Application method in flight control system virtual test verification based on data fusion

Abstract

The invention belongs to the technical field of data fusion, and particularly relates to an application method for virtual test verification of a flight control system based on data fusion. The method comprises the following steps: step 1, data acquisition; 2, data preprocessing; 3, constructing a convex optimization problem, and fusing and converting the data into convex quadratic programming; 4, solving fusion parameters; step 5, performing anomaly detection; aiming at a large-scale data fusion problem, the calculation complexity is simplified based on the convex optimization method, and the calculation efficiency is improved while the calculation accuracy is ensured; proper parameters and hyper-parameters are selected, and the parameters are further automatically selected and the algorithm is adjusted; in an actual flight control system, the data fusion algorithm of the invention has high real-time performance and real-time data processing capability. In conclusion, the method has a wide application prospect in virtual test verification of the flight control system.

Description

Technical Field

[0001] The invention belongs to the technical field of data fusion, and in particular relates to an application method based on data fusion in virtual test verification of a flight control system. Background Art

[0002] The flight control system is the core component of the aircraft, responsible for controlling the aircraft's attitude, flight path, and stability. In order to ensure the reliability and performance of the flight control system, a large number of tests and verifications are required before the actual flight. Traditional flight control system tests usually need to be conducted on actual aircraft, which is not only costly, but also poses safety risks and is difficult to control and test in complex scenarios. Therefore, it is of great significance to use virtual test technology to test and verify the flight control system.

[0003] Virtual testing is a test method based on computer simulation technology. It tests and verifies the flight control system by simulating the flight process of the aircraft and the operation of the flight control system in a computer environment. Compared with actual flight tests, virtual testing has the advantage of lower cost and can greatly save test costs and time costs. Through virtual testing, problems can be discovered and solved at an early stage, avoiding risks and losses that may be encountered in actual flight. Virtual testing can be carried out in a safe computer environment, avoiding safety risks and accidents that may exist in actual flight. Especially when testing new flight control systems or conducting tests on complex flight scenarios, virtual testing can ensure the safety of test personnel and equipment. Virtual testing can flexibly control test conditions and parameters, simulate various flight scenarios and abnormal situations to meet different test requirements. Test personnel can adjust and optimize the test according to actual needs to improve the efficiency and accuracy of the test. Virtual testing can generate a large amount of data, and can conduct comprehensive and in-depth analysis of the data, helping R&D personnel to deeply understand the performance and behavioral characteristics of the flight control system. In summary, virtual testing of flight control systems plays an important role in the development and improvement of aircraft, which can improve test efficiency, reduce costs and risks, and provide important support for the development and improvement of flight control systems.

[0004] In the virtual test verification of flight control system, data fusion is a crucial technology, which aims to improve the accuracy and reliability of flight status by integrating information from multiple sensors. Traditional data fusion methods are mainly based on statistics or machine learning techniques, such as Kalman filtering, particle filtering,

[0005] Extended Kalman filter, etc. Although these methods have achieved good results in some cases, they have certain limitations when dealing with nonlinearity, non-Gaussian noise and large amounts of data.

[0006] In recent years, data fusion methods based on convex optimization have gradually attracted attention and have been applied in virtual test verification of flight control systems. Compared with traditional methods, methods based on convex optimization have the following advantages: convex optimization problems have the characteristics of global optimal solutions, which can ensure the global optimal fusion results, thereby improving the accuracy and stability of data fusion; convex optimization theory has a rigorous mathematical foundation and proof, which can provide solid theoretical support for data fusion and ensure the reliability and stability of the algorithm; methods based on convex optimization can flexibly adapt to different data fusion scenarios and requirements, and have strong scalability and can handle large-scale data fusion problems; convex optimization methods can effectively handle nonlinear and non-Gaussian distributed data, and have wider applicability and stronger robustness.

[0007] Although the data fusion method based on convex optimization has many advantages, there are still some challenges and problems in the virtual test verification of flight control systems: for large-scale data fusion problems, the method based on convex optimization may face high computational complexity, and it is necessary to consider how to improve computational efficiency while ensuring accuracy; convex optimization methods usually require the selection of appropriate parameters and hyperparameters, which is a challenge for practical applications, and further research is needed on how to automatically select parameters and adjust algorithms; in actual flight control systems, data fusion algorithms need to have high real-time performance and the ability to process real-time data.

[0008] Therefore, how to meet the real-time requirements while ensuring accuracy is a problem that needs to be solved.

[0009] In summary, the data fusion method based on convex optimization has broad application prospects in the virtual test verification of flight control systems, but further research and optimization are still needed to solve the challenges and problems therein. Summary of the invention

[0010] Purpose of the invention: In view of the problem that the existing technology is difficult to meet the real-time requirements while ensuring accuracy, the present invention proposes an application method based on data fusion in virtual test verification of flight control system. The data fusion algorithm needs to have high real-time performance and strong processing ability for real-time data, which effectively meets the needs of virtual test verification of flight control system.

[0011] The technical solution of the present invention:

[0012] The application method of data fusion in virtual test verification of flight control system includes the following steps:

[0013] Step 1: Data collection;

[0014] Step 2: Data preprocessing;

[0015] Step 3: Convex optimization problem construction;

[0016] Step 4: solving fusion parameters;

[0017] Step 5: Anomaly detection.

[0018] Further, in step 1, flight data of the aircraft is collected from sensor data of the virtual test of the flight control system, and the data is converted into a csv format; the flight data includes but is not limited to: axial overload, lateral overload, normal overload, pitch angle rate, roll angle rate, yaw angle rate, angle of attack, sideslip angle, flight track angle, track inclination angle, true heading angle, pitch angle, roll angle, true airspeed, ground speed, static pressure, dynamic pressure, celestial speed, landing gear, aircraft center of gravity, aircraft weight, left canard rudder surface position, left ETU turning angle, left inner aileron rudder surface position, left front flap rudder surface inner position, left outer aileron rudder surface position, left rudder rudder surface position, right canard rudder surface position, right ETU turning angle, right inner aileron rudder surface position, right front flap rudder surface inner position, right outer aileron rudder surface position, and right rudder rudder surface position.

[0019] Furthermore, step 2 specifically resamples and normalizes the data in csv format, that is, resamples the flight data with a high sampling frequency, and then uses a normalization method to limit all flight data to the range of 0-1.

[0020] Furthermore, step 3 of the convex optimization problem construction is to transform the data fusion problem into a convex quadratic programming problem, including:

[0021] Build the fusion target:

[0022] The first fusion goal is to reduce the volatility of data. Variance is used to measure the volatility of flight data. The variance of flight data is estimated through matrix S, and a quadratic form is constructed to ensure the convexity of the constructor.

[0023] The second fusion objective is to maximize the change in the fault flight data, measured by the first-order difference in the data after the fault; using the matrix E i To evaluate the variation of flight data, and use the parameter γ ij The constraints of the fusion make the fused data show an overall upward trend after the failure;

[0024] Combine the fusion objectives to form the objective function, solve the objective function, and obtain the optimal fusion parameters:

[0025] After constructing the fusion target, the hyperparameters α, β, c are introduced ij Optimize the solution strategy, where α and β are used to balance the importance of the two fusion objectives, c ijUsed to adjust the weight of post-fault data; by combining the two fusion objectives, the objective function of the convex optimization problem is formed, and the objective function is solved to obtain the optimal fusion parameters, so as to make the optimal combination of multiple parameters with different noise, distribution characteristics and fault characterization speed to maximize the fault characteristics.

[0026] Furthermore, the objective function of the convex optimization problem is:

[0027]

[0028] swt T 1=1,E i W i ≤0……(2)

[0029] i=1,…,n,j=1,…,o i

[0030] Among them, α, β, c ij is a hyperparameter used to adjust the solution strategy of the optimization algorithm. Formula (1) is the first fusion goal, which is to evaluate the fluctuation of the fusion data by estimating the variance of the fusion data and to ensure the convexity of the objective function by constructing a quadratic form. Formula (2) is the second fusion goal, which is to combine the constraint term E i W i ≤0 maximizes the difference in the post-failure data. There are three hyperparameters in the equation, where α and β work together to adjust the importance of the two optimization objectives, and c ij With parameter γ ij They work together to adjust the differences in the post-fault data; hyperparameters α and β are obtained through empirical analysis, and the sum of these two parameters is a constant, so the appropriate parameter ratio can effectively adjust the optimization effect;

[0031] w is the fusion weight matrix, which determines the weights corresponding to each set of parameters in the fusion; X i represents the fault-free dataset used in training, where rows correspond to the collected data and columns correspond to parameters; o i is the number of measurement points of the post-fault data, and the S matrix is ​​a symmetric matrix with the following form:

[0032]

[0033] Where I is the identity matrix, J is the matrix of all 1s, and n i represents the number of data points collected without failure, and the matrix S is used to estimate the variance of the fused data; c ij is used as a hyperparameter to determine how much the post-failure data affects the overall optimization objective, and γ ij Used to maximize the difference between adjacent data after a failure.

[0034] Further, step 4: solving fusion parameters;

[0035] The objective function of the convex optimization problem is formed by introducing α, β, c ij The three hyperparameters form a multi-objective optimization problem:

[0036] The hyperparameter α is designed to enhance the smoothness of the fused data when there is no fault, and the hyperparameter β is designed to enhance the variation of the data after the fault. Through different configurations of α and β, the marginal effect is used to find the optimal hyperparameters of the multi-objective optimization problem.

[0037] After solving for the optimal hyperparameters, the interior point method is used to solve for the optimal fusion parameters.

[0038] Furthermore, the interior point method is used to solve the optimal fusion parameters:

[0039] The basic idea of ​​the interior point method is expressed as follows: For convex optimization problems, the problem is expressed as finding the minimum value of a convex function f(x) within a certain value range A;

[0040] Therefore, the objective function of the convex optimization problem is transformed into the following standard form of optimization equation:

[0041] min f(x)x∈A

[0042] stf(x)≤b

[0043] h(x)=0

[0044] Among them, f(x) is a hypothetical convex function, A is the domain of x, and f(x)≤b, h(x)=0, is the hypothetical constraint range. Then, the range A is determined by the constraints in the above equation, that is, x∈A is equivalent to the constraint; then, a barrier function g(x) is constructed to satisfy that the function value outside the range A is much larger than the function value within the range A, that is, it satisfies:

[0045]

[0046] The standard form of the optimization equation is a function under ideal conditions. The barrier function that satisfies the ideal condition usually does not exist. A logarithmic barrier function is used to approximate the barrier function under ideal conditions. The logarithmic barrier function is:

[0047] g(x)=μln(f(x))

[0048] where μ is a very small positive parameter, and g(x) is the value of x in the complex domain;

[0049] At this time, construct a new function h(x): h(x) = f(x) + g(x)

[0050] When g(x) = μln(f(x)), the function value of h(x) outside the range of A is much larger than the value of f(x). At the same time, the value of h(x) within the range of A is equal to the original value of f(x), that is, the minimum value of h(x) is the minimum value of f(x); that is, the optimization problem is equivalent to the following problem:

[0051] min h(x)x∈C

[0052] Where C is the field of complex numbers, which is equivalent to obtaining the constraints in the original problem and including them in the functional properties of the newly constructed function h(x);

[0053] Since the solution space of the new function h(x) is unbounded, the Newton method is used to find the minimum value of the new function h(x). The solution obtained is exactly equal to the minimum solution of the convex function f(x) of the original problem within a certain range A, that is, the minimum value of the new function h(x) is the optimal solution of the original optimization problem within the constraints. After obtaining the hyperparameters α, β, c i j Afterwards, the optimal fusion parameter w of the data is obtained by solving the optimal solution of the convex quadratic programming problem, thereby realizing data fusion.

[0054] Further, step 5 is to use the exponentially weighted moving average-cumulative and EWMA-cumulative sum algorithms to implement anomaly detection based on fused data.

[0055] Beneficial effects of the present invention:

[0056] The present invention proposes an application method based on data fusion in virtual test verification of flight control system. For large-scale data fusion problems, the method based on convex optimization simplifies the computational complexity, ensures the computational accuracy and improves the computational efficiency; selects appropriate parameters and hyperparameters, further automates the parameter selection and adjustment algorithm; in the actual flight control system, the data fusion algorithm of the present invention has high real-time performance and the ability to process real-time data. In summary, the method of the present invention has broad application prospects in virtual test verification of flight control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 is a flow chart of the present invention;

[0058] Figure 2 This is the flow chart of the flight control simulation system;

[0059] Figure 3 Schematic diagram of angle of attack prediction results and anomaly detection. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] Embodiment 1:

[0062] The present invention proposes an application method based on data fusion in virtual test verification of flight control system, which is aimed at virtual test verification of flight control system. The method flow is as follows: Figure 1 As shown, the present invention realizes the application of the convex optimization data fusion method in the virtual test verification of the flight control system through the following steps:

[0063] Step 1: Data collection: Collect flight data of N parameters from the sensor data of the virtual test of the flight control system and convert the data into csv format.

[0064] Step 2: Data preprocessing; Different sensors have different operating frequencies and data characteristics, which affect the quality of fused data. Through the data preprocessing link, the data of each sensor is converted into data with different characteristics to achieve the fusion of multi-sensor data. Data preprocessing mainly includes resampling and standardization to achieve data fusion with different sampling frequencies and eliminate the scale differences between different data. In this method, the fusion problem of multimodal data is simplified, and the problems of data reception and alignment in real situations are not considered, but the focus is on the design of the optimal fusion scheme. Power system data with higher sampling frequencies are resampled so that they can be trained and fused with control system data. Since the construction of convex optimization problems requires a positive definition of the matrix, which requires the final fused data set to have positive covariates, normalization is used to limit the data to the range of 0-1 for optimal fusion.

[0065] Step 3: Convex optimization problem construction: In order to obtain the best fusion effect and achieve fast and accurate anomaly detection, the data fusion problem is transformed into a convex quadratic programming problem, including:

[0066] Build the fusion target:

[0067] The first fusion goal is to reduce the volatility of data. Variance is used to measure the volatility of flight data. The variance of flight data is estimated through matrix S, and a quadratic form is constructed to ensure the convexity of the constructor.

[0068] The second fusion objective is to maximize the change in the fault flight data, measured by the first-order difference in the data after the fault; using the matrix E i To evaluate the variation of flight data, and use the parameter γ ijThe constraints of the fusion make the fused data show an overall upward trend after the failure;

[0069] Combine the fusion objectives to form the objective function, solve the objective function, and obtain the optimal fusion parameters:

[0070] After constructing the fusion target, the hyperparameters α, β, c are introduced ij Optimize the solution strategy, where α and β are used to balance the importance of the two fusion objectives, c ij It is used to adjust the weight of post-fault data; by combining the two fusion objectives, the objective function of the convex optimization problem is formed, and the objective function is solved to obtain the optimal fusion parameters, so that the optimal combination of multiple parameters with different noise, distribution characteristics and fault characterization speed can maximize the fault characteristics. Through the convex optimization problem objective function, the power system parameters with large observation noise and obvious fault characterization can be fused with the control system parameters with good observation quality to maximize the difference between the data before and after the fault.

[0071] The objective function of the convex optimization problem is:

[0072]

[0073] swt T 1=1,E i W i ≤0……(2)

[0074] i=1,…,n,j=1,…,o i

[0075] Among them, α, β, c ij is a hyperparameter used to adjust the solution strategy of the optimization algorithm. Formula (1) is the first fusion goal, which is to evaluate the fluctuation of the fusion data by estimating the variance of the fusion data and to ensure the convexity of the objective function by constructing a quadratic form. Formula (2) is the second fusion goal, which is to combine the constraint term E i W i ≤0 maximizes the difference in the post-failure data. There are three hyperparameters in the equation, where α and β work together to adjust the importance of the two optimization objectives, and c ij With parameter γ ij They work together to adjust the differences in the post-fault data; hyperparameters α and β are obtained through empirical analysis, and the sum of these two parameters is a constant, so the appropriate parameter ratio can effectively adjust the optimization effect;

[0076] w is the fusion weight matrix, which determines the weights corresponding to each set of parameters in the fusion; X i represents the fault-free dataset used in training, where rows correspond to the collected data and columns correspond to parameters; o iis the number of measurement points of the post-fault data, and the S matrix is ​​a symmetric matrix with the following form:

[0077]

[0078] Where I is the identity matrix, J is the matrix of all 1s, and n i represents the number of data points collected without failure, and the matrix S is used to estimate the variance of the fused data; c ij is used as a hyperparameter to determine how much the post-failure data affects the overall optimization objective, and γ ij Used to maximize the difference between adjacent data after a failure.

[0079] Step 4: solving fusion parameters;

[0080] The objective function of the convex optimization problem is formed by introducing α, β, c ij The three hyperparameters form a multi-objective optimization problem:

[0081] The hyperparameter α is designed to enhance the smoothness of the fused data when there is no fault, and the hyperparameter β is designed to enhance the variation of the data after the fault. Through different configurations of α and β, the marginal effect is used to find the optimal hyperparameters of the multi-objective optimization problem.

[0082] After solving for the optimal hyperparameters, the interior point method is used to solve for the optimal fusion parameters.

[0083] Furthermore, the interior point method is used to solve the optimal fusion parameters:

[0084] The basic idea of ​​the interior point method is expressed as follows: For convex optimization problems, the problem is expressed as finding the minimum value of a convex function f(x) within a certain value range A;

[0085] Therefore, the objective function of the convex optimization problem is transformed into the following standard form of optimization equation:

[0086] min f(x)x∈A

[0087] stf(x)≤b

[0088] h(x)=0

[0089] Among them, f(x) is a hypothetical convex function, A is the domain of x, and f(x)≤b, h(x)=0, is the hypothetical constraint range. Then, the range A is determined by the constraints in the above equation, that is, x∈A is equivalent to the constraint; then, a barrier function g(x) is constructed to satisfy that the function value outside the range A is much larger than the function value within the range A, that is, it satisfies:

[0090]

[0091] The standard form of the optimization equation is a function under ideal conditions. The barrier function that satisfies the ideal condition usually does not exist. A logarithmic barrier function is used to approximate the barrier function under ideal conditions. The logarithmic barrier function is:

[0092] g(x)=μln(f(x))

[0093] where μ is a very small positive parameter, and g(x) is the value of x in the complex domain;

[0094] At this time, construct a new function h(x): h(x) = f(x) + g(x)

[0095] When g(x) = μln(f(x)), the function value of h(x) outside the range of A is much larger than the value of f(x). At the same time, the value of h(x) within the range of A is equal to the original value of f(x), that is, the minimum value of h(x) is the minimum value of f(x); that is, the optimization problem is equivalent to the following problem:

[0096] min h(x)x∈C

[0097] Where C is the field of complex numbers, which is equivalent to obtaining the constraints in the original problem and including them in the functional properties of the newly constructed function h(x);

[0098] Since the solution space of the new function h(x) is unbounded, the Newton method is used to find the minimum value of the new function h(x). The solution obtained is exactly equal to the minimum solution of the convex function f(x) of the original problem within a certain range A, that is, the minimum value of the new function h(x) is the optimal solution of the original optimization problem within the constraints. After obtaining the hyperparameters α, β, c i j Afterwards, the optimal fusion parameter w of the data is obtained by solving the optimal solution of the convex quadratic programming problem, thereby realizing data fusion.

[0099] Step 5 is to use the exponentially weighted moving average-cumulative and EWMA-cumulative sum algorithms to implement anomaly detection based on fused data.

[0100] The optimization goal of the constructed optimization problem is to maximize the anomalies of normal data and fault data. At the same time, considering the real-time requirements of abnormality detection of the flight control system, the detection algorithm should be able to quickly calculate and realize detection, and be sensitive to faults appearing in the fused data. The method of the present invention uses the exponentially weighted moving average-cumulative sum EWMA-cumulative sum algorithm to realize anomaly detection based on fused data. The algorithm can monitor the data stream in real time, has low computational complexity, strong adaptability, and has the characteristics of smoothing parameters adjusting over time. It can always maintain monitoring under time-varying conditions, is suitable for the characteristics of fused data, can realize rapid diagnosis, and can also reflect the effect of data fusion.

[0101] Embodiment 2: An application method based on data fusion in virtual test verification of flight control system, comprising the following steps:

[0102] Step 1, such as Figure 2 As shown. Combined with the flight control system simulation, the time series data of the control system and the power system during the flight can be obtained. The sensor data of the virtual test of the flight control system is collected to obtain the flight data of each flight parameter under normal conditions. The flight parameters are axial overload, lateral overload, normal overload, pitch rate, roll rate, yaw rate, angle of attack, sideslip angle, flight track angle, track inclination angle, true heading angle, pitch angle, roll angle, true airspeed, ground speed, static pressure, dynamic pressure, celestial speed, landing gear, aircraft center of gravity, aircraft weight, left canard rudder position, left ETU turning angle, left inner aileron rudder position, left front flap rudder inner position, left outer aileron rudder position, left rudder rudder position, right canard rudder position, right ETU turning angle, right inner aileron rudder position, right front flap rudder inner position, right outer aileron rudder position, right rudder rudder position, etc.

[0103] Step 2: Normalize and slide window all data. First, calculate the maximum value x of each flight parameter. max and the minimum value x min , scale the original data x to the interval [0,1] to obtain the normalized flight data Then, the data of the training set and the test set are divided into samples by combining sliding windows and slicing, and the sliding step is set to 100. The historical slice data is used to predict the data of the navigation parameters to be tested at the current moment, and the sliding window sampling operation of the data is completed. t =(x t-1 ,x t-2 ,…x t-D ), where t is the timestamp of the flight data and D is the sliding step. t =(x t-1 ,x t-2 ,…x t-D ) as the data input into the model at time t, Y t =y t is the output target value of the model, that is, the predicted value of the parameter to be tested at time t.

[0104] Step 3, in order to obtain the best fusion effect and achieve fast and accurate anomaly detection, the fusion problem is transformed into a convex quadratic programming problem. The first fusion goal is to reduce the volatility of the data. This paper uses variance to measure the volatility of the data. The variance of the data is estimated by the matrix S, and a quadratic form is constructed based on it to ensure the convexity of the constructor. The second fusion goal is to maximize the change of the fault data, which is measured by the first-order difference in the data after the fault. This paper uses the matrix E iTo evaluate the amount of change in the data, and use the parameter γ ij After constructing the fusion target, the hyperparameters α, β, c are introduced. ij Optimize the solution strategy, where α and β are used to balance the importance of the two objectives, c ij Used to adjust the weight of post-fault data. By combining the two fusion objectives to form an objective function, solving the objective function, and obtaining the optimal fusion parameters, the optimal combination of multiple parameters with different noise, distribution characteristics, and fault characterization speeds can maximize the fault characteristics. Therefore, the objective function of the convex optimization problem proposed in this paper is as follows. Through this objective function, the power system parameters with large observation noise and obvious fault characterization can be fused with the control system parameters with good observation quality to maximize the difference between the data before and after the fault.

[0105]

[0106] swt T 1=1,E i W i ≤0

[0107] i=1,…,n,j=1,…,o i

[0108] In step 4, the objective function forms a multi-objective optimization problem by introducing three hyperparameters: the first hyperparameter is designed to enhance the smoothness of the fused data when there is no fault, and the second hyperparameter is designed to enhance the variation of the data after the fault. Through different configurations of α and β, the marginal effect is used to find the optimal hyperparameters of the multi-objective optimization problem. After solving the obtained hyperparameters, the interior point method can be used to solve the optimal fusion parameters to solve the optimization problem in the previous section. The boundary point method is a numerical optimization method used to solve convex optimization problems by searching for solutions in the interior of feasible solutions to avoid a large number of iterative steps that may occur in many exterior point methods.

[0109] Step 5: The simulation takes the flight control system as the research object. Considering that the angle sensor data in the flight control can characterize the occurrence of faults more quickly, the angle of attack θ is selected. attack , sideslip angle θ slide , pitch angle θ pitch , pitch angle rate w pitch , roll angular rate w row , yaw rate w yaw Six parameters are used as the fusion data set, which is composed as follows:

[0110] [θ attack ,θ slide ,θ pitch ,w pitch ,wrow ,w yaw ]

[0111] Step 6, the simulation time is set to 80s, and the angle of attack is affected by the shock wave and the spike anomaly is injected into the simulation model at 50s. This kind of fault is common in the flight control system, which may lead to inaccurate aircraft control or even failure of the flight control system. After selecting the fusion data set, the simulation platform is used to construct the training data, and the random fault factors and attitude control targets (initial attitude and end attitude) are obtained through Monte Carlo simulation to form 25 sets of data sets. Five groups in the data set were selected as training data sets, and twenty groups were selected as test data sets. The data set scenario configuration is shown in the following table. The parameter range of Monte Carlo sampling used to simulate the real environment is shown in the equation. Different fault factors affect the degree of fault, and different attitude angle control instructions change the trajectory and control scheme of the aircraft, making the simulation random and allowing the generalization ability of the algorithm to be verified.

[0112] 50s after the fault occurs, the parameters fluctuate to varying degrees. After data preprocessing, the optimal fusion parameters can be obtained by constructing and solving a convex quadratic programming problem, and the marginal effect is used to solve the change process of the above hyperparameter α. Since α can reflect the importance of the two components in the optimization objective function, this paper selects α = 0.7, because at this time the variance of the pre-fault data begins to slow down, and selecting this parameter can achieve the maximum gain of the objective function. After determining the hyperparameters, the optimal parameters are obtained by solving the optimization problem. The final detection results are as follows Figure 3 shown.

[0113] This embodiment proposes a data fusion method based on convex optimization to propose an anomaly detection method for power and control system data fusion. This method solves the problem of insufficient observation information inside the flight control system in real situations, and integrates data from other systems at the data level to achieve fast and accurate anomaly detection. Figure 3 It can be seen that the fused data can enhance the difference in data characteristics before and after the fault. Compared with before fusion, the accuracy rate is improved by 17.7% and the detection time is shortened by 24%. In addition, the algorithm has higher accuracy and faster speed than similar anomaly detection algorithms.

[0114] The above is only a specific embodiment of the present invention, and the present invention is described in detail. The unexplained part is a conventional technology. However, the protection scope of the present invention is not limited to this. Any changes or substitutions that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. The application method of data fusion in virtual test verification of flight control system is characterized by: The following steps are involved: Step 1: Data collection; Step 2: Data preprocessing; Step 3: Convex optimization problem construction, transforming data fusion into convex quadratic programming; Step 4: solving fusion parameters; Step 5: Anomaly detection.

2. The application method of data fusion in virtual test verification of flight control system according to claim 1 is characterized in that: In step 1, the flight data of the aircraft is collected from the sensor data of the virtual test of the flight control system, and the data is converted into a csv format; the flight data includes but is not limited to: axial overload, lateral overload, normal overload, pitch rate, roll rate, yaw rate, angle of attack, sideslip angle, flight track angle, track inclination angle, true heading angle, pitch angle, roll angle, true airspeed, ground speed, static pressure, dynamic pressure, celestial speed, landing gear, aircraft center of gravity, aircraft weight, left canard rudder surface position, left ETU turning angle, left inner aileron rudder surface position, left front flap rudder surface inner position, left outer aileron rudder surface position, left rudder rudder surface position, right canard rudder surface position, right ETU turning angle, right inner aileron rudder surface position, right front flap rudder surface inner position, right outer aileron rudder surface position, right rudder rudder surface position.

3. The application method of data fusion in virtual test verification of flight control system according to claim 2 is characterized in that: Step 2 specifically resamples and normalizes the data in csv format, that is, resamples the flight data with a high sampling frequency, and then uses the normalization method to limit all flight data to the range of 0-1.

4. The application method of data fusion in virtual test verification of flight control system according to claim 3 is characterized in that: Step 3 transforms the data fusion problem into a convex quadratic programming problem, including: Build the fusion target: The first fusion goal is to reduce the volatility of data. Variance is used to measure the volatility of flight data. The variance of flight data is estimated through matrix S, and a quadratic form is constructed to ensure the convexity of the constructor. The second fusion objective is to maximize the change in the fault flight data, measured by the first-order difference in the data after the fault; using the matrix E i To evaluate the variation of flight data, and use the parameter γ ij The constraints of the fusion make the fused data show an overall upward trend after the failure; Combine the fusion objectives to form the objective function, solve the objective function, and obtain the optimal fusion parameters: After constructing the fusion target, the hyperparameters α, β, c are introduced ij Optimize the solution strategy, where α and β are used to balance the importance of the two fusion objectives, c ij Used to adjust the weight of post-fault data; by combining the two fusion objectives, the objective function of the convex optimization problem is formed, and the objective function is solved to obtain the optimal fusion parameters, so as to make the optimal combination of multiple parameters with different noise, distribution characteristics and fault characterization speed to maximize the fault characteristics.

5. The application method of data fusion in virtual test verification of flight control system according to claim 4 is characterized in that: The objective function of the convex optimization problem is: s.t.w T 1=1,E i W i ≤0……(2) i=1,…,n,j=1,…,o i Among them, α, β, c ij is a hyperparameter used to adjust the solution strategy of the optimization algorithm. Formula (1) is the first fusion goal, which is to evaluate the fluctuation of the fusion data by estimating the variance of the fusion data and to ensure the convexity of the objective function by constructing a quadratic form. Formula (2) is the second fusion goal, which is to combine the constraint term E i W i ≤0 maximizes the difference in the post-failure data. There are three hyperparameters in the equation, where α and β work together to adjust the importance of the two optimization objectives, and c ij With parameter γ ij They work together to adjust the differences in the post-fault data; hyperparameters α and β are obtained through empirical analysis, and the sum of these two parameters is a constant, so the appropriate parameter ratio can effectively adjust the optimization effect; w is the fusion weight matrix, which determines the weights corresponding to each set of parameters in the fusion; X i represents the fault-free dataset used in training, where rows correspond to the collected data and columns correspond to parameters; o i is the number of measurement points of the post-fault data, and the S matrix is ​​a symmetric matrix with the following form: Where I is the identity matrix, J is the matrix of all 1s, and n i represents the number of data points collected without failure, and the matrix S is used to estimate the variance of the fused data; c ij is used as a hyperparameter to determine how much the post-failure data affects the overall optimization objective, and γ ij Used to maximize the difference between adjacent data after a failure.

6. The application method of data fusion in virtual test verification of flight control system according to claim 4 is characterized in that: Step 4: solving fusion parameters; The objective function of the convex optimization problem is formed by introducing α, β, c ij The three hyperparameters form a multi-objective optimization problem: The hyperparameter α is designed to enhance the smoothness of the fused data when there is no fault, and the hyperparameter β is designed to enhance the variation of the data after the fault. Through different configurations of α and β, the marginal effect is used to find the optimal hyperparameters of the multi-objective optimization problem. After solving for the optimal hyperparameters, the interior point method is used to solve for the optimal fusion parameters.

7. The application method of data fusion in virtual test verification of flight control system according to claim 6 is characterized in that: The interior point method is used to solve the optimal fusion parameters: The basic idea of ​​the interior point method is expressed as follows: For convex optimization problems, the problem is expressed as finding the minimum value of a convex function f(x) within a certain value range A; Therefore, the objective function of the convex optimization problem is transformed into the following standard form of optimization equation: min f(x)x∈A stf(x)≤b h(x)=0 Among them, f(x) is a hypothetical convex function, A is the domain of x, and f(x)≤b, h(x)=0, is the hypothetical constraint range. Then, the range A is determined by the constraints in the above equation, that is, x∈A is equivalent to the constraint; then, a barrier function g(x) is constructed to satisfy that the function value outside the range A is much larger than the function value within the range A, that is, it satisfies: The standard form of the optimization equation is a function under ideal conditions. The barrier function that satisfies the ideal condition usually does not exist. A logarithmic barrier function is used to approximate the barrier function under ideal conditions. The logarithmic barrier function is: g(x)=μln(f(x)) where μ is a very small positive parameter, and g(x) is the value of x in the complex domain; At this time, construct a new function h(x): h(x) = f(x) + g(x) When g(x) = μln(f(x)), the function value of h(x) outside the range of A is much larger than the value of f(x). At the same time, the value of h(x) within the range of A is equal to the original value of f(x), that is, the minimum value of h(x) is the minimum value of f(x); that is, the optimization problem is equivalent to the following problem: min h(x)x∈C where C is the field of complex numbers, which is equivalent to taking the constraints in the original problem and including them in the functional properties of the newly constructed function h(x).

8. The application method based on data fusion in virtual test verification of flight control system according to claim 7 is characterized in that: The solution space of the new function h(x) is unbounded. The minimum value of the new function h(x) is found by Newton's method. The obtained solution is equal to the minimum solution of the convex function f(x) of the original problem within a certain range A, that is, the minimum value of the new function h(x) is the optimal solution of the original optimization problem within the constraints. After obtaining the hyperparameters α, β, c ij Afterwards, the optimal fusion parameter w of the data is obtained by solving the optimal solution of the convex quadratic programming problem, thereby realizing data fusion.

9. The application method based on data fusion in virtual test verification of flight control system according to claim 7 is characterized in that: Step 5 is to use the exponentially weighted moving average-cumulative and EWMA-cumulative sum algorithms to implement anomaly detection based on fused data.

Images