TCN-LSTM-based flight control system multivariate data fusion virtual test verification method

Abstract

The invention belongs to the technical field of data fusion, and relates to a TCN-LSTM-based flight control system multivariate data fusion virtual test verification method, which comprises the following steps of: 1, acquiring flight data of a plurality of parameters from a flight control computer of an aircraft and a virtual test; 2, performing multivariate data signal level fusion; step 3, performing multivariate data correlation analysis; step 4, feature level data fusion based on TCN-LSTM is carried out; 5, verifying the multivariate data feature level fusion simulation model; according to the method, the relevance of multivariate output data in time and space dimensions is mined by fusing multivariate simulation data. And carrying out difference evaluation analysis on the multivariate reference data and the simulation data by adopting a CDF, and converting the difference evaluation result into a credibility grade evaluation result.

Description

Technical Field

[0001] The present invention belongs to the technical field of data fusion, and particularly relates to a virtual test verification method for multi-source data fusion of a flight control system based on TCN-LSTM. Background Art

[0002] Current aircraft have a high degree of integration, complex functions, and high costs. The flight control system is one of the core systems of an aircraft, and its design and performance directly affect flight safety, mission execution efficiency, and overall performance. Conducting virtual test research on the flight control system can perform system performance evaluation, fault mode analysis, design optimization, etc. in the initial design stage, improving the efficiency of the flight control system design, manufacturing, and maintenance processes and reducing costs. However, virtual tests often involve certain theoretical assumptions and model simplifications, which may lead to differences between virtual test results and actual situations.

[0003] According to the output result types of the virtual test simulation model, the verification methods for the flight control system simulation model can be mainly divided into static verification methods, time-domain verification methods, frequency-domain verification methods, and verification methods based on feature categories. Static verification methods focus on the statistical characteristics of data, and the verification methods include parameter estimation methods, hypothesis testing methods, Bayesian factors, evidence distances, and probability distribution difference methods, etc. Time-domain verification methods require the simulation time-series data to be consistent with the reference time-series data in terms of time series, and verify the simulation model by comparing the distance and shape between the two. Common methods include Theil inequality coefficient method, grey relational analysis method, relative mean absolute error, and Euclidean distance, etc. Frequency-domain verification methods are for periodic time-series data and mainly consist of power spectrum estimation and compatibility test. Power spectrum estimation verifies the simulation model by comparing the power distributions of the simulation data and the reference data at each frequency and analyzing the consistency of the statistical characteristics of their power spectral densities. Verification methods based on feature categories analyze the characteristics of the simulation data and the reference data and verify the simulation model based on this, including simulation model verification methods based on knowledge, triangular fuzzy numbers, data characteristics, and ensemble learning, etc.

[0004] Existing simulation verification methods cannot effectively mine the characteristics of the simulation data of the flight control system simulation model in the time and space dimensions and the correlation between data, resulting in inaccurate verification results of the simulation model, and also requiring a large amount of time and high economic costs. Summary of the Invention

[0005] Objective of the present invention: To solve the problems existing in the prior art, the present invention provides a virtual test verification method for multi-source data fusion based on TCN-LSTM, which fuses multi-source simulation data and mines the correlation of multi-source output data in the time and space dimensions. The CDF is used to evaluate and analyze the differences between the multi-source reference data and the simulation data, and convert them into the evaluation results of the credibility level.

[0006] Technical solution of the present invention:

[0007] A virtual test verification method for multi-source data fusion of a flight control system based on TCN-LSTM includes the following steps:

[0008] Step 1: Collect flight data of multiple parameters from the flight control computer of the aircraft and the virtual test.

[0009] Step 2: Perform multi-source data signal-level fusion.

[0010] Step 3: Perform multi-source data correlation analysis.

[0011] Step 4: Perform feature-level data fusion based on TCN-LSTM.

[0012] Step 5: Perform verification of the multi-source data feature-level fusion simulation model.

[0013] Further, in Step 1, flight data of multi-source parameters are collected from the flight control computer of the aircraft and the virtual test, including flight data of 20 dimensions in 5 categories of load, angular rate, angle, speed, and position.

[0014] Further, Step 2 specifically uses the signal-level data fusion method to fuse the flight data of multiple parameters output by the simulation model. The flight data of multiple parameters output by the simulation model includes the flight data output by the real reference system and the virtual test simulation model.

[0015] Further, the specific process of Step 2 is: Use S to represent the flight control system, S S and S R respectively represent the virtual test simulation model and the real reference system; Use O S and O R respectively represent the multi-source output data sets of the simulation model and the real reference system under the same input conditions. Therefore, O S fuses the multi-source output data of the simulation model, and O R fuses the multi-source output data of the real reference system, that is:

[0016]

[0017] O Sn and O RnDenote the flight data of any dimension of the simulation model and the real reference system respectively. Let \(i\) represent the length of the time series of the multi - output data, and \(t\) 1 , \(t\) 2 , …, \(t\) i represent the time points corresponding to the time series. Let \(n\) represent the dimension of the output data, that is, the number of acquisition sensors for the flight data of this dimension in the flight control system data acquisition.

[0018] Furthermore, in step 3, the Pearson correlation coefficient (PCC) is used to measure the correlation strength between the multi - variable data of the flight control system: when PCC is greater than 0, it indicates a positive correlation between the two groups of data, and the closer the value is to 1, the stronger the linear positive correlation; when PCC is less than 0, it indicates a negative correlation between the two groups of data, and the closer the value is to - 1, the stronger the linear negative correlation; when PCC is close to 0, it indicates a very weak correlation between the two groups of data, and there is no linear relationship between the data.

[0019] Furthermore, the Pearson correlation coefficient PCC calculates the correlation \(r\) between the flight data of this dimension of the simulation model and the flight data of this dimension of the real reference system through the following formula X，Y :

[0020]

[0021] where \(cov(X,Y)\) represents the covariance between data \(X\) i and \(Y\) i , \(\sigma\) X represents the standard deviation of data \(X\) i , \(\mu\) X represents the mean of data \(X\) i , \(\sigma\) Y represents the standard deviation of data \(Y\) i , \(\mu\) Y represents the mean of data \(Y\) i ; \(N\) represents the length of the flight data of this dimension of the simulation model and the flight data of this dimension of the real reference system.

[0022] Construct a data correlation matrix, and define that \(r\) X,Y \(\in[0.8,1]\) indicates a relatively high correlation between the data, \(r\) X,Y \(\in[0.2,0.8)\) indicates that there is a correlation between the data, \(r\) X,Y \(\in[0,0.2)\) indicates a weak correlation between the data. Combine the multi - output data combinations of the real reference system with relatively high correlations and the multi - output data combinations of the simulation model, and use them as the multi - output feature data set of the real reference system and the multi - output feature data set of the simulation model

[0023] Furthermore, step 4 specifically is to use the multi - output feature data set of the real reference system And the simulation model multivariate output feature dataset As the original input data of the data reconstruction neural network model, the reconstructed data o Sj ' and o Rj ', realize the feature-level fusion of multivariate data of flight control system.

[0024] Further, step 5 is specifically: calculate under the same input conditions o Rj ' and o Sj 'Multivariate output data, KS test was used to evaluate the o calculated under the same input conditions. Rj 'Multivariate output data and o Sj 'The CDF difference between the multivariate output data; finally the credibility level evaluation result C is obtained:

[0025]

[0026] Where P K-S represents the P value of KS test;

[0027] Furthermore, steps 3 to 5 are repeated to cyclically calculate the CDF difference after the fusion of other dimensional flight data of the simulation model and other dimensional flight data of the real reference system, so as to realize the verification of the multivariate data feature-level fusion simulation model of the flight control system.

[0028] Beneficial effects of the present invention: The present invention proposes a virtual test verification method for multivariate data fusion of a flight control system based on TCN-LSTM. Based on signal-level data fusion, the method fuses multivariate simulation data, effectively reduces the noise and error in a single simulation data source, and improves the accuracy of the input data. On this basis, the PCC correlation coefficient is used to perform correlation analysis on the multivariate output data of the simulation model and the real reference system under the same input conditions, and the data features are extracted respectively. The data feature-level fusion method based on TCN-LSTM is used to perform feature fusion, and the correlation of the multivariate output data in the time and space dimensions is mined. Finally, CDF is used to perform a difference evaluation analysis on the multivariate reference data and the simulation data, and it is converted into a credibility level evaluation result to realize the verification of the virtual test simulation model of the flight control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a flow chart of the method of the present invention;

[0030] Figure 2 It is the feature-level data fusion network architecture of TCN-LSTM. DETAILED DESCRIPTION

[0031] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0032] An embodiment of the present invention provides a virtual test verification method based on TCN-LSTM for fusing multi-source data, mining the correlation of multi-source output data in the time and space dimensions, and using CDF to perform differential evaluation and analysis on multi-source reference data and simulation data, and converting it into a credibility level evaluation result, which specifically includes the following steps:

[0033] Step 1: Collect flight data of multiple parameters from the flight control computer of the aircraft and virtual tests; specifically, collect flight data of multi-source parameters from the flight control computer of the aircraft and virtual tests, including 20 dimensions of flight data in 5 categories: load, angular rate, angle, speed, and position.

[0034] Step 2: Perform multi-source data signal-level fusion; traditional simulation model verification methods often only focus on single simulation and reference data, without comprehensively considering the relationship between multi-source data of the flight control system, and cannot effectively mine the characteristics of multi-source data in the time and space dimensions and the correlation between data. The present invention adopts a signal-level data fusion method to fuse the flight data of multiple parameters output by the simulation model. The flight data of multiple parameters output by the simulation model includes the flight data output by the real reference system and the virtual test simulation model. The multi-source output data after signal-level fusion reduces the noise and error in a single simulation data source, effectively improving the accuracy and reliability of the virtual test simulation model verification of the flight control system.

[0035] The specific process of Step 2 is as follows: Let S represent the flight control system, S S and S R represent the virtual test simulation model and the real reference system respectively; let O S and O R represent the multi-source output data sets of the simulation model and the real reference system under the same input conditions respectively. Therefore, O S fuses the multi-source output data of the simulation model, and O R fuses the multi-source output data of the real reference system, that is:

[0036]

[0037] O Sn and O RnDenote the flight data of any dimension of the simulation model and the real reference system respectively. Let \(i\) represent the length of the time series of the multi-output data, and \(t\) 1 , \(t\) 2 , …, \(t\) i represent the corresponding time points of the time series, and \(n\) represents the dimension of the output data, that is, the number of acquisition sensors for collecting the flight data of this dimension by the flight control system data acquisition.

[0038] Step 3: Conduct multi-data correlation analysis;

[0039] In Step 3, the Pearson correlation coefficient (PCC) is used to measure the correlation strength between the multi-data of the flight control system: when PCC > 0, it indicates a positive correlation between the two groups of data, and the closer the value is to 1, the stronger the linear positive correlation; when PCC < 0, it indicates a negative correlation between the two groups of data, and the closer the value is to -1, the stronger the linear negative correlation; when PCC is close to 0, it indicates a very weak correlation between the two groups of data, and there is no linear relationship between the data.

[0040] The Pearson correlation coefficient PCC calculates the correlation \(r\) between the flight data of this dimension of the simulation model and the flight data of this dimension of the real reference system through the following formula X，Y :

[0041]

[0042] where \(cov(X, Y)\) represents the covariance between data \(X\) i and \(Y\) i , \(\sigma\) X represents the standard deviation of data \(X\) i , \(\mu\) X represents the mean of data \(X\) i , \(\sigma\) Y represents the standard deviation of data \(Y\) i , \(\mu\) Y represents the mean of data \(Y\) i , and \(N\) represents the length of the flight data of this dimension of the simulation model and the flight data of this dimension of the real reference system;

[0043] Construct a data correlation matrix, and define \(r\) X,Y ∈[0.8, 1] indicating that there is a high correlation between the data, \(r\) X,Y ∈[0.2, 0.8) indicating that there is a correlation between the data, \(r\) X,Y ∈[0, 0.2) indicating that the correlation between the data is weak. Combine the multi-output data combinations of the real reference system with high correlations and the multi-output data combinations of the simulation model, and use them as the multi-output feature data set of the real reference system and the multi-output feature data set of the simulation model

[0044] Step 4: Perform feature-level data fusion based on TCN-LSTM;

[0045] The multi-output data of the flight control system is correlated in time and space. In the present invention, the Long Short-Term Memory (LSTM) neural network and the Temporal Convolutional Network (TCN) are introduced into the data reconstruction neural network model to capture the dependency relationships between multi-output data at different time points and extract high-level spatial features. As Figure 2 shown in the data reconstruction neural network model proposed in this paper, X represents the original input data, and X p represents the data after multi-data correlation analysis, and X' represents the reconstructed data. The input dimension of the model is n, and the output dimension is 1, that is, n multi-output data of the real reference system are used to reconstruct 1 data to achieve data feature-level fusion.

[0046] Specifically, in Step 4, the multi-output feature data set of the real reference system and the multi-output feature data set of the simulation model are used as the original input data of the data reconstruction neural network model, and the reconstructed data o Sj ' and o Rj ' are obtained respectively to achieve multi-data feature-level fusion of the flight control system.

[0047] In the above Step 4, the training optimization index of the feature-level data fusion model of TCN-LSTM is the Mean Squared Error (MSE), the optimizer used is Adam, the learning rate is set to 0.001, and the weight decay regularization technology (weight decay = 0.00001) is applied to prevent the model from overfitting.

[0048] Step 5: Verify the multi-data feature-level fusion simulation model.

[0049] Specifically, in Step 5: Calculate the multi-output data of o Rj ' and o Sj ' respectively under the same input conditions, and use the K-S test to evaluate the CDF difference between the multi-output data of o Rj ' and the multi-output data of o Sj ' calculated under the same input conditions; finally, obtain the credibility level evaluation result C:

[0050]

[0051] where P K-S represents the P value of the K-S test;

[0052] Repeat steps 3 to 5, and cyclically calculate the CDF difference after fusing the flight data of other dimensions of the simulation model and the flight data of other dimensions of the real reference system, so as to realize the verification of the multi-source data feature-level fusion simulation model of the flight control system.

[0053] The second embodiment of the present invention proposes a virtual test verification method for multi-source data fusion of a flight control system based on TCN-LSTM, including the following steps:

[0054] 1. Collect multiple flight parameters from the flight control computer of a certain type of aircraft and virtual tests, including load types (axial overload signal, lateral overload signal, normal overload signal), angular rate types (pitch angular rate signal, roll angular rate signal, yaw angular rate signal), angle types (angle of attack signal, sideslip angle signal, flight path angle, track inclination angle, true course angle signal, pitch angle signal, roll angle signal), speed types (true airspeed signal, ground speed signal, celestial velocity), position types (left aileron flap position, left rudder flap position, right aileron flap position, right rudder flap position), etc.

[0055] 2. By using the method of data signal-level fusion, we can obtain:

[0056]

[0057] 3. Multi-source data correlation analysis. Use PCC to measure the correlation strength between multi-source data of the flight control system. First, for a set of simulation output data o Sj , j ∈ [1, n] of the simulation model, the output data of its real reference system under the same input conditions is o Rj , j ∈ [1, n]; then, use the PCC correlation analysis method to calculate the correlation coefficients of o S in the multi-source output data O Sj of the simulation model and the real reference system respectively, and construct a data correlation matrix, as shown; R in O Rj Finally, define

[0058]

[0059] Finally, define indicating that there is a high correlation between data, indicating that there is a correlation between data, indicating that the correlation between data is weak. Respectively combine the above output data with high correlation into multi-source output feature datasets and

[0060] 4. Feature-level data fusion based on TCN-LSTM. Combine the above multi-source output feature datasets and As the original input data for the data reconstruction neural network model, the reconstructed data o Sj ' and o Rj ' are obtained respectively, realizing the multi-source data feature-level fusion of the flight control system.

[0061] 5. Verification of the multi-source data feature-level fusion simulation model. Calculate the CDF of o Rj ' and o Sj ' respectively under the same input conditions, and use the K-S test to evaluate the CDF difference between the two groups of data. Finally, the credibility level evaluation result C is obtained:

[0062]

[0063] where P K-S represents the P-value of the K-S test.

[0064] 6. Repeat steps 3 - 5 to calculate the CDF of the multi-source output data, realizing the verification of the multi-source data feature-level fusion simulation model of the flight control system.

[0065] As described above, only the specific embodiments of the present invention are described in detail, and the unelaborated parts are conventional technologies. However, the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. The protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A virtual test verification method for multivariate data fusion of flight control system based on TCN-LSTM, characterized in that: The following steps are involved: Step 1: Collect flight data of multiple parameters from the aircraft's flight control computer and virtual tests; Step 2: Perform multi-data signal level fusion; Step 3: Conduct multivariate data correlation analysis; Step 4: Perform feature-level data fusion based on TCN-LSTM; Step 5: Verify the multivariate data feature-level fusion simulation model.

2. The virtual test verification method for multivariate data fusion of a flight control system based on TCN-LSTM according to claim 1 is characterized in that: In step 1, multivariate parameter flight data is collected from the flight control computer of the aircraft and the virtual test, including flight data of 20 dimensions in five categories: load, angular rate, angle, speed, and position.

3. The virtual test verification method for multivariate data fusion of a flight control system based on TCN-LSTM according to claim 2 is characterized in that: Step 2 specifically adopts a signal-level data fusion method to fuse the flight data of multiple parameters output by the simulation model, and the flight data of multiple parameters output by the simulation model includes the flight data output by the real reference system and the virtual test simulation model.

4. The virtual test verification method for multivariate data fusion of a flight control system based on TCN-LSTM according to claim 3 is characterized in that: The specific process of step 2 is as follows: S represents the flight control system, S S and S R Respectively represent the virtual test simulation model and the real reference system; S and O R represent the multivariate output data sets of the simulation model and the real reference system under the same input conditions, so O S Fusion of simulation model multivariate output data, O R It integrates the multivariate output data of the real reference system, namely: O Sn and O Rn represents the flight data of any dimension of the simulation model and the real reference system, i represents the length of the multivariate output data time series, t1, t2, …, t i It represents the time point corresponding to the time series, and n represents the dimension of the output data, that is, the number of sensors used by the flight control system to collect flight data of this dimension.

5. The virtual test verification method for multivariate data fusion of a flight control system based on TCN-LSTM according to claim 4 is characterized in that: In step 3, the Pearson correlation coefficient PCC is used to measure the strength of the correlation between the multivariate data of the flight control system: when PCC is greater than 0, it means that there is a positive correlation between the two sets of data, and the closer the value is to 1, the stronger the linear positive correlation is; When PCC is less than 0, it means that there is a negative correlation between the two sets of data. The closer the value is to -1, the stronger the linear negative correlation is. When PCC is close to 0, it means that the correlation between the two sets of data is very weak and there is no linear relationship between the data.

6. The virtual test verification method for multivariate data fusion of a flight control system based on TCN-LSTM according to claim 5 is characterized in that: The Pearson correlation coefficient PCC is calculated by the following formula: X,Y : Among them, cov(X,Y) represents the data X i and Y i The covariance between X Represents data X i The standard deviation of X Represents data X i The mean value, σ Y Represents data Y i The standard deviation of Y Represents data Y i The mean value of , N represents the length of the flight data of the simulation model in this dimension and the flight data of the real reference system in this dimension; Construct the data correlation matrix and define r X,Y ∈[0.8,1] indicates that the data have a high correlation, r X,Y ∈[0.2,0.8) indicates that there is correlation between the data, r X,Y ∈[0,0.2) indicates that the correlation between the data is weak. The real reference system multivariate output data combination and the simulation model multivariate output data combination with high correlation are used as the real reference system multivariate output feature data set. And the simulation model multivariate output feature dataset 7. The virtual test verification method for multivariate data fusion of a flight control system based on TCN-LSTM according to claim 6 is characterized in that: Step 4 is to convert the real reference system multivariate output feature dataset And the simulation model multivariate output feature dataset As the original input data of the data reconstruction neural network model, the reconstructed data o Sj ' and o Rj ', realize the feature-level fusion of multivariate data of flight control system.

8. The virtual test verification method for multivariate data fusion of a flight control system based on TCN-LSTM according to claim 7 is characterized in that: Step 5 is as follows: Calculate the Rj ' and o Sj 'Multivariate output data, KS test was used to evaluate the o calculated under the same input conditions. Rj 'Multivariate output data and o Sj 'The CDF difference between the multivariate output data; finally the credibility level evaluation result C is obtained: Where P K-S represents the P value of KS test.

9. The virtual test verification method for multivariate data fusion of a flight control system based on TCN-LSTM according to claim 7 is characterized in that: Repeat steps 3 to 5, cyclically calculate the CDF difference after the fusion of other dimensional flight data of the simulation model and other dimensional flight data of the real reference system, and realize the verification of the multivariate data feature-level fusion simulation model of the flight control system.

Images