Multi-satellite cooperative intelligent task planning method and system

Abstract

The invention provides a multi-satellite cooperative intelligent task planning method and system, and relates to the technical field of multi-satellite cooperative task planning. An input sequence is processed through a task planning model to obtain a satellite task planning scheme; wherein the task planning model comprises an embedded layer, an encoder and a decoder, and the embedded layer obtains an embedded matrix according to an input sequence; the encoder performs bidirectional encoding on the embedded matrix to obtain an output vector, a hidden state and a cell state; and the decoder performs node selection through an LSTM and an attention mechanism according to the output vector, the hidden state, the cell state, the input sequence and the embedded matrix of the encoder to obtain a task planning scheme. A task planning model based on a sequence-to-sequence structure is constructed, a high-quality feasible solution can be quickly obtained, the process that a heuristic algorithm based on deep reinforcement learning needs to be repeatedly iterated and optimized is avoided, and the technical problem that a multi-satellite cooperative task planning method is low in efficiency is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-satellite collaborative mission planning, and in particular to a multi-satellite collaborative intelligent mission planning method and system. Background Art

[0002] Earth observation satellites (hereinafter referred to as satellites) use onboard imaging sensors to observe the earth's surface to obtain image information. Due to their unique advantages such as wide observation range and no restrictions on geographical conditions and national boundaries, they have been widely used in military reconnaissance, agricultural production, emergency rescue and other fields. As the number of satellite resources increases, user imaging needs are also expanding rapidly, showing an "oversubscription" phenomenon. It is urgent to improve the multi-satellite collaborative mission planning capabilities and give full play to satellite service capabilities to meet more imaging needs.

[0003] Multi-satellite collaborative mission planning aims to meet user needs, optimize the scheduling of imaging tasks based on multi-satellite observation capabilities and collaborative capabilities, reasonably allocate tasks to each satellite, and determine the satellite observation time. Current research mainly focuses on exact algorithms and heuristic algorithms. Although the exact algorithm can obtain the optimal solution to the problem, it is only suitable for optimization calculations of small-scale instances. The meta-heuristic algorithm based on iterative optimization can obtain a better solution, but iterative optimization requires a lot of computing time, while the construction heuristic algorithm can quickly calculate a feasible solution, but the quality of the solution is difficult to guarantee.

[0004] With the rapid development of new-generation information technology, artificial intelligence technology has been successfully applied to the problem of multi-satellite collaborative mission planning. Most existing algorithms adopt a combination of deep reinforcement learning and heuristic algorithms. Although this can improve the performance of the algorithm, it still requires a lot of iterative optimization time and cannot solve such problems in an end-to-end manner, making it difficult to improve the efficiency of multi-satellite collaborative mission planning. Summary of the invention

[0005] 1. Technical issues to be solved

[0006] In view of the deficiencies in the prior art, the present invention provides a multi-satellite collaborative intelligent mission planning method and system, which solves the technical problem of low efficiency of the multi-satellite collaborative mission planning method.

[0007] (II) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] In a first aspect, the present invention provides a multi-satellite collaborative intelligent mission planning method, comprising:

[0010] Obtain mission data and satellite data to determine all observation windows for the mission;

[0011] All observation windows of the mission are used as input sequences, and the input sequences are processed by the pre-built mission planning model to obtain the satellite mission planning scheme;

[0012] The pre-built task planning model includes an embedding layer, an encoder and a decoder:

[0013] The embedding layer is used to map the input sequence from low-dimensional features to high-dimensional features to obtain an embedding matrix;

[0014] The encoder is used to bidirectionally encode the embedding matrix to obtain the output vector, hidden state, and cell state;

[0015] The decoder is used to select nodes through LSTM and attention mechanism based on the output vector, hidden state, cell state, input sequence and embedding matrix of the encoder to obtain the task planning solution.

[0016] Preferably, the input sequence is X={x 1 ,x 2 ,...,x |TW|}, the element set of each node in the input sequence is defined as

[0017] Among them, t i represents the i-th task; s j represents the jth satellite; represents the observation start time of the j-th satellite for the i-th mission, represents the end time of observation of the i-th mission by the j-th satellite; v i represents the revenue of the ith task; |TW i | indicates that the current observation window corresponds to task t i The number of observation windows; Indicates the number of conflicting observation windows in the current observation window; Q j represents the maximum storage capacity of the jth satellite; Indicates that the jth satellite continuously performs mission t i and task t i′ The conversion time required.

[0018] Preferably, the encoder uses Bi-LSTM to embed the matrix Perform bidirectional encoding, including:

[0019] For forward encoding, we obtain (o n ,h n ,C n ),

[0020]

[0021] Among them,n is the output vector of the forward encoder, h n and C n is the hidden state and cell state of the last time step of the forward encoding;

[0022] For reverse encoding, we obtain

[0023]

[0024] in, is the output vector of the reverse encoding, and is the hidden state and cell state of the last time step in reverse encoding;

[0025] The data in both the forward and reverse directions are spliced ​​and merged to obtain the output information of the encoder

[0026] in is the output vector, h and C are the hidden state and cell state of the last time step of the Bi-LSTM.

[0027] Preferably, the specific decoding process of the decoder specifically includes:

[0028] S21, let the selected node set Generate a Mask vector based on the input sequence; including:

[0029] The length of the Mask matrix is ​​equal to the length of the input sequence nodes. Initially, the Mask values ​​corresponding to all nodes are set to 1;

[0030] S22, update the Mask vector based on equations (1)-(4);

[0031] After a node is selected, its corresponding Mask value is updated to 0. At the same time, based on the set of nodes whose corresponding Mask values ​​are 1 in the input sequence, it is determined whether constraints (1)-(4) are satisfied. If not, the Mask value corresponding to the node is updated to 0. If all the corresponding values ​​of the nodes in the Mask vector are 0, decoding is completed.

[0032]

[0033] in,

[0034] Formula (1) means that if task t i is observed by the jth satellite, then otherwise

[0035] Formula (2) indicates that each task is observed at most once;

[0036] Formula (3) indicates that a satellite must have a certain switching time to continuously observe two tasks;

[0037] Formula (4) indicates that the storage capacity of each satellite during the observation mission cannot exceed its maximum storage capacity;

[0038] S23, if the Mask vector is all 0, execute step S28, otherwise execute step S24;

[0039] S24, using output node y t The corresponding embedding layer output The hidden state h at the previous time step t-1 t-1 and cell state C t-1 Calculate the hidden state h at the current time step t t and cell state C t ;

[0040]

[0041] S25, calculate the alignment vector through the additive attention mechanism, and then adjust the encoder output matrix based on the alignment vector Perform weighted accumulation to obtain the attention distribution gc t :

[0042]

[0043] Where V g and W g is a trainable parameter;

[0044] S26. Use the attention mechanism to obtain the selection probability P(y) of all nodes in the input sequence t+1 |Y t ,X):

[0045]

[0046] Where Y t is the currently selected output node set, y t+1 The output node that may be selected at the next moment;

[0047] S27, according to P(y t+1 |Y t ,X) Select output node y t+1 , let Y=Y∪{y t+1}, Return to step S22;

[0048] S28. Output the selected node set Y as the task planning solution.

[0049] Preferably, the pre-built task planning model is trained and optimized through an Actor Critic algorithm.

[0050] In a second aspect, the present invention provides a multi-satellite collaborative intelligent mission planning system, comprising:

[0051] Data acquisition module, used to obtain mission data and satellite data and determine all observation windows of the mission;

[0052] The mission planning module is used to take all observation windows of the mission as input sequences, process the input sequences through the pre-built mission planning model, and obtain the satellite mission planning scheme;

[0053] The pre-built task planning model includes an embedding layer, an encoder and a decoder:

[0054] The embedding layer is used to map the input sequence from low-dimensional features to high-dimensional features to obtain an embedding matrix;

[0055] The encoder is used to bidirectionally encode the embedding matrix to obtain the output vector, hidden state, and cell state;

[0056] The decoder is used to select nodes through LSTM and attention mechanism based on the output vector, hidden state, cell state, input sequence and embedding matrix of the encoder to obtain the task planning solution.

[0057] Preferably, the specific decoding process of the decoder specifically includes:

[0058] S21, let the selected node set Generate a Mask vector based on the input sequence; including:

[0059] The length of the Mask matrix is ​​equal to the length of the input sequence nodes. Initially, the Mask values ​​corresponding to all nodes are set to 1;

[0060] S22, update the Mask vector based on equations (1)-(4);

[0061] After a node is selected, its corresponding Mask value is updated to 0. At the same time, based on the set of nodes whose corresponding Mask values ​​are 1 in the input sequence, it is determined whether constraints (1)-(4) are satisfied. If not, the Mask value corresponding to the node is updated to 0. If all the corresponding values ​​of the nodes in the Mask vector are 0, decoding is completed.

[0062]

[0063] in,

[0064] Formula (1) means that if task t i is observed by the jth satellite, then otherwise

[0065] Formula (2) indicates that each task is observed at most once;

[0066] Formula (3) indicates that a satellite must have a certain switching time to continuously observe two tasks;

[0067] Formula (4) indicates that the storage capacity of each satellite during the observation mission cannot exceed its maximum storage capacity;

[0068] S23, if the Mask vector is all 0, execute step S28, otherwise execute step S24;

[0069] S24, using output node y t The corresponding embedding layer output The hidden state h at the previous time step t-1 t-1 and cell state C t-1 Calculate the hidden state h at the current time step t t and cell state C t ;

[0070]

[0071] S25, calculate the alignment vector through the additive attention mechanism, and then adjust the encoder output matrix based on the alignment vector Perform weighted accumulation to obtain the attention distribution gc t :

[0072]

[0073]

[0074] Where V g and W g is a trainable parameter;

[0075] S26. Use the attention mechanism to obtain the selection probability P(y) of all nodes in the input sequence t+1 |Y t ,X):

[0076]

[0077] Where Y t is the currently selected output node set, y t+1 The output node that may be selected at the next moment;

[0078] S27, according to P(y t+1 |Y t ,X) Select output node y t+1, let Y=Y∪{y t+1}, Return to step S22;

[0079] S28. Output the selected node set Y as the task planning solution.

[0080] Preferably, the pre-built task planning model is trained and optimized through an Actor Critic algorithm.

[0081] In a third aspect, the present invention provides a computer-readable storage medium storing a computer program for multi-satellite collaborative intelligent mission planning, wherein the computer program enables a computer to execute the multi-satellite collaborative intelligent mission planning method as described above.

[0082] In a fourth aspect, an electronic device includes:

[0083] One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include methods for executing the multi-satellite collaborative intelligent mission planning method as described above.

[0084] (III) Beneficial effects

[0085] The present invention provides a multi-satellite collaborative intelligent mission planning method and system. Compared with the prior art, it has the following beneficial effects:

[0086] The present invention constructs a mission planning model based on a sequence-to-sequence structure, which can quickly obtain high-quality feasible solutions, avoid the repeated iterative optimization process required by the heuristic algorithm based on deep reinforcement learning, and solve the technical problem of low efficiency of multi-satellite collaborative mission planning methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0088] Figure 1 A block diagram of a multi-satellite collaborative intelligent mission planning method according to an embodiment of the present invention;

[0089] Figure 2 Update the schematic for the Mask vector. DETAILED DESCRIPTION

[0090] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0091] The embodiments of the present application provide a multi-satellite collaborative intelligent mission planning method and system to solve the technical problem of low efficiency of the multi-satellite collaborative mission planning method, and achieve the ability to quickly calculate feasible solutions similar to the construction of heuristic algorithms, avoiding the need for repeated iterative optimization of heuristic algorithms based on deep reinforcement learning. At the same time, it has strong search capabilities, can obtain high-quality feasible solutions, and generate better satellite mission planning solutions in a shorter computing time.

[0092] The technical solution in the embodiment of the present application is to solve the above technical problems, and the overall idea is as follows:

[0093] With the rapid development of new generation information technology, artificial intelligence technology has been successfully applied to the problem of multi-satellite collaborative mission planning. Although the existing heuristic algorithm based on deep reinforcement learning can improve the performance of the algorithm, it still requires a lot of iterative optimization time and cannot be solved in an end-to-end manner. The satellite mission planning method based on the end-to-end approach is only designed for a single satellite and is difficult to apply to the problem of multi-satellite collaborative mission planning. For this reason, the embodiment of the present invention carries out multi-satellite collaborative mission planning based on the idea of ​​artificial intelligence, solves the problem in an end-to-end manner, and improves the intelligence level of satellite mission planning.

[0094] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0095] The embodiment of the present invention provides a multi-satellite collaborative intelligent mission planning method, such as Figure 1 As shown, including:

[0096] S1. Obtain mission data and satellite data and determine all observation windows of the mission;

[0097] S2, taking all observation windows of the mission as input sequences, processing the input sequences through the pre-built mission planning model to obtain the satellite mission planning scheme;

[0098] The pre-built task planning model includes an embedding layer, an encoder and a decoder:

[0099] The embedding layer is used to map the input sequence from low-dimensional features to high-dimensional features to obtain an embedding matrix;

[0100] The encoder is used to bidirectionally encode the embedding matrix to obtain the output vector, hidden state, and cell state;

[0101] The decoder is used to select nodes through LSTM and attention mechanism based on the output vector, hidden state, cell state, input sequence and embedding matrix of the encoder to obtain the task planning solution.

[0102] The embodiment of the present invention constructs a mission planning model based on a sequence-to-sequence structure, which can quickly obtain high-quality feasible solutions, avoid the repeated iterative optimization process required by the heuristic algorithm based on deep reinforcement learning, and solve the technical problem of low efficiency of multi-satellite collaborative mission planning methods.

[0103] In one embodiment, S1, obtain mission data and satellite data, and determine all observation windows of the mission. The specific implementation process is as follows:

[0104] Obtain and define mission data and satellite data, and determine all observation windows for each mission based on the mission data.

[0105] Satellite data mainly includes:

[0106] S: satellite set, |S| represents the total number of satellites in the set, S = {s1,s2,...,s |S|};

[0107] s j : jth satellite;

[0108] Q j : The maximum storage capacity of the jth satellite;

[0109] α j : Storage occupancy per unit observation time of the j-th satellite.

[0110] Task data mainly includes:

[0111] T: task set, |T| represents the total number of tasks in the set, T={t1,t2,...,t |T|};

[0112] t i : the i-th task;

[0113] v i : The income of the i-th task;

[0114] The jth satellite continuously performs mission t i and task t i′ The time required for the changeover;

[0115] The i-th mission is in the observation window of the j-th satellite. represents the observation start time of the j-th satellite for the i-th mission, Indicates the end time of observation of the i-th mission by the j-th satellite;

[0116] |TW i |: The current observation window corresponds to task t i The number of observation windows;

[0117] The number of conflicting observation windows of the current observation window, i.e.

[0118] In one embodiment, S2, all observation windows of the mission are used as input sequences, and the input sequences are processed by a pre-built mission planning model to obtain a satellite mission planning solution. The specific implementation process is as follows:

[0119] The multi-satellite collaborative mission planning problem mainly involves planning the observation windows between satellites and missions. Therefore, all observation windows of the mission are taken as the input sequence X = {x 1 ,x 2 ,...,x |TW|}, the element set of each node is defined as

[0120] In an embodiment of the present invention, the pre-built task planning model includes an embedding layer, an encoder and a decoder.

[0121] The fully connected neural network is used as the embedding layer of the task planning model. The embedding layer is used to map the input sequence from low-dimensional features to high-dimensional features to obtain the embedding matrix Right now Where W e is the weight matrix of the embedding layer.

[0122] The encoder performs bidirectional encoding on the embedding matrix to obtain the output vector Hidden state h and cell state C. The specific implementation process is as follows:

[0123] The encoder uses a Bi-LSTM structure to embed the matrix For forward coding, we can obtain (o n ,h n ,C n ),

[0124]

[0125] Among them, n is the output vector of the forward encoder, h n and C nis the hidden state and cell state of the last time step of the forward encoding;

[0126] For reverse encoding, we can obtain

[0127]

[0128] in, is the output vector of the reverse encoding, and is the hidden state and cell state of the last time step in reverse encoding.

[0129] Since Bi-LSTM generates data in two directions, which is inconsistent with the decoder input, the data in the positive and negative directions are concatenated and fused to obtain the output information of the encoder. in is the output vector, h and C are the hidden state and cell state of the last time step of the Bi-LSTM.

[0130] The decoder outputs the vector Hidden state h, cell state C, input sequence X, embedding matrix sequence The information is used to select nodes through LSTM and attention mechanism to obtain the task planning scheme. The specific implementation process is as follows:

[0131] S21, let the selected node set Generate a Mask vector based on X;

[0132] In the embodiment of the present invention, the Mask vector represents various constraints, such as Figure 2 As shown, in the process of selecting the input sequence nodes, the Mask vector is dynamically updated through the constraints.

[0133] Initialize the Mask matrix:

[0134] The length of the Mask matrix is ​​equal to the length of the input sequence nodes, and the Mask values ​​corresponding to all nodes are initially set to 1.

[0135] S22, update the Mask vector based on equations (1)-(4);

[0136] Update the Mask vector:

[0137] After a node is selected, its corresponding Mask value is updated to 0. At the same time, based on the set of nodes whose corresponding Mask values ​​are 1 in the input sequence, it is determined whether constraints (1)-(4) are satisfied. If not, the Mask value corresponding to the node is updated to 0. If all the corresponding values ​​of all nodes in the Mask vector are 0, decoding is completed.

[0138]

[0139] in,

[0140] Formula (1) means that if task t i is observed by the jth satellite, then otherwise

[0141] Formula (2) indicates that each task is observed at most once;

[0142] Formula (3) indicates that a satellite must have a certain switching time to continuously observe two tasks;

[0143] Formula (4) indicates that the storage capacity of each satellite during the observation mission cannot exceed its maximum storage capacity;

[0144] S23, if the Mask vector is all 0, execute step S28, otherwise execute step S24;

[0145] S24, using output node y t The corresponding embedding layer output h at the previous time step t-1 t-1 and C t-1 Calculate h at the current time step t t and C t ;

[0146]

[0147] S25, calculate the alignment vector through the additive attention mechanism, and then adjust the encoder output matrix based on the alignment vector Perform weighted accumulation to obtain the attention distribution gc t :

[0148]

[0149]

[0150] Where V g and W g is a trainable parameter.

[0151] S26. Use the attention mechanism to obtain the selection probability P(y) of all nodes in the input sequence t+1 |Y t ,X):

[0152]

[0153] Where Y t is the currently selected output node set, y t+1The output node that may be selected at the next moment;

[0154] S27, according to P(y t+1 |Y t ,X) Select output node y t+1 , let Y=Y∪{y t+1}, Return to step S22.

[0155] S28. Output the selected node set Y as the task planning solution.

[0156] In the embodiment of the present invention, the pre-built task planning model is trained by the Actor Critic algorithm, and the Actor network is the task planning model constructed above. In the M sample sequences, the network is used to sample and select the mth sample sequence index to form a planning result Y m The Critic network is a one-dimensional convolutional network that shares an embedding layer with the Actor network. The network is used to evaluate the total mission benefit of the mth sample sequence to obtain an evaluation value V. The ActorCritic algorithm training model can effectively learn and improve the satellite mission planning strategy, so that the model can obtain a better satellite mission planning solution.

[0157] An embodiment of the present invention provides a multi-satellite collaborative intelligent mission planning system, including:

[0158] Data acquisition module, used to obtain mission data and satellite data and determine all observation windows of the mission;

[0159] The mission planning module is used to take all observation windows of the mission as input sequences, process the input sequences through the pre-built mission planning model, and obtain the satellite mission planning scheme;

[0160] The pre-built task planning model includes an embedding layer, an encoder and a decoder:

[0161] The embedding layer is used to map the input sequence from low-dimensional features to high-dimensional features to obtain an embedding matrix;

[0162] The encoder is used to bidirectionally encode the embedding matrix to obtain the output vector, hidden state, and cell state;

[0163] The decoder is used to select nodes through LSTM and attention mechanism based on the output vector, hidden state, cell state, input sequence and embedding matrix of the encoder to obtain the task planning solution.

[0164] It can be understood that the multi-satellite collaborative intelligent mission planning system provided by the embodiment of the present invention corresponds to the above-mentioned multi-satellite collaborative intelligent mission planning method, and the explanations, examples, beneficial effects and other parts of its relevant contents can refer to the corresponding contents in the multi-satellite collaborative intelligent mission planning method, which will not be repeated here.

[0165] An embodiment of the present invention also provides a computer-readable storage medium, which stores a computer program for multi-satellite collaborative intelligent mission planning, wherein the computer program enables a computer to execute the multi-satellite collaborative intelligent mission planning method as described above.

[0166] An embodiment of the present invention also provides an electronic device, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include a method for executing the multi-satellite collaborative intelligent mission planning method as described above.

[0167] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0168] 1. The task planning model constructed by the embodiment of the present invention is based on a sequence-to-sequence structure, which can quickly obtain high-quality feasible solutions, avoid the process of repeated iterative optimization required by the heuristic algorithm based on deep reinforcement learning, and solve the technical problem of low efficiency of multi-satellite collaborative task planning methods. At the same time, it has strong search capabilities, can obtain high-quality feasible solutions, and generate better satellite task planning solutions in a shorter computing time.

[0169] 2. The embodiment of the present invention establishes an embedding layer based on a fully connected network, maps the input sequence from a low-dimensional space to a high-dimensional vector space, enhances the representation capability of the observation window features, more fully expresses the complex temporal and spatial correlations between the observation windows, and enables the model to learn a generalized representation of the input data, which helps to improve the applicability of the model and support multi-satellite collaborative intelligent mission planning in different mission scenarios.

[0170] 3. The embodiment of the present invention establishes a Bi-LSTM encoder and uses the bidirectional propagation characteristics to encode the embedding matrix to obtain more comprehensive context information, which is convenient for the model to understand and learn the spatiotemporal correlation between observation windows. At the same time, it can process the past and future information of the input sequence, assist the model to effectively capture and memorize the long-term dependency between observation times, improve the rationality and effectiveness of observation window selection in the subsequent decoding process, and optimize the satellite mission planning scheme.

[0171] 4. The embodiment of the present invention designs a decoder that combines LSTM and attention mechanism, which selectively focuses on a specific part of the input sequence in each time step, maintains sensitivity to key information, makes the decoding process of the model more efficient, and can accurately select the observation window to be output, thereby achieving better resource allocation and time scheduling in complex satellite mission planning scenarios.

[0172] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0173] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-satellite collaborative intelligent mission planning method, characterized in that: include: Obtain mission data and satellite data to determine all observation windows for the mission; All observation windows of the mission are used as input sequences, and the input sequences are processed by the pre-built mission planning model to obtain the satellite mission planning scheme; The pre-built task planning model includes an embedding layer, an encoder and a decoder: The embedding layer is used to map the input sequence from low-dimensional features to high-dimensional features to obtain an embedding matrix; The encoder is used to bidirectionally encode the embedding matrix to obtain the output vector, hidden state, and cell state; The decoder is used to select nodes through LSTM and attention mechanism based on the output vector, hidden state, cell state, input sequence and embedding matrix of the encoder to obtain the task planning solution.

2. The multi-satellite collaborative intelligent mission planning method according to claim 1, characterized in that: The input sequence is X={x 1 ,x 2 ,...,x |TW| }, the element set of each node in the input sequence is defined as Among them, t i represents the i-th task; s j represents the jth satellite; represents the observation start time of the j-th satellite for the i-th mission, represents the end time of observation of the i-th mission by the j-th satellite; v i represents the revenue of the ith task; |TW i | indicates that the current observation window corresponds to task t i The number of observation windows; Indicates the number of conflicting observation windows in the current observation window; Q j represents the maximum storage capacity of the jth satellite; Indicates that the jth satellite continuously performs mission t i and task t i′ The conversion time required.

3. The multi-satellite collaborative intelligent mission planning method according to claim 1, characterized in that: The encoder uses a Bi-LSTM to embed the matrix Perform bidirectional encoding, including: For forward encoding, we obtain (o n ,h n ,C n ), Among them, n is the output vector of the forward encoder, h n and C n is the hidden state and cell state of the last time step of the forward encoding; For reverse encoding, we obtain in, is the output vector of the reverse encoding, and is the hidden state and cell state of the last time step in reverse encoding; The data in both the forward and reverse directions are spliced ​​and merged to obtain the output information of the encoder in is the output vector, h and C are the hidden state and cell state of the last time step of the Bi-LSTM.

4. The multi-satellite collaborative intelligent mission planning method according to any one of claims 1 to 3, characterized in that: The specific decoding process of the decoder specifically includes: S21, let the selected node set Generate a Mask vector based on the input sequence; including: The length of the Mask matrix is ​​equal to the length of the input sequence nodes. Initially, the Mask values ​​corresponding to all nodes are set to 1; S22, update the Mask vector based on equations (1)-(4); After a node is selected, its corresponding Mask value is updated to 0. At the same time, based on the set of nodes whose corresponding Mask values ​​are 1 in the input sequence, it is determined whether constraints (1)-(4) are satisfied. If not, the Mask value corresponding to the node is updated to 0. If all the corresponding values ​​of the nodes in the Mask vector are 0, decoding is completed. in, Formula (1) means that if task t i is observed by the jth satellite, then otherwise Formula (2) indicates that each task is observed at most once; Formula (3) indicates that a satellite must have a certain switching time to continuously observe two tasks; Formula (4) indicates that the storage capacity of each satellite during the observation mission cannot exceed its maximum storage capacity; S23, if the Mask vector is all 0, execute step S28, otherwise execute step S24; S24, using output node y t The corresponding embedding layer output The hidden state h at the previous time step t-1 t-1 and cell state C t-1 Calculate the hidden state h at the current time step t t and cell state C t ; S25, calculate the alignment vector through the additive attention mechanism, and then adjust the encoder output matrix based on the alignment vector Perform weighted accumulation to obtain the attention distribution gc t : Where V g and W g is a trainable parameter; S26. Use the attention mechanism to obtain the selection probability P(y) of all nodes in the input sequence t+1 |Y t ,X): Where Y t is the currently selected output node set, y t+1 The output node that may be selected at the next moment; S27, according to P(y t+1 |Y t ,X) Select output node y t+1 , let Y=Y∪{y t+1 }, Return to step S22; S28. Output the selected node set Y as the task planning solution.

5. The multi-satellite collaborative intelligent mission planning method according to any one of claims 1 to 3, characterized in that: The pre-built task planning model is trained and optimized through the Actor Critic algorithm.

6. A multi-satellite collaborative intelligent mission planning system, characterized in that: include: Data acquisition module, used to obtain mission data and satellite data and determine all observation windows of the mission; The mission planning module is used to take all observation windows of the mission as input sequences, process the input sequences through the pre-built mission planning model, and obtain the satellite mission planning scheme; The pre-built task planning model includes an embedding layer, an encoder and a decoder: The embedding layer is used to map the input sequence from low-dimensional features to high-dimensional features to obtain an embedding matrix; The encoder is used to bidirectionally encode the embedding matrix to obtain the output vector, hidden state, and cell state; The decoder is used to select nodes through LSTM and attention mechanism based on the output vector, hidden state, cell state, input sequence and embedding matrix of the encoder to obtain the task planning solution.

7. The multi-satellite collaborative intelligent mission planning system according to claim 6, characterized in that: The specific decoding process of the decoder specifically includes: S21, let the selected node set Generate a Mask vector based on the input sequence; including: The length of the Mask matrix is ​​equal to the length of the input sequence nodes. Initially, the Mask values ​​corresponding to all nodes are set to 1; S22, update the Mask vector based on equations (1)-(4); After a node is selected, its corresponding Mask value is updated to 0. At the same time, based on the set of nodes whose corresponding Mask values ​​are 1 in the input sequence, it is determined whether constraints (1)-(4) are satisfied. If not, the Mask value corresponding to the node is updated to 0. If all the corresponding values ​​of the nodes in the Mask vector are 0, decoding is completed. in, Formula (1) means that if task t i is observed by the jth satellite, then otherwise Formula (2) indicates that each task is observed at most once; Formula (3) indicates that a satellite must have a certain switching time to continuously observe two tasks; Formula (4) indicates that the storage capacity of each satellite during the observation mission cannot exceed its maximum storage capacity; S23, if the Mask vector is all 0, execute step S28, otherwise execute step S24; S24, using output node y t The corresponding embedding layer output The hidden state h at the previous time step t-1 t-1 and cell state C t-1 Calculate the hidden state h at the current time step t t and cell state C t ; S25, calculate the alignment vector through the additive attention mechanism, and then adjust the encoder output matrix based on the alignment vector Perform weighted accumulation to obtain the attention distribution gc t : Where V g and W g is a trainable parameter; S26. Use the attention mechanism to obtain the selection probability P(y) of all nodes in the input sequence t+1 |Y t ,X): Where Y t is the currently selected output node set, y t+1 The output node that may be selected at the next moment; S27, according to P(y t+1 |Y t ,X) Select output node y t+1 , let Y=Y∪{y t+1 }, Return to step S22; S28. Output the selected node set Y as the task planning solution.

8. The multi-satellite collaborative intelligent mission planning system according to claim 6 or 7, characterized in that: The pre-built task planning model is trained and optimized through the Actor Critic algorithm.

9. A computer-readable storage medium, characterized in that: It stores a computer program for multi-satellite collaborative intelligent mission planning, wherein the computer program enables a computer to execute the multi-satellite collaborative intelligent mission planning method as described in any one of claims 1 to 5.

10. An electronic device, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include a program for executing the multi-satellite collaborative intelligent mission planning method as described in any one of claims 1 to 5.

Images