A method for intention recognition and behavior prediction of non-cooperative targets in high orbit

Abstract

The present invention discloses a method for high-orbit non-cooperative target intention recognition and behavior prediction, including establishing and based on a universal description model of high-orbit non-cooperative targets, calculating the maneuvering sequence under various intentions, and constructing a target maneuvering behavior sequence library; identifying the maneuvering situation of the target, and matching the maneuvering situation with the target maneuvering behavior sequence library through a Bayesian classifier, and calculating the probability of various intentions; using the probability of the previous intention as the prior probability of the next Bayesian classifier processing, until the probability of a certain intention is greater than a specified threshold, the intention recognition is completed, and the target subsequent maneuvering behavior prediction results are extracted from the target maneuvering behavior sequence library. It overcomes the shortcomings of existing intention recognition methods that only target non-maneuvering targets; and achieves high-precision intention recognition and behavior prediction for maneuvering non-cooperative targets.

Description

Technical Field

[0001] The present invention relates to the field of aerospace technology, and in particular to a method for high-orbit non-cooperative target intention recognition and behavior prediction. Background Art

[0002] With the continuous deepening of space exploration and the continuous upgrading of space missions, the intention recognition and parameter estimation of space non-cooperative targets have become key technologies for on-orbit service missions. The difficulty of the intention recognition problem of non-cooperative targets lies in the lack of cognition of the intention of space targets and the great uncertainty of targets with maneuvering capabilities. In existing studies, most of them consider the intention recognition of non-maneuvering targets. Under the assumption that the target enters the mission orbit, the mission orbit type is classified as the intention, and artificial intelligence technology is used for identification. In the existing technology, the intention of the target is assumed to include "fly-by observation", "electromagnetic interference", "fly-by detection", "terminal flyby", "hover operation" and "no threat", a total of six cases. A deep neural network is used to identify the intention of aerial targets. The attributes of the aerial targets such as maneuver type, speed, altitude, position and the corresponding intention labels are input into the deep neural network model as samples, the network node weights are adjusted, and finally the trained model is used to predict the intention of the target. The existing technology has the problems of being unable to model the intention of non-cooperative targets with subsequent maneuvers and being unable to deal with the uncertainty caused by multiple maneuvers of the target. Therefore, a method for intention recognition and behavior prediction of high-orbit non-cooperative targets is urgently needed to solve the above problems. Summary of the invention

[0003] The purpose of the present invention is to provide a method for high-orbit non-cooperative target intention recognition and behavior prediction, which can effectively solve the problems existing in the above-mentioned prior art.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions: a method for high-orbit non-cooperative target intention recognition and behavior prediction, establishes and is based on a general description model of high-orbit non-cooperative targets, calculates maneuver sequences under various intentions, including maneuver time and pulse speed increment, and constructs a target maneuver behavior sequence library, and performs the following operations based on the target maneuver behavior sequence library:

[0005] Identify the target's maneuvers and match them with the target's maneuver behavior sequence library through a Bayesian classifier to calculate the probability of each type of intention; and

[0006] The probability of the previous intention is used as the prior probability for the subsequent Bayesian classifier processing until the probability of a certain intention is greater than the specified threshold. The intention recognition is completed, and the prediction results of the target's subsequent maneuvering behavior are extracted from the target maneuvering behavior sequence library.

[0007] Preferably, the orbital game intention of the non-cooperative target can be expressed as the process of the target spacecraft transferring from the original orbit to the mission orbit that forms a specific relative motion to itself. Therefore, the target orbital game process is decomposed into the transfer segment and the mission segment, and a general description model of high-orbit non-cooperative targets of "mission orbit" + "transfer orbit" is established, wherein the model includes N types of typical intentions of the target in the high-orbit scenario.

[0008] Preferably, for the mission segment: based on the CW relative dynamics model, the design is carried out according to the parameters such as the mission relative distance and relative speed constraints and the load working cycle, including various relative orbit styles such as fly-by and fly-by;

[0009] For the transfer segment: take the solved mission orbit as the rendezvous target, optimize the pulse number, pulse rules and performance indicators according to the game strategy, and obtain the transfer trajectory.

[0010] Preferably, during the target intention recognition process, the orbital state of the own spacecraft is , get the target's orbit information .

[0011] Preferably, according to N types of typical intentions, particle swarm optimization algorithm is used to perform optimization and solve respectively to obtain optimization variables that optimize the performance index; and

[0012] Based on the optimization variable results, the maneuver sequences under various intentions are calculated and a target maneuver behavior sequence library is constructed.

[0013] Preferably, the target maneuvering behavior sequence library is:

[0014] ;

[0015] Where T is the maneuvering time, is the speed increment, the superscript i indicates the sequence number, , the subscript is the maneuver number, and n represents the total number of maneuvers.

[0016] Preferably, the target is identified by a movement detection system, including the time of maneuver With maneuvering speed vector , denoted as ;

[0017] The Bayesian classifier is used to match and calculate the probability of various intents, including:

[0018] Calculate the matching difference:

[0019] ;

[0020] in, are normalized parameters for measuring time difference, velocity increment amplitude difference and velocity direction difference respectively;

[0021] Based on the difference , calculate the likelihood:

[0022] ;

[0023] in, is an adjustable parameter used to control the width of the distribution; assuming that the prior probability of each intention is equal, then:

[0024] ;

[0025] Based on the likelihood results, the Bayesian formula is calculated to obtain the posterior probabilities of various intentions:

[0026] ;

[0027] in, For all possible The sum of:

[0028] ;

[0029] Therefore, the recognition probability of each intention is obtained according to the maneuver matching:

[0030] .

[0031] Preferably, They are:

[0032] .

[0033] Preferably, the previous is the prior probability, and the probability of the next intention recognition is calculated ; and repeat until the jth maneuver, the probability of a certain intention is greater than the given probability threshold , then the target intention recognition is considered to be completed, and the subsequent maneuvers under this intention in the target maneuver behavior sequence library are the behavior prediction results for the target : .

[0034] Beneficial effects: The present invention establishes a universal description model of high-orbit non-cooperative targets of "mission orbit" + "transfer orbit" and defines typical intentions in high-orbit scenarios, thereby making up for the problem of missing intention definition of maneuvering targets in existing studies. The maneuvering situation is matched with the target maneuvering behavior sequence library through a Bayesian classifier, and the probability of each type of intention is calculated, which overcomes the deficiency of existing intention recognition methods that only target non-maneuvering targets. Numerical simulation results show that the recognition accuracy of this technology reaches 93%, and it can accurately predict the subsequent maneuvering behavior sequence of the target. The prediction error of the maneuvering time window does not exceed 427s (3σ), and the relative error of the maneuvering speed increment is 1.06% (3σ). The results show that the present invention achieves high-precision intention recognition and behavior prediction for maneuvering non-cooperative targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0036] In the attached picture:

[0037] Figure 1 is a flow chart of the intention recognition and behavior prediction method of the present invention;

[0038] Figure 2 It is a schematic diagram of the universal description model of the high-orbit non-cooperative target of the target "mission orbit" + "transfer orbit";

[0039] Figure 3 Schematic diagram of the four types of intent used in the test case;

[0040] Figure 4 Comparison of the true intent and intent recognition results in the Monte Carlo simulation results for the test case;

[0041] Figure 5 Box plot of the maneuver time error distribution in the Monte Carlo simulation of the test case;

[0042] Figure 6 Box plot of the relative error distribution of maneuvering speed increment in the Monte Carlo simulation of the test case. DETAILED DESCRIPTION

[0043] The embodiments of the present invention are described below in conjunction with the drawings in the embodiments of the present invention. The terms used in the implementation mode of the present invention are only used to explain the specific embodiments of the present invention, and are not intended to limit the present invention. The embodiments of the present application are described below in conjunction with the drawings.

[0044] Example: Figure 1 As shown, a high-orbit non-cooperative target intention recognition and behavior prediction method includes the following steps:

[0045] S1. Establish a universal description model of high-orbit non-cooperative targets of "mission orbit" + "transfer orbit" and define typical intentions in high-orbit scenarios;

[0046] Among them, the orbital game intention of the non-cooperative target can be expressed as the process of the target spacecraft transferring from the original orbit to the mission orbit that forms a specific relative motion to itself. Therefore, the target orbital game process is decomposed into the transfer segment and the mission segment, and a general description model of the high-orbit non-cooperative target of "mission orbit" + "transfer orbit" is established, such as Figure 2 As shown;

[0047] For the mission segment, the mission orbit is designed based on the CW relative dynamics model according to the mission relative distance and relative speed constraints and the payload duty cycle and other parameters, including various relative orbit styles such as flyby and flyby.

[0048] For the transfer orbit of the transfer segment: take the solved mission orbit as the rendezvous target, optimize the pulse number, pulse rules and performance indicators according to the game strategy, and obtain the transfer trajectory

[0049] Based on the above general description model of high-orbit non-cooperative targets, N types of typical intentions of targets in high-orbit scenarios are defined;

[0050] In a specific case, refer to Figure 3 As shown, there are four types of high-orbit intention models:

[0051] (1) Straight-line flyby detection: The mission orbit is designed as a straight-line flyby orbit, and the transfer trajectory is designed as a transfer orbit with the optimal comprehensive performance in terms of time and fuel consumption with a pulse number of three.

[0052] Establish a mathematical optimization model for straight-line flyby detection, and set the initial orbit state of the target spacecraft to , the terminal condition is , is a straight-line flyby trajectory; the pulse maneuvering time t and the pulse velocity vector As the optimization variable, It is expressed as follows:

[0053] ;

[0054] in is the velocity increment amplitude, is the angle between the projection of the velocity increment direction in the inertial system xoy and the positive direction of the x-axis of the engineering orbit system, is the angle between the velocity increment direction and the positive direction of the z-axis of the inertial system; to ensure that the rendezvous conditions are met, the velocity vectors of the last two maneuvers can be obtained by solving the Lambert problem:

[0055] ;

[0056] in, and are the position and velocity vectors before the n-1th maneuver, For the velocity vectors of the last two Lambert maneuvers, the upper bound constraint of the pulse amplitude is added in the form of a penalty term:

[0057] ;

[0058] in, is the symbolic function, is the single pulse amplitude constraint.

[0059] The number of pulses in the straight flyby detection transfer process is set to three, so the optimization variables are the time and velocity vector of the first pulse maneuver and the transfer time of the two Lambert maneuvers. and ,for , the optimal performance index is designed as follows:

[0060] ;

[0061] in, is the total duration of the task, is the weight coefficient, is the weight of the penalty term, which is generally given a larger value.

[0062] (2) Elliptical flyby reconnaissance: The mission trajectory is designed as an elliptical flyby orbit, and the transfer orbit is selected as the fuel-optimal orbit with a pulse number of three.

[0063] Similar to the straight-line flyby detection, , the difference is that the terminal condition is an elliptical orbit, and the optimal performance index is designed as follows:

[0064] ;

[0065] (3) Kinetic flyby: The mission orbit of the kinetic flyby is such that the terminal position coincides with the flyby target, and the transfer orbit is selected as a single-pulse Lambert maneuver.

[0066] The mathematical optimization model of kinetic flyover, with the pulse number as one, sets the optimization variable as , respectively, the waiting time before the maneuver and the flyover time after the maneuver. The pulse maneuver speed increment can be obtained by solving the Lambert problem. The design flyover speed performance index and penalty term are as follows

[0067] ;

[0068] in, is the relative velocity at the end moment, is the specified minimum relative speed; the final performance index is

[0069] ;

[0070] in, is the weight coefficient.

[0071] (4) Close-range service: The mission orbit is designed to have a position and velocity consistent with the service target, and the transfer orbit is designed to be a four-pulse coplanar rendezvous orbit. The first two pulse maneuvers are to transform from an out-of-plane orbit to a coplanar orbit, and the last two maneuvers are to achieve a coplanar rendezvous.

[0072] The mathematical optimization model of close-range service, the transfer orbit is designed as a four-pulse transfer. The first two Lambert maneuvers eliminate the difference between the orbital plane and the right ascension of the ascending node, and send it into the coplanar orbit of the target satellite. The last two Lambert maneuvers adjust the orbital altitude and phase difference to complete the rendezvous with the target satellite.

[0073] The optimization variables are ,in and The waiting time and transfer time of the two Lambert maneuvers are respectively: It is the difference in semi-major axis and orbital inclination between the target coplanar orbit and the own orbit.

[0074] The penalty term for the four Lambert maneuvers is:

[0075] ;

[0076] in, represents the pulse velocity vector of the fourth Lambert maneuver, and the optimal performance index is defined as follows:

[0077] .

[0078] S2. Obtain the orbital status of the target and the target itself through the navigation system, perform optimal transfer trajectory planning under several types of intentions, predict the maneuvering sequence of the target under various intentions, including the maneuvering time and the velocity vector of each maneuver, and build a target maneuvering behavior sequence library;

[0079] Based on the above four types of intentions, a target maneuver behavior sequence library is established, as follows:

[0080] During the target intention recognition process, the orbital state of the own spacecraft is , get the target's orbit information , respectively use the particle swarm optimization algorithm to optimize and solve, and obtain the optimization variables that make the performance index optimal ;

[0081] According to the optimization variables The maneuvering time and pulse speed increment of each maneuver under each intention are calculated with the four types of intention models, and the following target maneuvering behavior sequence library is established:

[0082] ;

[0083] Where, T is the maneuvering time, is the maneuvering speed increment, the superscript i indicates the plan number, , the subscript is the maneuver number, and n represents the total number of maneuvers.

[0084] S3, after the target makes its first maneuver, identify the maneuvering time and maneuvering speed vector of the target, match them with the target behavior sequence library through the Bayesian classifier, and calculate the probability of each type of intention;

[0085] Based on the four types of intent in the above case, the intent probability is calculated, including:

[0086] When the target makes its first maneuver, the friendly satellite identifies the target's maneuvering moment through the abnormal movement detection system. With maneuvering speed vector , denoted as ;

[0087] The identified target maneuver is matched with the first maneuver of each intention in the maneuver behavior sequence library through the Bayesian classifier. The smaller the matching difference, the higher the probability of the intention. The matching difference between two maneuvers is calculated using the following formula:

[0088] ;

[0089] in, are normalized parameters for measuring time difference, velocity increment amplitude difference and velocity direction difference respectively;

[0090] The final matching difference is: ;

[0091] For the difference , calculate the likelihood: ;

[0092] in, is an adjustable parameter used to control the width of the distribution; assuming that the prior probability of each intention is equal, then: ;

[0093] Based on the likelihood results, the Bayesian formula is calculated to obtain the posterior probabilities of various intentions:

[0094] ;

[0095] in, For all possible The sum of: ;

[0096] Therefore, the recognition probability of each intention is obtained according to the maneuver matching:

[0097] ;

[0098] S4. In the subsequent maneuvers of the target, the intention probability in the previous cycle is used as the prior probability of the Bayesian classifier in the next cycle, until the probability of a certain intention is greater than the specified threshold, the intention recognition is completed, and the prediction result of the subsequent maneuvering behavior of the target is extracted from the maneuvering behavior sequence library; specifically:

[0099] After the target maneuvers again, is the prior probability, repeat step S3 to get the second intention recognition probability ;

[0100] After each target maneuver, repeat the above operation until the probability of a certain intention is greater than the given probability threshold at the jth maneuver. , then the target intention recognition is considered to be completed, and the subsequent maneuvers under this intention in the target maneuver behavior sequence library are the behavior prediction results for the target :

[0101] .

[0102] In a specific case, simulation is performed based on the above method:

[0103] During the simulation, the own spacecraft is set to be in the geosynchronous orbit at the initial moment, and the orbital elements are [42167.8420, 0.0002, 0.0698, 0.8901, 0.1745, 1.832], and the orbital elements of the target orbit at the initial moment are [42153.4422, 0.0081, 0.0720, 0.8596, 0.1745, 1.8608];

[0104] Optimizing model parameters =0.01km / s, , , , , , , km / s.

[0105] The target intention is to serve at close range. Based on the mathematical model, the optimal variable is y=[0.1634, 0.3897, 0.4903, 0.4975, 0.3734, 0.6335], which solves the target's first pulse maneuver. ;

[0106] Our side identifies the target's intention, optimizes the trajectory according to four optimization models, and builds a target maneuver behavior sequence library.

[0107] set up , , the probabilities of identifying four types of intentions are [0.0049, 0.0059,1.2381*10-5, 0.9891], where the probability of the fourth intention is greater than , intention recognition is completed, the highest probability of intention recognition is the fourth intention, close range service, and the target’s true intention Figure 1 To.

[0108] Extract the target's subsequent three maneuvers in the maneuver behavior sequence library as the behavior prediction result

[0109] ;

[0110] Repeat the above process 100 times to perform Monte Carlo simulation and statistically analyze the results (statistical results are shown in Figure 2). Figure 4 As shown in the figure, the correct intention recognition was performed 93 times in 100 simulations, with a success rate of 93%. Under the 93 successful intention recognition cases, the prediction error of the maneuvering time window did not exceed 427s (3σ), and the box plot is shown in the figure. Figure 5 As shown in the figure, the average absolute percentage error of the maneuvering speed increment is 1.060% (3σ), and the box plot is shown in the figure. Figure 6 shown.

[0111] The embodiments of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments. For ordinary technicians in this technical field, after knowing the contents recorded in the present invention, they can make several equivalent changes and substitutions without departing from the principle of the present invention. These equivalent changes and substitutions should also be regarded as belonging to the protection scope of the present invention.

Claims

1. A method for high-orbit non-cooperative target intention recognition and behavior prediction, characterized by: A universal description model of high-orbit non-cooperative targets is established to calculate the maneuvering sequences under various intentions, and a target maneuvering behavior sequence library is constructed. The following operations are performed based on the target maneuvering behavior sequence library: Identify the target's maneuvers and match them with the target's maneuver behavior sequence library through a Bayesian classifier to calculate the probability of each type of intention; and The probability of the previous intention is used as the prior probability of the subsequent Bayesian classifier processing until the probability of a certain intention is greater than the specified threshold, the intention recognition is completed, and the prediction result of the target's subsequent maneuvering behavior is extracted from the target maneuvering behavior sequence library; The target maneuvering behavior sequence library is: ； Where T is the maneuvering time, is the speed increment, the superscript i indicates the sequence number, , the subscript is the maneuver number, and n represents the total number of maneuvers; Identify targets through abnormal movement detection system, including maneuvering time With maneuvering speed vector , denoted as ; The Bayesian classifier is used to match and calculate the probability of various intents, including: Calculate the matching difference: ； in, are normalized parameters for measuring time difference, velocity increment amplitude difference and velocity direction difference respectively; Based on the difference , calculate the likelihood: ； in, is an adjustable parameter used to control the width of the distribution; assuming that the prior probability of each intention is equal, then: ； Based on the likelihood results, the Bayesian formula is calculated to obtain the posterior probabilities of various intentions: ； in, For all possible The sum of: ； Therefore, the recognition probability of each intention is obtained according to the maneuver matching: 。 2. According to claim 1, a high-orbit non-cooperative target intention recognition and behavior prediction method is characterized by: The target orbit game process in the high-orbit non-cooperative target general description model includes: Mission segment: designed according to the mission relative distance and relative speed constraints and preset parameters; Transfer stage: Taking the solved mission orbit as the rendezvous target, the number of pulses, pulse rules and performance indicators are optimized according to the game strategy to obtain the transfer trajectory.

3. The method for high-orbit non-cooperative target intention recognition and behavior prediction according to claim 1, characterized in that: The particle swarm optimization algorithm is used to calculate various typical intentions in the general description model of high-orbit non-cooperative targets to obtain the optimization variables that optimize the performance indicators; as well as Based on the optimization variable results, the maneuver sequences under various intentions are calculated and a target maneuver behavior sequence library is constructed.

4. A high-orbit non-cooperative target intention recognition and behavior prediction method according to claim 1 or 3, characterized in that: The maneuver sequence includes time and pulse speed increments of the maneuvers.

5. The method for high-orbit non-cooperative target intention recognition and behavior prediction according to claim 4, characterized in that: They are: ； in, is the total duration of the task.

6. The method for high-orbit non-cooperative target intention recognition and behavior prediction according to claim 4, characterized in that: The previous one is the prior probability, and the probability of the next intention recognition is calculated ; and repeat until the jth maneuver, the probability of a certain intention is greater than the given probability threshold , then the target intention recognition is considered to be completed, and the subsequent maneuvers under this intention in the target maneuver behavior sequence library are the behavior prediction results for the target : .

Images