An online cooperative detection task planning method based on probability graph dynamic programming

Abstract

The application discloses an online cooperative detection task planning method based on a probability graph dynamic programming, and comprises the following steps: first, a search probability graph is constructed according to the position of a possible deployment area of a target obtained from a situation; second, a correlation matrix is constructed, clustering fusion is carried out according to the correlation matrix, the clustering center after the clustering fusion is taken as an initial pointing position of an infrared detector, the coverage of the infrared detector is calculated, and the search sub-area of the infrared detector is optimally distributed; and third, a periodic and reactive mixed search scheduling strategy is adopted, the pointing position of the infrared detector is optimally distributed every fixed period, and online cooperative detection task planning is carried out. The application solves the problems of low efficiency and poor generalization of online cooperative search and sensing decision in the prior art.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of red and blue sides counter deduction intelligent decision-making, and particularly relates to an online cooperative detection task planning method based on a probability graph dynamic programming. BACKGROUND

[0002] Adversarial artificial intelligence has emerged and achieved certain success. As early as 2011, IBM's "Waston" defeated the contestant with the highest cumulative score and the record holder of consecutive victories in the quiz show "Jeopardy!". Waston, whose core is IBM DeepQA technology, adopts an expert system approach to reasoning with the correlation analysis and statistical characteristics of big data. Its victory reflects the understanding, reasoning and solution of machine AI to complex problems in natural language, and its intelligent representation makes artificial intelligence face a major turning point. With the support of big data and high-performance computing, the breakthrough of deep learning technology provides conditions for the development of artificial intelligence.

[0003] Searching and discovering targets in the task area are the basis and premise for a series of subsequent combat activities. With the development of science and technology, the combat environment of modern war is becoming increasingly complex, and it is becoming more and more difficult to search and discover targets in time. Before the aircraft performs the search task, it only knows the geographical position information and boundary information of the search area, and has no information about the targets in the area. The aircraft performs the search task in a completely unknown complex area. Considering the limitations of single aircraft search, it is necessary to develop cooperative search research on multiple aircrafts.

[0004] Most of the current intelligent decision-making algorithms use optimization-based and prior knowledge-based methods. For the multi-target dynamic optimization problem in the whole process of red and blue sides deduction task timing, there are problems such as low decision-making efficiency and poor generalization. SUMMARY

[0005] The technical problem solved by the present application is to overcome the shortcomings of the prior art and provide an online cooperative detection task planning method based on probability graph dynamic programming, which solves the problems of missing coverage area, poor generalization and low decision-making efficiency in multi-target detection dynamic programming tasks.

[0006] The technical solution of the present application is:

[0007] The online cooperative detection task planning method based on probability graph dynamic programming comprises the following steps:

[0008] First, a search probability graph is constructed according to the position of the possible deployment area of the target obtained from the situation;

[0009] Secondly, the correlation matrix is constructed, and the clustering fusion is performed according to the correlation matrix, the clustering center after the clustering fusion is taken as the pointing position of the initial infrared detector, the coverage range of the infrared detector is calculated, and the search sub-area of the infrared detector is optimally distributed;

[0010] Thirdly, an online dynamic search scheduling strategy is adopted, the pointing position of the infrared detector is optimally distributed every fixed period, and the online cooperative detection task planning is performed.

[0011] In the first step, the construction process of the probability graph is as follows:

[0012] (1.1) The situation is acquired through the passive radar and infrared detector on the plurality of aircrafts, the acquired target information is converted to a unified geocentric coordinate system, and a possible target deployment area is initially obtained;

[0013] (1.2) It is assumed that the possible target deployment area is a planar rectangular search area in which a plurality of target groups are distributed, a plurality of aircrafts enter the search area to search for targets, the search area is divided into m*n quadrilateral grids, each quadrilateral grid is taken as a unit, and the set E={(i,j)|i=1,2,…,m;j=1,2,…,n} of all units is called a search environment, (i,j) represents a unit located in the i-th row and the j-th column, m is the maximum row boundary of the search field, and n is the maximum column boundary of the search field;

[0014] (1.3) P ij (k)∈[0,1] is defined for each unit (i,j), and P ij (k)∈[0,1] represents the probability that the unit (i,j) contains a target group at the k-th moment, and the search probability graph is represented as SPM={P ij (k)|(i,j)∈E}. In the process of searching, the search probability graph is updated in real time according to the detected information.

[0015] In the step (1.3), the method for updating the search probability graph is as follows:

[0016] The probability that the unit (i,j) contains a target group at the k-th moment is

[0017]

[0018] In the formula, P D is the detection probability of the infrared detector, and P F is the false alarm probability.

[0019] If the infrared detector does not detect a target, according to the Bayes theorem, P

[0020]

[0021] Since each time step is limited by the number of cells searched by the aircraft, only the searched cells are updated, and the target group existence probability of other unsearched cells remains unchanged;

[0022] Define a variable z(i,j,k) to represent whether there is a target in cell (i,j), that is

[0023]

[0024] θ represents the threshold value of target update. After the search is completed at a predetermined time, the target existence probability of each cell is calculated according to z(i,j,k), and the search probability map of the target possible deployment area is updated.

[0025] When the target and the infrared detector satisfy the following two conditions at the same time, the infrared detector can detect the target:

[0026] Condition 1: The target is within the detection range of the infrared detector. According to the infrared detector capability and target characteristics, the limit action distance R of the infrared detector is calculated. Only when the distance between the target and the infrared detector is less than R, the target is considered to be observable;

[0027] Condition 2: The target is not blocked by the earth from the infrared detector. Assuming that the earth is a sphere with equal radius, as long as the target is located within the conical region with the infrared detector as the apex and the half-cone angle α, it is considered to be observable; V is the distance from the center of the infrared detector field of view to the infrared detector,

[0028] The implementation of the second step is as follows:

[0029] Construct a topology graph, where the points represent targets, and assume that the searched target set is {t H ,t G},

[0030]

[0031] Where, represents the connectivity of targets t H and t G in the topology graph, 0 represents not connected, and 1 represents connected;

[0032] According to the correlation matrix, cluster fusion is performed to cluster the possible M targets and determine the best initial infrared detector pointing position.

[0033] The implementation of the cluster fusion according to the correlation matrix is as follows:

[0034] Step 1: Scan each target in the target set in turn. If the target t xAnd t y has a connectivity of 1, then in the current iteration, the target t x is counted in the cell, the target t y is counted in the cell;

[0035] Step 2: the target t x and t y are divided into the same cluster, and each cell is checked in turn, if the count value of the cell is greater than the preset threshold, the targets corresponding to the cells are divided into the same cluster;

[0036] Step 3: repeat steps 1-2 until the maximum number of iterations is reached, and the clustering fusion is completed.

[0037] The implementation mode of the third step is as follows:

[0038] (3.1) an online dynamic search scheduling strategy is adopted to decompose the cooperative detection task to obtain a meta-task sequence;

[0039] (3.2) a scheduling model is established for the meta-task;

[0040] (3.3) a cooperative detection task planning method is determined according to the scheduling model, the pointing position of the infrared detector is optimized in each fixed period, and the task planning is realized.

[0041] In step (3.1), the online dynamic search scheduling strategy is adopted to decompose the cooperative detection task to obtain the steps of the meta-task sequence as follows:

[0042] Step 1: calculate the visualization time period between all infrared detectors and targets;

[0043] Step 2: find out the start time and end time of all visualization time periods, and arrange these time points in ascending order according to time, then form a time period between two time points from front to back, so as to decompose the whole task into meta-tasks in each time period, and all time periods constitute a meta-task set;

[0044] Step 3: filter out all time periods with a length greater than D max from the time periods corresponding to the meta-task set, and continue to uniformly split them to ensure that the time period length is within the interval [D min , D max ]; D max is the longest time period of the meta-task, and D min is the shortest time period of the meta-task;

[0045] Step 4: filter the time periods obtained in step 3, and the time period length is less than D minThe meta task is deleted from the meta task set, and a new meta task set is obtained as the meta task sequence.

[0046] In step (3.2), the scheduling model is as follows:

[0047] 1. Decision variable

[0048] In time period d ab , the task of the a-th infrared detector detecting target b is the K-th meta task in the meta task sequence, and if b is detected, then Otherwise,

[0049] 2. Define a first objective function, which is to maximize the weighted detection K-th meta task revenue ab in time period d , that is:

[0050]

[0051] wherein, is the detection revenue value of the a-th infrared detector to target b;

[0052] 3. Define a second objective function, which is to maximize the number of detected targets in time period d ab , that is

[0053]

[0054]

[0055] represents the number of targets detected by the a-th infrared detector in time period d ab .

[0056] The implementation of the step (3.3) is as follows:

[0057] Step 0: Sort the meta tasks in the meta task sequence in chronological order and store them in the waiting execution task linked list, align the system time with the start time of the first meta task; solve the optimal solution of the first meta task according to the scheduling model, and set the initial value of K as 1;

[0058] Step 1: transfer the K-th meta task to the executing task linked list, execute the K-th meta task according to the obtained optimal solution, and increase the execution time of the K-th meta task by t p .

[0059] Step 2: if t p > t KIf yes, the Kth subtask is transferred to the task list to be executed, and step 3 is executed; otherwise, step 1 is returned.

[0060] Step 3: judging whether the Kth subtask is the last subtask, if yes, the task list to be executed is emptied, and the planning is terminated; otherwise, the value of K is added by 1, and step 4 is executed.

[0061] Step 4: arranging the subtasks in time sequence, storing into the task list to be executed, and solving the optimal solution of the Kth subtask according to the scheduling model of the subtask, repeating steps 1-3.

[0062] Compared with the prior art, the present application has the following beneficial effects:

[0063] The present application significantly reduces the probability of searching invalid areas during cooperative detection by constructing a probability graph and clustering targets, and therefore solves the problems of missing coverage area, poor generalization and low decision efficiency in the multi-target detection dynamic programming task. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 The flow chart of the present application;

[0065] Figure 2 The schematic diagram of the coverage range of the infrared detector of the present application;

[0066] Figure 3 The schematic diagram of the topological structure;

[0067] Figure 4 The schematic diagram of task decomposition. DETAILED DESCRIPTION

[0068] The present application proposes an online cooperative detection task planning method based on probability graph dynamic programming, as shown in Figure 1 The steps include:

[0069] Firstly, the time-space error of online detection of multiple aircrafts is considered, the detection coordinate systems of various aircrafts are unified, the information of infrared detectors and passive radars on the multiple aircrafts is converted into a unified Earth-Centered Earth-Fixed (ECEF) coordinate system, and time-space consistency is realized when the cooperative detection information is shared.

[0070] The process of time-space consistency of information sharing is as follows:

[0071] (1.1) determining the conversion relationship between the polar coordinate system of the infrared detector and the Cartesian coordinate system;

[0072] The infrared detector directly measures the target in polar coordinates (r, θ, η), where r is the target slant range measurement, θ is the azimuth measurement, and η is the elevation measurement. Based on the relationship between polar and Cartesian coordinate systems, the target's coordinates (x, y, z) in the infrared detector's local Cartesian coordinate system are:

[0073]

[0074] Where x, y, and z are the coordinates of the target in the X, Y, and Z axes of the Cartesian coordinate system, respectively.

[0075] (1.2) Determine the transformation relationship between the Cartesian coordinate system and the Northeast-Eastern-Universe (ENU) coordinate system.

[0076] Assume that there is an attitude angle between the Cartesian coordinate system and the northeast-sky coordinate system. in Let y be the yaw angle, ψ be the pitch angle, and η be the roll angle. The transformation from the Cartesian coordinate system to the Northeast-Sky coordinate system can be described as three rotations around the coordinate axes.

[0077] Assuming the target's coordinates in the Cartesian coordinate system are (x, y, z), then its position in the corresponding northeast-northeast coordinate system (x, y, z) is... NEU ,y NEU ,z NEU )for:

[0078]

[0079] Among them, A roll (k), A pitch (k), A head (k) represents the rotation matrix of the platform in the Cartesian coordinate system at time k about the X, Y, and Z axes, respectively, and its specific form is:

[0080]

[0081]

[0082]

[0083] (1.3) Determine the transformation relationship between the northeast-sky coordinate system and the geocentric-earth-fixed (ECEF) coordinate system.

[0084] Assuming the geographic coordinates are (L, λ, H), the target's coordinates in the northeast-central coordinate system are (x, λ, H). NEU ,y NEU ,z NEU Then, the position of the target in the Northeast-Eastern Sky coordinate system can be transformed to the ECEF coordinate system by rotation and translation:

[0085] The position (x s ,y s ,z s ) in the ECEF coordinate system is calculated by the following formula:

[0086]

[0087] wherein C is defined as:

[0088]

[0089] wherein E q represents the equatorial radius of the earth, and e is the first eccentricity of the earth.

[0090] The northeast celestial coordinate system is converted to the ECEF coordinate system through rotation and translation, and the conversion formula is:

[0091]

[0092] wherein T is the rotation matrix of the northeast celestial coordinate system to the parallel coordinate system of the ECEF coordinate system.

[0093] (1) Environment model

[0094] It is assumed that the search area is a planar rectangular search area, wherein M target groups are distributed, and N aircrafts enter the task area to search for targets. The area is divided into m x n quadrilateral grids, and the set E = {(i, j) | i = 1, 2,..., m; j = 1, 2,..., n} of all cells is called a search environment, and (i, j) represents a cell located in the i-th row and the j-th column.

[0095] (2) Search probability graph model

[0096] P ij (k)∈[0,1] is defined for each cell (i, j), and P ij (k)∈[0,1] represents the probability of the existence of a target group in the cell (i, j) at time k, and the search probability graph can be represented as SPM = {P ij (k) | (i, j) ∈ E}. In the process of searching, the search probability graph is updated in real time according to the detected information.

[0097] Considering the uncertainty of the measurement of the infrared detector, when the aircraft searches the cell (i, j) at time k, if the infrared detector detects a target group, according to the Bayes theorem, the probability of the existence of a target group in the cell at this time is

[0098]

[0099] P DP is the detection probability of the infrared detector (i.e., the probability that a target group exists and is detected by the infrared detector) F Pfa is the false alarm probability (i.e., the probability that no target group exists but is detected by the infrared detector).

[0100] If the infrared detector does not detect a target, then according to Bayes' theorem, there is

[0101]

[0102] Since the number of cells searched by the aircraft is limited at each time step, only the searched cells are updated, and the target group existence probability of other unsearched cells remains unchanged.

[0103] After the search is completed, a variable z(i,j,k) is defined to represent whether a target exists in the cell, i.e.,

[0104]

[0105] In the formula, θ is a predefined threshold.

[0106] (3) Infrared detector detection model of the aircraft

[0107] Considering the detection distance constraint of the infrared detector on the target, a maximum possible coverage area (coverage capability) is designed, and an infrared detector detection model is established.

[0108] The infrared detector performs dynamic search on the target during the flight of the aircraft, and the detection capability is mainly measured by the detection distance and the optical axis steering range. The infrared detector can only detect a small field of view within the angle range, and the pointing angle is the scanning angle of the small field of view optical imaging instrument. The optical axis steering range depends on the maximum angle that the load can be turned relative to the earth axis of the aircraft when the field of view is directed to the target area under the control of the optical axis pointing control.

[0109] After considering the performance of the infrared detector and the visibility conditions, it can be determined that the infrared detector can detect a given target within which time period, and the determination conditions are as follows:

[0110] Condition 1: The target is within the infrared detection range. According to the infrared detector capability and target characteristics, the limit action distance L of the infrared detector can be calculated. Only when the distance between the target and the infrared detector is less than L, the target is considered to be observable.

[0111] Condition 2: The target is not blocked by the earth from the infrared detector. Assuming that the earth is an equal radius sphere, as long as the target is located outside the circular cone region with the infrared detector as the apex and the half-cone angle α, it is considered to be observable (see Figure 2 ).

[0112] Thirdly, the region blockage coverage based on the correlation matrix is realized

[0113] In order to improve the cooperative coverage ability of multiple aircrafts, a sub-region division method based on correlation graph clustering is proposed, which divides the search region into a limited number of sub-regions and determines the clustering center point.

[0114] A topological graph is constructed, in which the points represent targets, and the searched target set is assumed to be {t H ,t G};

[0115]

[0116] Wherein, H=1, 2, ……M; G=1, 2, ……N; Indicates the connectivity of targets t H and t G in the topological graph, 0 indicates not connected, and 1 indicates connected.

[0117] According to the correlation matrix, the possible M targets are clustered to determine the best initial infrared detector pointing position.

[0118] Figure 3 The numbers in the figure represent target points. Two targets have a connection relationship, indicating that the connectivity is 1.

[0119] According to the correlation matrix, the possible multiple targets are clustered to determine the best initial detection point. The implementation of clustering fusion according to the correlation matrix is as follows:

[0120] Step 1: Scan each target in the target set in turn. If the connectivity of targets t x and t y is 1, then in the current iteration, the count of the cell where target t x is located is increased by 1, and the count of the cell where target t y is located is increased by 1.

[0121] Step 2: Targets t x and t y are divided into the same cluster. Check the count of each cell in turn. If the count value of a cell is greater than a pre-set threshold, the targets corresponding to these cells are divided into the same cluster.

[0122] Step 3: Repeat steps 1-2 until the maximum number of iterations is reached, and the clustering fusion is completed.

[0123] The above algorithm does not need to set any parameters, and the number of clusters can be automatically selected in the process of cluster fusion. And it can handle clusters of any shape. The clustering algorithm divides two data points according to the possibility of them appearing in the same cluster during cluster fusion, so as long as the distance between two data points is close enough, they will be divided into a cluster.

[0124] Fourthly, task planning under multi-aircraft cooperative perception is carried out.

[0125] (1) Task decomposition

[0126] When performing a multi-target detection task, there may be a multi-platform multi-infrared detector relay tracking. It is difficult to complete effective scheduling as a single task throughout the early warning process. By using the task decomposition method, the multi-infrared detector multi-target cooperative detection task is decomposed into a series of meta-tasks by decomposing the multi-attribute decision-making task decision into a series of meta-tasks.

[0127] When performing task decomposition, the following constraint conditions must be met:

[0128] ① The length of the time period of the meta-task cannot be too long to ensure the flexibility of task execution and prevent a certain meta-task from occupying critical infrared detector resources for a long time. Assuming that the longest time of the meta-task is D max ;

[0129] ② The length of the time period of the meta-task cannot be too short to ensure that the trajectory of the target can be predicted within the time period, and to avoid frequent handover between infrared detectors. Assuming that the shortest time period of the meta-task is D min .

[0130] According to the above constraints, the task decomposition is performed on multiple targets.

[0131] The detailed steps of task decomposition are as follows:

[0132] Step 1: Calculate the visualization time period between all infrared detectors and targets;

[0133] Step 2: Find the start and end times of all visualization time periods, and arrange these times in ascending order according to time, then form a time period between two times from front to back, thereby decomposing the entire task into a set of meta-tasks in time periods;

[0134] Step 3: According to constraint ①, all time periods with a length greater than D max are filtered out from the meta-task set, and they are further uniformly split to ensure that the time period length is within the interval [D min , D max ];

[0135] Step 4: According to constraint ②, the time period obtained in step 3 is filtered, and the meta task with a time period length less than D min is deleted from the meta task set.

[0136] As shown by the gray part in Figure 4 , the time period from T1 to T2 constitutes a meta task, the time period from T4 to T5 constitutes a meta task, the time period from T8 to T10 constitutes a meta task, and different meta tasks are in different time periods.

[0137] (2) Task scheduling model

[0138] After task decomposition, the execution of the entire task is converted into the execution of each meta task. In the time period corresponding to each meta task, the infrared detector and the target have two relationships: whether the target is in the visual time period of the infrared detector; and if the target is in the visual time period of the infrared detector, whether the infrared detector performs detection on the target. According to the above analysis, a meta task scheduling model is established.

[0139] Suppose there are M infrared detectors, the target capacity of the a-th infrared detector is B a , there are N targets, w b is the threat degree of target b, TY ab is the detection benefit degree of the a-th infrared detector to target b, TY ab = d ab / max(d ab ), where max(d ab ) is the maximum value of the visual time period length of the a-th infrared detector to target b; here, the target threat degree refers to the threat of multiple targets being sorted according to the main factors affecting the threat degree of the target, and the index reflecting the threat size of the target is obtained by using the threat evaluation method.

[0140] 1. Decision variable

[0141] In the time period d ab , the task of the a-th infrared detector to target b is the K-th meta task of the meta task sequence, if b is detected, then otherwise,

[0142] 2. Define the first objective function, which is to maximize the weighted detection benefit of the K-th meta task in the time period d ab , that is:

[0143]

[0144] wherein, ​The detection benefit value of the a-th infrared detector to the target b;

[0145] The detection benefit value of the infrared detector a to the target b Wherein, λ is a weight coefficient, e k The detection benefit compensation of the first k-1 meta tasks to the k-th meta task.

[0146] 3. Defining a second target function, the second target function is that the number of detected targets is as many as possible in the time period d ab

[0147]

[0148]

[0149] The number of targets detected by the a-th infrared detector in the time period d ab

[0150] Constraint condition

[0151] Resource constraint: the target capacity reserved by the a-th infrared detector is Then the resource constraint is satisfied:

[0152]

[0153] (3) Cooperative detection multi-task planning algorithm

[0154] The application designs an infrared detector cooperative detection task planning algorithm based on task decomposition, which meets the high detection benefit of task planning and takes into account the real-time demand of task planning.

[0155] Step 0: The meta tasks in the meta task sequence are sorted in chronological order, and stored in the waiting execution task linked list, and the system time is aligned with the start time of the first meta task; the optimal solution of the first meta task is solved according to the scheduling model, and the initial value of K is 1;

[0156] Step 1: The K-th meta task is transferred to the executing task linked list, and the K-th meta task is executed according to the optimal solution obtained, and the system time t p The execution time of the K-th meta task is increased;

[0157] Step 2: If t p >t K , the K-th meta task is transferred to the executed task linked list, and step 3 is executed; otherwise, return to step 1;

[0158] ​​Step 3: judging whether the Kth subtask is the last subtask, if yes, emptying the waiting execution task linked list, and then the planning terminates; otherwise, the value of K is added by 1, and step 4 is executed;

[0159] Step 4: arranging the subtasks in time sequence, storing into the waiting execution task linked list, solving the optimal solution of the Kth subtask according to the scheduling model of the subtask, and repeating steps 1-3.

[0160] The contents not described in detail in the specification of the present application belong to the known technology of the person skilled in the art.

Claims

1. An online cooperative exploration task planning method based on probabilistic graphical dynamic programming, characterized in that... The steps of this method include: The first step is to construct a search probability map based on the possible deployment areas of the target obtained from the situational awareness. The second step is to construct an association matrix, perform clustering and fusion based on the association matrix, use the cluster centers after clustering and fusion as the initial pointing positions of the infrared detectors, calculate the coverage range of the infrared detectors, and optimize the allocation of the search sub-regions of the infrared detectors. The third step involves adopting an online dynamic search and scheduling strategy to optimize the allocation of the pointing positions of infrared detectors in each fixed cycle, and to plan online collaborative detection tasks. The second step is implemented as follows: Construct a topological graph where points represent targets, and assume the set of targets found is {t}. H ,t G }, in, Represents target t in the topology graph H and t G The connectivity is denoted by 0, where 0 indicates no connectivity and 1 indicates connectivity. Clustering and fusion are performed based on the correlation matrix to cluster the M possible targets and determine the optimal initial pointing position of the infrared detector. The implementation method for cluster fusion based on the correlation matrix is ​​as follows: Step 1: Scan each target in the target set sequentially. If target t... x and t y If the connectivity is 1, then in the current iteration, the target t... x The cell count is incremented by 1, and the target t is... y Increment the cell count by 1; Step 2: Target t x and t y Once they are assigned to the same cluster, the count of each unit is checked sequentially. If the count of any unit is greater than a preset threshold, the targets corresponding to these units are assigned to the same cluster. Step 3: Repeat steps 1-2 until the maximum number of iterations is reached to complete the cluster fusion.

2. The online collaborative exploration task planning method based on probabilistic graphical dynamic programming according to claim 1, characterized in that: In the first step, the process of constructing the probability graph is as follows: (1.1) Situational awareness is acquired by using passive radar and infrared detectors on multiple aircraft, and the acquired target information is converted into a unified geocentric coordinate system to obtain the possible deployment area of ​​the target. (1.2) Assume that the possible deployment area of ​​the target is a planar rectangular search area, in which multiple target groups are distributed. After multiple aircraft enter the search area, they search for the target. The search area is divided into m×n quadrilateral grids, and each quadrilateral grid is a cell. The set of all cells E={(i,j)|i=1,2,…,m;j=1,2,…,n} is called the search environment. (i,j) represents the cell located in the i-th row and j-th column, m is the row boundary of the largest search field of view, and n is the column boundary of the largest search field of view. (1.3) Define P for each unit (i,j) ij (k)∈[0,1], P ij (k)∈[0,1] represents the probability that the target group exists in cell (i,j) at time k. The search probability graph is then represented as SPM={P ij (k)|(i,j)∈E}, the spacecraft updates the search probability map in real time based on the detected information during the search process.

3. The online cooperative exploration task planning method based on probabilistic graphical dynamic programming according to claim 2, characterized in that: In step (1.3), the method for updating the search probability graph is as follows: The probability that the target group exists in cell (i,j) at time k is In the formula P D P is the detection probability of the infrared detector. F This represents the probability of a false alarm. If the infrared detector does not detect the target, then according to Bayes' theorem, Since the number of cells that the aircraft can search for at each time step is limited, only the cells that have been searched are updated, while the probability of the target group existing for the other cells that have not been searched remains unchanged. Define a variable z(i,j,k) to represent whether the target exists in cell (i,j), i.e. θ represents the threshold for target updates. At the predetermined time, after the search is completed, the probability of the target existing in each cell is calculated based on z(i,j,k), and the search probability map of the possible deployment areas of the target is updated.

4. The online collaborative exploration task planning method based on probabilistic graphical dynamic programming according to claim 3, characterized in that, An infrared detector can detect a target when both the target and the infrared detector simultaneously meet the following two conditions: Condition 1: The target is within the detection range of the infrared detector. Based on the capabilities of the infrared detector and the characteristics of the target, the limit of the infrared detector's operating distance R is calculated. The target is considered observable only when the distance between the target and the infrared detector is less than R. Condition 2: The target and the infrared detector are not obstructed by the Earth. Assuming the Earth is a sphere of equal radius, the target is considered observable as long as it lies within a conical region with a semi-cone angle α and the infrared detector as its vertex; V is the distance from the center of the infrared detector's field of view to the infrared detector.

5. The online collaborative exploration task planning method based on probabilistic graphical dynamic programming according to claim 1, characterized in that, The third step is implemented as follows: (3.1) An online dynamic search and scheduling strategy is adopted to decompose the cooperative exploration task and obtain the meta-task sequence; (3.2) Establish a scheduling model for meta-tasks; (3.3) Determine the collaborative detection task planning method based on the scheduling model, optimize the allocation of the pointing position of the infrared detector in each fixed period, and realize task planning.

6. The online collaborative exploration task planning method based on probabilistic graphical dynamic programming according to claim 1, characterized in that, In step (3.1), the online dynamic search and scheduling strategy is used to decompose the cooperative exploration task and obtain the meta-task sequence. The steps are as follows: Step 1: Calculate the visualization time intervals between all infrared detectors and the target; Step 2: Find the start and end times of all visualized time periods, and arrange these times in ascending order. Then, form a time period by combining each pair of times from beginning to end, thereby decomposing the entire task into meta-tasks in each time period. All time periods constitute a meta-task set. Step 3: Filter out tasks with a length greater than D from the time period corresponding to the meta-task set. max For all time periods, continue to divide them evenly, ensuring that the length of each time period is within the interval [D]. min D max ] within; D max D represents the longest time period of the meta-task. min The shortest time period for the meta-task; Step 4: Filter the time periods obtained in Step 3, selecting those with a length less than D. min The metatasks are removed from the metatask set, and the resulting new metatask set is the metatask sequence.

7. The online cooperative exploration task planning method based on probabilistic graphical dynamic programming according to claim 6, characterized in that, In step (3.2), the scheduling model is as follows:

1. Decision variables During time period d ab Within the sequence, the task of the a-th infrared detector to detect target b is the k-th meta-task in the meta-task sequence. If b is detected, then... otherwise, 2. Define the first objective function, which is defined over the time interval d. ab Within, the benefit of weighted detection of the Kth meta-task is increased. To reach the maximum, that is: in, Let be the detection gain value of the a-th infrared detector for target b; 3. Define a second objective function, which is defined over a time interval d. ab Within, the number of detected targets should be as large as possible, that is... Indicates the time period d ab The number of targets detected by the a-th infrared detector.

8. The online cooperative exploration task planning method based on probabilistic graphical dynamic programming according to claim 7, characterized in that, The implementation method of step (3.3) is as follows: Step 0: Sort the metatasks in the metatask sequence according to their time order and store them in the waiting task list, aligning the system time with the start time of the first metatask; The optimal solution for the first meta-task is obtained based on the scheduling model, and the initial value of K is set to 1. Step 1: Transfer the Kth meta-task to the list of currently executing tasks, and execute the Kth meta-task according to the obtained optimal solution. Simultaneously, the system time t... p Increase the execution time of the Kth meta-task; Step 2: If t p >t K If the task is successful, transfer the Kth meta-task to the list of already executed tasks and proceed to step 3; otherwise, return to step 1. Step 3: Determine if the Kth meta-task is the last meta-task. If it is, clear the waiting task list and terminate the planning; otherwise, increment the value of K by 1 and proceed to step 4. Step 4: Arrange the meta-tasks in chronological order and store them in the waiting task list. Solve for the optimal solution of the Kth meta-task based on the meta-task scheduling model, and repeat steps 1-3.

Images