Gyroscope set state evaluation method based on KPCA-IBWO-KELM

Abstract

The invention relates to the technical field of gyroscopes state evaluation, and discloses a KPCA-IBWO-KELM-based gyroscopes state evaluation method, which comprises the following steps: acquiring detection port voltage signal data of a gyroscopes from a normal operation state to a fault state; carrying out dimension reduction processing on the collected data through kernel principal component analysis KPCA, and taking the data as a data set input by the model; dividing the model input data set into a test data set and a training data set; a white whale optimization algorithm BWO is improved, including introducing Logistic chaotic mapping in a BWO initialization stage, introducing a nonlinear inertia weight factor in a BWO development stage to enhance the global search ability of the algorithm, introducing a dynamic reverse learning strategy after the development stage is ended to jump out a local optimal solution, exploring more search spaces, obtaining an improved white whale optimization algorithm IBWO and the like; according to the method, the fault diagnosis process is more efficient and accurate, technical support is provided for maintenance of the gyroscopes, and the defect of blindness of parameter selection in the training process is overcome.

Description

Technical Field

[0001] The present invention relates to the technical field of gyroscope group state evaluation, and specifically to a method for evaluating the state of a gyroscope group based on KPCA-IBWO-KELM. Background Art

[0002] The gyroscope group is the core sensor of the inertial navigation system, fire control system, and vehicle body attitude monitoring system of armored vehicles, and is widely used in many key fields such as vehicle positioning, weapon stabilization, target tracking, and battlefield mobility enhancement. The stability of its working state and measurement accuracy directly affect the navigation ability, weapon shooting accuracy, and combat mission execution effect of armored vehicles. Therefore, effectively evaluating the state of the gyroscope group is of great significance for ensuring the battlefield survival ability and combat effectiveness of armored vehicles.

[0003] With the continuous development of industrial technology, more and more gyroscope health monitoring and fault detection methods have been proposed and applied to reduce the probability of system failure and improve the reliability of combat missions. Currently, data-driven health assessment methods have gradually become a research hotspot, including using means such as gyroscope signal feature extraction, vibration analysis, temperature monitoring, and noise assessment to comprehensively diagnose the operating state of gyroscopes. The health state of gyroscopes can usually be evaluated through indicators such as zero-bias drift, noise characteristics, temperature change effects, and dynamic errors. Among them: Zero-bias drift reflects the stability of the gyroscope during long-term operation and is crucial for navigation accuracy; Noise characteristics (such as power spectral density) can reveal the signal quality of the gyroscope and the health status of internal electronic components; Temperature impact assessment can be used to analyze the performance changes of the gyroscope in high and low temperature environments to ensure its reliability under extreme battlefield conditions; Dynamic error analysis can be used to monitor the performance degradation of the gyroscope in a severe motion or shock vibration environment.

[0004] However, in military armored vehicles, it is relatively difficult to obtain sample data on the health status of the gyroscope group, especially the negative samples (i.e., fault data) are scarce, resulting in the finiteness of training data; in addition, current state evaluation methods at home and abroad mostly use a single feature (such as only based on zero-bias drift or noise characteristics) for analysis, which to a certain extent affects the classification effect and may lead to misjudgment or missed judgment; at the same time, due to the extremely complex working environment of armored vehicles, factors such as severe vibration, high impact load, and strong electromagnetic interference will interfere with the gyroscope signal, making it difficult for existing single evaluation methods to comprehensively and accurately diagnose faults.

[0005] Therefore, it is urgent to study the gyroscope state evaluation method based on multi-source information fusion, comprehensively analyze various characteristics such as zero bias, noise, temperature, and vibration, so as to improve the accuracy and robustness of state evaluation. Summary of the Invention

[0006] Aiming at the above-mentioned shortcomings and deficiencies in the prior art, the present invention provides a gyroscope group state evaluation method based on KPCA-IBWO-KELM, improves the beluga whale optimization algorithm BWO, optimizes the key parameters of the kernel extreme learning machine KELM through the improved beluga whale optimization algorithm IBWO, and constructs an IBWO-KELM state evaluation model, which makes up for the defect of blindness in parameter selection during the training process, including the following steps: Step S01: Collect the detection port voltage signal data of the gyroscope group from the normal operation state to the fault state, and divide the collected data into five working states of the gyroscope group: healthy, good, attention, deterioration, and fault; Step S02: Perform dimensionality reduction processing on the collected data through kernel principal component analysis KPCA as the data set input to the model; divide the model input data set into a test data set and a training data set; Step S03: Improve the beluga whale optimization algorithm BWO, including introducing Logistic chaotic mapping in the initialization stage of BWO, introducing a non-linear inertia weight factor in the exploitation stage of BWO to enhance the global search ability of the algorithm and introducing a dynamic reverse learning strategy after the exploitation stage to jump out of the local optimum and explore more search spaces, and obtain the improved beluga whale optimization algorithm IBWO; Step S04: Use the improved beluga whale optimization algorithm IBWO to optimize the key parameters of the kernel extreme learning machine KELM, and construct an IBWO-KELM state evaluation model; Step S05: Use the training data set in Step S02 to train the IBWO-KELM state evaluation model in Step S04, output the optimal parameters, and construct the optimal IBWO-KELM state evaluation model; Step S06: Use the test data set in Step S02 as the input of the optimal IBWO-KELM state evaluation model trained in Step S05, and use the five working states of the gyroscope group as the output of the model to judge the working state of the gyroscope group.

[0007] The input samples in the data set are normalized to the range [-1, 1] through the mapminmax function in matlab.

[0008] In Step S03, Logistic chaotic mapping is introduced in the initialization stage of BWO, and the improved formula is: ; In the formula, Represents the initial position of the th beluga individual, represents the minimum boundary of the search space, represents the maximum boundary of the search space, represents the chaotic sequence value generated by the Logistic chaotic map, with a value range between (0,1), providing a high-quality random initialization distribution; In the BWO development stage, a non-linear inertia weight factor W is introduced to enhance the global search ability of the algorithm and avoid premature convergence to local optima; The formula for the non-linear weight factor W is: ; In the formula, represents the current iteration number, represents the maximum iteration number; The formula after introducing the non-linear inertia weight factor W in BWO is: ; In the formula: and respectively represent the positions of the beluga individuals and the beluga individual , represents the new position of the beluga individual , is the best position of the individual, and are both random values between (0,1), is the Lévy flight strategy, is the random jump factor; After the development stage ends, a dynamic reverse learning strategy is introduced to jump out of the local optimal solution: ; In the formula: represents the dynamic reverse solution, and are both random values between (0,1), represents the current solution, and respectively represent the upper and lower bounds of the actual search range of the current population.

[0009] Step S04 uses the improved beluga whale optimization algorithm IBWO to optimize the key parameters of the kernel extreme learning machine KELM. The optimized parameters include the optimal kernel function parameter and the penalty coefficient.

[0010] Compared with the prior art, the present invention has the following beneficial technical effects and advantages: The present invention improves the Beluga Whale Optimization Algorithm (BWO) by introducing the Logistic chaotic map, a non-linear inertia weight factor, and a dynamic reverse learning strategy, enhancing its global search ability, as well as its accuracy and efficiency in local exploitation, avoiding population singularity, and enabling higher applicability to different algorithm parameters; The improved Beluga Whale Optimization Algorithm is used to optimize the key parameters of the Kernel Extreme Learning Machine (KELM), improving the performance of the algorithm, making it more efficient and accurate in the process of state evaluation, and providing technical support for the maintenance of the gyroscope group. Description of the Drawings

[0011] Figure 1 It is a schematic diagram of the contribution rate and cumulative contribution rate of the sequential principal components of the KPCA algorithm; Figure 2 It is a comparison chart of the fitness values of the algorithm before and after improvement with the number of iterations; Figure 3 It is a diagram of the SVM classification result; Figure 4 It is a diagram of the WOA-KELM classification result; Figure 5 It is a diagram of the IBWO-KELM classification result. Detailed Embodiment

[0012] The present invention will be described in detail below, but the protection scope of the present invention is not limited by the drawings.

[0013] The present invention provides a method for gyroscope group state evaluation based on KPCA-IBWO-KELM, including the following steps: Step S01: Collect the detection port voltage signal data of the gyroscope group from the normal operation state to the fault state, and divide the collected data into five working states of the gyroscope group: healthy, good, attention, deterioration, and fault; Specifically, as people's requirements for equipment reliability are getting higher and higher, it continuously promotes the development of equipment maintenance theory. In order to more clearly determine the working state of the equipment, the concept of the health state in biology is introduced into equipment state evaluation. The number of health state levels to be divided should be appropriate. By comprehensively analyzing the operating state of the gyroscope group and the experience of industry experts and maintenance personnel, the state levels of the gyroscope group are divided into five categories: healthy, good, attention, deterioration, and fault. The state levels of the gyroscope group are as shown in Table 1 below: Status level Specific description Healthy The detection signals of the equipment are all within the specified range, far from the threshold, and the equipment can operate normally Good The detection signals of the equipment are all within the reference range, and some values fluctuate around 10% of the reference value, and none of them reach the threshold, and the equipment can operate normally Attention The detection signals of the equipment are basically qualified, all within the reference range, and some values fluctuate around 30% of the reference value. The equipment has a fault trend and should be prepared for maintenance Deterioration The detection signals of the equipment are all within the reference range, and some values fluctuate around 50% of the reference value. The fault trend is obvious and it should be repaired as soon as possible Fault The detection signals of the equipment reach or exceed 50% of the reference value, and the equipment cannot work normally and needs to be repaired immediately Table 1

[0014] Step S02: Perform dimensionality reduction processing on the collected data through Kernel Principal Component Analysis (KPCA) to obtain the data set for model input; divide the model input data set into a test data set and a training data set; The data collected by the gyroscope group has the characteristics of high dimensionality and non-linearity. If directly used as the input of the model, it will greatly increase the computational complexity of the model. At the same time, it will also make the model overly adapt to the noise in the data, resulting in the model being unable to learn all the features and ultimately causing the problem of overfitting. Therefore, the KPCA method is used to reduce the dimension of the original data, and the principal components with a cumulative contribution rate exceeding 90% are selected as the feature input, which can remove a large amount of redundant information and retain some key information.

[0015] As Figure 1 shown, KPCA sorts each sequential principal component according to the size of the eigenvalue, and calculates their contribution rate and cumulative contribution rate. As can be seen from Table 2, the cumulative contribution rate of the first six reaches 90.48%, which is greater than the set threshold of 90%, indicating that the first six principal components contain most of the information affecting the output signal of the gyroscope group. Therefore, the first six principal components are used as the input variables of the evaluation model, which reduces the computational amount and training time of the evaluation model to a certain extent and improves the efficiency; Principal component Contribution rate / % Cumulative contribution rate / % 1 20.66 20.66 2 19.87 40.53 3 18.23 58.75 4 16.75 75.51 5 8.023 83.53 6 6.948 90.48 Table 2

[0016] Specifically, 450 groups of data are extracted as experimental data, 300 groups of data are used as the training data set (the working states of five gyroscope groups, the first 60 groups of data in each group), and 150 data are used as the test data set (the working states of five gyroscope groups, the last 30 groups of data in each group).

[0017] There are 25 detection ports in the gyroscope group, including the signals input to the gyroscope by other modules and the signals output by the gyroscope itself. All the signals input to the gyroscope group by other modules are removed, and when collecting, it is ensured that the input quantity of other modules and the power supply module inside the gyroscope group are within the normal range.

[0018] There are a total of 10 collected output signals, and the constructed data set has a total of 10 dimensions. The collected data has the characteristics of non-linearity and high dimensionality. If the collected data is directly input into the evaluation model, it will affect the accuracy of the evaluation. Therefore, the KPCA dimensionality reduction processing is performed on the collected data.

[0019] More specifically, the dimensionality reduction processing of the data using KPCA includes the following steps: Step S201, construct a kernel matrix; Construct the original sample matrix , the number of samples is , the dimension of each sample is , using the non-linear function map the original samples from dimensions to dimensions in the space ( ) and map the original sample matrix Convert to : ; Wherein: is a sample in the low-dimensional space, is a sample in the high-dimensional space.

[0020] Step S202, decentralize the kernel matrix; The covariance matrix of is: ; wherein is a matrix of dimension is the transpose of is the mapping function.

[0021] Step S203, calculate the eigenvalues; According to the characteristic equation , obtain the eigenvalues of , the eigenvectors of ; Assume that when , there exists an eigenvector coefficient The eigenvector can be linearly represented as: ; Substitute the eigenvector into 's characteristic equation, and multiply both sides by the matrix ; ; Defines a kernel matrix , and any element in the matrix can be expressed as: ; can be transformed into: ; Obtain 's eigenvalues and eigenvectors, and then obtain 's normalized eigenvectors, then the principal component of the sample is: .

[0022] According to the eigenvalues of the kernel matrix , calculate the contribution rate of each principal component. When the contribution rate reaches ( Generally 85 - 95%), select the largest t eigenvalues and their corresponding eigenvectors : ;

[0023] The matrix after dimensionality reduction is: ; The input samples in the dataset are normalized to the range [-1, 1] using the mapminmax function in Matlab.

[0024] Step S03: Improve the Beluga Whale Optimization (BWO) algorithm. This includes introducing the Logistic chaotic map in the initialization stage of BWO, introducing a non-linear inertia weight factor in the exploitation stage to enhance the global search ability of the algorithm, and introducing a dynamic reverse learning strategy after the exploitation stage to jump out of the local optimal solution and explore more search spaces, resulting in the improved Beluga Whale Optimization (IBWO) algorithm. Specifically, in the Beluga Whale Optimization (BWO) algorithm, if a random initialization method is used, it may cause the phenomenon of beluga whales gathering, resulting in uneven distribution of beluga whale individuals. Using the Logistic chaotic map in the initialization population stage can increase the search space of each individual and enhance the search ability of the individual.

[0025] The initial position of the beluga whale population can be described as: ; In the formula: is the population size, is the variable dimension.

[0026] The formula for the Logistic chaotic map is: ; In the formula: is an adjustable parameter, , is the proportion of the population at time to the maximum possible population size, , .

[0027] After introducing the Logistic chaotic map, the improved formula is: ; In the formula, represents the initial position of the th beluga whale individual, represents the minimum boundary of the search space, represents the maximum boundary of the search space, represents the chaotic sequence value generated by the Logistic chaotic map, with a value range between (0, 1), providing a high-quality random initialization distribution.

[0028] The Beluga Whale Optimization algorithm includes three stages: exploration stage, exploitation stage, and whale fall stage.

[0029] The algorithm is based on the balance factor To determine whether the algorithm enters the exploration or exploitation stage, according to the probability factor of the occurrence of whale fall To determine whether the algorithm enters the whale fall stage

[0030] ; Wherein: represents the maximum number of iterations, represents the current iteration number; is a random number in (0, 1) and changes continuously as the iteration progresses; when the balance factor is less than 0.5, the algorithm transitions to the local exploitation stage, otherwise, the algorithm is in the global exploration stage. As the iteration progresses, the value of the balance factor gradually decreases, and the probability of the algorithm entering the exploitation stage increases

[0031] (1) The exploration stage is also the global search stage of the algorithm. In this stage, the algorithm simulates two beluga whales swimming in a synchronous or mirror image manner. The mathematical model can be expressed as: ; Wherein: represents the current iteration number, represents the beluga whale individual at the position in the represents a random integer in the represents the beluga whale individual at the position in the represents the beluga whale individual at the position in the and are both random values between (0, 1), and represent that the fins of the mirror image beluga whale face the water surface, and there are two choices for odd and even dimensions

[0032] (2) The exploitation stage is also the local search stage of the algorithm. In this stage, the selected beluga whale individuals share their position information with nearby beluga whales, and then search for food together. And the Levy flight strategy is introduced to enhance the local search ability of the algorithm. The mathematical model can be expressed as: ; ; ; ; Wherein: and respectively represent the beluga whale individual and the beluga individual position, represents the beluga individual new position, is the best position of the individual, and are both random values between (0, 1), is the random jump factor, is the Lévy flight strategy, is the probability function, and are both random numbers of the normal distribution, is the constant 1.5.

[0033] At this time, the nonlinear inertia weight factor W is introduced to enhance the global search ability of the algorithm: The formula for the nonlinear weight factor W is: ; In the formula, represents the current iteration number, represents the maximum iteration number; The formula of BWO after introducing the nonlinear inertia weight factor W is: ; In the formula: and respectively represent the positions of the beluga individuals and the beluga individual position, represents the beluga individual new position, is the best position of the individual, and are both random values between (0, 1), is the Lévy flight strategy, is the random jump factor.

[0034] After the development stage is over, the dynamic reverse learning strategy is introduced to jump out of the local optimal solution and explore more search spaces. The formula is as follows: ; In the formula: represents the dynamic reverse solution, and both represent random values between (0, 1), represents the current solution, and respectively represent the upper and lower bounds of the actual search range of the current population.

[0035] (3) When threatened with life, beluga whales will choose to migrate to other safe areas or sink to the bottom of the sea by whale-falling. To keep the population size unchanged, the beluga whale's position is updated based on the beluga whale's position and the step size of whale-falling. The mathematical model can be expressed as: ; ; ; In the formula: , and are both random values between (0, 1), is the step size of beluga whale falling, is the step factor, is the probability factor of whale-falling.

[0036] Step S04: Use the improved beluga whale optimization algorithm IBWO to optimize the key parameters of the kernel extreme learning machine KELM, the optimal kernel function parameter and the penalty coefficient to construct the IBWO-KELM state evaluation model; Specifically, the kernel extreme learning machine KELM introduces a kernel function on the basis of the extreme learning machine ELM, which not only retains the real-time performance of ELM, but also has strong generalization ability and stability. KELM is an efficient and fast classification model, which can enhance the classification ability by assigning higher weights to a smaller number of samples. The present invention uses the KELM algorithm as the basic classification algorithm.

[0037] The mathematical model of ELM can be expressed as: ; In the formula, represents the input of the network; represents the weight between the hidden layer and the output layer; represents the hidden layer matrix, represents the output function of the hidden layer.

[0038] In the kernel extreme learning machine, the mapping task in the traditional extreme learning machine is replaced by a Gaussian kernel function, which not only retains the real-time performance of ELM, but also has strong generalization ability and stability.

[0039] Select the radial basis kernel function as the kernel function of KELM: ; In the formula, is the input of the training set; is the input of the test set; is the width coefficient of the kernel function.

[0040] The output function of KELM based on the radial basis kernel function can be expressed as: ; In KELM, the kernel function parameter and the penalty coefficient have a significant impact on the performance of the model. They can affect the application range of the kernel function and have an impact on the stability performance of the model.

[0041] Step S05: Use the training data set in Step S02 to train the IBWO-KELM state evaluation model in Step S04, output the optimal parameters, and construct the optimal IBWO-KELM state evaluation model; Specifically, it includes the following steps: Step S501: Initialize the relevant parameters of IBWO and KELM, including the kernel function parameter of KELM and the penalty coefficient .

[0042] Specifically, the maximum number of iterations in the IBWO algorithm, the search dimension and the population size .

[0043] Optimize the kernel function parameter of KELM and the penalty coefficient , =2, =50, the search spaces of the kernel function parameter and the penalty coefficient are and respectively. Randomly generate the individuals of the beluga whale population, and the position of each beluga whale individual is the parameter combination of the KELM model.

[0044] Step S502: Use the accuracy rate as the fitness value of the model, and retain the best fitness value and the corresponding beluga whale individual.

[0045] Step S503: Judge whether the current number of iterations reaches the set maximum number of iterations. If it reaches the maximum number of iterations, output the optimal parameters and construct the optimal model. If it does not reach the maximum number of iterations, repeat Step S03 until the maximum number of iterations is reached, and then output the optimal parameters.

[0046] Step S504: Input the output optimal kernel function parameter and the penalty coefficient into the kernel extreme learning machine to establish the state evaluation model.

[0047] Step S06: Use the test data set in Step S02 as the input of the optimal IBWO-KELM state evaluation model trained in Step S05, and use the five working states of the gyroscope group as the output of the model to judge the working state of the gyroscope group.

[0048] To verify the classification performance of IBWO-KELM, compare IBWO-KELM and BWO-KELM experimentally, and use the accuracy rate as a function.

[0049] In the embodiment, the beluga whale optimization algorithm is used to find the optimal kernel function parameters and the penalty coefficient as the initial parameters of the KELM model.

[0050] Set the maximum number of iterations of the beluga whale optimization algorithm to 50, the population size to 30, and select the accuracy rate as the fitness value of the model. The fitness curves before and after the improvement of the beluga whale algorithm are as Figure 2 shown.

[0051] It can be seen from the figure that the BWO algorithm falls into the local optimum after 9 iterations until the end of the iteration, and the best fitness value is 94.7%; the fitness value of IBWO reaches the maximum at the 3rd iteration, the model converges to the optimum, and the best fitness value is 95.3%; the improved algorithm not only converges faster, but also has a higher classification accuracy rate.

[0052] Select 450 groups of data processed by KPCA as training samples to establish a KELM state evaluation model. To prove the advantages of this algorithm, compare it with the WOA-KELM model and SVM. Set the initial parameters of the two intelligent population algorithms to be the same. The parameters of SVM use the method of 5-fold cross-validation and grid search. The experimental results are as Figures 3 - 5 and Table 3 show: Model Accuracy rate / % Time / s SVM 89.33 8.62 WOA-KELM 91.2 9.65 IBWO-KELM 95.1 6.5 Table 3

[0053] In the present invention, the beluga whale optimization algorithm BWO is improved by introducing Logistic chaotic mapping, non-linear inertia weight factor and dynamic reverse learning strategy, enhancing its global search ability, as well as the accuracy and efficiency in local development, avoiding the singularity of the population, and being able to have higher applicability to different algorithm parameters.

[0054] Use the improved beluga whale optimization algorithm to optimize the key parameters of the kernel extreme learning machine KELM, improve the performance of the algorithm, and can be more efficient and accurate in the process of state evaluation, providing technical support for the maintenance of the gyroscope group.

[0055] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Any modifications, substitutions, and variations made by those of ordinary skill in the art to the above embodiments fall within the scope of the present invention.

[0056] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A method for evaluating the state of a gyroscope group based on KPCA-IBWO-KELM, characterized in that, It includes the following steps: Step S01: Collect the detection port voltage signal data of the gyroscope group from the normal operation state to the fault state, and divide the collected data into five working states of the gyroscope group: healthy, good, attention, deterioration, and fault. Step S02: Perform dimensionality reduction processing on the collected data through Kernel Principal Component Analysis (KPCA) to obtain the dataset for model input; divide the model input dataset into a test dataset and a training dataset. Step S03: Improve the Beluga Whale Optimization (BWO) algorithm, including introducing the Logistic chaotic map in the initialization stage of BWO, introducing a non-linear inertia weight factor in the exploitation stage of BWO to enhance the global search ability of the algorithm, and introducing a dynamic reverse learning strategy after the exploitation stage to jump out of the local optimum and explore more search spaces, resulting in the improved Beluga Whale Optimization algorithm (IBWO). Specifically, in step S03, the Logistic chaotic map is introduced in the initialization stage of BWO, and the improved formula is: ； In the formula, represents the initial position of the th beluga individual, represents the minimum boundary of the search space, represents the maximum boundary of the search space, represents the chaotic sequence value generated by the Logistic chaotic map, with a value range between (0, 1), providing a high-quality random initialization distribution; In the exploitation stage of BWO, a non-linear inertia weight factor W is introduced to enhance the global search ability of the algorithm and avoid falling into the local optimum prematurely. The formula for the non-linear weight factor W is: ； In the formula, represents the current iteration number, represents the maximum iteration number; The formula of BWO after introducing the non-linear inertia weight factor W is: ； Wherein: and respectively represent the positions of beluga individuals and beluga individual ; represents the new position of beluga individual ; is the optimal position of the individual, and are both random values between (0, 1), is the Lévy flight strategy, is the random jump factor; After the exploitation stage, a dynamic reverse learning strategy is introduced to jump out of the local optimum: ； In the formula: represents the dynamic reverse solution, and both represent random values between (0, 1), represents the current solution, and represent the upper and lower bounds of the actual search range of the current population respectively; Step S04: Use the improved Beluga Whale Optimization algorithm IBWO to optimize the key parameters of the Kernel Extreme Learning Machine (KELM), and construct an IBWO-KELM state evaluation model. Step S05: Use the training dataset in step S02 to train the IBWO-KELM state evaluation model in step S04, output the optimal parameters, and construct the optimal IBWO-KELM state evaluation model. Step S06: Use the test dataset in step S02 as the input of the optimal IBWO-KELM state evaluation model trained in step S05, and use the five working states of the gyroscope group as the output of the model to judge the working state of the gyroscope group.

2. The method for evaluating the state of a gyroscope group based on KPCA-IBWO-KELM according to claim 1, characterized in that: The input samples in the dataset are normalized to the range [-1, 1] through the mapminmax function in Matlab.

3. The method for evaluating the state of a gyroscope group based on KPCA-IBWO-KELM according to claim 1, characterized in that: In step S04, the improved Beluga Whale Optimization algorithm IBWO is used to optimize the key parameters of the Kernel Extreme Learning Machine (KELM). The optimized parameters include the optimal kernel function parameters and the penalty coefficient.

Images