Data processing method based on telemetry data, computer device and storage medium

Abstract

The application relates to the technical field of telemetry data processing, and specifically discloses a data processing method based on telemetry data, computer equipment and a storage medium. The method comprises the following steps: receiving a telemetry data stream sent by remote equipment and decoding to obtain multiple physical parameter sequences; inputting the physical parameter sequences in parallel to a graph convolution network and a bidirectional long short-term memory network to extract spatial topology correlation features of the physical parameters and time axis long-range dependence features of the physical parameters; fusing the spatial topology correlation features and the time axis long-range dependence features to generate a fusion feature vector; inputting the fusion feature vector into a dynamic threshold evaluator based on a variational autoencoder to reconstruct a telemetry feature distribution, and determining a deviation degree between the fusion feature vector of the original input and the reconstructed telemetry feature distribution to generate an anomaly score; and warning potential risk positions of the remote equipment according to the anomaly score. Thus, the false negative rate of hidden faults is effectively reduced.

Description

Technical Field

[0001] This application relates to the field of telemetry data processing technology, and in particular to a data processing method, computer equipment and storage medium based on telemetry data. Background Technology

[0002] With the rapid development of aerospace, deep space exploration, and monitoring technologies for large and complex industrial facilities, telemetry technology has become a crucial technology for sensing the operational status of remote equipment, assessing system health, and supporting ground-based decision-making. Telemetry data carries the physical parameters of equipment operation and is a prerequisite for ensuring the safe and stable operation of complex systems. Currently, the interpretation of telemetry data mainly relies on preset static threshold discrimination methods supplemented by human expert experience. That is, for each key technical indicator, technicians typically pre-set one or more upper and lower threshold levels based on design specifications, ground test data, and historical operational experience. When the real-time transmitted telemetry parameters exceed the preset range or deviate from the preset range, the monitoring system will trigger the corresponding alarm mechanism.

[0003] However, as application scenarios become increasingly complex and performance requirements rise, existing technologies still have some shortcomings. Specifically, due to the extremely high integration of large industrial systems, there are highly complex coupling relationships between various subsystems, and fluctuations in a single parameter are often insufficient to characterize the overall health of the system. Existing technologies treat each telemetry channel as an isolated signal source, ignoring the nonlinear correlations between parameters in both time and spatial dimensions. For example, under certain specific operating conditions, even if all individual parameters are within the preset acceptable threshold range, the coordinated drift of multiple parameters over a specific time series may already indicate a deep-seated systemic failure, posing a risk of missed detections. Summary of the Invention

[0004] Therefore, it is necessary to provide a data processing method, computer equipment, and storage medium based on telemetry data to address the above problems.

[0005] According to a first aspect of the embodiments of this application, a data processing method based on telemetry data is provided, comprising: Receive telemetry data streams sent by remote equipment and decode them to obtain multiple physical parameter sequences; The sequences of physical parameters are input in parallel into a graph convolutional network and a bidirectional long short-term memory network to extract the spatial topological correlation features of each physical parameter and the long-range temporal dependence features of each physical parameter. The spatial topological association features and the long-range time axis dependency features are fused to generate a fused feature vector; The fused feature vector is input into a dynamic threshold evaluator based on a variational autoencoder to reconstruct the telemetry feature distribution and determine the deviation between the original input fused feature vector and the reconstructed telemetry feature distribution to generate anomaly scores. Based on the abnormal score, an alert is issued for potential risk areas of the remote equipment.

[0006] In one embodiment, after the step of decoding to obtain a multi-path physical parameter sequence, the data processing method based on telemetry data further includes: The physical parameter sequences of each path are preprocessed, including adaptive noise reduction, missing value interpolation and completion, and dimension normalization.

[0007] In one embodiment, the adaptive noise reduction process includes: Perform multi-level wavelet decomposition on each of the physical parameter sequences to obtain wavelet coefficients in different frequency bands; The wavelet coefficients corresponding to the noise components are suppressed using a hard threshold function; The signal is reconstructed based on the suppressed wavelet coefficients.

[0008] In one embodiment, after the step of decoding to obtain multiple physical parameter sequences, the data processing method based on telemetry data further includes: assigning a corresponding timestamp to each of the physical parameter sequences; The missing value interpolation and completion includes: Monitor the continuity of timestamps in each of the aforementioned physical parameter sequences; When a discontinuity in timestamps is detected, the Lagrange interpolation polynomial is determined based on the length of the missing interval and the dynamic frequency response characteristics of the corresponding physical parameters. The completed data is generated by combining the calculated value of the Lagrange interpolation polynomial with the trend components of the corresponding physical parameters within the historical sampling period.

[0009] In one embodiment, the step of extracting the spatial topological association features of each physical parameter includes: Each physical parameter sequence is abstracted as a node in a graph structure; Obtain the mutual information entropy matrix between each parameter to construct the edge weight matrix between nodes, where mutual information is used to characterize the statistical dependency depth between parameters; Based on the graph structure and the edge weight matrix, a graph convolution operation is performed using an adjacency matrix and a degree matrix that include self-connections to extract the spatial topological association features of each physical parameter.

[0010] In one embodiment, the step of fusing the spatial topological association features and the long-range time-axis dependency features to generate a fused feature vector includes: Multiple attention heads are computed in parallel using a multi-head attention mechanism. Each attention head learns the weight contribution of different feature vectors through interactive operations of query matrix, key matrix and value matrix. Based on the distribution of weighted contributions, the spatial topological association features and the long-range time axis dependency features are weighted and fused together.

[0011] In one embodiment, the step of determining the deviation between the fused feature vector of the original input and the reconstructed telemetry feature distribution to generate anomaly scores includes: The Mahalanobis distance between the fused feature vector of the original input and the reconstructed telemetry feature distribution is calculated, and the calculated Mahalanobis distance is determined as the anomaly score.

[0012] In one embodiment, the data processing method based on telemetry data further includes: The current operational stage of the remote equipment is identified based on the received telemetry data stream; Extract the feature distribution benchmark corresponding to the current operational stage from the knowledge base to reconstruct the telemetry feature distribution.

[0013] According to a second aspect of the embodiments of this application, a computer device is provided, comprising: At least one processor; The system also includes a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the telemetry data processing method as described above.

[0014] According to a third aspect of the embodiments of this application, a computer-readable storage medium is provided, on which a computer program is stored, wherein when the computer program is executed by a processor, it implements the data processing method based on telemetry data as described above.

[0015] The data processing method based on telemetry data provided in this application uses a parallel architecture of graph convolutional networks and bidirectional long short-term memory networks to effectively extract the spatial topological correlation features and long-range time-axis dependency features between various physical parameters in the telemetry data, and then fuses them to obtain a fused feature vector. A dynamic threshold evaluator based on a variational autoencoder is then used to evaluate the fused feature vector to obtain an anomaly score, and based on the anomaly score, an alarm is triggered for potential risky parts of the controlled remote equipment. This solves the problem in existing technologies that cannot capture the cross-parameter nonlinear coupling laws within complex systems, effectively reducing the false alarm rate of latent faults. Furthermore, by setting a dynamic threshold mechanism based on the reconstruction probability of a variational autoencoder (VAE), replacing the traditional static hard threshold, the system can adapt to parameter baseline drift at different operating stages, effectively reducing the false alarm frequency caused by operating condition switching. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a data processing method based on telemetry data provided in an embodiment of this application; Figure 2 This is a partial flowchart of a data processing method based on telemetry data provided in an embodiment of this application; Figure 3 This is a partial flowchart of a data processing method based on telemetry data provided in an embodiment of this application; Figure 4 A partial flowchart of step S300 in a data processing method based on telemetry data provided in an embodiment of this application; Figure 5 A partial flowchart of step S500 in a data processing method based on telemetry data provided in an embodiment of this application; Figure 6 This is a partial flowchart of step S600 in a data processing method based on telemetry data provided in an embodiment of this application. Detailed Implementation

[0017] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings. Preferred embodiments of this application are shown in the drawings. However, this application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application.

[0018] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "joining," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise expressly limited. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0019] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] As described in the background section, with the rapid development of aerospace, deep space exploration, and large-scale complex industrial facility monitoring technologies, telemetry technology has become a crucial technology for sensing the operational status of remote equipment, assessing system health, and assisting ground-based decision-making. Telemetry data carries the physical parameters of equipment operation and is a prerequisite for ensuring the safe and stable operation of complex systems. Currently, the interpretation of telemetry data mainly relies on preset static threshold discrimination methods supplemented by human expert experience. That is, for each key technical indicator, technicians typically pre-set one or more upper and lower threshold levels based on design specifications, ground test data, and historical operational experience. When the real-time transmitted telemetry parameters exceed the preset range or deviate from the preset range, the monitoring system will trigger the corresponding alarm mechanism.

[0022] However, as application scenarios become increasingly complex and performance requirements rise, existing technologies still have some shortcomings. Specifically, due to the extremely high integration of large industrial systems, there are highly complex coupling relationships between various subsystems, and fluctuations in a single parameter are often insufficient to characterize the overall health of the system. Existing technologies treat each telemetry channel as an isolated signal source, ignoring the nonlinear correlations between parameters in both time and spatial dimensions. For example, under certain specific operating conditions, even if all individual parameters are within the preset acceptable threshold range, the coordinated drift of multiple parameters over a specific time series may already indicate a deep systemic fault, posing a risk of missed detections. Furthermore, with the exponential growth of sampling frequency and telemetry dimensions, the resulting data storage volume is enormous. Traditional interpretation methods often struggle to respond quickly and accurately to transient signals or weak trend evolution due to a lack of deep feature extraction capabilities.

[0023] To address the aforementioned issues, this application provides a data processing method, computer device, and computer-readable storage medium based on telemetry data.

[0024] Reference Figure 1 In one embodiment, a data processing method based on telemetry data is provided, comprising the following steps: Step S100: Receive the telemetry data stream sent by the remote equipment and decode it to obtain a multi-path physical parameter sequence.

[0025] Specifically, the raw Pulse Code Modulation (PCM) signal stream transmitted by remote equipment can be accessed via a physical interface. Then, the integrated hardware encoding logic based on a Field Programmable Gate Array (FPGA) is used to parse the unstructured bit stream into multiple separate physical parameter sequences (hereinafter referred to as physical parameter sequences) in real time. The hardware encoding logic refers to the synchronous execution of frame synchronization code group detection, subframe synchronization identification, and word synchronization verification after receiving the raw bit stream, in order to segment the continuous bit stream into telemetry frame structures with temporal meaning. One physical parameter sequence can include real-time monitored values ​​of physical parameters such as temperature, pressure, voltage, current, angular velocity, and switch status. After decoding multiple physical parameter sequences, each telemetry channel, i.e., each physical parameter sequence, can be assigned a corresponding feature identifier code and timestamp tag.

[0026] Furthermore, multiple parallel physical layer receiving units can be configured, and each physical layer receiving unit can be configured with an independent direct memory access (DMA) channel to directly write the parsed physical parameters into the circular buffer of the system's main memory, thereby ensuring zero data loss during the data reception process.

[0027] Step S300: Input the physical parameter sequences in parallel into a graph convolutional network and a bidirectional long short-term memory network to extract the spatial topological correlation features of each physical parameter and the long-range time-axis dependency features of each physical parameter.

[0028] In this embodiment, a heterogeneous computing AI inference engine can be constructed, which integrates a deep learning computing cluster and a pre-loaded spatiotemporal coupling situation assessment model. The spatiotemporal coupling situation assessment model employs a parallel architecture composed of a graph convolutional network (GCN) and a bidirectional long short-term memory network (Bi-LSTM). After decoding the sequences of physical parameters, the spatiotemporal coupling situation assessment model within the heterogeneous computing AI inference engine can be used to extract the temporal and spatial features of each physical parameter sequence. Specifically, the sequences of physical parameters are input in parallel to the graph convolutional network and the bidirectional long short-term memory network. The graph convolutional network is used to analyze the complex subsystem coupling relationships within the controlled object, thereby analyzing the coupling relationships between each physical parameter, i.e., extracting the spatial topological association features of each physical parameter. The bidirectional long short-term memory network can be used to extract the dependencies of each parameter over a long time span, i.e., extracting the long-range time-axis dependency features of each physical parameter.

[0029] Step S500: Fuse the spatial topological association features with the long-range time axis dependency features to generate a fused feature vector.

[0030] After extracting the spatial topological correlation features and long-range time axis dependence features of each physical parameter, they can be fused to obtain a fused feature vector for subsequent analysis.

[0031] Step S700: Input the fused feature vector into a dynamic threshold evaluator based on a variational autoencoder to reconstruct the telemetry feature distribution and determine the deviation between the original input fused feature vector and the reconstructed telemetry feature distribution to generate anomaly scores.

[0032] To achieve adaptability in telemetry data interpretation logic, this embodiment configures an adaptive dynamic threshold estimator on the backend of the heterogeneous computing AI inference engine. This dynamic threshold estimator abandons the traditional static upper and lower limit comparison method and adopts probability distribution modeling based on variational autoencoder (VAE). The dynamic threshold estimator receives the fused feature vector and reconstructs the input telemetry feature distribution through a generative model. Then, it compares the fused feature vector of the original input with the reconstructed telemetry feature distribution to determine the degree of offset between the two. Finally, it generates an anomaly score, the magnitude of which corresponds to the degree of offset.

[0033] Step S900: Based on the abnormal score, issue an alert for potential risk areas of the remote equipment.

[0034] Ultimately, alerts can be issued to potential risk areas of controlled remote equipment based on anomaly scores. Specifically, different anomaly scores can be used to grade and color-code potential risk areas within the human-computer interaction interface.

[0035] The data processing method based on telemetry data provided in this application uses a parallel architecture of graph convolutional networks and bidirectional long short-term memory networks to effectively extract the spatial topological correlation features and long-range time-axis dependency features between various physical parameters in the telemetry data, and then fuses them to obtain a fused feature vector. A dynamic threshold evaluator based on a variational autoencoder is then used to evaluate the fused feature vector to obtain an anomaly score, and based on the anomaly score, an alarm is triggered for potential risky parts of the controlled remote equipment. This solves the problem in existing technologies that cannot capture the cross-parameter nonlinear coupling laws within complex systems, effectively reducing the false alarm rate of latent faults. Furthermore, by setting a dynamic threshold mechanism based on the reconstruction probability of a variational autoencoder (VAE), replacing the traditional static hard threshold, the system can adapt to parameter baseline drift at different operating stages, effectively reducing the false alarm frequency caused by operating condition switching.

[0036] In one embodiment, after the step of decoding to obtain the multiple physical parameter sequences in step S100, the data processing method based on telemetry data may further include the following steps: Step S200: Preprocess the physical parameter sequences of each path. The preprocessing includes adaptive noise reduction, missing value interpolation and completion, and dimension normalization.

[0037] Furthermore, referring to Figure 2 The adaptive noise reduction process in step S200 may specifically include the following steps: Step S211: Perform multi-level wavelet decomposition on each of the physical parameter sequences to obtain wavelet coefficients for different frequency bands; Step S212: Suppress the wavelet coefficients corresponding to the noise components using a hard threshold function; Step S213: Reconstruct the signal based on the suppressed wavelet coefficients.

[0038] That is, in the first stage of the preprocessing pipeline, adaptive noise reduction processing based on an improved wavelet thresholding algorithm can be performed. In practical applications, firstly, multi-level wavelet decomposition is performed on each physical parameter sequence to select a wavelet basis function suitable for the characteristics of the telemetry signal, thereby obtaining wavelet coefficients in different frequency bands, including detail coefficients and approximation coefficients. Then, based on the noise distribution characteristics, the improved hard thresholding function is used to shrink the high-frequency detail coefficients. Finally, the signal is reconstructed from the suppressed coefficients, and the reconstructed signal function can be expressed as: in, This represents the reconstructed signal; j represents the "scaling degree" of the wavelet function, which determines the frequency / resolution of the signal analysis; k represents the movement of the wavelet function on the time axis, which determines the position of the signal analysis. These are the suppressed wavelet coefficients (i.e., the detail coefficients after threshold filtering). This is a set of basis functions obtained by scaling and translating the mother wavelet ψ(t). This processing method can effectively suppress glitches introduced by sensor drift or electromagnetic interference, while preserving the transient change characteristics in telemetry parameters that reflect the true physical state.

[0039] In one embodiment, after the step of decoding to obtain multiple physical parameter sequences in step S100, the data processing method based on telemetry data provided in this embodiment further includes: assigning a corresponding timestamp to each of the physical parameter sequences. Specifically, a high-precision timing reference can be obtained through a hardware timer and globally consistent time coordinates can be assigned to the parsed telemetry parameters, and the original code can be converted into engineering values ​​through a channel mapping table.

[0040] Correspondingly, refer to Figure 3 The missing value interpolation and completion in step S200 may specifically include the following steps: Step S221: Monitor the continuity of timestamps in each of the physical parameter sequences; Step S222: When a discontinuous timestamp is detected, the Lagrange interpolation polynomial is determined based on the length of the missing interval and the dynamic frequency response characteristics of the corresponding physical parameters. Step S223: Combine the calculated value of the Lagrange interpolation polynomial with the trend components of the corresponding physical parameters in the historical sampling period to generate complete data.

[0041] In the second stage of the preprocessing pipeline, a missing value interpolation and completion step can be performed to address packet loss or momentary link interruptions during telemetry data transmission. Specifically, a dynamic interpolation strategy based on the Lagrange interpolation method and the weighted fusion of historical mean trend terms can be employed. When a discontinuity in the timestamps of each physical parameter sequence is detected, the Lagrange interpolation polynomial can be automatically calculated based on the length of the missing interval in the timestamps and the dynamic frequency response characteristics of the corresponding physical parameter. Then, the calculated value of the Lagrange interpolation polynomial is combined with the trend components of the corresponding physical parameter within the historical sampling period to generate completed data. For example, the evolution trend of the channel corresponding to the physical parameter over the past 500 sampling channels can be incorporated. The generated completed data maintains the phase continuity and smoothness of higher-order derivatives of the signal, thereby avoiding misjudgments by the AI ​​model caused by sudden jumps in discrete sampling points. The generated completed data numerically satisfies the following weight allocation: in, The calculated value of the Lagrange interpolation polynomial is given. This refers to the trend component of the corresponding physical parameter within the historical sampling period, also known as the historical mean component. This is a smoothing factor. The above method ensures that the completed signal possesses phase continuity and smoothness in higher-order derivatives, preventing AI inference logic misjudgments caused by sudden jumps in discrete sampling points.

[0042] The final stage of the preprocessing pipeline in step S200 is to perform dimensional normalization, specifically nonlinear dimensional normalization. This can be achieved by using a standardized score (Z-Score) to map engineering values ​​with different dimensions and dynamic ranges to a normal distribution space with a mean of 0 and a variance of 1. The formula used is as follows: ,in: These are the dimensionless, identically distributed new variables obtained after standardization. Represents raw, dimensional engineering data points; The average value of all data for the corresponding variable; This metric measures the dispersion of variable data, representing the magnitude of data fluctuation around the mean. This method can eliminate the weight imbalance in the model caused by the difference in magnitude between voltage signals (volt levels) and temperature signals (Kelvin levels).

[0043] Reference Figure 4 In one embodiment, the step of extracting the spatial topological association features of each physical parameter in step S300 may further include the following steps: Step S311: Abstract each physical parameter sequence into a node in a graph structure; Step S312: Obtain the mutual information entropy matrix between each parameter to construct the edge weight matrix between nodes, where mutual information is used to characterize the statistical dependency depth between parameters; Step S313: Based on the graph structure and the edge weight matrix, perform graph convolution operation using the adjacency matrix and degree matrix containing self-connections to extract the spatial topological association features of each physical parameter.

[0044] Graph convolutional networks can be used to analyze the complex subsystem coupling relationships within a controlled object. Specifically, the telemetry parameters in each sequence of physical parameters can be abstracted as a node V in a graph structure G=(V,E). The weights of the edges E connecting nodes are determined by the mutual information entropy matrix W between the telemetry parameters (i.e., physical parameters). That is, the edge weights (weights of the connecting edges) matrix between nodes can be constructed based on the mutual information entropy matrix between the parameters. Mutual information calculation can reflect the statistical dependency depth between the telemetry parameters, and can be obtained by calculating the sum of the logarithms of the joint probability distribution and the marginal probability distribution between the parameters. The formula can be expressed as: in This represents the degree of coupling / dependence between telemetry parameters; This represents all actual sampled values ​​of parameter X; This represents all actual sampled values ​​of parameter Y; This represents the value of parameter X. And the probability of parameter Y taking the value y simultaneously; for: The marginal probability distribution; Let Y be the marginal probability distribution.

[0045] Next, an adjacency matrix A and a degree matrix D can be dynamically constructed. Based on the graph structure and edge weight matrix, graph convolution operations are performed using adjacency matrix A and degree matrix D. The state evolution law of the graph convolution layer is used to capture the entire situation, and feature aggregation is performed on the entire system state to extract the spatial topological correlation features of each physical parameter. The state evolution law follows a linear transformation based on an adjacency matrix containing self-connections and a nonlinear activation function. The inter-layer transfer formula in the above process is: in, For the first +1 layer of graph convolutional layer output feature matrix; It is a nonlinear activation function that introduces nonlinearity into the model and fits the complex nonlinear subsystem coupling relationships within the controlled object. It is an adjacency matrix that includes self-connections; For adjacency matrix with self-connection The corresponding degree matrix; For the first The input feature matrix of the layer graph convolutional layer; The weight matrix is ​​trainable. This method can not only monitor whether individual telemetry parameters exceed limits, but also identify "system-level anomaly patterns" composed of nonlinear coordinated fluctuations of multiple telemetry parameters, such as the inverse divergence between temperature and pressure.

[0046] In step S300, the step of extracting the long-range time-axis dependency features of each physical parameter employs a bidirectional long short-term memory (LSTM) network, which is responsible for extracting the dependencies of telemetry parameters over a long time span. The bidirectional LSTM network performs a bidirectional scan of the preprocessed physical parameter sequence through two recursive hidden layers, one forward and one backward. The hidden layer output h at each time step... t It combines the processing logic of input gate, forget gate and output gate to effectively capture the rate of change of system state, trend term and periodic component.

[0047] Reference Figure 5 In one embodiment, step S500, namely the step of fusing the spatial topological correlation features and the long-range time-axis dependency features to generate a fused feature vector, may specifically include the following steps: Step S510: Utilize a multi-head attention mechanism to compute multiple attention heads in parallel. Each attention head learns the weight contribution of different feature vectors through interactive operations of the query matrix, key matrix, and value matrix. Step S520: Based on the distribution of weight contribution, the spatial topological association features and the long-range time axis dependency features are weighted and spliced ​​together.

[0048] In this embodiment, a multi-head attention mechanism is introduced at the output of the parallel architecture consisting of a graph convolutional network and a bidirectional long short-term memory network. By computing multiple attention heads in parallel, the weight contribution of different feature vectors is automatically assigned. The specific formula can be expressed as: Where Q, K, and V represent the query matrix, key matrix, and value matrix, respectively; This is the transpose of the key matrix K; For the dimensions of Q and K. This mechanism ensures that the system can focus on the core telemetry channels that contribute the most to the assessment of the current state under complex dynamic conditions.

[0049] Then, based on the distribution of weight contributions, the spatial topological correlation feature vector and the long-range dependence feature vector of the time axis are weighted and spliced ​​together.

[0050] In this embodiment, the offline pre-training process in steps S300 and S500 can optimize the model parameters by minimizing the reconstruction error function, as shown in the following formula: in, These are reconstruction terms used to ensure that the model captures the distribution characteristics under normal operating conditions; The divergence term constrains the latent variable space as a regularization term; These are weighting coefficients used to adjust the importance of the KL divergence term; The model learns the actual distribution of latent variables from the original input X as an approximate posterior distribution. The prior distribution is defined by a standard distribution (usually the standard normal distribution), which constrains the canonical format of the latent variables.

[0051] In one embodiment, the step S700 of determining the deviation between the fused feature vector of the original input and the reconstructed telemetry feature distribution to generate an anomaly score may specifically include: calculating the Mahalanobis distance between the fused feature vector of the original input and the reconstructed telemetry feature distribution, and determining the calculated Mahalanobis distance as the anomaly score.

[0052] That is, in this embodiment, the adaptive dynamic threshold evaluator can evaluate based on the reconstructed probability distribution generated by the variational autoencoder. It reconstructs the telemetry feature distribution of the input through a generative model, calculates the Mahalanobis distance between the fused feature vector of the original input (i.e., the original real-time observation) and the reconstructed distribution, and uses this distance as the anomaly score S. The specific formula can be: in, To enable real-time observation of fused feature vectors; This is the feature mean vector of the controlled object (i.e., remote equipment) under historical healthy operating conditions; superscript Let be the vector transpose symbol, and ∑ be the corresponding covariance matrix. Since the anomaly score S considers the correlation between parameters, the system will automatically trigger an alarm when the system's operating state deviates from the upper limit of the probability confidence interval of the normal distribution. That is, when the anomaly score exceeds the upper limit of the dynamically generated probability confidence interval, an alarm will be automatically triggered.

[0053] In this embodiment, a multi-dimensional situational awareness alarm terminal can be configured, which serves as a human-computer interaction interface and uses a 3D visualization engine (such as Unity3D) to render a digital twin model of the controlled object in real time. The alarm terminal can use color-gradient mapping technology to mark risk areas on the model surface in real time based on the anomaly score level output by the inference engine. For example, when the sensor group shows an abnormal increase in Mahalanobis distance, the system will display a red highlight at the corresponding pump, valve, or pipeline location on the 3D model and display a pop-up window indicating the potential fault mechanism, thereby achieving a leap from abstract data to intuitive perception.

[0054] Reference Figure 6 In one embodiment, the data processing method based on telemetry data provided in this embodiment may further include the following steps: Step S610: Identify the current operating stage of the remote equipment based on the received telemetry data stream; Step S620: Extract the feature distribution benchmark corresponding to the current operating stage from the knowledge base for reconstructing the telemetry feature distribution.

[0055] Specifically, based on the received telemetry data stream and its characteristic distribution patterns, the current operational stage of the controlled object (i.e., remote equipment) can be identified, such as "startup," "steady-state operation," or "shutdown." Then, the adaptive dynamic threshold evaluator retrieves the characteristic distribution benchmark corresponding to the current operational stage from the knowledge base, and reconstructs the telemetry characteristic distribution based on this benchmark. In other words, the corresponding dynamic threshold benchmark can be dynamically adjusted according to the actual operational stage.

[0056] In this embodiment, when an anomaly score exceeds a threshold, a pre-configured knowledge base can be invoked. This knowledge base integrates an ontology-based logic inference engine. The inference engine uses descriptive logic to structurally represent massive historical fault patterns. It can then perform rule-based logical deduction (Rete algorithm) to search for similar fault mechanisms within the vast knowledge space and generate alarm messages containing fault location, type, and handling solutions. The multi-dimensional situational awareness alarm terminal can push alarm messages to the monitoring interface in real time and record expert confirmation of the alarm results. If an expert confirms a false alarm, the sample is stored in a negative sample database. An incremental learning mechanism is used to adjust the model weights in real time, enabling the continuous evolution of the aforementioned judgment logic. Therefore, by integrating the knowledge base and the incremental learning mechanism, the algorithm model can be dynamically optimized, effectively addressing the decision-making risks caused by alarm fatigue among monitoring personnel in environments with massive telemetry data.

[0057] Based on the same inventive concept, in another embodiment, a computer device is provided, including at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described data processing method based on telemetry data.

[0058] The computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores various types of data involved in the data processing method based on telemetry data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements a data processing method based on telemetry data.

[0059] Based on the same inventive concept, in another embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the method described above.

[0060] Based on the same inventive concept, in another embodiment, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the method described above.

[0061] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0062] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0063] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A data processing method based on telemetry data, characterized in that, include: Receive telemetry data streams sent by remote equipment and decode them to obtain multiple physical parameter sequences; The physical parameter sequences are input in parallel into a graph convolutional network and a bidirectional long short-term memory network to extract the spatial topological association features and the long-range time-axis dependency features of each physical parameter. The step of extracting the spatial topological association features of each physical parameter includes abstracting each physical parameter sequence into each node in a graph structure; obtaining the mutual information entropy matrix between each parameter to construct the edge weight matrix between nodes, wherein the mutual information is used to characterize the statistical dependency depth between parameters; and performing graph convolution operation based on the graph structure and the edge weight matrix, using an adjacency matrix and a degree matrix containing self-connections to extract the spatial topological association features of each physical parameter. The spatial topological association features and the long-range time-axis dependency features are fused to generate a fused feature vector. This includes using a multi-head attention mechanism to compute multiple attention heads in parallel. Each attention head learns the weight contribution of different feature vectors through interactive operations of query matrix, key matrix, and value matrix. Based on the distribution of weight contribution, the spatial topological association features and the long-range time-axis dependency features are weighted and fused together. The fused feature vector is input into a dynamic threshold evaluator based on a variational autoencoder to reconstruct the telemetry feature distribution and determine the deviation between the original input fused feature vector and the reconstructed telemetry feature distribution to generate anomaly scores. Based on the abnormal score, an alert is issued for potential risk areas of the remote equipment.

2. The data processing method based on telemetry data according to claim 1, characterized in that, After the step of decoding to obtain multiple physical parameter sequences, the data processing method based on telemetry data further includes: The physical parameter sequences of each path are preprocessed, including adaptive noise reduction, missing value interpolation and completion, and dimension normalization.

3. The data processing method based on telemetry data according to claim 2, characterized in that, The adaptive noise reduction process includes: Perform multi-level wavelet decomposition on each of the physical parameter sequences to obtain wavelet coefficients in different frequency bands; The wavelet coefficients corresponding to the noise components are suppressed using a hard threshold function; The signal is reconstructed based on the suppressed wavelet coefficients.

4. The data processing method based on telemetry data according to claim 2, characterized in that, After the step of decoding to obtain multiple physical parameter sequences, the data processing method based on telemetry data further includes: assigning a corresponding timestamp to each of the physical parameter sequences; The missing value interpolation and completion includes: Monitor the continuity of timestamps in each of the aforementioned physical parameter sequences; When a discontinuity in timestamps is detected, the Lagrange interpolation polynomial is determined based on the length of the missing interval and the dynamic frequency response characteristics of the corresponding physical parameters. The completed data is generated by combining the calculated value of the Lagrange interpolation polynomial with the trend components of the corresponding physical parameters within the historical sampling period.

5. The data processing method based on telemetry data according to claim 1, characterized in that, The step of determining the deviation between the fused feature vector of the original input and the reconstructed telemetry feature distribution to generate anomaly scores includes: The Mahalanobis distance between the fused feature vector of the original input and the reconstructed telemetry feature distribution is calculated, and the calculated Mahalanobis distance is determined as the anomaly score.

6. The data processing method based on telemetry data according to claim 1, characterized in that, The data processing method based on telemetry data also includes: The current operational stage of the remote equipment is identified based on the received telemetry data stream; Extract the feature distribution benchmark corresponding to the current operational stage from the knowledge base to reconstruct the telemetry feature distribution.

7. A computer device, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the data processing method based on telemetry data as described in any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the data processing method based on telemetry data as described in any one of claims 1 to 6.

Images