Active phased-array antenna deformation reconstruction method based on RMBO-XGBoost

Abstract

The invention discloses an active phased-array antenna deformation reconstruction method based on RMBO-XGBoost. The method comprises the following steps: forming individuals of a population in an RMBO algorithm by using parameters of an XGBoost algorithm, and setting an initial population of the RMBO algorithm; performing an exploration process of the RMBO algorithm to obtain a population after first update; forming a plurality of XGBoost networks to be trained; each XGBoost network is trained and tested, and an optimal XGBoost network and a first optimal individual are obtained; developing an RMBO algorithm by using the population updated for the first time to obtain a population updated for the second time; the XGBoost network is trained again, and an optimal XGBoost network and a second optimal individual after training are obtained; judging whether a threshold value is met or not, and if yes, inputting the to-be-reconstructed structural strain data into the current XGBoost network to obtain deformation data; and if not, returning to continue execution. According to the invention, the RMBO algorithm is combined to carry out parameter optimization on the XGBoost algorithm, the prediction precision of the XGBoost algorithm is improved, and the structure displacement field of the antenna structure under the load can be accurately obtained.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of antenna structure health monitoring, and particularly relates to an active phased array antenna deformation reconstruction method based on RMBO-XGBoost, which can be used for health monitoring of an active phased array antenna structure in a service environment and displacement field reconstruction of active phased array antenna structure deformation under service load. BACKGROUND

[0002] With the progress of national defense science and technology, radar equipment is developing in the direction of high integration, high performance and high mobility, which promotes the active phased array antenna to continuously improve the requirements for high-density array elements and lightweight, and causes the active phased array antenna structure to be increasingly sensitive to load and more likely to cause electrical performance decline due to environmental load. However, in the service environment, the structural deformation of the active phased array antenna is difficult to accurately evaluate, and thus accurate electrical performance compensation cannot be performed, which affects the performance of the radar. In order to accurately evaluate the deformation amount of the antenna structure, a high-precision active phased array antenna deformation reconstruction algorithm needs to be developed to accurately evaluate the structural displacement of the antenna in real time.

[0003] The active phased array antenna deformation reconstruction method is crucial for antenna structure monitoring, and the accuracy of the reconstruction algorithm affects the compensation of the electrical performance of the antenna and further affects the detection performance of the radar. The traditional structural deformation reconstruction algorithm, on the one hand, cannot meet the real-time requirement of antenna structure displacement reconstruction, that is, the displacement field of the antenna structure changes in real time under external load, and the real-time requirement of the structural displacement reconstruction method is high; on the other hand, the structural displacement field reconstruction precision is insufficient, and the deformation state of the antenna structure cannot be accurately reflected, which further affects the electrical compensation of the antenna. Therefore, it is an urgent problem to be solved to research a high-precision real-time active phased array antenna structural displacement field reconstruction algorithm and obtain real-time displacement field data of the antenna structure. SUMMARY

[0004] In order to solve the above problems in the prior art, the application provides an active phased array antenna deformation reconstruction method based on RMBO-XGBoost, which can accurately reconstruct the deformation displacement field of the active phased array antenna structure in real time and ensure the reliable service of the active phased array antenna. The technical problem to be solved by the application is solved by the following technical scheme: The application provides an active phased array antenna deformation reconstruction method based on RMBO-XGBoost, comprising: S1: using the parameters of the XGBoost algorithm to compose a plurality of individuals of a population in the RMBO algorithm, and setting an initial population of the RMBO algorithm; S2: performing an exploration process of the RMBO algorithm: generating a random number rand 1, determining whether the random number rand1 is less than a determination factor If yes, a small population search is performed to obtain a first updated population, and if no, a large population search is performed to obtain the first updated population; S3: multiple individuals in the first updated population are brought into an XGBoost network for parameter setting to form multiple XGBoost networks to be trained; S4: each XGBoost network is trained and tested by using an active phased array antenna data set to obtain an optimal XGBoost network after training and a corresponding first optimal individual; S5: a development process of the RMBO algorithm is performed by using the first updated population: a random number rand2 is generated, and it is determined whether the random number rand2 is less than a determination factor If yes, a small population attack is performed to obtain a second updated population, and if no, a large population attack is performed to obtain the second updated population; S6: multiple individuals in the second updated population are brought into the XGBoost network for retraining to obtain an optimal XGBoost network after training and a corresponding second optimal individual; S7: it is determined whether an RMSE value corresponding to the optimal XGBoost network after training meets a set threshold value, if yes, the second optimal individual is taken as a final parameter of the XGBoost network, and structure strain data to be reconstructed is input into the current XGBoost network to obtain corresponding deformation data; and if no, step S2 is returned to continue execution.

[0005] In an embodiment of the present application, the S1 comprises: S1.1: parameters num_trees, subsample, eta, max_depth and min_child_weight of the XGBoost algorithm are used to form individuals of a population in the RMBO algorithm; S1.2: ranges of the parameters num_trees, subsample, eta, max_depth and min_child_weight are set, and initial values of the groups are randomly generated n to form an initial population of the RMBO algorithm; S1.3: parameters of the RMBO algorithm are initialized, and a maximum iteration number, a maximum population size and a determination factor are set.

[0006] In an embodiment of the present application, the S1.2 comprises: Set the range of num_trees to [100, 1000], the range of subsample to [0.1, 1], the range of eta to [0.1, 1], the range of max_depth to [3, 10], and the range of min_child_weight to [1, 10]. The value is randomly selected from the range of values ​​for each parameter. n The initial population for the RMBO algorithm consists of 1,000 individuals, and the specific representation of the population is as follows:

[0007] in, Indicates the current population, In population X, the first... i The first of the individuals j One element, In population X, the first... n The dim-th element in each individual; The expression is: , in, ub and lb These are the upper and lower bounds of the corresponding parameters in the XGBoost algorithm, respectively, and rand is a random number between 0 and 1.

[0008] In one embodiment of the present invention, S2 includes: Generate random numbers in the range [0~1] rand 1. And determine the random number. rand 1 and the determination factor The size, if A small population search is performed, randomly selecting 2-5 individuals from the current population as a small population. This small population is then used to update the population for the first time, resulting in the updated population. The update formula for the small population search is expressed as: , in, Indicates the first t In the nth iteration i The value after the first update for each individual Indicates the first t In the nth iteration i The value of each individual before its first update. p This represents the number of individuals selected from the small population in the small population search, and is a random number between 2 and 5. This indicates the first small population selected in the small population search. m Individual, It is an individual randomly selected from the current population; like Perform a large-scale population search, randomly selecting from the current population. q Each individual, acting as a large population, performs the first update on every individual in the current population, resulting in the population after the first update. The update formula for large population search is expressed as: , in, q This indicates the number of individuals from the selected large population in the large population search, with a value ranging from 10 to... n .

[0009] In one embodiment of the present invention, S4 includes: S4.1: Use a portion of the active phased array antenna dataset as a training set and another portion as a test set. The active phased array antenna dataset is a large amount of collected historical data of antenna structures. Each set of historical data of antenna structures includes structural strain data and corresponding real deformation data. S4.2: Using the training set to... n Each of the XGBoost networks to be trained is trained to obtain... n Each trained XGBoost network takes the structural strain data from the test set as input and outputs the corresponding predicted structural deformation value. S4.3: Calculate the root mean square error (RMSE) between the predicted structural deformation output of each trained XGBoost network and the corresponding real deformation data in the test set, and compare them. n The RMSE value of each trained XGBoost network is selected as F. obj The individual with the smallest XGBoost network is selected as the first optimal individual. .

[0010] In one embodiment of the present invention, S5 includes: Generate random numbers in the range [0~1] rand 2. Determine the random number. rand Is 2 less than the decision factor? ,like Randomly select from the population after the first update Individuals form a small population to carry out a small-population attack, obtaining a second updated population. The update formula for the small-population attack is expressed as: , in, Indicates the first t In the nth iterationi a value of the i-th individual before the second update, i.e. equal to the value after the first update, denotes a value of the i-th individual in the first iteration, t denotes a value of the i-th individual after the second update in the first iteration, i denotes a value of the i-th individual after the second update in the first iteration, denotes the first optimal individual in the current iteration, , T denotes the maximum number of iterations; denotes the i-th individual randomly selected from the population after the first update, m denotes the i-th individual randomly selected from the population after the first update, denotes the number of individuals in the small population selected in the small population attack, = 2~5; if , a large population attack is performed by randomly selecting individuals from the population after the first update to form a large population, and a population after the second update is obtained, wherein the update formula of the large population attack is represented as: , wherein, denotes the number of individuals in the small population selected in the large population attack.

[0011] In an embodiment of the present application, the S6 comprises: S6.1: the parameter value of the individual of the population after the second update is brought into the XGBoost network to complete parameter setting, forming a second XGBoost network to be trained; S6.2: the active phased array antenna data set is used to train and test each XGBoost network respectively, and the optimal XGBoost network after training and the corresponding second optimal individual are obtained.

[0012] In an embodiment of the present application, the S7 comprises: whether the RMSE value corresponding to the optimal XGBoost network after training is less than a set threshold value, if yes, the second optimal individual is taken as the final parameter of the XGBoost network, the final parameter is brought into the XGBoost network as a final trained XGBoost network, and the structure strain data to be reconstructed is input into the final trained XGBoost network to obtain corresponding deformation data; if no, the iteration number t+1 is returned to step S2, and the iteration operation is continued based on the population after the second update obtained in step S5.

[0013] Another aspect of the present application provides a storage medium, wherein a computer program is stored in the storage medium, and the computer program is used to execute the steps of the RMBO-XGBoost based active phased array antenna deformation reconstruction method in any one of the above embodiments.

[0014] Another aspect of the present application provides an electronic device, characterized in that comprising a memory and a processor, the memory has stored therein a computer program, the processor invokes the computer program in the memory to implement the steps of the RMBO-XGBoost-based active phased array antenna deformation reconstruction method according to any one of the above embodiments.

[0015] Compared with the prior art, the present application has the following beneficial effects: 1. For the structural deformation displacement field reconstruction of a large-aperture lightweight active phased array antenna, the present application proposes an RMBO-XGBoost-based active phased array antenna deformation reconstruction method, which can accurately obtain the structural displacement field of the antenna structure under load. In order to more effectively solve the problems of insufficient accuracy, poor real-time performance, and serious dependence on finite element model accuracy of traditional structural reconstruction algorithms, the present application uses the XGBoost algorithm to establish an antenna structural displacement prediction based on discrete strain data, and jointly optimizes the parameters of the XGBoost algorithm with the RMBO algorithm, thereby improving the prediction accuracy of the XGBoost. The RMBO-XGBoost-based active phased array antenna deformation reconstruction method is superior to existing methods in terms of structural displacement reconstruction accuracy and real-time performance.

[0016] 2. The present application uses the RMBO-XGBoost algorithm to realize real-time reconstruction of the structural deformation displacement field of the active phased array antenna, solves the problem of displacement field reconstruction accuracy and real-time performance of the active phased array antenna, and can provide a reference for health monitoring of large flexible structures, ensuring the reliable operation of large equipment, and having important significance for health monitoring of space deployable antennas, large-aspect-ratio unmanned aerial vehicles and other weapon equipment that are easily affected by environmental loads during service.

[0017] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is a flowchart of an RMBO-XGBoost-based active phased array antenna deformation reconstruction method provided by an embodiment of the present application; Figure 2 is a detailed flowchart of an RMBO-XGBoost-based active phased array antenna deformation reconstruction method provided by an embodiment of the present application; Figure 3 is a structural schematic diagram of an active phased array antenna provided by an embodiment of the present application; Figure 4 is a comparison diagram of actual measurement and RMBO-XGBoost algorithm results provided by an embodiment of the present application; Figure 5The RMBO-XGBoost algorithm error cloud map is actually measured. DETAILED DESCRIPTION

[0019] In order to further illustrate the technical means and effects taken by the present application to achieve the predetermined object, the RMBO-XGBoost based active phased array antenna morphing reconstruction method according to the present application is described in detail below in combination with the drawings and specific embodiments.

[0020] The foregoing and other technical contents, features and effects of the present application can be clearly presented in the detailed description of the specific embodiments below in combination with the drawings. Through the description of the specific embodiments, the technical means and effects taken by the present application to achieve the predetermined object can be more deeply and specifically understood. However, the attached drawings are provided for reference and illustration only, and are not intended to limit the technical solutions of the present application.

[0021] It should be noted that, in this document, relational terms such as first and second and the like can only be used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that these entities or operations exist in any such actual relationship or order. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a list of elements does not necessarily include only those elements, but can include other elements not expressly listed or included. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus including the element.

[0022] The present application provides an RMBO-XGBoost based active phased array antenna morphing reconstruction method, as shown in Figure 1 and Figure 2 The active phased array antenna morphing reconstruction method comprises: S1: using the parameters of the XGBoost algorithm to form a plurality of individuals of a population in the RMBO, and setting an initial population of the RMBO algorithm.

[0023] XGBoost (eXtreme Gradient Boosting) is an optimized implementation framework of gradient boosting tree (GBDT), which significantly improves training efficiency, accuracy and scalability by engineering optimization and algorithm improvement while maintaining the core idea of GBDT (improving prediction ability by integrating weak learners (trees)). The essence of XGBoost is an additive ensemble model, which iteratively trains multiple decision trees (base learners) to combine weak learners into strong learners, and the core goal is to minimize the loss function of the predicted value and the true value.

[0024] RMBO (Red-billed blue magpie optimizer) is a novel metaheuristic algorithm inspired by the cooperative hunting behavior of red-billed blue magpies (including searching for, attacking prey, and storing food). It simulates the cooperative searching and hunting behavior of red-billed blue magpies in groups, dynamically adjusts the search step size with adaptive coefficients to balance exploration and exploitation capabilities, and uses a food storage mechanism to retain historical optimal solutions to enhance stability. Its core objective is to efficiently solve high-dimensional, computationally expensive, non-convex, or discontinuous optimization problems.

[0025] In this embodiment, step S1 specifically includes: S1.1: The parameters num_trees, subsample, eta, max_depth, and min_child_weight of the XGBoost algorithm are used to construct the individuals in the RMBO algorithm population. The dimension of each individual is dim=5, meaning dim represents the number of elements in each individual. Then, the th element in the population... i Individual X i The expression is: .

[0026] It should be noted that each individual in the population consists of five elements: num_trees, subsample, eta, max_depth, and min_child_weight. The values ​​of these elements may differ for each individual.

[0027] Among the core parameters of the XGBoost algorithm, `num_trees`, `subsample`, `eta`, `max_depth`, and `min_child_weight` are key parameters controlling model complexity, training efficiency, and generalization ability. `num_trees` represents the number of decision trees (base learners) in the ensemble model and is one of the core parameters of gradient boosting. The XGBoost algorithm iteratively trains multiple trees, gradually correcting the prediction errors of preceding models. `subsample` represents the proportion of randomly sampled samples (sampling without replacement) when training a single tree, a row-level regularization method. `eta` represents the contribution weight of each tree to the final prediction result (scaling factor), a core parameter controlling the model's convergence speed. `max_depth` represents the maximum depth of a single decision tree, directly controlling the structural complexity of the tree. `min_child_weight` represents the minimum sum of sample weights in a node, used to control tree branching to prevent overfitting.

[0028] S1.2: Set the value ranges for parameters num_trees, subsample, eta, max_depth, and min_child_weight, and generate them randomly. n The initial values ​​form the initial population for the RMBO algorithm.

[0029] Specifically, the values ​​for num_trees are set to [100, 1000], subsample to [0.1, 1], eta to [0.1, 1], max_depth to [3, 10], and min_child_weight to [1, 10]. These values ​​are then randomly generated. n group initial value as n Each individual is formed by randomly selecting values ​​from the range of each parameter. n Individual, this n The initial population for the RMBO algorithm consists of 1,000 individuals. The population is represented as follows:

[0030] in, This represents the current population, which contains... n There are *dim* individuals, where each row in the matrix represents an individual, and each individual comprises *dim* elements. For example... In population X, the first... i The first of the individuals j One element, In population X, the first... n The dim-th element in each individual, in this embodiment, dim=5.

[0031] The expression is: , in, ub and lb These are the upper and lower bounds of the corresponding parameters in the XGBoost algorithm, respectively, and rand is a random number between 0 and 1. For example, if the current... If the corresponding element is num_trees, then its ub and lb The values ​​are 1000 and 100 respectively. If the current The corresponding element is max_depth, then its ub and lb The values ​​are 10 and 3 respectively.

[0032] S1.3: initialize the parameters of the RMBO algorithm and set the initial values, the parameters of the RMBO algorithm include the maximum number of iterations of the algorithm, the maximum population size and the determination factor, the maximum population size is equal to the number of individuals in the initial population. In this embodiment, the maximum number of iterations T = 200, the maximum population size n = 50, and the determination factor .

[0033] S2: perform the exploration process of the RMBO algorithm: generate a random number rand 1, determine whether the random number rand 1 is less than the determination factor , if yes, perform small population search to obtain the first updated population, if no, perform large population search to obtain the first updated population.

[0034] In this step, a random number rand 1 in the range of [0~1] is randomly generated and the random number rand 1 is determined to be less than the determination factor , when , small population search is performed, that is, 2~5 individuals in the current population are randomly selected as a small population for optimization, and the average value of all individuals in the small population is used as a reference for mobile optimization of the current population, that is, each individual in the current population is updated to obtain the first updated population, wherein the update formula of the small population search is represented as: , wherein, t represents the t th iteration, represents the first updated value of the t th individual in the i th iteration, represents the first updated value of the t th individual in the i th iteration, p represents the number of individuals in the small population selected in the small population search, which is a random number between 2 and 5, represents the m th individual in the small population selected in the small population search, is a randomly selected individual in the current population, that is, a random selection is made from 1 to n n ( n is the maximum population size) in order to increase randomness.

[0035] When , large population search is performed, that is, a random number qEach individual is considered part of a large population. The average value of all individuals in the large population is used as a reference for moving and optimizing the current population. That is, each individual in the current population undergoes an initial update to obtain the population after the first update. The update formula for large population search is expressed as: , in, q This indicates the number of individuals from the selected large population in the large population search, with a value ranging from 10 to... n .

[0036] S3: Input multiple individuals from the population after the first update into the XGBoost network for parameter setting, forming multiple XGBoost networks to be trained.

[0037] Specifically, in step S2, based on the random number rand 1 and the criterion factor The comparison results of the sizes are relevant to the population. n Each individual performed a small-population search or a large-population search, obtaining the population after the first update. In this step, the population from the first update in step S2 is... n Each individual (each individual includes a set of num_trees, subsample, eta, max_depth, and min_child_weight values) is fed into the XGBoost network to establish different parameters. n Each XGBoost network consists of a set of parameters, namely num_trees, subsample, eta, max_depth, and min_child_weight, for each updated individual. These parameters are fed into the XGBoost network to form the desired network structure. n An XGBoost network with different parameter values.

[0038] S4: Use the active phased array antenna dataset to train and test each XGBoost network separately, and obtain the XGBoost network with the best performance after training and the corresponding individual.

[0039] Specifically, the active phased array antenna dataset can be a large collection of historical antenna structure data. Each set of historical antenna structure data includes structural strain data and corresponding actual deformation data. The structural strain data is collected by strain sensors on the antenna structure. It is assumed that the antenna structure is equipped with… s If there are 1 sensor, then the structural strain data is: The displacement data are the data of target points on the structure, obtained by photogrammetry, assuming that there are [displacement points] on the structure. w For each target point, the deformation data .

[0040] Step S4 in this embodiment specifically includes: S4.1: Use a portion of the active phased array antenna dataset as the training set and the other portion as the test set.

[0041] S4.2: Using the training set to pair n Each of the XGBoost networks to be trained is trained to obtain... n Each trained XGBoost network then inputs structural strain data from the test set into each trained XGBoost network to output the corresponding structural deformation prediction value.

[0042] S4.3: Calculate the root mean square error (RMSE) between the predicted structural deformation output of each trained XGBoost network and the corresponding real deformation data in the test set, and compare them. n The RMSE value of each trained XGBoost network is selected as F. obj The individual with the smallest XGBoost network is selected as the first optimal individual. .

[0043] Specifically, assuming the test set has k Group test data, which includes k Group structural strain data E With the corresponding k Real deformation data D The above k The strain data of the group structure is input into the system trained in step S4.2. n In each XGBoost network, the trained XGBoost network outputs a corresponding predicted structural deformation value. The average root mean square error (RMSE) is calculated using the actual deformation data and the predicted structural deformation value, as follows: , in, express k The first set of test data i The actual deformation data of each test data pair Indicates the use of the first i The predicted structural deformation values ​​obtained from the structural strain data of each test data pair are RMSE, which represents the RMSE of the XGBoost network after training.

[0044] Therefore, the RMSE corresponding to each trained XGBoost network can be obtained using the above formula, i.e., the... n Compare these RMSE values. n The size of each RMSE value, storing the minimum RMSE value F. objXGBoost network with the minimum RMSE value is saved. food , i.e. the current parameter value of the XGBoost network with the minimum RMSE value is saved.

[0045] S5: Development process of the RMBO algorithm using the first updated population: generate a random number rand2, and determine whether the random number rand2 is less than the determination factor If yes, perform a small population attack to obtain a second updated population, and if no, perform a large population attack to obtain a second updated population.

[0046] Specifically, rand 2 is a randomly generated random number in the range [0~1], and when , a small population attack is performed, and individuals in the first updated population are randomly selected to form a small population for attack, and a second updated population is obtained, i.e. on the basis of the first optimal individual , the average value of individuals in the small population is used as a reference for population mobile optimization, and a CF parameter is introduced to control the attack strength, wherein the update formula of the small population attack is represented as: , wherein, represents the value of the t th individual before the second update in the i th iteration, i.e. equal to the value after the first update, represents the value of the t th individual after the second update in the i th iteration, represents the first optimal individual in the current iteration, , T represents the maximum number of iterations, which changes with the number of iterations and gradually increases with the increase of the number of iterations, represents the m th individual randomly selected from the first updated population, represents a randomly generated standard normal distribution random number (mean value is 0, and variance is 1), represents the number of individuals in the small population selected in the small population attack, =2~5.

[0047] If , randomly select individuals in the first updated population to form a large population for large population attack to obtain a second updated population, wherein the update formula of the large population attack is represented as: , wherein, a number of individuals selected from the large population attack, = 10 n .

[0048] S6: Bring the plurality of individuals in the second updated population into the XGBoost network for further training to obtain the XGBoost network with the optimal performance after training and the corresponding second optimal individual.

[0049] Step S6 of the embodiment specifically comprises the following steps: S6.1: Bring the individual parameters of the second updated population into the XGBoost network to complete parameter setting, forming a plurality of XGBoost networks to be trained.

[0050] This step is similar to the operation process of step S3, which will not be repeated here.

[0051] S6.2: Train and test each XGBoost network using the active phased array antenna dataset to obtain the XGBoost network with the optimal performance after training and the corresponding second optimal individual.

[0052] For details, please refer to the first training process of steps S4.1-S4.3, which will not be repeated here, and finally obtain the minimum RMSE value F obj corresponding to the XGBoost network and the second optimal individual.

[0053] S7: Determine whether the RMSE value corresponding to the optimal XGBoost network after training meets the set threshold, if so, take the second optimal individual as the final parameter of the XGBoost network, input the structure strain data to be reconstructed into the current XGBoost network to obtain the corresponding deformation data; if not, return to step S2 for continuous execution.

[0054] Specifically, determine whether the RMSE value corresponding to the optimal XGBoost network after training is less than the set threshold, if so, take the second optimal individual as the final parameter of the XGBoost network, input the final parameter into the XGBoost network as the final trained XGBoost network; input the structure strain data to be reconstructed into the final trained XGBoost network to obtain the corresponding deformation data; if not, iterate the number t+1, return to step S2 to continue the iteration operation based on the second updated population obtained in step S5.

[0055] It should be noted that if the current iteration number has reached the maximum iteration number T, the iteration is ended, the second optimal individual obtained after the current iteration is ended is taken as the final parameter of the XGBoost network, and the final parameter is input into the XGBoost network as the final trained XGBoost network.

[0056] In order to verify the effect of the active phased array antenna deformation reconstruction method based on RMBO-XGBoost proposed in the application, an active phased array antenna structure model as shown in Figure 3 The active phased array antenna structure model as shown in the drawing comprises four parts: an antenna back frame, an actuator, an antenna panel and a horn antenna array. The back frame is a support structure for installing T / R components, power modules and other wave control components; the actuator is installed on the back frame to cause the antenna panel to deform; the antenna panel is used to install the array element array; and the horn antenna array is used to transmit and receive electromagnetic waves.

[0057] In order to verify the effectiveness of the active phased array antenna deformation reconstruction method of the application, the actuator is used to apply a load to the antenna array surface to cause the antenna array surface to deform, then a digital photogrammetry tool is used to measure the antenna array surface to obtain the actual deformation of the antenna, and finally the active phased array antenna deformation reconstruction method based on RMBO-XGBoost proposed in the application is used to calculate the deformation of the antenna array surface, and the results of the two methods are compared, as shown in Figure 4 It can be found that the surface has high coincidence, which indicates that the method of the application can meet the reconstruction requirements of the antenna array surface.

[0058] Further, in order to quantitatively analyze the effect of the method proposed in the application, an error cloud map is drawn, as shown in Figure 5 It can be seen that the maximum positive error is not more than 16 mm, the maximum negative error is not more than 15 mm, the areas with large displacement errors are mainly concentrated in the edge area of the antenna array surface, which does not affect the radiation performance of the antenna, the reconstruction error of the central area of the antenna array surface is not more than ±4 mm, which indicates that the method of the application is very meaningful in the deformation reconstruction of the antenna structure.

[0059] For the structural deformation displacement field reconstruction of a large-aperture lightweight active phased array antenna, the application proposes an active phased array antenna deformation reconstruction method based on RMBO-XGBoost, which can accurately obtain the structural displacement field of the antenna structure under load. In order to more effectively solve the problems of insufficient accuracy, poor real-time performance and serious dependence on the accuracy of the finite element model of the traditional structure reconstruction algorithm, the application uses the XGBoost algorithm to establish an antenna structure displacement prediction based on discrete strain data, and combines the RMBO algorithm to optimize the parameters of the XGBoost algorithm, thereby improving the prediction accuracy of the XGBoost. The active phased array antenna deformation reconstruction method based on RMBO-XGBoost is superior to existing methods in terms of structural displacement reconstruction accuracy and real-time performance.

[0060] The application can realize real-time reconstruction of the structural deformation displacement field of the active phased array antenna by using the RMBO-XGBoost algorithm, solves the problems of the reconstruction accuracy and real-time performance of the displacement field of the active phased array antenna, can provide a reference for health monitoring of large flexible structures, guarantees reliable operation of large equipment, and has important significance for health monitoring of space deployable antennas, large-aspect-ratio unmanned aerial vehicles and other weapon equipment that are easily affected by environmental loads during service.

[0061] Still another embodiment of the application provides a storage medium, in which a computer program is stored, the computer program being used to execute the steps of the active phased array antenna deformation reconstruction method based on RMBO-XGBoost described in the above embodiments. In still another aspect of the application, an electronic device is provided, which includes a memory and a processor, the memory storing a computer program, and the processor calling the computer program in the memory to realize the steps of the active phased array antenna deformation reconstruction method based on RMBO-XGBoost described in the above embodiments. Specifically, the integrated modules realized in the form of software functional modules described above can be stored in a computer-readable storage medium. The software functional modules described above are stored in a storage medium, including a plurality of instructions for causing an electronic device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute part of the steps of the method described in each embodiment of the application. And the aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0062] The above is a further detailed description of the application in combination with specific preferred embodiments, and the specific implementation of the application cannot be limited to these descriptions. For ordinary skilled persons in the technical field to which the application belongs, some simple deductions or replacements can be made without departing from the concept of the application, and all of them should be regarded as falling within the protection scope of the application.

Claims

1. A method for deformable reconfiguration of an active phased array antenna based on RMBO-XGBoost, characterized in that, include: S1: Use the parameters of the XGBoost algorithm to form multiple individuals in the population of the RMBO algorithm, and set the initial population of the RMBO algorithm. S2: Perform the exploration process of the RMBO algorithm: generate random numbers rand 1. Determine whether the random number rand1 is less than the decision factor. If yes, then perform a small population search to obtain the population after the first update; otherwise, perform a large population search to obtain the population after the first update. S3: Input multiple individuals from the population after the first update into the XGBoost network for parameter setting, forming multiple XGBoost networks to be trained; S4: Use the active phased array antenna dataset to train and test each XGBoost network separately, and obtain the best XGBoost network after training and the corresponding first best individual. S5: The process of developing the RMBO algorithm using the population after the first update: Generate a random number rand2, and determine whether the random number rand2 is less than the decision factor. If yes, launch a small-scale attack to obtain the second updated population; otherwise, launch a large-scale attack to obtain the second updated population. S6: Input multiple individuals from the population after the second update into the XGBoost network for retraining to obtain the best XGBoost network and the corresponding second best individual after training. S7: Determine whether the RMSE value corresponding to the best XGBoost network after training meets the set threshold. If yes, use the second best individual as the final parameter of the XGBoost network, input the strain data of the structure to be reconstructed into the current XGBoost network, and obtain the corresponding deformation data. If no, return to step S2 to continue execution.

2. The active phased array antenna deformation and reconstruction method based on RMBO-XGBoost according to claim 1, characterized in that, S1 includes: S1.1: The parameters num_trees, subsample, eta, max_depth, and min_child_weight of the XGBoost algorithm are used to form the individuals in the population of the RMBO algorithm; S1.2: Set the ranges for parameters num_trees, subsample, eta, max_depth, and min_child_weight, and generate them randomly. n The initial values ​​form the initial population for the RMBO algorithm; S1.3: Initialize the parameters of the RMBO algorithm and set the maximum number of iterations, the maximum population size, and the decision factor.

3. The active phased array antenna deformation and reconstruction method based on RMBO-XGBoost according to claim 2, characterized in that, S1.2 includes: Set the range of num_trees to [100, 1000], the range of subsample to [0.1, 1], the range of eta to [0.1, 1], the range of max_depth to [3, 10], and the range of min_child_weight to [1, 10]. The value is randomly selected from the range of values ​​for each parameter. n The initial population for the RMBO algorithm consists of 1,000 individuals, and the specific representation of the population is as follows: in, Indicates the current population, In population X, the first... i The first of the individuals j One element, In population X, the first... n The dim-th element in each individual; The expression is: , in, ub and lb These are the upper and lower bounds of the corresponding parameters in the XGBoost algorithm, respectively, and rand is a random number between 0 and 1.

4. The active phased array antenna deformation and reconstruction method based on RMBO-XGBoost according to claim 2, characterized in that, S2 includes: Generate random numbers in the range [0~1] rand 1. And determine the random number. rand 1 and the determination factor The size, if A small population search is performed, randomly selecting 2-5 individuals from the current population as a small population. This small population is then used to update the population for the first time, resulting in the updated population. The update formula for the small population search is expressed as: ， in, Indicates the first t In the nth iteration i The value after the first update for each individual Indicates the first t In the nth iteration i The value of each individual before its first update. p This represents the number of individuals selected from the small population in the small population search, and is a random number between 2 and 5. This indicates the first small population selected in the small population search. m Individual, It is an individual randomly selected from the current population; like Perform a large-scale population search, randomly selecting from the current population. q Each individual, acting as a large population, performs the first update on every individual in the current population, resulting in the population after the first update. The update formula for large population search is expressed as: ， in, q This indicates the number of individuals from the selected large population in the large population search, with a value ranging from 10 to... n .

5. The active phased array antenna deformation and reconfiguration method based on RMBO-XGBoost according to claim 2, characterized in that, S4 includes: S4.1: Use a portion of the active phased array antenna dataset as a training set and another portion as a test set. The active phased array antenna dataset is a large amount of collected historical data of antenna structures. Each set of historical data of antenna structures includes structural strain data and corresponding real deformation data. S4.2: Using the training set to... n Each of the XGBoost networks to be trained is trained to obtain... n Each trained XGBoost network takes the structural strain data from the test set as input and outputs the corresponding structural deformation prediction value. S4.3: Calculate the root mean square error (RMSE) between the predicted structural deformation output of each trained XGBoost network and the corresponding real deformation data in the test set, and compare them. n The RMSE value of each trained XGBoost network is selected as F. obj The individual with the smallest XGBoost network is selected as the first optimal individual. .

6. The active phased array antenna deformation and reconstruction method based on RMBO-XGBoost according to claim 5, characterized in that, S5 includes: Generate random numbers in the range [0~1] rand 2. Determine the random number. rand Is 2 less than the decision factor? ,like Randomly select from the population after the first update Individuals form a small population to carry out a small-population attack, resulting in a second updated population. The update formula for the small-population attack is expressed as: ， in, Indicates the first t In the nth iteration i The value of an individual before the second update is equal to the value after the first update. Indicates the first t In the nth iteration i The value of each individual after the second update This represents the first optimal individual in the current iteration. , T Indicates the maximum number of iterations; This indicates that the first [number]th ... m Individual, This indicates the number of individuals in the small population selected in a small population attack. =2~5; like Randomly select from the population after the first update Individuals form a large population to launch a large-scale attack, resulting in a second updated population. The update formula for the large-scale attack is expressed as: ， in, This indicates the number of small population individuals selected in a large population attack.

7. The active phased array antenna deformation and reconstruction method based on RMBO-XGBoost according to claim 2, characterized in that, S6 includes: S6.1: Input the individual parameter values ​​of the population after the second update into the XGBoost network to complete the parameter setting and form the second XGBoost network to be trained; S6.2: Train and test each XGBoost network using the active phased array antenna dataset to obtain the best XGBoost network after training and the corresponding second best individual.

8. The active phased array antenna deformation and reconstruction method based on RMBO-XGBoost according to claim 2, characterized in that, S7 includes: Determine whether the RMSE value corresponding to the optimal XGBoost network after training is less than a set threshold. If so, use the second optimal individual as the final parameter of the XGBoost network and input the final parameter into the XGBoost network as the finally trained XGBoost network. Input the strain data of the structure to be reconstructed into the finally trained XGBoost network to obtain the corresponding deformation data. If not, the iteration count is t+1, return to step S2, and continue the iteration operation based on the population updated for the second time obtained in step S5.

9. A storage medium, characterized in that, The storage medium stores a computer program for executing the steps of the active phased array antenna deformation and reconfiguration method based on RMBO-XGBoost as described in any one of claims 1 to 8.

10. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor, when calling the computer program in the memory, implements the steps of the active phased array antenna deformation and reconfiguration method based on any one of claims 1 to 8.

Images