Multi-dimensional radar resource scheduling method combining anti-interference and tracking under low interception probability
By adopting improved artificial bee colony algorithm and interactive multi-model-extended Kalman filtering in multi-dimensional radar resource scheduling, optimizing radar resource parameters, the problem of degradation of maneuver target tracking performance in the existing technology is solved, and the low interception probability is achieved Efficient tracking and anti-interference performance improvements.
Patent Information
- Application Number
- CN202410829112.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-25
- Publication Date
- 2024-11-08
AI Technical Summary
The prior art is difficult to directly apply to maneuver target tracking scenarios in scenarios restricted by interference, resulting in the problem of missing target tracks and degraded target state estimation performance.
Using a multi-dimensional radar resource scheduling method with combined anti-jamming and tracking at low interception probability, using the improved artificial bee colony algorithm and interactive multi-model-Expanded Carl Man filtering optimizes radar carrier frequency, bandwidth, dwell time, radiated power and frequency modulation slope to form the optimal CRMRM strategy.
Under the conditions of low interception probability, the RF resource parameters of multi-dimensional radar are optimized, the tracking performance of maneuverable targets is improved, and noise suppression interference is reduced, which solves the problems of target track loss and reduced state estimation performance, and improves the anti-interference of radar in complex environments ability.
Smart Images

Figure SMS_54 
Figure SMS_56 
Figure SMS_60
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of radar waveform optimization, and in particular to a multi-dimensional radar resource scheduling method for joint anti-interference and tracking under low intercept probability. Background Art
[0002] Due to the non-cooperative nature of target motion and the uncertainty of target source measurement, radar tracking of maneuvering targets is complex and difficult. With the development of electronic countermeasures technology and electromagnetic spectrum warfare, active electronic interference (such as noise suppression interference) in complex electromagnetic interference environments seriously affects radar performance. In addition to passive reception anti-interference and improved tracking filter algorithms, inspired by the closed-loop learning self-evolution cognitive radar architecture, researchers are concerned about how to use real-time environmental perception information and multi-scale prior knowledge (such as echo history data, target and electronic interference characteristics) to intelligently optimize transmission strategies and RF resource configuration to actively resist interference, so as to improve the performance of maneuvering target tracking in radar-interference source coexistence scenarios.
[0003] The active anti-interference cognitive transmission of the radar transmitter refers to the use of prior knowledge of radar targets and electronic interference, historical data of radar echoes, and real-time cognitive information of the battlefield environment to optimize the transmission strategy and RF resource configuration, and improve the anti-reconnaissance and anti-interference performance in complex electromagnetic confrontation environments. It can be divided into anti-interference waveform design and anti-interference resource parameter scheduling. According to different interference styles, the anti-interference waveform design can be divided into anti-suppression interference, anti-deception interference, and anti-smart interference. According to different anti-interference strategies, it can be divided into anti-interference waveform optimization based on avoidance strategy and anti-interference waveform optimization based on notch strategy. Based on the OODA (Observation-Orientation-Decision-Action, OODA) closed loop formed in a dynamic confrontation environment, the anti-interference RF resource parameter scheduling generates a more powerful anti-interference capability through the collaboration of multiple radar signal domains. It is also a research focus. It can be divided into waveform parameter intelligent scheduling, power resource intelligent scheduling, spectrum resource intelligent scheduling, etc., and is solved by using game theory, multi-armed bandit, reinforcement learning, and meta-learning methods.
[0004] Mathematically speaking, resource management for maneuvering target tracking can be formulated as a constrained optimization problem, which balances the contradiction between target tracking performance and limited system resources. It can be roughly divided into two resource management schemes, including maximizing target tracking performance under certain resource constraints, and minimizing system resource loss under the premise of meeting the predetermined target tracking performance requirements. The system resource elements involved include time-sensitive resources (including frequency band, bandwidth, waveform, transmission power, dwell time, directional pattern, etc.) and non-time-sensitive resources (including organizational link network architecture, network nodes, and trajectory planning).
[0005] However, both the existing resource management methods for maneuvering target tracking and the resource scheduling methods for anti-interference are difficult to be directly applied to maneuvering target tracking scenarios under interference-restricted scenarios, which leads to the problems of target track loss and reduced target state estimation performance. Summary of the invention
[0006] The purpose of the present invention is to propose a multi-dimensional radar resource scheduling method for joint anti-interference and tracking under low intercept probability in order to solve the problem that the existing scheduling method is difficult to directly apply to the maneuvering target tracking scenario under interference-restricted scenario, which leads to the loss of target track and the degradation of target state estimation performance.
[0007] The technical solution adopted by the present invention to solve the above technical problems is:
[0008] The multi-dimensional radar resource scheduling method for joint anti-interference and tracking under low probability of interception includes the following steps:
[0009] Randomly initialize the strategy function and the state value function v(b,w), where β represents the randomly initialized radar band occupation action, b represents the randomly initialized radar confidence state, represents the randomly initialized policy function parameters, w represents the randomly initialized state value function parameters, and the randomly initialized policy learning rate and the state value function learning rate α w ,in, Perform the following steps in sequence for time step k, k = 1, ..., N;
[0010] Step 1: Obtain radar echo data and use the radar echo data to construct the radar confidence state b at time step k-1 k-1 , the radar confidence state includes a target state and a radio frequency interference state;
[0011] Step 2: Radar confidence state b based on time step k-1 k-1 , using the policy function of time step k-1 Generate radar sub-band selection action And select actions for radar sub-bands Solve and get the optimal radar carrier frequency and optimal radar bandwidth
[0012] Step 3: Use the radar intercept probability threshold p th and radar confidence state b k-1 , respectively, at the maximum radar dwell time Radar radiated power under and at the minimum radar dwell time Radar radiated power under Among them, R k|k-1 represents the predicted distance between the radar and the target at time k-1;
[0013] Step 4: Based on the upper limit of radar radiated power and lower limit contrast and The smallest value is recorded as the radar radiation power that meets the RF resource constraint condition.
[0014] Step 5: Compare the maximum signal-to-interference-noise ratio criterion and Select the group with the largest value, which is the optimal radar dwell time. and optimal radiated power
[0015] Step 6: Action based on radar sub-band occupancy Generate real-time spectrum constraint function And the spectrum constraint function As the real-time spectrum constraint condition for waveform parameter optimization, w type Indicates the type of radar waveform, λ k represents the Gaussian pulse width, κ k Indicates the frequency modulation rate;
[0016] Step 7: Radar confidence state b based on time step k-1 k-1 Constructing a conditional predictive Bayesian risk lower bound for the MMSE criterion
[0017] Step 8: Generate resource constraint function R based on resource constraints Θ (λ k ,κ k w type );
[0018] Step 9: Predicting Bayesian risk based on conditions under the MMSE criterion Resource constraint function R Θ (λ k ,κ k w type ) and spectral constraint function Construct constrained optimization problems;
[0019] Step 10: Solve the optimization problem in step 9 and obtain the optimal Gaussian pulse width and the optimal FM slope
[0020] Step 11: Based on the optimal radar carrier frequency Optimal radar bandwidth Optimal radar dwell time Optimum radiated power Optimal Gaussian pulse width And the optimal frequency modulation slope Get the echo signal and obtain the spectrum measurement h in the echo signal k ;
[0021] Step 12: Generate overall target state estimation based on interactive multi-model-extended Kalman filter and the overall target state estimation error covariance P k|k , and the spectrum of the echo signal is measured by the classical energy detection method k Transformed into interference state estimate
[0022] Step 13: Estimation based on overall target state The overall target state estimation error covariance P k|k and interference state estimation Generate radar confidence status That is, the radar confidence state b k ;
[0023] Step 14: Based on radar confidence state b k-1 , radar confidence state b k and radar sub-band occupancy actions Generate reward function
[0024] Step 15: Based on the reward function and the state value function v(b k ,w k-1 ), construct the timing difference error
[0025] Step 16: Based on the timing difference error and the policy function Get the policy function parameters
[0026] Step 17: Based on the timing difference error and the state value function v(b k-1 ,w k-1 ), get the state value function parameter w k ;
[0027] Step 18: Based on the policy function parameters and the state value function parameter w k , construct the function and the state value function v(b k ,w k), and the policy function and the state value function v(b k ,w k ), repeat the above steps until convergence, get the optimal strategy function, and use the optimal strategy function to get the optimal CRMRM strategy.
[0028] Furthermore, the optimal radar carrier frequency It is expressed as:
[0029]
[0030] The optimal radar bandwidth It is expressed as:
[0031]
[0032] in, represents the binary indicator function, f c represents the radar center frequency, ΔB represents the frequency band width of a sub-band, Indicates radar sub-band occupation action β k The i-th element of , N represents the number of sub-bands into which the radio frequency band is equally divided.
[0033] Furthermore, the radar radiation power It is expressed as:
[0034]
[0035] The radar radiated power It is expressed as:
[0036]
[0037] The radar radiated power It is expressed as:
[0038]
[0039] Among them, p' fa represents the preset false alarm probability of the intercept receiver, erfc(·) represents the complementary error function, T I represents the search time of the intercept receiver, k0 represents the Boltzmann constant, T0 represents the effective noise temperature of the radar, B I represents the intercept receiver bandwidth, F I represents the noise factor of the intercept receiver, G t represents the radar transmitting antenna gain, G r represents the radar receiving antenna gain, G IP represents the radar receiver processing gain, λ t Indicates the radar wavelength.
[0040] Furthermore, the optimal radar dwell time and optimal radiated power It is expressed as:
[0041]
[0042] The spectrum constraint function It is expressed as:
[0043]
[0044] in, Indicates frequency band occupation action β k The corresponding bandwidth, B θ (λ k ,κ k |w type ) indicates that under the linear frequency modulation waveform, the Gaussian pulse width λ k and the frequency modulation rate κ k Determines the signal bandwidth.
[0045] Furthermore, the Bayesian risk lower bound It is expressed as:
[0046]
[0047] in, and denote the target state transition matrix at time step k-1 and the noise covariance matrix at time step k-1 respectively, M denotes the number of target tracking models, represents the target mixed state error covariance at time step k-1, expressed as:
[0048]
[0049] in, represents the mixed state estimate at time step k-1, represents the probability of the mixed model at time step k-1, Γ mi represents the model transition probability from model m to model i, and They represent the target state estimation corresponding to the target tracking model j at time step k-1 and the target state covariance matrix at time step k-1, μ k-1|k-1 (j) represents the model validity probability of tracking model j at time step k-1, Γ ji represents the model transition probability from model j to model i, H k|k-1 represents the Jacobian matrix of the radar measurement function h(·) at time step k-1 with respect to the target state prediction, represents the second-order derivative operator, represents the target state at time step k predicted at time step k-1, It is expressed as:
[0050]
[0051] in, represents the measurement error covariance matrix, c represents the speed of light, SINR k (β k ,P t,k ,T d,k ) is the signal to interference noise ratio at time k, expressed as:
[0052]
[0053] Among them, P t,k is the radar radiation power, T d,k is the radar dwell time, is the radial distance between the radar and the target, The radio frequency interference state c k The i-th element of is the radar sub-band occupation action β k The i-th element of T r represents the radar pulse repetition interval, G RP represents the radar receiver processing gain, σ represents the radar cross-section of the target, L represents the generalized loss factor, P I It represents the interference signal power from the same frequency suppression noise interference source calculated by the interference equation;
[0054] The resource constraint function R Θ (λ k ,κ k w type ) is expressed as;
[0055]
[0056] in, and Respectively represent the radar waveform type w type The corresponding Gaussian pulse width λ k The minimum and maximum values of and Respectively represent the radar waveform type w type The corresponding minimum and maximum values of the frequency modulation slope.
[0057] Furthermore, the constrained optimization problem is expressed as:
[0058]
[0059] in, represents the optimal value of the radar Gaussian pulse width, Indicates the optimal value of the radar frequency modulation slope.
[0060] Furthermore, in step 10, solving the optimization problem in step 9 is performed by improving the artificial bee colony algorithm, and the specific steps for solving the optimization problem in step 9 are:
[0061] Step 1: Initialize the number SN of employed bees, follower bees, and observer bees, and randomly generate the initial nectar source within the range of variable values. The random initialization process is expressed as:
[0062]
[0063] Among them, x i,d represents the i-th nectar source X i The j-th dimension value, X i ={x i,1 ,x i,2 ,...,x i,D}, i∈{1,2,...,SN}, SN represents the population size, d∈{1,2,...,D}, D represents the dimension of the problem to be solved, that is, the number of variables to be optimized, and Respectively represent x i,d The upper and lower bounds of the value,
[0064] Step 2: Calculate the fitness of the nectar source and update the nectar source location, expressed as:
[0065]
[0066] Among them, x i Indicates the current nectar source location, x k , represents the randomly selected neighborhood honey source location, where k∈{1,2,...,SN}, k≠i, is a random number in [-1,1], v i,j Represents the new nectar source V generated by employed bees in each cycle i = {v i,1 ,v i,2 ,...,v i,D}, R i,j It represents the optimization function of nectar source location selection, which is expressed as:
[0067]
[0068] Among them, a, b are constants, f i represents the objective function value corresponding to the i-th employed bee, f krepresents the objective function value corresponding to the kth employed bee. The objective function f is the conditional risk Bayesian lower bound. Fitness function fit i It is expressed as:
[0069]
[0070] Then, calculate the new nectar source V i = {v i,1 ,v i,2 ,...,v i,D}, compared with the new nectar source V i = {v i,1 ,v i,2 ,...,v i,D} and Old Honey Source X i ={x i,1 ,x i,2 ,...,x i,D}, compare the fitness of the new nectar source with the old nectar source, and retain the nectar source with greater fitness as the new nectar source;
[0071] Step 3: Based on the fitness of employed bees in selecting nectar sources, the observer bees calculate the probability of selecting the i-th nectar source Among them, fit i is the fitness function corresponding to the i-th employed bee, i∈{1,2,...,SN}, fit q is the fitness function corresponding to the qth employed bee, q∈{1,2,...,SN}. Based on the above nectar source selection probability, the observation bee uses the roulette algorithm to select the nectar source v selected by the SN employed bees. i,d The optimal v i,d , let the optimal v i,d Equal to x i,d , based on x i,d Substitute into the formula Update the location of the honey source to generate a new honey source v i,d ,based on Calculate the fitness value and compare the new v i,d and x i,d The fitness function is used to determine the new v i,d Is it better than x? i,d , if the new v i,d Better than x i,d , then use the new v i,d Replace x i,d Otherwise, keep x i,d , thus obtaining SN updated solution vectors V i = {v i,1 ,v i,2 ,...,v i,d ,...,vi,D};
[0072] Step 4: If the solution vectors of both the employed bee stage and the observer bee stage are updated after two cycles, and α has not reached the maximum number of cycles LIMIT, the scout bee uses the formula Randomly generate a new nectar source x i,d To replace the honey source v that has not been updated for a long time i,d ;
[0073] Step 5: Determine whether the result converges or reaches the maximum number of cycles LIMIT. If so, the iteration ends and the optimal solution is output, which is the optimal Gaussian pulse width. and the optimal FM slope Otherwise, let α=α+1 and enter the hired bee stage to continue the iterative cycle.
[0074] Furthermore, the interactive multi-model-extended Kalman filter generates an overall target state estimate and the overall target state estimation error covariance P k|k , specifically including the following steps:
[0075] Step A: Based on the innovation covariance Kalman Gain and model probability And according to the classic Kalman filter algorithm, the target state estimation corresponding to each tracking model i is calculated and the target state estimation error covariance It is expressed as:
[0076]
[0077] Among them, z k are the radar measurements of the target’s radial position and radial velocity, represents the Jacobian matrix of the radar measurement function h(·) with respect to the radar target, represents the target state transition matrix of the radar at time step k-1;
[0078] Step B: Calculate the model likelihood probability corresponding to each tracking model i and the conditional model probability It is expressed as:
[0079]
[0080] Step C: Target state estimation by fusing each tracking model and the target state estimation error covariance Based on the conditional model probability Generate an overall target state estimate and the overall target state estimation error covariance P k|k , expressed as:
[0081]
[0082] Among them, E{x k |z k} represents the radar target state x k Given the radar measurement z k , E{·} represents the operator for finding the mathematical expectation;
[0083] Step D: Use the classical energy detection method to measure the spectrum h k Transformed into interference state estimate Finally, based on the overall target state estimation The overall target state estimation error covariance P k|k and interference state estimation Generate radar confidence status
[0084] Furthermore, the reward function It is expressed as:
[0085]
[0086] Among them, α1 and α2 represent user-defined weight factors.
[0087] Furthermore, the specific steps of step seventeen are:
[0088] Based on the reward function and the state value function v(b k ,w k-1 ), construct the timing difference error It is expressed as:
[0089]
[0090] Where γ represents the discount factor;
[0091] Based on the timing difference error and the policy function Update the policy function parameters It is expressed as:
[0092]
[0093] in, represents the learning rate of radar frequency band occupancy strategy update, Indicates the radar frequency band occupation strategy About parameters The gradient of
[0094] Based on the timing difference error and the state value function v(b k-1 ,w k-1 ), update the state value function parameter w k , expressed as:
[0095]
[0096] Among them, α w represents the learning rate of the state value function update, It means to find the state value function v(b k-1 ,w k-1 ) is the gradient of w with respect to the parameter w.
[0097] The beneficial effects of the present invention are:
[0098] Under the condition of low probability of interception, the proposed CRMRM strategy can jointly optimize the multi-dimensional radar RF resource parameters (including radar carrier frequency, bandwidth, dwell time, radiation power, Gaussian pulse width and frequency modulation slope), improve the maneuvering target tracking performance, that is, solve the problem of target track loss and target state estimation performance degradation in the existing technology. At the same time, it reduces the noise suppression interference caused by other in-band noise suppression interference sources.
[0099] The traditional single signal domain resource optimization strategy for a specific single task is difficult to cope with the increasingly complex radar detection environment. By leveraging the reusable information and coupling relationship between multi-stage tasks, the proposed CRMRM strategy is a multi-dimensional RF resource integrated scheduling technology for joint multi-tasks, which improves the linkage multi-task processing capability by considering multi-dimensional signal domain collaborative optimization.
[0100] In order to solve the underlying real-time nonlinear, high-dimensional, non-convex, multi-variable and multi-constrained optimization problem, the proposed CRMRM method is a multi-stage iterative optimization algorithm driven by multi-scale prior knowledge. The original CRMRM strategy is decomposed into two sub-strategies and solved in sequence, thereby deconstructing the solution process of discrete variables (i.e., radar sub-band occupancy action) and continuous variables (i.e., dwell time, radiated power, pulse width, and frequency modulation slope). First, based on the offline prior knowledge obtained from real-time training, the temporal difference speaker critic (TDAC) algorithm can well cope with the lagged rewards and future multi-step effects in the optimization solution of the band occupancy sub-strategy through the continuous interaction of strategy update and value function update based on the temporal difference error. Subsequently, in order to avoid falling into the local optimal solution by directly using the optimization algorithm when solving the waveform optimization strategy, based on the two proposed theorems, this application uses an analytical method to jointly optimize the radar average radiated power and dwell time, reducing the variable dimension and computational complexity in subsequent optimization. Subsequently, through iteration and interaction between swarms, the improved artificial bee colony (IABC) algorithm balances global search and local optimization and finally converges to the optimal pulse width and frequency modulation slope, while satisfying the waveform parameter constraints and the real-time spectrum constraints generated by the radar sub-band occupancy action. DETAILED DESCRIPTION
[0101] It should be particularly noted that, in the absence of conflict, the various embodiments disclosed in this application can be combined with each other.
[0102] Specific implementation method 1: Specifically describe this implementation method. The multi-dimensional radar resource scheduling method for joint anti-interference and tracking under low intercept probability described in this implementation method includes the following steps:
[0103] Randomly initialize the policy function and state value function, randomly initialize the policy learning rate and the state value function learning rate α w ,in Initialize the number of employed bees and follower bees SN, the number of cycles α = 1, the number of iteration terminations (of the improved artificial bee colony algorithm) LIMIT, and randomly initialize the honey source within the variable value range. Based on the radar waveform type and RF resource parameters, a radar RF resource library is constructed, where the waveform type w type For the specified linear frequency modulation signal, the types of RF resource parameters involved include carrier frequency, bandwidth, dwell time, radiated power, Gaussian pulse width and frequency modulation slope. The upper and lower limits of the radar dwell time are set as and The upper and lower limits of radar radiation power are set as and The upper and lower limits of the Gaussian pulse width are set as and The upper and lower limits of the frequency modulation slope are set as and Set the radar intercept probability threshold p th .
[0104] Randomly initialize the strategy function and the state value function v(b,w), where β represents the randomly initialized radar band occupation action, b represents the randomly initialized radar confidence state, represents the randomly initialized policy function parameters, w represents the randomly initialized state value function parameters, and the randomly initialized policy learning rate and the state value function learning rate α w ,in, The following steps are performed sequentially for time step k, k=1,...,N; hereinafter, the subscript k represents time step k, and the subscript k-1 represents the previous time step, i.e., time step k-1.
[0105] Step 1: Obtain radar echo processing data, and construct the radar confidence state b at time step k-1 based on the received processing results of the radar echo data. k-1 ; The radar confidence state includes the target state and the radio frequency interference state, that is, in represents the interference state estimate of the RF interference, represents the target state estimate at time step k-1, P k-1|k-1 represents the target state error covariance at time step k-1, Represents the normal distribution identifier.
[0106] Step 2: For radar confidence state b k-1 , based on the policy function at time step k-1 Generate radar sub-band occupancy action Right now And select actions based on radar sub-band Calculate the optimal radar carrier frequency and optimal radar bandwidth Expressed as
[0107]
[0108] in represents the binary indicator function, f c Indicates the radar center frequency.
[0109] Step 3: Optimization of joint radar dwell time and radiated power based on analytical method, including the following steps:
[0110] (1) Based on the radar intercept probability threshold p th and radar confidence state b k-1 , calculated at the maximum radar dwell time Radar radiated power under It is expressed as:
[0111]
[0112] where p' fa represents the preset false alarm probability of the intercept receiver, erfc(·) represents the complementary error function, T I represents the search time of the intercept receiver, k0 is the Boltzmann constant, T0 is the effective noise temperature of the radar, B I To capture the receiver bandwidth, F I represents the noise factor of the intercept receiver, G t is the radar transmitting antenna gain, G r is the radar receiving antenna gain, G IP is the radar receiver processing gain, λ t is the radar wavelength, R k|k-1 represents the predicted distance between the radar and the target at time step k-1.
[0113] (2) Based on the radar intercept probability threshold p th and radar confidence state b k-1 , calculate the minimum radar dwell time Radar radiated power under It is expressed as:
[0114]
[0115] (3) To avoid Exceeding the preset radar radiation power limit Calculate the radar radiation power that meets the RF resource constraints at this time It is expressed as:
[0116]
[0117] (4) Since the radar receiving signal-to-noise ratio is the radar dwell time T d,k and radiated power P t,k Under the maximum signal-to-interference-noise ratio criterion, compare and The larger of the two sets of products is selected as the optimized radar dwell time. and radiated power It is expressed as:
[0118]
[0119] Step 4: Joint Gaussian pulse width and frequency modulation slope optimization based on artificial bee colony algorithm. The specific steps are as follows:
[0120] (1) Action based on radar sub-band occupancy Generate real-time spectrum constraint function And the spectrum constraint function As a real-time spectrum constraint for waveform parameter optimization, It is expressed as:
[0121]
[0122] Where ΔB represents the bandwidth of a sub-band, and the frequency band occupancy action β k The corresponding bandwidth can be expressed as In the case of linear frequency modulation waveform, the Gaussian pulse width λ k , frequency modulation rate κ k Determined signal bandwidth B θ (λ k ,κ k |w type ) can be expressed as B θ (λ k ,κ k w type )=λ k ×κ k .
[0123] (2) Based on radar confidence state b k-1 , construct the conditional prediction Bayesian risk lower bound of the MMSE criterion It is expressed as:
[0124]
[0125] in and denote the target state transition matrix at time step k-1 and the noise covariance matrix at time step k-1 respectively, M denotes the number of target tracking models, represents the target mixed state error covariance at time step k-1, calculated as
[0126]
[0127] in and They represent the target state estimation corresponding to the target tracking model j at time step k-1 and the target state covariance matrix at time step k-1, μ k-1|k-1 (j) is the model validity probability of tracking model j at time step k-1, Γ ji represents the model transition probability from model j to model i. Represents the Jacobian matrix of the measurement function at time step k-1 with respect to the target state prediction, where It can be expressed as:
[0128]
[0129] in represents the measurement error covariance matrix, which is calculated as
[0130]
[0131] The signal to noise ratio at time k can be expressed as:
[0132]
[0133] Where T r Represents the radar pulse repetition interval (PRI), G RP represents the radar receiver processing gain, σ represents the radar cross section (RCS) of the target, k0 represents the Boltzmann constant, T0 represents the effective noise temperature of the radar, L represents the generalized loss factor, P I It represents the interference signal power from the same frequency suppression noise interference source calculated by the interference equation.
[0134] (3) Generate resource constraint function R based on specific resource constraints Θ (λ k ,κ k |w type ), R Θ (λ k ,κ k |w type ) is expressed as:
[0135]
[0136] (4) Bayesian risk prediction based on the MMSE criterion Resource constraint function R Θ (λ k ,κ k |w type ) and the spectrum constraint function C B (λ k ,κ k β k ,w type ), construct a constrained optimization problem, which can be expressed as:
[0137]
[0138] Step 5: Use the improved artificial bee colony (IABC) algorithm to solve the above constrained optimization problem. In the iterative process, each honey source represents a potential feasible solution for the Gaussian pulse width and frequency modulation slope. According to the positions of the global optimal particle and the individual optimal particle, the speed and position of the particle are continuously adjusted. The number of cycles α = 1, and the maximum number of iterations is LIMIT. The specific steps of the IABC algorithm are as follows:
[0139] (1) Population initialization and nectar source initialization. Initialize the number SN of employed bees, follower bees, and observer bees, and randomly generate the initial nectar source within the range of variable values. The random initialization process is expressed as:
[0140]
[0141] where x i,d represents the i-th nectar source X i The j-th dimension value, X i ={x i,1 ,x i,2 ,...,x i,D}, i∈{1,2,...,SN}, SN represents the population size, d∈{1,2,...,D}, D is the dimension of the problem to be solved (i.e., the number of variables to be optimized), and Respectively represent x i,d Initializing the nectar source is to assign initial values within a range to all dimensions of each nectar source using the above formula, thereby randomly generating SN initial nectar sources.
[0142] (2) Hired bee stage. Calculate the fitness of the nectar source and update the location of the nectar source, expressed as:
[0143]
[0144] where x i Indicates the current nectar source location, x k is a randomly selected neighborhood honey source location, where k∈{1,2,...,SN}, k≠i, is a random number in [-1,1], v i,j is the new nectar source V generated by hired bees in each cycle i = {v i,1 ,v i,2 ,...,v i,D}, R i,j The optimization function for selecting the honey source location is expressed as:
[0145]
[0146] Where a and b are constants, f iis the objective function value corresponding to the i-th employed bee, f k is the objective function value corresponding to the kth employed bee, and the objective function f is the conditional risk Bayesian lower bound The fitness function is fit i , expressed as
[0147]
[0148] Then, calculate the new nectar source V i = {v i,1 ,v i,2 ,...,v i,D}, compared with the new nectar source V i = {v i,1 ,v i,2 ,...,v i,D} and Old Honey Source X i ={x i,1 ,x i,2 ,...,x i,D}, compare the fitness of the new nectar source with the old nectar source, and retain the nectar source with greater fitness as the new nectar source.
[0149] (3) Observer bee stage. Based on the fitness of employed bees in selecting nectar sources, the observer bees calculate the probability of selecting the i-th nectar source. where fit i is the fitness function corresponding to the i-th employed bee, i∈{1,2,...,SN}, fit q is the fitness function corresponding to the qth employed bee, q∈{1,2,...,SN}; based on the above nectar source selection probability, the observation bee uses the roulette algorithm to select the nectar source v selected by the SN employed bees. i,d The optimal v i,d , let the optimal v i,d Equal to x i,d Based on x i,d Substitute into the formula Update the location of the honey source to generate a new honey source v i,d .based on Calculate the fitness value and compare the new v i,d And the old v i,d (i.e. x i,d )’s fitness function to determine the new v i,d Is it better than the old v i,d , then use the new v i,d Replace the old v i,d Otherwise, keep the old v i,d , thus obtaining SN updated solution vectors V i = {v i,1 ,v i,2 ,...,vi,d ,...,v i,D}
[0150] (4) Scouting bee stage. The scout bee stage needs to constantly scout the solution vectors in the employed bee stage and the observer bee stage. If the solution vectors in both the employed bee stage and the observer bee stage are updated after two cycles, and α does not reach the maximum number of cycles LIMIT at this time, the algorithm falls into a local optimal solution. In this case, the scout bee uses the formula Randomly generate a new nectar source x i,d To replace the honey source v that has not been updated for a long time i,d .
[0151] (5) Termination condition judgment. If the result converges or reaches the maximum number of cycles LIMIT, the iteration ends and the optimal solution is output, that is, the optimal Gaussian pulse width and frequency modulation slope; otherwise, α = α + 1 is set to enter the hired bee stage to continue the iteration cycle.
[0152] In summary, in the improved artificial bee colony algorithm, employed bees and follower bees develop food sources, while scout bees control the exploration process of food sources in the algorithm, thus ensuring the robustness of the algorithm. From the above process, through the iteration and interaction between employed bees, observer bees and scout bees, the optimal Gaussian pulse width is finally converged. and FM slope
[0153] Step 6: According to the optimal carrier frequency obtained above Optimal bandwidth Optimal residence time Optimum radiated power Optimal Gaussian pulse width Optimal FM slope Get the corresponding radar echo signal, and perform matched filtering and back-end radar data processing according to the corresponding radar echo signal, and then update the radar confidence state. Here we generate the overall target state estimate based on the interactive multi-model-extended Kalman filter (IMM-EKF) and the overall target state estimation error covariance P k|k , which includes the following four steps:
[0154] (1) First, based on the new information covariance Kalman Gain and model probability According to the classic Kalman filter algorithm, the target state estimation corresponding to each tracking model i is calculated and the target state estimation error covariance It can be expressed as
[0155]
[0156] (2) Further, calculate the model likelihood probability corresponding to each tracking model i and the conditional model probability It can be expressed as
[0157]
[0158] (3) Further, by fusing the target state estimation of each tracking model and the target state estimation error covariance Based on the conditional model probability Generate an overall target state estimate and the overall target state estimation error covariance P k|k , which can be expressed as
[0159]
[0160] (4) Finally, the spectrum measurement h is converted to k Transformed into interference state estimate Finally, based on the overall target state estimation The overall target state estimation error covariance P k|k and interference state estimation Generate radar confidence status
[0161] Step 7: Based on radar confidence state b k-1 , updated radar confidence state b k and frequency band occupancy actions Generate reward function It can be expressed as:
[0162]
[0163] Where α1 and α2 are user-defined weight factors, is a binary indicator function. The second term in the above formula corresponds to the bandwidth-related reward. It can be seen that the more sub-bands are occupied, the larger the bandwidth and the higher the ranging resolution. The third term is the reward for the highest sub-band occupied by the radar, which aims to reduce the distortion loss caused by non-essential frequency band switching.
[0164] Step 8: Based on the reward function and radar sub-band occupancy actions Perform strategy update and value function update based on temporal difference error. The specific steps are as follows:
[0165] (1) Based on reward function and the state value function v(b k ,wk-1 ), construct the timing difference error It is expressed as:
[0166]
[0167] (2) Based on the timing difference error and the policy function Update the policy function parameters It is expressed as:
[0168]
[0169] (3) Based on the timing difference error and the state value function v(b k-1 ,w k-1 ), update the state value function parameter w k , expressed as:
[0170]
[0171] In summary, by utilizing the feedback information in the target tracking cycle, the proposed CRMRM scheme sequentially performs frequency domain resource optimization based on TDAC, joint dwell time and pulse width optimization based on analytical method, and joint pulse width and frequency modulation slope optimization based on IABC until converging to the optimal CRMRM strategy.
[0172] The traditional single signal domain resource optimization strategy for a specific single task is difficult to cope with the increasingly complex radar detection environment. By leveraging the reusable information and coupling relationship between multi-stage tasks, the multi-dimensional RF resource integrated scheduling technology for joint multi-tasks improves the linkage multi-task processing capability by considering the multi-dimensional signal domain collaborative optimization, which is a potential breakthrough point. From the perspective of application needs and technological development, there are still the following challenges and difficulties. Specifically:
[0173] (1) From the perspective of radar tasks, the existing resource scheduling criterion function is difficult to characterize the comprehensive performance of joint multi-tasks in complex detection scenarios (including low interception performance and maneuvering target tracking performance). It is necessary to take into account the performance indicators of different tasks, radar resource limitations and real-time environmental constraints, especially the coupling relationship between different tasks and potential multi-task resource conflicts.
[0174] (2) From the perspective of resource elements, solve the high-dimensional, non-convex multi-constrained optimization problem corresponding to the resource scheduling strategy for multi-signal domain collaboration. It is necessary to decompose the optimization strategy based on the characteristics of the criterion function and seek collaborative phased processing at the algorithm level, including random optimization methods, deterministic optimization methods, and analytical methods.
[0175] (3) From the perspective of information utilization, the analysis and utilization of multi-domain prior information is insufficient. It is necessary to consider reusable information based on the coupling and association between tasks, as well as multi-scale prior knowledge, including real-time environmental perception information (such as target motion status and RF interference status) and offline long-term memory (such as trained radar agents and feature databases).
[0176] The purpose of this application is to collaboratively optimize radar multi-dimensional RF resource parameters (including carrier frequency, bandwidth, radiated power, dwell time, pulse width, frequency modulation slope, etc.) under low intercept probability, improve maneuvering target tracking performance and reduce main lobe suppression interference. This application constructs a cognitive radar multi-dimensional RF resource management strategy (CRMRM) for joint anti-multi-source interference and maneuvering target tracking. As far as we know, there is no multi-dimensional resource scheduling method for joint anti-noise suppression interference and maneuvering target tracking under such low intercept probability conditions. The technical problems to be solved include:
[0177] (1) First, the basic idea of the cognitive radar multi-dimensional RF resource management (CRMRM) strategy is to jointly optimize multi-dimensional RF resources by minimizing the CRMRM-oriented criterion function, which integrates the tracking-oriented sub-reward, the band occupancy-oriented sub-reward, and the multi-component constraint terms. Based on the interactive multi-model-extended Kalman filter framework, the first term is constructed by the conditional prediction Bayesian risk lower bound for maneuvering target tracking under the minimum mean square error criterion, including the prediction information matrix and the Fisher information matrix. Based on the RF interference state transition and the maneuvering target state, the second term considers the trade-off between bandwidth utilization and interference avoidance, as well as the impact of the current band occupancy action on the future multi-step reward. The third term (multi-component constraint term) integrates the radar multi-dimensional RF resource constraints, the real-time spectrum constraints generated by the band occupancy action, and the pre-set radar interception probability threshold to characterize the radar's RF stealth performance requirements.
[0178] (2) Second, to solve the underlying real-time nonlinear, high-dimensional, non-convex, multi-variable and multi-constrained optimization problem, we propose a multi-stage iterative optimization algorithm driven by multi-scale prior knowledge through the analysis of the above objective function characteristics. The original CRMRM strategy is decomposed into two sub-strategies and solved in sequence, thereby deconstructing the solution process of discrete variables (i.e., band occupancy action) and continuous variables (i.e., dwell time, radiated power, pulse width, and frequency modulation slope). First, based on the offline prior knowledge obtained from real-time training, the temporal difference Actor-Critic (TDAC) algorithm can well deal with the lagged rewards and future multi-step effects when optimizing the band occupancy sub-strategy through the continuous interaction of strategy update and value function update based on temporal difference error. Subsequently, in order to avoid falling into the local optimal solution by directly using the optimization algorithm when solving the waveform optimization strategy, based on the two proposed theorems, we use an analytical method to jointly optimize the radar average radiated power and dwell time, reducing the variable (optimization) dimension and computational complexity. Subsequently, through iteration and interaction between bee colonies, the improved artificial bee colony algorithm balances global search and local optimization and finally converges to the optimal pulse width and frequency modulation slope, while satisfying the waveform parameter constraints and the real-time spectrum constraints generated by the frequency band occupancy action.
[0179] It should be noted that the specific implementation is only an explanation and description of the technical solution of the present invention, and cannot be used to limit the scope of protection of the rights. Any partial changes made according to the claims and description of the present invention should still fall within the scope of protection of the present invention.
Claims
1. A multi-dimensional radar resource scheduling method for joint anti-interference and tracking under low probability of interception, characterized in that The following steps are involved: Randomly initialize the strategy function and the state value function v(b,w), where β represents the randomly initialized radar band occupation action, b represents the randomly initialized radar confidence state, represents the randomly initialized policy function parameters, w represents the randomly initialized state value function parameters, and the randomly initialized policy learning rate and the state value function learning rate α w ,in, Perform the following steps in sequence for time step k, k = 1, ..., N; Step 1: Obtain radar echo data and use the radar echo data to construct the radar confidence state b at time step k-1 k-1 , the radar confidence state includes a target state and a radio frequency interference state; Step 2: Radar confidence state b based on time step k-1 k-1 , using the policy function of time step k-1 Generate radar sub-band selection action And select actions for radar sub-bands Solve and get the optimal radar carrier frequency and optimal radar bandwidth Step 3: Use the radar intercept probability threshold p th and radar confidence state b k-1 , respectively, at the maximum radar dwell time Radar radiated power under and at the minimum radar dwell time Radar radiated power under Among them, R k|k-1 represents the predicted distance between the radar and the target at time k-1; Step 4: Based on the upper limit of radar radiated power and lower limit contrast and The smallest value is recorded as the radar radiation power that meets the RF resource constraint condition. Step 5: Compare the maximum signal-to-interference-noise ratio criterion and Select the group with the largest value, which is the optimal radar dwell time. and optimal radiated power Step 6: Action based on radar sub-band occupancy Generate real-time spectrum constraint function And the spectrum constraint function As the real-time spectrum constraint condition for waveform parameter optimization, w type Indicates the type of radar waveform, λ k represents the Gaussian pulse width, κ k Indicates the frequency modulation rate; Step 7: Radar confidence state b based on time step k-1 k-1 Constructing a conditional predictive Bayesian risk lower bound for the MMSE criterion Step 8: Generate resource constraint function R based on resource constraints Θ (λ k ,κ k |w type ); Step 9: Predicting Bayesian risk based on conditions under the MMSE criterion Resource constraint function R Θ (λ k ,κ k |w type ) and spectral constraint function Construct constrained optimization problems; Step 10: Solve the optimization problem in step 9 and obtain the optimal Gaussian pulse width and the optimal FM slope Step 11: Based on the optimal radar carrier frequency Optimal radar bandwidth Optimal radar dwell time Optimum radiated power Optimal Gaussian pulse width And the optimal frequency modulation slope Get the echo signal and obtain the spectrum measurement h in the echo signal k ; Step 12: Generate overall target state estimation based on interactive multi-model-extended Kalman filter and the overall target state estimation error covariance P k|k , and the spectrum of the echo signal is measured by the classical energy detection method k Transformed into interference state estimate Step 13: Estimation based on overall target state The overall target state estimation error covariance P k|k and interference state estimation Generate radar confidence status That is, the radar confidence state b k ; Step 14: Based on radar confidence state b k-1 , radar confidence state b k and radar sub-band occupancy actions Generate reward function Step 15: Based on the reward function and the state value function v(b k ,w k-1 ), construct the timing difference error Step 16: Based on the timing difference error and the policy function Get the policy function parameters Step 17: Based on the timing difference error and the state value function v(b k-1 ,w k-1 ), get the state value function parameter w k ; Step 18: Based on the policy function parameters and the state value function parameter w k , construct the function and the state value function v(b k ,w k ), and the policy function and the state value function v(b k ,w k ), repeat the above steps until convergence, get the optimal strategy function, and use the optimal strategy function to get the optimal CRMRM strategy.
2. The multi-dimensional radar resource scheduling method for joint anti-interference and tracking under low probability of intercept according to claim 1 is characterized in that The optimal radar carrier frequency It is expressed as: The optimal radar bandwidth It is expressed as: in, represents the binary indicator function, f c represents the radar center frequency, ΔB represents the frequency band width of a sub-band, Indicates radar sub-band occupation action β k The i-th element of , N represents the number of sub-bands into which the radio frequency band is equally divided.
3. The multi-dimensional radar resource scheduling method for joint anti-interference and tracking under low probability of intercept according to claim 2 is characterized in that The radar radiated power It is expressed as: The radar radiated power It is expressed as: The radar radiated power It is expressed as: Among them, p' fa represents the preset false alarm probability of the intercept receiver, erfc(·) represents the complementary error function, T I represents the search time of the intercept receiver, k0 represents the Boltzmann constant, T0 represents the effective noise temperature of the radar, B I represents the intercept receiver bandwidth, F I represents the noise factor of the intercept receiver, G t represents the radar transmitting antenna gain, G r represents the radar receiving antenna gain, G IP represents the radar receiver processing gain, λ t Indicates the radar wavelength.
4. The multi-dimensional radar resource scheduling method for joint anti-interference and tracking under low probability of intercept according to claim 3 is characterized in thatThe optimal radar dwell time and optimal radiated power It is expressed as: The spectrum constraint function It is expressed as: in, Indicates the frequency band occupation action β k The corresponding bandwidth, B θ (λ k ,κ k |w type ) indicates that under the linear frequency modulation waveform, the Gaussian pulse width λ k and the frequency modulation rate κ k Determines the signal bandwidth.
5. The multi-dimensional radar resource scheduling method for joint anti-interference and tracking under low probability of intercept according to claim 4 is characterized in that The Bayesian risk lower bound It is expressed as: in, and denote the target state transition matrix at time step k-1 and the noise covariance matrix at time step k-1 respectively, M denotes the number of target tracking models, represents the target mixed state error covariance at time step k-1, expressed as: in, represents the mixed state estimate at time step k-1, represents the probability of the mixed model at time step k-1, Γ mi represents the model transition probability from model m to model i, and They represent the target state estimation corresponding to the target tracking model j at time step k-1 and the target state covariance matrix at time step k-1, μ k-1|k-1 (j) represents the model validity probability of tracking model j at time step k-1, Γ ji represents the model transition probability from model j to model i, H k|k-1 represents the Jacobian matrix of the radar measurement function h(·) at time step k-1 with respect to the target state prediction, represents the second-order derivative operator, represents the target state at time step k predicted at time step k-1, It is expressed as: in, represents the measurement error covariance matrix, c represents the speed of light, SINR k (β k ,P t,k ,T d,k ) is the signal-to-interference-noise ratio at time k, expressed as: Among them, P t,k is the radar radiation power, T d,k is the radar dwell time, is the radial distance between the radar and the target, The radio frequency interference state is c k The i-th element of is the radar sub-band occupation action β k The i-th element of T r represents the radar pulse repetition interval, G RP represents the radar receiver processing gain, σ represents the radar cross-section of the target, L represents the generalized loss factor, P I It represents the interference signal power from the same frequency suppression noise interference source calculated by the interference equation; The resource constraint function R Θ (λ k ,κ k |w type ) is expressed as; in, and Respectively represent the radar waveform type w type The corresponding Gaussian pulse width λ k The minimum and maximum values of and Respectively represent the radar waveform type w type The corresponding minimum and maximum values of the frequency modulation slope.
6. The multi-dimensional radar resource scheduling method for joint anti-interference and tracking under low probability of interception according to claim 5 is characterized in that The constrained optimization problem is expressed as: in, represents the optimal value of the radar Gaussian pulse width, Indicates the optimal value of the radar frequency modulation slope.
7. The multi-dimensional radar resource scheduling method for joint anti-interference and tracking under low probability of interception according to claim 6 is characterized in that In step 10, solving the optimization problem in step 9 is performed by improving the artificial bee colony algorithm. The specific steps for solving the optimization problem in step 9 are: Step 1: Initialize the number SN of employed bees, follower bees, and observer bees, and randomly generate the initial nectar source within the range of variable values. The random initialization process is expressed as: Among them, x i,d represents the i-th nectar source X i The j-th dimension value, X i ={x i,1 ,x i,2 ,...,x i,D }, i∈{1,2,...,SN}, SN represents the population size, d∈{1,2,...,D}, D represents the dimension of the problem to be solved, that is, the number of variables to be optimized, and Respectively represent x i,d The upper and lower bounds of the value, Step 2: Calculate the fitness of the nectar source and update the nectar source location, expressed as: Among them, x i Indicates the current nectar source location, x k , represents the randomly selected neighborhood honey source location, where k∈{1,2,...,SN}, k≠i, is a random number in [-1,1], v i,j Represents the new nectar source V generated by employed bees in each cycle i = {v i,1 ,v i,2 ,...,v i,D }, R i,j It represents the optimization function of nectar source location selection, which is expressed as: Among them, a, b are constants, f i represents the objective function value corresponding to the i-th employed bee, f k represents the objective function value corresponding to the kth employed bee. The objective function f is the conditional risk Bayesian lower bound. Fitness function fit i It is expressed as: Then, calculate the new nectar source V i = {v i,1 ,v i,2 ,...,v i,D }, compared with the new nectar source V i = {v i,1 ,v i,2 ,...,v i,D } and Old Honey Source X i ={x i,1 ,x i,2 ,...,x i,D }, compare the fitness of the new nectar source with the old nectar source, and retain the nectar source with greater fitness as the new nectar source; Step 3: Based on the fitness of employed bees in selecting nectar sources, the observer bees calculate the probability of selecting the i-th nectar source Among them, fit i is the fitness function corresponding to the i-th employed bee, i∈{1,2,...,SN}, fit q is the fitness function corresponding to the qth employed bee, q∈{1,2,...,SN}. Based on the above nectar source selection probability, the observation bee uses the roulette algorithm to select the nectar source v selected by the SN employed bees. i,d The optimal v i,d , let the optimal v i,d Equal to x i,d , based on x i,d Substitute into the formula Update the location of the honey source to generate a new honey source v i,d ,based on Calculate the fitness value and compare the new v i,d and x i,d The fitness function is used to determine the new v i,d Is it better than x? i,d , if the new v i,d Better than x i,d , then use the new v i,d Replace x i,d Otherwise, keep x i,d , thus obtaining SN updated solution vectors V i = {v i,1 ,v i,2 ,...,v i,d ,...,v i,D }; Step 4: If the solution vectors of both the employed bee stage and the observer bee stage are updated after two cycles, and α has not reached the maximum number of cycles LIMIT, the scout bee uses the formula Randomly generate a new nectar source x i,d To replace the honey source v that has not been updated for a long time i,d ; Step 5: Determine whether the result converges or reaches the maximum number of cycles LIMIT. If so, the iteration ends and the optimal solution is output, which is the optimal Gaussian pulse width. and the optimal FM slope Otherwise, let α=α+1 and enter the hired bee stage to continue the iterative cycle.
8. The multi-dimensional radar resource scheduling method for joint anti-interference and tracking under low probability of interception according to claim 7 is characterized in that The interactive multi-model-extended Kalman filter generates an overall target state estimate and the overall target state estimation error covariance P k|k , specifically including the following steps: Step A: Based on the innovation covariance Kalman Gain and model probability And according to the classic Kalman filter algorithm, the target state estimation corresponding to each tracking model i is calculated and the target state estimation error covariance It is expressed as: Among them, z k are the radar measurements of the target’s radial position and radial velocity, represents the Jacobian matrix of the radar measurement function h(·) with respect to the radar target, represents the target state transition matrix of the radar at time step k-1; Step B: Calculate the model likelihood probability corresponding to each tracking model i and the conditional model probability It is expressed as: Step C: Target state estimation by fusing each tracking model and the target state estimation error covariance Based on the conditional model probability Generate an overall target state estimate and the overall target state estimation error covariance P k|k , expressed as: Among them, E{x k |z k } represents the radar target state x k Given the radar measurement z k , E{·} represents the operator for finding the mathematical expectation; Step D: Use the classical energy detection method to measure the spectrum h k Transformed into interference state estimate Finally, based on the overall target state estimation The overall target state estimation error covariance P k|k and interference state estimation Generate radar confidence status 9. The multi-dimensional radar resource scheduling method for joint anti-interference and tracking under low probability of interception according to claim 8 is characterized in that The reward function It is expressed as: Among them, α1 and α2 represent user-defined weight factors.
10. The multi-dimensional radar resource scheduling method for joint anti-interference and tracking under low probability of intercept according to claim 9 is characterized in that The specific steps of step seventeen are: Based on reward function and the state value function v(b k ,w k-1 ), construct the timing difference error It is expressed as: Where γ represents the discount factor; Based on the timing difference error and the policy function Update the policy function parameters It is expressed as: in, represents the learning rate of radar frequency band occupancy strategy update, Indicates the radar frequency band occupation strategy About parameters The gradient of Based on the timing difference error and the state value function v(b k-1 ,w k-1 ), update the state value function parameter w k , expressed as: Among them, α w represents the learning rate of the state value function update, It means to find the state value function v(b k-1 ,w k-1 ) is the gradient of w with respect to the parameter w.
Citation Information
Cited By
Low-interception probability radar waveform design method and device
CN121091255A
