Vehicle body gyroscope group fault prediction method based on IAOA-XGBoost

Abstract

The invention relates to the technical field of artificial intelligence fault diagnosis, and discloses a vehicle body gyroscope group fault prediction method based on IAOA-XGBoost, which comprises the following steps: carrying out dimensionality reduction and normalization preprocessing on acquired data through a KPCA algorithm, introducing a local chaotic mapping strategy in an AOA initialization stage to expand the range of an initial candidate solution position, and carrying out fault prediction on the initial candidate solution position; the method comprises the following steps: introducing a non-inertial control factor into a mathematical function accelerator of the AOA for improvement, judging whether the current AOA is in an exploration stage or a development stage according to the improved mathematical function accelerator, and updating a candidate solution position; and performing parameter optimization on the extreme gradient boosting tree XGBoost by adopting an improved arithmetic optimization algorithm IAOA, constructing an IAOA-XGBoost fault diagnosis model, inputting the test data set into the IAOA-XGBoost fault diagnosis model, performing fault diagnosis on the vehicle body speed gyroscope, and outputting a prediction result. According to the method, the defect of blindness of parameter selection in the training process is overcome, the prediction precision of the regression prediction model is improved, and higher prediction precision and practicability are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence fault diagnosis, and in particular to a vehicle body gyroscope group fault prediction method based on IAOA-XGBoost. Background Art

[0002] The vehicle body velocity gyroscope is an important component of the gun control system. The vehicle body velocity gyroscope can sense the magnitude and direction of the angular velocity of the vehicle body's vibration in the turret's rotation plane, and convert it into a stable feedforward signal in the horizontal direction, thereby improving the stability accuracy of the automatic working condition of the gun control system in the horizontal direction. Fault prediction can ensure its effective firepower output capability, improve battlefield combat capability, and maximize combat capability.

[0003] At present, various artificial intelligence-based algorithms have been widely used in the field of fault diagnosis. The arbitrariness of kernel function selection and the limitation of large-scale training of support vector machines lead to the lack of accuracy of prediction results. The difficulty of expert system knowledge acquisition and the limitation of knowledge base on storage cannot guarantee the efficiency and correctness of prediction results. The fault tree analysis method has large calculation scale, cumbersome work and excessive memory usage, which also leads to slower running speed in fault prediction.

[0004] Compared with other algorithms, the extreme gradient boosting tree algorithm (XGBoost) has the advantages of interpretability and robustness, which perfectly solves the problem of complex internal mechanisms; XGBoost has strong generalization and expression capabilities, which also makes the fault prediction results more accurate. Through the integrated learning method, it has improved the limitations of other machine learning algorithms in the field of fault diagnosis to a certain extent, such as weak generalization ability. Summary of the invention

[0005] In view of the shortcomings of the fault analysis algorithm in the prior art, such as large calculation scale, high complexity, excessive memory usage, cumbersome work, and lack of accuracy in prediction results, the technical solution adopted by the present invention is: a vehicle body gyroscope group fault prediction method based on IAOA-XGBoost, comprising the following steps: S1, collect data of the pin signal of the vehicle body speed gyroscope; S2. Perform dimension reduction and normalization preprocessing on the collected data through kernel principal component analysis (KPCA), screen out the data set that can be used as model input, and divide the model input data set into a test data set and a training data set; S3. Improvements to the arithmetic optimization algorithm AOA, including introducing the lterative chaos mapping strategy in the AOA initialization phase to expand the range of the initial candidate solution position, optimizing the position update in the initialization phase, and calculating the fitness; AOA mathematical function accelerator Introduce non-inertial control factors for improvement, and use the improved mathematical function accelerator Determine whether the current stage is the exploration stage or the development stage of AOA, update the candidate solution position, and obtain the improved arithmetic optimization algorithm IAOA; Specifically: S301, initialization phase: Generate a uniformly distributed candidate solution. The initial candidate solution distribution can be obtained by the following formula: , (2-1) Where: The value is between [0,1]; For the i The location of each solution in space; and are the upper and lower bounds respectively; At this point, the lterative chaotic mapping strategy is introduced to expand the range of the initial candidate solution position, and the position update is optimized in the initialization phase. The optimized position update is shown in the expression: , (2-2) Where: For the i+ 1 location of the solution in space; a is a control parameter, taking values ​​in [0,1]; S302, select mathematical function accelerator , (2-3) In the formula, It is an accelerator for mathematical functions; T and t They are the maximum number of iterations and the current number of iterations respectively; min and max are the minimum and maximum values ​​of the acceleration function respectively; At this time, the non-inertial control factor is introduced to improve the search ability of the algorithm. The improved formula is as follows: , (2-4) In the formula, For the improved math function accelerator; is the sinusoidal control factor; S303, Setting r 1 , r 2 and r 3 is a random parameter, when , which is currently in the exploration phase. The exploration phase uses a multiplication strategy or a division strategy to perform a global search. r 2 To determine the location: , (2-5) In the formula, for the new location; is the best position in the current iteration number t; is the search process control coefficient; is the minimum value; and are the upper and lower bounds respectively; P MO For the mathematical optimizer probability, the expression is as follows: , (2-6) In the formula, α is the sensitivity coefficient. The higher the α value, the better the sensitivity. P MO The greater the impact; S304, when At present, it is in the development stage, and the algorithm is locally optimized through addition strategy or subtraction strategy. r 3 To determine the location: , (2-7); S4, using the improved arithmetic optimization algorithm IAOA to optimize the parameters of the extreme gradient boosting tree XGBoost, constructing the IAOA-XGBoost fault diagnosis model, and training the fault diagnosis model with the training data set in S2; S5. Input the test data set in S2 into the IAOA-XGBoost fault diagnosis model, perform fault diagnosis on the vehicle body speed gyroscope, and output the prediction results.

[0006] The improved arithmetic optimization algorithm IOA in step S4 performs parameter optimization on the extreme gradient boosting tree XGBoost, specifically: introducing the lterative chaos mapping strategy in the AOA initialization phase to expand the range of the initial candidate solution position, optimizing the position update in the initialization phase, and calculating the fitness; judge r 1 With improved math function accelerator If r 1 < , then it is the exploration phase, and the position is updated through formula (2-5); if r 1 > , then it is the development stage, and the position is updated through formula (2-7); update the current optimal position and determine the current number of iterations t Whether the maximum number of iterations has been reached T , if the current number of iterations t The maximum number of iterations was not reached T, re-execute S3 to update the current optimal position. If the current number of iterations is t Reached the maximum number of iterations T , and obtain the optimal parameter combination.

[0007] The optimization parameters include: learning rate, minimum leaf node sample weight sum, maximum tree depth, node classification parameters, sample sampling rate and feature column sampling.

[0008] Compared with the prior art, the present invention has the following beneficial technical effects and advantages: 1. Using the KPCA algorithm to preprocess data is more effective than other comprehensive evaluation algorithms (hierarchical analysis method, grey correlation analysis, etc.) in processing nonlinear data, mapping data to high-dimensional space through kernel techniques, and capturing complex structures. It can also reduce dimensions, remove redundant features, improve the performance of subsequent models, and has a good effect in noise reduction; 2. Compared with support vector machines (SVMs) and decision trees, the extreme gradient boosting tree (XGBoost) can handle various types of data, including discrete and continuous values; it can use some robust loss functions; it can handle nonlinear data; and it has a higher prediction accuracy with relatively few parameter adjustments; 3. The main parameters of the extreme gradient boosting tree (XGBoost) are optimized by using the improved arithmetic optimization algorithm (IAOA), which makes up for the blindness of parameter selection during the training process and improves the prediction accuracy of the regression prediction model. Compared with the decision tree, support vector machine, and GBDT algorithm experiments, IAOA-XGBoost has higher prediction accuracy and practicality. DETAILED DESCRIPTION

[0009] The present invention is described in detail below.

[0010] The present invention provides a vehicle body gyroscope group fault prediction method based on IAOA-XGBoost, comprising the following steps: S1. Collecting data of the pin signal of the vehicle body speed gyroscope: The research object is the vehicle body speed gyroscope of a certain type of tank. The data value of the pin signal of the vehicle body speed gyroscope is collected through the equipment test bench, and the data value of the port number is collected as the initial data of the experiment.

[0011] S2. Perform dimension reduction and normalization preprocessing on the collected data through the KPCA algorithm to screen out the data set that can be used as model input; divide the model input data set into a test data set and a training data set; Specifically, the KPCA algorithm is used to perform dimension reduction and normalization preprocessing on the collected data as follows: S201, construct a decision matrix: suppose n sample data with m dimensions, and obtain an n*m decision matrix X, ,(1-1) in is a vector, n is a positive integer; The matrix X Mapped to a d-dimensional high-dimensional space, the mapping relationship is as follows: ,(1-2) in R m is the low-dimensional space before mapping, R d is the high-dimensional space after mapping, and the new kernel matrix after mapping is: ,(1-3) S202. Center the kernel matrix: return the sample data to the zero vector. After dimensionality increase, the covariance matrix is: ,(1-4) Where C is a d-dimensional matrix; for The transposed matrix of is the i-th mapping function, for The transposed matrix of .

[0012] S203, eigenvalue solution: According to the eigenvalue solution formula: ,(1-5) In the formula, p is the eigenvector, is the characteristic coefficient. Substitute formula (1-4) into formula (1-5) and omit the coefficient to obtain formula (1-6): ,(1-6) Divide both sides of formula (1-6) by , we get formula (1-7): ,(1-7) In the formula, The linear combination of is represented by the eigenvector: ,(1-8) In the formula, As a vector, substitute formula (1-8) into formula (1-6) and multiply both sides by Get formula (1-9): ,(1-9) make is a symmetric semi-positive definite matrix, , then the above formula can be simplified to: ,(1-10) The non-zero eigenvalues ​​obtained by solving formula (1-10) are equivalent to the non-zero eigenvalues ​​of formula (1-5). p After unitization, we have: ,(1-11).

[0013] S204, projecting the data processed in step S203 onto the selected principal component to form a new low-dimensional representation of the data in the feature vector p Projection z on to get the dimension-reduced data: ,(1-12).

[0014] S205, using a normalization formula to normalize the obtained dimension reduction data, the formula is: ,(1-13) In the formula, represents the normalized data. represents the normalized data, Represents the maximum value of the normalized data, Indicates the minimum value of the normalized data.

[0015] The arithmetic optimization algorithm uses the concept of mixed mathematical operations, including the exploration phase and the development phase, which correspond to the multiplication and division operation strategies and the addition and subtraction operation strategies respectively, and perform global search and local optimization respectively. The algorithm uses the mathematical function accelerator to flexibly select the optimization strategy. It is a new meta-heuristic optimization algorithm, but the algorithm still has defects in convergence speed and easy to fall into the local optimal solution. Therefore, the arithmetic optimization algorithm AOA is improved.

[0016] S3. Improvements to the arithmetic optimization algorithm AOA, including introducing the lterative chaos mapping strategy in the AOA initialization phase to expand the range of the initial candidate solution position, optimizing the position update in the initialization phase, and calculating the fitness; AOA mathematical function accelerator The non-inertial control factor is introduced to improve the algorithm's search capability, and the improved mathematical function accelerator is used Determine whether the current phase is the exploration phase or the development phase of the AOA, and update the candidate solution positions.

[0017] The specific steps include: S301, initialization phase: Generate a uniformly distributed candidate solution. The initial candidate solution distribution can be obtained by the following formula: , (2-1) Where: The value is between [0,1]; For the i The location of each solution in space; and are the upper and lower bounds respectively; The diversity of the initial population plays a vital role in the convergence speed and convergence accuracy of the algorithm. In order to further expand the scope of the initial population to improve the local search ability and make it easier to find the optimal solution, the lterative chaotic mapping strategy is introduced to expand the scope of the initial candidate solution position and optimize the position update in the initialization phase. The optimized position update is shown in the expression: , (2-2) Where: For the i+ 1 location of the solution in space; a is a control parameter, taking values ​​in [0,1]; S302, select mathematical function accelerator , (2-3) In the formula, It is an accelerator for mathematical functions; T and t They are the maximum number of iterations and the current number of iterations respectively; min and max are the minimum and maximum values ​​of the acceleration function respectively; There are instabilities in global search and local development, which leads to inaccuracies in finding the optimal solution. The value of plays a decisive role. When the value of increases, the local development ability of the algorithm is enhanced; when When the value of decreases, the global search ability of the algorithm is enhanced. The value range is between 0.2 and 1. If the changes are too fast, Entering the local development stage too early, non-inertia control factors are introduced for improvement. Here, sinusoidal control factors are used to make The value of increases slowly in the early stage to better perform global search, and increases rapidly in the later stage to enter the local development stage faster and enhance the search ability of the algorithm. The improved formula is as follows: , (2-4) In the formula, For the improved math function accelerator; is the sinusoidal control factor; S303, Setting r1 , r 2 and r 3 is a random parameter, when When , we are currently in the exploration phase. In the exploration phase, we use the multiplication strategy or the division strategy to perform a global search. Since the two strategies have high discreteness, r 2 To determine the location: , (2-5) In the formula, for the new location; is the best position in the current iteration number t; is the search process control coefficient; is the minimum value; and are the upper and lower bounds respectively; P MO For the mathematical optimizer probability, the expression is as follows: , (2-6) In the formula, α is the sensitivity coefficient. The higher the α value, the better the sensitivity. P MO The greater the impact; S304, when At present, it is in the development stage, and the algorithm is locally optimized through the addition strategy or the subtraction strategy. Since these two strategies are characterized by low dispersion, the random number r 3 To determine the location: , (2-7).

[0018] The parameters involved in modeling the extreme gradient boosting tree algorithm mainly include the learning rate (eta), which is used to improve the robustness and robustness of the model. By reducing the weight of each step, the effect of the model is guaranteed. When updating the weight of the leaf node, multiply it by this coefficient to avoid too large a step length, because the larger the value of this parameter, the less likely it is to fail to converge; the minimum leaf node sample weight (min_child_weight) is used to avoid overfitting. When its value is large, it can prevent the model from learning local special samples, but if this value is too high, it will lead to underfitting; the maximum depth of the tree (max_depth) is used to avoid overfitting. The larger the depth value, the better. The larger the value, the more specific and local samples the model will learn. The node classification parameter (gamma) is used to determine that the node split will only be performed when the loss reduction caused by the split is greater than gamma, which can effectively avoid overfitting. The sample sampling rate (subsample) is reduced, and the algorithm will be more conservative to avoid overfitting. However, if this value is set too small, it may cause underfitting. Feature column sampling (colsample_bytree) is used to control overfitting and control the proportion of columns randomly sampled by each tree (each column is a feature), that is, to control the sampling of features to prevent overfitting.

[0019] S4. Use the improved arithmetic optimization algorithm IAOA to optimize the parameters of the extreme gradient boosting tree XGBoost, build the IAOA-XGBoost fault diagnosis model, introduce the lterative chaos mapping strategy in the AOA initialization stage to expand the range of the initial candidate solution position, optimize the position update in the initialization stage, and calculate the fitness; judge r 1 With improved math function accelerator If r 1 < , then it is the exploration phase, and the position is updated through formula (2-5); if r 1 > , then it is the development stage, and the position is updated through formula (2-7); update the current optimal position and determine the current number of iterations t Whether the maximum number of iterations has been reached T , if the current number of iterations t The maximum number of iterations was not reached T , re-execute S3 to update the current optimal position. If the current number of iterations is t Reached the maximum number of iterations T , and obtain the optimal parameter combination.

[0020] S5. Input the test data set into the IAOA-XGBoost fault diagnosis model, perform fault diagnosis on the vehicle body speed gyroscope, and output the prediction results.

[0021] Finally, the improved arithmetic optimization algorithm optimized extreme gradient boosting tree (IAOA-XGBoost) was used to predict the fault of the body gyroscope of the tank gun control box, and the accuracy was significantly improved compared with the AOA-XGBoost prediction model. This model can effectively predict the fault of the obtained body gyroscope pin signal.

[0022] 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 principles of the present invention should be included in the protection scope of the present invention.

Claims

1. The vehicle body gyroscope group fault prediction method based on IAOA-XGBoost is characterized by: The following steps are involved: S1, collect data of the pin signal of the vehicle body speed gyroscope; S2. Perform dimension reduction and normalization preprocessing on the collected data through kernel principal component analysis (KPCA), select the data set that can be used as model input, and divide the model input data set into a test data set and a training data set; S3. Improvements to the arithmetic optimization algorithm AOA, including introducing the lterative chaos mapping strategy in the AOA initialization phase to expand the range of the initial candidate solution position, optimizing the position update in the initialization phase, and calculating the fitness; AOA mathematical function accelerator Introduce non-inertial control factors for improvement, and use the improved mathematical function accelerator Determine whether the current phase is the exploration phase or the development phase of AOA, update the candidate solution position, and obtain the improved arithmetic optimization algorithm IAOA; Specifically: S301, initialization phase: Generate a uniformly distributed candidate solution. The initial candidate solution distribution can be obtained by the following formula: (2-1)， Where: The value is between [0,1]; For the i The location of each solution in space; and are the upper and lower bounds respectively; At this point, the lterative chaotic mapping strategy is introduced to expand the range of the initial candidate solution position, and the position update is optimized in the initialization phase. The optimized position update is shown in the expression: (2-2)， Where: For the i+ 1 The location of a solution in space; a is a control parameter, taking values ​​in [0,1]; S302, select mathematical function accelerator (2-3), In the formula, It is an accelerator for mathematical functions; T and t They are the maximum number of iterations and the current number of iterations respectively; min and max are the minimum and maximum values ​​of the acceleration function respectively; At this time, the non-inertial control factor is introduced to improve the search ability of the algorithm. The improved formula is as follows: (2-4)， In the formula, For the improved math function accelerator; is the sinusoidal control factor; S303, Setting r 1. r 2 and r 3 is a random parameter. , which is currently in the exploration phase. The exploration phase uses a multiplication strategy or a division strategy to perform a global search. r 2 to determine the location: (2-5)， In the formula, for the new location; is the best position in the current iteration number t; is the search process control coefficient; is the minimum value; and are the upper and lower bounds respectively; P MO For the mathematical optimizer probability, the expression is as follows: (2-6)， In the formula, α is the sensitivity coefficient. The higher the α value, the better the sensitivity. P MO The greater the impact; S304, when At present, it is in the development stage, and the algorithm is locally optimized through addition strategy or subtraction strategy. r 3 to determine the location: (2-7)； S4, using the improved arithmetic optimization algorithm IAOA to optimize the parameters of the extreme gradient boosting tree XGBoost, constructing the IAOA-XGBoost fault diagnosis model, and training the fault diagnosis model with the training data set in S2; S5. Input the test data set in S2 into the IAOA-XGBoost fault diagnosis model, perform fault diagnosis on the vehicle body speed gyroscope, and output the prediction results.

2. The vehicle body gyroscope group fault prediction method based on IAOA-XGBoost according to claim 1 is characterized in that: The improved arithmetic optimization algorithm IOA in step S4 performs parameter optimization on the extreme gradient boosting tree XGBoost, specifically: introducing the lterative chaotic mapping strategy in the AOA initialization phase to expand the range of the initial candidate solution position, optimizing the position update in the initialization phase, and calculating the fitness; judge r 1 and improved math function accelerator If r 1< , then it is the exploration phase, and the position is updated by formula (2-5); if r 1> , then it is the development stage, and the position is updated through formula (2-7); update the current optimal position and determine the current number of iterations t Whether the maximum number of iterations has been reached T , if the current number of iterations t The maximum number of iterations was not reached T , re-execute S3 to update the current optimal position. If the current number of iterations is t Reached the maximum number of iterations T , and obtain the optimal parameter combination.

3. The vehicle body gyroscope group fault prediction method based on IAOA-XGBoost according to claim 2 is characterized in that: The parameters for optimizing the extreme gradient boosting tree XGBoost include: learning rate, minimum leaf node sample weight and, maximum depth of the tree, node classification parameters, sample sampling rate, and feature column sampling.

Images