnetic signal of the detection block, and Smax is the maximum value of the variance data in the time domain characteristics of the electromagnetic signal of the detection block; The criticality of the detection signal is determined based on the level of the detection block and the criticality coefficient of the electromagnetic signal within the detection block: For high-level detection blocks, the detection signals of the first 20% of the critical coefficients of the electromagnetic signals in the detection block are set as critical detection signals, and the detection signals of the last 5% are set as non-critical detection signals; For the middle-level detection block, the detection signals with the top 15% of the critical coefficients of the electromagnetic signals in the detection block are set as critical detection signals, and the detection signals with the bottom 10% are set as non-critical detection signals; For low-level detection blocks, the detection signals of the first 10% of the critical coefficients of the electromagnetic signals in the detection block are set as critical detection signals, and the detection signals of the last 15% are set as non-critical detection signals; The non-critical detection signal data in the detection block is retained in the edge node, the critical detection signal data in the detection block is sent to the cloud data center in real time, and other signal data in the detection block is sent to the cloud data center every preset time period.

2. The system for real-time processing and intelligent analysis of ground-space time-frequency electromagnetic data according to claim 1, characterized in that: The front-end data processing module performs preliminary electromagnetic signal processing and feature extraction through the edge computing node, including the following steps: Pre-process the electromagnetic signals in the ground environment, aviation environment and semi-aviation environment, and remove background noise and interference signals through band-pass filters; Extract features of electromagnetic signals, including time domain feature extraction and frequency domain feature extraction; Among them, various electromagnetic signals are extracted for time domain feature extraction, including instantaneous amplitude, mean and variance data; Frequency domain features are extracted for various electromagnetic signals respectively. The time domain signals are converted into frequency domain through FFT fast Fourier transform. The amplitude square of the frequency components of various electromagnetic signals is calculated through spectrum analysis to obtain the power spectrum. The power spectrum is divided by the total duration of the sampling time interval to obtain the energy distribution of various electromagnetic signals.

3. The real-time processing and intelligent analysis system for ground-space time-frequency electromagnetic data according to claim 1 is characterized in that: The time-frequency analysis module decomposes the broadband signal into multiple sub-bands by wavelet packet transform, performs time-frequency analysis on the electromagnetic signal of each sub-band by STFT short-time Fourier transform, and extracts the characteristics of each frequency sub-band, including the following steps: The time-frequency analysis module constructs a wavelet packet tree based on the Haar wavelet basis using the wavelet packet transform method; Extract decomposed subbands from the wavelet packet tree; Select the window function and window length through STFT short-time Fourier transform; Slide the selected window on each sub-band electromagnetic signal and perform Fourier transform at each window position; By performing Fourier transform on the signal in each window, the time-frequency diagram of each sub-band electromagnetic signal is obtained; According to the time-frequency diagram of each sub-band electromagnetic signal, the characteristics of each frequency sub-band are extracted, including amplitude spectrum, phase spectrum and energy spectrum; According to the energy spectrum of each frequency sub-band extracted, the PSD power spectrum density graph is drawn with the frequency component of the electromagnetic signal as the X-axis and the energy distribution on the frequency component of the electromagnetic signal as the Y-axis; The main frequency components and their harmonics are identified based on the power spectrum density graph, and the main frequency, bandwidth and harmonic ratio data of the sub-band are extracted as the characteristics of each frequency sub-band.

4. The system for real-time processing and intelligent analysis of ground-space time-frequency electromagnetic data according to claim 3, characterized in that: The wavelet packet transform method is used to construct a wavelet packet tree based on the Haar wavelet basis, which includes the following steps: Perform a wavelet transform on the original signal to obtain the low-frequency part and the high-frequency part. The low-frequency part is passed through a low-pass filter to generate the approximate coefficient, and the high-frequency part is passed through a high-pass filter to generate the detail coefficient. After performing a DWT discrete wavelet transform on the original signal, the initial low-frequency coefficients and high-frequency coefficients are obtained; The low-frequency part is further subjected to wavelet transform, and the low-frequency part is subjected to a second DWT discrete wavelet transform to obtain the secondary low-frequency coefficients and high-frequency coefficients; This process is repeated to decompose each low-frequency subband until the predetermined number of layers is reached.

5. The system for real-time processing and intelligent analysis of ground-to-space time-frequency electromagnetic data according to claim 1, characterized in that: The prediction and analysis module uses the K-means clustering algorithm and deep learning algorithm to classify, predict, and detect anomalies of key data, including the following steps: Obtain the main frequency, bandwidth and harmonic ratio data of the sub-band extracted by the time-frequency analysis module; All features are integrated into a data table, where each row represents a sample and each column represents a feature; The key data is classified by similarity using the K-means clustering algorithm, and the feature data of the sub-band is divided into different similarity categories; Obtain the features of each frequency subband extracted by the time-frequency analysis module, including the amplitude spectrum, phase spectrum, and energy spectrum of the subband. Using deep learning algorithms, build anomaly detection and prediction models for different similarity classifications. Through the anomaly detection and prediction models of different similarity classifications, the abnormal conditions of the sub-band are detected and predicted.

6. The system for real-time processing and intelligent analysis of ground-space time-frequency electromagnetic data according to claim 5, characterized in that: The K-means clustering algorithm is used to classify the key data by similarity, and the feature data of the sub-bands are divided into different similarity categories, including the following steps: S1: Standardize the feature data in the data table; S2: Calculate the silhouette coefficients under different K values by using the silhouette coefficient method, and select the K value with the highest silhouette coefficient as the final choice; S3: Randomly select K initial cluster centers; S4: For each sub-band feature data in the data table, calculate the distance between it and all cluster centers, and assign it to the cluster where the nearest cluster center is located; S5: Calculate the mean of each sub-band feature data in all data tables in the current cluster according to the position of each sub-band feature data in all data tables in the current cluster, and update the position of the mean of each sub-band feature data in all data tables in the current cluster as the cluster center; S6: Repeat steps S4 and S5 until the change in the latest updated cluster center position is less than the preset distance, and use the feature data of each sub-band in all data tables in each cluster as data in the same similarity classification.

7. The system for real-time processing and intelligent analysis of ground-to-space time-frequency electromagnetic data according to claim 5, characterized in that: Through deep learning algorithms, we build models for anomaly detection and prediction for different similarity classifications, including the following steps: Acquire historical data of sub-band feature data of different similarity classifications, and image data of amplitude spectrum, phase spectrum, and energy spectrum of corresponding sub-bands in the sub-band feature data of different similarity classifications; Divide historical data into data at the time of anomaly, data in a preset time period before the anomaly, and normal data, and add data labels to the corresponding data according to the classification; The neural network model is trained by adding labeled amplitude spectrum, phase spectrum and energy spectrum image data of different similarity classifications to build anomaly detection and prediction models of different similarity classifications.

Images