A method and system for data synchronization processing of a test device

Abstract

The present invention discloses a method and system for data synchronization processing of a test device, specifically relating to the technical field of data synchronization, and is used to solve the problem of insufficient adaptive optimization of data synchronization; the present invention constructs a port data association graph, optimizes the connection relationship between ports based on a time-varying adjacency matrix, extracts port data representations using a graph neural network, optimizes the data synchronization strategy, analyzes the influencing factors of data synchronization errors based on a causal relationship model, identifies the propagation path of data deviation between ports, calculates the data drift amount, uses causal inference to optimize the synchronization window, corrects the data drift, constructs a unified time scale, extracts modal information based on feature encoding, uses a self-attention mechanism to model cross-modal data associations, optimizes feature matching between ports, improves the accuracy of data fusion, realizes high-precision synchronization alignment and fusion of port data, and provides higher-quality data input for data analysis and anomaly detection of test devices.

Description

Technical Field

[0001] The present invention relates to the technical field of data synchronization, and more specifically, to a method and system for data synchronization processing of test equipment. Background Art

[0002] With the wide application of intelligent test equipment, multi-port test systems have become an important means for high-precision data acquisition and analysis, and are widely used in fields such as power system status monitoring, industrial automation detection, wireless communication testing, and intelligent manufacturing. In these application scenarios, multiple test ports usually collect multi-modal data such as voltage, current, spectrum, temperature, vibration, and images simultaneously to comprehensively evaluate the operating state of the system.

[0003] Deficiencies of the prior art: The current multi-port test equipment lacks the ability of adaptive optimization in data synchronization strategies, resulting in unstable data alignment accuracy. In practical applications, due to factors such as data transmission delay, clock drift, different sampling frequencies, and environmental interference between different ports, once the test environment changes, it is difficult to dynamically adjust the synchronization strategy, leading to the gradual accumulation of time alignment errors between data, thereby affecting the accuracy of data fusion and subsequent analysis. In addition, most current synchronization optimization methods rely on simple rules or fixed threshold adjustments, lacking in-depth analysis of data synchronization errors between different ports, resulting in the synchronization effect depending on manual settings, with poor adaptability, and thus affecting the original test process. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method and system for data synchronization processing of test equipment to solve the problem of insufficient adaptive optimization of data synchronization in the above background art.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for data synchronization processing of test equipment, comprising the following steps:

[0007] Construct a port data association graph based on the data of each port of the test equipment, determine the connection relationship between ports based on the time-varying adjacency matrix, extract port data representations using a graph neural network, and optimize the data synchronization strategy;

[0008] Construct a causal relationship model, analyze the influencing factors of data synchronization errors, identify the propagation path of data deviation between ports, calculate the data drift amount, and optimize the synchronization window based on causal inference to correct data drift;

[0009] Perform time alignment on the port data, construct a unified time scale, extract modal key information based on feature encoding, determine cross-modal data associations based on the self-attention mechanism, and optimize feature matching between ports.

[0010] In a preferred embodiment, a port data association graph is constructed based on the data of each port of the test device. The specific process is as follows:

[0011] Construct a port data association graph of the test device, define the port set and edge set of the test device. The edge set E is determined by the correlation between port data, representing the dynamic connection relationship between ports. The edge set is expressed as: , where is the data correlation function of ports i and j at time t, is the correlation threshold;

[0012] Construct a time-varying adjacency matrix A(t), and the expression is: , where controls the correlation decay rate, is the association weight between ports i and j.

[0013] In a preferred embodiment, based on the time-varying adjacency matrix, determine the connection relationship between ports. The specific process is as follows:

[0014] Calculate the time-varying covariance matrix of the ports to determine the correlation between port data:

[0015] The calculation expression of the time-varying covariance matrix is: , where and respectively represent the data of ports i and j within the time window T.

[0016] In a preferred embodiment, use a graph neural network to extract port data representations and optimize the data synchronization strategy. The specific steps are as follows:

[0017] Use a graph neural network to extract port data representations. The initial port data features are , and X(t) represents the raw data collected by the multi-port test device at time t;

[0018] Construct a graph neural network model and use graph convolution for feature update: , where is the node feature of the l-th layer, is the learned weight matrix, is a non-linear activation function;

[0019] Train the graph neural network model using the loss function: , and optimize through gradient descent , and output the port data representation ;

[0020] Calculate the optimal synchronization window: , where D is a data consistency metric function;

[0021] Calculate the synchronization parameters based on the port data representation to minimize data drift, and output the optimized data synchronization strategy for port optimization.

[0022] In a preferred embodiment, a causal relationship model is constructed to analyze the influencing factors of data synchronization errors and identify the propagation path of data deviation between ports. The specific process is as follows:

[0023] Define the set of causal variables: , where is the data flow of port i at time t, is the data flow of port N; is the time synchronization error, representing the clock drift between ports; is the signal interference intensity, representing the signal influence degree between ports; is the device error, representing the systematic deviation of sensors or measurement devices;

[0024] Use the structure learning algorithm based on the probabilistic graphical model to automatically learn the causal relationship and obtain the causal relationship graph. The nodes of the causal relationship graph represent the data and influencing factors of each port, and the edges represent the causal relationship;

[0025] After constructing the causal relationship graph, use transfer entropy to calculate the influence degree of data deviation;

[0026] The expression of transfer entropy is: , where H(X) represents the information entropy of the data flow, represents the influence intensity of the data of port i on the future value of port j, and are the data flows of ports i and j respectively;

[0027] By calculating the transfer entropy between all ports, a data deviation propagation matrix is constructed: , where represents the influence intensity of the data of port k on the future value of port j.

[0028] In a preferred embodiment, calculate the data drift amount and optimize the synchronization window based on causal inference to correct the data drift. The specific process is as follows:

[0029] Determine the propagation path according to the data deviation propagation matrix constructed based on the transfer entropy between ports, and use causal intervention to calculate the data drift amount:

[0030] Take the deviation between the theoretical value and the actual observed value of the port data as the data drift amount;

[0031] Calculate the contribution amount of data deviation based on Bayesian inference. The expression is: , is the posterior probability, representing the data drift amount of port i after the actual observed value X is known. The probability of actual occurrence; is the existence of data drift amount of port i The probability of observing the actual observed value X during the process, is the data drift amount The prior probability of is the total probability of the actual observed value X occurring;

[0032] Define the synchronization error function: , where is the data synchronization window, is the relationship function between the window size and the drift error;

[0033] Use the optimization method to solve the window size with the minimum synchronization error: ;

[0034] Select the smallest synchronization window to minimize the data drift error.

[0035] In a preferred embodiment, time alignment is performed on the port data, a unified time scale is constructed, modal key information is extracted based on feature encoding, cross-modal data association is determined based on the self-attention mechanism, and feature matching between ports is optimized. The specific process is as follows:

[0036] Perform time alignment on the port data, use linear interpolation for the time series signal data; keep the nearest value for the low-frequency data; perform Fourier interpolation on the frequency domain data;

[0037] Use Transformer position encoding to identify data at different time steps and construct a Token sequence;

[0038] Construct a modal feature encoding network, and use independent feature encoders to extract features for different modalities;

[0039] Obtain the feature representation before fusion, calculate the feature vectors of all modalities, use the self-attention mechanism to model the cross-modal relationship, and generate a fusion representation, and optimize the feature matching between ports according to the fusion representation.

[0040] A test device data synchronization processing system for implementing the above-mentioned test device data synchronization processing method, including:

[0041] A strategy determination module for constructing a port data association graph based on the data of each port of the test device, determining the connection relationship between ports based on the time-varying adjacency matrix, extracting port data representations using a graph neural network, and optimizing the data synchronization strategy;

[0042] A data correction analysis module, which is used to build a causal relationship model, analyze the influencing factors of data synchronization errors, identify the propagation path of data deviation between ports, calculate the data drift amount, and optimize the synchronization window based on causal inference to correct data drift;

[0043] A port synchronization optimization module, which is used to perform time alignment on port data, build a unified time scale, extract modal key information based on feature encoding, determine cross-modal data associations based on the self-attention mechanism, and optimize feature matching between ports.

[0044] The technical effects and advantages of the present invention:

[0045] By constructing a port data association graph, the present invention optimizes the connection relationship between ports based on a time-varying adjacency matrix, extracts port data representations using a graph neural network, optimizes the data synchronization strategy, improves the accuracy of data alignment, analyzes the influencing factors of data synchronization errors based on a causal relationship model, identifies the propagation path of data deviation between ports, calculates the data drift amount, uses causal inference to optimize the synchronization window, dynamically corrects data drift, improves data consistency, performs time alignment on port data, builds a unified time scale, extracts modal key information based on feature encoding, uses the self-attention mechanism to model cross-modal data associations, optimizes feature matching between ports, improves the accuracy of data fusion. The present invention can dynamically adjust data synchronization parameters, reduce data drift errors, achieve high-precision synchronization alignment and fusion of port data, improve data consistency, provide high-quality data input for data analysis, anomaly detection, and prediction of test equipment, and enhance the stability and intelligent level of test equipment. Description of the Drawings

[0046] Figure 1 It is a flowchart of a method for data synchronization processing of a test equipment according to the present invention.

[0047] Figure 2 It is a schematic structural diagram of a data synchronization processing system of a test equipment according to the present invention. Detailed Embodiments

[0048] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0049] Embodiment 1: As Figure 1 shown, a method for data synchronization processing of a test equipment includes the following steps:

[0050] Construct a port data association graph based on the data of each port of the test device, determine the connection relationship between ports based on the time-varying adjacency matrix, use a graph neural network to extract port data representations, and optimize the data synchronization strategy;

[0051] Construct a causal relationship model, analyze the influencing factors of data synchronization errors, identify the propagation path of data deviation between ports, calculate the data drift amount, and optimize the synchronization window based on causal inference to correct data drift;

[0052] Perform time alignment on port data, construct a unified time scale, extract modal key information based on feature encoding, determine cross-modal data associations based on the self-attention mechanism, and optimize feature matching between ports.

[0053] During the data synchronization process of a multi-port test device, the measurement data of each port often has complex spatial correlations and time dependencies. These correlations may be caused by factors such as device layout, signal propagation path, correlation of measured physical quantities, and environmental interference. If the correlation relationship between ports is not fully modeled, it may lead to insufficient optimization of the data synchronization strategy and a decrease in synchronization accuracy.

[0054] Step 1, perform multi-port data association modeling based on a graph neural network, and the specific steps are as follows:

[0055] In a test device, multiple ports simultaneously measure different data streams, such as current, voltage, spectrum, etc. There may be certain correlations in the data of these ports. For example, adjacent ports may measure different parts of the same signal, so the data streams of adjacent ports have a high correlation. Due to different signal propagation paths, there may be strong or weak time correlations between some ports, and the associations between ports may be time-varying, that is, the correlations between data in different time periods may change;

[0056] Construct the port data association graph of the test device, and the specific process is as follows:

[0057] Define the port set. Suppose the test device has N ports, and the data stream measured by each port is , then define the port set: , where, represents the i-th port and is a time series signal;

[0058] Define the edge set. The edge set E is determined by the correlation between port data and represents the dynamic connection relationship between ports: , where, is the data correlation function between ports i and j at time t, that is, it represents the correlation between the data streams of port i and port j at time t. Common methods include Pearson correlation coefficient, mutual information, or cross-correlation function; is the correlation threshold;

[0059] The correlation threshold is set to 0.7 or 0.8. Only when the value of the correlation function is greater than the correlation threshold, it is considered that there is a significant association between port i and port j, and thus an edge connection is established during graph construction;

[0060] When constructing the time-varying adjacency matrix A(t), if there is a significant correlation between port i and j, then set as the association weight between port i and j, otherwise set it to zero, that is, define the edge weight as: , where controls the correlation decay rate (e.g., ), which is used to amplify or suppress the influence of correlation on the weight;

[0061] By assigning lower decay values (higher weights) to port pairs with high correlation (larger values), while directly setting low correlation to zero, only the information of strong associations is retained in the graph.

[0062] By calculating the correlation and covariance between ports, the time-varying adjacency matrix can reflect the real-time association degree of signals between ports. Ports with high correlation have larger weights in the graph, meaning that their data changes tend to be consistent.

[0063] After completing the above steps, the port data association graph G(V, E) and the time-varying adjacency matrix A(t) are output.

[0064] Calculate the time-varying covariance matrix of ports. The correlation of multi-port data is not fixed but changes with time. Therefore, a time-varying covariance matrix is constructed to dynamically calculate the correlation (connection relationship) between port data and adjust the data synchronization strategy:

[0065] Let the data of port i and j within the time window T be and , and define the covariance function: , by calculating the covariance of data of different ports, the time-varying covariance matrix is obtained;

[0066] If the time-varying covariance matrix changes significantly, then adjust the adjacency matrix A(t): ;

[0067] By comparing the covariance at adjacent moments, the trend of data correlation changing with time can be captured; when the change is significant, it indicates that there may be a synchronization deviation between the two ports, and the updated lower weight reflects this inconsistency, which is convenient for subsequent detection and compensation.

[0068] The output is the optimized time-varying adjacency matrix, which is used for graph neural network modeling.

[0069] Train a graph neural network (GNN) to extract port data representations;

[0070] Build a GNN model, with the initial port data features set as , where \(X(t)\) represents the raw data collected by a multi-port test device at time \(t\). Feature updates are performed using graph convolutional (GCN): , where is the node feature of the \(l\)-th layer, is the learned weight matrix, is a non-linear activation function (such as ReLU);

[0071] The GNN model is trained using a loss function: , which is optimized by gradient descent to output the port data representation , which is used for data synchronization optimization. The weights of each layer are optimized by gradient descent to minimize the representation differences of ports with high weights (high correlations).

[0072] Based on the port data representations learned by the GNN, the data synchronization strategy can be optimized to dynamically adjust the synchronization window size and improve data synchronization accuracy;

[0073] Calculate the optimal synchronization window: , where \(D\) is a data consistency metric function. Synchronization parameters are calculated based on the port data representation to minimize data drift, and then an optimized data synchronization strategy is output;

[0074] For example, the data synchronization strategy dynamically adjusts the synchronization window (optimizes the synchronization window size). A fixed synchronization window may lead to excessive data alignment errors. If the window is too large, it may introduce irrelevant data; if the window is too small, it may result in missing valid data. Calculating the optimal synchronization window based on the port data representation enables the synchronization window to dynamically adapt to changes in the data stream, improving synchronization accuracy. It can also automatically adjust the timestamps of the data to align the data more precisely and reduce synchronization errors;

[0075] A specific embodiment is as follows: Assume a test device has 4 ports, and the specific data and parameter settings are as follows:

[0076] The port data is set such that each port is a time series of length 100 (e.g., voltage signal), with a sampling frequency of 100 Hz, a correlation threshold of , an attenuation parameter , a time window \(T\) of ten seconds and a step size of one second;

[0077] At a certain moment, the correlations calculated using the Pearson correlation coefficient between each pair of ports are as follows (assuming the data has been normalized). The port correlation values are: , , , , , ;

[0078] According to the correlation threshold analysis, a connection is established only when . Then, for ports 1 and 2: , then ; for ports 1 and 3: ; for ports 1 and 4: ; for ports 2 and 3: , ; for ports 2 and 4 it is 0; for ports 3 and 4: , then ;

[0079] Finally, the time-varying adjacency matrix (symmetric matrix) is obtained: ;

[0080] Perform graph neural network processing. The 100-dimensional time series of each port is used as a row vector to form a 4×100 matrix. Assume that the weight matrix of the first layer is 100×64 (randomly initialized), and the activation function uses ReLU. Then, a 4×64 matrix is obtained, and each row is the representation of the port after graph convolution. Design multiple layers (for example, a total of 2 or 3 layers) of graph convolution, and finally obtain the fused port features . At this time, each row of contains the comprehensive information of the port considering its time-varying association with other ports.

[0081] Based on the data consistency information reflected among the ports, it can guide the adjustment of the synchronization window (for example, select the optimal synchronization window, dynamically compensate the timestamps of each port, align the multi-port data in time, and thus achieve overall data synchronization.

[0082] Through the graph convolution layer, the original data of each port and its dynamic relationship are fused into a unified feature representation. This representation contains both the local information of the single-port signal and captures the global relationship among the ports. The finally obtained port representation can be used to evaluate the synchronization status among the ports - if the representations of highly correlated ports in the port representation are quite different, it indicates that there are data drift or time delay problems. Based on this information, the system can dynamically adjust the synchronization window and the timestamp correction amount to achieve high-precision alignment and synchronization of the data.

[0083] Step 2, perform causal inference-driven data consistency analysis;

[0084] After completing Step 1, the dynamic association graph of the data of each port of the test device has been constructed, and the topological representation of the multi-port data has been extracted through GNN. However, during the multi-port data fusion process, data consistency deviations may still occur, that is, the data of some ports may have problems such as drift, mismatch, and error accumulation. These problems may be caused by various factors such as clock drift, data transmission delay, device error, and signal interference;

[0085] Construct a causal relationship model between the ports of the test device, and define the set of causal variables. That is, assume that the test device has N ports, and the data stream measured by each port is When constructing the causal relationship model, it is necessary to define the set of causal variables: where, is the data stream of port i at time t, is the data stream of port N; is the time synchronization error, representing the synchronization deviation caused by the clock drift between ports; is the signal interference intensity, measuring the signal influence degree between ports; is the device error, representing the systematic deviation or noise level of the sensor or measurement device;

[0086] Construct a causal relationship graph G=(V, E), and use a structure learning algorithm based on the probabilistic graph model (such as the PC algorithm, GIES algorithm) to automatically learn the causal relationship and obtain the causal relationship graph G(V, E). Among them, the node V represents the data of each port and the influencing factors, and the edge E represents the causal relationship, that is, how the data of a certain port affects other ports. For example, how the data change of a certain port is affected by the time error, interference intensity or device error, or how the data abnormality of one port will transmit and affect other ports.

[0087] After constructing the causal relationship graph, further quantify the propagation direction and influence range of the data deviation between ports. For example, if the data abnormality of a certain port will affect multiple other ports, then the data consistency problem of this port is more serious and needs to be optimized first. Therefore, use transfer entropy to calculate the influence degree of data deviation. The specific steps are as follows:

[0088] Calculate the information flow direction between data streams. That is, assume that the data streams of ports i and j are respectively and , and the transfer entropy is defined as: where, represents the information entropy of the data stream, represents the influence intensity of the data of port i on the future value of port j;

[0089] If the uncertainty drops significantly after the data stream is added to port i, it indicates that For has a strong information contribution, that is, there is a strong causal influence;

[0090] By calculating the transfer entropy between all ports, a data deviation propagation matrix is constructed: , this matrix is used to identify the key influencing ports, that is, the ports with the most serious data deviation propagation. The greater the transfer entropy, the more serious the data deviation. For example, represents the relative contribution of port i when affecting the future state of port j. The greater the transfer entropy, the easier it is for the data deviation to be propagated by this port, thus providing a basis for identifying the key influencing ports.

[0091] After identifying the propagation path of the data deviation, quantitatively evaluate the true impact of the data drift for compensation. Use causal intervention to calculate the data drift amount. The specific steps are as follows:

[0092] Define the data drift amount. Let the data drift amount of port i be: , this formula represents the deviation between the theoretical expected value and the actual observed value of the port data under the condition of eliminating the time error ; represents the theoretical expected value under artificial intervention (that is, eliminating the clock drift, making ), is the actually observed expected value;

[0093] Based on Bayesian inference, calculate the true contribution value of the data deviation. The contribution value expression of the data deviation is: , where is the posterior probability, indicating the probability that the data drift amount of port i actually occurs after the actual observed value X is known; is the probability of observing the data X when there is a data drift amount in port i, is the prior probability of the data drift amount , is the total probability of the actual observed value X occurring; Through Bayesian inference, quantitatively analyze the credibility of the drift of each port, and then perform subsequent synchronization compensation.

[0094] The data of different ports may have different optimal alignment windows. If the synchronization window is too large, it may cause irrelevant data to be aligned, increasing the data noise; if the synchronization window is too small, it may cause data loss. Therefore, it is necessary to calculate the optimal synchronization window size to minimize the data drift error;

[0095] Let the data drift amount of port i be , then the synchronization error function can be defined as: , where is the data synchronization window size, is the relationship function between the window size and the drift error (obtained by fitting historical data);

[0096] Using an optimization method, solve for the window size that minimizes the synchronization error: , the selected synchronization window can minimize the data drift error, thus enabling optimal data alignment and synchronization.

[0097] Furthermore, after adjusting the synchronization window, there may still be clock drift in some ports. The optimal time compensation amount can be further calculated to correct the data timestamp, making the data alignment more accurate.

[0098] For example, a test device has 4 ports. The signals collected by each port are voltage data, the sampling rate is 100Hz, and each signal has 1000 sampling points. The specific implementation steps are as follows:

[0099] The voltage time series obtained by the four ports respectively ; The time synchronization error is set to approximately 0.05 seconds, and the signal interference intensity is 0.2, and the device error is 0.01 (after normalization); The candidate range of the synchronization window: from 2 seconds to 10 seconds; The fitting function is a linear relationship ;

[0100] Perform preliminary statistics on the data of each port and related factors, and use the PC algorithm to automatically learn the causal structure to obtain the following example relationship: and are greatly affected by the time synchronization error; and affect each other and are also interfered by the signal interference intensity;

[0101] Calculate the entropy between each port as follows: ; ; ; Other combinations are slightly lower;

[0102] Construct the propagation matrix T as follows: , and so on; It can be identified from the matrix that, for example, the influence of port 2 on port 3 is the greatest, indicating that the data synchronization accuracy of these two ports is particularly critical;

[0103] Use Bayes' formula to calculate the posterior probability of data drift for each port. For example, assuming it is relatively high for port 2, it means that the data drift problem of port 2 is relatively serious and needs to be compensated first;

[0104] According to the aforementioned data drift and fitting function, calculate the synchronization error function. In this example, if w seconds and seconds can both achieve relatively small errors. According to other synchronization requirements (such as data coverage rate, real-time requirements, etc.), seconds can be selected as the optimal synchronization window;

[0105] Data synchronization is achieved through causal inference methods. First, structure learning is used to determine the causal relationships between the data of each port and their influencing factors, and identify which factors (such as time error, signal interference, device error) have the greatest impact on data consistency. Subsequently, through quantitative calculations of transfer entropy and data drift, the propagation path of data deviation and the inconsistencies in data synchronization among ports can be captured. Then, the synchronization error function and optimization methods are used to automatically select the optimal synchronization window and time compensation amount, so as to finely align and compensate the data, achieve the consistency of multi-port data in time and value, and further achieve high-precision and real-time data synchronization.

[0106] Step 3, perform multi-modal data fusion based on cross-modal Transformer;

[0107] After completing the causal inference-driven data consistency analysis in Step 2, the sources of data consistency deviation have been identified, and the data synchronization strategy has been adjusted based on causal inference. However, in the application scenarios of multi-port test equipment, the data may contain different modalities (such as voltage, current, spectrum, image, temperature, displacement, etc.). There are non-linear correlations and temporal dependencies between these modalities, making it difficult to effectively capture the complex relationships between different modality data, resulting in information loss or fusion distortion;

[0108] The feature spaces of different modality data are different. For example, the signals of voltage and current have different physical units numerically and cannot be directly used for numerical operations;

[0109] The time scales of the data are different. For example, some sensors (such as temperature sensors) sample at a low frequency, while other sensors (such as spectrum analyzers) sample at a high frequency, resulting in difficulties in aligning the data in time.

[0110] The data of each port usually have large differences in the time dimension and numerical scale. For example, the data of some ports are high-frequency signals, while the data of some ports are low-frequency measurement values and need to be aligned;

[0111] For time-series signal data (such as voltage and current), linear interpolation is used; for low-frequency data, the nearest value is maintained; for frequency-domain data, Fourier interpolation is performed;

[0112] A Token is a data unit used for Transformer processing. In different tasks and data types, assume the data Token structure at each time step is as follows: , where is the data value after time alignment, representing the standard time points obtained through interpolation or resampling. For example, for high-frequency and low-frequency signals, different interpolation strategies (such as linear interpolation, Fourier interpolation, or keeping the nearest value) are used to ensure that the data is expressed on the same time basis; is the time position encoding, generated using sine and cosine functions; is the modality index, used to distinguish different data modalities (e.g., voltage is 1, current is 2, spectrum is 3, image is 4, etc.). Modality information can be added in the form of an embedding vector to enable subsequent networks to distinguish data from different sources;

[0113] Use Transformer position encoding to identify data at different time steps and construct a Token sequence , where M is the total number of time steps, and then process it.

[0114] The data feature distributions of different modalities are different. Encode the features of the data to extract the key information within the modality;

[0115] Construct a modality feature encoding network, using independent feature encoders to extract features for different modalities: ;

[0116] For example, for time series signals, use a 1D convolutional network (a convolutional neural network for processing one-dimensional sequence data) to extract local time patterns; for image data, use a convolutional neural network (CNN) to extract features, and for spectrum data, use the Fourier transform (FFT) to extract frequency domain features.

[0117] Obtain the feature representation before fusion and calculate the feature vectors of all modalities: , which is used for the Transformer to perform cross-modal information fusion.

[0118] Use the self-attention mechanism to model the cross-modal relationships and generate a fused representation, that is, perform cross-modal fusion on these Tokens and calculate the self-attention weights: , where is the Query vector, is the Key vector, is the normalization factor of the Key dimension, is the query transformation matrix that projects the input features into the query space so that the Query of different data points has a suitable feature representation form; is the key transformation matrix that projects the input features into the key space so that the Key of different data points has a suitable feature representation form;

[0119] The Query key (query vector) represents the information that the current data point wants to ask other data points. The Key (key vector) represents the features of the current data point for other data points to match the query. The Value (value vector) stores the content information of this data point and is used for the final calculation of the weighted output.

[0120] Through weighted sum calculation: , where is the Value vector, is the value transformation matrix.

[0121] Update the cross-modal representation of the port data, and calculate the new fused data representation as: , where H is the original fused data representation and is reconstructed using the decoder g: ; where is the fused data output. Then, according to the differences in the fused representation within different time windows;

[0122] The fused data representation is an optimized feature vector or high-dimensional data representation containing multi-port and multi-modal information. This fused data representation has the characteristics of time alignment, modal feature complementarity, information enhancement, etc.;

[0123] By analyzing, abnormal patterns, mutation points or abnormal relationships between modalities in the data can be detected, improving the accuracy of anomaly detection. For example, in power system monitoring, if the current signal of a certain port is abnormal while the spectrum signal and temperature signal are normal, the fused data representation can reveal the deviations between these modalities and help accurately locate the cause of the fault. Since the fused data representation fuses data from different modalities, it can provide more comprehensive information for the data prediction model and improve the prediction accuracy;

[0124] After tokenizing the data from different ports and different modalities, the self-attention mechanism of the Transformer can automatically capture the non-linear and time-series related relationships between modalities. Each modality data carries the original signal and time position information. During the fusion process, the model can learn which time steps and which modalities have complementary or redundant information, thus forming a unified fused representation. The cross-modal Transformer can fuse the redundant information from different signal sources. Even if there are abnormalities or noises in a certain modality data, it can be compensated by other modality data, overall improving the robustness of the data synchronization process;

[0125] By elaborately designing the specific formulas and parameters of Token composition, positional encoding, linear transformation matrix, attention weight calculation, and synchronization window optimization, an example is given to illustrate how to use cross-modal Transformer to fuse information from different data sources, thereby providing a reliable basis for dynamically adjusting the data synchronization strategy and ultimately achieving high-precision and real-time data synchronization processing.

[0126] In device health monitoring, a prediction model can be trained based on the fused data representation to predict possible abnormal states that the device may exhibit in the future. In an intelligent decision-making system, the fused data can be used to determine whether the current state meets the expectations or to give optimization suggestions.

[0127] In this step, the self-attention mechanism is used to model the non-linear relationship between different modalities, automatically learn the correlation between data, realize information interaction and feature fusion, perform time alignment and Tokenization on multi-modal data, map the data of different ports to a unified time axis, and convert it into a unified Token representation. The modal feature encoding is used to extract the key information of different modalities, and the adaptive attention mechanism is used to calculate the weights between different data points, enabling the Transformer to automatically focus on important features, improve the fusion accuracy, output a cross-modal unified representation, and ensure that the data of all ports can be optimally matched in the high-dimensional feature space, providing high-precision input for subsequent data analysis and synchronization strategies.

[0128] It should be noted that the thresholds involved in the embodiments can be determined according to specific scenarios and requirements.

[0129] The present invention optimizes the connection relationship between ports based on a time-varying adjacency matrix by constructing a port data association graph, extracts port data representations using a graph neural network, optimizes the data synchronization strategy, improves the accuracy of data alignment, analyzes the influencing factors of data synchronization errors based on a causal relationship model, identifies the propagation path of data deviation between ports, calculates the data drift amount, optimizes the synchronization window using causal inference, dynamically corrects the data drift, improves data consistency, performs time alignment on port data, constructs a unified time scale, extracts modal key information based on feature encoding, models cross-modal data associations using the self-attention mechanism, optimizes feature matching between ports, improves the accuracy of data fusion. The present invention can dynamically adjust data synchronization parameters, reduce data drift errors, achieve high-precision synchronization alignment and fusion of port data, improve data consistency, provide high-quality data input for data analysis, anomaly detection, and prediction of test equipment, and enhance the stability and intelligence level of test equipment.

[0130] Embodiment 2: A data synchronization processing system for a test device, as Figure 2 shown, specifically includes:

[0131] A policy determination module, configured to construct a port data association graph based on the data of each port of the test device, determine the connection relationship between ports based on a time-varying adjacency matrix, extract port data representations using a graph neural network, and optimize the data synchronization policy;

[0132] A data correction and analysis module, configured to construct a causal relationship model, analyze the influencing factors of data synchronization errors, identify the propagation path of data deviation between ports, calculate the data drift amount, and optimize the synchronization window based on causal inference to correct data drift;

[0133] A port synchronization optimization module, configured to perform time alignment on port data, construct a unified time scale, extract modal key information based on feature encoding, determine cross-modal data associations based on the self-attention mechanism, and optimize feature matching between ports.

[0134] The above formulas are all dimensionless and take their numerical calculations. Specific dimensionless methods can use various means such as standardization, which will not be elaborated here. The formula is obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0135] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or data center that contains one or more collections of available media. The available media can be magnetic media (such as floppy disks, ATA hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state ATA hard disk.

[0136] It should be understood that in various embodiments of the present application, the magnitudes of the serial numbers of the above processes do not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0137] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0138] In several embodiments provided by this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of devices or units can be in electrical, mechanical, or other forms.

[0139] The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units. They can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0140] In addition, the functional units in each embodiment of this application can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0141] As described above, this is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claimed rights.

Claims

1. A method for synchronously processing test equipment data, characterized in that: The steps include: A port data association graph is constructed based on the data of each port of the test equipment, and the connection relationship between ports is determined based on the time-varying adjacency matrix. The graph neural network is used to extract the port data representation and optimize the data synchronization strategy. Build a causal relationship model, analyze the factors affecting data synchronization errors, identify the propagation path of data deviations between ports, calculate the amount of data drift, and optimize the synchronization window based on causal inference to correct data drift; Time-align the port data, build a unified time scale, extract key modal information based on feature encoding, determine cross-modal data association based on the self-attention mechanism, and optimize feature matching between ports; Construct a causal relationship model, analyze the influencing factors of data synchronization errors, and identify the propagation path of data deviations between ports. The specific process is as follows: Define the set of causal variables: , where X N is the data flow of port N; is the time synchronization error, indicating the clock drift between ports; is the signal interference strength, indicating the degree of signal influence between ports; is the equipment error, which represents the systematic deviation of the sensor or measuring device; Use a structural learning algorithm based on a probabilistic graphical model to automatically learn causal relationships and obtain a causal relationship graph. The nodes of the causal relationship graph represent the data of each port and the influencing factors, and the edges represent the causal relationships. After constructing the causal relationship diagram, transfer entropy is used to calculate the impact of data deviation; The transfer entropy expression is: , where H(X) represents the information entropy of the data stream, Indicates the impact strength of the data of port i on the future value of port j, and are the data flows of ports i and j at time t respectively; By calculating the transfer entropy between all ports, the data deviation propagation matrix is ​​constructed: , where Indicates the impact strength of the data of port k on the future value of port j; Calculate the data drift, optimize the synchronization window based on causal inference, and correct the data drift. The specific process is as follows: The propagation path is determined by the data deviation propagation matrix constructed based on the transfer entropy between ports, and the data drift is calculated using causal intervention: The deviation between the theoretical value and the actual observed value of the port data is taken as the data drift; The contribution of data deviation is calculated based on Bayesian inference, and the expression is: , is the posterior probability, which indicates the data drift of port i after the actual observation value X is known. The probability of it actually happening; is the data drift of port i The probability that the process observes the actual observation value X, is the data drift The prior probability of is the total probability of the actual observed value X occurring; Define the synchronization error function: ,in, It is the data synchronization window. It is a function of the relationship between window size and drift error; Use the optimization method to find the window size with the smallest synchronization error: ,in, is the minimum synchronization window; The smallest synchronization window is chosen to minimize data drift errors.

2. A method for synchronously processing test equipment data according to claim 1, characterized in that: Construct a port data association diagram based on the data of each port of the test equipment. The specific process is as follows: Construct a port data association graph for the test equipment, define the port set and edge set of the test equipment, the edge set E is determined by the correlation between the port data, and represents the dynamic connection relationship between the ports. The edge set is expressed as: ,in, is the data correlation function between ports i and j at time t, is the correlation threshold; Construct the time-varying adjacency matrix A(t), the expression is: ,in, Controls the rate at which correlation decays, is the association weight between ports i and j.

3. A method for synchronously processing test equipment data according to claim 2, characterized in that: And the connection relationship between ports is determined based on the time-varying adjacency matrix. The specific process is as follows: Calculate the time-varying covariance matrix of the ports to determine the correlation between the port data: The calculation expression of the time-varying covariance matrix is: ,in, and Represent the data of port i and j in time window T respectively.

4. A method for synchronously processing test equipment data according to claim 3, characterized in that: Use graph neural network to extract port data representation and optimize data synchronization strategy. The specific steps are as follows: The graph neural network is used to extract the port data representation. The initial port data features are: , X(t) represents the original data collected by the multi-port test equipment at time t; Build a graph neural network model and use graph convolution for feature update: ,in, is the node feature of the lth layer, is the learned weight matrix, is a nonlinear activation function; The loss function used for training the graph neural network model is: , optimized by gradient descent , output port data characterization ; Calculate the optimal synchronization window: , where D is the data consistency measurement function; The synchronization parameters are calculated based on the port data characterization to minimize data drift, and the optimized data synchronization strategy is output for port optimization.

5. A method for synchronously processing test equipment data according to claim 4, characterized in that: Time-align the port data, build a unified time scale, extract modal information based on feature encoding, determine cross-modal data association based on the self-attention mechanism, and optimize feature matching between ports. The specific process is as follows: Time-align the port data and use linear interpolation for the timing signal data; keep the nearest value for the low-frequency data; and perform Fourier interpolation on the frequency domain data; Use Transformer position encoding to identify data at different time steps and construct a Token sequence; Construct a modal feature encoding network and use independent feature encoders to extract features from different modalities; Obtain the feature representation before fusion, calculate the feature vectors of all modalities, use the self-attention mechanism to model the cross-modal relationship, generate a fused representation, and optimize the feature matching between ports based on the fused representation.

6. A test equipment data synchronization processing system, used to implement a test equipment data synchronization processing method according to any one of claims 1 to 5, characterized in that: include: The strategy determination module is used to construct a port data association graph based on the data of each port of the test equipment, determine the connection relationship between ports based on the time-varying adjacency matrix, use the graph neural network to extract the port data representation, and optimize the data synchronization strategy; The data correction analysis module is used to build a causal relationship model, analyze the influencing factors of data synchronization errors, identify the propagation path of data deviations between ports, calculate the data drift, and optimize the synchronization window based on causal inference to correct data drift; The port synchronization optimization module is used to time-align port data, build a unified time scale, extract modal key information based on feature encoding, determine cross-modal data association based on the self-attention mechanism, and optimize feature matching between ports.

Images