An aircraft state data prediction method, device and equipment

Abstract

This application relates to the field of aircraft data processing and discloses a method, apparatus, and device for predicting aircraft state data. The method includes: acquiring historical operational data of the aircraft; performing high-dimensional feature mapping on the historical operational data and injecting temporal position encoding to obtain a feature vector matrix; performing feature correlation transformation on the feature vector matrix to obtain a feature correlation matrix; generating initial prediction data of the aircraft based on a pre-trained state prediction model, according to the feature vector matrix and a specified prediction duration; performing correlation state correction on the initial prediction data according to the feature correlation matrix to obtain corrected prediction data; and performing low-dimensional feature mapping on the corrected prediction data to obtain target state data of the aircraft. The target state data represents the reference operational state of the aircraft in the future within the specified prediction duration. The technical solution provided by this application can improve the accuracy and reliability of aircraft state prediction.

Description

Technical Field

[0001] This application relates to the field of aircraft data processing, and in particular to a method, apparatus and equipment for predicting the state data of an aircraft. Background Technology

[0002] As a commonly used type of modern aircraft, the state data of six-degree-of-freedom aircraft exhibits high nonlinearity and cross-coupling. By accurately predicting the future state of the aircraft, decision-making references can be provided for the trajectory planning and attitude control of the aircraft.

[0003] In related technologies, aircraft state data are typically predicted using physical mechanism-based modeling methods or data-driven deep learning methods. However, in practical applications, the former relies on known parameters and is susceptible to model mismatch due to complex operating conditions, while the latter, although capable of learning and fitting nonlinear characteristics, does not consider the coupling characteristics of state data and long-term state changes, resulting in lower accuracy of the predicted state data.

[0004] Therefore, improving the accuracy and reliability of aircraft status prediction has become a key research focus in the field of aircraft data processing. Summary of the Invention

[0005] This application provides a method, apparatus, and device for predicting the state data of an aircraft, which can improve the accuracy and reliability of aircraft state prediction.

[0006] This application provides a method for predicting the state data of an aircraft. The method includes: acquiring historical operational data of the aircraft, wherein the historical operational data characterizes the state information and control information of the aircraft within a specified historical time period; performing high-dimensional feature mapping on the historical operational data to obtain embedded feature vectors; injecting temporal position encoding into the embedded feature vectors to obtain a feature vector matrix, wherein the feature vector matrix is ​​characterized by a feature dimension and a time dimension; performing feature association transformation on the feature vector matrix to obtain a feature association matrix, wherein the feature association matrix reflects the feature association relationships between feature dimensions and between the time dimension; generating initial prediction data of the aircraft based on a pre-trained state prediction model, according to the feature vector matrix and a specified prediction time period; performing associated state correction on the initial prediction data according to the feature association matrix to obtain corrected prediction data; and performing low-dimensional feature mapping on the corrected prediction data to obtain target state data of the aircraft, wherein the target state data characterizes the future state information of the aircraft within a specified prediction time period.

[0007] In one implementation, the state prediction model is trained based on a simulation dataset generated by the mechanism simulation model; wherein: the mechanism simulation model is characterized by a state-space equation, which is constructed based on an input state vector, an observation output vector, and an excitation vector; wherein: the input state vector is determined based on a differential equation, which is used to characterize the dynamic characteristics of the aircraft; the observation output vector is used to construct the simulation dataset; and the excitation vector is determined based on a transfer function, which is used to characterize different operating conditions of the aircraft.

[0008] In one embodiment, performing feature association transformation on the feature vector matrix to obtain a feature association matrix includes: performing feature association processing on the feature vector matrix based on a preset set of reference matrices to obtain feature association data, wherein the reference categories of each reference matrix in the set of reference matrices are different; and performing feature transformation processing on the feature association data to obtain the feature association matrix, wherein the feature vector matrix and the feature association matrix have the same dimension.

[0009] In one implementation, feature association processing is performed on the feature vector matrix based on a preset set of reference matrices to obtain feature-associated data. This includes: performing a linear transformation on the feature vector matrix based on any reference matrix in the set of reference matrices to obtain a set of feature matrices for the feature vector matrix, wherein the set of feature matrices includes multiple feature matrices corresponding to the reference matrices; for any feature matrix in the set of feature matrices, performing feature weighting on the feature matrix with other feature matrices to obtain dimensional association data corresponding to the feature matrix; and normalizing the dimensional association data corresponding to each feature matrix to use the normalized dimensional association data as feature association data.

[0010] In one embodiment, performing feature transformation processing on the feature-related data to obtain the feature-related matrix includes: expanding the dimensions of the feature-related data to obtain expanded dimension data; performing function mapping on the expanded dimension data; and performing dimension restoration and normalization processing on the expanded dimension data after function mapping to obtain the feature-related matrix.

[0011] In one implementation, generating initial prediction data for the aircraft based on a pre-trained state prediction model, according to the feature vector matrix and a specified prediction duration, includes: inputting the feature vector matrix into the state prediction model to generate state prediction data for a specified prediction duration; and performing feature association processing on the state prediction data based on a preset set of reference matrices to obtain the initial prediction data.

[0012] In one embodiment, correcting the association state of the initial prediction data based on the feature association matrix to obtain corrected prediction data includes: performing cross-association processing on the feature association matrix and the initial prediction data based on a preset set of reference matrices to obtain initial corrected data corresponding to the initial prediction data; performing feature transformation processing on the initial corrected data, and using the initial corrected data after feature transformation processing as the corrected prediction data.

[0013] In one implementation, cross-correlation processing of the feature correlation matrix and the initial prediction data based on a preset reference matrix set includes: determining a first feature matrix set of the feature correlation matrix and a second feature matrix set of the initial prediction data based on the preset reference matrix set; for any second feature matrix in the second feature matrix set, performing correlation feature weighting on the second feature matrix and the first feature matrix set to obtain dimensional correlation data corresponding to the second feature matrix; and normalizing the dimensional correlation data corresponding to each second feature matrix to use the normalized dimensional correlation data as initial correction data.

[0014] A second aspect of this application provides an apparatus for predicting the state data of an aircraft. The apparatus includes: a historical data acquisition unit for acquiring historical operational data of the aircraft, wherein the historical operational data characterizes the state information and control information of the aircraft within a specified historical time period; a historical data processing unit for performing high-dimensional feature mapping on the historical operational data to obtain embedded feature vectors, injecting temporal position encoding into the embedded feature vectors to obtain a feature vector matrix, wherein the feature vector matrix is ​​characterized by feature dimensions and time dimensions, performing feature correlation transformation on the feature vector matrix to obtain a feature correlation matrix, wherein the feature correlation matrix reflects the feature correlation relationships between feature dimensions and between the time dimension; a state data prediction unit for generating initial prediction data of the aircraft based on a pre-trained state prediction model, according to the feature vector matrix and a specified prediction time period, and performing correlation state correction on the initial prediction data according to the feature correlation matrix to obtain corrected prediction data; and a state data processing unit for performing low-dimensional feature mapping on the corrected prediction data to obtain target state data of the aircraft, wherein the target state data characterizes the reference operational state of the aircraft in the future within a specified prediction time period.

[0015] A third aspect of this application provides a computer device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions, the computer device being used to implement the aircraft state data prediction method described in the first aspect.

[0016] The technical solution provided in one or more embodiments of this application, based on the deep and long-term dependencies of the aircraft, infers and corrects the long-term state data of the aircraft to improve the accuracy and reliability of aircraft state prediction. Specifically, high-dimensional feature mapping, temporal position encoding injection, and feature association transformation are performed on historical operating data containing state and control information to obtain a feature association matrix containing deep features, which characterizes the coupling relationship between temporal context information and multi-dimensional features, thereby maintaining stable prediction performance under different complex operating conditions. Furthermore, based on the deep features extracted from the feature association matrix, the initial prediction data of rapid prediction is corrected for associated states, significantly improving the prediction accuracy of aircraft state data. At the same time, through the step-by-step processing of prediction and correction, the generation of prediction bias is effectively avoided in long-term prediction, thereby improving the accuracy and reliability of aircraft state prediction.

[0017] It is evident that the technical solution provided in this application can improve the accuracy and reliability of aircraft state prediction and provide a basis for aircraft control decisions. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 A schematic diagram illustrating the steps of a method for predicting the state data of an aircraft provided in this application. Figure 2 This is a schematic diagram illustrating the steps of performing feature correlation transformation on an eigenvector matrix according to an embodiment of this application; Figure 3 This is a schematic diagram of a feature association transformation method provided in one embodiment of this application; Figure 4 This application provides a method step diagram for correcting the associated state of initial prediction data according to an embodiment of the present application; Figure 5 This is a schematic diagram illustrating a method for cross-association processing provided in one embodiment of this application; Figure 6 A schematic diagram of the structure of an aircraft state data prediction device provided in one embodiment of this application; Figure 7 This is a schematic diagram of the structure of a computer device provided in one embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] Furthermore, the use of terms such as "first," "second," etc., in this application is for descriptive purposes only and should not be construed as indicating or implying their 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 embodiments in this application, unless otherwise stated, "multiple" means two or more. Additionally, the use of "based on" or "according to" implies openness and inclusiveness, because processes, steps, calculations, or other actions "based on" or "according to" one or more of the stated conditions or values ​​may in practice be based on additional conditions or beyond the stated values.

[0022] With the continuous development of the aerospace industry, the prediction of state parameters for six-degree-of-freedom (DOF) aircraft, which can move in six independent directions in three-dimensional space, has attracted widespread attention. The state space of a six-DOF aircraft is typically represented by twelve dimensions, encompassing the motion states of position, velocity, attitude, and angular velocity, thus exhibiting high nonlinearity, time-varying characteristics, and coupling. Specifically, the twelve-dimensional state variables are interdependent; predicting a single variable requires considering the influence of other variables. For example, attitude changes directly affect the velocity direction, while velocity changes affect aerodynamic forces, thereby altering attitude. Furthermore, the aircraft's dynamic parameters change continuously over time, with different flight phases (e.g., takeoff, cruise, maneuvering, and landing) exhibiting different dynamic characteristics, making it difficult to capture parameter changes under various operating conditions. Accurately predicting the future state of the aircraft allows for adjustments to trajectory planning, attitude control, and other decisions, ensuring the safe and stable operation of the aircraft.

[0023] In related technologies, aircraft state data are typically predicted using physical mechanism-based modeling methods or data-driven deep learning methods. Among them, physical mechanism-based modeling methods deduce changes in aircraft state by solving mathematical models constructed based on the aerodynamics and kinematics principles of the aircraft. This approach relies on known aerodynamic parameters, moments of inertia, and external disturbances, and is easily affected by complex operating conditions and environmental disturbances, leading to model mismatch and making it difficult to adapt to highly dynamic and strongly coupled flight scenarios.

[0024] Current mainstream data-driven deep learning methods, while able to fit nonlinear features by learning the implicit mapping relationship from the state space to dynamic parameters, do not fully consider the inherently high coupling characteristics of aircraft state data. Furthermore, in long-term prediction, they struggle to effectively capture the relative positional relationships and long-term dependencies between time steps, easily leading to state drift and error accumulation, resulting in severe attenuation of prediction accuracy and insufficient reliability. In addition, when processing historical operational data, they often only focus on the changing trends of state information, neglecting the integration and utilization of control information. This makes it difficult to fully explore the deep coupling relationship between state and control information, and lacks explicit constraints on the physical laws of the aircraft, easily generating outliers that violate dynamic logic during prediction, thus limiting prediction accuracy.

[0025] In view of the above, this application provides one or more embodiments of an aircraft state data prediction method, apparatus and device, which can solve the above problems, enhance the in-depth mining and dynamic correction of historical operating data, and improve the accuracy and reliability of aircraft state prediction.

[0026] Firstly, please refer to Figure 1 One embodiment of this application provides a method for predicting the state data of an aircraft, which may include the following steps: S1: Obtain historical operational data of the aircraft, wherein the historical operational data characterizes the historical operational status and control information of the aircraft within a specified historical period.

[0027] The aforementioned historical operational data is pre-divided into time steps of a specified historical duration, typically represented as a vector sequence. Since the aircraft's operational state is influenced by time, processing the instantaneous operational data at each time step using vector sequences facilitates accurate control of the changing trends of the instantaneous operational data at each time step, thereby improving the prediction accuracy of subsequent state data. The instantaneous operational data at each time step represents both the aircraft's state information and control information at that corresponding time step. For example, starting from the beginning of the sequence, historical operational data is truncated in a sliding manner at one time step length to form a data sequence of a specified historical duration, which serves as the historical operational data after time step division.

[0028] In a six-degree-of-freedom aircraft, the aforementioned historical operational data is characterized using 15 dimensions. Among these, the state information within the historical operational data is typically characterized using 12 state dimensions, specifically including the state relative to the ground three-dimensional coordinate system during flight. coordinate, coordinates and Coordinates, and the velocity components of the aircraft's velocity vector in the three directions. , and and the roll angle of the aircraft Pitch angle and yaw angle and the rates of rotation around the coordinate axes at the three angles mentioned above. , , Historical operational data typically represents control information in three dimensions: the control parameters of the aircraft relative to each of the three velocity components. , and .

[0029] S3: Perform high-dimensional feature mapping on the historical running data to obtain embedded feature vectors, and inject temporal position encoding into the embedded feature vectors to obtain a feature vector matrix, wherein the feature vector matrix is ​​characterized by feature dimension and time dimension.

[0030] The aforementioned feature vector matrix is ​​a high-dimensional feature representation of historical running data with temporal position encoding. Specifically, the historical running data is mapped to the instantaneous running data of each time step in a temporal order using high-dimensional feature mapping, and the resulting embedded feature vectors are injected with temporal position encoding to obtain a feature vector matrix characterized by the feature dimension and the time dimension. The aforementioned temporal position encoding is used to determine the temporal arrangement among the embedded feature vectors. For example, high-dimensional feature mapping can be performed using convolutional neural networks, recurrent neural networks, or graph neural networks, or it can be performed using a model pre-trained based on simulation data. Optionally, the aforementioned feature vector matrix is ​​represented in the form of a two-dimensional matrix, that is, the embedded feature vectors of all time steps after high-dimensional feature mapping are stacked in chronological order after being injected with temporal position encoding to form a two-dimensional matrix with the time dimension and the feature dimension as the dimensional directions.

[0031] Since historical operational data consists of low-dimensional physical quantities with different dimensions, mapping it to a high-dimensional feature space facilitates the learning of richer feature representations across different dimensions. For example, for the instantaneous operational data at each time step, a linear transformation can be used to map it to a preset feature dimension, such as mapping 15-dimensional historical operational data to 256 dimensions. Because there are complex interactions between state data at different time steps, injecting temporal position codes into the embedded feature vectors corresponding to each mapped time step can improve the detection accuracy and temporal reliability of the state data. For instance, generating a temporal position code for each time step and concatenating it with the mapped embedded feature vectors can distinguish historical operational data from different time steps.

[0032] S5: Perform feature correlation transformation on the feature vector matrix to obtain a feature correlation matrix, which is used to reflect the feature correlation relationship between feature dimensions and between time dimensions; The aforementioned feature correlation matrix is ​​a feature representation of the feature vector matrix after feature correlation transformation. Its feature dimensions are the same as the feature vector matrix, but the feature vectors at each time step contain the contextual information of the entire matrix. There are complex dependencies between the various feature dimensions of the aforementioned feature vector matrix, and complex interactions exist in the time dimension (specifically between different time steps). For example, the current velocity of the aircraft is not only related to the current thrust but also closely related to the velocity change trend over a previous period. Therefore, it is necessary to perform feature correlation transformation on the feature vector matrix to capture feature correlations from both the feature dimension and the time dimension, that is, to capture the dependencies between different features and the interactions of the same feature between different time steps, so that the feature correlations are fully explored.

[0033] The aforementioned feature association transformation operation can be understood as learning the temporal patterns, coupling relationships between different dimensions, and feature evolution laws in the feature vector matrix through a specific neural network structure, thereby generating a feature representation containing rich contextual information. For example, algorithms such as Kalman filtering and wavelet transform can be used to decompose and reconstruct the feature vector matrix along the time dimension at multiple scales. Alternatively, neural networks can be used to mine feature couplings, or features can be extracted using models pre-trained based on simulation data.

[0034] S7: Based on the pre-trained state prediction model, generate the initial prediction data for the aircraft according to the feature vector matrix and the specified prediction duration; The aforementioned eigenvector matrix is ​​used to quickly generate preliminary, low-precision initial prediction data. This initial prediction data is represented in matrix form, with the same feature dimensions as the eigenvector matrix. Specifically, this initial prediction data is a prediction sequence of state data arranged in time steps for a specified prediction duration. The state data at each time step represents the aircraft's future state and control information at that moment. It should be noted that the control information is only related to human control factors, and the control information in the predicted state is usually the same as that in historical operational data. It should also be noted that the specified prediction duration may not be the same as the specified historical duration; that is, the time dimensions of the feature correlation matrix and the initial prediction data may differ (time step lengths may differ).

[0035] In this embodiment, based on the eigenvector matrix and a pre-trained state prediction model, the operational data within a specified prediction period is inferred through the state prediction model. The aforementioned eigenvector matrix contains basic information and temporal relationships of the original physical quantities. Initial predictions based on the eigenvector matrix can effectively capture motion patterns, thereby inferring the operational state at subsequent times. Specifically, the eigenvector matrix is ​​input into the state prediction model, and the initial prediction data is generated through the state prediction model. This can be understood as establishing a direct mapping relationship from a past time period to a future time period. The aforementioned state prediction model needs to satisfy the dynamic laws of the aircraft, and can be pre-constructed and trained based on the dynamic equations.

[0036] S9: Based on the feature correlation matrix, the initial prediction data is corrected for its correlation state to obtain corrected prediction data; The aforementioned feature correlation matrix is ​​used to calibrate the initial prediction data based on the extracted feature correlations. Through the step-by-step processing of prediction and adjustment, the deep information in historical operating data can be better utilized, thereby improving the prediction accuracy and reliability of state data.

[0037] Since the initial prediction data does not fully integrate the deep coupling information between different time steps, it is necessary to perform correlation state correction on the initial prediction data based on the feature correlation matrix to improve the accuracy and reliability of state data prediction. Specifically, the correlation state correction operation can be understood as correcting the initial prediction data according to the feature correlation relationship represented by the feature correlation matrix obtained in step S7, making it more consistent with the actual dynamic characteristics of the aircraft. For example, if the initial prediction data shows that the aircraft will suddenly change its direction of motion at some point in the future, but it can be determined based on the corresponding feature correlation relationship that this sudden change does not conform to the dynamic characteristics of the aircraft, the state correction mechanism will suppress this unreasonable prediction trend. For example, a correction network based on residual learning can be used to perform correlation state correction on the initial prediction data, or a dynamic feature fusion based on a cross-attention mechanism can be used to perform correlation state correction, or a model pre-trained based on simulation data can be used to perform correlation state correction. Compared with directly using the feature correlation matrix for data prediction, the step-by-step processing of prediction and correction significantly reduces the computational load of the prediction data while improving the accuracy of the prediction data.

[0038] S11: Perform low-dimensional feature mapping on the corrected prediction data to obtain the target state data of the aircraft, wherein the target state data characterizes the future state information of the aircraft within a specified prediction time.

[0039] The revised prediction data mentioned above still resides in a high-dimensional feature space, such as 256 dimensions. Each dimension does not correspond to a specific physical meaning but rather encodes complex, nonlinear temporal patterns and physical relationships. Therefore, it is necessary to map the revised prediction data to an output format suitable for aircraft state data, serving as the target state data output. For example, matrix multiplication projects the high-dimensional features into a low-dimensional physical space, in the opposite direction to the linear mapping used in step S3. This ensures that the relative relationships in the feature space are accurately reflected in the physical space.

[0040] It should be noted that the above-mentioned corrected prediction data only retains the feature mapping of state information during the mapping process. Since control information, as a quality emitted by the aircraft control system, is an external input, it does not need to be used as a prediction target. For example, if historical operational data includes state information in 12 state dimensions and control information in 3 control dimensions, then the target state data only includes 12 state dimensions to constitute the aircraft's state space. It should be noted that the above-mentioned historical operational data, feature vector matrix, and feature correlation matrix all determine the number of time steps based on a specified historical duration (or are represented by the time dimension), and the above-mentioned initial prediction data, corrected prediction data, and target state data all determine the number of time steps based on a specified prediction duration (or are represented by the time dimension). The above-mentioned target state data serves as the final output sequence of the aircraft's future states, used for decision inputs in aircraft flight control, attitude monitoring, etc.

[0041] Based on the above ideas, the technical solution provided in this embodiment of the application predicts and corrects the long-term state data of the aircraft based on its deep and long-term dependencies, thereby improving the accuracy and reliability of aircraft state prediction. Specifically, high-dimensional feature mapping, temporal position encoding, and feature association transformation are sequentially performed on historical operation data containing state and control information to obtain a feature association matrix containing feature relationships. This matrix represents the temporal context information and multi-dimensional feature coupling relationships during historical operation, thus maintaining stable prediction performance under different complex operating conditions. Furthermore, based on the feature association relationships extracted from the feature association matrix, the initial prediction data for rapid prediction is corrected for associated states, significantly improving the prediction accuracy of aircraft state data. At the same time, through the step-by-step processing of prediction and correction, prediction bias is effectively avoided in long-term prediction, thereby improving the accuracy and reliability of aircraft state prediction.

[0042] It should be noted that conventional aircraft prediction methods typically employ multi-level feature extraction models to progressively extract features from historical aircraft data, and then predict the outcome based on the extracted features. The predicted results are usually compared to pre-set standard values, and the model is corrected based on the error. However, in this application, after obtaining the feature vector matrix represented by the feature dimension and time dimension, initial prediction data for the aircraft is directly generated based on this feature vector matrix and a specified prediction duration. Then, a feature correlation transformation is performed on this feature vector matrix to obtain a feature correlation matrix that reflects the feature relationships between feature dimensions and the time dimension. Using this feature correlation matrix, further correlation state corrections can be made to the initial prediction data, resulting in accurate corrected prediction data. Therefore, the technical solution provided in this application does not pre-set standard values ​​but instead achieves unsupervised aircraft state prediction through an adaptive correction method using the feature vector matrix and the feature correlation matrix. Compared to conventional aircraft prediction methods, the technical solution provided in this application has a higher convergence speed and is less prone to overfitting.

[0043] In one implementation, based on step S3 above, the historical running data includes instantaneous running data from multiple consecutive time steps, with the number of time steps determined based on a specified historical duration and time step length. Furthermore, for any instantaneous running data at any time step, a high-dimensional feature mapping is performed on the instantaneous running data, mapping the historical running data to a high-dimensional feature space to obtain the corresponding embedded feature vector. Further, the temporal position code of the instantaneous running data at the current time step is determined, and the temporal position code is injected into the corresponding embedded feature vector to obtain the instantaneous feature vector of the current time step. The instantaneous feature vectors of each time step are arranged in temporal order to form the aforementioned feature vector matrix.

[0044] The instantaneous running data mentioned above can be regarded as a feature vector with 15 data dimensions at the current time step. The embedded feature vector is a feature vector with a preset feature dimension (e.g., 256 dimensions). After being injected by temporal position encoding, the dimension of the instantaneous feature vector does not change, that is, the preset feature dimension is still maintained. The instantaneous feature vectors of each time step are combined to form a feature vector matrix (represented by the preset feature dimension × preset historical duration as the dimension).

[0045] In one embodiment, sine-cosine position coding is used as the timing position coding as follows: Determine the timing position coding. ,in, Position of time step , To preset the historical duration, Indexing for feature dimensions , To define the preset feature dimensions, each position is encoded into a... A dimensional vector. The embedded feature vector is concatenated with the temporal position code to obtain the corresponding instantaneous feature vector, thus obtaining the feature vector matrix.

[0046] The technical solution provided in this embodiment achieves unified representation of data from different dimensions and the introduction of temporal information by injecting high-dimensional feature mapping and temporal position encoding into instantaneous running data at any time step in historical running data. Specifically, historical running data includes multi-dimensional instantaneous running data at multiple time points. Feature embedding of historical running data is performed through high-dimensional feature mapping, providing a representational basis for subsequent learning of rich features across different dimensions. Simultaneously, by injecting temporal position encoding into data features at different time steps, it is easy to distinguish data from different time steps and understand their sequential relationship, providing a standardized input representation for subsequent feature extraction and state prediction, effectively improving the temporal reliability and detection accuracy of state data prediction.

[0047] In one implementation, please refer to Figure 2 Based on step S5 above, the feature association transformation operation includes feature association processing and feature transformation processing. Feature extraction is performed on the feature vector matrix to obtain the feature association matrix, specifically including the following steps: S51: Based on a preset set of reference matrices, feature association processing is performed on the feature vector matrix to obtain feature association data, wherein the reference categories of each reference matrix in the aforementioned set of reference matrices are different; S53: Perform feature transformation processing on the above feature association data to obtain the feature association matrix, wherein the eigenvector matrix and the feature association matrix have the same dimension.

[0048] The aforementioned feature association processing is used to capture the correlations between different dimensions and time steps within the feature vector matrix. The aforementioned set of reference matrices is a predefined collection of multiple parameter matrices, including a first reference matrix, a second reference matrix, and a third reference matrix. Each reference matrix has a different reference category to learn and focus on different feature subspaces or different dependencies. For example, the first reference matrix is ​​used as an index reference for the feature vector matrix, the second reference matrix is ​​used as a query information reference for the feature vector matrix, and the third reference matrix is ​​used as an actual information reference for the feature vector matrix. Matrix operations are performed on the feature vector matrix according to the different reference matrices to obtain feature association data.

[0049] The aforementioned feature transformation process can be understood as performing nonlinear mapping and dimensionality normalization (e.g., dimensional expansion and dimensionality reduction) on the associated features to enhance the complex feature representation of the associated data. Since feature association processing primarily captures linear relationships, while aircraft dynamics are inherently nonlinear, nonlinear mapping is employed. Dimensional expansion and nonlinear mapping enhance the feature representation of the data, followed by dimensionality reduction and normalization operations to ensure the stability and consistency of the output. During the nonlinear mapping process, a nonlinear activation function, such as the ReLU function, is introduced to learn more complex feature representations, resulting in a feature association matrix containing deep-level coupling information.

[0050] In this embodiment, in step S51 above, the feature vector matrix is ​​subjected to feature association processing based on a preset reference matrix set to obtain feature association data, including the following steps: S511: Based on any reference matrix in the reference matrix set, perform a linear transformation on the eigenvector matrix to obtain the eigenvector matrix eigenma set, wherein the eigenma set includes multiple eigenma matrices corresponding to the reference matrix; S513: For any feature matrix in the feature matrix set, perform feature weighting on the feature matrix with other feature matrices to obtain the dimension-related data corresponding to the feature matrix; S515: Normalize the dimension-related data corresponding to each feature matrix so that the normalized dimension-related data can be used as feature-related data.

[0051] In one embodiment, see Figure 3 , eigenvector matrix Divide the data into time steps to obtain instantaneous runtime data for multiple time steps. , , For each instantaneous running data, it is used as an input vector. , with each reference matrix ( , , The first, second, and third reference matrices, respectively, are subjected to linear transformation to obtain the feature matrix of the instantaneous running data at the current time step. , , The above feature matrix characterizes , , The three types of eigenvectors correspond to the transformation results of each reference matrix.

[0052] Furthermore, performing correlated feature weighting on the feature matrix includes: for any feature matrix Combine it with the feature matrices of all other time steps The correlation degree is calculated in the following manner: ,in, Representing feature dimension, Characteristic matrix of Class feature vectors For other characteristic matrices , Class feature vectors. For each time step, the transient runtime data is weighted according to its correlation with other time steps to obtain the dimensional correlation data for the current time step. That is, the corresponding feature matrix ( , , ) Corresponding dimension related data ( , , Furthermore, the dimensional correlation data is normalized and represented as follows: The dimension-related data, which is normalized and whose vector values ​​are kept between 0 and 1, is used as the feature-related data.

[0053] In one embodiment, the feature vector matrix is ​​partitioned according to data dimensions, and feature association processing is performed on each resulting data block. For example, if the feature vector matrix has 256 dimensions, it is divided into eight 32-dimensional data blocks. Specifically, when calculating the association degree for each data block... The feature dimension is 32. Before normalizing the dimensional correlation data, residual joins are performed on the dimensional correlation data to avoid gradient vanishing of the dimensional correlation data. The dimensional correlation data corresponding to each database at each time step are then concatenated.

[0054] In one implementation, in step S33 above, performing feature transformation processing on the feature-related data to obtain the feature-related matrix includes: expanding the dimension of the feature-related data, that is, mapping the feature-related data to a higher dimension to obtain expanded dimension data; and performing function mapping on the expanded dimension data, that is, applying a nonlinear function (such as the ReLU function) to the expanded dimension data to fit a complex function. Further, performing dimension restoration and normalization processing on the expanded dimension data after function mapping, wherein the dimension restoration maps the expanded dimension data back to the feature dimensions of the feature-related data. Further, performing normalization processing on the expanded dimension data after dimension restoration, exemplarily represented as: To keep the vector values ​​between 0 and 1, Represent the feature association data to obtain the feature association matrix. .

[0055] The technical solution provided in this embodiment transforms the feature vector matrix into a feature correlation matrix containing deeply coupled information by sequentially performing feature correlation processing and feature transformation processing on the feature vector matrix. Specifically, the feature correlation processing, including linear transformation and weighted correlation features, effectively captures the complex dependencies and coupling features between different dimensions and time steps within the feature vector matrix. The feature transformation processing, including dimensionality expansion, nonlinear mapping, and normalization, further enhances the nonlinear expressive power of the features and ensures the stability of the output, thereby improving the fitting ability to complex dynamic characteristics. This provides a feature foundation containing physical information and contextual correlation for subsequent correlation state correction based on the feature correlation matrix, thus enabling accurate prediction of aircraft state data.

[0056] In one implementation, please refer to Figure 4 Based on step S9 above, the initial prediction data is corrected according to the feature correlation matrix to obtain corrected prediction data, including the following steps: S91: Based on a preset set of reference matrices, perform cross-correlation processing on the feature correlation matrix and the initial prediction data to obtain the initial corrected data corresponding to the initial prediction data; S93: Perform feature transformation on the initial corrected data, and use the initial corrected data after feature transformation as the corrected prediction data.

[0057] In this embodiment, unlike the feature association processing operation in step S51 described above, feature association processing is used for information interaction within the data to extract feature association relationships. Cross-association processing is based on the feature association matrix and the initial prediction data, and is used to calibrate and correct the initial prediction data according to the feature association relationships between feature dimensions and time dimension represented in the feature association matrix. Compared to performing feature association processing on the initial prediction data itself, the initial prediction data after cross-association processing is closer to the actual physical laws of the aircraft. For example, it avoids the generation of abnormal data such as sudden velocity changes and position jumps, enhances the physical consistency of the corrected prediction data, and thus improves the authenticity and reliability of the state data obtained in subsequent predictions.

[0058] In this embodiment, based on step S91 above, the cross-correlation processing of the feature correlation matrix and the initial prediction data based on a preset set of reference matrices includes the following steps: S911: Based on a preset set of reference matrices, determine a first set of feature matrices for the feature correlation matrix and a second set of feature matrices for the initial prediction data; S913: For any second feature matrix in the second feature matrix set, perform associated feature weighting on the second feature matrix and the first feature matrix set to obtain the dimension associated data corresponding to the second feature matrix; S915: Normalize the dimension-related data corresponding to each second feature matrix, and use the normalized dimension-related data as the initial correction data.

[0059] In one embodiment, see Figure 5 , feature correlation matrix Divide the data into time steps to obtain the first instantaneous running data of multiple time steps. , , ), and combine it with each reference matrix ( , , Let (representing the first reference matrix, the second reference matrix, and the third reference matrix respectively) undergo a linear transformation to obtain the first set of characteristic matrices of the characteristic correlation matrix. , including the first feature matrix , , The initial prediction data Divide the data into time steps to obtain the second instantaneous running data for multiple time steps. , , ), and combine it with each reference matrix ( , , The first, second, and third reference matrices (represented by the first, second, and third reference matrices, respectively) undergo a linear transformation to obtain the second feature matrix set of the initial prediction data. , including the second feature matrix , , .

[0060] Furthermore, the weighted association between the second feature matrix and the set of first feature matrices includes: for any second feature matrix... Combine it with the first feature matrix of all other time steps The correlation degree is calculated in the following manner: ,in, Representing feature dimension, For the second characteristic matrix Class feature vectors For other characteristic matrices , Feature vectors. For the second transient data at each time step, weights are applied according to the correlation degree between the second feature matrix and the first feature matrix at other time steps to obtain the dimensional correlation data of the second feature matrix at the current time step. , , Furthermore, the dimensional correlation data is normalized and represented as follows: The vector values ​​are kept between 0 and 1, and the normalized dimension correlation data is used as the initial correction data.

[0061] In this embodiment, based on step S93 above, feature transformation processing is performed on the initial corrected data, and the initial corrected data after feature transformation is used as the corrected prediction data. This includes: expanding the dimensions of the initial corrected data, that is, mapping the initial corrected data to a higher dimension to obtain corresponding expanded dimension data; and performing function mapping on the expanded dimension data, that is, applying a nonlinear function (such as the ReLU function) to the expanded dimension data to fit a complex function. Further, dimension restoration and normalization processing are performed on the expanded dimension data after function mapping. The dimension restoration process maps the expanded dimension data back to the feature dimensions of the initial corrected data. Further, the initial corrected data after dimension restoration is normalized, and the data obtained after normalization is used as the corrected prediction data.

[0062] The technical solution provided in this embodiment improves the accuracy of state data prediction by performing cross-correlation processing and feature transformation processing on the initial prediction data to correct its associated state. Specifically, based on a preset reference set, cross-correlation processing is performed on the initial prediction data and the feature correlation matrix. This allows for point-by-point calibration of the initial prediction data based on the historical deep features contained in the feature correlation matrix, improving the accuracy and reliability of state prediction. Furthermore, feature transformation processing is applied to the initial corrected data obtained from the cross-correlation processing to further enhance the nonlinear expression and output reliability of the corrected data, thereby achieving accurate and reliable prediction of the aircraft's future long-term state data.

[0063] In one implementation, based on step S7 above, generating initial prediction data for the aircraft based on a pre-trained state prediction model, according to the feature vector matrix and a specified prediction duration, includes: acquiring the pre-trained state prediction model; inputting the feature vector matrix into the state prediction model; and generating state prediction data for a specified prediction duration using the state prediction model. The state prediction model can be a trained neural network model used to map historical state sequences to future state sequences, establishing a mapping relationship from past state data to future state data. State prediction data for a specified prediction duration is generated by providing a feature vector matrix with a specified historical duration.

[0064] Furthermore, based on a preset set of reference matrices, feature association processing is performed on the state prediction data to obtain initial prediction data. The aforementioned set of reference matrices consists of a predefined first reference matrix, a second reference matrix, and a third reference matrix, each with a different reference category. For example, the feature association processing includes performing a linear transformation on the state prediction data based on any one of the reference matrices in the set to obtain a set of feature matrices corresponding to the state prediction data; for any feature matrix in the set, performing feature weighting and association with other feature matrices to obtain the feature prediction data corresponding to that feature matrix; and normalizing the feature prediction data of each feature matrix corresponding to the state prediction data, using the normalized feature prediction data as the initial prediction data.

[0065] The aforementioned state prediction model is trained on a simulation dataset generated by a mechanistic simulation model. This mechanistic simulation model is pre-constructed based on the dynamic characteristics of the aircraft and is used to simulate the physical characteristics of a six-degree-of-freedom aircraft. The state prediction model can be constructed using a neural network to fully learn the state change patterns of the simulation data under different operating conditions. The dynamic characteristics are characterized by the state control equations and differential equations of the mechanistic simulation model, enabling the trained state prediction model to possess prior physical knowledge. This improves the accuracy and rationality of the initial prediction, resulting in more realistic and reliable final state data.

[0066] In this embodiment, control signals under different operating conditions are processed through a transfer function determined by environmental parameters to collect response signal sequences. These sequences are then input into a mechanistic simulation model, which outputs state data. Based on this data, a simulation dataset for a state prediction model is constructed. The control signals are composite signals generated from ramp signals, pulse signals, sinusoidal signals, step signals, or any combination thereof. These control signals are randomly sampled and generated within a preset range to ensure coverage of the operating conditions required for aircraft control, as well as various extreme conditions.

[0067] In one implementation, the above-described mechanism simulation model is characterized by state-space equations. These state-space equations are constructed based on an input state vector, an observation output vector, and an excitation vector. The input state vector is determined based on a differential equation, which characterizes the dynamic characteristics of the aircraft. The observation output vector is used to construct the simulation dataset. The excitation vector is determined based on a transfer function, which characterizes different operating conditions of the aircraft. Understandably, the above-described mechanism simulation model is a composite form of state-space equations, transfer functions, and differential equations.

[0068] In one embodiment, the above state-space equation can be understood as a model that takes the aircraft's input state vector, the excitation vector used for control, and the environmental parameter vector as inputs, and the observation output vector as output. For example, it can be represented as: ,in, , As time goes by Continuously updated, in the form of, , .in, , The input state vector and observation output vector represent the actual state and observed output of the aircraft at time t (both characterized by 12 state dimensions). The aforementioned observation output vector serves as a vector representation of the state data output by the mechanistic simulation model and can be used to construct a simulation dataset for the state prediction model. The excitation vector, used to simulate the aircraft's operation under different conditions, is specifically the vector obtained at time t from a sequence of vectors derived from multiple control signals after processing by a transfer function determined by environmental parameters. The environmental observation vector can represent observation noise. and It is a nonlinear function, specifically a state function constructed through differential equations and an observation function of the sensor.

[0069] In this embodiment, the excitation vector is characterized by three control quantities. Specifically, control signals of different types, amplitudes, and frequencies are input to simulate the control outputs under different operating conditions, thereby obtaining the three control quantities of the aircraft relative to the velocity direction. , and The above transfer function is expressed as: ,in, For the damping ratio, It is the frequency of undamped natural oscillation. As a Laplace operator, the above transfer function can characterize the different operating conditions that the aircraft may be in.

[0070] In this embodiment, the aforementioned nonlinear function A state transition function is composed of multiple differential equations. These differential equations characterize the dynamics of the aircraft and provide physical constraints for the aircraft's state evolution. Specifically, the state transition function composed of multiple differential equations is expressed as follows: ,in, 、 、 These are the three-dimensional coordinate systems of the aircraft relative to the ground. axis, shaft and Displacement of the axis, , , They are respectively 、 、 The derivative of , and These are the three control variables for the aircraft relative to the velocity direction. The pitch angle of the aircraft. The yaw angle of the aircraft. Let be the velocity of the aircraft. The aforementioned nonlinear function... The sensor observation function, composed of multiple differential equations, can be viewed as the output of the state vector from the previous time step. Introducing observation noise The output value, after retaining several significant digits based on the sensor's accuracy, is used as the current moment. Observation output .

[0071] The technical solution provided in this embodiment trains a state prediction model based on a simulation dataset generated from a mechanistic simulation model, and achieves accurate prediction of initial prediction data based on the trained state prediction model. Specifically, the aforementioned mechanistic simulation model is pre-constructed according to the dynamic characteristics of the aircraft, and its output state data is collected to construct a simulation dataset covering various operating conditions and extreme conditions. This allows the state prediction model to fully learn the state evolution laws of the aircraft under different operating conditions and the prior physical knowledge of the aircraft. The initial prediction data generated by the state prediction model obtained through the above training method not only possesses the ability to quickly map but also contains the dynamic characteristics and physical laws of the aircraft, effectively improving the accuracy and reliability of the initial prediction.

[0072] In one implementation, before embedding and encoding historical operational data, anomaly detection is performed on the historical operational data to check for missing or outlier values. Missing or outlier data is then filled in as follows: ,in, For time step Instantaneous state data at that time The time step is defined as [value]. Further, the imputed historical operational data undergoes data normalization and denormalization, meaning that state variables at different scales are grouped to the same scale (mean 0, variance 1). By pre-impacting the historical operational data, potential noise and missing data in the original data are effectively eliminated, providing a complete data foundation for subsequent embedding encoding, feature extraction, and state prediction, thereby enhancing the robustness of state prediction.

[0073] Please see Figure 6 This application also provides an aircraft state data prediction device, the device comprising: The historical data acquisition unit 100 is used to acquire the historical operation data of the aircraft, wherein the historical operation data represents the status information and control information of the aircraft within a specified historical time period; The historical data processing unit 200 is used to perform high-dimensional feature mapping on the historical running data to obtain embedded feature vectors, inject temporal position codes into the embedded feature vectors to obtain a feature vector matrix, wherein the feature vector matrix is ​​characterized by feature dimension and time dimension, and perform feature association transformation on the feature vector matrix to obtain a feature association matrix, wherein the feature association matrix is ​​used to reflect the feature association relationship between feature dimensions and time dimension. The state data prediction unit 300 is used to generate initial prediction data of the aircraft based on a pre-trained state prediction model, according to the feature vector matrix and a specified prediction duration, and to perform associated state correction on the initial prediction data according to the feature correlation matrix to obtain corrected prediction data. The state data processing unit 400 is used to perform low-dimensional feature mapping on the corrected prediction data to obtain the target state data of the aircraft, wherein the target state data characterizes the future state information of the aircraft within a specified prediction time.

[0074] in, In one embodiment, the historical data acquisition unit 100 is specifically used to acquire the historical operation data of the aircraft, which includes instantaneous operation data of multiple consecutive time steps. The historical operation data is subjected to anomaly detection, and if the anomaly detection result indicates the presence of missing or abnormal data, the missing or abnormal data is filled in.

[0075] In one embodiment, the historical data processing unit 200 is specifically configured to perform high-dimensional feature mapping on the instantaneous running data at any time step to obtain the embedded feature vector of the current time step, determine the temporal position code of the instantaneous running data at the current time step, inject the temporal position code into the corresponding embedded feature vector to obtain the instantaneous feature vector of the current time step, arrange the instantaneous feature vectors of each time step in temporal order, use the arranged instantaneous feature vectors as the feature vector matrix, perform feature association processing on the feature vector matrix based on a preset reference matrix set to obtain feature association data, and perform feature transformation processing on the feature association data to obtain the feature association matrix.

[0076] In one embodiment, the state data prediction unit 300 is specifically used to acquire a pre-trained state prediction model, input a feature vector matrix into the state prediction model, generate state prediction data for a specified prediction duration, perform feature association processing on the state prediction data based on a preset set of reference matrices to obtain initial prediction data, perform cross-association processing on the feature association matrix and the initial prediction data based on the preset set of reference matrices to obtain initial corrected data corresponding to the initial prediction data, perform feature transformation processing on the initial corrected data, and use the initial corrected data after feature transformation processing as corrected prediction data.

[0077] In one embodiment, the state data processing unit 400 is specifically used to perform low-dimensional feature mapping on the corrected prediction data, and use the corrected prediction data after low-dimensional feature mapping as target state data, so as to map the corrected prediction data to an output form suitable for aircraft state data. The mapping direction of the low-dimensional feature mapping is opposite to the mapping direction in the high-dimensional feature mapping process, and the target state data is used as the final output sequence of future aircraft states.

[0078] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0079] The aircraft state data prediction device in this application embodiment is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, or other devices that can provide the above functions.

[0080] Please see Figure 7 , Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application, such as... Figure 7As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 7 Take a processor 10 as an example.

[0081] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0082] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0083] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0084] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0085] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0086] This application also provides a computer-readable storage medium. The methods described in this application can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the methods shown in the above embodiments are implemented.

[0087] The apparatus or module described in the above embodiments can be implemented by a computer chip or entity, or by a product with a certain function. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0088] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0089] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer devices. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0090] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer devices according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0091] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0092] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0093] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0094] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0095] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.

[0096] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method of predicting state data of an aircraft, characterized by, The method includes: Acquire historical operational data of the aircraft, wherein the historical operational data characterizes the status and control information of the aircraft within a specified historical time period; High-dimensional feature mapping is performed on the historical operation data to obtain embedded feature vectors, and temporal position codes are injected into the embedded feature vectors to obtain a feature vector matrix, wherein the feature vector matrix is ​​characterized by feature dimension and time dimension; The feature vector matrix is ​​subjected to feature association transformation to obtain a feature association matrix, which is used to reflect the feature association relationship between feature dimensions and between time dimensions; Based on a pre-trained state prediction model, initial prediction data for the aircraft is generated according to the feature vector matrix and the specified prediction duration. Based on the feature correlation matrix, the initial prediction data is corrected for its correlation state to obtain corrected prediction data; The corrected prediction data is subjected to low-dimensional feature mapping to obtain the target state data of the aircraft, which represents the future state information of the aircraft within a specified prediction time.

2. The method of claim 1, wherein, The state prediction model is trained based on a simulation dataset generated from the mechanism simulation model; wherein: The mechanism simulation model is characterized by state-space equations, which are constructed based on the input state vector, the observation output vector, and the excitation vector. The input state vector is determined based on a differential equation, which is used to characterize the dynamic characteristics of the aircraft. The observation output vector is used to construct the simulation dataset. The excitation vector is determined based on a transfer function, which is used to characterize the different operating conditions of the aircraft.

3. The method of claim 1, wherein, Performing a feature correlation transformation on the eigenvector matrix to obtain a feature correlation matrix includes: Based on a preset set of reference matrices, feature association processing is performed on the feature vector matrix to obtain feature association data, wherein the reference categories of each reference matrix in the set of reference matrices are different; The feature association data is subjected to feature transformation processing to obtain the feature association matrix, wherein the feature vector matrix and the feature association matrix have the same dimension.

4. The method of claim 3, wherein, Based on a preset set of reference matrices, feature association processing is performed on the feature vector matrix to obtain feature association data, including: Based on any reference matrix in the set of reference matrices, a linear transformation is performed on the eigenvector matrix to obtain a set of eigenvector matrices, wherein the set of eigenvector matrices includes multiple eigenma matrices corresponding to the reference matrices; For any feature matrix in the set of feature matrices, the feature matrix is ​​associated with other feature matrices and weighted accordingly to obtain the dimension-related data corresponding to the feature matrix. The dimensional correlation data corresponding to each feature matrix is ​​normalized so that the normalized dimensional correlation data can be used as the feature correlation data.

5. The method of claim 3, wherein, The feature association data is subjected to feature transformation processing to obtain the feature association matrix, including: The feature-related data is augmented to obtain augmented dimension data, and the augmented dimension data is then mapped using a function. The expanded dimension data after function mapping is subjected to dimension restoration and normalization to obtain the feature correlation matrix.

6. The method of claim 1, wherein, Based on a pre-trained state prediction model, the initial prediction data for the aircraft is generated according to the feature vector matrix and a specified prediction duration, including: The feature vector matrix is ​​input into the state prediction model to generate state prediction data for a specified prediction duration; Based on a preset set of reference matrices, feature association processing is performed on the state prediction data to obtain the initial prediction data.

7. The method of claim 1, wherein, Based on the feature correlation matrix, the initial prediction data is corrected for its correlation state to obtain corrected prediction data, including: Based on a preset set of reference matrices, the feature correlation matrix and the initial prediction data are cross-correlated to obtain the initial corrected data corresponding to the initial prediction data. The initial corrected data is subjected to feature transformation processing, and the initial corrected data after feature transformation processing is used as the corrected prediction data.

8. The method of claim 7, wherein, Based on a preset set of reference matrices, the cross-correlation processing of the feature correlation matrix and the initial prediction data includes: Based on a preset set of reference matrices, a first set of feature matrices for the feature correlation matrix is ​​determined, and a second set of feature matrices for the initial prediction data is determined. For any second feature matrix in the second feature matrix set, the second feature matrix is ​​associated with the first feature matrix set by performing a weighted feature aggregation to obtain the dimension-related data corresponding to the second feature matrix. The dimensional correlation data corresponding to each second feature matrix are normalized, and the normalized dimensional correlation data is used as the initial correction data.

9. An apparatus for predicting state data of an aircraft, characterized by comprising: The device includes: The historical data acquisition unit is used to acquire the historical operation data of the aircraft, wherein the historical operation data represents the status information and control information of the aircraft within a specified historical period. The historical data processing unit is used to perform high-dimensional feature mapping on the historical running data to obtain embedded feature vectors, inject temporal position codes into the embedded feature vectors to obtain a feature vector matrix, wherein the feature vector matrix is ​​characterized by feature dimension and time dimension, and perform feature association transformation on the feature vector matrix to obtain a feature association matrix, wherein the feature association matrix is ​​used to reflect the feature association relationship between feature dimensions and between time dimension; The state data prediction unit is used to generate initial prediction data for the aircraft based on a pre-trained state prediction model, according to the feature vector matrix and a specified prediction duration, and to perform associated state correction on the initial prediction data according to the feature correlation matrix to obtain corrected prediction data. A state data processing unit is used to perform low-dimensional feature mapping on the corrected prediction data to obtain the target state data of the aircraft, wherein the target state data characterizes the reference operating state of the aircraft in the future within a specified prediction time.

10. A computer device, comprising: include: A memory and a processor are communicatively connected, the memory storing computer instructions, and the processor executing the computer instructions to perform the aircraft state data prediction method according to any one of claims 1 to 8.

Images