Telemetry data analysis method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention discloses a telemetry data analysis method and device, electronic equipment and a storage medium, and relates to the technical field of aerospace telemetry data analysis, and the method comprises the steps: obtaining telemetry data of a carrier rocket, and carrying out the processing, classification and marking of the telemetry data to obtain a data set; designing a feature extraction network according to a long short-term memory (LSTM) network model, and constructing a telemetry data analysis model in combination with a graph fusion network designed according to a graph neural network model GNN; performing iterative training on the telemetry data analysis model according to the set training parameters, and performing time sequence feature extraction, fault correlation feature extraction, feature fusion and comprehensive analysis on to-be-analyzed data by adopting the trained model to obtain a prediction result; and performing comparative analysis, anomaly detection and fault prediction on the prediction result according to the actual measurement data to obtain a fault diagnosis result, and displaying the fault diagnosis result. The problems that in the prior art, massive data are difficult to process, and potential relations are difficult to mine are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of space telemetry data analysis, and in particular to a telemetry data analysis method and device, electronic equipment and a storage medium. BACKGROUND

[0002] With the rapid development of space technology, the development and launch of carrier rockets are increasingly heavy, and the amount of telemetry data generated during ground testing and flight is growing explosively. These telemetry data not only contain key information such as rocket attitude information and system working parameters, but also are an important basis for task decision-making, fault diagnosis and system performance evaluation.

[0003] However, with the increasing number of carrier rocket models, telemetry data presents significant characteristics such as large data volume, diverse data types and complex data relationships, which pose unprecedented high requirements for telemetry data analysis and comparison. Traditional telemetry data analysis methods often rely on manual rule-based interpretation and simple threshold interpretation statistical analysis. These methods are not up to the task when dealing with current massive, complex and strongly correlated telemetry data.

[0004] In addition, manual interpretation is not only inefficient, but also susceptible to subjective factors, making it difficult to ensure the accuracy and timeliness of data analysis; simple threshold interpretation cannot effectively mine the potential relationships between data, making it difficult to cope with complex and variable fault scenarios. Traditional methods often lack effective data classification and feature extraction methods when faced with large-scale telemetry data, resulting in low data analysis efficiency and inability to provide strong support for task decision-making.

[0005] Therefore, there is an urgent need for a telemetry data analysis method that can handle different types of space missions and complex and variable operating conditions, and improve the accuracy and efficiency of data analysis. SUMMARY

[0006] The telemetry data analysis method provided by the embodiments of the present application solves the problem that the prior art cannot handle massive complex data, has low efficiency, poor accuracy and cannot mine the potential relationships between data. The technical solution is as follows:

[0007] According to one aspect of the present application, a telemetry data analysis method, the method comprising: acquiring telemetry data of a launch vehicle, performing frame count continuity checking, data splicing, normalization processing on the telemetry data, and classifying and labeling according to data physical characteristics and historical failure records to obtain a dataset of time series data samples; the dataset includes a training set and a test set; a feature extraction network is designed according to a long short-term memory network model LSTM, and a graph fusion network is designed according to a graph neural network model GNN to construct a telemetry data analysis model; the feature extraction network includes 6 layers of hidden layers, and the graph fusion network includes 3 layers of hidden layers; the telemetry data analysis model is iteratively trained and performance tested according to the set training parameters until the set requirements are met, the trained telemetry data analysis model is used to extract time series features, extract fault correlation features, fuse features, and comprehensively analyze the to-be-analyzed data to obtain a prediction result; the prediction result is compared and analyzed, abnormality is detected, and failure is predicted according to the actual measurement data to obtain a fault diagnosis result, and the fault diagnosis result is displayed; the prediction result includes a fault probability, a normal probability, and a component fault probability.

[0008] In one embodiment, the frame count continuity checking, data splicing, and normalization processing of the telemetry data are achieved by the following steps: removing modulation data and abnormal data, checking the signal error code rate, frame count continuity, and integrity of the telemetry data, and retaining the telemetry data that meets the set requirements in terms of signal error code rate, frame count continuity, and integrity; when multiple devices relay to complete the telemetry task, the telemetry data is sorted and spliced according to the measurement arc segments of each device, and is normalized to the interval [0, 1] or [-1, 1]; the splicing ensures parameter alignment.

[0009] In one embodiment, the classification and labeling according to data physical characteristics and historical failure records to obtain a dataset of time series data samples are achieved by the following steps: classifying the telemetry data according to physical characteristics; the physical characteristics include environmental parameters, electrical system related single machine operating voltage, video images, and control system data; based on historical failure records, the telemetry data is divided into normal data and fault data, different labels are set for the telemetry data according to the classification results to obtain a dataset, and the dataset is divided into a training set and a test set according to the set proportion by using stratified sampling method.

[0010] In one of the embodiments, the 6-layer hidden layer of the feature extraction network comprises 1 deformation layer, 2 LSTM layers, 1 flattening layer, 1 addition layer and 1 fully connected layer; the hidden layer of the graph fusion network comprises 2 GNN layers and 1 multiplication layer; the feature extraction network is used to process long-term dependence in time series data; the graph fusion network is used to mine complex correlation between different parameters.

[0011] In one of the embodiments, the iterative training and performance testing of the telemetry data analysis model on the dataset according to the set training parameters until the set requirements are met are achieved by setting the learning rate to 0.01, the number of iterations to 300, the batch size to 128, the neuron data of the hidden layer of the feature extraction network to 128, using binary cross-entropy as the loss function; using the Adam adaptive learning rate optimization algorithm, dynamically adjusting the hyperparameters using the back propagation algorithm, and adding the sum of squares of model weight parameters as a penalty term in the loss function using the L2 regularization technique; calculating the performance index on the validation set every 5 iterations during the training process, and stopping the training if the performance index on the validation set does not improve for 10 consecutive times, to obtain the trained telemetry data analysis model.

[0012] In one of the embodiments, the trained telemetry data analysis model is used to perform time series feature extraction, fault correlation feature extraction, feature fusion and comprehensive analysis on the data to be analyzed to obtain the prediction result by the following steps: the first layer of the LSTM layer is used to perform preliminary feature extraction and screening on the data to be analyzed, the second layer of the LSTM layer is used to mine the long-term time series dependence and change law of the data to be analyzed to obtain a three-dimensional feature tensor, and the flattening layer is used to convert it into a two-dimensional feature matrix to obtain time series features; the input layer of the graph fusion network is used to regard the data to be analyzed as nodes and the correlation between data as edges to construct a graph structure according to the labels of the data, the GNN layer is used to collect parameter features of other nodes within a set range of each node and weighted sum based on parameter correlation strength to obtain fault correlation features; the multiplication layer is used to perform element-level multiplication of the time series features and the fault correlation features, and element-level addition with auxiliary features to obtain fusion features, the fully connected layer is used to perform nonlinear transformation and comprehensive analysis on the fusion features to obtain key features, and the output layer is used to obtain the prediction result according to the key features.

[0013] In one embodiment, the fault diagnosis result is obtained by comparing and analyzing the prediction result with the actual measurement data, detecting anomalies, and predicting faults through the following steps: comparing the prediction result with the actual measurement data, calculating the deviation between the predicted value and the actual value, setting a deviation threshold, and determining an anomaly when the deviation exceeds the deviation threshold, triggering an alarm, and marking the abnormal time and abnormal parameters; performing feature matching between the abnormal parameters and a pre-established fault mode library, determining the fault type and fault location by combining parameter correlation changes, generating a fault diagnosis report, predicting the fault development path and remaining available time based on historical telemetry data, and generating processing suggestions by combining expert rule base.

[0014] According to one aspect of the present invention, a telemetry data analysis device includes: a data processing module for acquiring telemetry data from a launch vehicle, performing frame count continuity checks, data splicing, and normalization on the telemetry data, and classifying and labeling the data according to its physical characteristics and historical fault records to obtain a dataset of time-series data samples; the dataset includes a training set and a test set; and a model building module for designing a feature extraction network based on a Long Short-Term Memory (LSTM) network model, and constructing a telemetry data analysis model by combining it with a graph fusion network designed based on a Graph Neural Network (GNN) model; the feature extraction network includes 6 hidden layers, and the graph fusion network... The network includes three hidden layers; a model training and application module, used to iteratively train and test the telemetry data analysis model using the dataset according to the set training parameters until the set requirements are met, and to use the trained telemetry data analysis model to perform time-series feature extraction, fault correlation feature extraction, feature fusion and comprehensive analysis on the data to be analyzed to obtain prediction results; a result analysis module, used to compare and analyze the prediction results, detect anomalies and predict faults based on actual measurement data to obtain fault diagnosis results, and to display the fault diagnosis results; the prediction results include fault probability, normal probability and fault probability of each component.

[0015] According to one aspect of the present invention, an electronic device includes at least one processor and at least one memory, wherein computer-readable instructions are stored on the memory; the computer-readable instructions are executed by one or more of the processors to cause the electronic device to implement the telemetry data analysis method as described above.

[0016] According to one aspect of the present invention, a storage medium has computer-readable instructions stored thereon, which are executed by one or more processors to implement the telemetry data analysis method as described above.

[0017] The beneficial effects of the technical solution provided by this invention are:

[0018] In the technical solution, the telemetry data of the launch vehicle is first acquired, frame count continuity check, data splicing and normalization processing are performed, and classification and labeling are performed according to physical characteristics and historical fault records to form a data set. Then, a feature extraction network (LSTM core) containing 6 layers of hidden layers and a graph fusion network (GNN core) containing 3 layers of hidden layers are designed to construct a telemetry data analysis model. The model is iteratively trained using the data set, and a binary classification cross-entropy loss function, an Adam optimization algorithm and an L2 regularization technique are used until the set performance requirement is met. After training, the model can efficiently extract time sequence features and fault correlation features, and output the prediction results of the fault probability, the normal probability and the fault probability of each component through feature fusion and comprehensive analysis. Further, by comparing the prediction results with the actual measurement data, abnormal detection and fault prediction are performed, and a fault diagnosis report is generated by combining fault mode library matching and parameter correlation analysis, and disposal suggestions are provided. The method significantly improves the accuracy and efficiency of telemetry data analysis, and provides strong support for fault diagnosis and task decision of the launch vehicle. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the description of the embodiments of the present application will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0020] Figure 1 is a flowchart of a telemetry data analysis method according to an exemplary embodiment;

[0021] Figure 2 is a block diagram of a telemetry data analysis device according to an exemplary embodiment;

[0022] Figure 3 is a hardware structure diagram of an electronic device according to an exemplary embodiment;

[0023] Figure 4 is a block diagram of an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION

[0024] The embodiments of the present application will be described in detail below, and examples of the embodiments are shown in the drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are only used to explain the present application, and cannot be interpreted as a limitation of the present application.

[0025] Those skilled in the art of the present technology can understand that the singular forms "a", "an" and "the" used herein also include the plural forms unless specifically stated otherwise. It should be further understood that the use of the term "comprising" in the specification of the present disclosure means that the features, integers, steps, operations, elements, and / or components described therein are present, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is "connected" or "coupled" to another element, it can be directly connected or coupled to the other element, or there can be an intermediate element. In addition, "connected" or "coupled" used herein can include wireless connection or wireless coupling. The phrase "and / or" used herein includes all or any one of the associated listed items and all combinations thereof.

[0026] The present application provides a telemetry data analysis method, by constructing an intelligent telemetry data analysis model fused with LSTM and GNN, efficient feature extraction and fault prediction of launch vehicle telemetry data are realized, the problems of low efficiency, poor accuracy and inability to mine potential relationships in processing massive complex data by traditional methods are solved, and the fault diagnosis capability is improved. The telemetry data analysis method is suitable for a telemetry data analysis device, which can be an electronic device. The telemetry data analysis method in the present application can be applied to various scenarios, such as space launch vehicle telemetry data analysis, etc.

[0027] Please refer to Figure 1 The telemetry data analysis method provided by the present application can be applied to an electronic device.

[0028] In the following method embodiment, in order to facilitate description, the execution subject of each step of the method is taken as an example to be described as an electronic device, but this does not constitute a specific limitation.

[0029] As Figure 1 indicated, the method can include the following steps:

[0030] Step 110, acquiring telemetry data of a launch vehicle, performing frame count continuity check, data splicing, normalization processing on the telemetry data, and classifying and labeling according to data physical characteristics and historical fault records to obtain a data set of time series data samples.

[0031] In one possible implementation, the modulation data and abnormal data are removed, the signal error code rate, frame count continuity and integrity of the telemetry data are checked, and the telemetry data satisfying the set requirements in signal error code rate, frame count continuity and integrity are retained; when multiple devices relay to complete the telemetry task, the telemetry data is sorted and spliced according to the measurement arc segments of each device, and is normalized to the interval [0, 1] or [-1, 1], and the parameter alignment is ensured during splicing.

[0032] The dataset includes a training set and a test set.

[0033] In one possible implementation, the telemetry data is classified according to its physical characteristics. Based on historical fault records, the telemetry data is divided into normal data and fault data. Different labels are set for the telemetry data according to the classification results to obtain the dataset. The dataset is then divided into training set and test set according to a set ratio using stratified sampling.

[0034] The physical characteristics include environmental parameters, the operating voltage of individual electrical systems, video images, control system data, etc., which are not specified here.

[0035] Specifically, for the raw telemetry data stored by each telemetry data receiving device, a comprehensive check is performed on the continuity of frame counts during the receiving and storage process of each device. If the frame counts are discontinuous, it indicates that the data may be lost or erroneous, requiring further investigation to determine the cause and remove any abrupt or abnormal data. Simultaneously, the bit error rate (BER) of the data is carefully checked; only data with a BER meeting the set requirements (e.g., below 0.1%) is retained to ensure the integrity and continuity of the input data.

[0036] Furthermore, when multiple telemetry data receiving devices relay the telemetry task, the telemetry data is precisely sorted according to the measurement arc segments assigned to each device. Matching points in the telemetry data are then found and spliced ​​together, ensuring strict alignment of frame counts or parameters representing time points. If measurement arc segments overlap, data from ground telemetry data receiving devices with low bit error rates and continuous frame counts without jumps are prioritized, eliminating scattered segments as much as possible and fusing well-measured data. Subsequently, telemetry data of different ranges and dimensions are normalized to the [0,1] or [-1,1] interval. For example, for temperature data, if its range is -50℃ to 150℃, it is normalized to the [0,1] interval to eliminate dimensional differences between data features and improve the convergence speed and stability of subsequent model training.

[0037] Furthermore, based on the physical characteristics of the data, such as environmental parameters (temperature, pressure, shock, vibration, noise, etc.), the operating voltage of relevant individual units in the electrical system, video images, and the flight attitude and positioning information of the launch vehicle related to the control system, the data is meticulously classified. Simultaneously, based on historical fault records, the data is divided into normal data and fault data, and fault data is tagged with faulty components. A stratified sampling method is used to ensure that the ratio of normal / faulty data is consistent with the total dataset (e.g., normal data accounts for 80% and faulty data accounts for 20%), dividing the dataset into training and test sets, typically allocating 80% of the data to the training set and 20% to the test set.

[0038] In this invention, the embodiments effectively improve data quality through comprehensive data inspection and cleaning; better integrate measurement data in data stitching; and improve model training efficiency by uniformly normalizing data to a specific interval. Data classification and labeling are based on detailed physical characteristics and historical fault records, providing a higher-quality dataset for subsequent model training.

[0039] In the above process, the embodiments of the present invention ensure high-quality input data through rigorous data inspection and cleaning, providing a reliable foundation for subsequent model training and reducing model errors caused by data issues. Data splicing and normalization improve data usability and model training efficiency, enabling the model to converge faster. Detailed data classification and labeling provide a rich and structured dataset for subsequent model training, helping the model learn more accurate features and improving the accuracy of data analysis.

[0040] Step 120: Design a feature extraction network based on the Long Short-Term Memory (LSTM) network model, and construct a telemetry data analysis model by combining it with a graph fusion network designed based on the Graph Neural Network (GNN) model.

[0041] In one possible implementation, the feature extraction network includes 6 hidden layers, and the graph fusion network includes 3 hidden layers. The 6 hidden layers of the feature extraction network include 1 deformable layer, 2 LSTM layers, 1 flattening layer, 1 summing layer, and 1 fully connected layer; the hidden layers of the graph fusion network include 2 GNN layers and 1 multiplication layer. The feature extraction network is used to process long-term dependencies in time series data; the graph fusion network is used to mine complex relationships between different parameters.

[0042] Specifically, a feature extraction network is designed based on the Long Short-Term Memory (LSTM) network model. This network consists of 1 input layer, 6 hidden layers (1 deformable layer, 2 LSTM layers, 1 flattening layer, 1 summing layer, 1 fully connected layer), and 1 fully connected output layer.

[0043] The system consists of several layers. The input layer receives telemetry data after classification, normalization, and fusion, converting the data into a format recognizable by the model. The deformation layer adjusts the dimensions and shape of the input data, converting two-dimensional data into the three-dimensional time-series data format required by the LSTM layer. For example, it converts "4×10" two-dimensional data into a "10×4" three-dimensional time-series format to match the input requirements of the LSTM layer. The first LSTM layer performs preliminary feature extraction on the time-series data, using the Sigmoid activation function to control information flow and the tanh function to generate candidate information, filtering and retaining key short-term time-series features. The second LSTM layer, building on the first layer, further explores the long-term temporal dependencies and variation patterns of parameters at multiple consecutive moments during the launch vehicle's flight. The flattening layer converts the three-dimensional feature tensor output by the LSTM layer into a two-dimensional feature matrix, retaining all extracted time-series feature information. The addition layer adds the flattened time-series features element-wise with other auxiliary features (such as feature codes corresponding to data labels), achieving preliminary fusion of multi-dimensional features. The fully connected layer, through a large number of neuron connections, performs nonlinear transformation and comprehensive analysis on the fusion features output by the additive layer, and selects the key features that have the greatest impact on the data interpretation results.

[0044] Specifically, a graph fusion network is designed based on a graph neural network (GNN) model. The graph fusion network consists of one input layer and three hidden layers (two GNN layers and one multiplication layer). The input layer receives telemetry parameters (such as temperature, pressure, voltage, etc.) categorized by subsystem / professional. Each parameter is treated as a graph node, and the relationships between parameters (such as the correlation between temperature and pressure) are treated as edges, thus constructing a graph-structured data.

[0045] In this process, the first GNN layer collects the one-hop neighbor parameter features (directly related parameters) for each parameter node. These neighbor features are then fused into aggregated features through weighted summation (weights calculated based on parameter association strength), and combined with the node's own features to generate new features. The second GNN layer, building upon the first layer, collects the two-hop neighbor parameter features (neighboring parameters of neighboring parameters), repeating the "aggregation-update" process to uncover deeper parameter associations. The multiplication layer performs element-wise multiplication of the associated features output from the GNN layer with the temporal features output from the feature extraction network, achieving deep fusion of temporal and associated features.

[0046] In this embodiment of the invention, when processing telemetry data, an intelligent joint model of LSTM and GNN neural networks is constructed, which fully utilizes the advantages of LSTM in processing time series data and the ability of GNN in mining parameter correlations. The deformation layer in the feature extraction network realizes automatic adjustment of data dimensions, which greatly improves efficiency and reduces information loss. The graph fusion network constructs graph-structured data, transforming parameter analysis from "isolated analysis" to "correlation analysis," enabling the discovery of more potential fault correlation clues.

[0047] In the above process, the embodiments of the present invention construct complex feature extraction networks and graph fusion networks, enabling the model to simultaneously capture the temporal and correlational features of the data, thereby improving the comprehensiveness and accuracy of data analysis. This design allows the model to better adapt to the strong temporal sequence and complex parameter correlations of launch vehicle telemetry data, providing a more reliable basis for subsequent fault diagnosis.

[0048] Step 130: According to the set training parameters, iteratively train and test the telemetry data analysis model with the dataset until the set requirements are met. Use the trained telemetry data analysis model to perform time series feature extraction, fault correlation feature extraction, feature fusion and comprehensive analysis on the data to be analyzed to obtain the prediction results.

[0049] In one possible implementation, the learning rate is set to 0.01, the number of iterations to 300, the batch size to 128, and the number of neurons in the hidden layer of the feature extraction network to 128. Binary cross-entropy is used as the loss function. The Adam adaptive learning rate optimization algorithm is used, and the backpropagation algorithm is used to dynamically adjust the hyperparameters. L2 regularization is used to add the sum of squares of the model weight parameters as a penalty term to the loss function. During training, the performance index is calculated on the validation set every 5 iterations. If the performance index does not improve on the validation set for 10 consecutive iterations, training is stopped, and a trained telemetry data analysis model is obtained.

[0050] In one possible implementation, the first LSTM layer performs preliminary feature extraction and filtering on the data to be analyzed. The second LSTM layer mines the long-term temporal dependencies and change patterns of the data to be analyzed to obtain a three-dimensional feature tensor. The tensor is then flattened into a two-dimensional feature matrix to obtain temporal features. The input layer of the graph fusion network treats the data to be analyzed as nodes according to the data labels and the relationships between the data as edges to construct a graph structure. The GNN layer collects the parameter features of other nodes within a set range of each node and obtains the fault association features by weighted summation based on the parameter association strength.

[0051] Furthermore, the temporal features and fault association features are multiplied element-wise by the multiplication layer and added element-wise with the auxiliary features to obtain the fused features. The fused features are then subjected to nonlinear transformation and comprehensive analysis by the fully connected layer to obtain the key features. Finally, the prediction results are obtained by the output layer based on the key features.

[0052] Auxiliary features refer to feature types that provide additional information for model analysis, besides time-series features and fault correlation features. These mainly include feature codes corresponding to data labels (such as normal / fault category markers), static attributes of parameters (such as equipment type and measurement unit), or mission context information (such as launch phase identifiers). These features are combined with time-series-correlation fusion features through element-level addition to enhance the model's understanding of the data background and mission scenario, thereby improving the accuracy of fault prediction.

[0053] Specifically, the model is iteratively trained using the training set, which is divided into a training set and a validation set. Performance metrics, such as accuracy and loss function value, are calculated on the validation set every five iterations. If the performance metrics do not improve or even decrease on the validation set for 10 consecutive iterations, training is stopped, and the current optimal model is saved. For example, if the accuracy fluctuates by less than 0.1% and shows no improvement trend over 10 consecutive validation iterations, the model is considered to have reached a good state, and training is stopped to avoid overfitting.

[0054] This invention improves model training efficiency and performance by employing general empirical values ​​for neural network hyperparameters and optimizing them based on the characteristics of telemetry data. Regarding model validation, by calculating performance metrics on the validation set every five iterations and stopping training based on metric changes, the problem of overfitting or underfitting is effectively avoided.

[0055] In the above process, the reasonable training parameter settings and rigorous model validation procedures of this invention ensure the convergence and generalization ability of the model. The introduction of the Adam optimization algorithm and L2 regularization further improves the model training efficiency and stability, enabling the model to better adapt to different datasets and task scenarios, and improving the model's reliability and practicality.

[0056] Step 140: Based on the actual measurement data, compare and analyze the prediction results, detect anomalies, and predict faults to obtain fault diagnosis results, and then display the fault diagnosis results.

[0057] In one possible implementation, the predicted results are compared with the actual measurement data to calculate the deviation between the predicted and actual values. A deviation threshold is set, and when the deviation exceeds the threshold, it is judged as an anomaly, triggering an alarm and marking the abnormal time and abnormal parameters. The abnormal parameters are matched with a pre-established fault mode library for feature matching. The fault type and fault location are determined by combining the changes in parameter correlation, and a fault diagnosis report is generated. Based on historical telemetry data, the fault development path and remaining available time are predicted, and processing suggestions are generated by combining expert rule base.

[0058] The prediction results include failure probability, normal probability, and failure probability of each component, etc., which are not specified here.

[0059] Specifically, the model's predictions are compared point-by-point with the actual measurement data to calculate the deviation between the predicted and actual values. A reasonable deviation threshold is set, such as 5%, and when the deviation exceeds the threshold, it is considered an anomaly. By comparing the predicted and actual measurement values ​​and analyzing changes in the correlation between parameters, anomaly detection of telemetry data is achieved.

[0060] Furthermore, the intermediate layer feature vectors of the model at abnormal moments are extracted and matched with a pre-established fault mode library. Combined with changes in parameter correlations, the fault type and possible fault location are determined. For example, if the temperature parameter is abnormal and its correlation with the pressure parameter changes, and similar feature matches exist in the fault mode library, the fault type can be determined to be related to the temperature-pressure system, and the possible faulty component can be further located. A detailed fault diagnosis report is generated, including information such as fault type, fault location, and possible causes.

[0061] Furthermore, by learning parameter change trends based on historical telemetry data, the model can capture the changing trends and characteristics of various rocket system parameters before a failure occurs. When telemetry data shows characteristics similar to historical failure data or the model detects an abnormal state, it predicts the possible failure type and timing. Combining a pre-established failure mode library and inference rules, it performs failure diagnosis and identifies the cause of the failure. Based on the failure development trend and remaining available time, and in conjunction with an expert rule base, it generates handling suggestions, such as adjusting flight parameters and conducting system checks, providing decision-makers with references for response measures.

[0062] This invention, by comprehensively considering both predicted value deviations and changes in parameter correlations, can more accurately detect anomalies. In fault diagnosis, by extracting feature vectors from the intermediate layer of the model and matching them with a fault mode library, combined with changes in parameter correlations, it can more accurately determine the fault type and location. Trend prediction and handling suggestions can provide decision-makers with more scientific and timely decision-making basis.

[0063] In the above process, the embodiments of the present invention achieve accurate anomaly detection and fault diagnosis of launch vehicle telemetry data through detailed comparative analysis of prediction results and fault feature matching. The provision of trend predictions and handling suggestions further enhances the scientific rigor and timeliness of decision-making, providing strong support for the safe operation of the launch vehicle and reducing mission risks.

[0064] Through the aforementioned process, this invention achieves efficient and intelligent analysis of launch vehicle telemetry data via rigorous data preprocessing and classification, complex model construction, meticulous model training and optimization, and comprehensive prediction result analysis and fault diagnosis. This method significantly improves the accuracy and efficiency of data analysis, providing strong support for launch vehicle fault diagnosis and mission decision-making, and has significant practical application value. Furthermore, compared with existing technologies, this invention features significant improvements and innovations in each step, better adapting to the complex characteristics of aerospace telemetry data and meeting the high demands of aerospace missions for data analysis.

[0065] In one exemplary embodiment, the process of processing 1TB of telemetry data (containing 10 million time-series data) of a certain type of launch vehicle is demonstrated by the telemetry data analysis method of the present invention.

[0066] Specifically, it may include the following steps:

[0067] Step S1: Telemetry data acquisition and preprocessing.

[0068] Specifically, three telemetry data receiving devices were selected and deployed in different geographical locations to ensure coverage of the measurement arc along the entire rocket flight path. Parameters including pressure, temperature, flow rate, and rotational speed were collected through the telemetry data receiving devices. The acquisition frequency was set to 10Hz (data was collected once every 0.1 seconds), and the acquisition time was approximately 10 minutes, resulting in a total of 6,000 data points.

[0069] Furthermore, the frame count continuity of the raw data stored by each telemetry data receiving device is checked, and data with high frame count continuity is prioritized. When multiple devices are collecting data in relay, the telemetry data is sorted according to the measurement arc segment, and matching points are found for editing and splicing, ensuring that the frame count or the parameter representing the time of measurement is aligned during splicing. In the case of overlapping measurement arc segments, data with low bit error rate and continuous frame count without jump points are prioritized, scattered segments are removed, and data with intact measurements are merged.

[0070] Subsequently, the Min-Max normalization algorithm was used to normalize the data to the [0,1] interval, where the minimum and maximum values ​​are the extreme values ​​of the historical normal data for each parameter. Finally, the data was classified into "normal data" and "faulty data," and labels for faulty components were added to the faulty data to generate a labeled dataset file.

[0071] In the above process, the embodiments of the present invention ensure high-quality input data through rigorous frame count continuity checks and bit error rate filtering. Simultaneously, automated splicing and normalization processes improve the efficiency and accuracy of data preprocessing, providing a reliable data foundation for subsequent model training.

[0072] Step S2: Dataset partitioning and feature extraction.

[0073] Specifically, approximately 200 abrupt and outlier data entries were removed, retaining 5800 valid data entries. A stratified sampling method was used to ensure the normal / faulty data ratio remained consistent with the total dataset (normal data comprised 80% of the total dataset, and faulty data comprised 20%). 80% of the data (4640 entries) was allocated to the training dataset, and 20% (1160 entries) to the test dataset. The test dataset was stored separately and used only for final model evaluation, not for training. Of the 4640 training data entries, they were divided in a 7:3 ratio into a training set (3248 entries) and a validation set (1392 entries) for hyperparameter optimization and overfitting monitoring during model training.

[0074] Furthermore, the feature extraction network input layer receives time-series data samples of pressure, temperature, flow rate, and rotational speed, with an input dimension of 4 (number of parameters) × 10 (time-series window length) and a data type of float32. The deformation layer converts the input "4×10" two-dimensional data into a "10×4" three-dimensional time-series format, achieving dimension transformation through the view function.

[0075] Furthermore, the first LSTM layer extracts short-term temporal features, with 64 hidden neurons and an activation function of Sigmoid (gated) + tanh (candidate features), with a dropout probability of 0.2. The second LSTM layer extracts long-term temporal features, with 128 hidden neurons and the same activation function of Sigmoid + tanh, with a dropout probability of 0.2. The flattening layer preserves the 128-dimensional vector output from the second LSTM layer, directly passing the LSTM features.

[0076] In the above process, the embodiments of the present invention ensure the consistency of data distribution between the training set and the test set through stratified sampling, thereby improving the accuracy of model evaluation. The design of deformable layers and LSTM layers in the feature extraction network enables the model to effectively capture the short-term and long-term temporal features of the data, providing rich feature information for subsequent correlation analysis and fault diagnosis.

[0077] Step S3: Graph structure construction and association analysis.

[0078] Specifically, the graph fusion network input layer constructs an undirected graph based on four parameters (pressure, temperature, flow rate, and rotational speed), with edge weights calculated based on parameter correlations from historical data. The number of nodes is four core parameters, the number of edges is six (there is a correlation edge between every two parameters), and the node feature dimension is 128 (consistent with the output dimension of the LSTM layer).

[0079] Furthermore, a `torch_geometric.data.Data` object (x: 4×128, edge_index: 2×6, edge_weight: 6×1) is constructed. Both the first and second GNN layers use a weighted summation aggregation method, outputting node feature dimensions of 128. The multiplication layer performs element-wise multiplication between the associated features output by the GNN layers and the temporal features output by the feature extraction network, achieving deep fusion of temporal and associated features.

[0080] Furthermore, the summation layer element-wise adds the flattened temporal features to the faulty component label features. The input feature extraction network uses 128-dimensional features plus 4-dimensional parameter label encoding, and the output dimension is set to 128. The fully connected layer filters features crucial for fault prediction, reduces the feature dimension, sets the number of neurons to 64, uses the ReLU+Softmax activation function, and sets the dropout probability to 0.3. The final output layer outputs the fault prediction result in the temporal dimension, with 2 neurons (fault / normal), the Softmax activation function, and an output dimension of n+2 dimensions (fault probability, normal probability, and fault probability of n components).

[0081] In the above process, this embodiment of the invention explicitly expresses the correlation between parameters by constructing graph-structured data, enabling the model to uncover complex correlations between parameters. The introduction of GNN layers allows the model to identify nonlinear correlations, and the correlation analysis depth reaches two layers, enabling the discovery of more potential fault correlation clues. The design of multiplication and addition layers achieves deep fusion of temporal features and correlation features, improving the accuracy of fault prediction.

[0082] Step S4: Model training and optimization.

[0083] Specifically, the hyperparameters are set as follows: learning rate = 0.01, number of iterations = 300, batch size = 128; LSTM hidden layer neurons = 128, GNN layer number = 2; loss function = binary cross-entropy loss function, optimizer = Adam (lr = 0.01, weight_decay = 1e-4, L2 regularization).

[0084] Further, the training and validation sets are loaded. During the training phase, the model is trained by inputting the training set data, forward propagation is used to calculate the prediction results, the loss function is calculated, and backpropagation updates the model parameters. During the validation phase, validation set data is input into the model every 5 iterations, and the validation set loss and accuracy are calculated. If the validation set accuracy shows no improvement or even decreases after 10 consecutive iterations, `torch.save(model.state_dict(), 'best_model.pth')` is called to save the optimal model.

[0085] In the above process, the embodiments of the present invention ensure the convergence and generalization ability of the model through reasonable hyperparameter settings and a rigorous model validation process. The introduction of the Adam optimization algorithm and L2 regularization technology further improves the model training efficiency and stability, avoiding overfitting and underfitting problems.

[0086] Step S5: Output and application of results.

[0087] Specifically, the same preprocessing procedure as the training data is performed on the 1160 test data. The test data is input, the forward propagation is used to calculate the output result (n+2-dimensional probability vector), and the category with the highest probability is taken as the final prediction result.

[0088] Furthermore, a fault probability threshold is set. When the fault probability output by the model exceeds the threshold, an alarm is triggered. Simultaneously, the deviation between the predicted and actual measured values ​​is calculated. If the deviation exceeds 5%, it is marked as a high-risk anomaly. When a high-risk anomaly alarm is triggered, the system automatically recommends countermeasures, and decision-makers formulate a handling plan based on the recommendations. Combining the pre-established fault mode library, the system compares the fault component probabilities output by the model with the rules in the fault mode library to output the cause of the fault. The system analyzes the flight phases with high energy consumption in the test data, uses telemetry data to pinpoint the cause, proposes optimization suggestions, and provides feedback to the design team. Finally, visual charts such as line graphs, network diagrams, and pie charts are generated, and the visual charts, statistical results, and fault cause analysis are integrated into a PDF analysis report, which is sent to designers and mission decision-makers.

[0089] Specifically, under the same hardware environment (CPU: Intel Xeon Gold 6330, GPU: NVIDIA A100), the comparison data of the embodiments of the present invention with traditional manual interpretation, single CNN model, single LSTM model, and single GNN model in terms of total processing time, fault identification accuracy, fault root cause location accuracy, and false alarm rate are shown in Table 1 below.

[0090] Table 1 Comparison test results of the embodiments of the present invention and the prior art

[0091]

[0092] As shown in Table 1, the fusion model proposed in this embodiment of the invention, through automated splicing and normalization algorithms in the data preprocessing stage, has a total processing time of only 9 hours, which is 10 times more efficient than traditional manual methods and nearly 1 times more efficient than traditional machine learning models, enabling efficient processing of massive telemetry data. Because the fusion model simultaneously captures temporal features (LSTM) and parameter association features (GNN), it improves fault identification accuracy by 30% compared to traditional methods and by 13%-17% compared to single models; fault root cause location accuracy is improved by 58% compared to traditional methods and by 28%-33% compared to single models; and the false alarm rate is reduced by 17% compared to traditional methods and by 10%-12% compared to single models. Its ability to accurately identify faults and locate root causes is significantly superior to other models.

[0093] Through the above process, this embodiment of the invention, taking the telemetry data analysis of a certain type of launch vehicle as an example, elaborates on the specific implementation steps of the telemetry data analysis method based on an intelligent model. Through rigorous data acquisition and preprocessing, scientific dataset partitioning and feature extraction, complex graph structure construction and correlation analysis, meticulous model training and optimization, and comprehensive result output and application, efficient and intelligent analysis of launch vehicle telemetry data is achieved. This method significantly improves the accuracy and efficiency of data analysis, provides strong support for launch vehicle fault diagnosis and mission decision-making, and has significant practical application value.

[0094] The following are embodiments of the apparatus of the present invention, which can be used to execute the telemetry data analysis method involved in the present invention. For details not disclosed in the embodiments of the apparatus of the present invention, please refer to the method embodiments of the telemetry data analysis method involved in the present invention.

[0095] Please see Figure 2 This invention provides a telemetry data analysis device 800.

[0096] The telemetry data analysis device 800 includes, but is not limited to: a data processing module 810, a model building module 830, a model training and application module 850, and a result analysis module 870.

[0097] The data processing module 810 is used to acquire telemetry data of the launch vehicle, perform frame count continuity checks, data splicing, and normalization on the telemetry data, and classify and label the data according to the physical characteristics of the data and historical fault records to obtain a dataset of time-series data samples; the dataset includes a training set and a test set.

[0098] The model building module 830 is used to design a feature extraction network based on the Long Short-Term Memory (LSTM) network model and combine it with a graph fusion network designed based on the Graph Neural Network (GNN) model to build a telemetry data analysis model; the feature extraction network includes 6 hidden layers and the graph fusion network includes 3 hidden layers.

[0099] The model training and application module 850 is used to iteratively train and test the telemetry data analysis model using a dataset according to the set training parameters until the set requirements are met. The trained telemetry data analysis model is then used to extract time-series features, extract fault correlation features, fuse features, and perform comprehensive analysis on the data to be analyzed to obtain prediction results.

[0100] The results analysis module 870 is used to compare and analyze the prediction results based on the actual measurement data, detect anomalies and predict faults to obtain fault diagnosis results, and display the fault diagnosis results; the prediction results include the fault probability, normal probability and the fault probability of each component.

[0101] It should be noted that the telemetry data analysis provided in the above embodiments is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed. That is, the internal structure of the telemetry data analysis device will be divided into different functional modules to complete all or part of the functions described above.

[0102] Furthermore, the telemetry data analysis device and telemetry data analysis method provided in the above embodiments belong to the same concept, and the specific way in which each module performs its operation has been described in detail in the method embodiments, and will not be repeated here.

[0103] Figure 3 A schematic diagram of the structure of an electronic device according to an exemplary embodiment is shown.

[0104] It should be noted that this electronic device is merely an example adapted to the present invention and should not be construed as providing any limitation on the scope of use of the present invention. Furthermore, this electronic device should not be interpreted as requiring or depending on having... Figure 3 One or more components of the exemplary electronic device 2000 shown.

[0105] The hardware structure of electronic devices 2000 can vary significantly due to differences in configuration or performance, such as... Figure 3 As shown, the electronic device 2000 includes: a power supply 210, an interface 230, at least one memory 250, and at least one central processing unit (CPU) 270.

[0106] Specifically, power supply 210 is used to provide operating voltage for various hardware devices on electronic device 2000.

[0107] Interface 230 includes at least one wired or wireless network interface 231 for interacting with external devices. Of course, in other examples adapted to this invention, interface 230 may further include at least one serial-to-parallel conversion interface 233, at least one input / output interface 235, and at least one USB interface 237, etc. Figure 3 As shown, this does not constitute a specific limitation.

[0108] The memory 250 serves as a carrier for resource storage and can be a read-only memory, random access memory, disk, or optical disk, etc. The resources stored on it include the operating system 251, application programs 253, and data 255, etc., and the storage method can be temporary storage or permanent storage.

[0109] The operating system 251 is used to manage and control the various hardware devices and application programs 253 on the electronic device 2000, so as to enable the central processing unit 270 to perform calculations and processing on the massive data 255 in the memory 250. It can be Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0110] Application 253 is a computer-readable instruction based on operating system 251 that performs at least one specific task, and may include at least one module ( Figure 3 (Not shown), each module may contain computer-readable instructions for electronic device 2000. For example, the telemetry data analysis device may be considered as application program 253 deployed on electronic device 2000.

[0111] Data 255 may be signal information, etc., and is stored in memory 250.

[0112] The central processing unit 270 may include one or more processors and is configured to communicate with the memory 250 via at least one communication bus to read computer-readable instructions stored in the memory 250, thereby performing operations and processing on the massive amounts of data 255 stored in the memory 250. For example, a telemetry data analysis method may be implemented by the central processing unit 270 reading a series of computer-readable instructions stored in the memory 250.

[0113] Furthermore, the present invention can also be implemented through hardware circuits or a combination of hardware circuits and software. Therefore, the implementation of the present invention is not limited to any specific hardware circuit, software, or combination thereof.

[0114] Please see Figure 4 This invention provides an electronic device 4000, which may include: a desktop computer, a laptop computer, a server, etc., with sensor recognition capabilities.

[0115] existFigure 4 In this context, the electronic device 4000 includes at least one processor 4001 and at least one memory 4003.

[0116] The data interaction between the processor 4001 and the memory 4003 can be achieved through at least one communication bus 4002. This communication bus 4002 may include a path for transmitting data between the processor 4001 and the memory 4003. The communication bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The communication bus 4002 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0117] Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 4004 is not limited to one type, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of the present invention.

[0118] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 4001 may also be a combination that implements computational functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0119] The memory 4003 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program instructions or code in the form of instructions or data structures and accessible by the electronic device 4000, but not limited thereto.

[0120] The memory 4003 stores computer-readable instructions, and the processor 4001 can read the computer-readable instructions stored in the memory 4003 through the communication bus 4002.

[0121] The computer-readable instructions are executed by one or more processors 4001 to implement the telemetry data analysis methods in the above embodiments.

[0122] Furthermore, this embodiment of the invention provides a storage medium storing computer-readable instructions, which are executed by one or more processors to implement the telemetry data analysis method described above.

[0123] This invention provides a computer program product, which includes computer-readable instructions stored in a storage medium. One or more processors of an electronic device read the computer-readable instructions from the storage medium, load and execute the computer-readable instructions, thereby enabling the electronic device to implement the telemetry data analysis method as described above.

[0124] Compared with related technologies, the beneficial effects of the present invention are:

[0125] 1. This invention improves analysis efficiency and adaptability; by automatically learning complex patterns and rules in telemetry data through a neural network intelligent model, this method exhibits strong adaptability and generalization ability. The intelligent model can be trained and adjusted according to different aerospace scenarios and mission requirements. Whether it is a different type of launch vehicle or a different stage of aerospace mission, effective data analysis can be achieved by adjusting model parameters and training data.

[0126] 2. This invention possesses the ability to efficiently process complex data. Addressing the characteristics of aerospace telemetry data—large volume, diverse types, and complex relationships—the neural network intelligent model, through its complex network structure and powerful nonlinear fitting capabilities, effectively processes complex data and uncovers potential relationships between data points. This improves the accuracy and efficiency of data analysis, providing strong support for fault diagnosis and system optimization. Especially in complex interpretation scenarios such as frequency anomalies and frame drops, its performance is significantly improved compared to traditional methods.

[0127] 3. This invention provides a basis for performance optimization; based on data analysis results, this invention can evaluate the performance of space missions and propose targeted optimization suggestions. This helps improve the overall performance and efficiency of space systems and reduce mission costs. For example, by analyzing telemetry data, it is possible to identify high-energy-consumption phases during rocket flight and then propose optimization design suggestions.

[0128] 4. This invention lowers the interpretation threshold; the method does not rely on the experience and expertise of professional interpreters, but rather uses a data-driven approach to analysis, reducing the professional requirements for interpreters. This allows more people to participate in telemetry data analysis, while reducing subjective interference from human factors on the interpretation results, thus improving the objectivity and accuracy of data analysis.

[0129] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0130] The above description is only a partial embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method of telemetry data analysis, characterized by, The method comprises: Obtaining telemetry data of a carrier rocket, performing frame count continuity checking, data splicing, normalization processing on the telemetry data, and classifying and labeling a dataset of time series data samples according to data physical characteristics and historical fault records; the dataset comprises a training set and a test set; Designing a feature extraction network according to a long short-term memory network model LSTM, and constructing a telemetry data analysis model in combination with a graph fusion network designed according to a graph neural network model GNN; the feature extraction network comprises 6 hidden layers, and the graph fusion network comprises 3 hidden layers; Iteratively training and performance testing the telemetry data analysis model through the dataset according to set training parameters until a set requirement is reached, and using the trained telemetry data analysis model to perform time series feature extraction, fault correlation feature extraction, feature fusion and comprehensive analysis on to-be-analyzed data to obtain a prediction result; Comparatively analyzing, abnormally detecting and fault predicting the prediction result according to actual measurement data to obtain a fault diagnosis result, and displaying the fault diagnosis result; the prediction result comprises a fault probability, a normal probability and a component fault probability.

2. The method of telematics data analysis of claim 1, wherein, The frame count continuity checking, data splicing and normalization processing on the telemetry data comprise: Removing modulated data and abnormal data, checking signal error code rate, frame count continuity and integrity of the telemetry data, and retaining the telemetry data that meets a set requirement in terms of the signal error code rate, frame count continuity and integrity; When multiple devices relay to complete a telemetry task, sorting and splicing the telemetry data according to measurement arc segments of the devices, and normalizing the telemetry data to the interval [0, 1] or [-1, 1]; the splicing is performed to ensure parameter alignment.

3. The method of telematics data analysis of claim 1, wherein, The classifying and labeling of the dataset of time series data samples according to data physical characteristics and historical fault records comprises: Classifying the telemetry data according to physical characteristics; the physical characteristics comprise environmental parameters, electrical system related single machine operating voltage, video images and control system data; Based on historical fault records, the telemetry data is divided into normal data and fault data, different labels are set for the telemetry data according to the classification result to obtain a dataset, and the dataset is divided into a training set and a test set at a set proportion by using stratified sampling.

4. The method of telematics data analysis of claim 1, wherein, The 6 hidden layers of the feature extraction network comprise 1 deformation layer, 2 LSTM layers, 1 flattening layer, 1 addition layer and 1 fully connected layer; the hidden layers of the graph fusion network comprise 2 GNN layers and 1 multiplication layer; the feature extraction network is used to process long-term dependence in time series data; and the graph fusion network is used to mine complex correlation between different parameters.

5. The method of telematics data analysis of claim 1, wherein, The iteratively training and performance testing of the telemetry data analysis model through the dataset according to the set training parameters until the set requirement is reached comprise: Setting a learning rate to 0.01, setting an iteration number to 300, setting a batch size to 128, setting neuron data of a hidden layer of the feature extraction network to 128, and using binary cross-entropy as a loss function; An Adam adaptive learning rate optimization algorithm is used, a back propagation algorithm is used to dynamically adjust hyperparameters, and an L2 regularization technique is used to add the sum of squares of model weight parameters in the loss function as a penalty term. During the training process, the performance indicator is calculated on the validation set every 5 iterations, and if the performance indicator on the validation set does not improve for 10 consecutive times, the training is stopped, and the trained telemetry data analysis model is obtained.

6. The method of telematics data analysis of claim 4, wherein, The trained telemetry data analysis model is used to extract time sequence features, fault correlation features, fuse features, and comprehensively analyze the to-be-analyzed data to obtain a prediction result, including: The first layer of the LSTM layer is used to preliminarily extract and screen features from the to-be-analyzed data, the second layer of the LSTM layer is used to mine long-term time sequence dependence and change rules of the to-be-analyzed data to obtain a three-dimensional feature tensor, and the flattening layer is used to convert the three-dimensional feature tensor into a two-dimensional feature matrix to obtain time sequence features; The input layer of the graph fusion network is used to regard the to-be-analyzed data as nodes according to the labels of the data, and the association relationship between the data is regarded as an edge to construct a graph structure, the GNN layer is used to collect parameter features of other nodes within a set range of each node and weightedly sum the parameter features based on parameter correlation strength to obtain fault correlation features; The multiplication layer is used to element-level multiply the time sequence features and the fault correlation features, and element-level add auxiliary features to obtain fused features, the fully connected layer is used to nonlinearly transform and comprehensively analyze the fused features to obtain key features, and the output layer is used to obtain a prediction result according to the key features.

7. The method of telemetry data analysis of claim 1, wherein, The prediction result is compared with the actual measurement data, the deviation between the predicted value and the actual value is calculated, and the deviation threshold is set. When the deviation exceeds the deviation threshold, it is determined to be abnormal, an alarm is triggered, and the abnormal time and abnormal parameters are marked. The abnormal parameters are matched with the pre-established fault mode library, the fault type and fault location are determined based on the change of the parameter correlation relationship, a fault diagnosis report is generated, the fault development path and the remaining available time are predicted based on the historical telemetry data, and the processing suggestions are generated based on the expert rule library. The device comprises:

8. A telemetric data analysis device, characterized by A data processing module is configured to acquire telemetry data of a launch vehicle, perform frame count continuity checking, data splicing, and normalization processing on the telemetry data, and classify and label the data according to physical characteristics and historical fault records to obtain a dataset of time sequence data samples; the dataset includes a training set and a test set; A model construction module is configured to design a feature extraction network according to a long short-term memory network model (LSTM) and construct a telemetry data analysis model in combination with a graph fusion network designed according to a graph neural network model (GNN); the feature extraction network includes six hidden layers, and the graph fusion network includes three hidden layers. ​ A model training and application module is configured to iteratively train and test the performance of the telemetry data analysis model by using the data set according to the set training parameters until a set requirement is reached, and to use the trained telemetry data analysis model to perform time series feature extraction, fault correlation feature extraction, feature fusion and comprehensive analysis on to-be-analyzed data to obtain a prediction result. A result analysis module is configured to perform comparative analysis, abnormality detection and fault prediction on the prediction result according to actual measurement data to obtain a fault diagnosis result, and to display the fault diagnosis result; the prediction result includes a fault probability, a normal probability and a component fault probability.

9. An electronic device, comprising: Comprise: At least one processor and at least one memory, The memory has computer readable instructions stored thereon; The computer readable instructions are executed by one or more processors to enable the electronic device to implement the telemetry data analysis method of any one of claims 1 to 7.

10. A storage medium having stored thereon computer readable instructions, characterized in that, The computer readable instructions are executed by one or more processors to implement the telemetry data analysis method of any one of claims 1 to 7.

Images