Long-time scheduling method and system for movable active and passive sensor in clutter environment

Abstract

This invention provides a long-term scheduling method and system for mobile active and passive sensors in cluttered environments. The method includes: determining a sensor system scheduling model; establishing a multi-target state model and a multi-target measurement model based on POMDP and RFS theories; calculating the estimated states of multiple targets using the GM-PHD smoothing filtering algorithm, and using OSPA distance as a metric to predict the long-term tracking accuracy of each sensor for multiple targets; predicting the long-term radiation cost of sensors based on a radiation risk quantification method using signal power density; creating an optimization function; and determining a sensor scheduling scheme based on the predicted long-term tracking accuracy of multiple targets and the predicted long-term radiation cost of sensors, combined with the constraints of the minimum operating time of the sensors and the set tracking task duration. This invention can effectively control radiation costs and has significant advantages compared to other methods.

Description

Technical Field

[0001] This embodiment relates to the field of sensor scheduling technology, and in particular to a long-term scheduling method and system for mobile active and passive sensors in cluttered environments. Background Technology

[0002] In modern warfare, constrained by the complex electromagnetic environment, the effectiveness of various target detection sensors heavily relies on resource allocation and information fusion. Sensor scheduling technology has become a research hotspot in target tracking. Due to the characteristics of passive sensors—non-radiating signals and high angle measurement accuracy—and the risk avoidance advantages of mobile platforms, this research proposes a scheduling method suitable for mobile active and passive sensor systems. This method can effectively improve target tracking accuracy, reduce usage costs, and enhance the survivability of sensor systems.

[0003] Existing active sensors, such as radar, acquire information about a target's spatial position and motion by radiating energy signals such as electromagnetic waves. However, the application of radiation warning receivers can help targets intercept radiation signals, identify the sensor, and pinpoint its location, thereby enabling attacks. Therefore, active sensors carry a certain radiation risk when tracking targets, and the cost of using them due to this risk is called radiation cost.

[0004] In sensor scheduling for target tracking, the prediction of optimization metrics and the estimation of target states both rely on the application of target tracking algorithms. However, in real-world target tracking environments, the large number of targets makes complete observation difficult, and environmental clutter introduces significant uncertainty. To address this, classic multi-target tracking algorithms such as Joint Probabilistic Data Association (JPDA) and Multiple Hypothesis Tracking (MHT) use data association to solve the uncertainty problem, turning the scheduling solution into a nondeterministic polynomial problem with high computational cost and slow response time, failing to meet the timeliness requirements of target tracking. To address this issue, existing technologies propose a method for mobile single-sensor control problems that uses the Optimal Subpattern Assignment (OSA) distance as a metric for tracking accuracy and employs a Labeled Multi-Bernoulli (LMB) filter to estimate the target state, significantly improving tracking accuracy. For the optimization of the single indicator of tracking accuracy, control schemes were obtained by using short-time scheduling. However, when optimizing for radiation risk, due to the contradictory relationship between tracking accuracy and radiation risk, simply performing optimal scheduling will lead to a serious reduction in tracking accuracy. Therefore, the scheduling must take the effective balance between the two as the optimization goal.

[0005] However, short-term scheduling is based solely on single-step gains. While its performance in optimizing a single metric is outstanding, it cannot adequately meet the requirements of balancing multiple metrics. In contrast, long-term scheduling is based on the sum of scheduling gains over a period of time, and its performance in balancing optimization is superior to that of short-term scheduling.

[0006] In view of this, there is an urgent need to provide a long-term scheduling method for mobile active and passive sensors in clutter environments that can effectively control radiation costs in response to the above problems. Summary of the Invention

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is to provide a long-term scheduling method for mobile active and passive sensors in clutter environments, comprising the following steps:

[0008] Determine the sensor system scheduling model;

[0009] For multi-sensor multi-target tracking applications in cluttered environments, a multi-target state model and a multi-target measurement model are established based on POMDP and RFS theories.

[0010] The GM-PHD smoothing filtering algorithm is used to calculate the estimated state of multiple targets, and the OSPA distance is used as a metric to predict the long-term tracking accuracy of each sensor for multiple targets; the radiation risk quantification method based on signal power density is used to predict the long-term radiation cost of the sensors.

[0011] An optimization function is created with the objective of achieving an effective balance between target tracking accuracy and sensor radiation cost.

[0012] Based on the predicted values ​​of multi-target long-term tracking accuracy and sensor long-term radiation cost, and combined with the constraints of the minimum working time of the sensor and the set tracking task duration, the sensor scheduling scheme is determined.

[0013] The sensor system includes multiple mobile platforms, each equipped with at least one active sensor and one passive sensor.

[0014] This invention also provides a long-term scheduling system for mobile active and passive sensors in clutter environments, comprising:

[0015] A sensor system with multiple mobile platforms, each platform having at least one active sensor and one passive sensor;

[0016] Sensor system scheduling model establishment unit: used to determine the sensor system scheduling model.

[0017] Multi-objective state model and measurement model establishment unit: used to establish multi-objective state models and multi-objective measurement models based on POMDP theory and RFS theory;

[0018] Target optimization function creation unit: Creates an optimization function with the objective of achieving an effective balance between target tracking accuracy and sensor radiation cost.

[0019] Multi-target long-term tracking accuracy prediction unit: The GM-PHD smoothing filter algorithm is used to calculate the estimated state of multiple targets, and the OSPA distance is used as the metric to predict the long-term tracking accuracy of multiple targets by each sensor.

[0020] Long-term radiation cost prediction unit: The radiation risk quantification method based on signal power density is used to predict the long-term radiation cost value of the sensor.

[0021] Sensor scheduling scheme solution and execution unit: Based on the prediction values ​​of multi-target long-term tracking accuracy and sensor long-term radiation cost, and combined with the constraints of the minimum working time of the sensor and the set tracking task duration, the sensor scheduling scheme is determined.

[0022] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements the long-term scheduling method for mobile active and passive sensors in a clutter environment as described above.

[0023] The present invention also provides a computer-readable storage medium storing a computer program, characterized in that, when the computer program is executed by a processor, it implements the long-term scheduling method for mobile active and passive sensors in a clutter environment as described above.

[0024] This invention, based on a constructed sensor scheduling model, mathematically describes the motion state and measurement results of multiple targets, as well as sensor scheduling actions. Simultaneously, based on the working principle of radar and the concept of interception probability, an improved radiation risk quantification method is proposed. Subsequently, a Gaussian mixture probability hypothesis density filtering algorithm is used to predict long-term tracking accuracy, the proposed improved quantification method is used to predict long-term radiation cost, and an improved gray wolf optimization algorithm is used to solve the sensor scheduling scheme. Finally, the scheduling scheme is executed to obtain multi-target measurement information, and a joint generalized label multi-Bernoulli filtering algorithm is used to calculate the estimated target state. This scheduling method, while ensuring tracking accuracy, can effectively control radiation cost, exhibiting significant advantages compared to other methods. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A schematic diagram of the long-term scheduling method for mobile active and passive sensors in a clutter environment provided by the present invention;

[0027] Figure 2 This is a schematic diagram of the operation of the active and passive sensor system based on a mobile platform in the method provided by the present invention;

[0028] Figure 3 The method provided by the present invention N R =2 、N θ Schematic diagram of the sensor platform maneuver scheme when =8;

[0029] Figure 4 This invention provides a flowchart of the long-term scheduling process in the method.

[0030] Figure 5 The simulation experiment provided by this invention shows the actual motion trajectories of multiple targets and the observed trajectory diagrams obtained by scheduling using NMAPS.

[0031] Figure 6 This is a sensor scheduling sequence diagram in the simulation experiment provided by the present invention;

[0032] Figure 7 Comparison of OSPA and instantaneous OSPA distance obtained by different scheduling methods in the simulation experiment provided by this invention when performing tracking accuracy optimization scheduling;

[0033] Figure 8 A comparison chart of CRC and instantaneous radiation cost obtained when different scheduling methods are used to optimize tracking accuracy in the simulation experiment provided by this invention;

[0034] Figure 9 A schematic diagram of the structure of the long-term scheduling system for mobile active and passive sensors in a clutter environment provided by the present invention;

[0035] Figure 10 A schematic block diagram of a computer device provided for this invention. Detailed Implementation

[0036] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this specification.

[0037] The present invention will now be described in detail with reference to specific embodiments and accompanying drawings.

[0038] Method Implementation Examples

[0039] According to embodiments of the present invention, a long-term scheduling method for mobile active and passive sensors in clutter environments is provided, such as... Figure 1 The diagram shows a flowchart of a long-term scheduling method for mobile active and passive sensors in a clutter environment according to an embodiment of the present invention. The method includes the following steps:

[0040] Step 1: Determine the sensor system and its scheduling model;

[0041] Step 2: Establish a multi-objective state model and a multi-objective measurement model based on POMDP (Partially Observable Markov Decision Process) theory and RFS (Random Finite Set) theory;

[0042] Step 3: The GM-PHD smoothing filtering algorithm is used to calculate the estimated state of multiple targets, and the OSPA distance is used as the metric to predict the long-term tracking accuracy of each sensor for multiple targets; the radiation risk quantification method based on signal power density is used to predict the long-term radiation cost of the sensors.

[0043] Step 4: Create an optimization function with the goal of achieving an effective balance between target tracking accuracy and sensor radiation cost;

[0044] Step 5: Based on the predicted values ​​of multi-target long-term tracking accuracy and sensor long-term radiation cost, and combined with the constraints of minimum working time and the set tracking task duration, determine the sensor scheduling scheme.

[0045] The central idea of ​​this method embodiment is to propose a long-term scheduling method for radiation control in multi-target tracking under cluttered environments, based on a mobile active and passive sensor system. First, a scheduling model is established to mathematically describe the motion state and measurement results of multiple targets, as well as sensor scheduling actions. Simultaneously, based on the working principle of radar and the idea of ​​interception probability, an improved radiation risk quantification method is proposed. Then, a Gaussian mixture probability hypothesis density filtering algorithm is used to predict long-term tracking accuracy, the proposed improved quantification method is used to predict long-term radiation cost, and an improved gray wolf optimization algorithm is used to solve the sensor scheduling scheme. Finally, the scheduling scheme is executed to obtain multi-target measurement information, and a joint generalized label multi-Bernoulli filtering algorithm is used to calculate the estimated target state. Simulation experiments show that the scheduling method proposed in this embodiment can effectively control radiation cost while ensuring tracking accuracy, exhibiting significant advantages compared to other methods. The following detailed description of the above technical solution of this method embodiment, in conjunction with the accompanying drawings, further illustrates this approach.

[0046] The objective of this method embodiment is to study aerial target tracking, which is a sensor scheduling problem, wherein, for example... Figure 2As shown, in Cartesian coordinates, multiple active and passive sensors based on mobile platforms track multiple targets. Assuming the targets move at near-uniform linear velocity, the sensor system consists of N mobile platforms, each equipped with at least one active sensor (radar) and one passive sensor (infrared detector). Since active sensors face radiation risks when observing targets, including being identified and even having their positions locked by the targets, a scheduling method is researched to improve the survivability of the sensor system, achieve accurate tracking of multiple targets, and effectively control the radiation risks of the sensor system.

[0047] In this example, the scheduling actions of the sensor system are specified. ,in Assign a plan to the platform. For platform mobility solutions. Define the platform allocation scheme. ,in or This indicates whether to schedule the active sensors on platform n to track the target at time k, and stipulates that only one sensor platform is scheduled at each sampling time.

[0048] Assumption And the scheduled platform n is in k The position at time 1 is Then the position of platform n at time k can be updated as follows:

[0049] (1)

[0050] in, It is the sensor that controls the speed. Indicates the platform's maneuvering direction selection. This indicates the platform's maneuver speed selection; there are a total of maneuver options. N R ×N θ +1 type (including sensors that remain stationary), then The value can be:

[0051] (2)

[0052] like Figure 3 As shown, N R =2 、N θ A mobile embodiment of the sensor platform when =8.

[0053] In this embodiment, step two establishes a multi-objective state model and a multi-objective measurement model based on POMDP (Partially Observable Markov Decision Process) theory and RFS (Random Finite Set) theory, specifically including the following implementation steps and calculation process.

[0054] 1. Multi-objective state model

[0055] definition k Multi-target motion state at all times:

[0056] (3)

[0057] Among them, the state information of each target , contains the target m Position and velocity components in the X and Y directions; for k The number of targets at any given time.

[0058] Known k The multi-target motion state at time t is And each target in k+ The probability of survival at time 1 is and with probability density Transition to a new state Then the target can be in k+ The behavior at time 1 is modeled as RFS: Its value is equal to when the target is destroyed. When the target survives, it equals ;at the same time k+ At time 1, new goals still exist, including k Derivative objectives of the constant survival objective and k+ The new goal at moment 1, therefore k+ Multi-objective state at time 1 It can be given by the union of surviving objectives, derived objectives, and newly formed objectives:

[0059] (4)

[0060] in, express k+ The survival rate (RFS) of the target at time 1. Indicates the survival target The derived target RFS, express k+The new target RFS at time 1. The motion characteristics of each target follow a Linear Gaussian Multi-target (LGM) model, then the target Markov transition probability density satisfies:

[0061] (5)

[0062] in, The mean is Covariance is P The Gaussian probability density function; This is the state transition matrix; Let be the process noise covariance matrix.

[0063] 2. Multi-target measurement model

[0064] The state definition obtained by the sensor system observing multiple targets is as follows:

[0065] (6)

[0066] in, for k The time sensor system observes information about the target m The measured values, due to target omissions and clutter interference, result in an incorrect number of targets observed. It is not necessarily equal to the true value. .

[0067] Considering the possibility of missed detections by the sensor system in target observation, let's assume the target state... The detection probability is and with likelihood Generate measurement values Then in k Every moment, every state Generate RFS: Its value is equal to the target when it is missed. When the target is detectable, it equals In addition, the sensor also receives interference from clutter, so the multi-target measurements received by the sensor are subject to interference. Including measurements from the target and clutter interference :

[0068] (7)

[0069] Since the sensor measurement model must also conform to LGM, the multi-target measurement likelihood is:

[0070] (8)

[0071] in, For measurement matrix; This is the measurement noise covariance matrix.

[0072] Due to the operating characteristics of active and passive sensors k Active sensors at all times n 1 Observation Target m The obtained measurement information includes the target slant range and azimuth:

[0073] (9)

[0074] passive sensors n 2 Observation Targets m The only measurement information obtained is the azimuth angle:

[0075] (10)

[0076] in, k Time sensor n The position is ,Target m The position is ,sensor n and target m The straight-line distance between them is azimuth angle is , Indicates sensor n Additive Gaussian measurement noise when observing a target.

[0077] In this preferred embodiment, step three employs the GM-PHD smoothing filter algorithm to calculate the estimated state of multiple targets, and uses the OSPA distance as a metric to predict the long-term tracking accuracy of each sensor for multiple targets. The specific analysis and calculation process is as follows:

[0078] S311. Determine the initial state of the sensor system parameters;

[0079] In this embodiment, in the decision-making time domain [ k , k + h Within [1], the sensor system scheduling action is as follows: Among them, the platform allocation scheme Platform mobility solution ; k The target estimated state at time 1 is The sensor platform is located at Meanwhile, since passive sensors do not pose a radiation risk, all passive sensors are required to operate continuously with consistent performance and equal azimuth measurement noise.

[0080] S312. Calculate the predicted value of the target state at each time point in the decision-making time domain;

[0081] according to k Target estimation state at time 1 And the established multi-objective state model can yield the following results. k Target state prediction value at time 1 Therefore, it can be based on The predicted values ​​of the target state at each time point within the decision-making time domain are obtained. .

[0082] S313. Calculate the predicted position of the sensor platform at each time point based on the target state prediction value of the sensor platform at each time point in the decision time domain.

[0083] k At any time, according to the platform's allocation plan Unscheduled platforms do not maneuver and their positions remain unchanged. .

[0084] Combining equation (1), if k Mobility Plan for Time Platform If the scheduled platform does not move, its position remains unchanged. .if If the dispatching platform initiates a maneuver, it will update according to the following formula. k Time and location:

[0085] (11)

[0086] Among them, if ,but ;like ,but And so on, if ,but .

[0087] Similarly, according to the platform's allocation scheme Platform mobility solution ,as well as k Predicted positions on various platforms at any time The predicted position of the sensor platform at each moment in the decision-making time domain can be obtained. .

[0088] S314. Calculate the target measurement prediction value based on the scheduling actions, target state prediction value, and sensor platform prediction position at each moment in the decision time domain.

[0089] According to the scheduling action Target state prediction value Sensor platform predicts location Calculate the target measurement prediction value Specifically, it can be divided into the following situations (here, we will use...). k (Taking one moment as an example, the same applies to other moments within the decision-making time domain.)

[0090] ① The target measurement prediction value is calculated by fusing active and passive sensor measurement information from the dispatching platform using the following formula:

[0091] (12)

[0092] in, express The active sensors on the scheduling platform constantly predict values ​​based on the target state. The obtained target slant distance measurement prediction value, The mean is 0 and the covariance is The noise in the active sensor slant range measurement of the dispatched platform. and Determined by the following formula:

[0093] (13)

[0094] in, , The predicted target azimuth angle measurement and the azimuth angle measurement noise covariance are obtained from the passive sensors on the dispatching platform. , It consists of the target azimuth measurement prediction value and azimuth measurement noise covariance obtained by the active sensors on the scheduling platform. It is the noise of the azimuth measurement after fusion. The covariance.

[0095] ② Passive positioning using unscheduled platform sensors;

[0096] This example illustrates passive localization using two unscheduled platforms' passive sensors. For platforms with more than two sensors, passive localization is performed by combining two sensors and averaging the results. The passive sensors on the unscheduled platforms predict values ​​based on the target state. The obtained target measurement prediction values ​​are as follows:

[0097] (14)

[0098] in, and These are the target azimuth angle measurement predictions obtained from the passive sensors on unscheduled platforms 1 and 2, respectively. The mean is 0 and the covariance is The passive sensor azimuth measurement noise. Based on and This allows us to obtain the predicted Cartesian coordinates of the target:

[0099] (15)

[0100] in, and These are the predicted locations for unscheduled platforms 1 and 2, respectively. Therefore, the target measurement prediction value obtained from passive localization by the unscheduled platforms is:

[0101] (16)

[0102] in, The mean is 0 and the covariance is The noise of passive positioning slant range measurement The mean is 0 and the covariance is Passive positioning azimuth measurement noise:

[0103] (17)

[0104] in, and This refers to the slant range measurement matrix and azimuth measurement matrix under passive positioning conditions. For the measurement matrix of the passive sensor, and ,but :

[0105] (18)

[0106] (19)

[0107] ③ System measurement information fusion

[0108] When using only two passive sensors for passive positioning, the predicted slope distance measurement has a large error, so only the predicted angle measurement is used. and By fusing the data, we obtain the predicted value of the target measurement information from the sensor system:

[0109] S315. Based on the target measurement prediction value obtained in step S314, calculate the target estimated state prediction value at each time in the decision time domain.

[0110] (20)

[0111] Based on the characteristics and scheduling actions of the GM-PHD algorithm Independent filtering is performed on all sensor platforms, and the weights of the Gaussian component model parameters are set during the prediction stage. w mean m Covariance PDuring filtering, each platform retrieves the parameters of the system's Gaussian component model and substitutes them into the target measurement prediction value. The target estimated state prediction value is calculated. .

[0112] S316. Based on the calculated target estimated state prediction value, determine the OSPA distance at each time point in the decision time domain, and calculate the long-term tracking accuracy prediction value in the decision time domain based on this. k At time, the predicted value of the target state Based on this, calculate the estimated state prediction value of the target. OSPA distance Similarly, the OSPA distance at each time point within the decision-making time domain can be obtained. The long-term tracking accuracy prediction value in the decision time domain is calculated based on this, as shown in the following formula:

[0113] (twenty one).

[0114] Furthermore, the implementation steps of step three of this embodiment, which is based on the radiation risk quantification method according to signal power density, are as follows:

[0115] Since the radiation risk of active sensors is a product of their outward radiated electromagnetic waves, the radiation risk of sensors can be assessed based on the basic radar formula and the concept of probability of intercept. Quantification is performed. If... k Time sensor n and target m The distance between them is Then the sensor n Emit electromagnetic waves to illuminate the target m The power density at that point is:

[0116] (twenty two)

[0117] in The radiated power of the sensor. This represents the sensor's transmit gain.

[0118] sensor n Irradiate the target m The radiation signal at that location can always be detected by the alarm receiver, but only when the power density reaches the minimum detectable signal threshold. Only then will radiation risks be identified and marked. Combining equation (22), the target can be obtained. m For sensors n The maximum effective distance for radiation alarms is:

[0119] (twenty three)

[0120] At the same time, due to the interception probability α The definition of is:

[0121] (twenty four)

[0122] Then it can be considered that when hour, The radiation alarm receiver cannot recognize the signal, meaning that the impact of radiation risk is not considered in the dispatching process. And when... hour, Then the radiation alarm receiver can identify the radiation signal and mark the radiation risk. And with Further reduction in interception probability α The probability of the radiation signal being intercepted and the platform's location being locked as the platform rises. The value also increases accordingly. This allows us to define the sensor radiation risk. :

[0123] (25)

[0124] Where B is the radiation risk quantification coefficient.

[0125] The method for predicting the long-term radiation cost of a sensor based on the radiation risk quantification method using signal power density includes the following steps:

[0126] S321. Determine the initial state of the sensor system parameters; initialize the current time. t=k ;

[0127] In the decision-making time domain [ k , k + h Within 1], sensor system scheduling actions Among them, the platform allocation scheme Platform mobility solution , k Sensor platform position at time 1 Based on steps S312 and S313, the predicted position of the sensor platform is obtained. and target estimated state prediction value .

[0128] S322. Calculate the predicted distance between the scheduled platform and the target at each time point in the decision-making time domain;

[0129] k At any given time, the predicted location of the scheduling platform is Target estimated state prediction value:

[0130] (26)

[0131] The predicted distance from the scheduled platform to the target is:

[0132] (27)

[0133] Similarly, according to the platform's allocation scheme Predicted location and target estimated state prediction value This allows us to obtain the predicted distance from the scheduled platform to the target at each moment within the decision-making time domain. .

[0134] S323. Calculate the predicted radiation risk value of active sensors at each moment in the decision-making time domain.

[0135] The maximum effective range for radiation alarms targeting active sensors is specified as follows: . k At any given time, the predicted distance from the scheduled platform to all targets is... Combined with formula (25), the predicted radiation risk value is calculated. .

[0136] like , that is, the target m The radiation alarm receiver cannot recognize the signals from the scheduled active sensors and does not consider the impact of radiation risks. .

[0137] like , that is, the target m The radiation alarm receiver can identify the signals of the scheduled active sensors and obtain... k The target is constantly tracked by the dispatching platform. m The predicted radiation risk at that time is as follows:

[0138] .

[0139] Select The maximum value in is used as k The predicted radiation risk value generated when the target is tracked by the scheduling platform at any time:

[0140] (28)

[0141] Similarly, based on the predicted distance from the scheduled platform to the target... The predicted radiation risk values ​​of the sensor system at each time point within the decision-making time domain are obtained. .

[0142] S324. Calculate the predicted value of long-term radiation cost based on the radiation risk prediction value of the sensor system;

[0143] Based on the radiation risk prediction value of the sensor system Calculate the predicted long-term radiation cost within the decision-making time domain:

[0144] (29).

[0145] In this embodiment, based on the above calculation process, in step four, the optimization function is created with the objective of achieving an effective balance between target tracking accuracy and sensor radiation cost. The specific objective optimization function is as follows:

[0146] (30)

[0147] in, These are the weighting coefficients. h For the decision duration of long-term scheduling, To determine long-term tracking accuracy in the decision-making time domain, This is the cost of long-term radiation within the decision-making time domain.

[0148] In step five of this embodiment, the decision-making basis is the predicted value of the long-term tracking accuracy of multiple targets and the predicted value of the long-term radiation cost of the sensor. Combined with the constraint of the minimum working time and the set tracking task duration, the sensor scheduling scheme is determined.

[0149] The idea behind the sensor scheduling method in this embodiment is to achieve long-term target tracking accuracy prediction and long-term radiation cost prediction. Based on the target optimization function (26), an improved gray wolf optimization algorithm is used to solve the scheduling scheme, execute the target tracking task, update the sensor position and obtain target measurement information according to the scheduling scheme, and finally estimate the target state using the Joint-GLMB filtering algorithm, as detailed below. Figure 2 As shown in the embodiment of this method, the tracking task duration is set to H.

[0150] In this embodiment, to adapt to the characteristics of solving long-term sensor scheduling schemes, improvements are made to the Grey Wolf Optimization algorithm (GWO) regarding the wolf pack initialization and search mechanisms. The improved GWO algorithm is then used to solve the scheduling scheme, specifically including the following execution steps:

[0151] A1. Wolf pack initialization;

[0152] Assuming the current is k At time [time], the improved GWO algorithm is used to obtain [ k , k+h 1] The optimal scheduling scheme in the time domain. The number of wolves is defined as follows: W The number of algorithm iterations is It The search target, i.e., the scheduling scheme, is a 3D matrix:

[0153] (31)

[0154] in, P The proposed scheduling platform at each time point has a value range of [1, ...]. N Integers within ] Q The proposed maneuver schemes for the scheduling platform at each time point are defined, with values ​​ranging from [1, ...]. N R N θ If the integer is within the range of +1, then the search space size is ( N ( N R N θ +1)) h .

[0155] Before the search begins, the wolf pack positions are initialized. Each wolf's current position represents the current scheduling scheme it has found. The value of each element in the initial positions is a random integer within the range of possible values.

[0156] (32)

[0157] in, Λ[1, N ] represents a value in [1, N A random integer within the range of ].

[0158] When using random initialization to generate the initial population, a small wolf pack makes it difficult to guarantee good population diversity, while increasing the wolf pack size leads to a significant increase in computational complexity. Therefore, to improve the diversity of the initial population and the search coverage, the initial positions of two wolves in the pack are changed to the lower and upper bounds of the scheduling scheme, respectively:

[0159] (33)

[0160] (34)

[0161] A2. Calculate the fitness value of each wolf at the current iteration number;

[0162] Based on the above, calculate the long-term tracking accuracy prediction value for the position of each wolf at the current iteration number. and long-term radiation cost prediction And, combined with the objective optimization function (26), calculate the fitness value of each wolf:

[0163] (35)

[0164] A3. Update and record the fitness values ​​of the wolf pack. α Wolf, β wolves and δ Wolf;

[0165] Based on the wolf pack's fitness value Fi Select and record α Wolf, β wolves and δ Wolves, specifically the three wolves with the smallest fitness values ​​up to the current iteration, satisfy the following conditions: The remaining wolves are defined as ω The update method for wolves is shown in Table 1.

[0166] Table 1 α Wolf, β wolves and δ Wolf Update

[0167]

[0168] A4. Update the wolf pack's location;

[0169] Since all elements in the position of each wolf must be integers within the range of values ​​when solving the scheduling scheme, the position updated after each iteration in the standard gray wolf optimization algorithm is defined as the proposed search position. After adaptive processing of the proposed search position, the wolf position is updated again.

[0170] If the proposed search position exceeds the boundary, the corresponding upper and lower boundary values ​​are used; if the proposed search position is within the range, it is rounded to the nearest integer. Meanwhile, since the wolf's position is the current scheduling scheme obtained by that wolf's search, it is considered a... It is a 3D matrix, so the wolf's position is updated based on each element in the matrix.

[0171] (36)

[0172] in, , , LoU for The intended search location, due to Wolf, wolves and The wolf's position represents the optimal scheduling scheme for the current iteration, therefore:

[0173] (37)

[0174] (38)

[0175] (39)

[0176] (40)

[0177] in, Convergence factor: ; It is a random number in [0,1].

[0178] A5. Based on the number of iterations, perform an iterative loop to obtain a sensor scheduling scheme;

[0179] like it <It Then jump to A2 to continue iterating until... it = It When, output ,get[ k,k+h 1] Sensor scheduling scheme in the time domain.

[0180] Based on the above, such as Figure 4 As shown, the specific implementation process of step five in this embodiment is as follows:

[0181] S1. Initialize the current time. t=k Wolf pack initialization;

[0182] k At time 1, the multi-objective estimated state is: The sensor location is The long-term scheduling decision duration is h The GM-PHD filtering algorithm is used to predict long-term tracking accuracy, and an improved gray wolf optimization algorithm is employed to solve the scheduling scheme, including the number of wolves. W Only, number of iterations It Second-rate.

[0183] S2. Solve the scheduling scheme based on the improved Grey Wolf optimization algorithm;

[0184] Based on the improved gray wolf optimization algorithm, k At time, substitute the target estimated state and sensor position Iterative calculation It Next, obtain [ k,k+h 1] Long-term scheduling scheme in the time domain And update the parameters of the GM-PHD filtered Gaussian component model of the system in the prediction phase;

[0185] S3, Execute the scheduling plan;

[0186] 1) Update the sensor platform location;

[0187] According to step S313 and k Time scheduling scheme The sensor platform location has been updated to .

[0188] 2) Obtain target measurement information;

[0189] according to Schedule the corresponding sensors to work and obtain k Target measurement information at time .

[0190] 3) Calculate the estimated state of the target;

[0191] Will Substituting into the Joint-GLMB filtering algorithm, calculate k Target estimation state at time 1 And update the parameters of the Joint-GLMB filter Gaussian component model during the execution phase of the system.

[0192] 4) Based on the selected sensor's minimum operating time, calculate [the following] within the decision time domain using a loop. k,k+h 1] Target estimation state at each time point in the time domain;

[0193] Similarly, based on the scheduling actions ,calculate[ k,k+h 1] Target estimation state at each time point in the time domain .

[0194] S4. Determine if the tracking task duration has been reached. H ,like k+h 1< H ,but k+h Make decisions again at any time. k+h Target estimation state at time 1 Sensor platform location Substitute the values ​​into step S2 to solve for the scheduling scheme and calculate the target estimated state.

[0195] like k+h 1≥ H If the task ends, the scheduling process is complete.

[0196] The effectiveness of the method in this embodiment will be verified through simulation experiments below.

[0197] In this simulation experiment, it is assumed that the sensor system consists of 3 movable platforms, tracking 4 targets of the same type. The initial positions of the platforms are... N 1(0m, 500m), N 2( 500m, 400m),N 3(500m, 400m), sampling interval =1s, without loss of generality, the measurement noise covariance of the active sensors on the three platforms are as follows: R Ra1 =diag([5m;0.001rad]), R Ra2 =diag([10m; 0.001 rad]), R Ra3 = diag([15m; 0.001rad]). Correspondingly, the radiation warning range of the three active sensors relative to four similar targets is: =500m, =400m, = 300m, meaning that sensors with better tracking performance are more easily identified by the target's radiation alarm receiver. The measurement noise covariance of a passive sensor is R Ir =10 5 rad. Control speed of the sensor platform. =5m / s, maneuver direction selection =8, speed selection =2.

[0198] The targets all adopt near-uniform linear motion, with the following initial states: M 1(0m,0m / s,800m, 14m / s), M 2(800m, 14m / s, 0m, 7m / s M 3( 800m, 14m / s, 200m 7m / s), M 4(800m, 14m / s, 600m (9m / s). Target survival probability. =0.99, detection probability =0.98; OSPA distance cutoff parameter c =100, order parameter p =1. The RFS of clutter interference in the observation environment is modeled as a Poisson distribution, and its intensity is... ,in u ( Z k () is the probability density function of a uniform distribution; V =( 1000m, 1000m) × ( The sensor monitoring area is defined as 1000m, 1000m. =3 represents the average number of clutter particles per unit area. The wolf pack size in the gray wolf optimization algorithm. W =5, number of algorithm iterations It= 5. Total simulation time H =100s, and the simulation result is the average of 100 Monte Carlo experiments.

[0199] (1) Performance analysis of scheduling methods

[0200] To fully verify the performance of the proposed non-myopic scheduling method for mobile active / passive sensors (NMAPS), it is compared with the myopic scheduling method for mobile active / passive sensors (MMAPS), the non-myopic scheduling method for active / passive sensors (NAPS), and the non-myopics scheduling method for mobile active sensors (NMAS).

[0201] The target tracking accuracy is measured by the average OSPA distance, positioning error OSPA-L, and potential error OSPA-C within the simulation time. The cumulative radiation cost (CRC) within the simulation time is used to evaluate the scheduling method's control effect on radiation risk. The average runtime (TIME) of a single simulation (100 sampling intervals) is used to evaluate the real-time performance of the scheduling method. For ease of description, the "average OSPA distance, OSPA-L, and OSPA-C within the simulation time" will be replaced by "OSPA," "OSPA-L," and "OSPA-C" respectively, with "OSPA distance" specifically used to describe the optimal sub-mode allocation distance at each time step. Using NMAPS to track the target, the resulting multi-target motion trajectory and scheduling sequence are as follows: Figure 5 , 6 As shown.

[0202] like Figure 7 , 8 As shown, this represents the long-term scheduling decision duration. hWhen the time is 2s, the OSPA and instantaneous OSPA distance are obtained when different scheduling methods are used for optimal tracking accuracy scheduling, and the CRC and instantaneous radiation cost are obtained when radiation cost is optimized scheduling.

[0203] To further verify the superiority and stability of the proposed NMAPS, decision durations were taken respectively. h Balanced optimization scheduling is performed at 2s, 3s, and 4s. Table 2 below shows the weighting coefficients. α= Performance of each scheduling method for balanced optimization scheduling under a value of 0.5.

[0204]

[0205] Because long-term scheduling has better optimization performance than short-term scheduling, therefore Figure 7 , 8 In this embodiment, the OSPA and CRC obtained by NMAPS in the optimal scheduling of tracking accuracy are lower than those obtained by MMAPS in the optimal scheduling of radiation cost. Moreover, in Table 2, the OSPA and CRC obtained by NMAPS in the balanced optimal scheduling under the three decision durations are also lower than those of MMAPS. h At 2s, the single simulation time was reduced by 32.37%, fully demonstrating the effectiveness of NMAPS. Meanwhile, in h =2s and h At 3s, the results obtained by NMAPS are basically the same, but h At 4s, the CRC increases significantly, because the performance of the optimization algorithm will increase with the decision duration. h The increase and decrease indicate that in W =5、 It =5、 h Under the condition of 2s, the proposed NMAPS basically obtained the optimal solution.

[0206] Compared to the three long-term scheduling methods, the proposed NMAPS can change the minimum distance between the scheduled platform and the target by maneuvering the sensor platform. This allows for advantages in scheduling performance; such as Figure 6 As shown, when optimizing the tracking accuracy scheduling, NMAPS achieves a slightly better OSPA than NAPS, while... Figure 7 When optimizing scheduling based on radiation cost, NMAPS yields a significantly lower CRC than NAPS, indicating that platform maneuvering effectively reduces sensor radiation risk, but its improvement on tracking accuracy is limited. In contrast, the proposed NMAPS improves target tracking accuracy and reduces OSPA distance at various time points through passive localization and information fusion, thus achieving better scheduling performance. Figure 7As shown, the OSPA obtained by NMAPS tracking accuracy optimization scheduling is significantly better than MMAS; however, in... Figure 8 When optimizing the scheduling of radiation costs, since NMAS also has the same mobility, the CRC obtained by NMAPS does not have a significant advantage.

[0207] Therefore, as shown in Table 2, NMAPS has advantages in tracking accuracy and maneuverability compared to NAPS and NMAS. When performing balanced optimization scheduling under three decision durations, NMAPS achieves the best control effect on radiation risk while ensuring tracking accuracy, and the lowest CRC is obtained from the scheduling.

[0208] At the same time, as can be seen from the preceding text, with h With the increase in the number of scheduling options, the performance of long-term scheduling balance optimization should improve, and the cumulative radiation cost under the optimal scheduling scheme should be further reduced. However, due to the size of the search space, the number of selectable scheduling schemes ( N ( N R N θ +1)) h It will exhibit exponential growth, in the size of the wolf pack W Number of algorithm iterations It Without changing the underlying principles, the performance of the improved gray wolf optimization algorithm decreases, and the resulting scheduling scheme deteriorates. Therefore, in Table 2, h The CRC increases significantly at 4s. Increasing the CRC can be considered. W or It This approach addresses the problem, but it increases system response time, impacting the real-time performance of sensor scheduling. Therefore, given the current wolfpack size and iteration count, the decision-making time using the proposed NMAPS for scheduling is [not specified]. h The time should generally not exceed 4 seconds.

[0209] Overall, NMAPS achieved better optimization results than other scheduling methods under different decision duration conditions, demonstrating good adaptability and robustness; simultaneously, in terms of parameters... W =5、 It =5、 h With a time limit of 2 seconds, the optimal solution for the scheduling scheme can be obtained, achieving the expected goal.

[0210] (2) Performance analysis of the improved gray wolf optimization algorithm;

[0211] This simulation experiment aims to verify the scheduling performance of the improved Gray Wolf Optimization algorithm (IGWO). The Gray Wolf Optimization algorithm (GWO), Particle Swarm Optimization algorithm (PSO), and improved Hill Climbing algorithm (IHC) were respectively substituted into NMAPS for comparative experiments. The decision-making time for long-term scheduling was measured. h =2s, the wolf pack size of GWO is 5, the number of particles of PSO and IHC is 5, the number of iterations is also 5, and all other simulation conditions are the same.

[0212] Since the number of particles and iterations are the same in all four optimization algorithms, the results of single-index optimal scheduling are used as the evaluation standard for comparing the performance of the optimization algorithms. That is, the smaller the TIME, OSPA, or CRC, the better the solution capability of the corresponding optimization algorithm. As shown in Tables 3 and 4, the results of tracking accuracy optimal scheduling and radiation cost optimal scheduling are obtained by using different optimization algorithms.

[0213] As shown in Tables 3 and 4, when NMAPS adopts the proposed IGWO solution for the scheduling scheme, the OSPA and TIME are lowest for tracking accuracy optimization scheduling, and the CRC and TIME are lowest for radiation cost optimization scheduling. This indicates that under the same particle number and iteration count, IGWO has better optimization performance than other optimization algorithms, yielding the optimal solution, while the runtime of a single simulation is also acceptable. The simulation experiments in this section demonstrate the advantages of the improved Grey Wolf optimization algorithm, effectively improving the efficiency of the scheduling scheme solution. Furthermore, by combining it with various solution algorithms, the feasibility of the proposed NMAPS is further proven, enabling stable and effective sensor scheduling.

[0214]

[0215] This embodiment proposes a long-term scheduling method for multi-target tracking in mobile active and passive sensor systems. The method consists of two main stages. In the first stage, based on the GM-PHD filtering algorithm for long-term tracking accuracy prediction and the radiation cost prediction using a radiation risk quantification method based on signal power density, an improved gray wolf optimization algorithm is used to search for the optimal scheduling scheme. In the second stage, the sensor platform position is updated according to the scheduling scheme, and target measurement information is processed using measurement information fusion and passive sensor localization methods. The Joint-GLMB filtering algorithm is then used to obtain the estimated target state. Finally, simulation experiments are conducted to analyze the performance of the scheduling method and optimization algorithm. By comparing with three different scheduling methods, the effectiveness of the proposed scheduling method under different optimization objectives is verified, its performance advantages are further analyzed, and suggestions for the decision-making time of long-term scheduling are given based on the changes in scheduling performance. Under the same experimental conditions, using different optimization algorithms for comparison, the advantages of the improved gray wolf optimization algorithm in solving scheduling schemes are fully verified, effectively improving computational efficiency and accuracy.

[0216] System Implementation Examples

[0217] According to embodiments of the present invention, a long-term scheduling system for mobile active and passive sensors in clutter environments is provided, such as... Figure 9 As shown, this is a schematic block diagram of a long-term scheduling system for mobile active and passive sensors in a clutter environment according to an embodiment of the present invention. The long-term scheduling system for mobile active and passive sensors in a clutter environment according to this embodiment of the present invention includes:

[0218] Multiple active and passive sensors based on mobile platforms are used to track multiple targets. Assuming that the targets are moving in a near-uniform linear motion, the system includes N mobile platforms, each equipped with at least one active sensor (radar) and one passive sensor.

[0219] Sensor system scheduling model establishment unit: used to determine the sensor system scheduling model.

[0220] Multi-objective state model and measurement model establishment unit: used to establish multi-objective state models and multi-objective measurement models based on POMDP theory and RFS theory;

[0221] Target optimization function creation unit: Creates an optimization function with the objective of achieving an effective balance between target tracking accuracy and sensor radiation cost.

[0222] Multi-target long-term tracking accuracy prediction unit: The GM-PHD smoothing filter algorithm is used to calculate the estimated state of multiple targets, and the OSPA distance is used as the metric to predict the long-term tracking accuracy of multiple targets by each sensor.

[0223] Long-term radiation cost prediction unit: The radiation risk quantification method based on signal power density is used to predict the long-term radiation cost value of the sensor.

[0224] Sensor scheduling scheme solution and execution unit: Based on the prediction values ​​of multi-target long-term tracking accuracy and sensor long-term radiation cost, and combined with the constraints of the minimum working time of the sensor and the set tracking task duration, the sensor scheduling scheme is determined.

[0225] In this system embodiment, the sensor system scheduling model established by the sensor system scheduling model establishment unit is as follows:

[0226] Specify the scheduling actions of the sensor system ,in Assign a plan to the platform. For platform mobility solutions. Define the platform allocation scheme. ,in or This indicates whether to schedule the active sensors on platform n to track the target at time k, and stipulates that only one sensor platform is scheduled at each sampling time.

[0227] Assumption And the scheduled platform n is in k The position at time 1 is Then the position of platform n at time k can be updated as follows:

[0228] (41)

[0229] in, It is the sensor that controls the speed. Indicates the platform's maneuvering direction selection. This indicates the platform's maneuver speed selection; there are a total of maneuver options. N R ×N θ +1 type (including sensors that remain stationary), then The value can be:

[0230] (42)

[0231] In this embodiment, the multi-target state model and the multi-target measurement model can be specifically referred to equations (3) to (10) in the above method embodiment, and will not be repeated here.

[0232] The objective optimization function establishment unit establishes an objective optimization function with the goal of balancing the sensor's tracking accuracy and radiation cost. The specific objective optimization function is as follows:

[0233] Based on the sensor's minimum operating time hBased on this, the objective optimization function is constructed as follows:

[0234] (43)

[0235] in, These are the weighting coefficients. For long-term tracking accuracy, This comes at the cost of long-term radiation.

[0236] In this system embodiment, the process of the multi-target long-term tracking accuracy prediction unit to predict the multi-target long-term tracking accuracy refers to the process of steps S311-S316 in the above method embodiment; it will not be repeated here.

[0237] In this system embodiment, the specific implementation process of the multi-step radiation cost prediction unit by using the radiation effect as an indicator to quantify the radiation cost of the sensor is referred to formulas (22) to (25) in the above method embodiment, and all the contents of steps S321-S324, which will not be elaborated here.

[0238] The sensor scheduling scheme solution and execution unit implements the sensor scheduling method based on the realization of long-term target tracking accuracy prediction and long-term radiation cost prediction. According to the target optimization function, an improved gray wolf optimization algorithm is used to solve the scheduling scheme, execute the target tracking task, update the sensor position according to the scheduling scheme, obtain target measurement information, and finally estimate the target state using the Joint-GLMB filtering algorithm. In this embodiment, the tracking task duration is set to H. The improved gray wolf optimization algorithm for solving the scheduling scheme refers to steps A1-A5 in the above embodiment. The specific steps for solving and executing the sensor scheduling scheme in this embodiment refer to steps S1-S4, specifically including the following steps:

[0239] B1. Initialize the current time. t=k Wolf pack initialization;

[0240] k At time 1, the multi-objective estimated state is: The sensor location is The long-term scheduling decision duration is h The GM-PHD filtering algorithm is used to predict long-term tracking accuracy, and an improved gray wolf optimization algorithm is employed to solve the scheduling scheme, including the number of wolves. W Only, number of iterations It Second-rate.

[0241] B2. Solve the scheduling scheme based on the improved Grey Wolf optimization algorithm;

[0242] Based on the improved gray wolf optimization algorithm, kAt the moment, substitute into the target estimated state and the sensor position , perform iterative calculation It times to obtain k,k+h 1) Long-term scheduling plan within the time domain , and update the GM-PHD filtering Gaussian component model parameters of the system in the prediction stage;

[0243] B3. Execute the scheduling plan;

[0244] 1) Update the sensor platform position;

[0245] According to step S313 and k the scheduling plan at the moment , the sensor platform position is updated to .

[0246] 2) Obtain the target measurement information;

[0247] According to , schedule the corresponding sensors to work and obtain k the target measurement information at the moment .

[0248] 3) Calculate the target estimated state;

[0249] Substitute into the Joint-GLMB filtering algorithm to calculate k the target estimated state at the moment , and update the Joint-GLMB filtering Gaussian component model parameters of the system in the execution stage.

[0250] 4) According to the selected minimum working duration of the sensors, loop within the decision time domain to calculate k,k+h 1) The target estimated state at each moment within the time domain;

[0251] Similarly, according to the scheduling action , calculate k,k+h 1) The target estimated state at each moment within the time domain .

[0252] B4. Judge whether the tracking task duration H is reached. If k+h 1 < H, then k+h make a new decision at the moment, and substitute k+h the target estimated state at the moment 1 , the sensor platform position into step B2 to solve the scheduling plan and calculate the target estimated state.

[0253] like k+h If 1 ≥ H, then the scheduling task ends.

[0254] It should be noted that the system embodiments of the present invention are method embodiments corresponding to the above methods. The specific operations of the processing steps of each module or unit can be understood by referring to the description of the method embodiments, and will not be repeated here.

[0255] like Figure 10 As shown, the present invention also provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the long-term scheduling method for mobile active and passive sensors in a clutter environment as described in the above embodiments, or when the computer program is executed by a processor, it implements the long-term scheduling method for mobile active and passive sensors in a clutter environment as described in the above embodiments.

[0256] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0257] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A long-term scheduling method for mobile active and passive sensors in cluttered environments, characterized in that, Includes the following steps: Determine the sensor system scheduling model; For multi-sensor multi-target tracking applications in cluttered environments, a multi-target state model and a multi-target measurement model are established based on POMDP and RFS theories. The GM-PHD smoothing filtering algorithm is used to calculate the estimated state of multiple targets, and the OSPA distance is used as a metric to predict the long-term tracking accuracy of each sensor for multiple targets; the radiation risk quantification method based on signal power density is used to predict the long-term radiation cost of the sensors. An optimization function is created with the objective of achieving an effective balance between target tracking accuracy and target sensor radiation cost. Based on the predicted values ​​of multi-target long-term tracking accuracy and sensor long-term radiation cost, and combined with the constraints of the minimum working time of the sensor and the set tracking task duration, the sensor scheduling scheme is determined. The sensor system includes multiple mobile platforms, each equipped with at least one active sensor and one passive sensor. The decision-making process, based on the predicted accuracy of multi-target long-term tracking and the predicted radiation cost of sensors, combined with the constraints of minimum working time and the set tracking task duration, specifically includes the following steps: S1. Initialize the current time. t=k Wolf pack initialization, including initializing target estimation state and sensor platform position; S2. Solve the scheduling scheme based on the improved Grey Wolf optimization algorithm; S3. Based on the minimum tracking time of the sensor, decide on the time-domain loop and execute the scheduling scheme; S4. Determine if the tracking task duration has been reached. H ,like k+h 1< H ,but k+h Make decisions again at any time. k+h Target estimation state at time 1 Sensor platform location Substitute into step S1 to solve for the scheduling scheme and calculate the target estimated state; if k+h- 1≥ H If so, the scheduling task ends; in, h This is the minimum operating time of the sensor. H Track the duration of the target task; The scheduling scheme is solved using the improved Grey Wolf optimization algorithm, which includes the following execution steps: A1. Wolf pack initialization: Change the initial positions of two wolves in the wolf pack to the lower and upper bounds of the scheduling scheme, respectively. A2. Calculate the fitness value of each wolf at the current iteration number; A3. Update and record the fitness values ​​of the wolf pack. α Wolf, β wolves and δ Wolf; A4. Update the wolf pack position; the position after each iteration is defined as the proposed search position. If the proposed search position exceeds the boundary, take the corresponding upper and lower boundary values; if the proposed search position is within the range of values, round it to the nearest integer. A5. Based on the number of iterations, perform an iterative loop to obtain a sensor scheduling scheme; The GM-PHD smoothing filtering algorithm is used to calculate the estimated state of multiple targets, and the OSPA distance is used as a metric to predict the long-term tracking accuracy of each sensor for multiple targets. The specific analysis and calculation process is as follows: Determine the initial state of the sensor system parameters; Calculate the predicted target state value at each time point within the decision-making time domain; Calculate the predicted position of the sensor platform at each time point based on the target state prediction value of the sensor platform at each time point within the decision time domain. The target measurement prediction value is calculated based on the scheduling actions, target state prediction value, and sensor platform prediction position at each moment in the decision time domain. Based on the calculated target measurement prediction values, calculate the target estimated state prediction values ​​at each time point in the decision time domain; Based on the calculated target estimated state prediction value, the OSPA distance at each time point in the decision time domain is determined, and the long-term tracking accuracy prediction value in the decision time domain is calculated based on this.

2. The long-term scheduling method for mobile active and passive sensors in a clutter environment as described in claim 1, characterized in that, Step S2 specifically includes: Update the sensor platform position; obtain target measurement information; calculate the current target estimated state; based on the selected minimum working time of the sensor, make a decision on the time domain loop and calculate the target estimated state at each time point in the time domain.

3. The long-term scheduling method for mobile active and passive sensors in a clutter environment as described in claim 1, characterized in that, Sensor radiation risk As shown in the following formula: （1） in, For sensors n Emit electromagnetic waves to illuminate the target m Power density at the location; for k Time sensor n and target m The distance between them For the goal m For sensors n The maximum effective distance of the radiation alarm; The radiated power of the sensor. B is the emission gain of the sensor; B is the radiation risk quantification coefficient.

4. The long-term scheduling method for mobile active and passive sensors in a clutter environment as described in claim 1, characterized in that, The radiation risk quantification method based on signal power density for predicting the long-term radiation cost of sensors includes the following steps: determining the initial state of sensor system parameters; Calculate the predicted distance between the scheduled platform and the target at each time point within the decision-making time domain; Calculate the predicted radiation risk values ​​of active sensors at each moment in the decision-making time domain; Based on the predicted radiation risk value of the sensor system, calculate the predicted long-term radiation cost of the sensor.

5. A long-term scheduling system for mobile active and passive sensors in clutter environments, used to implement the long-term scheduling method for mobile active and passive sensors in clutter environments as described in any one of claims 1 to 4, characterized in that, This includes a sensor system with multiple mobile platforms, each platform equipped with at least one active sensor and one passive sensor; Sensor system scheduling model establishment unit: used to determine the sensor system scheduling model; Multi-objective state model and measurement model establishment unit: used to establish multi-objective state models and multi-objective measurement models based on POMDP theory and RFS theory; Target optimization function establishment unit: Creates an optimization function with the objective of achieving an effective balance between target tracking accuracy and target sensor radiation cost; Multi-target long-term tracking accuracy prediction unit: The GM-PHD smoothing filter algorithm is used to calculate the estimated state of multiple targets, and the OSPA distance is used as the metric to realize the prediction of the long-term tracking accuracy value of multiple targets by each sensor. Long-term radiation cost prediction unit: The radiation risk quantification method based on signal power density is used to predict the long-term radiation cost value of the sensor; Sensor scheduling scheme solution and execution unit: Based on the prediction values ​​of multi-target long-term tracking accuracy and sensor long-term radiation cost, and combined with the constraints of the minimum working time of the sensor and the set tracking task duration, the sensor scheduling scheme is determined.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the long-term scheduling method for mobile active and passive sensors in clutter environments as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the long-term scheduling method for mobile active and passive sensors in clutter environments as described in any one of claims 1 to 4.

Images