Dynamic response reconstruction-oriented wireless sensor dimensionality reduction sparse configuration evolution method

Abstract

The invention discloses a dynamic response reconstruction-oriented wireless sensor dimensionality reduction sparse configuration evolution method, and belongs to the technical field of wireless sensor networks. The method comprises the following steps: constructing a feature selection driving sparse optimization model based on compressed sensing, and optimizing a sensor configuration weight by taking an aggregation function of minimizing residual adaptive regularization and a quadratic form accumulation term as a target; a dimension reduction differential evolution algorithm is provided, the dimension of a decision vector is dynamically reduced through an adaptive restart strategy, and the convergence efficiency and the global search capability are improved; and finally, the number of sensors is remarkably reduced while high-precision prediction of unreachable node response is ensured. The method is suitable for health monitoring in complex projects such as spacecraft structures, and has the advantages of low configuration cost, high prediction precision and high applicability.

Description

Technical Field

[0001] This invention belongs to the field of wireless sensor network technology, specifically relating to a dimensionality reduction sparse configuration evolution method for wireless sensors oriented towards dynamic response reconstruction. Background Technology

[0002] Structural health monitoring systems are crucial for ensuring the safe operation of spacecraft in orbit. They monitor the response data of critical nodes through the deployment of wireless sensor networks to identify and assess structural condition. However, spacecraft structures generally exhibit lightweight design requirements, non-uniform geometric topologies, and extreme multiphysics coupling environments (such as high-energy particle radiation, drastic temperature changes, and high-frequency vibrations). This results in many critical load-bearing components (such as main load-bearing sections and propulsion interfaces) and sensitive areas (such as thermal protection systems and precision instrument mounting locations) being unable to directly mount sensors, forming "unreachable nodes." These nodes typically bear major loads or possess important functions, and their dynamic response data is crucial for assessing structural integrity, docking accuracy, and connection reliability. Therefore, achieving accurate prediction of their responses has become an urgent engineering challenge.

[0003] Existing wireless sensor optimization configuration methods are mostly based on modal kinetic energy ranking, modal confidence criteria, or evolutionary algorithms (such as particle swarm optimization). While these methods have achieved some success in overall structure monitoring, they generally rely on accurate finite element models. Model discrepancies can easily lead to insufficient generalization ability and make it difficult to focus on predicting the responses of unreachable nodes. Furthermore, traditional methods often suffer from low computational efficiency and a tendency to get trapped in local optima when dealing with high-dimensional, nonlinear, and strongly coupled dynamic characteristics.

[0004] In recent years, compressed sensing theory has provided a new approach to reconstructing original signals based on limited observation data, promoting the application of data-driven methods in structural response prediction. For example, while methods combining neural networks or Gaussian process regression can effectively capture the nonlinear characteristics of a system, they do not significantly reduce the number of sensors, failing to meet the practical requirements of spacecraft for sparse sensor configuration and lightweight design. Furthermore, existing sparse optimization models (such as L1 / L2 regularization) still have shortcomings in handling high-dimensional nonlinear feature interactions and balancing feature selection and prediction performance, and most studies have not fully considered the computational complexity issues in high-dimensional optimization, resulting in low algorithm efficiency.

[0005] Therefore, there is an urgent need to develop a sparse sensor optimization configuration method that can balance high prediction accuracy, low sensor configuration quantity, strong generalization ability and high computational efficiency to support the practical engineering application of spacecraft structural health monitoring systems. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a dimensionality reduction sparse configuration evolution method for wireless sensors oriented towards dynamic response reconstruction. Combining feature selection and dimensionality reduction differential evolution strategies, it has advantages such as significantly reducing the number of wireless sensor configurations, obtaining the optimal wireless sensor configuration positions, and maintaining high-precision unreachable node structure response prediction capabilities.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A dimension-reduction sparse configuration evolutionary method for wireless sensors oriented towards dynamic response reconstruction, the method comprising:

[0009] Step 1: Determine the number of candidate nodes and the sensor type to be configured for each candidate node, number the sensor locations, and set the sensor weight threshold, compression ratio threshold, and maximum number of evaluations.

[0010] Step 2: Construct a state-space model of the spacecraft structure, perform finite element simulation by applying random loads of different sizes, and generate a dynamic response sample dataset;

[0011] Step 3: Based on the response sample dataset, establish the mapping relationship between the target node response and the candidate node response, and construct a sparse optimization model with the optimization objective of minimizing the aggregation function of the fusion residual adaptive regularization and the quadratic cumulative term.

[0012] Step 4: Solve the sparse optimization model using the dimensionality reduction differential evolution algorithm to obtain the optimal configuration weights for multiple sensor positions;

[0013] Step 5: Determine if the maximum number of evaluations has been reached. If not, return to step 4 to continue iterating; otherwise, proceed to step 6.

[0014] Step 6: Compare the weight threshold with the optimal configuration weight of multiple sensor positions, filter out key sensor positions, eliminate redundant positions, and calculate the compression ratio of the current sensor configuration.

[0015] Step 7: If the compression ratio is greater than the compression ratio threshold, retain the key sensor positions and reduce the dimension of the optimization variables, then return to step 4; otherwise, output the optimal sensor configuration scheme.

[0016] Furthermore, in step 1, the sensor types include: measuring velocity and displacement.

[0017] Furthermore, in step 2, the state-space model is derived from the second-order system equations of structural dynamics and is used to simulate the displacement and velocity response of the structure under external excitation; the random load is composed of a reference load vector and Gaussian white noise superimposed on it.

[0018] Furthermore, in step 3, the sparse optimization model sets the configuration weights corresponding to the sensor positions of each candidate node as the decision vector. When the weighted sum of the decision vector and the optimization objective is minimized, the configuration weight vector obtains the optimal value, thus obtaining the optimal configuration of the wireless sensor. Here, the residual adaptive regularization is used to dynamically adjust the penalty weights of different sensor positions based on the fluctuation of the prediction residuals, and the quadratic cumulative term is used to suppress the repeated selection of highly correlated features by considering the interaction effects between different sensor responses.

[0019] Furthermore, in step 4, the specific process of the dimensionality reduction differential evolution algorithm includes:

[0020] During the initialization phase, multiple sensor configuration weight vectors are randomly generated to form an initial population;

[0021] During the evolutionary phase, different differential mutation strategies are assigned to each individual in the current population to generate subpopulations, and crossover, mutation, and selection operations are sequentially performed on each subpopulation to generate new individuals; in each iteration, the fitness of the newly generated individuals is evaluated according to the optimization objective function, and the global evaluation count is updated.

[0022] Furthermore, in step 6: the optimal configuration weights of the multiple sensor locations are compared with a preset weight threshold, and candidate locations corresponding to sensors with weight values ​​greater than the weight threshold are retained and recorded as key sensor locations that significantly contribute to the prediction of the target node response; redundant locations corresponding to sensors with weight values ​​lower than the weight threshold are eliminated; and the compression ratio of the current wireless sensor configuration is calculated based on the number of key sensor locations selected and the total number of configurable locations.

[0023] Furthermore, in step 7: if the compression ratio is greater than the compression ratio threshold, only the decision variables corresponding to the selected key sensor locations are retained, and the configuration weights of the remaining candidate locations are reset to zero; thereby reducing the search space dimension of the optimization problem, and returning to step 4 to restart the differential evolution optimization process based on the dimensionality-reduced variable space; through iterative dimensionality reduction and restart optimization, the optimal sparse configuration is gradually approached until the compression ratio meets the requirements.

[0024] Furthermore, the method also includes dividing the dynamic response sample dataset into a training set and a test set, and using the test set to evaluate the performance of the obtained sensor configuration scheme.

[0025] In a second aspect, the present invention provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned wireless sensor dimensionality reduction sparse configuration evolution method oriented towards dynamic response reconstruction.

[0026] Thirdly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned method for dimensionality reduction and sparse configuration evolution of wireless sensors oriented towards dynamic response reconstruction.

[0027] The beneficial effects of this invention are as follows:

[0028] First, by constructing a sparse optimization model that integrates residual adaptive regularization and quadratic cumulative terms, efficient screening and weight optimization of sensor locations are achieved. This model can dynamically adjust the regularization strength based on the residual variance, improving its adaptability to nonlinear features. At the same time, it suppresses redundancy and related features through quadratic constraints, effectively improving prediction stability and accuracy.

[0029] Second, a dimensionality reduction differential evolution algorithm is proposed, which introduces an adaptive restart mechanism based on configuration weights. During the optimization process, the search space is gradually reduced, which significantly reduces the computational complexity of high-dimensional nonlinear problems and avoids the problems of traditional gradient-based methods being prone to getting trapped in local optima and slow convergence, thus greatly improving optimization efficiency.

[0030] Third, this method ultimately enables high-precision reconstruction of the dynamic response of unreachable nodes with an extremely low number of sensors. It significantly reduces the deployment cost and system complexity of wireless sensors, providing an efficient and reliable solution for health monitoring of complex structures such as spacecraft, and possesses strong engineering applicability and widespread application value. Attached Figure Description

[0031] Figure 1 This is a flowchart of the dimensionality reduction and sparse configuration evolution method for wireless sensors oriented towards dynamic response reconstruction according to the present invention.

[0032] Figure 2 This is a schematic diagram of a truss structure;

[0033] Figure 3 This is a comparison chart of predictions after the initial screening;

[0034] Figure 4 A comparison chart of predictions under the final wireless sensor configuration method. Detailed Implementation

[0035] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0036] This invention constructs a sparse monitoring optimization model based on a structural dynamics finite element model, aiming at optimal wireless sensor configuration. During the optimization process, a feature selection mechanism is introduced to evaluate and screen the importance of candidate wireless sensor locations. For example... Figure 1 As shown, the dimensionality reduction sparse configuration evolution method for wireless sensors oriented towards dynamic response reconstruction of this invention includes:

[0037] Step 1: Determine the number of candidate nodes for configuring the wireless sensor. Each candidate node is to be configured with two wireless sensors for measuring velocity and displacement, numbered sequentially from 1 to... Each candidate location for a wireless sensor is assigned a number. A weight threshold for the wireless sensor is then determined. Wireless sensor compression ratio threshold and maximum number of evaluations .

[0038] Determine as Figure 2 The truss model shown contains 40 nodes, each of which is configured with two wireless sensors for measuring velocity and displacement. This results in a total of 80 candidate locations for the wireless sensors. A weight threshold is set for each wireless sensor. Wireless sensor compression ratio threshold Maximum number of evaluations Based on the structural dynamics of the truss, the velocity of node 22, the initial data point with the highest fluctuation, is selected as the data to be predicted. It is assumed that node 22 is unreachable and cannot be configured with any wireless sensors; therefore, the problem dimension is... .

[0039] Because the selection of points is general and universal, it can be verified that this method can be extended to spacecraft for effective prediction of unreachable nodes.

[0040] Step 2: Construct a state-space model and simulate the state-space system. Apply loads of different magnitudes to the same finite element model, generate response samples as a dataset through multiple random samplings, and divide it into a training set and a test set. The training set is used for pre-training in steps 4 to 7 to obtain the optimal configuration of the wireless sensor; the test set is used in step 7.4 to evaluate the performance indicators of the configuration scheme. The specific steps are as follows:

[0041] Step 2.1: The material and geometric parameters of the truss model are set as follows: material density ; cross-sectional area Elastic modulus Damping coefficient .

[0042] The dynamic response of a spacecraft structure under external excitation is simulated based on the classical second-order system equations of structural dynamics. The equations are as follows:

[0043] ,

[0044] in, , and These represent the mass, damping, and stiffness matrices of the truss structure based on the above parameters, respectively. It is a time variable; The generalized displacement response vector of the structure; For generalized velocity, which is the first derivative of displacement, it represents the velocity response of each degree of freedom; For generalized acceleration, it represents the acceleration response of each degree of freedom; For external stimulus input.

[0045] Step 2.2: Transform the above second-order system equations into the following first-order state-space form:

[0046] ,

[0047] Wherein, the state vector This represents the displacement and velocity response of the structure in each degree of freedom; Represents the rate of change of the system's state. System matrix. With input matrix The structure is as follows:

[0048] ,

[0049] Step 2.3: Simulate the state-space system by applying loads of different magnitudes to the same finite element model for multiple random samplings. This yields the complete dynamic response of the system, which includes the time-varying displacement and velocity of all nodes.

[0050] ,

[0051] in, The reference load vector and the reference time-domain load function are as follows:

[0052] ,

[0053] In the truss example, external excitation Acting on such Figure 2 The y-direction of position 26 shown is expressed as follows after discretization: t is a time variable, discretized as ; The number of discrete points in time; Noise intensity coefficient; is a Gaussian white noise matrix; ⨀ represents the Hadamard product.

[0054] Since the response of a truss structure is characterized by displacement and velocity, as shown in the figure Figure 2 The displacement and velocity data for each node shown in Figures 1-40 are arranged and renumbered from 1-80, i.e., nodes. Displacement data number Speed ​​data number .

[0055] A finite element simulation model was used to generate 1000 sets of response samples by repeatedly exciting the model under random loads 1000 times. Each set of samples contains 80 observation degrees of freedom and their corresponding excitations, forming a three-dimensional tensor. To illustrate the sample structure, Table 1 shows the structural response vector of the first group of samples as an example, which is a... The slice represents the instantaneous response under 80 degrees of freedom in a set of samples. It is a matrix composed of the state vectors of the 40 nodes of the truss corresponding to the external excitation in step 2, that is, at each time step... Down middle and The values ​​are respectively used as column vectors to form a matrix.

[0056] Table 1

[0057]

[0058] The samples are divided into training and test sets. Based on the training set and the proposed feature selection-driven sparse optimization objective model and dimensionality reduction differential evolution algorithm, the optimal candidate position weights can be found (i.e., ... ); The test set found on the training set Performance evaluations were conducted to verify the accuracy of the wireless sensor configuration in predicting unreachable nodes.

[0059] Step 3: Based on the response samples from Step 2, construct a mapping relationship between the target response (i.e., unreachable nodes) and the responses of candidate nodes. This model sets the configuration weights corresponding to each candidate position as the decision vector, with the optimization objective being to minimize the optimization function composed of residual adaptive regularization and quadratic cumulative terms. The goal is to minimize the weighted sum of these three elements to construct a feature selection-driven sparse optimization objective model. That is, the configuration weight vector of the selected position is optimal when the weighted sum is minimized, thus obtaining the optimal configuration of the wireless sensor. The specific steps are as follows:

[0060] Step 3.1: Construct the predicted dynamic response of unreachable nodes State vectors of other candidate nodes The mapping relationship between them, that is, the mapping relationship between the velocity data of position 22 and other candidate position state data:

[0061] ,

[0062] in, Configure a weight vector for the optimal sensor. It is a matrix composed of the state vectors of 39 candidate nodes corresponding to 150 sampling points under real excitation, that is, at each sampling time. Down middle and The values ​​are respectively used as column vectors to form a matrix.

[0063] Step 3.2: Introduce a sparse optimization function that integrates adaptive regularization of residuals and quadratic accumulation terms:

[0064] ,

[0065] in, From step 2 The corresponding external stimulus A matrix composed of the state vectors of candidate nodes The number of time-domain sampling points. For the real dynamic response of unreachable nodes, Let be the decision variable to be optimized, i.e., the weight vector corresponding to the candidate sensor nodes. This is the residual vector for each iteration; For residual variance; correlation coefficient matrix , medium elements , express The Middle The first line List of responses; This is residual adaptive regularization; This refers to the cumulative term of a quadratic form. Specifically, Given the state vectors of 39 candidate nodes (excluding node 22) with known displacement and velocity data under 1000 external excitations. This represents the actual dynamic response of the velocity of node 22 across 1000 samples.

[0066] Step 4: Based on the training set obtained in Step 2 and the high-dimensional nonlinear single-objective optimization model constructed in Step 3, solve the problem using the proposed dimensionality reduction differential evolution algorithm. Initialize the configuration weight vectors for candidate positions, with each configuration weight scheme corresponding to an individual. Randomly generate multiple configuration weight schemes to form a configuration weight vector set as the initial population. Evaluate the fitness of the initial population based on the sparse optimization objective function described in Step 3 and update the function evaluation count. Randomly sample each individual in the current population and assign different differential evolution mutation strategies. Perform crossover, mutation, and selection operations sequentially on each subpopulation to generate new individuals. Evaluate the population fitness again based on the objective function in Step 3 and update the function evaluation count.

[0067] Step 5: If the function is evaluated a certain number of times Less than the maximum number of evaluations If not, return to step 4 and update the weight configuration vector set; otherwise, continue with the subsequent steps.

[0068] Step 6: Based on the optimal configuration weights of candidate locations calculated in Step 4 and the set wireless sensor weight threshold, candidate locations with weights greater than the threshold are selected and considered as key locations that significantly contribute to the prediction of unreachable node responses. Redundant and irrelevant locations are eliminated, thus obtaining the optimal wireless sensor configuration scheme under this differential evolution. The wireless sensor compression ratio is calculated based on the number of selected wireless sensors and the total number of configurable wireless sensor locations. Specifically:

[0069] Step 6.1: Table 2 shows the optimal configuration weights for the wireless sensors calculated in Step 4. If a number is not marked, it means that the weight corresponding to that number is 0, and the same applies below. Since position 22 is a position to be predicted, data 43 and 44 are not included in the optimization after renumbering, resulting in a total of 78 valid data points, of which 28 have a weight of 0.

[0070] Table 2

[0071]

[0072] Step 6.2: Based on the set weight threshold The results were compared with the actual threshold of each wireless sensor. The key wireless sensor locations that significantly contribute to the target prediction function (i.e., locations with values ​​greater than the threshold) were retained, while redundant and irrelevant locations were removed, as shown in Table 3. The wireless sensor compression ratio was calculated as follows:

[0073] ,

[0074] in, Filter out the number of wireless sensors, This represents the total number of configurable wireless sensor locations. The smaller the value, the sparser the selected wireless sensors, and the higher the model efficiency.

[0075] Table 3

[0076]

[0077] Step 7: If the wireless sensor compression ratio is greater than the compression ratio threshold, reconstruct the optimization variables based on the selected key sensor locations to reduce the dimensionality of the search space. Return to Step 4, restart the optimization problem, and perform differential evolution again; otherwise, output the current optimal candidate location weight vector and select the corresponding optimal wireless sensor configuration scheme as the final result. Furthermore, in the example, the wireless sensor compression ratio is greater than the threshold in the initial iteration. The specific steps are as follows:

[0078] Step 7.1: Retain the key wireless sensor positions that significantly contribute to the target prediction function selected in Step 6. Reset the weights of candidate positions that were not selected to zero. Update the wireless sensor configuration scheme to the velocities of positions 6, 24, 25, 26, 27, 28, 30, and 34, and the displacements of positions 30, 31, 32, 34, 36, and 40, corresponding to the numbers 12, 48, 50, 52, 54, 56, 59, 60, 61, 63, 67, 68, 71, and 79.

[0079] Step 7.2: Remove the zero-set positions from the decision vector and retain the selected key wireless sensor positions as candidate positions for the next round of iteration. In this way, the optimization variable space is reconstructed to achieve dimensionality reduction of the search space, reduce computational complexity and accelerate the optimization process.

[0080] Step 7.3: Based on the dimensionality-reduced search space obtained in Step 7.2, return to Step 4 to restart optimization. Through optimization iterations in the dimensionality-reduced search space, gradually approach the wireless sensor compression ratio threshold, and finally obtain the optimal wireless sensor configuration scheme.

[0081] Furthermore, in the final optimization result step 7, the wireless sensor compression ratio is less than the threshold. Once optimization is complete, the latest wireless sensor configuration scheme and corresponding wireless sensor configuration weights are output as the final result, as shown in Table 4. Final Threshold The displacement data from the wireless sensor at position 25 and the velocity data from the wireless sensor at position 30 were selected, meaning that configuring wireless sensors at positions 25 and 30 would enable effective prediction.

[0082] Table 4

[0083]

[0084] Step 7.4: Verify the performance of the solution based on the test set and test metrics. Specifically, the classic root mean square error (RMSE) metric is introduced to reflect the deviation between the predicted and actual values, thus evaluating the model's accuracy. The RMSE formula is as follows:

[0085] ,

[0086] in, For the sample size, For the first The true value of each sample For the first The predicted values ​​for each sample are used, and a smaller RMSE index indicates better model predictive performance. The calculation results show that the RMSE decreased from 0.0121±0.0002 to 0.0005±0.0001. Figure 3This is a comparison chart of predictions after the initial screening, showing the comparison between the actual values ​​and predicted values ​​of all test set samples. In the chart, the light gray solid line and the dark gray thick solid line represent the actual value response curve and the mean curve of each sample, respectively; the light gray dashed line and the black thick dashed line represent the predicted value response curve and the mean curve of each sample, respectively. Figure 4 The prediction comparison chart under the final wireless sensor configuration method is drawn in the same way as... Figure 3 The results of the method performance tests show that this method can significantly improve prediction accuracy while significantly reducing the compression ratio of wireless sensors.

[0087] In summary, this invention combines feature selection and dimensionality reduction differential evolution strategies to effectively solve the monitoring problem of areas inaccessible to wireless sensors in spacecraft structural health monitoring. It achieves high-precision structural response prediction with a very small number of wireless sensors, providing a low-cost, high-precision health monitoring solution for complex engineering structures such as spacecraft.

[0088] A feature selection mechanism is introduced during the optimization process to evaluate and filter the importance of candidate wireless sensor locations. A novel sparse optimization function is proposed by fusing residual adaptive regularization and quadratic accumulation terms. The residual adaptive regularization is implemented through residual variance. Adaptive adjustment of regularization coefficients dynamically handles different feature coefficients, thereby improving flexibility and accuracy in nonlinear complex problems. In regions of high data volatility, residual adaptive regularization can individually assign penalty coefficients to ensure prediction accuracy. The quadratic accumulation term effectively reduces the selection of redundant features by introducing quadratic constraints on feature interactions, optimizes the feature space, and thus reduces the impact of correlation between neighboring wireless sensors on feature selection results. It also actively suppresses the repeated selection of highly correlated wireless sensor data, removing redundant features. The fusion of these two approaches not only achieves a high compression ratio for wireless sensors and reduces configuration costs, but also reduces the noise superposition effect of highly correlated data, thereby improving the stability and performance of the model.

[0089] This invention proposes a dimensionality-reduction differential evolution algorithm that constructs an adaptive restart strategy for decision vector reconstruction to rapidly reduce the dimensionality of the search space, thus solving the significant limitations of traditional gradient descent-based optimization strategies in handling high-dimensional nonlinear problems. Traditional methods approximate the optimal solution by calculating the gradient of the loss function. However, when the model exhibits nonlinearity, high-dimensional correlation, and non-stationary characteristics, traditional gradient descent methods are prone to getting trapped in local optima and failing to converge to the global optimum. Furthermore, the computational complexity of gradient calculation increases dramatically with increasing dimensionality, leading to low convergence efficiency and potential redundancy in wireless sensor configuration. Therefore, this invention constructs a dimensionality-reduction differential evolution algorithm that utilizes a multi-strategy collaborative optimization and reconstruction dimensionality reduction mechanism. Overall, it achieves dimensionality reduction and reconstruction of the optimal solution through a process of "global exploration - local search - dimensionality reduction acceleration," reducing model complexity and improving convergence speed. Simultaneously, regarding computational complexity, assuming a population size of... Problem dimensions Subpopulation size Feature selection cycle Each subpopulation needs to complete the mutation and crossover operations. The total complexity of n vector operations is O(n). Fitness evaluation is performed in each iteration. There are solutions, with a complexity of [number] times. ,in Cost per evaluation; each feature selection Performing contribution calculation and dimension filtering once in a proxy, complexity The complexity of SQP in local search is... However, it only triggers with a low probability in the later stages. Considering the above complexity calculations, the worst-case total time complexity is: ,in Feature selection followed by dimensionality reduction and reconstruction significantly reduces the computational cost of later iterations.

[0090] In summary, this invention combines a sparse wireless sensor configuration optimization model for unreachable nodes with a dimensionality-reducing differential evolution algorithm, successfully overcoming the limitations of traditional methods in handling complex nonlinear characteristic relationships and significantly improving the sparsity of wireless sensor configuration. This method effectively solves the monitoring challenge of wireless sensor-inaccessible areas in spacecraft structural health monitoring, achieving high-precision structural response prediction with a minimal number of wireless sensors, and providing a low-cost, high-precision health monitoring solution for complex engineering structures such as spacecraft.

[0091] In a second aspect, the present invention provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned wireless sensor dimensionality reduction sparse configuration evolution method oriented towards dynamic response reconstruction.

[0092] Thirdly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned method for dimensionality reduction and sparse configuration evolution of wireless sensors oriented towards dynamic response reconstruction.

[0093] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dimension-reduction sparse configuration evolutionary method for wireless sensors oriented towards dynamic response reconstruction, characterized in that, The method includes: Step 1: Determine the number of candidate nodes and the sensor type to be configured for each candidate node, number the sensor locations, and set the sensor weight threshold, compression ratio threshold, and maximum number of evaluations. Step 2: Construct a state-space model of the spacecraft structure, perform finite element simulation by applying random loads of different sizes, and generate a dynamic response sample dataset; Step 3: Based on the response sample dataset, establish the mapping relationship between the target node response and the candidate node response, and construct a sparse optimization model with the optimization objective of minimizing the aggregation function of the fusion residual adaptive regularization and the quadratic cumulative term. Step 4: Solve the sparse optimization model using the dimensionality reduction differential evolution algorithm to obtain the optimal configuration weights for multiple sensor positions; Step 5: Determine if the maximum number of evaluations has been reached. If not, return to step 4 to continue iterating; otherwise, proceed to step 6. Step 6: Compare the weight threshold with the optimal configuration weight of multiple sensor positions, filter out key sensor positions, eliminate redundant positions, and calculate the compression ratio of the current sensor configuration. Step 7: If the compression ratio is greater than the compression ratio threshold, retain the key sensor positions and reduce the dimension of the optimization variables, then return to step 4; otherwise, output the optimal sensor configuration scheme.

2. The evolutionary method for dimensionality reduction and sparse configuration of wireless sensors oriented towards dynamic response reconstruction according to claim 1, characterized in that, In step 1, the sensor types include: measuring velocity and displacement.

3. The evolutionary method for dimensionality reduction and sparse configuration of wireless sensors oriented towards dynamic response reconstruction according to claim 1, characterized in that, In step 2, the state-space model is derived from the second-order system equations of structural dynamics and is used to simulate the displacement and velocity response of the structure under external excitation; the random load is composed of a reference load vector and Gaussian white noise superimposed on it.

4. The evolutionary method for dimensionality reduction and sparse configuration of wireless sensors oriented towards dynamic response reconstruction according to claim 1, characterized in that, In step 3, the sparse optimization model sets the configuration weights corresponding to the sensor positions of each candidate node as the decision vector. When the weighted sum of the decision vector and the optimization objective is minimized, the configuration weight vector obtains the optimal value, thus obtaining the optimal configuration of the wireless sensor. The residual adaptive regularization is used to dynamically adjust the penalty weights of different sensor positions based on the fluctuation of the prediction residual. The quadratic cumulative term is used to suppress the repeated selection of highly correlated features by considering the interaction effect between different sensor responses.

5. The evolutionary method for dimensionality reduction and sparse configuration of wireless sensors oriented towards dynamic response reconstruction according to claim 1, characterized in that, In step 4, the specific process of the dimensionality reduction differential evolution algorithm includes: During the initialization phase, multiple sensor configuration weight vectors are randomly generated to form an initial population; During the evolutionary phase, different differential mutation strategies are assigned to each individual in the current population to generate subpopulations, and crossover, mutation, and selection operations are sequentially performed on each subpopulation to generate new individuals; in each iteration, the fitness of the newly generated individuals is evaluated according to the optimization objective function, and the global evaluation count is updated.

6. The dimensionality reduction and sparse configuration evolution method for wireless sensors oriented towards dynamic response reconstruction according to claim 1, characterized in that, In step 6: the optimal configuration weights of the multiple sensor locations are compared with a preset weight threshold; candidate locations corresponding to sensors with weight values ​​greater than the weight threshold are retained and recorded as key sensor locations that significantly contribute to the prediction of the target node response; and sensor locations that are removed are discarded. Redundant locations corresponding to sensors with weight values ​​lower than the weight threshold; the compression ratio of the current wireless sensor configuration is calculated based on the number of key sensor locations selected and the total number of configurable locations.

7. The dimensionality reduction and sparse configuration evolution method for wireless sensors oriented towards dynamic response reconstruction according to claim 1, characterized in that, In step 7: if the compression ratio is greater than the compression ratio threshold, only the decision variables corresponding to the selected key sensor locations are retained, and the configuration weights of the remaining candidate locations are reset to zero; in this way, the search space dimension of the optimization problem is reduced, and the differential evolution optimization process is restarted based on the variable space after dimensionality reduction; through iterative dimensionality reduction and restart optimization, the optimal sparse configuration is gradually approached until the compression ratio meets the requirements.

8. The evolutionary method for dimensionality reduction and sparse configuration of wireless sensors oriented towards dynamic response reconstruction according to claim 1, characterized in that, The method further includes dividing the dynamic response sample dataset into a training set and a test set, and using the test set to evaluate the performance of the obtained sensor configuration scheme.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When one or more programs are executed by the one or more processors, the one or more processors implement the wireless sensor dimensionality reduction sparse configuration evolution method for dynamic response reconstruction as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, It stores executable instructions that, when executed by a processor, enable the processor to implement the dimensionality reduction sparse configuration evolution method for wireless sensors oriented towards dynamic response reconstruction as described in any one of claims 1-8.

Images