A time-discrete phased array radar beam gain scheduling method

Abstract

The application discloses a time-discrete phased array radar beam gain scheduling method, and belongs to the field of radar resource scheduling. First, a scheduling benefit function of all tasks in a scheduling time slot is established; each task is decomposed into a plurality of subtasks executed at different time slice start moments, and the scheduling benefit of each subtask is calculated; then, according to the principle that the execution time of each task does not cross each other and each subtask decomposed from the same task is mutually exclusive, a mutual exclusion matrix is established, and the scheduling problem is converted into a minimum weighted independent set problem in graph theory; finally, a continuous Hopfield neural network is used to solve the problem, and a scheduling result is obtained. The application is mainly used for reasonably allocating beam residence resources for different tasks of a phased array radar under the condition that time resources are limited, can consider the urgency and timeliness of all tasks in the scheduling time slot, globally optimizes the beam residence sequence, is convenient for hardware parallel calculation, and can obtain good scheduling benefit with a small time cost.

Description

Technical Field

[0001] This invention belongs to the field of radar resource scheduling, and in particular relates to a method for scheduling beam gain of time-discrete phased array radar. Background Technology

[0002] In the modern information-based battlefield environment, radar systems face increasingly complex usage scenarios. Phased array radars, through electronic scanning and flexible beam scheduling, have achieved a breakthrough in multi-functional integrated combat capabilities, enabling them to simultaneously perform multiple tasks such as search, tracking, and guidance. However, this also brings the challenge of rationally allocating radar beam dwell resources among multiple tasks.

[0003] Ting C et al. proposed a scheduling method based on scheduling gain. By constructing a multivariate scheduling benefit function, it can take into account importance, urgency and timeliness to a certain extent. However, its scheduling algorithm is locally greedy and does not comprehensively consider the global scheduling effect.

[0004] Yang Shanchao et al. proposed a phased array radar task scheduling algorithm based on value optimization. They constructed schedulers for tracking and search tasks respectively, and considered factors such as task timeliness. Compared with traditional methods, it can improve scheduling performance to a certain extent. However, this scheduling algorithm still does not perform scheduling from a global perspective of the entire scheduling time slot, and cannot determine the mathematical superiority of the scheduling result.

[0005] Hao L et al. proposed a task scheduling algorithm based on genetic algorithm, which can perform global optimization of scheduling strategy. However, the algorithm requires repeated genetic iteration, which is complex to implement, has a high computational cost, and it is difficult to mathematically guarantee its optimization effect. It often fails to converge to a good local optimum. Summary of the Invention

[0006] To address the problem mentioned in the background that existing scheduling algorithms struggle to balance global scheduling effectiveness and computational cost, this invention proposes a time-discrete phased array radar beam gain scheduling method. During task scheduling, this invention discretizes the scheduling time slots into several time slices, enabling the method to exhaustively enumerate all actual execution times of a task. This transforms the scheduling problem into solving a maximum weighted independent set problem. Introducing a continuous Hopfield neural network allows the problem to quickly converge to a local optimum, and the entire method is easily accelerated using hardware parallelization. The specific details of this invention are as follows:

[0007] A method for beam gain scheduling in a time-discrete phased array radar includes the following steps:

[0008] Step 1: Discretly construct a set of scheduling subtasks; for all tasks to be scheduled, decompose each task into M subtasks based on time slices, with each subtask having the same dwell time;

[0009] Step 2: Establish a scheduling benefit vector; Based on the principles of importance, urgency, and timeliness of phased array radar beam stationing scheduling, establish a scheduling benefit function for each subtask, calculate the scheduling benefits of all subtasks, and combine them to form a scheduling benefit vector.

[0010] Step 3: Determine the start and end times of each subtask; calculate the termination time of each subtask based on its actual execution time and dwell time, and construct the actual execution time vector and the termination execution time vector;

[0011] Step 4: Construct a mutual exclusion matrix to show the mutual exclusion relationship between subtasks based on the time relationship between each subtask in the actual execution time vector and the termination execution time vector;

[0012] Step 5: Construct a minimum weighted independent set optimization problem based on the scheduling reward vector and the execution vector of the scheduled subtasks;

[0013] Step 6: Construct a Hopfield neural network to iteratively optimize the execution vectors of the scheduled subtasks and obtain the final execution scheme for the scheduled subtasks.

[0014] Furthermore, for all tasks to be scheduled, the time slices spanned by each task's time window are calculated, thereby decomposing each task to be scheduled into M subtasks. Each subtask is actually executed at the start of a different time slice, but has the same dwell time. For tasks whose time window ends later than the end of the scheduling time slot, an additional delayed task can be decomposed.

[0015] The actual execution time of each subtask satisfies the condition that there is only one subtask at the current time, and its actual execution time is within the time slice spanned by each task's time window. The time slice spanned by each task's time window is within the ± time window range of the task's expected execution time. The undiscrete dwell time of task i is divided by the time slice width to obtain the discrete dwell time of task i.

[0016] Furthermore, the scheduling benefit function for each subtask includes: an importance benefit function, calculated based on task priority; an urgency benefit function, calculated by the ratio of the latest execution time of the task to the end time of the scheduling slot; and a timeliness benefit function, calculated based on the difference between the actual and expected execution time of the subtask.

[0017] Furthermore, the minimum weighted independent set optimization problem is expressed as follows:

[0018] ;

[0019] in, Let M be the scheduling reward vector composed of each scheduling subtask, where M is the total number of subtasks. It is a mutually exclusive matrix; This is the execution vector for scheduled subtasks, indicating whether each scheduled subtask is executed. Execution, then its first Each component Conversely, it is 0.

[0020] Furthermore, step 6 is detailed as follows:

[0021] Build with A Hopfield neural network with 100 neurons, the input-output relationship of each neuron is as follows:

[0022] ;

[0023] in, For the first The state vector of each neuron This represents the time derivative of the state vector of the m-th neuron. To preset parameters, It is a non-linear Sigmoid function;

[0024] Then, construct the energy function E:

[0025] ;

[0026] This yields the execution vector of the scheduled subtasks. Update strategy:

[0027] ;

[0028] in, For the first The state vector of each iteration For the first The execution vector of the scheduled subtasks in step-by-step iteration. for The vector formed Represents the Hadamard product; This represents the product of the derivative of the energy function with respect to the execution vector of the scheduled subtasks in the k-th iteration and the time iteration step size;

[0029] Perform iterative optimization, and output the binarized execution vector of the scheduled subtask after the iteration stops.

[0030] Furthermore, the stopping iteration criterion is as follows: when the absolute value of the change in energy function between two adjacent iteration steps is less than a minimum threshold or the number of iterations is greater than... When the time is right, stop iterative optimization.

[0031] Furthermore, each component of the scheduling benefit vector corresponds one-to-one with a subtask, storing the scheduling benefit of each subtask.

[0032] Calculate the termination execution time of each subtask based on its actual execution time and dwell time. If the subtask is a delayed task, the termination execution time is the same as the actual execution time; for normal subtasks, the termination execution time = actual execution time + dwell time. Then, construct the actual execution time vector and the termination execution time vector according to the subtask arrangement order of the scheduling reward vector.

[0033] Furthermore, the mutual exclusion matrix is ​​established in the following manner:

[0034] Based on the total number of subtasks M, establish Mutual exclusion matrix ,in Subtasks sub-tasks Whether there is a conflict; if there is a conflict, the value is 1, otherwise it is 0.

[0035] for Whether a task is in conflict is determined according to the following rules: if the time windows of two subtasks overlap, they are considered to be in conflict; any two different subtasks derived from the same task are considered to be in conflict; other than the above two cases, they are considered not to be in conflict. In this situation, This means that each subtask does not conflict with itself. This results in a symmetric 0-1 mutual exclusion matrix with zeros on the diagonal. .

[0036] The beneficial effects of this invention are as follows:

[0037] This invention discretizes the phased array radar task scheduling time slots into several time slices and decomposes all tasks to be scheduled into several sub-tasks, using 0-1 vectors to represent the execution status of each sub-task. By pre-calculating the scheduling benefits of each sub-task and establishing a scheduling benefit vector, the total scheduling benefit is successfully simplified to the form of a vector inner product. At the same time, a conflict matrix between each sub-task is constructed, successfully describing the time constraints between each task to be scheduled in a quadratic form within the discretized scheduling time slots. This transforms the complex scheduling optimization problem into a maximum weighted independent set problem in graph theory, greatly reducing the complexity of the phased array radar task scheduling model. A continuous Hopfield neural network is introduced to solve the simplified maximum weighted independent set problem, enabling this invention to achieve better phased array radar scheduling benefits at a lower time cost compared to traditional genetic algorithms. Attached Figure Description

[0038] Figure 1This is a flowchart of the method of the present invention.

[0039] Figure 2 This is a schematic diagram of discrete decomposition scheduling tasks.

[0040] Figure 3 This is a sequence diagram of tasks to be scheduled.

[0041] Figure 4 This is a sequence diagram of the scheduling results. Detailed Implementation

[0042] This invention employs simulation experiments to demonstrate the effectiveness of the proposed method. All steps and conclusions of this invention have been verified on the Matlab 2022b simulation platform. To enable those skilled in the art to understand the invention, it is further described below with reference to the accompanying drawings.

[0043] A time-discrete phased array radar beam gain scheduling method, such as Figure 1 As shown, it includes the following steps:

[0044] Step 1: Discretly construct the set of scheduling subtasks;

[0045] In this invention, a randomly generated task sequence is used in the simulation environment to replace all the tasks to be scheduled in a scheduling time slot in the actual situation. The parameters of the scheduling task generator are shown in Table 1. Among them, for task priority, expected execution time, dwell time and time window, the generator randomly generates task parameters with uniform or equal probability distribution within the set range.

[0046] Table 1: Simulation Parameters of the Scheduling Task Generator

[0047] Parameter name numerical value or range <![CDATA[Task priority p i > {1,2,3,4} Scheduling time slot width SI 100ms Number of time slices 100 <![CDATA[Residence time w i_cts > (0ms, 10ms) <![CDATA[Task expected execution time dt i > <![CDATA[[0,SI-w i_cts ]]]> <![CDATA[Time window w i > (0.5ms, 10ms) Total number of tasks N 50

[0048] like Figure 2 As shown, the entire scheduling time slot is divided into several scheduling time slices at equal intervals. For all tasks to be scheduled, the time slice spanned by each task's time window is calculated. In this way, each task to be scheduled is decomposed into several sub-tasks. The actual execution time of each sub-task is located at the beginning of a different time slice, but they have the same dwell time. For tasks whose time window ends later than the end of the scheduling time slot, an additional delayed task can be decomposed.

[0049] Each subtask is executed at a time when there is only one subtask, and its actual execution time is within the time slice spanned by each task's time window. The time slice spanned by each task's time window is within the range of ± the expected execution time of the task, which can be expressed by the following formula:

[0050]

[0051] and

[0052] in, For subtasks The actual execution time; simultaneously, the undiscrete dwell time of task i is divided by the time slice width to obtain the discrete dwell time of task i, as expressed by the following formula:

[0053]

[0054] in, For the task The dwell time after discretization For the task Undispersed dwell time, This represents the width of the time slice.

[0055] Step 2: Establish the scheduling benefit vector;

[0056] Based on the principles of importance, urgency, and timeliness in phased array radar beam dwell scheduling, a scheduling benefit function is established for each subtask, including an importance benefit function calculated based on task priority; an urgency benefit function calculated by the ratio of the latest execution time of the task to the end time of the scheduling slot; and a timeliness benefit function calculated based on the difference between the actual and expected execution time of the subtask. The specific formula is as follows:

[0057]

[0058]

[0059] in For subtasks The scheduling benefit function, It's the task number. , Number the subtasks, subtasks , This represents the total number of tasks within the scheduled time slot.

[0060] For importance payoff function, Prioritize this task. For the pre-set state The attenuation factor for the interval, taken here. When a task is successfully executed within this scheduling slot, it will receive [benefits / rewards]. The scheduling benefit, if delayed to the next time slot, is multiplied by a decay factor. ;

[0061] For the urgency payoff function, The expected execution time for this task. This is the time window for the task. For the scheduling time slot length, For the pre-defined positive real number parameter, here we take... ;

[0062] For the timeliness revenue function, For subtasks The actual execution time For the pre-defined positive real number parameter, here we take... .

[0063] After calculating the scheduling benefits of all subtasks, they are combined to form a scheduling benefit vector. The vector dimension M is the total number of subtasks; each component in the scheduling reward vector corresponds one-to-one with a subtask, storing the scheduling reward of each subtask.

[0064] Step 3: Determine the start and end times of each subtask;

[0065] First, based on the actual execution time of each subtask and length of stay Calculate its termination execution time. If the subtask is a delayed task, the termination execution time is the same as the actual execution time; for general subtasks, the termination execution time = actual execution time + dwell time. Then, following step 2, calculate the termination execution time according to the scheduling benefit vector. The order of subtasks is used to construct the actual execution time vector. and termination execution time vector .

[0066] Step 4: Construct a mutual exclusion matrix;

[0067] Based on the total number of subtasks ,Establish Mutual exclusion matrix ,in Subtasks sub-tasks Whether there is a conflict; if there is a conflict, the value is 1; otherwise, it is 0. Whether a task conflicts with another task is considered a conflict is determined according to the following rules: 1. If the time windows of two subtasks overlap, they are considered to be in conflict; 2. Any two different subtasks derived from the same task conflict with each other; 3. Other cases are considered not to conflict. It is worth noting that for… In this situation, This means that each subtask does not conflict with itself. From this, we can obtain a symmetric 0-1 mutually exclusive matrix with 0s on the diagonal. .

[0068] Step 5: Construct the minimum weighted independent set optimization problem;

[0069] Based on the results of the previous steps, we construct the following minimum weighted independent set optimization problem:

[0070]

[0071] in, This is the scheduling reward vector composed of each scheduling subtask. Let the vector represent whether each scheduled subtask is executed, if the subtask Execution, then its first Each component Conversely, it is 0.

[0072] Step 6: Iterative optimization of the Hopfield neural network:

[0073] Build with A Hopfield neural network with 100 neurons, the input-output relationship of each neuron is as follows:

[0074]

[0075] in, For the first The state vector of each neuron This represents the time derivative of the state vector of the m-th neuron. To preset parameters, The sigmoid function is a non-linear function, and its expression is:

[0076]

[0077] in, The parameters are pre-defined, and x is a referential variable.

[0078] Then, construct the energy function E:

[0079]

[0080] This yields the execution vector of the scheduled subtasks. Update strategy:

[0081]

[0082] in, For the first The state vector of each iteration For the first The execution vector of the scheduled subtasks in step-by-step iteration. for The resulting vector, in this example, is set to a vector of all 1s. Represents the Hadamard product; This represents the product of the derivative of the energy function with respect to the execution vector of the scheduled subtasks in the k-th iteration and the time iteration step size; in this example, it applies to all... The system is randomly initialized using a normal distribution with a mean of 0 and a variance of 1, and the iteration step size is... .

[0083] Set a stopping criterion: when the absolute value of the change in the energy function between two adjacent iteration steps is less than a minimum threshold. or the number of iterations is greater than Stop iterative optimization when the time comes. In this example... , After the iteration is complete, the vector will be executed. Binarization involves setting components greater than or equal to 0.5 to 1, and vice versa.

[0084] To demonstrate the effectiveness of this invention, simulations were performed on the Matlab 2022b simulation platform.

[0085] Figure 3 It is a sequence of tasks randomly generated by the task generator, where each bar represents a task, the left edge of the bar represents the expected execution time of the task, and the width of the bar represents the required dwell time of the task. Since a large number of tasks overlap and intersect with each other, the scheduler needs to adjust the execution time, discard or delay some tasks.

[0086] Figure 4 (a) in the table shows the scheduling result of the genetic algorithm, with a total scheduling benefit of 80.17, and 21 tasks were successfully executed in this time slot. Figure 4 (b) in the figure represents the scheduling result of this invention. The total scheduling benefit is 97.41, and 28 tasks were successfully executed within this time slot. The scheduling benefit and the success rate of execution are higher.

[0087] Those skilled in the art can make relevant applications based on the time-discrete phased array radar beam dwell scheduling method disclosed in this invention, and the relevant knowledge is still within the protection scope of this invention.

[0088] It is understood that the present invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1. A method for beam gain scheduling in a time-discrete phased array radar, characterized in that, Includes the following steps: Step 1: Discretly construct the set of scheduling subtasks; For all tasks to be scheduled, each task is decomposed into M subtasks based on a time slice, and each subtask has the same dwell time. Step 2: Establish a scheduling benefit vector; Based on the principles of importance, urgency and timeliness of phased array radar beam stationing scheduling, establish a scheduling benefit function for each subtask, calculate the scheduling benefits of all subtasks, and combine them to form a scheduling benefit vector. The scheduling benefit function for each subtask includes: an importance benefit function, calculated based on task priority; an urgency benefit function, calculated by the ratio of the latest execution time of the task to the end time of the scheduling slot; and a timeliness benefit function, calculated based on the difference between the actual and expected execution time of the subtask. Step 3: Determine the start and end times of each subtask; calculate the termination time of each subtask based on its actual execution time and dwell time, and construct the actual execution time vector and the termination execution time vector; Step 4: Construct a mutual exclusion matrix to show the mutual exclusion relationship between subtasks based on the time relationship between each subtask in the actual execution time vector and the termination execution time vector; Step 5: Construct a minimum weighted independent set optimization problem based on the scheduling reward vector and the scheduling subtask execution vector; the minimum weighted independent set optimization problem is expressed as follows: ； in, Let M be the scheduling reward vector composed of each scheduling subtask, where M is the total number of subtasks. It is a mutually exclusive matrix; This is the execution vector for scheduled subtasks, indicating whether each scheduled subtask is executed. Execution, then its first Each component Conversely, it is 0; Step 6: Construct a Hopfield neural network to iteratively optimize the execution vectors of the scheduled subtasks, obtaining the final execution scheme for the scheduled subtasks; Step 6 is detailed below: Build with A Hopfield neural network with 100 neurons, the input-output relationship of each neuron is as follows: ； in, For the first The state vector of each neuron This represents the derivative of the state vector of the m-th neuron with respect to time. To preset parameters, It is a non-linear Sigmoid function; Then, construct the energy function E: ； This yields the execution vector of the scheduled subtasks. Update strategy: ； in, For the first The state vector of each iteration For the first The execution vector of the scheduled subtasks in step-by-step iteration. for The vector formed Represents the Hadamard product; This represents the product of the derivative of the energy function with respect to the execution vector of the scheduled subtasks in the k-th iteration and the time iteration step size; Perform iterative optimization, and output the binarized execution vector of the scheduled subtasks after satisfying the stopping iteration criterion; When the absolute value of the change in energy function between two adjacent iteration steps is less than the minimum threshold or the number of iterations is greater than... Stop iterative optimization when the time is right.

2. The time-discrete phased array radar beam gain scheduling method according to claim 1, characterized in that, Step 1 is described in detail as follows: For all tasks to be scheduled, calculate the time slices spanned by each task's time window, thereby decomposing each task to be scheduled into M subtasks. Each subtask is actually executed at the start of a different time slice, but has the same dwell time. For tasks whose time window ends later than the end of the scheduling time slot, an additional delayed task is decomposed. The actual execution time of each subtask satisfies the condition that there is only one subtask at the current time, and its actual execution time is within the time slice spanned by each task's time window. The time slice spanned by each task's time window is within the ± time window range of the task's expected execution time. The undiscrete dwell time of task i is divided by the time slice width to obtain the discrete dwell time of task i.

3. The time-discrete phased array radar beam gain scheduling method according to claim 2, characterized in that, The scheduling benefit vector corresponds one-to-one with each subtask, storing the scheduling benefit of each subtask. The termination execution time is calculated based on the actual execution time and dwell time of each subtask. If the subtask is a delayed task, the termination execution time is the same as the actual execution time. For normal subtasks, the termination execution time = actual execution time + dwell time. Then, according to the order of subtasks in the scheduling benefit vector, construct the actual execution time vector and the termination execution time vector.

4. The time-discrete phased array radar beam gain scheduling method according to claim 3, characterized in that, The mutual exclusion matrix is ​​established in the following way: Based on the total number of subtasks M, establish Mutual exclusion matrix ,in Subtasks sub-tasks Whether there is a conflict; if there is a conflict, the value is 1, otherwise it is 0. for Whether a task is in conflict is determined according to the following rules: if the time windows of two subtasks overlap, they are considered to be in conflict; any two different subtasks derived from the same task are considered to be in conflict; other than the above two cases, they are considered not to be in conflict. In this situation, This means that each subtask does not conflict with itself; thus, a symmetric 0-1 mutual exclusion matrix with 0s on the diagonal is obtained. .

Images