Multi-sensor long-term scheduling method based on bilateral joint risk control

Abstract

The invention discloses a multi-sensor long-term scheduling method based on bilateral joint risk control, and the method comprises the steps: firstly constructing a multi-sensor multi-target scene, and obtaining data related to sensors and targets; secondly, on the basis of data related to a sensor and a target, a bilateral joint risk control model is constructed by integrating a target identification risk and a target long-term tracking cost; and finally, solving the bilateral joint risk control model by using an improved artificial bee colony algorithm, and obtaining a multi-sensor long-term scheduling distribution scheme through one-to-one correspondence of obtained solutions. According to the method, the situation that the measurement information of the sensors is affected due to the fact that accurate risk assessment among the targets cannot be achieved due to the fact that the number of the targets is variable is made up, the risk detection capacity of the sensors and the risk estimation calculation efficiency are improved, and the optimal multi-sensor scheduling scheme is obtained.

Description

Technical Field

[0001] The present invention relates to the fields of sensor management, optimization algorithms, etc., and more specifically, to a multi-sensor long-term scheduling method based on bilateral joint risk control. Background Art

[0002] With the acceleration of the informatization process, there are more and more information resources of multi-sensors. Using only the detection performance of the sensors themselves as the standard for battlefield risk assessment has far been unable to meet the needs of combat. To further improve the objectivity and diversity of battlefield situation risk assessment, comprehensive evaluation should be carried out on various aspects of risks to better guarantee the judgment of battlefield risks. However, the battlefield environment is complex, and the number of targets changes under the situation of being attacked and threatened. In addition, the previous risk theory did not consider the impact of the change in the number of targets on battlefield risk assessment, and the ability to judge battlefield risks and reasonably schedule multi-sensor resources in the scenario of target number change.

[0003] The problem that the multi-sensor system needs to solve is how to evaluate and control various risks it faces under different environmental conditions, so as to maximize the combat mission effectiveness and make the system operate safely and reliably. This method can effectively improve the combat effectiveness and confrontation ability of the system under various complex environmental conditions, making the combat efficiency of the system higher and the robustness stronger. In recent years, domestic and foreign research has successively achieved many results in the research on the problem of multi-sensor collaborative tracking and allocation with risk control as the core. However, the existing research is still insufficient in aspects such as risk modeling and optimization algorithms. In particular, how to effectively and efficiently achieve risk assessment and resource allocation under the cooperation of multiple targets and multiple sensors is an urgent problem to be solved. Summary of the Invention

[0004] The purpose of the present invention is to propose a multi-sensor long-term scheduling method based on bilateral joint risk control in view of the deficiencies of the existing research in aspects such as risk modeling and optimization algorithms, which effectively improves the risk detection ability of sensors and the efficiency of risk estimation calculation.

[0005] The present invention mainly solves the problem of multi-sensor long-term scheduling based on bilateral joint risk control through the following methods: comprehensively considering the new target recognition risk and the long-term tracking cost of the target, establishing a bilateral joint risk control model, and imposing constraints on the radiation risk of the sensors. Using the improved artificial bee colony algorithm to solve the objective function, dynamically solving the multi-sensor scheduling resources, and obtaining the optimal multi-sensor scheduling scheme.

[0006] The specific implementation steps of the present invention are as follows:

[0007] S1: Construct a multi-sensor multi-target scenario and obtain data related to sensors and targets. This includes the sampling period of the sensors, the number of sensors, the lower bound of the distance corresponding to the minimum measurement error variance of the sensors and the corresponding optimal signal-to-noise ratio, the path loss exponent of the sensors, the covariance matrix of the process noise of the sensors, the measurement noise variance of the sensors; data related to the targets: the number of targets, the turning rate of the targets, the motion model of the targets; scenario information: the number of samples for Monte Carlo sampling, the simulation time, the initial filtering value, and the initial value of the filtering error covariance matrix.

[0008] S2: Based on the data related to sensors and targets, construct a bilateral joint risk control model. Identify the occurrence of targets in the targets of S1 to obtain the recognition risk of the targets, then calculate the threat risk of the existing targets, and then obtain the long-term tracking cost of the targets according to the long-term prediction. Finally, with the threat risk of the targets as the constraint condition, comprehensively construct a bilateral joint risk control model considering the target recognition risk and the long-term tracking cost of the targets. Specifically, it includes:

[0009] S2.1: Calculate the recognition risk of the targets. Assume that there are types of targets. Use multi-sensors to measure the target state. At time, use the filter to estimate the state of the target type at the next moment, so as to obtain the estimated value of the target type state at time. Use multi-sensors to measure the state information of the targets. After obtaining the target state measurement at time, use the Bayesian formula to update the measurement of the target state estimated value.

[0010] When the estimated target type state does not match the true target type, it means that the target type is misjudged. At this time, the target will damage the measurement of the sensors. In order to more reasonably evaluate the loss value caused to the sensor measurement when the target type is misrecognized, define the loss matrix ; use the risk definition to calculate the target recognition risk;

[0011] S2.2: Calculate the threat risk of the targets. Define the target threat levels as three levels: 1, 2, and 3. Use the threat degree to form the threat degree level information matrix . Use the target speed information, distance information, and target aggressiveness information measured by the sensors to comprehensively evaluate the threat degree of the targets. Use the hidden Markov theory to establish a threat level transition matrix, and obtain the target threat level transition matrix at time and the observation matrix of the sensors for the target speed information, distance information, and type information at time; assume that the threat degree level information matrix of the targets at the th time is The threat degree belief state of the target at the moment estimated according to the above model 。 。

[0012] When the threat degree of the target itself is relatively high, it will cause greater damage to the measurement of the sensor. In order to reasonably evaluate the threat loss of the target, a cost matrix is defined 。Calculate the threat risk of the target using the risk definition.

[0013] S2.3: Calculate the long-term tracking cost of the target. PCRLB (Posterior Cramér-Rao Lower Bound) is the inverse matrix of the Fisher information matrix. The Fisher information matrix consists of two parts, the prior information matrix of the system state and the information gain matrix brought by measurement update. Taking the trace of the estimated PCRLB within the future time as the long-term tracking risk index of the target, it can be modeled in two aspects: one is the immediate time caused by the current decision, and the other is the cumulative impact of the current decision on the subsequent tracking process. The state vector to be estimated, the state transition matrix of the target, the noise variance matrix representing the target state, and the measurement noise variance matrix are used to obtain the long-term tracking risk cost of the target at the moment with a prediction time domain length of 。

[0014] S2.4: Considering the detection capabilities and tracking capabilities limitations of each radar, obtain the recognition risk of the target from S2.1, obtain the threat risk of the target from S2.2, obtain the long-term tracking risk cost of the target from S2.3, and then combine the comprehensive risks such as the long-term tracking cost, recognition, and threat to establish a bilateral joint risk control model.

[0015] S3: Use the improved artificial bee colony algorithm to solve the bilateral joint risk control model constructed in S2. In the population initialization process, use the maximum-minimum distance product method to replace the traditional initialization method, design a new fitness function, and add a global guiding factor to the position update formula to accelerate the convergence speed of the algorithm and improve the solution quality of the algorithm. Finally, obtain the allocation scheme of the multi-sensor long-term scheduling corresponding one by one to the solutions obtained by the algorithm.

[0016] Advantages of the present invention: The present invention combines the results of long-term prediction fusion in the method. For the long-term scheduling problem of bilateral joint risk control, it makes up for the situation that due to the variable number of targets, accurate risk assessment between targets cannot be achieved, which in turn affects the sensor measurement information. It enables the sensor to consider more risk factors it receives during decision-making, can effectively optimize its subsequent decision-making behavior, and can achieve stable and effective tracking of the target in an environment with a variable number of targets. It achieves the purpose of efficient and effective control of risk control in a dynamic target environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is the multi-sensor bilateral joint risk control model diagram of the present invention;

[0018] Figure 2 This is the flowchart of the method of the present invention;

[0019] Figure 3 This is the comparison effect diagram of the iteration times between the present invention and other optimization algorithms;

[0020] Figure 4 This is the comparison effect diagram of the risk values between the present invention and other optimization algorithms;

[0021] Figure 5 This is the comparison effect diagram of the position estimation errors under different prediction step lengths of the present invention. Specific Embodiments

[0022] The following further describes the specific embodiments of the present invention in conjunction with the accompanying drawings.

[0023] Although the following detailed description provides an in-depth understanding of the present invention, the present invention can be implemented in many different ways, which may be different from those described in the text. Professionals can make appropriate adjustments and changes according to the core principles of the present invention without departing from its essence. Therefore, the protection scope of the present invention should not be limited to the specific embodiments detailed in the text.

[0024] Specifically, it includes the following:

[0025] S1: Construct a scenario and obtain data: There are moving targets in the detection area. The targets make counterclockwise turning motions (turning rate is ) in the previous period of time, and make clockwise turning motions (turning rate is ) in the subsequent period of time. The other simulation time targets perform uniform linear motions. The receiving source is sensors. The measurement sampling period of the sensors is , the number of samples for Monte Carlo sampling is , and the simulation time is . The lower bound of the distance corresponding to the minimum measurement error variance is , the optimal signal-to-noise ratio corresponding to the shortest distance is , the path loss exponent is , the covariance matrix of the process noise is , represents the measurement noise variance of sensor based on the reference distance . Under angular measurement , under distance measurement , the initial values of the filtering and the filtering error covariance matrix are respectively , .

[0026] S2: Construct a multi-sensor bilateral joint risk control model, as shown in Figure 1 . Identify the occurrence of the target in the target of S1 to obtain the identification risk of the target, then calculate the threat risk of the existing target, and then obtain the long-term tracking cost of the target according to the long-term prediction. Finally, construct a bilateral joint risk control model with the target threat risk as the constraint condition, comprehensively considering the target identification risk and the target long-term tracking cost; specifically including:

[0027] S2.1: Target identification risk:

[0028] Assume that there are types of enemy incoming targets, then the state types of the target are:

[0029]

[0030] Among them, represents the probability that the target type is the th class.

[0031] Use multi-sensors to measure the target state. The measurement matrix of the target type by multi-sensors is:

[0032]

[0033] Among them, represents the true type of the target is while the observed type is with a probability.

[0034] At time, use the filter to estimate the state of the target type at the next moment, so as to obtain the time target type state estimate value.

[0035]

[0036] Among them, represents time the probability estimate value that the target type state is the th class.

[0037] Use multi-sensors to measure the state information of the target, that is, use the measurement matrix ,, after obtaining the time target state measurement, use the Bayesian formula to update the measurement of the target state estimate value, and each element is updated according to the observation value formula:

[0038]

[0039] Among them, represents the probability that the state estimate of the target type at time is the th class while the measured target type is the th class. represents the predicted value of the probability that the state estimate of the target type at time is the

[0040] When the state estimate of the target type does not match the true target type, it represents an error in target type judgment. At this time, the enemy target will damage the measurement of the sensor. In order to more reasonably evaluate the loss value caused to the sensor measurement when the target type recognition is incorrect, a loss matrix is defined as:

[0041] Among them, represents the loss caused when the true target type is while the estimated type is . When , that is, when the true target type is the same as the target estimated type, the loss is 0.

[0042] Using the risk definition (the recognition risk is obtained by multiplying the loss matrix by the probability of recognition error), the target recognition risk value is obtained:

[0043] Among them, represents the target recognition risk value of the estimated type of the target at time being the th class.

[0044] S2.2: Target threat risk

[0045] Use the Markov model to construct a target threat level model. The target threat level is defined as three levels: 1, 2, and 3, where 1 represents a low threat level, 2 represents a medium threat level, and 3 represents a high threat level. Use the threat degree to form a threat degree level information matrix . At time, the threat degree level information matrix of the target is:

[0046]

[0047] Among them, represents that at time, the threat degree of the target is Probability, .

[0048] Comprehensively evaluate the threat level of a target by using the target speed information, distance information, and target aggressiveness information measured by sensors , and use the Markov model to divide the speed information into 1 (low speed), 2 (medium speed), and 3 (high speed), the distance information into 1 (short distance), 2 (medium distance), and 3 (long distance), and the target aggressiveness information into 1 (weak), 2 (medium), and 3 (strong). Use the hidden Markov theory to establish a threat level transition matrix, Threat level transition matrix of the target at time

[0049]

[0050] At time, the observation matrix of the sensor for the target speed information, distance information, and type information:

[0051]

[0052] where represents the observed value. Assume that the threat level information matrix of the target at the th time is , and estimate the threat level information matrix of the target at the th time according to the threat level information matrix, threat level transition matrix, and observation matrix. The specific estimation steps are as follows.

[0053] (1) Threat prediction

[0054] At the th time, the estimated value of the threat level information state matrix of the target:

[0055]

[0056] where represents the estimated value of the threat level information matrix of the target at the th time obtained at the th time; , and , , are the observed values of the target speed, distance, and aggressiveness at the rd to th times respectively. Since the observed value at the th time cannot be obtained at the th time, the threat level of the target at the th time is predicted:

[0057]

[0058] Among them, represents the target obtained at the time, and the threat level information matrix at the

[0059] (2) Measurement update

[0060] Use multi-sensors to measure the motion information of the target. At the time, after obtaining the observed value of the target, according to Bayes' formula and the first-order Markov theory, measure and update the predicted matrix of the target threat level state at the time, and then obtain the target threat level information matrix at the time:

[0061]

[0062]

[0063] When the threat level of the target itself is relatively high, it will cause greater damage to the measurement of the sensor. In order to reasonably evaluate the target threat risk, using the risk definition, it is crucial to evaluate the threat loss value of the target. In order to reasonably evaluate the target threat loss, define the cost matrix :

[0064]

[0065] Among them, represents the loss caused when the true value of the threat level is and the estimated value is .

[0066] Use the risk definition (the threat risk is obtained by multiplying the cost matrix by the probability of the target threat level information) to obtain the target threat risk value:

[0067]

[0068] Among them, represents the risk value when the target threat level is the th class at the time.

[0069] S2.3: Calculate the long-term tracking cost of the target. PCRLB, as the inverse matrix of the Fisher information matrix, strictly defines the theoretical lower bound of the state estimation accuracy. Specifically, it includes the state vector to be estimated, represents the state transition matrix of the target, and respectively represent the noise variance matrix of the target state and the measurement noise variance matrix, is expressed as the Fisher information matrix, represents the Jacobian matrix of the observation function (i.e., the first-order partial derivative matrix of the measurement equation). Based on the calculation of the PCRLB theory, the complete expression of the Fisher information matrix can be obtained:

[0070]

[0071] Thus, the mathematical formula of the posterior Cramér-Rao lower bound PCRLB can be defined as:

[0072]

[0073]

[0074] where is defined as the trace of the PCRLB. Taking the trace of the estimated PCRLB in the future time as the target long-term tracking risk cost, the target long-term tracking cost can be expressed in two aspects: one is the immediate cost caused by the current decision, and the other is the cumulative impact of the current decision on the subsequent tracking process;

[0075]

[0076] In the formula, represents the target state estimation at time and is the prediction time domain length. represents the target long-term tracking risk cost at time

[0077] with a prediction time domain length of S2.4: Considering the detection capabilities and tracking ability limitations of each radar, the recognition risk of the target is obtained from S2.1, the threat risk of the target is obtained from S2.2, and the target long-term tracking cost is obtained from S2.3. Combining the long-term tracking cost, recognition, and threat and other comprehensive risks, a multi-sensor long-term scheduling model based on bilateral joint risk control is established. By optimizing the allocation of sensor resources to track the target to minimize the joint risk cost, the loss of multi-sensor measurement caused by target recognition errors can be reduced, and the threat risk of existing targets to sensors can be controlled through the multi-sensor scheduling scheme. Since active sensors will also radiate signals while detecting target information, the position exposure of the sensors is aggravated. In order to ensure the joint risk management of the battlefield while improving the concealment and security of its own detection system, the threat risk brought by the sensors tracking the target is restricted, and thus the multi-sensor management objective function is constructed as follows:

[0078]

[0079]

[0080]

[0081]

[0082]

[0083] Among them, is the combined risk cost index, which is calculated in step S2.3, and is calculated in step S2.1; and are the normalized weight coefficients of the target tracking cost and the identification risk index; indicates that a target can be tracked by at most one sensor at the same time; also, due to physical limitations, a sensor can track at most one target; is the radar detection constraint. When the target is within the detection range of the radar , otherwise , . is calculated in step S2.2 and is the target threat risk threshold.

[0084] S3: Use the improved artificial bee colony algorithm to solve the bilateral joint risk control model. In the population initialization process of step S3.1, the maximum minimum distance product method is used to replace the traditional initialization method. In the position update formula of step S3.2, a global guiding factor is added. In step S3.3, a new fitness function is designed. In step S3.4, a better solution of the search algorithm is searched to accelerate the convergence speed of the algorithm and improve the solution quality of the algorithm. Finally, the allocation scheme of the multi-sensor long-term scheduling is obtained one-to-one corresponding to the solution obtained by the algorithm.

[0085] S3.1: Initialization by the maximum minimum distance product method (initialization stage)

[0086] The setting of the initial population is an important parameter that affects the overall convergence rate and solution quality of the evolutionary algorithm. The traditional methods have disadvantages such as strong randomness and uneven initial populations. Therefore, the present invention proposes to initialize the bee colony by using the maximum minimum distance product method, which inherits the advantages of the maximum minimum distance method and the maximum distance product method and solves the disadvantages of the traditional methods.

[0087] represents the set of all data, is the number of the initial bee colony, is the number of the initial points, is the set of the initial points to be added, is for storage each element in and the array of the product results of each element in . The basic idea of this algorithm is to use the product result to select the area with dense points as the starting point, making the distribution of the starting points sparser. It not only reduces the dependence on parameters but also eliminates the disadvantages such as too dense or too close to the periphery of the initial points, greatly improving the quality of its initialization.

[0088] S3.2: Local Search of Employed Bees (Employed Bee Phase)

[0089] After the initialization phase in step S3.1, when using the traditional position update rule, the neighborhood search process is prone to problems of randomness and local minima. Therefore, the present invention proposes to correct the position update formula by combining a global guiding factor as follows:

[0090]

[0091] where is the newly generated position near the current solution , and are both the -th dimension of another solution randomly selected from the population, is a random number generated by a random formula. ; is the weight of the global guiding factor, is the current optimal solution.

[0092] After adding the global guiding factor , the search process becomes more directional. The global guiding factor can adjust the size of the search step according to the change in the distance between the current position and the optimal position, thereby improving the search speed and the ability to jump out of the local optimal solution. At the same time, the addition of the weight can make the search range more reasonable, achieving a better balance between global search and local exploitation.

[0093] S3.3: Selection Mechanism of Scout Bees (Scout Bee Phase) After the employed bee search phase in step S3.2, the scout bees select solutions according to the quality of the food sources (fitness value ). The calculation formula for the selection probability is:

[0094]

[0095] where is the population size, is the fitness value of the solution .

[0096] Among them, the fitness function determines the search direction and convergence of the algorithm. The present invention uses the iterative process of the artificial bee colony algorithm and the idea of K-means clustering to propose a new fitness function:

[0097]

[0098] Among them, represents the number of points in the th class, represents the sum of the distances from the objects within the th class to the center point . This function comprehensively considers the number of points and the within-class distance, avoiding the limitations brought by a single index.

[0099] For example, when the within-class distance is the same but the number of points is different, or the number of points is the same but the within-class distance is different, the traditional method may not be able to distinguish the quality, while the fitness function of the present invention can effectively distinguish the quality and guide the search direction of the algorithm. In the same situation, this function can also provide a more accurate evaluation, thus improving the search efficiency of the algorithm.

[0100] S3.4: Random exploration of scout bees (scout bee stage)

[0101] After the observation bee stage in step S3.3, when a certain solution has not been improved after multiple iterations, the scout bee will abandon this solution and randomly generate a new solution:

[0102]

[0103] Among them, and are respectively the upper and lower bounds of the th dimension of the solution space, and rand(0,1) is a random number within the range of [0,1].

[0104] The new solution finally obtained by the algorithm determines the decision action of the multi-sensor at the next moment, thereby further obtaining the allocation scheme of the final long-term multi-sensor scheduling.

[0105] The flowchart of the method of the present invention is as Figure 2 visible.

[0106] The present invention provides a multi-sensor long-term scheduling method based on bilateral joint risk control, which makes up for the situation that the accurate risk assessment between targets cannot be achieved due to the variable number of targets, thus affecting the sensor measurement information. By comprehensively considering the recognition risk of newly emerging targets and the long-term tracking cost of targets, a bilateral joint risk model is proposed to minimize the joint risk, a target function under multi-sensor management is formed, the multi-sensor radiation risk is constrained, and an improved artificial bee colony algorithm is used for optimization and solution, thereby completing the optimization of the multi-sensor allocation scheme. The method can realize the safety of sensors and can stably and effectively track targets in an environment with variable numbers of targets.

[0107] The comparison effect diagram of the number of iterations of the present invention and other optimization algorithms is as Figure 3 shown. In order to better verify the performance of the algorithm given by the present invention, at moments, the basic artificial bee colony algorithm, the improved artificial bee colony algorithm, the particle swarm algorithm, and the genetic algorithm are used to solve the multi-sensor - multi-target allocation scheme for the target tracking task, and a large number of simulation experiments are carried out. During these processes, Monte Carlo experiments are performed, and the average value is taken. The experimental results show that the algorithm has the characteristics of fast convergence speed, good solution quality, and strong solution robustness. For complex optimization problems, the algorithm can be used as an effective means.

[0108] The comparison effect diagram of the risk values of the present invention and other optimization algorithms is as Figure 4 shown. During the entire simulation process, the risk value of the improved artificial bee colony algorithm is lower than that of the other three algorithms. During the simulation process, the joint risk value of the improved artificial bee colony algorithm is reduced by 7.22% compared with the original artificial bee colony algorithm, 8.58% lower than the heuristic algorithm, and 4.03% lower than the genetic algorithm. The heuristic algorithm performs multi-sensor resource scheduling based on distance and does not consider the continuous change of risk values, so it cannot effectively manage the changing risk values. Therefore, compared with the other three algorithms, it has the greatest risk and cannot ensure effective control of the risk values. When the original artificial bee colony algorithm solves high-dimensional problems, the convergence speed is relatively slow; it is easy to fall into local optimal solutions when it is complex. The genetic algorithm is relatively effective in risk control during the entire simulation process. Compared with the other three algorithms, during the entire simulation process, the total risk value of the improved artificial bee colony algorithm is the lowest, indicating that the risk value control of the algorithm proposed in this chapter is relatively effective. During the entire simulation process, it can effectively balance the radiation risks of multiple sensors and the target tracking error, better manage the joint risk, and effectively track the target.

[0109] The comparison effect diagram of the position estimation error of the present invention under different prediction steps is as Figure 5As shown, it can be seen that the long-term scheduling method proposed by the present invention has the optimal performance in multi-sensor multi-target cooperative positioning and tracking, while the positioning and tracking performance of the sensors under the myopic scheduling strategy is slightly worse. Under the method of the present invention, as the prediction step length increases (from 1 to 15), the positioning error of the sensors for the target decreases. Among them, the positioning and tracking performance of the target within the range of the prediction step length from 2 to 10 first improves and then deteriorates. However, when the prediction step length is greater than 10, as the prediction step length increases, the positioning and tracking error of the sensors for the target increases. This is because the fuzzy maneuverability of the target itself causes the system to be unable to accurately predict the future state of the target, and a longer prediction step length instead makes the performance of target positioning and tracking worse.

Claims

1. A multi-sensor forward scheduling method based on bilateral joint risk control, characterized in that: The steps include: S1: Build a multi-sensor multi-target scenario and obtain data related to sensors and targets; S2: Based on the data related to sensors and targets, the bilateral joint risk control model is constructed by integrating the target identification risk and the target long-term tracking cost; S3: The improved artificial bee colony algorithm is used to solve the bilateral joint risk control model, and the obtained solutions correspond one-to-one to the allocation scheme of multi-sensor forward scheduling.

2. The multi-sensor forward scheduling method based on bilateral joint risk control according to claim 1 is characterized in that: The sensor-related data in step S1 include the sampling period of the sensor, the number of sensors, the lower bound of the distance corresponding to the minimum measurement error variance of the sensor and the optimal signal-to-noise ratio, the path loss index of the sensor, the covariance matrix of the process noise of the sensor, and the measurement noise variance of the sensor; the target-related data include the number of targets, the turning rate of the target, and the motion model of the target; scene information: the number of samples of Monte Carlo sampling, the simulation time, the initial value of the filter, and the initial value of the filter error covariance matrix.

3. The multi-sensor forward scheduling method based on bilateral joint risk control according to claim 2 is characterized in that: The construction of the bilateral joint risk control model is specifically as follows: the appearance of new targets in the targets of S1 is calculated to obtain the identification risk of the new targets, the threat risk of the existing targets is calculated, and then the long-term tracking cost of the targets is obtained based on long-term predictions. Finally, the bilateral joint risk control model is constructed by taking the target threat risk as a constraint condition and comprehensively considering the target identification risk and the target long-term tracking cost.

4. The multi-sensor forward scheduling method based on bilateral joint risk control according to claim 3 is characterized in that: The specific implementation process of step S2 is as follows: S2.1: Use multiple sensors to measure the target state. The filter is used to estimate the state of the target type at the next moment, and the The target type state estimation value at the moment; using multiple sensors to measure the state information of the target, we can get After measuring the target state at a certain moment, the target state estimate is updated using the Bayesian formula; the loss matrix is ​​defined as , calculate the target identification risk value; S2.2: Define target threat level and use threat level Threat Level Information Matrix ; The target's threat level is comprehensively assessed using the target speed information, distance information, and target aggressiveness information measured by the sensor. , using hidden Markov theory to establish the threat level transfer matrix, we get The target threat level transfer matrix at each moment and At this moment, the sensor's observation matrix of target speed information, distance information, and type information is obtained; according to the threat level information matrix, threat level transfer matrix, and observation matrix estimation, we can get The threat level information matrix of the target at each moment; define the cost matrix , where the elements Indicates that when the true value of the threat level is The estimated value is The loss caused by using the cost matrix Elements and The corresponding elements in the target threat level information matrix at each moment are multiplied and summed to obtain the target threat risk value; S2.3: The posterior Cramer-Rao lower bound PCRLB is taken as the inverse matrix of the Fisher information matrix, and the trace of the estimated PCRLB in the future is taken as the target long-term tracking risk indicator, which is modeled as a two-component superposition structure: the first is the immediate cost term caused by the current decision, and the second is the cumulative impact term of the current decision on the subsequent tracking process; The state vector to be estimated, the state transfer matrix of the target, the noise variance matrix representing the target state, and the measurement noise variance matrix are obtained through the posterior Cramer-Rao lower bound PCRLB. At time, the prediction time domain length is The goal is to track the risk cost in the long term; S2.4: Considering the detection and tracking capability limitations of each radar, a bilateral joint risk control model is established based on the target identification risk, target threat risk and target long-term tracking risk costs.

5. The multi-sensor forward scheduling method based on bilateral joint risk control according to claim 4 is characterized in that: The specific implementation process of step S2.1 is as follows: Set the status type to , where the elements Indicates that the target type is The probability of the class is obtained by using multiple sensors to measure the target state and obtain the measurement matrix of the target type by multiple sensors , where the elements The actual type of the target is The observation type is The probability of exist At this moment, the filter is used to estimate the state of the target type at the next moment, and the The target type state estimate at the moment; using the measurement matrix , in getting After measuring the target state at a certain moment, the target state estimate is updated using the Bayesian formula. When the target type state estimation does not match the actual target type, it means that the target type is misjudged. At this time, the enemy target damages the sensor's measurement. The loss value caused to the sensor measurement is defined as the loss matrix , where the elements Indicates that when the target real type is The estimated type is The loss caused by When , that is, when the target true type is the same as the target estimated type, the loss is 0; Using the loss matrix , and obtain the target identification risk value: , Representative The estimation type of the moment target is The target identification risk value of the class, represent The target type status is The probability estimate of the class.

6. The multi-sensor forward scheduling method based on bilateral joint risk control according to claim 5 is characterized in that: The step S2.4 is specifically implemented as follows: Based on the target identification risk, target threat risk, target long-term tracking cost, combined long-term tracking cost, identification and threat and other comprehensive risks, a multi-sensor long-term scheduling model based on bilateral joint risk control is established. By optimizing the allocation of sensor resources to track targets, the joint risk cost is minimized, the loss of target identification errors to multi-sensor measurements is reduced, and the threat risk of existing targets to sensors is controlled through multi-sensor scheduling schemes. The threat risk brought by sensor tracking targets is constrained, and the multi-sensor management objective function is constructed: , ,in, is the joint risk cost indicator, For the The time domain length of the moment prediction is The goal is to track the risk cost in the long term. for The estimation type of the moment target is The target identification risk value of the class; and Normalized weight coefficients for target tracking cost and identification risk indicators; It means that a target can be tracked by at most one sensor at a time; For radar detection constraints, when the target On the radar The detection range ,otherwise , For the The target threat risk value at all times, is the target threat risk threshold.

7. The multi-sensor forward scheduling method based on bilateral joint risk control according to claim 6 is characterized in that: The step S3 is implemented as follows: in the population initialization process, the maximum minimum distance product method is used to replace the traditional initialization method, and a new fitness function is designed, and a global guidance factor is added to the position update formula. The final solution is a one-to-one correspondence to obtain a multi-sensor long-term scheduling allocation scheme, which is as follows: S3.1: Initialize the swarm using the method of maximizing the minimum distance product; S3.2: Employed bee search phase: After the initialization phase, a modified position update formula combining the global guidance factor is proposed: ,in, For the current solution Newly generated locations nearby, and are all other solutions randomly selected from the population. dimension, ; is the weight of the global guidance factor, is the current optimal solution; S3.3: Observation bee stage: After the employed bee search stage, the selection probability The calculation formula is: ,in, is the population size, The fitness value is , Indicates The number of points in the class, Indicates Objects within a class to the center point The sum of the distances; S3.4: Scouting bee stage: After the observation bee stage, when a solution is not improved after multiple iterations, the scout bee abandons the solution and randomly generates a new solution, and finally determines the decision action of the multi-sensor at the next moment and obtains the final multi-sensor scheduling solution.

Images