An efficient optimization method based on multivariate vector controlled cross-eye interference technology

Abstract

This invention discloses an efficient optimization method based on multivariate vector controlled cross-eye interference technology, which belongs to the field of wireless communication technology. In the optimization stage of generating interference antenna target points, the present invention adopts an embedded hybrid algorithm, selects the PSO algorithm as the dominant one, and integrates the crossover and mutation ideas unique to the GA algorithm into the optimization process. By expanding the population diversity through crossover and mutation, the algorithm is not easily trapped in the local optimum during optimization, and can quickly narrow the solution space within the global parameter range, effectively improving the optimization accuracy. In the process of selectively screening amplitude and phase parameters, the present invention introduces a rapid modeling scheme based on neural networks to reduce the overall time consumption of forward numerical modeling and eliminate feed parameter combinations that do not meet the accuracy requirements.

Description

Technical Field

[0001] The invention relates to a high-efficiency optimization algorithm based on a multi-element vector controlled cross-eye interference technology, belonging to the technical field of wireless communications. Background Art

[0002] Currently, cross-eye jamming is considered the most effective method for jamming monopulse radars. However, effective cross-eye jamming requires that the two jamming signals be of nearly equal amplitude and opposite phase, a requirement that cannot be guaranteed with current hardware. To address these limitations, a multi-element vector-controlled cross-eye jamming technique has been proposed. This technique increases the jamming system's freedom by introducing multiple antennas, allowing the synthesized "false targets" to reside in a two-dimensional plane, significantly increasing the range and breadth of jamming options. To achieve effective jamming, multi-element vector-controlled cross-eye jamming requires the amplitude and phase of the multiple antennas. Therefore, the accuracy of the feed amplitude and phase of the jamming antenna array is crucial. Otherwise, not only will jamming be ineffective, but the jammer's own aircraft or ship could even become a beacon. Furthermore, missile speeds have generally reached several times the speed of sound, with advanced intercontinental missiles reaching up to twenty times the speed of sound. Therefore, in the field of electronic countermeasures, rapid and accurate response to incoming missiles is crucial, and efficient and high-precision control algorithms are crucial.

[0003] Currently, cross-eye jamming technology based on multi-element vector control uses the Poynting vector theorem to select the amplitude and phase of multiple antenna elements and calculate the location of the equivalent combined interference center. However, the large number of amplitude and phase combinations involved increases the computational complexity. Quickly finding the amplitude and phase center point that is hardware-implementable among all these combinations is crucial for successful jamming. Therefore, a combination of optimization algorithms and neural networks is used to accelerate execution. This optimization algorithm optimization significantly improves computational time, computational complexity, and accuracy. However, each individual algorithm has its own strengths and weaknesses, and hybrid algorithms are increasingly being used to leverage their strengths and overcome their weaknesses, accelerating computational speed and improving accuracy. Machine learning, after decades of rapid development, has developed a large number of models and a robust algorithmic foundation. It has significantly addressed the challenges of high memory usage and computational time consumption, and is playing an increasingly important role in solving practical problems. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an efficient optimization algorithm based on multi-element vector controlled cross-eye jamming technology, which is used to solve the problem that when an aircraft or ship faces an attack from precision-guided weapons, it can quickly, efficiently and accurately calculate the feed amplitude and phase of each antenna unit, generate a false target to lure the enemy's seeker, and thus effectively interfere with the single-pulse radar.

[0005] The present invention is achieved through the following technical solutions:

[0006] An efficient optimization method based on multivariate vector controlled cross-eye interference technology, comprising:

[0007] Step 1: Set the target point where interference is required;

[0008] Step 2: Based on the set target point position, the Poynting vector method combined with the particle swarm-genetic hybrid optimization algorithm is used to extract the initial amplitude and phase parameters of the interference antenna group;

[0009] Step 2.1: Based on the selected target point A(x,y), set the interference antenna model size and normalize the interference antenna model parameters; set the optimization range of the amplitude and phase parameters based on prior information, where the maximum range of the amplitude is (0,1) and the phase is (0,2π); set the population size, rated number of iterations, and target accuracy;

[0010] Step 2.2: Define the error between the forward solution and the target position parameter as the population cost function of the algorithm, that is, the loss function:

[0011] C(m)=||S(m)-S goal ||2

[0012] in, is the interference antenna feeding amplitude and phase parameters to be optimized, S(m) is the interference antenna forward response, S goal is the target point location parameter, A n represents the feeding amplitude of the nth antenna to be optimized, Indicates the feeding phase of the nth antenna to be optimized; brings the particle population members into the objective function to calculate the fitness value of each particle, evaluates the fitness of each particle, and updates the historical individual optimal pbest and historical global extreme value gbest of each particle;

[0013] Step 2.3: Update the particle speed and position according to the position and speed update formula of the particle swarm algorithm, and evaluate the particle fitness value;

[0014] The position and velocity update formula of the particle swarm algorithm in step 2.3 is:

[0015]

[0016]

[0017] in, gbest represents the best result obtained by searching for the i-th particle up to the t-th generation. (t) is the best solution of all particle swarms so far, They are the current position and speed of the i-th particle, c1 and c2 are learning factors, r1 and r2 are random numbers on [0,1], and ω is the inertia weight.

[0018] Step 2.4: Select a certain proportion of individuals in the population and optimize the new population according to a certain crossover probability;

[0019] Step 2.5: Select a certain proportion of individuals in the population and perform mutation operations according to certain rules to generate new mutant individuals;

[0020] Step 2.6: Evaluate whether the optimal value of the population after the above optimization operation has reached the rated convergence number or meets the target point optimization accuracy. If not, return to step 2.3 to continue a new round of optimization. If the output conditions are met, output multiple optimal solutions of the population, one of which is the initial feed amplitude and initial phase information of an interference antenna group.

[0021] Step 3: Input the initial feeding amplitude and initial phase information into the trained GA-BP or PSO-SVM multivariate vector synthesis network to generate the fitting coordinates of the equivalent radiation center;

[0022] Step 4: Calculate the fitness value. The fitness function is defined as the error between the fitted coordinates of the equivalent radiation center and the true position, as follows:

[0023]

[0024] Among them, n is the number of target point position parameters output by the network, y i is the true position of the equivalent radiation center; i is the predicted target point position of the i-th node, k is the coefficient, and abs(·) represents the absolute value;

[0025] According to the calculated error, filter out the feeding amplitude and phase information that meet the accuracy requirements; if none of them meet the requirements, return to step 2 and recalculate;

[0026] Step 5: Output the feeding amplitude and phase parameters that meet the hardware requirements.

[0027] Furthermore, the GA-BP multi-vector synthesis network training process in step 3 specifically includes:

[0028] Step 3.1: Perform data preprocessing on the initial amplitude and phase parameters, determine the BP network structure, and create a BP neural network.

[0029] Step 3.2: Use genetic algorithm to encode the initial value and calculate the fitness value;

[0030] Step 3.3: Determine whether the fitness value meets the termination condition; if so, use the weight threshold of the population optimized by the GA algorithm as the initial parameter of the multivariate vector synthesis BP network; if not, select, crossover, and mutate the initial value to generate a new population and proceed to the next iteration;

[0031] Step 3.4: Calculate the BP network output error and determine whether the termination condition is met; if so, complete the GA-BP network modeling; if not, update the weight threshold and perform the next iteration until the accuracy requirement is met.

[0032] Furthermore, the PSO-SVM multi-vector synthesis network training process in step 3 specifically includes:

[0033] Step 3.1: Preprocess the sample data;

[0034] Step 3.2: Initialize parameters c and g and set the parameter optimization range;

[0035] The c parameter is the penalty factor of the SVM, which represents the balance between network training complexity and error tolerance. If c is too high, the model complexity will increase, leading to overfitting. If c is too small, the decision plane transition will be smooth, resulting in underfitting, affecting training accuracy.

[0036] g is the kernel function parameter of the SVM network. It determines the mapping complexity and feature space distribution. The value of g affects the number of support vectors, and the number of support vectors affects the training speed of the SVM. The more support vectors, the slower the training speed, and the fewer support vectors, the faster the training speed.

[0037] Step 3.3: Calculate the population fitness value;

[0038] Step 3.4: Update the c and g values ​​according to the PSO optimization algorithm steps described in step 3.2;

[0039] Step 3.5: Determine whether the termination condition is met. If so, proceed to the next step; otherwise, go to step 3.3.

[0040] Step 3.6: Bring the optimized c and g parameters to SVM for regression model training.

[0041] Because the feed amplitude for each antenna is in the range of (0, 1) and the phase is (0, 2π), the computational complexity is too high when performing calculations for multiple antennas. Step 2 roughly determines the amplitude and phase for each antenna. Then, the resulting initial amplitude and phase parameters are fed into the trained neural network in step 3 to calculate the equivalent synthesis center (the location of the target point). Step 4 determines whether the error between the target point and the true value of the feed amplitude and phase parameters meets the accuracy requirements. If not, the next iteration is performed. If the accuracy requirements are met, step 5 determines whether the hardware conditions are met (because the hardware cannot output every amplitude and phase). If so, the results are output; otherwise, the next iteration is performed. (The trained neural network is used to replace multiple repeated iterations of electromagnetic simulation to improve accuracy, which can significantly reduce calculation time.)

[0042] The present invention has the following advantages and beneficial effects:

[0043] 1. In the optimization stage of generating interference antenna target points, the present invention adopts an embedded hybrid algorithm, selects the PSO algorithm as the dominant one, and integrates the crossover and mutation ideas unique to the GA algorithm into the optimization process. By expanding the population diversity through crossover and mutation, the algorithm is not easily trapped in the local optimum during optimization, and can quickly narrow the solution space within the global parameter range, effectively improving the optimization accuracy.

[0044] 2. The present invention introduces a neural network-based rapid modeling solution in the process of preferentially screening amplitude and phase parameters to reduce the overall time consumption of forward numerical modeling and eliminate feed parameter combinations that do not meet the accuracy requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The drawings described herein are used to provide a further understanding of the implementation of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0046] Figure 1 Flow chart of the optimization method of the present invention.

[0047] Figure 2 The present invention is a flowchart of the target point optimization using the particle swarm-genetic hybrid optimization algorithm (PSO-GA) in step 1 of the optimization method.

[0048] Figure 3 (a) GA-BP network optimization structure diagram.

[0049] Figure 3 (b) is a flowchart of GA-BP network training in step 2 of the optimization method of the present invention.

[0050] Figure 4 The flowchart of PSO-SVM network training is used in step 2 of the optimization method of the present invention.

[0051] Figure 5 (a) is the normalized model of three-antenna array.

[0052] Figure 5 (b) shows the convergence characteristics of the four test points.

[0053] Figure 6 (a) Prediction error of BP network without genetic algorithm optimization.

[0054] Figure 6 (b) is the prediction error of the BP network optimized by genetic algorithm.

[0055] Figure 7 (a) The fitting performance of the SVM network without PSO optimization for the multivariate vector synthesis model.

[0056] Figure 7 (b) The fitting performance of the SVM network after PSO optimization for the multivariate vector synthesis model. DETAILED DESCRIPTION

[0057] The technical solution of the present invention is described in detail below with reference to the embodiments and drawings.

[0058] To address the limited accuracy and efficiency of existing target optimization inversion methods, as well as the low efficiency of forward numerical modeling in optimal screening, this paper proposes a high-efficiency optimization algorithm based on multivariate vector-controlled cross-eye interference technology. In the optimization process of interference target points, this paper uses the Poynting vector method combined with a particle swarm-genetic hybrid optimization algorithm (PSO-GA) to efficiently and accurately extract the initial amplitude and phase parameters of the interference antenna. Furthermore, a GA-BP / PSO-SVM multivariate vector synthesis network is introduced to optimally screen the initial amplitude and phase parameters, greatly improving the accuracy and speed of screening.

[0059] like Figure 1 As shown, the method of this embodiment specifically includes the following steps:

[0060] Step 1: Set the target point where interference is required;

[0061] Step 2: Based on the set target point position, the Poynting vector method combined with the particle swarm-genetic hybrid optimization algorithm (PSO-GA) is used to extract the initial amplitude and phase parameters of the interference antenna;

[0062] Step 3: Bring the initial amplitude and phase parameters into the trained GA-BP / PSO-SVM multivariate vector synthesis network to generate the fitting coordinates of the equivalent radiation center;

[0063] Step 4: Calculate the fitness value. The fitness function is defined as the error between the fitted coordinates of the equivalent radiation center and the true position, as follows:

[0064]

[0065] Where n is the number of target point position parameters output by the network, y i is the true position of the equivalent radiation center; i is the predicted target point position of the i-th node; k is the coefficient.

[0066] Step 5: Output the feeding amplitude and phase parameters that meet the hardware requirements;

[0067] Figure 2 The specific steps of the PSO-GA hybrid optimization algorithm used in step 2 of the present invention are shown. The application in this embodiment is as follows:

[0068] like Figure 5 As shown in (a), in this embodiment, a three-antenna array model is selected and normalized. Four typical radiation centers within and outside the three-antenna array region are selected as optimization targets for analysis: (-0.1, 0.5), (1.1, 0.5), (0.5, -0.1), and (0.5, 0.2), located in different characteristic regions.

[0069] The population size is set to 120, the maximum number of iterations is controlled at 30 steps, and the optimization range of amplitude and phase parameters is (0,0,0,0,0,0) to (1,1,1,π,π,π). In each iteration, all particles in the population are first optimized by PSO, and then enter the GA optimization phase, with a crossover probability of 70% and a mutation probability of 30%.

[0070] Figure 5 (b) is the convergence characteristics of the four sampling points based on the PSO-GA hybrid algorithm. It can be seen from the figure that the optimal fitness of the swarm particles decreases rapidly with the increase of the number of iterations, and the accuracy requirement of 10-6 can be met after less than 10 iterative optimizations.

[0071] The inversion performance of different methods was compared by controlling the inversion accuracy to 10-6 and the maximum number of iterations to 30. The results are shown in Tables 1 and 2.

[0072] Table 1 Comparison of PSO-GA optimization results for four typical target points (fixed error of 30 iterations)

[0073]

[0074] Table 2 Comparison of PSO-GA optimization results for four typical target points (fixed optimization accuracy 10-6)

[0075]

[0076] From the results, it can be seen that the PSO-GA hybrid optimization algorithm is much more accurate than the single PSO algorithm when the number of iterations is the same when searching for the target point. When the optimization accuracy is fixed, the number of iterations required is also less and the time spent is shorter.

[0077] Figure 3 (a) and (b) are specific steps of the GA-BP multivariate vector synthesis network used in step 3 of the present invention, and are applied in this embodiment as follows:

[0078] Step 3.1: Preprocess the initial amplitude and phase parameters, create a BP neural network, and determine the BP network structure: the input layer has 6 nodes (the amplitude and phase vector x of the interference antenna = {A1, A2, A3, φ1, φ2, φ3}), and the neural network output vector is the target point coordinates calculated by electromagnetic simulation under different parameter combinations. That is, the output layer consists of two neurons. The three-antenna array BP network to be optimized has two hidden layers, with 10 and 6 nodes respectively.

[0079] Step 3.2: Use genetic algorithm to encode the initial value and calculate the fitness value.

[0080] Among them, the number of parameters to be optimized by the GA algorithm obtained from step 2 is 120, that is, the encoding length is 120.

[0081] The fitness function is defined as the error between the fitted coordinates of the equivalent radiation center and the true position, as follows:

[0082]

[0083] Among them, n is the number of target point position parameters output by the network, y i is the true position of the equivalent radiation center; i is the predicted target point position of the i-th node; k is the coefficient.

[0084] Step 3.3: Determine whether the fitness value meets the termination criteria. If so, use the weight threshold of the population optimized by the GA algorithm as the initial parameters for the multivariate vector synthesis BP network. If not, select the initial values, set the crossover probability to 70% and the mutation probability to 30%, generate a new population, and proceed to the next iteration.

[0085] Step 3.4: Calculate the BP network output error and determine whether the termination condition is met. If so, complete the GA-BP network modeling and output the optimally selected amplitude and phase parameters. If not, update the weight threshold and proceed to the next iteration until the accuracy requirement is met.

[0086] In this embodiment, only the case where the interfering antenna feed amplitude is variable is considered. The number of weight thresholds to be optimized is set to 120, and the same number of iteration steps is limited to 50. The prediction performance of the BP network for the equivalent radiation center of the three-antenna array before and after GA algorithm optimization is compared and analyzed.

[0087] like Figure 6 As shown in (a), the training error of the BP network without GA algorithm optimization is (10-5).

[0088] like Figure 6 As shown in (b), the training error of the BP network optimized by the GA algorithm is (10-6).

[0089] Therefore, using the GA algorithm to provide optimized initial values ​​can enhance network performance. It avoids the defect of traditional network training methods that are prone to falling into local minima, improves the convergence speed, and also increases the prediction accuracy, overcoming the uncertainty of traditional network performance affected by parameters.

[0090] Figure 4 The specific steps of the present invention in step 2 using the PSO-SVM multivariate vector synthesis network to preferentially screen the amplitude and phase parameters are as follows in this embodiment:

[0091] Set the optimization interval of parameters c and g to [2-2, 22], bring the initial amplitude and phase parameters obtained in step 2 into the trained PSO-SVM network model, and output the feeding amplitude and phase parameters that meet the hardware requirements.

[0092] In order to verify the fitting performance of the SVM network optimized by PSO for the multivariate vector synthesis model, the same three-interference antenna array sample was selected and SVM and PSO-SVM training were performed respectively. The simulation results are as follows:

[0093] Figure 7 (a) Actual and network prediction graphs of the SVM network test set without PSO optimization.

[0094] Figure 7 (b) Actual and network prediction graphs of the SVM network test set after PSO optimization.

[0095] By using SVM to model the multivariate vector synthesis, and simulating the model under different training strategies based on the same three-interference antenna array dataset, the following comparison can be obtained, as shown in Table 3.

[0096] Table 3 Comparison of SVM network performance with different optimization schemes

[0097] SVM optimization strategy Support Vector Machine PSO Optimization of SVM Mean square error 0.00113256 8.6304e-06 Correlation coefficient 0.989491 0.999707

[0098] Comparing the data in the table, we find that the SVM network based on PSO parameter optimization achieves higher fitting accuracy for the regression modeling of the equivalent radiation center of a three-antenna array. Considering only the unit feed amplitude, the mean square error between the predicted target point location and the true radiation center coordinates reaches the order of 10⁻⁶. Furthermore, the PSO algorithm does not need to traverse all parameter points within the c and g parameter adjustment step grid, saving memory usage and computation time. Therefore, the PSO-SVM-based multivariate vector synthesis network model has greater application value in the process of optimizing the initial feed amplitude and phase parameters.

[0099] In order to verify the effectiveness of the present invention, this embodiment provides a comparison of optimization results of some radiation target points, as shown in Table 4.

[0100] Table 4 Comparison of network model results of multivariate vector synthesis

[0101]

[0102]

[0103] By comparing the data in the table, it is found that the present invention has higher calculation accuracy for each target point, thereby proving the feasibility of the proposed optimization scheme.

[0104] Through the description of this embodiment, it can be effectively proved that the efficient optimization algorithm of the multivariate vector synthesis technology proposed in this paper has higher inversion accuracy and computational efficiency.

Claims

1. An efficient optimization method based on multi-element vector controlled cross-eye interference technology, comprising: Step 1: Set the target point where interference is required; Step 2: Based on the set target point position, the Poynting vector method combined with the particle swarm-genetic hybrid optimization algorithm is used to extract the initial amplitude and phase parameters of the interference antenna group; Step 2.1: Based on the selected target point A(x,y), set the interference antenna model size and normalize the interference antenna model parameters; set the optimization range of the amplitude and phase parameters based on prior information, where the maximum range of the amplitude is (0,1) and the phase is (0,2π); set the population size, rated number of iterations, and target accuracy; Step 2.2: Define the error between the forward solution and the target position parameter as the population cost function of the algorithm, that is, the loss function: C(m)=||S(m)-S goal ||2 in, is the interference antenna feeding amplitude and phase parameters to be optimized, S(m) is the interference antenna forward response, S goal is the target point location parameter, A n represents the feeding amplitude of the nth antenna to be optimized, Indicates the feeding phase of the nth antenna to be optimized; brings the particle population members into the objective function to calculate the fitness value of each particle, evaluates the fitness of each particle, and updates the historical individual optimal pbest and historical global extreme value gbest of each particle; Step 2.3: Update the particle speed and position according to the position and speed update formula of the particle swarm algorithm, and evaluate the particle fitness value; The position and velocity update formula of the particle swarm algorithm in step 2.3 is: in, gbest represents the best result obtained by searching for the i-th particle up to the t-th generation. (t) is the best solution of all particle swarms so far, V i (t) are the current position and speed of the i-th particle, c1 and c2 are learning factors, r1 and r2 are random numbers in [0,1], and ω is the inertia weight; Step 2.4: Select a certain proportion of individuals in the population and optimize the new population according to a certain crossover probability; Step 2.5: Select a certain proportion of individuals in the population and perform mutation operations according to certain rules to generate new mutant individuals; Step 2.6: Evaluate whether the optimal value of the population after the above optimization operation has reached the rated convergence number or meets the target point optimization accuracy. If not, return to step 2.3 to continue a new round of optimization. If the output conditions are met, output multiple optimal solutions of the population, one of which is the initial feed amplitude and initial phase information of an interference antenna group. Step 3: Input the initial feeding amplitude and initial phase information into the trained GA-BP or PSO-SVM multivariate vector synthesis network to generate the fitting coordinates of the equivalent radiation center; Step 4: Calculate the fitness value. The fitness function is defined as the error between the fitted coordinates of the equivalent radiation center and the true position, as follows: Among them, n is the number of target point position parameters output by the network, y i is the true position of the equivalent radiation center; i is the predicted target point position of the i-th node, k is the coefficient, and abs(·) represents the absolute value; According to the calculated error, filter out the feeding amplitude and phase information that meet the accuracy requirements; if none of them meet the requirements, return to step 2 and recalculate; Step 5: Output the feeding amplitude and phase parameters that meet the hardware requirements.

2. The efficient optimization method based on multi-element vector controlled cross-eye interference technology according to claim 1, characterized in that: The GA-BP multi-vector synthesis network training process in step 3 specifically includes: Step 3.1: Perform data preprocessing on the initial amplitude and phase parameters, determine the BP network structure, and create a BP neural network; Step 3.2: Use genetic algorithm to encode the initial value and calculate the fitness value; Step 3.3: Determine whether the fitness value meets the termination condition; if so, use the weight threshold of the population optimized by the GA algorithm as the initial parameter of the multivariate vector synthesis BP network; if not, select, crossover, and mutate the initial value to generate a new population and proceed to the next iteration; Step 3.4: Calculate the BP network output error and determine whether the termination condition is met; if so, complete the GA-BP network modeling; if not, update the weight threshold and perform the next iteration until the accuracy requirement is met.

3. The efficient optimization method based on multi-element vector controlled cross-eye interference technology according to claim 1, characterized in that: The PSO-SVM multi-vector synthesis network training process in step 3 specifically includes: Step 3.1: Preprocess the sample data; Step 3.2: Initialize parameters c and g and set the parameter optimization range; The c parameter is the penalty factor of the SVM, which represents the balance between network training complexity and error tolerance. If c is too high, the model complexity will increase, leading to overfitting. If c is too small, the decision plane transition will be smooth, resulting in underfitting, affecting training accuracy. g is the kernel function parameter of the SVM network, which determines the mapping complexity and feature space distribution. The value of g affects the number of support vectors, and the number of support vectors affects the training speed of the SVM. The more support vectors, the slower the training speed, and the fewer support vectors, the faster the training speed. Step 3.3: Calculate the population fitness value; Step 3.4: Update the c and g values ​​according to the PSO optimization algorithm steps described in step 3.2; Step 3.5: Determine whether the termination condition is met. If so, proceed to the next step; otherwise, go to step 3.

3. Step 3.6: Bring the optimized c and g parameters to SVM for regression model training.

Images