Multi-satellite cooperative remote communication task joint planning method for ecological monitoring

Through multi-agent learning methods and fully connected neural networks, the satellite resource scheduling is optimized, and the resource allocation conflicts and unbalanced coverage problems in multiple large-scale ecological area task scheduling are solved, and efficient ecological monitoring and data transmission are achieved.

CN118984180APending Publication Date: 2024-11-19XIDIAN UNIV
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202411056892.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-02
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

The existing technology is difficult to effectively schedule satellite missions in multiple large-scale ecological regions, resulting in resource allocation conflicts, affecting the immediate response capabilities of satellite missions and the effectiveness of the entire satellite network, and failing to fully consider the spatial correlation and ecological processes between regions. Continuity, resulting in blind spots covering or imbalance in repeated observations.

Method used

Using multi-agent learning method, each large-scale ecological area to be observed is used as an agent, and distributed scheduling is performed according to its own urgency and observation status, and a strategy function and action-value function are constructed through a fully connected neural network. Achieve the spatial decomposition and data return mechanism of regional goals, and optimize the dynamic scheduling and data transmission of satellite resources.

Benefits of technology

Continuous, comprehensive and efficient monitoring of multiple ecological areas has been achieved, satellite resource utilization and data transmission efficiency have been improved, data transmission wait time and retransmission times have been reduced, and timely recycling and analysis of key ecological data has been ensured.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118984180A_ABST
    Figure CN118984180A_ABST
Patent Text Reader

Abstract

The invention discloses an ecological monitoring-oriented communication and remote control integrated task joint planning method, and mainly solves the problems that the prior art cannot be suitable for a plurality of large-scale regional scenes and does not consider that satellite data is transmitted back to a ground station. According to the implementation scheme, a satellite-region target network is constructed; decomposing a regional target into observation strips, and constructing a subtask set according to information of the observation strips; constructing a state space and an action space according to the subtask set; constructing a strategy function and an action value function; detecting whether the selected satellites of the regional targets at the same moment conflict; and obtaining a next moment state St + 1 and a t moment award Rt according to the detected conflict, updating a strategy function and an action value function according to the two parameters, obtaining a task scheduling strategy, and matching the satellite with the ground station. The method can efficiently complete the joint scheduling of the remote communication integrated task of a plurality of large-scale area targets, and can be used for the intelligent scheduling of a satellite network for a plurality of large-scale ecological area targets.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of satellite communication technology, and in particular relates to a joint planning method for multi-satellite collaborative remote communication tasks for ecological monitoring, which can be used for the intelligent scheduling of multiple large-scale ecological regional targets by satellite networks to obtain higher regional observation coverage efficiency, while efficiently transmitting back observation data to maximize the amount of data transmitted back. Background Art

[0002] In the context of the era of accelerated globalization, satellite mission resource scheduling technology, as the core of the "space-ground integration" information network, has put forward more stringent standards for satellite mission scheduling strategies. Although the traditional scheduling mechanism can rely on the optimization model to efficiently allocate satellite resources when dealing with isolated point targets or single-region tasks, it will be greatly limited in diversified regions, especially complex tasks such as ecological environment monitoring that require large-scale and high-precision coverage. The current trend has shifted from single-target observation to comprehensive observation involving multiple large areas, requiring satellites to complete comprehensive and detailed coverage within a strict time frame, while taking into account the process of satellite data transmission back to the ground station, maximizing the use of satellite and ground station resources, and reducing the waiting time and retransmission times of data transmission. This undoubtedly puts a test on the optimal use of satellite resources. With the increase in the number of satellites and the continuous expansion of the monitoring target area, resource allocation conflicts have become increasingly prominent within the limited observation window, threatening the immediate response capability of satellite missions and the effectiveness of the entire satellite network control system. In view of this, it is particularly important to develop a joint planning method for remote sensing missions designed for multi-satellite collaborative observation of large areas.

[0003] At present, there are very few studies on the scheduling methods of regional target tasks for multiple satellites, and the schemes considered rarely consider multiple large-scale ecological regions to be observed at the same time, resulting in a very single number of scheduled tasks, which is not practical. Therefore, it is necessary to design a joint scheduling method for multiple ecological regions under different satellite resources; in addition, when satellites observe large-scale ecological regions, there are few specific observation postures of different satellites, especially observation angles, to dynamically disassemble regional targets. Therefore, it is necessary to consider the attributes of different satellites such as bandwidth and observation angle to disassemble regional targets and be able to complete efficient resource scheduling according to the split regional strips, and at the same time design an efficient resource transmission mechanism to provide efficient satellite response services for multi-region observation needs.

[0004] In their paper "A hierarchical classification planning strategy for multi-target repeated observation tasks", Zhang Yaoyuan et al. innovatively introduced a hierarchical classification task planning strategy to solve the problem of earth observation satellites tracking point targets under field of view constraints. This strategy carefully evaluates the contribution of each point target in task planning, dynamically assigns priority levels according to the time window characteristics of the target, adopts a step-by-step classification method, gradually organizes task categories according to priority, and interactively iterates between classification and planning activities until the planning task is successfully completed. In this step-by-step construction process, the strategy cleverly incorporates the instant reduction mechanism of the conflict window, and continuously and effectively resolves the mutual constraints of conflicting tasks in the planning link, providing convenience for the rapid task layout of the greedy search algorithm. However, this method only targets point targets and fails to fully consider the spatial correlation between regions and the continuity of ecological processes. As a result, when facing regional targets, it is impossible to perform regional division and task planning, thereby ignoring the mutual influence and overall effect between regions; especially when dealing with complex shapes and large areas, there will be problems of coverage blind spots or uneven repeated observations.

[0005] In their paper "Discussion on Imaging Satellite Mission Planning Model and Algorithm for Point Group and Large Area Targets", Yuan Jianbo et al. focused on coping with new complex imaging requirements such as point groups and large areas, aiming to improve satellite resource utilization efficiency and mission results. They constructed a satellite planning model and algorithm framework specifically for point group and large area imaging tasks. For large areas, the framework quantifies coverage through a gridding method, calculates based on the grid center point, and assumes that different satellites and their payload imaging strips can be seamlessly spliced, ignoring potential overlaps or invalid observations in the splicing, thereby establishing a mission benefit model based on grid coverage. An innovative simulated annealing taboo algorithm with adaptive temperature control is designed, which enhances the global exploration of the algorithm through fluctuating temperature control and internal loop iteration adjustment strategies, and integrates taboo strategies to enhance local optimization capabilities, providing a set of adaptive and efficient solution paths for the model. Although this method shows efficiency and innovation when dealing with a single large area target, its algorithm cannot be fully extended to the situation of dealing with multiple regional targets at the same time, and has limitations. At the same time, the process of transmitting data back to the ground station was not taken into consideration. When facing multiple areas, with limited satellite resources, it was impossible to balance the coverage requirements of multiple targets, resulting in mission conflicts and inability to efficiently transmit data back. Summary of the invention

[0006] The purpose of the present invention is to address the deficiencies of the above-mentioned prior art and propose a joint planning method for multi-satellite collaborative remote communication missions for ecological monitoring. Multiple regions are striped according to relevant parameters of different satellites, with the goal of maximizing regional coverage, to achieve dynamic scheduling and real-time matching of satellite resources, thereby obtaining a better satellite scheduling scheme under multiple regional targets, and at the same time designing an efficient return mechanism to maximize the use of satellite and ground station resources and ensure the maximization of data transmitted back to the ground station.

[0007] To achieve the above object, the technical solution of the present invention comprises the following steps:

[0008] (1) Constructing a satellite network to be planned including a satellite set S and a regional target task set G, and a visible time window;

[0009] (2) Based on the satellite network to be planned, the strip information of each large-scale ecological area target is obtained by the satellite, and the meta-task is generated;

[0010] (3) Construct the observation meta-task set O of the satellite at different observation angles and divide it into meta-task sets O corresponding to multiple different regions. j ={o 1j ,o 2j ···o ij ···o mj}, where o ij Represents the meta-task generated by the i-th satellite observing the j-th region; i∈{1,2,...,m}, j∈{1,2,...,n}, m is the total number of satellites, and n is the total number of target regions to be observed;

[0011] (4) Set the input state space S and output action space of the task learning agent

[0012] (5) According to the state space S and the output action space Use the fully connected neural network FCN to build the policy function for each meta-task scheduling stage and the action-value function

[0013] Among them, θ represents the meta-task o ij The training parameters of the corresponding strategy function, represents the t-th time slot meta-task o in the state space S ij Status information, Representing the action space Select meta-task o in the tth time slot ij Scheduling strategy;

[0014] (6) Check whether the satellites selected at the same time conflict:

[0015] If the satellites selected at the same time conflict, the action with higher mission urgency and lower satellite energy consumption will be selected for execution;

[0016] If the satellites selected at the same time do not conflict, all actions will be executed to obtain status information

[0017] (7) Obtaining the strategy function for task resource matching

[0018] 7.1) Treat each regional target as an intelligent agent;

[0019] 7.2) Input to the policy function The probability of each action is obtained from the Get the state at time t+1 The reward for the action chosen by the winning agent v The reward for the action chosen by the losing agent l The reward for the action chosen by the conflict-free agent k and guides the selection of actions at time t+1;

[0020] 7.3) Input training experience into parameterized action-value function Get the expected future return y i , and then evaluate the action value and guide the strategy function Update the parameters θ until the policy function Converge and obtain the strategy function for task resource matching

[0021] (8) Input the meta-task corresponding to each regional target into the task resource matching strategy function Get the scheduling strategy for all regional target tasks;

[0022] (9) Based on the satellite selected in the scheduling strategy, the data is transmitted back to the selected ground station on the ground.

[0023] Compared with the prior art, the present invention has the following advantages:

[0024] 1. The present invention calculates the strips that can be scanned by each satellite at a specific time based on its unique orbital parameters and sensor pitch angle, splits the target area according to the scanned strips, and obtains multiple subtasks. Based on the information of these strips, a suitable satellite combination is selected, so that each satellite can effectively observe the corresponding area, which not only realizes the spatial decomposition of regional targets, but also makes full use of the observation advantages brought by the different pitch angles of satellites, thereby improving the flexibility, accuracy and resource utilization of the entire multi-satellite collaborative observation system, and maximizing the observation area of ​​the target area. It is particularly suitable for continuous, comprehensive and efficient monitoring tasks of multiple ecological regional targets in a wide area.

[0025] 2. The present invention adopts multi-agent learning, taking each large-scale ecological region to be observed as an intelligent agent, and distributing and scheduling each regional target according to its own urgency and observation status. Therefore, it can be oriented to multiple ecological regions, breaking through the traditional single-point or single-region observation mode, better simulating the situation of limited resources, numerous targets and mutual dependence between targets, and realizing simultaneous or orderly rotation observation of multiple ecological regions, thus improving the breadth and coverage of ecological monitoring.

[0026] 3. Since the present invention designs an efficient mechanism for transmitting data back to the ground station while scheduling tasks for multiple large-scale areas, it can respond to large-scale regional target tasks to be observed in real time and build a highly adaptive and scalable satellite observation network. Therefore, it can maximize the use of satellite and ground station resources, reduce the waiting time and retransmission times for data transmission, ensure the timely recovery and analysis of key ecological data, and provide strong technical support for environmental monitoring and protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a flow chart of the implementation of the present invention;

[0028] Figure 2 It is a schematic diagram of a large-scale regional target scheduling scenario constructed in the present invention;

[0029] Figure 3 It is a schematic diagram of the geometric relationship of satellite scanning strip coverage constructed in the present invention;

[0030] Figure 4 It is a schematic diagram of the process of performing agent learning in the present invention. DETAILED DESCRIPTION

[0031] The embodiments of the present invention are further described in detail below with reference to the accompanying drawings.

[0032] Reference Figure 1 , the implementation steps of this example are as follows:

[0033] Step 1: Build the satellite network to be planned and the visible time window.

[0034] 1.1) Select m satellites to form the satellite set S, and select n large-scale regional targets to form the task set G:

[0035] S={s1,s2,...s i ...,s m}

[0036] G={g1,g2,...,g j ,..,g n}

[0037] where s i represents the i-th satellite, g j represents the jth area to be observed, i∈{1,2,...,m}, j∈{1,2,...,n}, m is the number of satellites, and n is the number of target areas to be observed;

[0038] 1.2) The satellite set S and the target set G are sparsely distributed to form a satellite network to be planned, such as Figure 2 As shown;

[0039] 1.3) Calculate the i-th satellite s i For the jth target g in the observed area j The starting point of observation ij and the end observation point end ij The time length between is taken as the visible time window obs ij :

[0040] obs ij ={start ij ,end ij}.

[0041] Step 2: Based on the satellite network to be planned, obtain the strip information of each large-scale ecological area target scanned by the satellite and generate a meta-task.

[0042] 2.1) Select 4 boundary points in each target area to form the target g j The boundary point set P j :

[0043] P j = {P j1 (x1,y1),P j2 (x2,y2),P j3 (x3,y3),P j4 (x4,y4)}

[0044] Among them, (x1, y1), (x2, y2), (x3, y3), (x4, y4) are the longitude and latitude coordinates of the four boundary points of the region respectively;

[0045] 2.2) The visible time window obs between the i-th satellite and the j-th target area ij Next, calculate the satellite s i For regional targets j The minimum effective observation angle g s (i,j), maximum angle g e (i,j):

[0046]

[0047] where g min (i,j),g max (i, j) are the minimum and maximum angles of the i-th satellite pointing to the j-th vertex of the target area, respectively. i , maxg i Satellites i The minimum and maximum side view angles, Δg i For satellites i The field of view of the remote sensor;

[0048] 2.3) Construct the geometric relationship of satellite scanning strip coverage:

[0049] Reference Figure 3 , the specific implementation of this step is as follows:

[0050] Assume the satellite position at time t is point S, point O is the center of the earth, point A is the subsatellite point, and draw a perpendicular line from point A to the equatorial plane at point C;

[0051] At the start of observation time start ij In the figure, with SO as the center, the intersection points of the satellite line of sight forward deviation angle θ and backward deviation angle θ with the earth's surface are marked as L and R respectively;

[0052] At the end time point end ij In the following figure, with SO as the center, the intersection points of the satellite line of sight forward deviation angle θ and backward deviation angle θ with the earth surface are L' and R' respectively;

[0053] 2.4) Strip the target area at different angles g', that is, g' from g s (i,j) starts with an offset λ and gradually increases to g e (i, j), and calculate the latitude of point R at the observation angle g' Longitude λ R and the latitude of point L Longitude λ L :

[0054]

[0055] in, Start observation time ij The latitude of the subsatellite point below the satellite, i is the satellite's orbital inclination, h is the satellite's orbital altitude, and r is the radius of the earth.

[0056]

[0057] 2.5) Calculate the latitude of R' using the same formula as step 2.4) Longitude λ R' and the latitude of point L' Longitude λ L' ;

[0058] 2.6) Using the longitude and latitude coordinates of R, L, R', L' Form a scan strip.

[0059] Step 3: Construct the meta-task set O.

[0060] 3.1) Using satellites i , Regional Target j , the satellite's observation angle g', the satellite's start observation time for the target area start ij , end observation time end ij and scan strips Meta-task ij ;

[0061] 3.2) All meta-tasks form a meta-task set O:

[0062] O={o 11 ,o 12 ···o ij ···o mn}

[0063] Where m is the total number of satellites, n is the total number of target areas to be observed, i∈{1,2,...,m}, j∈{1,2,...,n};

[0064] 3.3) In order to adapt to multi-agent learning, the meta-task set O is divided into n different meta-task sets according to different targets in the observed area, so as to facilitate the subsequent action selection of the multi-agents. j Represents the meta-task corresponding to the j-th region:

[0065]

[0066] O j ={o1j ,o 2j ···o ij ···o mj}.

[0067] Step 4: Construct the input state space S and output action space of the task learning agent

[0068] 4.1) Using the energy E of the i-th satellite i , the urgency of the j-th regional target U j , the observation information M of the j-th regional target j The state S that constitutes the jth region target j :

[0069] S j ={E i ,U j ,M j}

[0070] Among them, M j is a binary number, M j = 1 means the jth region target is observed, M j =0 means that the target in the jth region has not been observed;

[0071] 4.2) Use the state of each regional target to form the state space S:

[0072] S={S1,S2,···S j ,···,S m};

[0073] 4.3) Using the meta-task O corresponding to the jth region in step 3.3) j Constitute the action space of the jth region

[0074]

[0075] Step 5: Build a strategy function and the action-value function

[0076] 5.1) Use the state space S as the input layer, the activation function ReLU as the activation layer, and the softmax function as the output layer to construct a fully connected neural network FCN1;

[0077] 5.2) Using neural network FCN1 to form the strategy function of the meta-task scheduling stage

[0078]

[0079] Among them, θ represents the meta-task o ij The training parameters of the corresponding strategy function, represents the state information of the tth time slot in the state space S, Representing the action space Select meta-task o in the tth time slot ij The scheduling strategy of , P(·|·) represents the conditional probability;

[0080] 5.3) and As the input layer, ReLU as the activation layer, and softmax function as the output layer, a fully connected neural network FCN2 is constructed;

[0081] 5.4) In order to evaluate the quality of the actions taken by the policy function, use this FCN2 to construct the meta-task o ij Action-value function in the scheduling phase

[0082]

[0083] in represents the meta-task o in the tth time slot ij Take Action , E(·) means the expected reward.

[0084] Step 6: Check whether the satellites selected at the same time conflict and update their status.

[0085] 6.1) Conflict Detection

[0086] Traverse all agents and determine whether any two agents j1 and j2 conflict:

[0087] If the same satellite is selected and It is considered a conflict, and the pair of agents (j1, j2) is added to the set C, and 6.2 is executed, where are the time windows of agents j1 and j2 and satellite i, respectively, and C is an empty set;

[0088] Otherwise, the agent does not detect a conflict, and the satellite that schedules the agent k is s. k , respectively update the satellite energy status to: Update the target observation status to: Update the urgency status to:

[0089] 6.2) Update the status in case of conflict:

[0090] 6.2.1) For the conflicting pair (j1, j2) ∈ C, calculate the ratio of the energy consumption of the satellites selected by agents j1 and j2 to the urgency of the agents η j2 :

[0091]

[0092] in They represent the current urgency of agents j1 and j2 at time t, respectively. Respectively represent the current satellite selected by agents j1 and j2 Observe the energy consumption required.

[0093] 6.2.2) Comparison Choose the side with the higher ratio as the winner:

[0094] if Then agent j1 wins, and execute 6.2.3)

[0095] Otherwise, agent j1 fails; execute 6.2.4)

[0096] 6.2.3) For the winning agent v, dispatch the satellite s of the agent v And update, that is, update the satellite energy status to: Update the target observation status to: Update the urgency status to:

[0097] 6.2.4) For the failed agent l, dispatch the satellite s of the agent l And update, that is, update the satellite energy status to: Update the target observation status to: Update the urgency status to: Where UF is the emergency factor, which represents the increase in the emergency level when the regional target is not observed.

[0098] Step 7: According to the action value function, the strategy function Train and finally get the task matching strategy function

[0099] Reference Figure 4 , the implementation of this step is as follows:

[0100] 7.1) Treat each target area as an agent j represents the agent corresponding to the jth target area, n is the total number of target areas, j∈{1,2,...,n};

[0101] 7.2) Calculate the state information of the t+1th time slot And the reward R after the agent takes action:

[0102] 7.2.1) The status information of the tth time slot Input to the policy function The probability of each action is obtained from the neural network FCN1;

[0103] 7.2.2) Each agent outputs an action according to the action probability, and determines the winning agent, the losing agent and the conflict-free agent based on the action, and obtains the satellite s at time t+1 i With regional target g j Status

[0104]

[0105] in represents the selected satellite s at time t+1 i The energy represents the urgency of the j-th regional target at time t+1, represents the observation state of the j-th regional target at time t+1;

[0106] 7.2.3) Calculate the reward R for the action taken by the winning agent v v , the reward R for the action taken by the losing agent f f , the reward R for the action taken by the conflict-free agent u u :

[0107]

[0108] Among them, e(s v ) represents the satellite s selected by the agent v energy consumption, λ(0<λ<1) is the penalty factor for energy consumption, which is used to guide the agent to choose actions with low energy consumption.

[0109] is the urgency of the v-th regional target at time t,

[0110] I(b v ,g v ) indicates scanning strip b v With regional target g v The intersection area of

[0111] Area(g v ) represents the regional target g v The area of

[0112] β is a positive weighting factor that balances the impact of urgency and scanning effectiveness, ensuring that both are appropriately weighted in the reward calculation.

[0113] γ is a positive penalty factor that emphasizes the negative impact of failing to observe high-urgency targets. Unobserved targets with high urgency are punished more severely, prompting the agent to pay more attention to urgent tasks in future decisions.

[0114] s u Indicates the satellite selected when there is no conflict.

[0115] Indicates the urgency of the target in the observation area at time t,

[0116] δ is the weight factor of the time window efficiency reward, which is used to encourage efficient completion of tasks within the time window, taking into account the effective utilization of resources and the overall optimization goal of the system.

[0117] end iu , start iu are the end time point and start time point of the i-th satellite observing the u-th regional target respectively;

[0118] 7.3) Using parameterized action-value functions Get the expected future return y i , and guides the policy function Update the training parameters θ of the meta-task:

[0119] 7.3.1) Set the collection of initial training experience for task scheduling in the current cycle:

[0120]

[0121] in, represents the empty set, Indicates that the collected task is o ij A collection of empirical data for allocating resources;

[0122] 7.3.2) The current state information S at time t t , Resource Scheduling Action The reward R for taking the action t And the status information S of the meta-task at time t+1 t+1 , stored in the collection In, it is expressed as:

[0123]

[0124] 7.3.3) Set up collection The maximum number of training data that can be stored in is MS. Repeat step 7.3.2) until The amount of data in reaches the maximum number MS;

[0125] 7.3.4) The state information S at time t t , resource scheduling action at time t The reward R for taking action at time t t And the status information S of the meta-task at time t+1 t+1 , input to In the function, we get the expected future return y of the jth agent j :

[0126]

[0127] where χ is the discount factor, is the parameterized action-value function at time t+1, is the satellite s at time t+1 i With regional target g j status, It is the action space Select meta-task o at time t+1 ij The scheduling strategy, is the parameter of the jth agent at time t+1;

[0128] 7.3.5) Based on the expected future reward y of the jth agent j , and get the mean square error loss function of the jth agent

[0129]

[0130] in Indicates from The state information S at time t randomly selected from t , resource scheduling action at time t The reward R for taking action at time t t And the state information S at time t+1 t+1 ; E[·] means to find the expectation;

[0131] 7.3.6) According to the mean square error loss function Get the network parameters for the t+1th time slot

[0132]

[0133] in is the network parameter of the tth time slot, α is the learning rate, is the loss function About parameters The gradient of

[0134] 7.3.7) Use the policy gradient method to update the policy function Parameter θ, get θ';

[0135]

[0136] in At time t, the state S t Next agent takes action The probability of is the jth agent in The expected future return corresponding to the nth piece of information selected;

[0137] 7.3.8) Repeat steps 7.3.6) and 7.3.7) until the mean square error loss function of all training cycle meta-tasks is All converge, and we get the strategy function for task resource matching.

[0138] Step 8: Set the satellite s at time t i With regional target g j Status Input to the strategy function for task resource matching In the output, the optimal scheduling strategy for regional target tasks is output, that is, the specific information of all regions observed by satellites.

[0139] Step 9: Based on the satellite selected in the scheduling strategy, the data is transmitted back to the selected ground station on the ground.

[0140] The existing methods for solving the problem of transmitting data back to the selected ground station include simulated annealing algorithm, multi-objective genetic algorithm, branch and bound method, etc. This example uses but is not limited to particle swarm algorithm, and its specific implementation is as follows:

[0141] 9.1) The satellite set I and ground station set K selected in the scheduling strategy are:

[0142] I={s1,s2···s i ···s m'}

[0143] K={k1,k2···k j ···k n'}

[0144] where s i represents the i-th satellite selected, i∈{1,2,···m'}, m' is the maximum number of selected satellites, k j represents the jth ground station, j∈{1,2,···n′}, n′ is the maximum number of selected ground stations;

[0145] 9.2) Set the position vector x at time t p (t) and the velocity vector v p (t):

[0146] x p (t) = {x 11 (t),x 12 (t),···x ij (t)···x m'n' (t)}

[0147] v p (t) = {v 11 (t),v 12 (t),···v ij (t)···v m'n' (t)}

[0148] Among them, x ij (t) is the satellite s at time t i The amount of data transmitted to the jth ground station, x ij ≤D max , D max is the maximum amount of data transmitted by the satellite to the ground station; v ij (t) is the satellite s at time t i The rate of change of the amount of data transmitted to the jth ground station, t = 1, 2···l, where l is the maximum number of time slots;

[0149] 9.3) According to the position vector x at time t p (t) and the velocity vector v p (t) Define particle p:

[0150] p = [x p (t),v p (t)];

[0151] 9.4) According to the satellite s at time t i The amount of data transmitted to the jth ground station is x ij (t) Establish the objective function F(t):

[0152]

[0153] 9.5) Calculate the objective function value of all particle position vectors;

[0154] 9.6) Take the position vector with the maximum objective function value of particle p as the individual optimal solution of particle p The position vector with the largest objective function value of all particles is taken as the global optimal solution x gbest ;

[0155] 9.7) Based on the individual optimal solution and the global optimal solution x gbest , update the position vector x of particle p at time t p (t) and velocity vector v p (t), get the position vector x at time t+1 p (t+1) and the velocity vector v p (t+1):

[0156] x p (t+1)=x p (t)+v p (t+1)

[0157]

[0158] Where w is the inertia weight, c1 and c2 are two acceleration constants with different values, and r1 and r2 are two random factors with different values;

[0159] 9.8) Establish the constraints of the objective function:

[0160]

[0161] x ij (t) = min(x ij (t),D max )

[0162] 9.9) Determine the satellite s at time t in the position vector of particle p i The amount of data transmitted to the jth ground station is x ij (t) Whether the constraints in 9.8) are met:

[0163] If satisfied, proceed to step 9.7).

[0164] If it is not satisfied, a new particle is randomly selected to replace it;

[0165] 9.10) Repeat steps 9.5)-9.9) until the position vector x of all particles at time t is p (t) and velocity vector v p (t) The number of updates reaches the maximum number M, and the global optimal solution x is obtained gbest .

[0166] The above description is only a specific example of the present invention and does not constitute any limitation to the present invention. It is obvious that for professionals in this field, after understanding the content and principles of the present invention, they may make various modifications and changes in form and details without departing from the principles and structures of the present invention. However, these modifications and changes based on the ideas of the present invention are still within the scope of protection of the claims of the present invention.

[0167] It should be noted that the step numbers in the specification and claims of the present invention are only for a clear description of the implementation scheme of the present invention, and the order of the step numbers is not limited.

Claims

1. A joint planning method for multi-satellite coordinated remote sensing missions for ecological monitoring. It is characterized in that The steps include: (1) Constructing a satellite network to be planned including a satellite set S and a regional target task set G, and a visible time window set; (2) Based on the satellite network to be planned, the strip information of each large-scale ecological area target is obtained by the satellite, and the meta-task is generated; (3) Construct the observation meta-task set O of the satellite at different observation angles and divide it into meta-task sets O corresponding to multiple different regions. j ={o 1j ,o 2j ···o ij ···o mj }, where o ij Represents the meta-task generated by the i-th satellite observing the j-th region; i∈{1,2,...,m}, j∈{1,2,...,n}, m is the total number of satellites, and n is the total number of target regions to be observed; (4) Set the input state space S and output action space of the task learning agent (5) According to the state space S and the output action space Use the fully connected neural network FCN to build the policy function for each meta-task scheduling stage and the action-value function Among them, θ represents the meta-task o ij The training parameters of the corresponding strategy function, represents the t-th time slot meta-task o in the state space S ij Status information, Representing the action space Select meta-task o in the tth time slot ij Scheduling strategy; (6) Check whether the satellites selected at the same time conflict: If the satellites selected at the same time conflict, the action with higher mission urgency and lower satellite energy consumption will be selected for execution; If the satellites selected at the same time do not conflict, all actions will be executed to obtain status information (7) Obtaining the strategy function for task resource matching 7a) Treat each regional target as an intelligent agent; 7b) Input to the policy function The probability of each action is obtained from the Get the state at time t+1 The reward for the action chosen by the winning agent v The reward for the action chosen by the losing agent l The reward for the action chosen by the conflict-free agent k and guides the selection of actions at time t+1; 7c) Input the training experience into the parameterized action-value function Get the expected future return y i , and then evaluate the action value and guide the strategy function Update the parameters θ until the policy function Converge and obtain the strategy function for task resource matching (8) Input the meta-task corresponding to each regional target into the task resource matching strategy function Get the scheduling strategy for all regional target tasks; (9) Based on the satellite selected in the scheduling strategy, the data is transmitted back to the selected ground station on the ground.

2. The method according to claim 1, Features: The satellite set S and target task set G in step (1) are respectively expressed as follows: S={s 1 ,s 2 ,···,s i ,···,s m } G={g 1 ,g 2 ,···,g j ,···,g n } where s i represents the i-th satellite, g j represents the jth area to be observed, i∈{1,2,...,m}, j∈{1,2,...,n}, m is the number of satellites, n is the number of target areas to be observed, and the visible time window Obs of each satellite and each target area is: Obs={obs 11 ,...,obs ij ,...obs mn }, obs ij ={start ij ,end ij } is the visible time window of the i-th satellite and the j-th target area, start ij 、end ij are the start and end observation time points of the i-th satellite and the j-th target area, respectively.

3. The method according to claim 1, It is characterized in that In step (2), based on the satellite network to be planned, the strip information of each large-scale ecological area target scanned by the satellite is obtained, and the meta-task is generated, which is implemented as follows: 2a) Select 4 boundary points in each target area to form the target g j The boundary point set P j : P j ={P j1 (x 1 ,y 1 ),P j2 (x 2 ,y 2 ),P j3 (x 3 ,y 3 ),P j4 (x 4 ,y 4 )} Where (x 1 ,y 1 ),(x 2 ,y 2 ),(x 3 ,y 3 ),(x 4 ,y 4 ) are the latitude and longitude coordinates of the four boundary points of the region; 2b) The visible time window obs between the i-th satellite and the j-th target area ij Next, calculate the satellite s i For regional targets j The minimum effective observation angle g s (i,j), maximum angle g e (i,j): where g min (i,j),g max (i, j) are the minimum and maximum angles of the i-th satellite pointing to the j-th vertex of the target area, respectively. i , maxg i Satellites i The minimum and maximum side view angles, Δg i For satellites i The field of view of the remote sensor; 2c) Let the satellite position at time t be point S, point O be the center of the earth, point A be the subsatellite point, draw a perpendicular line from point A to the equatorial plane at point C, and start the observation at the start time start. ij In the following example, with SO as the center, the intersection points of the satellite line of sight forward deviation angle θ and backward deviation angle θ with the earth's surface are marked as L and R respectively; at the end time point end ij In the following figure, with SO as the center, the intersection points of the satellite line of sight forward deviation angle θ and backward deviation angle θ with the earth surface are L' and R' respectively; 2d) Strip the target area at different angles g', i.e. g' from g s (i,j) starts with an offset λ and gradually increases to g e (i, j), and calculate the latitude of point R at the observation angle g' Longitude λ R and the latitude of point L Longitude λ L : in, Start observation time ij The latitude of the subsatellite point below the satellite, i is the orbital inclination of the satellite, h is the orbital height of the satellite, r is the radius of the earth, 2e) Calculate the latitude of R' using the same formula as step 2d) Longitude λ R' and the latitude of point L' Longitude λ L' ; Using the longitude and latitude coordinates of R, L, R', L' Constitute a scanning strip; 2f) Use of satellites i , Regional Target j , the satellite's observation angle g', the satellite's start observation time for the target area start ij , end observation time end ij and scan strips Constitute a meta-task o ij .

4. The method according to claim 1, Features In step (4), the input state space S and output action space of the agent learning task are constructed They are respectively expressed as follows: S={S 1 ,S 2 ,···S j ,···,S m } Where S j ={E i ,U j ,M j } represents the state of the agent corresponding to the j-th regional target in the environment, E i represents the energy of the ith satellite, U j represents the urgency of the jth regional target, M j Represents the binary state information of the j-th regional target, M j = 1 means the jth region target is observed, M j =0 means that the target in the jth region has not been observed.

5. The method according to claim 1, Features: The step (5) constructs the strategy function for each meta-task scheduling phase and the action-value function Among them, θ represents the meta-task o ij The training parameters of the corresponding strategy function, represents the meta-task o at time t in the state space S ij Status information, Representing the action space Select meta-task o in the tth time slot ij The scheduling strategy of , P(·|·) represents the conditional probability; represents the meta-task o in the tth time slot ij Take Action , E(·) means the expected reward.

6. The method according to claim 1, Features In step (6), detect whether the satellites selected at the same time conflict and update the status: 6a) Conflict Detection Traverse all agents, for any two agents j 1 and j 2 : If the same satellite is selected and It is marked as a conflict, and the pair of agents (j 1 ,j 2 ) is added to set C and 6b) is executed, where They are agent j 1 and j 2 The time window with satellite i, C is an empty set; Otherwise, the agent does not detect a conflict, and the satellite that schedules the agent k is s. k , respectively update the satellite energy status to: The target observation status is updated as: Urgency status update: 6b) For the conflicting agent j 1 ,j 2 Status updates with satellites: 6b1) For the conflicting agent pair (j 1 ,j 2 )∈C, computing agent j 1 and j 2 Ratio of energy consumption of selected satellites to agent urgency in They represent agent j at time t respectively. 1 ,j 2 The current urgency, Represents agent j 1 ,j 2 Select current satellite Energy consumption required for observation; 6b2) Comparison Choose the side with the higher ratio as the winner: if Then agent j 1 Win, execute step 6b3); Agent j 2 If failed, go to step 6b4) On the contrary, agent j 2 Win, execute step 6b3); Agent j 1 If failed, go to step 6b4); 6b3) The winning agent v dispatches the satellite s of the agent v : Update satellite energy status: Update the target observation status: Update urgency status: 6b4) The failed agent l dispatches the satellite s of the agent l : Update satellite energy status: Update the target observation status: Update urgency status:

7. The method according to claim 1, It is characterized in that Step 7b) gets the state at time t+1 It is expressed as follows: in represents the satellite s at time t+1 i energy, represents the urgency of the j-th regional target at time t+1, Represents the observation state of the j-th regional target at time t+1.

8. The method according to claim 1, It is characterized in that The reward for the action selected by the winning agent v obtained in step 7b) The reward for the action chosen by the losing agent l The reward for the action chosen by the conflict-free agent k They are respectively expressed as follows: Where e(s v ) represents the satellite s selected by the winning agent v v energy consumption, λ is the penalty factor for energy consumption, 0<λ<1, represents the urgency of the vth regional target at time t, I(b v ,g v ) indicates scanning strip b v With regional target g v The intersection area, Area(g v ) represents the regional target g v The area of ​​​​the ,ρ is a positive weight factor, is the urgency of the failed agent l, γ is a positive penalty factor, s k represents the satellite selected by agent k without conflict, represents the urgency of agent k without conflict at time t, δ is the weight factor of time window efficiency reward, end iu , start iu They represent the end time point and the start time point of the i-th satellite observing the k-th regional target respectively.

9. The method according to claim 1, It is characterized in that Step 7c) Input the training experience into the parameterized action-value function Get the expected future return y i : 7c1) Set the set of task training experience in the initial state 7c2) The state information S of each agent at the current time t t , Resource Scheduling Action The reward R for taking the action t And the status information S of the meta-task at time t+1 t+1 , stored in the collection In, it is expressed as: 7c3) Set up collection The maximum number of training data that can be stored in is MS. Repeat 7c2) to accumulate experience until The number of experience data in reaches the maximum number MS; 7c4) Each agent learns separately and parameterizes its action value function at time t+1 as in: Represents the meta-task o at time t+1 in the state space S ij Status information, Representing the action space Select meta-task o at time t+1 ij The scheduling strategy, is the parameter at time t+1, i∈{1,2,...,m}, j∈{1,2,...,n}, m is the number of satellites, and n is the number of regional targets; 7c5) The state information S of each agent in t , Resource Scheduling Action The reward R for taking the action t And the status information S of the meta-task at time t+1 t+1 , input to In the above example, we get the expected future return y of the jth agent. j : Where R t is the reward at time t, and χ is the discount factor.

10. The method according to claim 1, It is characterized in that Step 7c) Using expected future returns y j Guidance strategy function Update the parameter θ, which can be achieved as follows: 7c6) Based on the expected future return y j , we get the mean square error loss function in Indicates from Randomly extract the state information S of each agent t , Resource Scheduling Action The reward R for taking the action t And the status information S of the meta-task at time t+1 t+1 ; 7c7) Based on mean square error loss function Get the network parameters at time t+1 in is the network parameter at time t, α is the learning rate, is the loss function About parameters gradient. 7c8) Based on the expected future return y j Update strategy function Parameter θ, get the training parameter θ'; 7c9) Repeat 7c7) and 7c8) until the mean square error loss function of each agent is All converge, making the policy function The training parameter θ' no longer changes, and the strategy function for task resource matching is obtained 11. The method according to claim 1, It is characterized in that In step (9), according to the satellite selected in the scheduling strategy, the data is transmitted back to the ground station selected on the ground, which is implemented as follows: 9a) The satellite set I and ground station set K selected in the scheduling strategy: I={s 1 ,s 2 ···s i ···s m' } K={k 1 ,k 2 ···k j ···k n' } where s i represents the i-th satellite selected, i∈{1,2,···m'}, m' is the maximum number of selected satellites, k j represents the jth ground station, j∈{1,2,···n′}, n′ is the maximum number of selected ground stations; 9b) Set the position vector x at time t p (t) and the velocity vector v p (t): x p (t)={x 11 (t),x 12 (t),···x ij (t)···x m'n' (t)} v p (t)={v 11 (t),v 12 (t),···v ij (t)···v m'n' (t)} Among them, x ij (t) is the satellite s at time t i To ground station k j The amount of data transferred, x ij ≤D max , D max is the maximum amount of data transmitted by the satellite to the ground station; v ij (t) is the satellite s at time t i To ground station k j The rate of change of the amount of data transmitted, t = 1, 2···l, where l is the maximum number of time slots; 9c) According to the position vector x at time t p (t) and the velocity vector v p (t) Define particle p: 9d) According to the satellite s at time t i To ground station k j The amount of data transferred x ij (t) Establish the objective function: 9e) Take the maximum position of the objective function value of particle p as the individual optimal solution of particle p The maximum position of the objective function value of all particles is taken as the global optimal solution x gbest ; 9f) Update the particle position vector x at time t p (t) and velocity vector v p (t), get the position vector x at time t+1 p (t+1) and velocity vector v p (t+1): x p (t+1)=x p (t)+v p (t+1) Where w is the inertia weight, c 1 , c 2 are two acceleration constants with different values, r 1 , r 2 are two random factors with different values; 9g) Establish the constraints of the objective function: x ij (t)=min(x ij (t),D max ) 9h) Determine the x of particle p ij (t) Whether the constraints in 9g) are met: If satisfied, execute 9f). If it is not satisfied, a new particle is randomly selected to replace it; 9i) Repeat steps 9d)-9h) until the position vector x of all particles at time t is p (t) and velocity vector v p (t) The number of updates reaches the maximum number M, and the global optimal solution x is obtained gbest .

Citation Information

Cited By

  • Planning method and device of giant satellite base measurement and control data transmission task, electronic equipment and medium

    CN120106478A

  • Satellite remote control instruction automatic generation system based on artificial intelligence

    CN120454826A

  • An automatic satellite remote control command generation system based on artificial intelligence

    CN120454826B