A mission planning method for geosynchronous orbit SAR satellite based on target trajectory

Abstract

The present invention discloses a method for geosynchronous orbit SAR satellite mission planning based on target trajectory, and belongs to the field of satellite mission planning technology. The method includes the steps of obtaining historical trajectory of observed targets, data preprocessing, prediction model building, model training and verification, prediction data processing, and mission planning based on greedy algorithm. The present invention uses a trajectory prediction method in the environment of geosynchronous orbit synthetic aperture radar satellite, which improves the efficiency and reliability of mission planning.

Description

Technical Field

[0001] The present invention relates to the field of satellite control technology, in particular to satellite mission planning, and in particular to a geosynchronous orbit SAR satellite mission planning method based on target trajectory. Background Art

[0002] Compared with optical satellites, synthetic aperture radar (SAR) satellites have all-day and all-weather high-resolution earth observation capabilities, so synthetic aperture radar satellites have a faster response capability to natural disasters. In August 2023, Land Exploration No. 4 01, the world's first geosynchronous orbit SAR satellite, successfully entered its working orbit, which can effectively meet the needs of disaster emergency monitoring. After a natural disaster occurs, by observing the movement information of vehicles, ships, aircraft and other means of transportation in the disaster-stricken areas, the outside world can promptly grasp the movement and dispatch of disaster-stricken people and rescue personnel, and improve rescue efficiency. In addition, in multi-target observation tasks, when different targets have a strong correlation, the value of multiple related targets at the same time is often higher than the sum of the values ​​of each target calculated separately.

[0003] When the communication system is damaged, it becomes extremely inefficient and difficult to send disaster area information from within the disaster area. Therefore, it is particularly important to observe the disaster area around the clock through SAR satellites. How to use limited satellite resources to achieve the highest value of observation is an important issue facing satellite mission planning. Thanks to the popularity of mobile devices, the target trajectory prediction problem can be learned using a large amount of historical data. Using neural network models to predict the trajectory of mobile targets can play a strong auxiliary role in satellite mission planning, thereby making full use of satellite resources.

[0004] Therefore, there is an urgent need for a geosynchronous orbit SAR satellite mission planning method that can quickly plan observation missions in the face of natural disasters. Summary of the invention

[0005] The purpose of the present invention is to fill the gap in the prior art and provide a geosynchronous orbit SAR satellite mission planning method based on target trajectory to achieve rapid mission planning for geosynchronous orbit synthetic aperture radar satellites.

[0006] The technical solution adopted by the present invention is:

[0007] A method for planning a geosynchronous orbit SAR satellite mission based on target trajectories is used to generate an observation mission sequence, wherein each observation mission observes all observation targets within a wave position coverage area through a wave position of a satellite radar; the method comprises the following steps:

[0008] Step 1: Obtain the user's observation requirements, including observation period, observation range and observation target, calculate each wave position of the satellite radar according to the observation range, and obtain the target category, target individual, historical trajectory, target observation value and target observation difficulty of each observation target;

[0009] Step 2: Set the upper limit of each observation task to the maximum exposure time required for all observation targets at each wave position. , set 0.5 seconds as the basic duration, and the total duration of the total observation task is the duration of the user's required observation period The observation time interval of each observation task is divided into time periods with the basic duration as the step length. part;

[0010] Step 3: Construct a trajectory probability prediction model based on a gated recurrent unit, train the model, and obtain a trained trajectory probability prediction model;

[0011] Step 4: Set the observation task number ;

[0012] Step 5: Arrange the historical trajectories of each observed target as input data, input them into the trained trajectory probability prediction model, and calculate the probability of each observed target in the first The distribution probability of each time period and each wave position during an observation mission ,in, is the wave position number, is the number of the time period in the observation time series, is a target individual in the target category X;

[0013] Step 6: For Observation tasks, calculate the average expected observation value of each wave position in different time periods;

[0014] Step 7: Based on the greedy algorithm, gradually select the wave position and task observation duration with the highest average expected observation value, determine each observation task one by one, output the observation task sequence, and complete the task planning.

[0015] Furthermore, in step 1, the target categories are classified according to the minimum distinguishable categories of the targets; the target observation values ​​include the individual observation values ​​of the targets of each category, the joint gain observation values ​​between different numbers of targets of each category, and the joint gain observation values ​​between different numbers of targets of different categories. ,in is the individual observation value of each category of targets, the joint gain observation value between different numbers of targets in each category, and the joint gain observation value number between different numbers of targets in different categories, totaling kind.

[0016] Furthermore, in step 3, the trajectory probability prediction model takes the historical trajectory data of the target as input, the predicted trajectory data of the target as the intermediate output, and the distribution probability of the observed target in each time period and each wave position as the final output. The trajectory probability prediction model includes an input layer, a GRU hidden layer, an intermediate output layer and a final output layer. The input layer of the trajectory probability prediction model receives input data, that is, an information vector group of multiple time steps. The activation function of the GRU hidden layer is the tanh function, and the cyclic Dropout ratio is between 0.2 and 0.5. The intermediate output layer generates an intermediate vector group, and the final output layer is a Dense layer with a Softmax activation function.

[0017] Furthermore, in step 3, the specific method of training the model is:

[0018] Obtain the historical trajectory of the target and map the target coordinates at each moment in the trajectory into wave position numbers;

[0019] The time value of each moment in the trajectory , target coordinates , Wave position number corresponding to the target coordinates , Target speed Merge into the information vector of the current target ;

[0020] For each moment in the trajectory, all information vectors at and before that moment are used as input data of a set of training data, and all information vectors after that moment are used as output data of the intermediate output layer of the set of training data, thereby obtaining multiple sets of training data;

[0021] The trajectory probability prediction model is trained using the above training data. The trajectory probability prediction model uses the Adam optimizer, and the loss function is the classification cross entropy loss function. When the loss no longer decreases within the preset number of times, the training is stopped to obtain the trained trajectory probability prediction model.

[0022] Furthermore, the specific method of step 6 is:

[0023] Step 601: Generate target categories The observation probability matrix of the target individual in ;in For is a diagonal matrix with diagonal elements, is the observation success rate function, used to calculate the irradiation time at the standard wave position Next, in The probability of finding a target in target category X at the wave position;

[0024] Step 602: Establishing a historical value equilibrium function :

[0025]

[0026] in, is a coefficient less than 0, is a constant;

[0027] Step 603: Calculate the value correction matrix of the target individual observation probability:

[0028]

[0029] in, For the Target categories in observation tasks The observation probability diagonal matrix of , in particular, is a zero matrix;

[0030] Step 604: Generate Species containing A diagonal matrix with 1s and all other elements are 0 ;in, For the target category The total number of target individuals in, m=1,2,3...Y, is the number of the diagonal matrix;

[0031] Step 605: Calculation The expected function of the observation probability of a target individual is:

[0032] ,

[0033] in, is the unit diagonal matrix;

[0034] Step 606: Calculate the Among the observation tasks, Wave position and time period Next Target Observation Value The observation probability expectation function :

[0035] ;

[0036] Among them, x represents the number of the target category, ;

[0037] Step 607: Calculate the Among the observation tasks, The wave position in the time period The expected observation value function ;

[0038] Step 608: Calculate the Among the observation tasks, The wave position in the time period The average expected observation value function ;

[0039] Step 609: Generate The average expected observation value matrix of observation tasks , is a matrix with B rows and 2T columns, where B is the total number of wave positions and 2T is the total number of time periods in the observation time series. The element in row b and column t is .

[0040] Further, in step 601, the observation success rate function for:

[0041]

[0042] in, The observed target category is The target observation difficulty coefficient is For the The wave position in the time period The standard wave position exposure time is calculated as follows:

[0043]

[0044] in, for The mapping coefficient of the wave position to the equivalent irradiation of the standard wave position of the sub-satellite point.

[0045] Furthermore, the specific method of step 7 is:

[0046] Step 701: Select the average expected observation value matrix Maximum value Wave position and time period , generate candidate observation tasks, calculation ;

[0047] Step 702: If , execute step 710 to step 711, otherwise continue with step 703;

[0048] Step 703: If , then execute step 704 to step 705, otherwise, execute step 709 to step 711; To correct the threshold, >1;

[0049] Step 704: Only in step 704, let the historical value equilibrium function is 0; recalculate the average expected observation value matrix , select The maximum value in ;

[0050] Step 705: If , then execute step 706 to step 711; otherwise, execute step 709 to step 711;

[0051] Step 706: Select middle Time period ,make , select middle Maximum value of the column ,make ; Select middle Time period ,make , select middle Maximum value of the column ,make ;

[0052] Step 707: When and When , then repeat step 706; otherwise, let , select middle Maximum value of the column Wave position and time period , generate candidate observation tasks, calculation ,make , select middle Maximum value of the column Wave position and time period , generate candidate observation tasks, calculation ;

[0053] Step 708: When or When middle Maximum value of the column Wave position and time period , generate candidate observation tasks, calculation , select middle Maximum value of the column Wave position and time period , generate candidate observation tasks, calculation ;

[0054] Step 709: Output candidate The observation task is An observation task, the wave position of which is , the total duration is the basic duration times;

[0055] Step 710: Calculate the remaining time , where the initial remaining time ;

[0056] Step 711: If , , repeat step 5 to step 7; otherwise, execute step 712 to step 713;

[0057] Step 712: Output candidate The observation task is An observation task, the wave position of which is , the total duration is the basic duration times;

[0058] Step 713: All observation tasks are organized into an observation task sequence to complete task planning.

[0059] The beneficial effects of the present invention are:

[0060] 1. The present invention uses a trajectory prediction method in the environment of a geosynchronous orbit synthetic aperture radar satellite to improve the efficiency and reliability of mission planning.

[0061] 2. By constructing a trajectory probability prediction model based on the gated recurrent unit (GRU), the trajectory of the target can be predicted more accurately, thereby optimizing the allocation of observation tasks and maximizing the utilization of satellite resources.

[0062] 3. In multi-target observation tasks, by calculating the joint gain observation value of related targets, the value of multi-target observation can be maximized, which is higher than the sum of the values ​​of observing each target separately, thus improving the overall benefit of the observation task.

[0063] 4. After a natural disaster occurs, relying on the SAR satellite's all-day and all-weather observation capabilities, satellite observation missions can be quickly and accurately planned to obtain detailed information on the disaster area in a timely manner, assist in rescue dispatch, improve rescue efficiency, and better protect the lives and property of the affected people. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 is a flow chart of an embodiment of the present invention;

[0065] Figure 2 A schematic diagram of calculating expected value for processing probability data of the present invention;

[0066] Figure 3 It is a schematic diagram of generating a task sequence based on a greedy algorithm according to the present invention. DETAILED DESCRIPTION

[0067] The specific implementation modes of the present invention are further described in detail below with reference to the accompanying drawings and examples.

[0068] A method for planning a geosynchronous orbit SAR satellite mission based on target trajectories is used to generate an observation mission sequence, wherein each observation mission observes all observation targets within a wave position coverage area through a wave position of a satellite radar; the method comprises the following steps:

[0069] Step 1: Obtain user observation requirements, including observation period, observation range and observation target, calculate each wave position of the satellite radar according to the observation range, and obtain the target category, target individual, historical trajectory, target observation value and target observation difficulty of each observation target; the target category is classified according to the minimum distinguishable type of the target; the target observation value includes the individual observation value of each category of targets, the joint gain observation value between different numbers of targets in each category, and the joint gain observation value between different numbers of targets in different categories. ,in is the individual observation value of each category of targets, the joint gain observation value between different numbers of targets in each category, and the joint gain observation value number between different numbers of targets in different categories, totaling kind.

[0070] Step 2: Set the upper limit of each observation task to the maximum exposure time required for all observation targets at each wave position. , set 0.5 seconds as the basic duration, and the total duration of the total observation task is the duration of the user's required observation period The observation time interval of each observation task is divided into time periods with the basic duration as the step length. part;

[0071] Step 3: Construct a trajectory probability prediction model based on gated recurrent units, train the model, and obtain a trained trajectory probability prediction model; the trajectory probability prediction model takes the historical trajectory data of the target as input, the predicted trajectory data of the target as intermediate output, and the distribution probability of the observed target in each time period and each wave position as the final output. The trajectory probability prediction model includes an input layer, a GRU hidden layer, an intermediate output layer, and a final output layer. The input layer of the trajectory probability prediction model receives input data, that is, a group of information vectors for multiple time steps. The activation function of the GRU hidden layer is the tanh function, and the cyclic Dropout ratio is between 0.2 and 0.5. The intermediate output layer generates an intermediate vector group, and the final output layer is a Dense layer with a Softmax activation function. The specific method for training the model is:

[0072] Obtain the historical trajectory of the target and map the target coordinates at each moment in the trajectory into wave position numbers;

[0073] The time value of each moment in the trajectory , target coordinates , Wave position number corresponding to the target coordinates , Target speed Merge into the information vector of the current target ;

[0074] For each moment in the trajectory, all information vectors at and before that moment are used as input data of a set of training data, and all information vectors after that moment are used as output data of the intermediate output layer of the set of training data, thereby obtaining multiple sets of training data;

[0075] The trajectory probability prediction model is trained using the above training data. The trajectory probability prediction model uses the Adam optimizer, and the loss function is the classification cross entropy loss function. When the loss no longer decreases within the preset number of times, the training is stopped to obtain the trained trajectory probability prediction model.

[0076] Step 4: Set the observation task number ;

[0077] Step 5: Arrange the historical trajectories of each observed target as input data, input them into the trained trajectory probability prediction model, and calculate the probability of each observed target in the first The distribution probability of each time period and each wave position during an observation mission ,in, is the wave position number, is the number of the time period in the observation time series, is a target individual in the target category X;

[0078] Step 6: For observation tasks, and calculate the average expected observation value of each wave position in different time periods; the specific method is:

[0079] Step 601: Generate target categories The observation probability matrix of the target individual in ;in For is a diagonal matrix with diagonal elements, is the observation success rate function, used to calculate the irradiation time at the standard wave position Next, the probability of finding a target in target category X;

[0080] Step 602: Establishing a historical value equilibrium function :

[0081]

[0082] in, is a coefficient less than 0, is a constant;

[0083] Step 603: Calculate the value correction matrix of the target individual observation probability:

[0084]

[0085] in, For the Target categories in observation tasks The observation probability diagonal matrix of , in particular, is a zero matrix;

[0086] Step 604: Generate Species containing A diagonal matrix with 1s and all other elements are 0 ;in, For the target category The total number of target individuals in, m=1,2,3...Y, is the number of the diagonal matrix;

[0087] Step 605: Calculation The expected function of the observation probability of a target individual is:

[0088] ,

[0089] in, is the unit diagonal matrix;

[0090] Step 606: Calculate the Among the observation tasks, Wave position and time period Next Target Observation Value The observation probability expectation function :

[0091] ;

[0092] Among them, x represents the number of the target category, ;

[0093] Step 607: Calculate the Among the observation tasks, The wave position in the time period The expected observation value function ;

[0094] Step 608: Calculate the Among the observation tasks, The wave position in the time period The average expected observation value function ;

[0095] Step 609: Generate The average expected observation value matrix of observation tasks , is a matrix with B rows and 2T columns, where B is the total number of wave positions and 2T is the total number of time periods in the observation time series. The element in row b and column t is .

[0096] Further, in step 601, the observation success rate function for:

[0097]

[0098] in, The observed target category is The target observation difficulty coefficient is For the The wave position in the time period The standard wave position exposure time is calculated as follows:

[0099]

[0100] in, for The mapping coefficient of the wave position to the equivalent irradiation of the standard wave position of the sub-satellite point.

[0101] Step 7: Based on the greedy algorithm, gradually select the wave position and task observation duration with the highest average expected observation value, determine each observation task one by one, output the observation task sequence, and complete the task planning; the specific method is:

[0102] Step 701: Select the average expected observation value matrix Maximum value Wave position and time period , generate candidate observation tasks, calculation ;

[0103] Step 702: If , execute step 710 to step 711, otherwise continue with step 703;

[0104] Step 703: If , then execute step 704 to step 705, otherwise, execute step 709 to step 711; To correct the threshold, >1;

[0105] Step 704: In this step, let the historical value equilibrium function To 0, recalculate the average expected observation value matrix , select The maximum value in ;

[0106] Step 705: If , then execute step 706 to step 711; otherwise, execute step 709 to step 711;

[0107] Step 706: Select middle Time period ,make , select middle Maximum value of the column ,make ; Select middle Time period ,make , select middle Maximum value of the column ,make ;

[0108] Step 707: When and When , then repeat step 706; otherwise, let , select middle Maximum value of the column Wave position and time period , generate candidate observation tasks, calculation ,make , select middle Maximum value of the column Wave position and time period , generate candidate observation tasks, calculation ;

[0109] Step 708: When or When middle Maximum value of the column Wave position and time period , generate candidate observation tasks, calculation , select middle Maximum value of the column Wave position and time period , generate candidate observation tasks, calculation ;

[0110] Step 709: Output candidate The observation task is An observation task, the wave position of which is , the total duration is the basic duration times;

[0111] Step 710: Calculate the remaining time , where the initial remaining time ;

[0112] Step 711: If , , repeat step 5 to step 7; otherwise, execute step 712 to step 713;

[0113] Step 712: Output candidate The observation task is An observation task, the wave position of which is , the total duration is the basic duration times;

[0114] Step 713: All observation tasks are organized into an observation task sequence to complete task planning.

[0115] Here is a more specific example:

[0116] A geosynchronous orbit SAR satellite mission planning method based on target trajectory, such as Figure 1-3 As shown, the specific steps are as follows:

[0117] Step 1: Obtain user observation requirements, including observation period, observation range and observation target, and calculate satellite radar wave position according to the observation range; obtain observation task target data list, the data list includes: target category is recorded as X, target individual is recorded as (X, Y), historical trajectory, target observation value, target observation difficulty; the target category is classified according to the minimum distinguishable type of the target; the target observation value includes the individual observation value of each category of targets, the joint gain observation value between different numbers of targets in each category, and the joint gain observation value between different numbers of targets in different categories ,in is the individual observation value of each category of targets, the joint gain observation value between different numbers of targets in each category, and the joint gain observation value number between different numbers of targets in different categories, totaling kind;

[0118] Step 2: record the mission observation sequence as a; record the satellite radar wave position as b, and the total number as B; set the upper limit of each observation mission to the maximum value of the exposure time of each wave position required for all observation targets , set 0.5 seconds as the basic duration, and the total duration of the total observation task is the duration of the user's required observation period The observation time interval of each observation task is divided into time periods with the basic duration as the step length. part;

[0119] Step 3: Establish coordinate-to-wavelength mapping function and establish observation time correction function for different wavelengths , establish the observation success rate time function The time correction function converts the irradiation time of different wave positions into the irradiation time of the standard wave position of the sub-satellite point, and the success rate time function is a function of the irradiation time and the probability of finding the target in the standard wave positions of different types of targets;

[0120] Step 4: Construct a trajectory probability prediction model based on the gated recurrent unit GRU. The model takes the historical trajectory data of the target as input, the predicted trajectory data of the target as the intermediate output, and the distribution probability of the observed target in each time period and each wave position as the final output; for example, set 64 hidden units, and gradually increase them according to demand; set the input sequence length to accept variable-length sequences; set the GRU hidden layer activation function to the tanh function to adapt to time series data; set the cyclic Dropout ratio between 0.2 and 0.5, and randomly disable 20% to 50% of neurons to prevent overfitting of the neural network. The ratio value can be adjusted according to the amount of data; the intermediate output layer generates an intermediate vector group and outputs a prediction information vector group; the final output layer is set to a Dense layer with a Softmax activation function to convert the trajectory prediction information vector group into a target position probability distribution; set the Adam optimizer to ensure model stability and adjust the learning rate according to demand; use the classification cross entropy loss function to make the probability distribution of the model output closer to the true probability; use the early stopping technique to stop training if it no longer decreases within several iterations;

[0121] Step 5: Obtain the historical trajectory of the target and perform preprocessing, and map the target coordinates at each moment in the trajectory to the wave position number; for example, obtain the real observation data set in the historical mission, map the target trajectory coordinates to the wave position number through the wave position function in step 3, and merge the wave position data and target information data at each moment into the target information vector at the current moment , scale the trajectory data to the range of [0, 1] or [−1, 1] to adapt to the model input, use the information vector group of a target at each moment and before that moment as the input data of a set of training data, and use the information vector group after that moment as the output data of the intermediate output layer of the set of training data, and obtain multiple sets of training data;

[0122] Step 6: trajectory probability prediction model training: Use the trajectory probability prediction model in step 4 to train based on the training data set in step 5. When the model meets the preset conditions, save the current model parameters and obtain the trained trajectory probability prediction model as the prediction model.

[0123] Step 7: Set the observation task number ;

[0124] Step 8: Input the mission target data and historical trajectory data into the trajectory probability prediction model respectively, and calculate the probability of each target in the The distribution probability of each wave position in each time period during an observation mission is recorded as ,in, is the wave position number, is the number of the time period in the observation time series, is a target individual in the target category X;

[0125] Step 9: For observation tasks, calculate the Wave position, Observational values ​​for different time periods The expected time function of the observed probability ; The steps are as follows:

[0126] Step 901: Generate The individual observation probability time matrix of the class ;in For is a diagonal matrix with diagonal elements; is the observation success rate function, used to calculate the irradiation time at the standard wave position The probability of finding a target in target category X is: The observed target category is The target observation difficulty coefficient is for The mapping coefficient of the wave position to the equivalent irradiation of the standard wave position of the sub-satellite point;

[0127] Step 902: Establishing a historical value equilibrium function :

[0128]

[0129] in, is a coefficient less than 0, is a constant;

[0130] Step 903: Calculate the value correction matrix of time probability:

[0131] ,

[0132] in, For the In the observation mission plan The class observation probability diagonal matrix, in particular, is a zero matrix;

[0133] Step 904: Generate Species containing A diagonal matrix with 1s and all other elements are 0 ;in, For the target category The total number of target individuals in, m=1,2,3...Y, is the number of the diagonal matrix;

[0134] Step 905: Calculation The expected time function of individual observation probability is:

[0135] ,

[0136] in, is the unit diagonal matrix;

[0137] Step 906: Calculate the Among the observation tasks, Wave position and time period Next Target Observation Value The observation probability expectation function :

[0138] ;

[0139] Among them, x represents the number of the target category, ;

[0140] Step 10: Calculate the The second observation, Expected value time function of the wave position

[0141] ;

[0142] Step 11: Calculate the The second observation, The wave position in the time period The expected average value of time function ;

[0143] Step 12: Generate the The average expected observation value matrix of observation tasks , is a matrix with B rows and 2T columns, where B is the total number of wave positions and 2T is the total number of time periods in the observation time series. The element in row b and column t is ;

[0144] Step 13: Based on the greedy algorithm, gradually select the wave position and duration with the highest average observation value to generate a task sequence; the steps are as follows:

[0145] Step 1301: In the In the observations, select the average expected observation value matrix Maximum value Wave position and time period , generate candidate observation tasks, calculation ;

[0146] Step 1302: When When , execute step 1310 to step 1311, otherwise continue with step 1303;

[0147] Step 1303: If , execute steps 1304 to 1305; otherwise, execute steps 1309 to 1311; wherein To correct the threshold, >1;

[0148] Step 1304: In this step, let the historical value equilibrium function To 0, recalculate the average expected observation value matrix , select The maximum value in ;

[0149] Step 1305: If , execute steps 1306 to 1311; otherwise, execute steps 1309 to 1311;

[0150] Step 1306: Select middle Time period ,make , select middle Maximum value of the column ,make ; Select middle Time period ,make , select middle Maximum value of the column ,make ;

[0151] Step 1307: When and When , then repeat step 1306; otherwise, let , select middle Maximum value of the column Wave position and time period , generate candidate observation tasks, calculation ,make , select middle Maximum value of the column Wave position and time period , generate candidate observation tasks, calculation

[0152] Step 1308: When or When middle Maximum value of the column Wave position and time period , generate candidate observation tasks, calculation , select middle Maximum value of the column Wave position and time period , generate candidate observation tasks, calculation ;

[0153] Step 1309: Output candidate The observation task is An observation task, the wave position of which is , the total duration is the basic duration times;

[0154] Step 1310: Calculate remaining time , where the initial remaining time ;

[0155] Step 1311: If , , repeat step 8 to step 13; otherwise, execute step 1312 to step 1313;

[0156] Step 1312: Output candidate The observation task is An observation task, the wave position of which is , the total duration is the basic duration times;

[0157] Step 1313: All observation tasks are organized into an observation task sequence to complete task planning.

[0158] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for planning a geosynchronous orbit SAR satellite mission based on target trajectories, for generating an observation mission sequence, wherein each observation mission observes all observation targets within a wave position coverage area through a wave position of a satellite radar; characterized in that: The following steps are involved: Step 1: Obtain the user's observation requirements, including observation period, observation range and observation target, calculate each wave position of the satellite radar according to the observation range, and obtain the target category, target individual, historical trajectory, target observation value and target observation difficulty of each observation target; Step 2: Set the upper limit of each observation task to the maximum exposure time required for all observation targets at each wave position. , set 0.5 seconds as the basic duration, and the total duration of the total observation task is the duration of the user's required observation period The observation time interval of each observation task is divided into time periods with the basic duration as the step length. part; Step 3: Construct a trajectory probability prediction model based on a gated recurrent unit, train the model, and obtain a trained trajectory probability prediction model; wherein the trajectory probability prediction model takes the historical trajectory data of the target as input, the predicted trajectory data of the target as intermediate output, and the distribution probability of the observed target in each time period and each wave position as the final output. The trajectory probability prediction model includes an input layer, a GRU hidden layer, an intermediate output layer, and a final output layer. The input layer of the trajectory probability prediction model receives input data, that is, an information vector group of multiple time steps. The activation function of the GRU hidden layer is the tanh function, and the cyclic Dropout ratio is between 0.2 and 0.

5. The intermediate output layer generates an intermediate vector group, and the final output layer is a Dense layer with a Softmax activation function. Step 4: Set the observation task number ; Step 5: Arrange the historical trajectories of each observed target as input data, input them into the trained trajectory probability prediction model, and calculate the probability of each observed target in the first The distribution probability of each time period and each wave position during an observation mission ,in, is the wave position number, is the number of the time period in the observation time series, is a target individual in the target category X; Step 6: For Observation tasks, calculate the average expected observation value of each wave position in different time periods; Step 7: Based on the greedy algorithm, gradually select the wave position and task observation duration with the highest average expected observation value, determine each observation task one by one, output the observation task sequence, and complete the task planning.

2. A method for geosynchronous orbit SAR satellite mission planning based on target trajectory according to claim 1, characterized in that: In step 1, the target categories are classified according to the minimum distinguishable categories of the targets; the target observation values ​​include the individual observation values ​​of the targets of each category, the joint gain observation values ​​between different numbers of targets of each category, and the joint gain observation values ​​between different numbers of targets of different categories. ,in is the individual observation value of each category of targets, the joint gain observation value between different numbers of targets in each category, and the joint gain observation value number between different numbers of targets in different categories, totaling kind.

3. A method for geosynchronous orbit SAR satellite mission planning based on target trajectory according to claim 2, characterized in that: In step 3, the specific method of training the model is: Obtain the historical trajectory of the target and map the target coordinates at each moment in the trajectory into wave position numbers; The time value of each moment in the trajectory , target coordinates , the wave position number corresponding to the target coordinates , Target speed Merge into the information vector of the current target ; For each moment in the trajectory, all information vectors at and before that moment are used as input data of a set of training data, and all information vectors after that moment are used as output data of the intermediate output layer of the set of training data, thereby obtaining multiple sets of training data; The trajectory probability prediction model is trained using the above training data. The trajectory probability prediction model uses the Adam optimizer, and the loss function is the classification cross entropy loss function. When the loss no longer decreases within the preset number of times, the training is stopped to obtain the trained trajectory probability prediction model.

4. The method for geosynchronous orbit SAR satellite mission planning based on target trajectory according to claim 3, characterized in that: The specific method of step 6 is: Step 601: Generate target categories The observation probability matrix of the target individual in ;in For is a diagonal matrix with diagonal elements, is the observation success rate function, used to calculate the irradiation time at the standard wave position Next, in The probability of finding a target in target category X at the wave position; Step 602: Establishing a historical value equilibrium function : ； in, is a coefficient less than 0, is a constant; Step 603: Calculate the value correction matrix of the target individual observation probability: ； in, For the Target categories in observation tasks The observation probability diagonal matrix of , in particular, is a zero matrix; Step 604: Generate Species containing A diagonal matrix with 1s and all other elements are 0 ;in, For the target category The total number of target individuals in, m=1,2,3...Y, is the number of the diagonal matrix; Step 605: Calculation The expected function of the observation probability of a target individual is: ， in, is the unit diagonal matrix; Step 606: Calculate the Among the observation tasks, Wave position and time period Next Target Observation Value The observation probability expectation function : ； Among them, x represents the number of the target category, ; Step 607: Calculate the Among the observation tasks, The wave position in the time period The expected observation value function ; Step 608: Calculate the Among the observation tasks, The wave position in the time period The average expected observation value function ; Step 609: Generate The average expected observation value matrix of observation tasks , is a matrix with B rows and 2T columns, where B is the total number of wave positions and 2T is the total number of time periods in the observation time series. The element in row b and column t is .

5. The method for geosynchronous orbit SAR satellite mission planning based on target trajectory according to claim 4, characterized in that: In step 601, observe the success rate function for: ； in, The observed target category is The target observation difficulty coefficient is For the The wave position in the time period The standard wave position exposure time is calculated as follows: ； in, for The mapping coefficient of the wave position to the equivalent irradiation of the standard wave position of the sub-satellite point.

6. A method for geosynchronous orbit SAR satellite mission planning based on target trajectory according to claim 5, characterized in that: The specific method of step 7 is: Step 701: Select the average expected observation value matrix Maximum value Wave position and time period , generate candidate observation tasks, calculation ; Step 702: If , then execute step 710 to step 711, otherwise continue with step 703; Step 703: If , then execute step 704 to step 705, otherwise, execute step 709 to step 711; To correct the threshold, >1; Step 704: Only in step 704, let the historical value equilibrium function is 0; recalculate the average expected observation value matrix , select The maximum value in ; Step 705: If , then execute step 706 to step 711; Otherwise, execute step 709 to step 711; Step 706: Select middle Time period ,make , select middle Maximum value of the column ,make ; Select middle Time period ,make , select middle Maximum value of the column ,make ; Step 707: When and When , then repeat step 706; otherwise, let , select middle Maximum value of the column Wave position and time period , generate candidate observation tasks, calculation ,make , select middle Maximum value of the column Wave position and time period , generate candidate observation tasks, calculation ; Step 708: When or When middle Maximum value of the column Wave position and time period , generate candidate observation tasks, calculation , select middle Maximum value of the column Wave position and time period , generate candidate observation tasks, calculation ; Step 709: Output candidate The observation task is An observation task, the wave position of which is , the total duration is the basic duration times; Step 710: Calculate the remaining time , where the initial remaining time ; Step 711: If , , repeat steps 5 to 7; Otherwise, execute step 712 to step 713; Step 712: Output candidate The observation task is An observation task, the wave position of which is , the total duration is the basic duration times; Step 713: All observation tasks are organized into an observation task sequence to complete task planning.

Images