Multi-channel data synchronous acquisition and optimization processing method

Abstract

The invention discloses a multi-channel data synchronous acquisition and optimization processing method, and relates to the technical field of data processing, and the method comprises the steps: controlling a multi-channel sensor to synchronously acquire data through a synchronous trigger signal, carrying out the preprocessing of the acquired multi-channel data, and generating a multi-channel data matrix; constructing an initial separation matrix, dynamically adjusting a learning rate based on a mutual information value, iteratively optimizing the separation matrix through a negentropy maximization criterion, realizing initial separation of multi-channel mixed signals, and screening source signals of a target frequency band; performing modal decomposition on the target source signal, extracting an extreme point and fitting an envelope line, and iteratively screening modal components meeting a preset condition to obtain an effective modal component set; the modal components are classified into cross-channel common modals or channel exclusive modals based on cosine similarity, weights are distributed according to the signal-to-noise ratio and weighted fusion is carried out, a result is output after the quality of the fusion signal is evaluated, and the signal-to-noise ratio and stability of the fusion signal are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method for synchronous acquisition and optimization of multi-channel data. Background Technology

[0002] Multi-channel data synchronous acquisition and processing technology has wide applications in fields such as industrial monitoring, medical diagnosis, and environmental monitoring. Existing technologies typically use multi-sensor systems for data acquisition and achieve temporal consistency of multi-channel data through synchronous trigger signals. In terms of signal processing, methods such as independent component analysis (ICA) and mode decomposition are widely used for the separation and feature extraction of multi-channel signals.

[0003] However, existing technologies still suffer from insufficient time offset correction accuracy during multi-channel data synchronous acquisition, resulting in unsatisfactory time alignment of multi-channel data. Traditional signal separation methods are prone to slow convergence or poor separation when processing highly correlated signals, affecting the accuracy of subsequent signal analysis.

[0004] Furthermore, in the signal optimization processing stage, existing technologies are not precise enough in distinguishing between cross-channel common modes and channel-specific modes, resulting in insufficient signal-to-noise ratio and stability of the fused signal. At the same time, existing methods lack dynamic adjustment mechanisms in the selection and weight allocation of modal components, making it difficult to adapt to the needs of different application scenarios and limiting the versatility and practicality of the methods. Summary of the Invention

[0005] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method for synchronous acquisition and optimization of multi-channel data. By employing high-precision synchronous trigger signals and dynamic time offset correction technology, it solves the problem of insufficient timing alignment accuracy of multi-channel data. The separation matrix optimization method based on dynamically adjusting the learning rate according to mutual information values ​​improves the separation efficiency and accuracy of highly correlated signals. By combining mode decomposition and cross-channel mode classification mechanisms, it achieves accurate differentiation and weighted fusion of common modes and specific modes, significantly improving the signal-to-noise ratio and stability of the fused signal.

[0006] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: a method for multi-channel data synchronous acquisition and optimization processing, comprising: By controlling the multi-channel sensors to synchronously acquire data through synchronous triggering signals, the acquired multi-channel data is preprocessed to generate a multi-channel data matrix. An initial separation matrix is ​​constructed, the learning rate is dynamically adjusted based on the mutual information value, and the separation matrix is ​​iteratively optimized through the negative entropy maximization criterion to achieve the initial separation of multi-channel mixed signals, and the source signals of the target frequency band are selected to form a target source signal set; Perform modal decomposition on the target source signal, extract the extreme points and fit the envelope curve, and iteratively screen the modal components that meet the preset conditions to obtain the set of effective modal components; Classify the modal components into cross-channel common modes or channel-exclusive modes based on cosine similarity, assign weights according to the signal-to-noise ratio and perform weighted fusion, and output the result after evaluating the quality of the fused signal.

[0007] Furthermore, multi-channel data synchronous acquisition is achieved by building an acquisition system including M sensors, a synchronous trigger module, and a high-speed data cache module, where M≥2 and is determined according to the monitoring target dimension; the sampling rate f is set to 2 to 5 times the highest frequency of the target signal, and the synchronous trigger module outputs a homologous trigger signal; the start timestamps of each channel are extracted through the built-in clock chip of the synchronous trigger module, and taking the start timestamp of the first channel as the reference, the linear interpolation algorithm is used to correct the time offset of other channels.

[0008] Furthermore, data preprocessing includes centering, normalization, and outlier repair. Centering subtracts the mean of all sampling points in the channel from each sampling point data; normalization linearly maps all data to the interval [-1, 1]; the 3σ criterion is used to identify outliers, and for outliers or missing data points, they are repaired by the sliding window weighted average method.

[0009] Furthermore, the initial separation matrix is M×M dimension, and the matrix elements are generated by Gaussian random distribution with a mean of 0 and a variance of 1 and are orthogonally processed; the independence judgment threshold is set to 0.001 to 0.01, the maximum number of iterations is 50 to 200 times, and the adaptive learning rate value range is [0.01, 0.1].

[0010] Furthermore, the mutual information value MI of any two-channel data is calculated using the K-nearest neighbor method, and the learning rate η is dynamically adjusted based on MI. When MI≥M1, η takes 0.08 to 0.1; when M2<MI<M1, η takes 0.04 to 0.08; when MI≤M2, η takes 0.01 to 0.04, where M1 and M2 are preset mutual information thresholds; the iterative optimization is based on the negative entropy maximization criterion: multiply the multi-channel data matrix by the initial separation matrix to obtain the output signal matrix, convert the output signal through the Sigmoid activation function, update the separation matrix using the natural gradient descent method, perform normalization processing after each update, calculate the difference between the current separation matrix and the separation matrix of the previous iteration, and stop the iteration when the F norm of this difference is less than the independence judgment threshold or the maximum number of iterations is reached. <0000​Furthermore, the Welch method is used to calculate the power spectral density (PSD) of the source signal. The signal is divided into L segments with an overlap rate of 50% between adjacent segments. After applying a Hanning window, the PSD curve is obtained by fast Fourier transform. The frequency range corresponding to the main peak frequency is extracted from the PSD curve and compared with the preset target frequency band. Source signals with a matching degree ≥85% are selected to form the target source signal set, and interference source signals with a matching degree <85% are eliminated.

[0012] Furthermore, by traversing the signals of all channels in the target source signal set, the extreme points, including local maxima and local minima, are extracted using the sliding window method to obtain the global joint maxima set and joint minima set; The joint maximum and joint minimum point sets are fitted by cubic spline interpolation to obtain the upper and lower envelopes. The mean line is obtained by averaging them. The mean line is obtained by subtracting the target source signal of each channel from the mean line. Repeat the iteration until the intermediate signal satisfies the IMF condition: the difference between the number of extreme points and the number of zero crossings is ≤1, and the absolute value of the mean of the upper and lower envelopes at any time is ≤1% of the maximum value of the absolute value of the intermediate signal. The intermediate signal at this time is the IMF component. Subtract the IMF component from the target source signal to obtain the residual signal. Use the residual signal as the new input signal and repeat the iteration to obtain K IMF components. Stop the decomposition when the residual signal has only one extreme point or no extreme point; otherwise, continue the iteration.

[0013] Furthermore, the Pearson correlation coefficient between each IMF component in each channel and the corresponding target source signal is calculated; IMF components with Pearson correlation coefficients ≥ correlation threshold are selected to form the effective modal component set of the channel, and low-correlation modal components with Pearson correlation coefficients < correlation threshold are removed; finally, M effective modal component sets are obtained.

[0014] Furthermore, the cosine similarity of IMF components of the same order in any two channels is calculated. Modal components with cosine similarity ≥ similarity threshold are classified as cross-channel shared modes, and those with cosine similarity < similarity threshold are classified as channel-specific modes. The signal-to-noise ratio of each modal component is calculated, and the final weights of cross-channel shared modes and channel-specific modes are assigned based on the signal-to-noise ratio.

[0015] Furthermore, the fused signal of each channel is obtained by weighted summation of the effective modal components according to the final weights, and the global fused signal is the arithmetic mean of the fused signals of all channels. The quality of the fused signal is evaluated by three indicators, including signal-to-noise ratio (SNR), root mean square error (RMSE), and peak signal-to-noise ratio (PSNR). If the indicators are not met, the learning rate is adjusted.

[0016] (III) Beneficial Effects This invention provides a method for multi-channel synchronous data acquisition and optimization processing, which has the following beneficial effects: (1) By controlling the synchronous acquisition of data by multi-channel sensors through synchronous triggering signals and combining high-precision timestamp correction technology, the problem of insufficient time alignment accuracy of multi-channel data is effectively solved. In the preprocessing stage, centralization, normalization and outlier repair are adopted, which significantly improves the consistency and reliability of data. By correcting the time offset through linear interpolation algorithm, the strict synchronization of multi-channel data is ensured, laying a high-quality data foundation for subsequent signal separation and optimization processing, thereby improving the accuracy and stability of the overall method.

[0017] (2) By constructing an initial separation matrix and dynamically adjusting the learning rate based on mutual information values, and iteratively optimizing the separation matrix in combination with the negative entropy maximization criterion, the separation efficiency and accuracy of multi-channel mixed signals are significantly improved. The dynamic learning rate mechanism is adaptively adjusted according to the signal correlation, which not only accelerates the convergence speed of highly correlated signals, but also ensures the stability of low-correlation signals. By screening target frequency band signals through the Welch method, interference sources are effectively eliminated, providing a high-quality source signal set for subsequent mode decomposition and fusion, thereby enhancing the robustness and adaptability of the overall method.

[0018] (3) By accurately extracting extreme points using the sliding window method and density clustering, and fitting the envelope using cubic spline interpolation, high-precision mode decomposition of the target source signal was achieved. By iteratively screening mode components that meet the IMF conditions, key feature components in the signal were effectively separated. Furthermore, the low correlation components dominated by noise were eliminated using the Pearson correlation coefficient, which significantly improved the accuracy and reliability of mode decomposition. The final set of effective mode components provided high-quality input for subsequent cross-channel mode classification and fusion, enhancing the robustness and adaptability of the overall method.

[0019] (4) By accurately distinguishing cross-channel shared modes and channel-specific modes through cosine similarity, and dynamically allocating weights based on signal-to-noise ratio, the accuracy and stability of the fused signal are significantly improved. A weighted fusion strategy is adopted to optimize and integrate effective modal components, ensuring that high signal-to-noise ratio components receive higher weights. At the same time, the fusion quality is strictly evaluated through three indicators: SNR, RMSE and PSNR. If the standard is not met, the learning rate is adaptively adjusted to optimize the separation process. The final output single-channel and global fused signals have both high signal-to-noise ratio and low error, which are suitable for multi-source data collaborative analysis in complex scenarios, enhancing the practicality and reliability of the method. Attached Figure Description

[0020] Figure 1 This is a schematic diagram illustrating the steps of the multi-channel data synchronous acquisition and optimization processing method of the present invention; Figure 2This is a schematic diagram of the multi-channel data synchronous acquisition and optimization processing method of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figures 1-2 This invention provides a method for multi-channel data synchronous acquisition and optimization processing, comprising the following steps: Step 1: Control the multi-channel sensors to synchronously acquire data by using a synchronous trigger signal, preprocess the acquired multi-channel data, and generate a multi-channel data matrix; Step one includes the following: Step 101: Build a multi-channel synchronous acquisition system, configure M sensors, a synchronous trigger module and a high-speed data buffer module, M≥2, and determine the specific number according to the requirements of the monitoring target dimension; set a uniform sampling rate f, specifically determined as 2 to 5 times the highest frequency of the target signal; output the same trigger signal to all sensors through the synchronous trigger module to achieve consistent sensor trigger timing; Step 102: Send a start command to the synchronization trigger module via the host computer. The synchronization trigger module generates a standard trigger pulse signal and sends it synchronously to all sensors. After receiving the trigger signal, each sensor simultaneously collects multi-dimensional data of the target object, such as mechanical vibration, acoustic signals, and electrical parameters. The collected multi-channel data is stored in real time to the high-speed data cache module via the data transmission bus. Step 103: Extract the start timestamp and sampling point timestamp of the data collected by each channel through the clock chip built into the synchronization trigger module. Using the start timestamp of the first channel as the reference time point, calculate the time offset between the start timestamp and the reference time point of other channels. For channel data with offset ≠ 0, use a linear interpolation algorithm to correct the time coordinate of the sampling points of the offset channels. Step 104: Calculate the mean of all sampling points for the synchronous data frames of each channel, and subtract this mean from the data of each sampling point to complete the centering operation; calculate the maximum value Max and minimum value Min of the centered data for this channel, and use the formula x_norm=(x_cent-Min) / (Max-Min)×2-1, where x_cent is the centered data and x_norm is the normalized data, to linearly map all data to the interval [-1, 1] to complete the normalization operation; output the multi-channel data matrix X after normalization, with dimensions N×M, where N is the total number of sampling points and M is the number of channels; Step 105: For each channel of data in the data matrix X, outlier identification is performed using the 3σ criterion. The standard deviation σ and mean μ of the data for that channel are calculated. If the sampled data is within the range of [mean - 3σ, mean + 3σ], it is determined to be impulse noise or an outlier. For the identified impulse noise points or missing data points, a sliding window weighted average method is used for repair. A sliding window with a length of 5 sampling points is constructed with the outlier as the center. The effective data points within the window are assigned weights according to the principle that the closer they are to the outlier, the greater the weight. The weights are 0.1, 0.3, 0.3, and 0.1 respectively. The weighted average is calculated to replace the outlier.

[0023] When using this method, refer to steps 101 to 105: By controlling the synchronous acquisition of data from multiple channels using a synchronous trigger signal and combining it with high-precision timestamp correction technology, the problem of insufficient timing alignment accuracy of multi-channel data is effectively solved. In the preprocessing stage, centralization, normalization, and outlier repair are adopted, which significantly improves the consistency and reliability of the data. By correcting the time offset through a linear interpolation algorithm, strict synchronization of multi-channel data is ensured, laying a high-quality data foundation for subsequent signal separation and optimization processing, thereby improving the accuracy and stability of the overall method.

[0024] Step 2: Construct an initial separation matrix, dynamically adjust the learning rate based on the mutual information value, iteratively optimize the separation matrix through the negative entropy maximization criterion, achieve the initial separation of multi-channel mixed signals, and select source signals of the target frequency band to form a target source signal set; Step two includes the following: Step 201: Based on the multi-channel data matrix X, with dimensions N×M, where N is the total number of sampling points and M is the number of channels, construct an initial M×M separation matrix. Each element in the matrix is ​​generated using a Gaussian random distribution with a mean of 0 and a variance of 1. Orthogonalization is performed during the generation process to ensure the matrix remains intact. Reversible; set an independence judgment threshold. The value ranges from 0.001 to 0.01, and is adjusted based on the correlation of the multi-channel data; the higher the data correlation, the higher the value. Take a smaller value, such as 0.001. The lower the correlation, the take a larger value, such as 0.01; set the maximum number of iterations to be 50 - 200 times, and at the same time initialize the adaptive learning rate whose value range is [0.01, 0.1]; Step 202: Use the K-nearest neighbor method to calculate the mutual information value MI of any two-channel data in the data matrix X. For any two-channel data sets and , and a ≠ b, respectively construct the K-nearest neighbor probability density function, and combine with the mutual information definition formula to calculate the mutual information value, where H is the information entropy, calculated by integrating the probability density function; dynamically adjust the learning rate η according to the calculated mutual information value: when MI ≥ M1, it means that the two-channel data is highly correlated, take 0.08 - 0.1 to accelerate the convergence speed; when M2 < MI < M1, it means that the two-channel data is moderately correlated, take 0.04 - 0.08 to balance the convergence speed and stability; when MI ≤ M2, it means that the two-channel data is lowly correlated, take 0.01 - 0.04 to avoid iterative oscillation, M1 > M2, M1 and M2 are preset mutual information thresholds, M1 is defaulted to 0.8, M2 is defaulted to 0.4, and set according to specific situations; Step 203: Iteratively optimize the separation matrix based on the negative entropy maximization criterion: calculate the output signal matrix at the k-th iteration. The matrix multiplication is performed according to the row-by-column rule to obtain M output signal vectors with dimensions of N×1; apply the Sigmoid activation function to the elements in each output signal vector, and after conversion, obtain the activation signal matrix ; calculate the outer product of the activation signal matrix and the transpose of the output signal matrix , T represents the transpose, and the expectation is obtained by averaging N sampling points; substitute the expected result into the natural gradient descent update formula , where I is the M×M identity matrix to ensure the invertibility of the separation matrix during the iteration process. After each update, is normalized; Step 204: Calculate the Frobenius norm of the difference between the (k + 1)-th iteration separation matrix and the k-th separation matrix . The Frobenius norm is calculated according to the formula , is the element of matrix A; if the Frobenius norm is less than the preset threshold or the number of iterations reaches the maximum number of iterations Immediately stop the iteration and save the current optimized separation matrix W; use W to perform a linear transformation Y=W×X on the data matrix X to obtain an N×M source signal matrix Y, where each column of the matrix corresponds to an independent source signal, thus achieving the initial separation of multi-channel mixed signals; Step 205: Calculate each source signal in the source signal matrix Y using the Welch method. The power spectral density (PSD). ,Will The dataset is divided into L segments, each with 512 sampling points. The overlap between adjacent segments is set to 50%. A Hanning window is applied to each segment to suppress spectral leakage. The time-domain signal is converted to the frequency-domain signal using a Fast Fourier Transform (FFT). The PSD curve is obtained by averaging the frequency domain signals of all segments. The frequency range [F_low, F_high] corresponding to the main peak frequency is extracted from the PSD curve and compared with the preset target signal frequency band to determine the overlap band width. The target signal frequency band is set according to the specific application scenario; for example, in mechanical equipment vibration monitoring, the target fault signal frequency band is set to 50~500Hz. The ratio of the overlap band width to the target band width is calculated as the matching degree. Source signals with a matching degree ≥ 85% are selected to form the target source signal set. Source signals with a matching degree of <85% are identified as interference source signals and eliminated.

[0025] When using this method, refer to steps 201 to 205: By constructing an initial separation matrix and dynamically adjusting the learning rate based on mutual information values, and iteratively optimizing the separation matrix using the negative entropy maximization criterion, the separation efficiency and accuracy of multi-channel mixed signals are significantly improved. The dynamic learning rate mechanism adaptively adjusts according to signal correlation, which not only accelerates the convergence speed of highly correlated signals but also ensures the stability of low-correlation signals. By using the Welch method to screen target frequency band signals, interference sources are effectively eliminated, providing a high-quality set of source signals for subsequent mode decomposition and fusion, thereby enhancing the robustness and adaptability of the overall method.

[0026] Step 3: Perform modal decomposition on the target source signal, extract extreme points and fit the envelope, and obtain the effective modal component set by iteratively screening the modal components that meet the preset conditions; Step three includes the following: Step 301: Analyze the target source signal set The target source signal contains M channels, each with a length of N. The number of modal components K is determined by the energy proportion method. Fourier transform is performed on the target source signal of each channel to calculate the energy proportion of different frequency components. The number of frequency components corresponding to the cumulative energy proportion reaching 95% is the initial value of K. Step 302: Traverse the target source signal set For signals from all channels, extreme points are extracted using the sliding window method. The window length is set to 1 / 10 of the sampling rate f (e.g., 100 sampling points when f = 1kHz), and the window step size is set to 1 / 20 of the sampling rate f (e.g., 50 sampling points when f = 1kHz). For the signal within each window, local maxima and local minima are determined by comparing the amplitudes of adjacent sampling points. Local maxima are defined as having an amplitude greater than their left and right adjacent points, and local minima are defined as having an amplitude less than their left and right adjacent points. After summing the extreme points from all windows, density clustering is used to remove duplicates. The clustering radius is set to 5 times the sampling interval (e.g., 5ms when the sampling interval is 1ms). Extreme points with a distance less than this radius are merged into one point, ultimately yielding a global joint maxima set and a joint minima set. Step 303: Fit the joint maximum point set using cubic spline interpolation to obtain the upper envelope e_max(t) shared by all channels, where t is time, with the time axis as the x-axis and the maximum point amplitude as the y-axis; similarly, fit the joint minimum point set to obtain the lower envelope e_min(t); calculate the mean line m(t) = (e_max(t) + e_min(t)) / 2, where m is the target source signal for each channel. Subtracting m(t) yields the intermediate signal. Repeat the iterative process of "extracting extreme points → fitting upper and lower envelopes → calculating the mean line → obtaining intermediate signals" until the k-th intermediate signal is obtained. The IMF conditions are met: the difference between the number of extreme points and the number of zero crossings is ≤1, and the absolute value of the mean of the upper and lower envelopes at any time is ≤ `max` represents the maximum value. This is the first IMF component of the channel. , where i represents the i-th channel; Step 304: Transfer the target source signal Subtract IMF component Obtain the residual signal The residual signal As a new input signal, repeat steps 302-303 to sequentially filter out the second IMF component, the third IMF component, and so on, until the Kth IMF component. ; Determine the residual signal after each iteration Is it a monotonic signal or a constant? Check if the residual signal has only one extreme point or no extreme points. If the residual signals of all channels meet this condition, stop the decomposition; otherwise, continue the iteration. The residual signal obtained after K mode decompositions of the i-th channel; Step 305: Calculate the IMF component and corresponding target source signal in each channel. The Pearson correlation coefficient is used to select IMF components with Pearson correlation coefficients ≥ the correlation threshold to form the effective modal component set of the channel. Low-correlation modal components with Pearson correlation coefficients < the correlation threshold are removed and identified as noise-dominated components. The above operation is repeated for all channels to finally obtain M effective modal component sets. The correlation threshold is 0.7 by default and can be set according to specific circumstances.

[0027] When using this method, refer to steps 301 to 305: By accurately extracting extreme points using the sliding window method and density clustering, and fitting the envelope using cubic spline interpolation, high-precision mode decomposition of the target source signal was achieved. By iteratively screening mode components that meet the IMF condition, key feature components in the signal were effectively separated. Furthermore, Pearson correlation coefficient was used to eliminate low-correlation components dominated by noise, significantly improving the accuracy and reliability of mode decomposition. The final set of effective mode components provides high-quality input for subsequent cross-channel mode classification and fusion, enhancing the robustness and adaptability of the overall method.

[0028] Step 4: Based on cosine similarity, classify the modal components into cross-channel shared modes or channel-specific modes, assign weights according to the signal-to-noise ratio and perform weighted fusion, evaluate the quality of the fused signal and output the result.

[0029] Step four includes the following: Step 401: For the set of M valid modal components, calculate the cosine similarity between IMF components of the same order in any two channels. The closer the value is to 1, the more similar the features of the two modal components are. Construct an M×M similarity matrix S based on the cosine similarity of the same order modes in all channels. Classify the modal components in the matrix with cosine similarity ≥ similarity threshold into the cross-channel shared modal set, and classify the modal components with cosine similarity < similarity threshold into the channel-specific modal set. The default similarity threshold is 0.8, which can be set according to the specific situation. Step 402: Calculate the signal-to-noise ratio (SNR) for each mode component in the cross-channel shared mode set and the channel-specific mode set. SNR = 10lg(signal power / noise power). Then, use the residual signal... Determine the noise power, noise power = Calculate the signal power of the modal components. Signal power = p is the modal order. Assign basic weights based on SNR values ,satisfy For modal components in the cross-channel shared modal set, their final weights l; For the modal components in the channel-specific modal set, the final weights ; Step 403: For each channel i, sum up all components in its set of effective modal components with the final weights to obtain the fusion signal of this channel . Meanwhile, calculate the arithmetic mean of the fusion signals of all channels as the global fusion signal, obtaining the set of single-channel fusion signals and the global fusion signal. The set of single-channel fusion signals is the set of fusion signals of M channels; Step 404: Use three indicators to evaluate the quality of the fusion signal. The three indicators include signal-to-noise ratio SNR, root mean square error RMSE, and peak signal-to-noise ratio PSNR. The calculation method of the signal-to-noise ratio SNR is the same as in Step 402, and it is required that SNR≥25dB; the root mean square error RMSE is calculated by the formula , and it is required that RMSE≤0.05; the peak signal-to-noise ratio PSNR is calculated according to the formula , and it is required that PSNR≥30dB. If all three indicators are met, it is determined that the fusion signal is qualified; otherwise, when SNR does not meet the standard, return to Step 202 to adjust the learning rate ; Step 405: Store the qualified set of single-channel fusion signals, the global fusion signal, and the key parameters in the processing process, including the sampling rate f, the separation matrix W, the number of modal components K, and the SNR values of each channel, in the database. Meanwhile, push the fusion signal to the terminal monitoring platform in real time.

[0030] When in use, combine the content of Step 401 to Step 405: Accurately distinguish cross-channel common modes and channel-specific modes through cosine similarity, and dynamically allocate weights based on the signal-to-noise ratio, significantly improving the accuracy and stability of the fusion signal. Adopt a weighted fusion strategy to optimize and integrate effective modal components, ensuring that high signal-to-noise ratio components obtain higher weights. At the same time, strictly evaluate the fusion quality through three indicators of SNR, RMSE, and PSNR. When not meeting the standard, adaptively adjust the learning rate to optimize the separation process. The finally output single-channel and global fusion signals have both high signal-to-noise ratio and low error, and are suitable for multi-source data collaborative analysis in complex scenarios, enhancing the practicability and reliability of the method.

[0031] In the application, several formulas involved are calculated by taking their numerical values after dimensionless, and the formula is obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The coefficients in the formula are set by those skilled in the art according to the actual situation.

[0032] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, and combinations thereof. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0033] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0034] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for multi-channel synchronous data acquisition and optimization processing, characterized in that: It includes: Controlling the multi-channel sensor to synchronously collect data through a synchronous trigger signal, preprocessing the collected multi-channel data, and generating a multi-channel data matrix; Constructing an initial separation matrix, dynamically adjusting the learning rate based on the mutual information value, iteratively optimizing the separation matrix through the negative entropy maximization criterion, achieving the preliminary separation of multi-channel mixed signals, and screening the source signals in the target frequency band to form a target source signal set; Performing modal decomposition on the target source signal, extracting extreme points and fitting the envelope line, and iteratively screening the modal components that meet the preset conditions to obtain an effective modal component set; Classifying the modal components into cross-channel common modes or channel-exclusive modes based on cosine similarity, assigning weights according to the signal-to-noise ratio and performing weighted fusion, and outputting the result after evaluating the quality of the fused signal.

2. The method for multi-channel data synchronous acquisition and optimization processing according to claim 1, characterized in that: The multi-channel data synchronous acquisition is realized by building an acquisition system including M sensors, a synchronous trigger module and a high-speed data cache module, where M≥2 and is determined according to the monitoring target dimension; the sampling rate f is set at 2 to 5 times the highest frequency of the target signal, and the synchronous trigger module outputs a homologous trigger signal; the start timestamp of each channel is extracted through the built-in clock chip of the synchronous trigger module, and based on the start timestamp of the first channel, the linear interpolation algorithm is used to correct the time offset of other channels.

3. The method for multi-channel data synchronous acquisition and optimization processing according to claim 2, characterized in that: The data preprocessing includes centering, normalization and outlier repair. Centering subtracts the mean of all sampling points in the channel from the data of each sampling point; Normalization linearly maps all data to the interval [-1, 1]; the 3σ criterion is used to identify outliers, and for outliers or missing data points, they are repaired by the sliding window weighted average method.

4. The method for multi-channel data synchronous acquisition and optimization processing according to claim 1, characterized in that: The initial separation matrix is M×M dimension, and the matrix elements are generated by Gaussian random distribution with a mean of 0 and a variance of 1 and are orthogonally processed; the independence judgment threshold is set to 0.001~0.01, the maximum number of iterations is 50~200 times, and the adaptive learning rate value range is [0.01, 0.1].

5. The method for multi-channel data synchronous acquisition and optimization processing according to claim 4, characterized in that: The K-nearest neighbor method is used to calculate the mutual information value MI of any two-channel data, and the learning rate η is dynamically adjusted based on MI. When MI≥M1, η takes 0.

6. The method for multi-channel data synchronous acquisition and optimization processing according to claim 5, characterized in that: ​ 7. The method for multi-channel data synchronous acquisition and optimization processing according to claim 1, characterized in that: The signals of all channels in the target source signal set are traversed, and the extreme points, including local maxima and local minima, are extracted using the sliding window method to obtain the global joint maxima set and joint minima set; The joint maximum and joint minimum point sets are fitted by cubic spline interpolation to obtain the upper and lower envelopes. The mean line is obtained by averaging them. The mean line is obtained by subtracting the target source signal of each channel from the mean line. Repeat the iteration until the intermediate signal satisfies the IMF condition: the difference between the number of extreme points and the number of zero crossings is ≤1, and the absolute value of the mean of the upper and lower envelopes at any time is ≤1% of the maximum value of the absolute value of the intermediate signal. The intermediate signal at this time is the IMF component. Subtract the IMF component from the target source signal to obtain the residual signal. Use the residual signal as the new input signal and repeat the iteration to obtain K IMF components. Stop the decomposition when the residual signal has only one extreme point or no extreme point; otherwise, continue the iteration.

8. The method for multi-channel data synchronous acquisition and optimization processing according to claim 7, characterized in that: Calculate the Pearson correlation coefficient between each IMF component in each channel and the corresponding target source signal; select IMF components with Pearson correlation coefficient ≥ correlation threshold to form the effective modal component set of the channel, and remove low-correlation modal components with Pearson correlation coefficient < correlation threshold; finally, obtain M effective modal component sets.

9. The method for multi-channel data synchronous acquisition and optimization processing according to claim 1, characterized in that: Calculate the cosine similarity between IMF components of the same order in any two channels. Modal components with cosine similarity ≥ similarity threshold are classified as cross-channel shared modes, and those with cosine similarity < similarity threshold are classified as channel-specific modes. Calculate the signal-to-noise ratio (SNR) of each modal component, and assign the final weights of cross-channel shared modes and channel-specific modes based on the SNR.

10. The method for multi-channel data synchronous acquisition and optimization processing according to claim 9, characterized in that: The fused signal for each channel is obtained by weighting and summing the effective modal components according to their final weights. The global fused signal is the arithmetic mean of the fused signals of all channels. The quality of the fused signal is evaluated by three metrics, including signal-to-noise ratio (SNR), root mean square error (RMSE), and peak signal-to-noise ratio (PSNR). If the metric is not met, the learning rate is adjusted.

Images