Danger point search method for aircraft automatic landing simulation based on cluster analysis

Abstract

The present invention discloses a dangerous point search method for aircraft automatic landing simulation based on cluster analysis, which comprises S1 performing a preset number of Monte Carlo simulations, and storing sample points with dangerous output results into a dangerous sample set; S2 normalizing the sample points in the dangerous sample set; S3 clustering the sample points in the dangerous sample set, and calculating the center point of each category after clustering; S4 generating random sample points, and adding the random sample points to the sample set when the random sample points meet the conditions; S5 judging whether the number of random sample points in the sample set reaches the preset number of added points, and if so, entering S6, otherwise returning to S4; S6 simulating the random sample points, and storing the random sample points with dangerous output results into the dangerous sample set; S7 judging whether the number of sample points in the dangerous sample set reaches the preset number of samples, and if so, entering S8, otherwise returning to S3; S8 denormalizing the sample points in the dangerous sample set to obtain the final dangerous sample set.

Description

Technical Field

[0001] The invention relates to a safety assessment technology for an aircraft automatic landing system, and in particular to a cluster analysis-based aircraft automatic landing simulation dangerous point search method. Background Art

[0002] Aircraft automatic landing system airworthiness certification requires passing the aircraft automatic landing system safety assessment. In terms of improving the safety of aircraft automatic landing systems, the optimization of the automatic landing control system is particularly critical: optimizing the automatic landing control system can improve the aircraft's ability to respond to gusts, turbulence, weight and center of gravity changes during automatic landing, thereby improving the safety of the aircraft's automatic landing system.

[0003] Researchers have determined the safety of the automatic landing system by conducting a large number of Monte Carlo simulation analyses on the automatic landing system. In most cases, the simulation results are safe. However, sometimes the simulation results may appear dangerous, and the adjustable input parameters of this simulation analysis corresponding to such simulation results are called dangerous sample points. Researchers will use optimization algorithms to locally optimize the aircraft automatic landing control system based on these dangerous sample points, thereby further improving the performance of the automatic landing control system to improve the safety of the automatic landing system. However, as the optimization proceeds, the safety of the automatic landing system becomes higher and higher, and the number of simulations required for Monte Carlo simulation to generate dangerous sample points will also increase, that is, the computational cost will increase (the simulation time required will be long).

[0004] Therefore, how to obtain dangerous sample points through fewer sample points is the need for optimizing the current aircraft automatic landing control system, the need to improve the safety of the aircraft automatic landing system, and thus the need to assist in the airworthiness certification of the aircraft automatic landing system. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present invention provides a cluster analysis-based dangerous point search method for aircraft automatic landing simulation, which solves the problem that it takes a long time to generate dangerous sample points in the prior art.

[0006] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0007] A method for searching dangerous points in aircraft automatic landing simulation based on cluster analysis is provided, which comprises the following steps:

[0008] S1. Obtain several groups of control parameters for automatic landing of an aircraft as sample points, perform a preset number of Monte Carlo simulations, and store the sample points with dangerous output results into a dangerous sample set;

[0009] S2, normalize the sample points in the dangerous sample set, and then enter step S3;

[0010] S3, clustering the sample points in the dangerous sample set using a clustering algorithm based on a distance threshold, and calculating the center point of each category after clustering;

[0011] S4, randomly generate a random sample point within 0-1, and when the distance between the same random sample point and the center point of all categories is within a preset range, add it to the sample set;

[0012] S5, determine whether the number of random sample points in the sample set reaches the preset number of points, if so, proceed to step S6, otherwise return to step S4;

[0013] S6, using a matlab simulation system to simulate random sample points in the sample set, and storing the random sample points with dangerous output results into the dangerous sample set;

[0014] S7, determine whether the number of sample points in the dangerous sample set reaches the preset number of samples, if so, proceed to step S8, otherwise return to step S3;

[0015] S8. Perform a denormalization operation on all sample points in the dangerous sample set and map them back to the upper and lower bounds specified by the MATLAB simulation system to obtain the final dangerous sample set.

[0016] Furthermore, before step S8, the method further includes:

[0017] A1. Calculate the Euclidean distance between each sample point in the risk sample set and all remaining sample points, and calculate the average Euclidean distance of all Euclidean distances of the same sample point;

[0018] A2, calculating the average value of the average Euclidean distances of all sample points, and determining whether the average value is less than a preset dispersion, if so, proceeding to step A3, otherwise proceeding to step S8;

[0019] A3. Cluster the sample points in the dangerous sample set using a clustering algorithm based on a distance threshold, and randomly delete a preset number of sample points in each category after clustering, and then return to step S3.

[0020] Furthermore, the preset range is ( d , r * d ), r is the expansion scale factor, d is the distance threshold, and its expression is:

[0021]

[0022] in, n is the dimension of the sample points.

[0023] Furthermore, the method of clustering the sample points in the dangerous sample set using a clustering algorithm based on a distance threshold includes:

[0024] B1. Calculate the distance threshold according to the dimension of the sample point:

[0025]

[0026] in, d is the distance threshold; n is the dimension of the sample point;

[0027] B2. Select an unclassified sample point and calculate the distance between it and the remaining unclassified sample points. Mark the sample points corresponding to the distances less than the distance threshold as the same category as the selected sample points.

[0028] B3. Determine whether all sample points in the dangerous sample set have been classified. If so, complete clustering; otherwise, return to step B2.

[0029] Furthermore, the normalized expression is:

[0030]

[0031] in, and are the sample points before and after normalization respectively; and are the lower and upper limits of the sample points respectively; n is the dimension of the sample point;

[0032] The expression for performing the inverse normalization operation on the sample points in the dangerous sample set in step S7 is:

[0033]

[0034] in, x It is a random sample point added to the dangerous sample set; For x The sample points obtained by inverse normalization.

[0035] Further, the control parameters include gust conditions, aircraft mass, aircraft center of gravity position, runway altitude, runway inclination, ambient temperature, glide slope and localizer offset;

[0036] The outputs of the Monte Carlo simulation include the aircraft's altitude 60m after entering the runway, the distance of the landing point relative to the runway threshold, the vertical speed of the aircraft at the landing point, the distance of the aircraft at the landing point from the runway centerline, the roll angle of the aircraft at the landing point, and the sideslip angle of the aircraft at the landing point.

[0037] The beneficial effects of the present invention are as follows: this scheme collects sample points based on the center point of each category, making full use of the clustering of the distribution of dangerous sample points themselves, and performing simulation to obtain sample points based on this. Compared with the Monte Carlo method, it is easier to find dangerous sample points, thereby improving the efficiency of obtaining dangerous sample points, that is, shortening the time to generate dangerous sample points.

[0038] By setting a preset range, this scheme can ensure that the collected sample points are not too close to the reference point (center point), which can avoid the dangerous sample points from being too concentrated, thereby ensuring the quality of the dangerous sample point set. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flow chart of the dangerous point search method for aircraft automatic landing simulation based on cluster analysis. DETAILED DESCRIPTION

[0040] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0041] refer to Figure 1 , Figure 1 The flowchart of the aircraft automatic landing simulation dangerous point search method based on cluster analysis is shown; Figure 1 As shown, the method S includes steps S1 to S8.

[0042] In step S1, several groups of control parameters of the aircraft during automatic landing are obtained as sample points, and a preset number of Monte Carlo simulations are performed, and the sample points with dangerous output results are stored in a dangerous sample set;

[0043] During implementation, the preferred control parameters of this scheme include gust conditions, aircraft mass, aircraft center of gravity position, runway altitude, runway inclination, ambient temperature, glide slope and locator offset; each control parameter needs to set its parameter symbol, distribution form, mean, standard deviation, maximum value and minimum value when inputting, and the specific distribution can be referred to Table 1.

[0044] Table 1 Distribution of control parameters

[0045]

[0046] The symbol kts in Table 1 is the unit of wind speed. ft is the unit of length, feet, tIt is the unit of weight, ton. T, H, L and R all indicate the direction of wind speed. μA Indicates current, microampere.

[0047] The output of the Monte Carlo simulation includes the height of the aircraft 60m after entering the runway, the distance of the landing point relative to the runway entrance, the vertical speed of the aircraft at the landing point, the distance between the aircraft at the landing point and the runway centerline, the roll angle of the aircraft at the landing point, and the sideslip angle of the aircraft at the landing point. Each output parameter includes the corresponding parameter symbol, hazard type, and hazard probability. The information covered by each output parameter can be referred to Table 2.

[0048] Table 2 Information covered by the parameters of the Monte Carlo simulation output

[0049]

[0050] In step S2, the sample points in the dangerous sample set are normalized:

[0051]

[0052] in, and are the sample points before and after normalization respectively; and are the lower and upper limits of the sample points respectively; n is the dimension of the sample points.

[0053] In step S3, a clustering algorithm based on a distance threshold is used to cluster the sample points in the dangerous sample set, and the center point of each category after clustering is calculated;

[0054] During implementation, the scheme preferably uses a distance threshold-based clustering algorithm to cluster sample points in the dangerous sample set, including:

[0055] B1. Calculate the distance threshold according to the dimension of the sample point:

[0056]

[0057] in, d is the distance threshold; n is the dimension of the sample point, that is, the number of parameters included in the control parameters. If it includes 9, then n =9.

[0058] B2. Select an unclassified sample point and calculate the distance between it and the remaining unclassified sample points. Mark the sample points corresponding to the distances less than the distance threshold as the same category as the selected sample points.

[0059] B3. Determine whether all sample points in the dangerous sample set have been classified. If so, complete clustering; otherwise, return to step B2.

[0060] In step S4, a random sample point is randomly generated within 0-1, and when the distance between the same random sample point and the center point of all categories is within the preset range, it is added to the sample set; wherein the preset range is ( d , r * d ), r is the expansion scale factor, d is the distance threshold. The expansion ratio is an adjustable parameter, which changes with the dimension. n The increase can be appropriately reduced. r Should be greater than 1. r The size is 1.01.

[0061] In step S5, it is determined whether the number of random sample points in the sample set reaches the preset number of added points. If so, the process proceeds to step S6, otherwise it returns to step S4; the preset number of added points is preferably 100.

[0062] In step S6, the random sample points in the sample set are simulated using a matlab simulation system, and the random sample points with dangerous output results are stored in the dangerous sample set;

[0063] In step S7, it is determined whether the number of sample points in the dangerous sample set reaches the preset number of samples. If so, the process proceeds to step S8, otherwise, it returns to step S3;

[0064] In step S8, all sample points in the dangerous sample set are denormalized and mapped back to the upper and lower bounds specified by the MATLAB simulation system to obtain the final dangerous sample set; the expression for the denormalization operation is:

[0065]

[0066] in, x It is a random sample point added to the dangerous sample set; For x The sample points obtained by inverse normalization.

[0067] In one embodiment of the present invention, before step S8, the method further includes:

[0068] A1. Calculate the Euclidean distance between each sample point in the risk sample set and all remaining sample points, and calculate the average Euclidean distance of all Euclidean distances of the same sample point;

[0069] A2, calculating the average value of the average Euclidean distances of all sample points, and determining whether the average value is less than a preset dispersion, if so, proceeding to step A3, otherwise proceeding to step S8;

[0070] A3. Cluster the sample points in the dangerous sample set using a clustering algorithm based on a distance threshold, and randomly delete a preset number of sample points in each category after clustering, and then return to step S3.

[0071] The average value in step A1 represents the degree of dispersion between the sample points in the dangerous sample point set. The larger the value, the greater the degree of dispersion. The greater the degree of dispersion, the higher the quality of the dangerous sample point set. This scheme can further ensure the quality of the selected dangerous sample points by determining the degree of dispersion.

[0072] The following is a description of the effect of the danger point generation method of this scheme with specific examples:

[0073] Taking the nine-dimensional matlab aircraft automatic landing simulation system based on the open-loop nonlinear model of a large civil aircraft as an example: according to the input control parameters, 2 dangerous sample points can be obtained by performing 10,000 Monte Carlo simulations; 29 dangerous sample points are obtained by performing 100,000 Monte Carlo simulations; the average value of these 29 dangerous sample points calculated by the method of steps A1 to A3 of this scheme is 0.8661. 264 dangerous sample points are obtained by performing 1 million Monte Carlo simulations, and the average value of the 264 dangerous sample points calculated by the method of steps A1 to A3 of this scheme is 0.8978.

[0074] Using the two dangerous sample points obtained based on the above-mentioned 10,000 small Monte Carlo simulations as the basis, steps S1 to S8 are executed, and a total of 3,800 simulation calculations are performed to obtain a total of 30 dangerous sample points (28 dangerous sample points are generated by 3,800 simulations of this scheme, and 2 dangerous sample points are generated by 10,000 Monte Carlo simulations). The average value of these 30 dangerous sample points is calculated using the method of steps A1 to A3 of this scheme, which is 0.8983. Compared with the dangerous sample points obtained by 100,000 Monte Carlo simulations, the average value of the 30 dangerous sample points obtained by this scheme is larger, that is, the discreteness of the dangerous sample points is better, which shows that the quality of the dangerous sample points generated by this scheme is higher than that directly generated by Monte Carlo simulation and it is more efficient.

[0075] Then, using the two dangerous sample points obtained based on the above-mentioned 10,000 small Monte Carlo simulations as the basis, steps S1 to S8 are executed, and a total of 43,100 simulation calculations are performed to obtain a total of 265 dangerous sample points (263+2 as above), and the average value obtained by the method of steps A1 to A3 of this scheme is 0.9568. Compared with the dangerous sample points obtained by 100,000 Monte Carlo simulations, the average value of the dangerous sample points obtained by this scheme is larger, the quality is better, and the efficiency is higher.

[0076] To sum up, the reason why this scheme can achieve such excellent results is mainly because it takes into account the clustering of the distribution of dangerous sample points themselves, and at the same time uses cluster analysis and high-dimensional space ring range to ensure that the distribution of dangerous sample points will not be too clustered, thereby ensuring the quality of the dangerous sample point set.

Claims

1. A dangerous point search method for aircraft automatic landing simulation based on cluster analysis, characterized in that: Includes steps: S1. Obtain several groups of control parameters for automatic landing of an aircraft as sample points, perform a preset number of Monte Carlo simulations, and store the sample points with dangerous output results into a dangerous sample set; S2, normalize the sample points in the dangerous sample set, and then enter step S3; S3, clustering the sample points in the dangerous sample set using a clustering algorithm based on a distance threshold, and calculating the center point of each category after clustering; S4, randomly generate a random sample point within 0-1, and when the distance between the same random sample point and the center point of all categories is within a preset range, add it to the sample set; S5, determine whether the number of random sample points in the sample set reaches the preset number of points, if so, proceed to step S6, otherwise return to step S4; S6, using a matlab simulation system to simulate random sample points in the sample set, and storing the random sample points with dangerous output results into the dangerous sample set; S7, determine whether the number of sample points in the dangerous sample set reaches the preset number of samples, if so, proceed to step S8, otherwise return to step S3; S8. Perform a denormalization operation on all sample points in the dangerous sample set and map them back to the upper and lower bounds specified by the MATLAB simulation system to obtain the final dangerous sample set.

2. The method for searching dangerous points in an aircraft automatic landing simulation according to claim 1, characterized in that: Before step S8, the method further includes: A1. Calculate the Euclidean distance between each sample point in the risk sample set and all remaining sample points, and calculate the average Euclidean distance of all Euclidean distances of the same sample point; A2, calculating the average value of the average Euclidean distances of all sample points, and determining whether the average value is less than a preset dispersion, if so, proceeding to step A3, otherwise proceeding to step S8; A3. Cluster the sample points in the dangerous sample set using a clustering algorithm based on a distance threshold, and randomly delete a preset number of sample points in each category after clustering, and then return to step S3.

3. The method for searching dangerous points in an aircraft automatic landing simulation according to claim 1, characterized in that: The preset range is ( d , r * d ), r is the expansion scale factor, d is the distance threshold, and its expression is: in, n is the dimension of the sample points.

4. The method for searching dangerous points in an aircraft automatic landing simulation according to claim 1, characterized in that: The method of clustering sample points in the dangerous sample set using a clustering algorithm based on a distance threshold includes: B1. Calculate the distance threshold according to the dimension of the sample point: in, d is the distance threshold; n is the dimension of the sample point; B2. Select an unclassified sample point and calculate the distance between it and the remaining unclassified sample points. Mark the sample points corresponding to the distances less than the distance threshold as the same category as the selected sample points. B3. Determine whether all sample points in the dangerous sample set have been classified. If so, complete clustering; otherwise, return to step B2.

5. The method for searching dangerous points in an aircraft automatic landing simulation according to claim 1, characterized in that: The normalized expression is: in, and are the sample points before and after normalization respectively; and are the lower and upper limits of the sample points respectively; n is the dimension of the sample point; The expression for performing the inverse normalization operation on the sample points in the dangerous sample set in step S7 is: in, x It is a random sample point added to the dangerous sample set; For x The sample points obtained by inverse normalization.

6. The method for searching dangerous points in an aircraft automatic landing simulation according to any one of claims 1 to 5, characterized in that: The control parameters include gust conditions, aircraft mass, aircraft center of gravity position, runway altitude, runway inclination, ambient temperature, glide slope and localizer offset; The outputs of the Monte Carlo simulation include the aircraft's altitude 60m after entering the runway, the distance of the landing point relative to the runway threshold, the vertical speed of the aircraft at the landing point, the distance of the aircraft at the landing point from the runway centerline, the roll angle of the aircraft at the landing point, and the sideslip angle of the aircraft at the landing point.

Images