Cluster target optimization semi-physical simulation method based on cascaded BP neural network

Abstract

The invention provides a cascaded BP neural network-based cluster target optimization semi-physical simulation method and system, and the method comprises the steps: determining a cluster target value scoring index according to a simulation task demand, and constructing a scoring sample training set with wide score distribution in combination with simulation experience; constructing an initial cascade BP neural network, and training the cascade BP neural network through the scoring sample training set; judging whether the distribution of the current cluster target exceeds the range of a simulation antenna group, if so, traversing all antenna groups of the simulation cluster target through a position relation, calculating a scoring index of each alternative scheme, inputting the score index into the trained cascade BP neural network, outputting a scoring result of each scheme, and taking the highest score as a final simulation scheme; and if not, directly simulating all targets. According to the method, the cluster target simulation effect is optimized under the condition that existing simulation resources are limited, the current optimal simulation antenna group is intelligently decided, the network structure is simple, and the processing speed is high.

Description

Technical Field

[0001] This invention relates to the field of hardware-in-the-loop (HIL) simulation technology, specifically to a hardware-in-the-loop simulation method and system for cluster target optimization based on a cascaded BP neural network. More particularly, it relates to a cluster target optimization simulation method for a hardware-in-the-loop simulation system, which mainly uses a cascaded BP neural network to achieve intelligent optimization of the antenna group scheme for the inner field simulation of radar cluster targets. Background Technology

[0002] With the rapid development of technologies such as electronic information, artificial intelligence, and unmanned systems, intelligent swarm warfare, represented by drone swarms, is gradually moving from conceptual research to practical application. The concept of distributed collaborative warfare is bringing disruptive changes to traditional combat models. Correspondingly, facing increasingly complex battlefield electromagnetic environments, accurately analyzing and identifying complex swarm target information has become a key issue in radar system development. For radar hardware-in-the-loop simulation systems, achieving swarm target simulation under limited indoor simulation resources is crucial for verifying radar performance.

[0003] Traditional radar hardware-in-the-loop (HILL) systems, which use a signal simulation source and a fed array for position control, can only generate a single target and connect to a single antenna output per channel. This means that increasing the number of targets requires adding more hardware channels, and the high cost and complexity limit the improvement of cluster target simulation capabilities. Existing improved methods use digital array units to synthesize multi-target signals in the digital domain and employ digital amplitude and phase modulation to independently control the position of each target within the antenna array, replacing the traditional precise position control module. This breaks through the limitation of the number of simulated targets per channel. However, the number of antennas that can output simultaneously is still determined by the number of channels, and the range of simulating targets on the array is limited by the hardware resources of the digital array units.

[0004] Patent document CN112947119B discloses a radio frequency hardware-in-the-loop (HILL) simulation digital array implementation system and method. This system uses a signal simulation unit to perform amplitude and phase modulation on signals in the baseband digital domain, enabling time-domain superposition output and individual position control of multiple target signals within a single triplet. It can simulate radar surface targets within the triplet range. However, for the simulation requirements of large-scale cluster targets, this method still has implementation limitations; limited digital array resources restrict the simultaneous signal simulation of large-scale triplet antennas.

[0005] In the final stages of radar hardware-in-the-loop simulation, as the target approaches, the line-of-sight angle of the cluster targets within the tested radar beam continuously expands. This means the positional distribution of each target on the antenna array rapidly spreads. The more dispersed the cluster spacing and the closer the remaining distance when the simulation stops, the larger the spread range in the final stage. Therefore, to ensure complete simulation of the cluster targets, a sufficient number of signal channels must be set up to support a sufficiently large antenna array. Covering the entire array area typically requires hundreds of channels, which significantly increases the demand for digital array elements, making it difficult to achieve for experimental systems with limited resources. Under this spread state, the value of peripheral targets within the cluster to the radar rapidly decreases, while core targets dominate the experimental results.

[0006] Therefore, in the final stage of the simulation where the range exceeds the limit, the most critical part of the cluster target is selected for simulation. Based on the possible combinations of array antennas, multiple alternative simulation ranges can be formed. The effectiveness of these alternative simulation schemes varies. How to quickly and accurately select the optimal scheme in the simulation process is the problem that this invention focuses on. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the purpose of this invention is to provide a hardware-in-the-loop simulation method and system for cluster target optimization based on cascaded BP neural networks.

[0008] A hardware-in-the-loop simulation method for cluster target optimization based on a cascaded BP neural network, provided by the present invention, includes:

[0009] Step S1: Determine the cluster target value scoring index according to the simulation task requirements;

[0010] Step S2: Based on the cluster target value scoring index and simulation experience, construct a scoring sample training set with a wide scoring distribution;

[0011] Step S3: Construct the initial cascaded BP neural network and train the cascaded BP neural network using the scoring sample training set;

[0012] Step S4: Determine whether the current cluster target distribution exceeds the range of the simulated antenna group. If so, traverse all antenna groups of the simulated cluster target through positional relationships, calculate the scoring index of each alternative scheme, and then proceed to step S5; otherwise, directly simulate all targets.

[0013] Step S5: Input the scoring indicators of all candidate schemes into the trained cascaded BP neural network, output the scoring results of each scheme, and take the one with the highest score as the final simulation scheme.

[0014] Preferably, step S1 includes:

[0015] Step S1.1: Set up a simulation target value analysis index system, and determine the quantitative indicators that affect the priority score of the antenna group scheme according to the specific requirements of the simulation test model;

[0016] Step S1.2: Determine the formula for calculating the overall score index of the scheme from all the target parameters included in the scheme.

[0017] Preferably, the cascaded BP neural network includes an input layer, two hidden layers, and an output layer;

[0018] The number of hidden layer neural units are 4 and 2, respectively;

[0019] The term cascading refers to each layer in a network receiving input from all the layers preceding it.

[0020] Preferably, step S4 includes numbering all targets. When the cluster exceeds the limit, all possible antenna group selections should be quickly listed according to their relative positions, and the target numbers included in each scheme should be given.

[0021] Preferably, during the simulation phase, after the radar system is powered on at the end of the simulation, i.e., in the first frame when the target simulation command is about to be updated, it is determined in real time whether the cluster target exceeds the limit. When it exceeds the limit, the network is used to score and optimize the selection scheme of all antenna groups when the limit is exceeded. In each subsequent frame of the simulation, the target included in the optimized scheme of the previous frame is the latest cluster, and it is re-determined whether it has spread beyond the limit again. If so, the optimization steps are repeated.

[0022] A hardware-in-the-loop simulation system for cluster target optimization based on a cascaded BP neural network, according to the present invention, includes:

[0023] Module M1: Determines the cluster target value scoring index based on the simulation task requirements;

[0024] Module M2: Based on the cluster target value scoring index and simulation experience, construct a scoring sample training set with a wide scoring distribution;

[0025] Module M3: Construct the initial cascaded BP neural network and train the cascaded BP neural network using the scoring sample training set;

[0026] Module M4: Determine whether the current cluster target distribution exceeds the range of the simulated antenna group. If so, traverse all antenna groups of the simulated cluster target by positional relationship, calculate the scoring index of each alternative scheme, and then execute Module M5; otherwise, directly simulate all targets.

[0027] Module M5: Input the scoring metrics of all alternative schemes into the trained cascaded BP neural network, output the scoring results of each scheme, and take the one with the highest score as the final simulation scheme.

[0028] Preferably, the module M1 includes:

[0029] Module M1.1: Set up a simulation target value analysis index system, and determine quantitative indicators that affect the priority score of antenna group schemes based on the specific requirements of the simulation test model;

[0030] Module M1.2: Determines the formula for calculating the overall score index of the scheme from all the target parameters included in the scheme.

[0031] Preferably, the cascaded BP neural network includes an input layer, two hidden layers, and an output layer;

[0032] The number of hidden layer neural units are 4 and 2, respectively;

[0033] The term cascading refers to each layer in a network receiving input from all the layers preceding it.

[0034] Preferably, module M4 includes numbering all targets, and when the cluster exceeds the limit, it should quickly list all possible antenna group selections based on their relative positions and give the target numbers included in each scheme.

[0035] Preferably, during the simulation phase, after the radar system is powered on at the end of the simulation, in the first frame when the target simulation command is about to be updated, it is determined in real time whether the cluster target exceeds the limit. When it exceeds the limit, the network is used to score and optimize the selection scheme of all antenna groups when the limit is exceeded. In each subsequent frame of the simulation, the target included in the optimized scheme of the previous frame is the latest cluster, and it is re-determined whether it has spread beyond the limit again. If so, the optimization is repeated.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] This invention utilizes a cascaded backpropagation (BP) neural network to score and optimize cluster target simulation schemes. By fully training the network with a scoring training set, the network parameters are made to fit as closely as possible to the criteria for judging the priority value of the simulated model for the cluster target. This enables rapid scoring and ranking of cluster target over-limit simulation schemes during the simulation process, ultimately achieving optimal cluster target simulation results under limited channel resources. It achieves optimal cluster target simulation results under limited existing simulation resources, intelligently decides the best simulation antenna group, and features a simple network structure and fast processing speed. Attached Figure Description

[0038] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0039] Figure 1 The preferred semi-physical simulation basic working block diagram for the cluster target of this invention;

[0040] Figure 2 This is a schematic diagram of the cascaded BP neural network structure in this invention;

[0041] Figure 3 This is a schematic diagram of the real-time scoring and optimization process of a network in a frame of simulated terminal cluster diffusion in an embodiment of the present invention.

[0042] Figure 4 This is a schematic diagram of the real-time scoring optimization simulation output result of a certain frame of the network in the simulation of terminal cluster diffusion in an embodiment of the present invention. Detailed Implementation

[0043] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0044] The implementation of this invention's hardware-in-the-loop simulation method for cluster target optimization based on a cascaded BP neural network includes a scoring index system and calculation method, a cascaded BP neural network, a training set and network training, real-time scheme generation and index calculation, network scoring, and scheme optimization. Specifically, the scoring index system and calculation method are determined first, serving as the basis for subsequently constructing the network training set and network scoring input. A cascaded BP neural network is established, where each neuron receives input from all preceding layers, providing stronger fitting and generalization capabilities. Based on the target value evaluation index of the model and simulation experience, a sufficiently distributed scoring sample is constructed, and the initial network is thoroughly trained. During the simulation process, the cluster diffusion exceeding the limit is judged in real time, all possible simulation antenna group schemes are traversed, and their scoring indices are calculated. All scheme indices are input into the trained network to quickly score and rank the candidate schemes, and simulation is performed using the highest-scoring scheme.

[0045] Example 1

[0046] According to the present invention, a semi-physical simulation method for cluster target optimization based on a cascaded BP neural network is provided, such as... Figure 1 As shown, it includes:

[0047] Step S1: Determine the cluster target value scoring index according to the simulation task requirements. Step S1 includes:

[0048] Step S1.1: Set up a simulation target value analysis index system, and clarify the quantitative indicators that affect the priority score of the antenna group scheme according to the specific requirements of the simulation test model.

[0049] Step S1.2: Determine the formula for calculating the overall score index of the scheme from all the target parameters included in the scheme.

[0050] Step S2: Based on the cluster target value scoring index and simulation experience, construct a scoring sample training set with a wide scoring distribution. For example, set uniformly distributed scoring samples in the range of 0 to 100 points, and determine the scoring index values ​​and corresponding expected scores based on human evaluation experience and model scoring standards, thereby forming the scoring training set.

[0051] Step S3: Construct the initial cascaded BP neural network and train it using the scoring sample training set. Determine the network's input and output formats based on the characteristics of the preferred scoring task selected in the simulation, and complete the network layer construction and parameter initialization. Figure 2 As shown, the cascaded backpropagation (BP) neural network includes an input layer, two hidden layers, and an output layer. The number of neurons in the hidden layers are 4 and 2, respectively. Cascading means that each layer in the network receives input from all the preceding layers.

[0052] Step S4: Determine whether the current cluster target distribution exceeds the range of the simulated antenna group. If it does, quickly traverse all possible antenna groups of the simulated cluster targets based on their positional relationships, calculate the scoring index of each alternative scheme, and then proceed to step S5. If not, directly simulate all targets. Specifically, number all targets. When the cluster exceeds the limit, quickly list all possible antenna group selections based on their relative positional relationships and provide the target numbers included in each scheme.

[0053] Step S5: Input the scoring indicators of all candidate schemes into the trained cascaded BP neural network, quickly output the scoring results of each scheme, and take the one with the highest score as the final simulation scheme. That is, the scoring indicators calculated for each scheme are input into the trained cascaded BP neural network, which outputs the score corresponding to each scheme, and selects the target number of the scheme with the highest score for simulation. Specifically, after the radar system is powered on at the end of the simulation, in the first frame before the target simulation command is updated, it is determined in real time whether the cluster target exceeds the limit. If so, the network is used to select the scheme for all antenna groups when the limit is exceeded and optimize the scoring. In each subsequent frame of the simulation, the targets included in the optimized scheme of the previous frame are the latest cluster, and it is re-determined whether the spread exceeds the limit again. If so, the optimization steps are repeated.

[0054] Furthermore, the hardware-in-the-loop simulation method for cluster target optimization based on cascaded BP neural networks of the present invention is specifically described below in conjunction with practical application scenarios:

[0055] Suppose we have a four-channel output digital array unit that can simultaneously output signals from four antennas in the array. With the help of a switching matrix, it is possible to simulate targets using any four-element antenna on the array. The number of cluster targets is 10. At the end of the simulation, the cluster spreads beyond the range of the four elements, so it is necessary to select the optimal four-element antenna scheme. According to the method of this invention, a semi-physical simulation of target optimization is performed.

[0056] First, before simulation, the target value scoring indicators need to be clearly defined according to the specific requirements of the model. This embodiment takes four scoring indicators as an example, including the proportion of targets in the previous frame (C), target dispersion (D), target power (P), and target velocity (V). The corresponding indicator calculation formulas are defined as follows:

[0057]

[0058] Where, num now num represents the number of objectives included in the current plan. pre |T represents the number of targets simulated in the previous frame. n T m | represents the distance between any two targets in the current scheme, d is the distance between antennas, and P i v i P represents the power and velocity of the i-th target included in the current scheme. j v j This represents the power and velocity of the j-th target in the previous frame simulation.

[0059] Then, a training set of scoring samples is constructed based on the scoring indicators, as shown in Table 1:

[0060] Table 1

[0061]

[0062]

[0063] The scoring range is 0 to 100, and the scoring follows these principles:

[0064] 1) Ensure that the target simulation between frames is as continuous and smooth as possible. Schemes with a higher proportion of targets from the previous frame will receive higher scores and should be given priority.

[0065] 2) Under the same conditions, the scheme with a smaller diffusion trend and lower cluster target dispersion scores higher and should be given priority.

[0066] 3) Schemes with higher overall power levels in the cluster target receive higher scores and should be prioritized.

[0067] 4) Solutions with higher overall speed levels of the cluster target receive higher scores and should be given priority.

[0068] Next, based on the target optimal input scoring index and output scoring value data format, the network structure settings, such as the number of neurons in the input layer, hidden layer, and output layer, and the connection methods between layers, are determined to complete the initial construction of the cascaded BP neural network. Appropriate training parameters, such as the network training function, performance function, learning rate, and maximum number of iterations, are selected. The network is trained using the training set, and the training performance results are monitored. The training parameters are adjusted and optimized to minimize the training error and ensure the network fits the desired scoring standard as closely as possible. Figure 2 As shown, the network consists of an input layer, two hidden layers, and an output layer. The number of neurons in the hidden layers are 4 and 2, respectively. Cascading means that each layer in the network receives input from all the preceding layers. The network takes four rating metrics as input and outputs a single rating value. The previously constructed training set samples are input into the network for training, allowing the network parameters to fully fit the desired rating criteria.

[0069] Then, after the network training is complete, it can be used for real-time scoring in simulations. A real-time simulation scheme generation method is designed so that, at the beginning of each frame after the radar is powered on, the relationship between the antenna groups and the target position can be used to determine if the cluster exceeds limits, and all antenna group schemes and corresponding scoring indicators can be automatically generated. For example... Figure 3 As shown in the figure, taking a frame of the simulated cluster diffusion at the end as an example, the real-time scoring and optimization process of the network is introduced. In the figure, A represents the antenna and T represents the target. The simulation range of the four-tuple in the previous frame included targets numbered 1, 2, 3, 8, 9, and 10 in the cluster. In the current frame, due to the rapid diffusion of the cluster, the original four-tuple can only simulate targets numbered 1, 2, and 10. All possible four-tuple selection schemes at this time are listed and the corresponding scoring indicators are calculated. All possible four-tuple selection schemes are shown in Table 2:

[0070] Table 2

[0071]

[0072] Finally, all the obtained scheme scoring metrics are input into the trained scoring network to obtain the scheme scores and ranking results. The scheme with the highest score is selected to update the target simulation system control commands, and the target numbers included in that scheme are updated to the latest cluster. This completes the target optimization simulation within one frame. Figure 4 As shown, the scoring performance of the network was simulated and verified in MATLAB. The scoring indicators of the above six alternative schemes were input into the trained neural network, thereby determining that the target simulation for this frame should select the quadruplet with antenna numbers A5, A6, A1, and A7. The network scoring results met the predetermined scoring criteria, the scheme ranking results met expectations, and the scoring output speed was fast.

[0073] During the research and development of this invention, constructing the training set required a comprehensive integration of target value evaluation criteria and simulation engineering experience, as well as the design of a sufficiently large number of representative scoring samples. This resulted in a relatively complex training process. However, the neural network was pre-trained before simulation, and the training steps did not need to be repeated if the evaluation criteria remained unchanged. Using the network for real-time scoring optimization was very convenient and efficient. This invention fully learns the simulation value scoring relationship of cluster targets through a cascaded BP neural network, and then quickly outputs the simulation value scores of each simulation candidate scheme during the simulation process. It accurately ranks and optimizes all possible antenna group schemes, providing the optimal solution for the radar hardware-in-the-loop simulation system to simulate cluster targets with excessive diffusion.

[0074] The purpose of this invention is to overcome the problems of important target loss and inter-frame target mutation when existing simulation systems face the problem of cluster target diffusion exceeding limits. It provides an antenna group optimization method for simulating cluster targets in a radar hardware-in-the-loop simulation system. By radiating target signals through antenna groups on the antenna array, a trained cascaded BP neural network is used to output scores of all antenna group schemes involved in the cluster range, thereby obtaining the antenna group scheme with the highest simulation value, improving the continuity of cluster target simulation, and making full use of the simulation capabilities of limited system resources.

[0075] Example 2

[0076] The present invention also provides a semi-physical simulation system for cluster target optimization based on a cascaded BP neural network. The semi-physical simulation system for cluster target optimization based on a cascaded BP neural network can be implemented by executing the process steps of the semi-physical simulation method for cluster target optimization based on a cascaded BP neural network. That is, those skilled in the art can understand the semi-physical simulation method for cluster target optimization based on a cascaded BP neural network as a preferred embodiment of the semi-physical simulation system for cluster target optimization based on a cascaded BP neural network.

[0077] According to the present invention, a hardware-in-the-loop simulation system for cluster target optimization based on a cascaded BP neural network includes: Module M1: determining cluster target value scoring indicators according to simulation task requirements; Module M1 includes: Module M1.1: setting a simulation target value analysis index system, and determining quantitative indicators affecting the priority scoring of antenna group schemes according to the specific requirements of the simulation test model; Module M1.2: determining the formula for calculating the overall score index of the scheme from all target parameters included in the scheme; Module M2: constructing a scoring sample training set with a wide score distribution based on the cluster target value scoring indicators and simulation experience; Module M3: constructing an initial cascaded BP neural network, and training the cascaded BP neural network through the scoring sample training set; the cascaded BP neural network includes an input layer, two hidden layers, and an output layer; the number of neurons in the hidden layers are 4 and 2, respectively; the cascading refers to each layer in the network receiving input from all the preceding layers. Module M4: Determines whether the current cluster target distribution exceeds the range of the simulated antenna groups. If so, it iterates through all antenna groups of the simulated cluster targets based on their positional relationships, calculates the scoring index of each alternative scheme, and then executes Module M5. If not, it directly simulates all targets. Module M4 includes numbering all targets; when the cluster exceeds the limit, it quickly lists all possible antenna group selections based on their relative positional relationships and provides the target numbers included in each scheme. Module M5: Inputs the scoring indexes of all alternative schemes into a trained cascaded BP neural network, outputs the scoring results of each scheme, and takes the highest score as the final simulation scheme.

[0078] During the simulation phase, after the radar system is powered on at the end of the simulation, in the first frame before the target simulation command is updated, it is determined in real time whether the cluster target exceeds the limit. When it exceeds the limit, the network is used to score and optimize the selection scheme of all antenna groups when the limit is exceeded. In each subsequent frame of the simulation, the target included in the optimized scheme of the previous frame is the latest cluster, and it is re-determined whether it has spread beyond the limit again. If so, the optimization is repeated.

[0079] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0080] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A hardware-in-the-loop simulation method for cluster target optimization based on a cascaded BP neural network, characterized in that, include: Step S1: Determine the cluster target value scoring index according to the simulation task requirements; Step S2: Based on the cluster target value scoring index and simulation experience, construct a scoring sample training set with a wide scoring distribution; Step S3: Construct the initial cascaded BP neural network and train the cascaded BP neural network using the scoring sample training set; Step S4: Determine whether the current cluster target distribution exceeds the range of the simulated antenna group. If so, traverse all antenna groups of the simulated cluster target through positional relationships, calculate the scoring index of each alternative scheme, and then proceed to step S5; otherwise, directly simulate all targets. Step S5: Input the scoring indicators of all candidate schemes into the trained cascaded BP neural network, output the scoring results of each scheme, and take the one with the highest score as the final simulation scheme.

2. The hardware-in-the-loop simulation method for cluster target optimization based on cascaded BP neural networks according to claim 1, characterized in that, Step S1 includes: Step S1.1: Set up a simulation target value analysis index system, and determine the quantitative indicators that affect the priority score of the antenna group scheme according to the specific requirements of the simulation test model; Step S1.2: Determine the formula for calculating the overall score index of the scheme from all the target parameters included in the scheme.

3. The hardware-in-the-loop simulation method for cluster target optimization based on cascaded BP neural networks according to claim 1, characterized in that, The cascaded BP neural network includes an input layer, two hidden layers, and an output layer; The number of hidden layer neural units are 4 and 2, respectively; The term cascading refers to each layer in a network receiving input from all the layers preceding it.

4. The hardware-in-the-loop simulation method for cluster target optimization based on cascaded BP neural networks according to claim 1, characterized in that, Step S4 includes numbering all targets. When the cluster exceeds the limit, all possible antenna group selections should be quickly listed according to their relative positions, and the target numbers included in each scheme should be given.

5. The hardware-in-the-loop simulation method for cluster target optimization based on cascaded BP neural networks according to claim 1, characterized in that, During the simulation phase, after the radar system is powered on at the end of the simulation, in the first frame before the target simulation command is updated, it is determined in real time whether the cluster target exceeds the limit. When it exceeds the limit, the network is used to score and optimize the selection scheme of all antenna groups when the limit is exceeded. In each subsequent frame of the simulation, the target included in the optimized scheme of the previous frame is the latest cluster, and it is re-determined whether it has spread beyond the limit again. If so, the optimization steps are repeated.

6. A hardware-in-the-loop simulation system for cluster target optimization based on a cascaded BP neural network, characterized in that, include: Module M1: Determines the cluster target value scoring index based on the simulation task requirements; Module M2: Based on the cluster target value scoring index and simulation experience, construct a scoring sample training set with a wide scoring distribution; Module M3: Construct the initial cascaded BP neural network and train the cascaded BP neural network using the scoring sample training set; Module M4: Determine whether the current cluster target distribution exceeds the range of the simulated antenna group. If so, traverse all antenna groups of the simulated cluster target by positional relationship, calculate the scoring index of each alternative scheme, and then execute Module M5; otherwise, directly simulate all targets. Module M5: Input the scoring metrics of all alternative schemes into the trained cascaded BP neural network, output the scoring results of each scheme, and take the one with the highest score as the final simulation scheme.

7. The hardware-in-the-loop simulation system for cluster target optimization based on a cascaded BP neural network according to claim 6, characterized in that, The module M1 includes: Module M1.1: Set up a simulation target value analysis index system, and determine quantitative indicators that affect the priority score of antenna group schemes based on the specific requirements of the simulation test model; Module M1.2: Determines the formula for calculating the overall score index of the scheme from all the target parameters included in the scheme.

8. The hardware-in-the-loop simulation system for cluster target optimization based on a cascaded BP neural network according to claim 6, characterized in that, The cascaded BP neural network includes an input layer, two hidden layers, and an output layer; The number of hidden layer neural units are 4 and 2, respectively; The term cascading refers to each layer in a network receiving input from all the layers preceding it.

9. The hardware-in-the-loop simulation system for cluster target optimization based on a cascaded BP neural network according to claim 6, characterized in that, The module M4 includes numbering all targets, and when the cluster exceeds the limit, it should quickly list all possible antenna group selections based on their relative positions and provide the target numbers included in each scheme.

10. The hardware-in-the-loop simulation system for cluster target optimization based on a cascaded BP neural network according to claim 6, characterized in that, During the simulation phase, after the radar system is powered on at the end of the simulation, in the first frame before the target simulation command is updated, it is determined in real time whether the cluster target exceeds the limit. When it exceeds the limit, the network is used to score and optimize the selection scheme of all antenna groups when the limit is exceeded. In each subsequent frame of the simulation, the target included in the optimized scheme of the previous frame is the latest cluster, and it is re-determined whether it has spread beyond the limit again. If so, the optimization is repeated.

Images