A gyroscope group fault prediction method based on MBKA-GBDT

Abstract

The present invention relates to the field of artificial intelligence fault diagnosis technology, and discloses a gyroscope group fault prediction method based on MBKA-GBDT, comprising the steps of: collecting data of gyroscope group pin signals; preprocessing the collected data, screening pin signal data with high correlation as a model input data set; improving the population initialization phase, attack phase and migration phase of the black kite optimization algorithm BKA, using the improved black kite optimization algorithm MBKA to perform parameter optimization on the gradient boosting decision tree GBDT, constructing an MBKA-GBDT fault diagnosis model, inputting a test data set into the fault diagnosis model, performing fault diagnosis on the gyroscope group, and outputting a prediction result. The present invention reduces data redundancy, reduces data dimensions, and is conducive to improving the prediction efficiency of the model, improves the black kite algorithm, solves its weak convergence, avoids the problem that its initialization population is unevenly distributed, and is easy to fall into a local optimal solution, and improves the accuracy of model prediction.

Description

Technical Field

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

[0002] In the gun control system, the gyroscope group is an important component, which consists of a rate gyroscope, a temperature control board, a power board, and a detection board. The gyroscope group is used in the fire control system to accurately measure the angular position and angular velocity of the turret and the gun, and provide the system with a drive signal to give the gun a new position. Its performance directly affects the accuracy and reliability of the gun. In a dynamic and changing battlefield environment, the stability and accuracy of the gyroscope become key factors in improving the reaction speed and hit rate of the gun system. However, the gyroscope group may suffer from a variety of faults during operation, such as mechanical wear, aging of electronic components, and software failures. These faults may cause the positioning accuracy of the gun control system to decrease or even fail completely. Effective prediction and diagnosis of gyroscope group faults can not only improve battlefield combat movement capabilities and firepower output, maximize combat capabilities, but also help extend the service life of equipment and reduce maintenance costs. Therefore, it is particularly important to predict the faults of the gyroscope group.

[0003] In recent years, many artificial intelligence algorithms have been widely used in the field of fault diagnosis. The fault tree algorithm requires accurate fault probability data, which may be difficult to obtain in actual operation, and the calculation scale is large and complex, the work will be very cumbersome, and the fault prediction operation speed will be slowed down. The randomness and limitations of the support vector machine in the selection of kernel functions and penalty factor parameters will lead to the lack of accuracy of the prediction results. The BP neural network will have the problem of poor generalization ability, which will affect the prediction accuracy.

[0004] Compared with other algorithms, the Gradient Boosting Decision Tree (GBDT) is used as a weak learner. Its advantages of interpretability and robustness perfectly solve the internal mechanism problem. It can handle nonlinear related data very well, which is very important for the complex dynamic changes in the gyroscope group. It can automatically select features, which helps to understand the key factors of fault occurrence. It also has good generalization and expression capabilities, making the fault prediction results more accurate. 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, cumbersome work, and lack of accuracy in prediction results, the technical solution adopted by the present invention is: a gyroscope group fault prediction method based on MBKA-GBDT, characterized in that it includes the following steps:

[0006] S1, collect data of gyroscope group pin signal;

[0007] S2. Preprocess the collected data by grey correlation analysis method, and select pin signal data with high correlation as the model input data set;

[0008] Divide the model input dataset into a test dataset and a training dataset;

[0009] S3. Improve the black kite optimization algorithm BKA, including introducing Tent chaos mapping in the BKA population initialization stage to calculate individual fitness; introducing Levy flight strategy in the BKA attack stage; introducing spiral search strategy in the BKA migration stage;

[0010] S4. Use the improved black kite optimization algorithm MBKA to optimize the parameters of the gradient boosting decision tree GBDT, build the MBKA-GBDT fault diagnosis model, and train the training data set;

[0011] S5. Input the test data set into the MBKA-GBDT prediction model, perform fault diagnosis on the gyroscope group, and output the prediction results.

[0012] In step S2, the data collected in step S1 is subjected to grey correlation analysis to filter out the pin data that has a greater impact on the gyroscope group and remove the data that has a smaller impact on the system, thereby reducing the redundancy between the data and simplifying the state feature information. The model input data set is specifically:

[0013] S201, determine the analysis sequence:

[0014] Define the main behavior or target sequence of the research object as the reference sequence Y ;

[0015] (1)

[0016] Define the sequence of factors that affect the reference sequence Y as the comparison sequence X i ;

[0017] (2)

[0018] In the formula, i =1,2,…n; n is an unknown integer; k is an integer between 1 and n;

[0019] S202, data normalization processing: solve the problem that the original data units and orders of magnitude may be different:

[0020] (3)

[0021] (4)

[0022] In the formula, For reference sequence Y In the k The value of a point; To compare the sequence X i In the k The value of a point; is the normalized reference sequence Y In the k The value of a point; To compare the sequences after normalization X i In the k The value of a point;

[0023] S203, calculate the correlation coefficient: calculate the reference sequence Y Compare each sequence X i The difference sequence between

[0024] (5)

[0025] In the formula, For reference sequence Y Compare with sequence X i In between k The difference of points;

[0026] Determine the grey relational coefficient:

[0027] (6)

[0028] In the formula, For the i Comparison sequence X i With reference sequence Y In the k The correlation coefficient of each point; is the resolution coefficient;

[0029] S204, calculate the correlation: calculate the correlation between each subsequence and the reference sequence Y The average correlation of:

[0030] (7)

[0031] In the formula, To compare the sequence X i With reference sequence Y The overall correlation between

[0032] S205, analysis and decision: sorting is performed according to the calculated correlation degree. The greater the correlation degree, the greater the impact. Based on the result of the correlation degree, corresponding decisions or further analysis are made.

[0033] Step S3 is specifically:

[0034] S301, population initialization: introduce Tent chaos mapping, initialize the population, calculate individual fitness, and the mathematical model is:

[0035] (8)

[0036] Where τ is the number of mappings; X (τ) is the value of the τth mapping function in the range [0-1];

[0037] S302, updating the positions of individuals in the population;

[0038] (9)

[0039] (10)

[0040] In the formula, For the i A black kite t The iteration j The location of the dimension; r A random number between 0 and 1; P A constant of 0.9 is used as the probability threshold to determine the behavior selection; η Iteration attenuation coefficient is used to adjust the individual position update step size; t is the current iteration number; T Total number of iterations;

[0041] S303. Introduce the Levy flight strategy in the BKA attack phase to enhance the global search capability in the solution space; the improved new formula is:

[0042] (11)

[0043] (12)

[0044] (13)

[0045] In the formula, L ( s ) is the Levy flight function; is a number in (1, 3); s is the flight step length; is the Gamma function;

[0046] S304, updating the position of the population leader;

[0047] (14)

[0048] (15)

[0049] In the formula, The position of the leader of the black-winged kite in the jth dimension at the tth iteration is the current optimal solution position; is the fitness value of the current position; is the fitness value of the random position; r is a random number between 0 and 1; m is a scaling factor used to adjust the attraction or repulsion of the solution toward the optimal solution or other positions; is a random number generated from the Cauchy distribution;

[0050] S305. Introducing a spiral search strategy in the BKA migration phase to increase the possibility of finding a global optimal solution;

[0051] The improved new formula is:

[0052] (16)

[0053] (17)

[0054] In the formula, is an exponential function used to adjust the expansion or contraction of the spiral. c and l Control the tightness and periodicity of the spiral respectively.

[0055] In step S4, the improved black kite optimization algorithm MBKA is used to optimize the parameters of the gradient boosting decision tree GBDT. Specifically, the attack behavior determined by the population and the migration behavior determined by the population leader are updated simultaneously;

[0056] like p < r , then update the individual's location information according to formula (11); if p ≥ r , then update the individual location information according to formula (12);

[0057] like F k < F r , then according to formula (16), the leader is replaced, the migration behavior is abandoned, the individual position information is updated, and a new leader is selected; if F k ≥ F r , then continue to execute the migration behavior instructions according to formula (17) and update the leader's position information;

[0058] Determine whether the maximum number of iterations has been reached. If t < T， Output the optimal parameter combination; if t ≥ T， Re-execute S4 and update the positions of the population individuals and the population leader again.

[0059] Compared with the prior art, the present invention has the following beneficial technical effects and advantages:

[0060] 1. Use grey correlation analysis to pre-process the data, filter out the pin data that has a greater impact on the gyroscope group and remove the data that has a smaller impact on the system, reduce the redundancy between data, and simplify the state feature information, which is conducive to improving the subsequent prediction efficiency. Compared with other pre-processing methods, grey correlation is more suitable for processing small sample and incomplete information data such as the gyroscope group;

[0061] 2. The black kite algorithm is improved by using Tent chaos mapping, spiral search strategy and Levy flight strategy, which can solve its weak convergence, avoid the problem of uneven distribution of the initial population, and easily fall into the local optimal solution, and improve the accuracy of subsequent prediction results;

[0062] 3. The main parameters of the gradient boosting decision tree (GBDT) are optimized by using the improved black kite optimization algorithm (MBKA), which makes up for the blindness of parameter selection during the training process and improves the prediction accuracy and practicality of the prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a main framework diagram of a gyroscope group fault prediction method based on MBKA-GBDT of the present invention. DETAILED DESCRIPTION

[0064] The present invention is described in detail below in conjunction with the accompanying drawings, but the protection scope of the present invention is not limited by the accompanying drawings.

[0065] The present invention provides a gyroscope group fault prediction method based on MBKA-GBDT, such as Figure 1 As shown: the following steps are included:

[0066] S1, collect data of gyroscope group pin signal;

[0067] Specifically, the research object is the gyroscope group of a certain type of tank fire control system, and the pin data of the gyroscope group is collected through the fire control system electrical component test bench.

[0068] S2. Preprocess the collected data by grey correlation analysis method, and select pin signal data with high correlation as the model input data set;

[0069] Divide the model input dataset into a test dataset and a training dataset;

[0070] Specifically, the data collected by S1 is subjected to grey correlation analysis to screen out the pin data that has a greater impact on the gyroscope group and remove the data that has a smaller impact on the system, thereby reducing the redundancy between the data and simplifying the state feature information. The model input data set is specifically:

[0071] S201, determine the analysis sequence:

[0072] Define the main behavior or target sequence of the research object as the reference sequence Y ;

[0073] (1)

[0074] Define the sequence of factors that affect the reference sequence Y as the comparison sequence X i ;

[0075] (2)

[0076] In the formula, i =1,2,…n; n is an unknown integer; k is an integer between 1 and n;

[0077] S202, data normalization processing: solve the problem that the original data units and orders of magnitude may be different:

[0078] (3)

[0079] (4)

[0080] In the formula, For reference sequence Y In the k The value of a point; To compare the sequence X i In the k The value of a point; is the normalized reference sequence Y In the k The value of a point; To compare the sequences after normalization X i In the k The value of a point;

[0081] S203, calculate the correlation coefficient: calculate the reference sequence Y Compare each sequence X iThe difference sequence between

[0082] (5)

[0083] In the formula, For reference sequence Y Compare with sequence X i In between k The difference of points;

[0084] Determine the grey relational coefficient:

[0085] (6)

[0086] In the formula, For the i Comparison sequence X i With reference sequence Y In the k The correlation coefficient of each point; is the resolution coefficient;

[0087] S204, calculate the correlation: calculate the correlation between each subsequence and the reference sequence Y The average correlation of:

[0088] (7)

[0089] In the formula, To compare the sequence X i With reference sequence Y The overall correlation between

[0090] S205, analysis and decision: sorting is performed according to the calculated correlation degree. The greater the correlation degree, the greater the impact. Based on the result of the correlation degree, corresponding decisions or further analysis are made.

[0091] S3. Improve the black kite optimization algorithm BKA, including introducing Tent chaos mapping in the BKA population initialization stage to calculate individual fitness; introducing Levy flight strategy in the BKA attack stage; introducing spiral search strategy in the BKA migration stage;

[0092] S301. Population initialization: Introduce Tent chaotic mapping, initialize the population, and calculate individual fitness. Since Tent chaotic mapping can produce highly unpredictable and random behaviors, the generated sequence has good statistical properties and uniform distribution, which helps to improve the global search capability of the Black Kite algorithm.

[0093] The mathematical model is:

[0094] (8)

[0095] Where τ is the number of mappings; X (τ) is the value of the τth mapping function in the range [0-1];

[0096] S302, updating the positions of individuals in the population;

[0097] As a predator, the black kite will adjust the angle of its wings and tail according to the wind speed, hover to observe the prey and then attack quickly. In this stage, there are two modes of hovering attack or diving attack behavior, which are used for global search. The update formula is:

[0098] (9)

[0099] (10)

[0100] In the formula, For the i A black kite t The first iteration j The location of the dimension; r A random number between 0 and 1; P A constant of 0.9 is used as the probability threshold to determine the behavior selection; η Iteration attenuation coefficient is used to adjust the individual position update step size; t is the current iteration number; T Total number of iterations;

[0101] S303. Introduce the Levy flight strategy in the BKA attack phase. The Levy flight strategy allows the algorithm to make occasional long-distance jumps and more short-distance local searches during the iteration process, which can enhance the algorithm's global search capability in the solution space.

[0102] The improved new formula is:

[0103] (11)

[0104] (12)

[0105] (13)

[0106] In the formula, L ( s ) is the Levy flight function; is a number in (1, 3); s is the flight step length; is the Gamma function;

[0107] S304, updating the position of the population leader;

[0108] The migration behavior of black-winged kites is generally led by the leader bird. The BKA algorithm incorporates the leader strategy based on this feature: if the current population fitness value is smaller than the random population fitness value, the current leader will give up the leader position and join the migrating population. On the contrary, if the current population fitness value is larger than the random population fitness value, the leader will continue to lead the population to ensure migration to a place with rich resources. The update formula is:

[0109] (14)

[0110] (15)

[0111] In the formula, The position of the leader of the black-winged kite in the jth dimension at the tth iteration is the current optimal solution position; is the fitness value of the current position; is the fitness value of the random position; r is a random number between 0 and 1; m is a scaling factor used to adjust the attraction or repulsion of the solution toward the optimal solution or other positions; is a random number generated from the Cauchy distribution, and the Cauchy mutation strategy is introduced to increase the randomness of the migration behavior.

[0112] S305. Introduce a spiral search strategy in the BKA migration phase. The spiral search strategy can effectively expand the search space, especially in complex or multi-peak optimization problems, helping the algorithm to explore multiple potential good areas, thereby increasing the possibility of finding the global optimal solution.

[0113] The improved new formula is:

[0114] (16)

[0115] (17)

[0116] In the formula, is an exponential function used to adjust the expansion or contraction of the spiral. c and l Control the tightness and periodicity of the spiral respectively.

[0117] S4. Use the improved black kite optimization algorithm MBKA to optimize the parameters of the gradient boosting decision tree GBDT, build the MBKA-GBDT fault diagnosis model, and train the training data set;

[0118] Specifically, constructing a gradient boosted decision tree GBDT includes:

[0119] (1) Initialize the model

[0120] At the beginning, the model uses a simple predictor as a benchmark, usually the mean of the data or the category with the highest probability. This initial model is called ,in x represents a feature vector.

[0121] (2) Iterative enhancement

[0122] (a) Calculate the residual

[0123] Residual Indicates the current model The true target value y i The difference between them is calculated as:

[0124] (18)

[0125] In the formula, For the t The residual of the round iterative calculation; L is the loss function; y i，t For the i The true target value of samples; x i，t For the i The feature vector of samples; For the t -Model prediction value after 1 round of iteration;

[0126] (b) Training a new decision tree

[0127] Use the residuals from the previous step , as the target to train a new decision tree ;

[0128] (c) Update the model

[0129] (19)

[0130] In the formula, F t ( x ) is the t-th model; v Is the step size factor, which is used to avoid overfitting and speed up the local convergence of the algorithm. v The value range is , under the same learning effect, if v If the value is smaller, more iterations may be required, and if it is too large, the optimal point will be skipped;

[0131] The hybrid strategy improved black kite optimization algorithm proposed in this paper can adjust the step size factor v Make reasonable choices;

[0132] (3) Update the final model

[0133] After T rounds of iterations, the final model is the accumulation of decision tree predictions in all iterations, expressed as:

[0134] (20)

[0135] Where t is the current iteration number and T is the maximum iteration number.

[0136] S4. Use the improved black kite optimization algorithm MBKA to optimize the parameters of the gradient boosting decision tree GBDT to build a fault prediction model, specifically:

[0137] Initialization parameters: The optimization parameters involved in the GBDT algorithm include learning rate, number of trees, maximum depth of trees, number of iterations, etc.

[0138] Initialize the population according to formula (8), calculate the fitness value of the individuals in the population, and determine the leader;

[0139] Simultaneously update the aggressive behavior determined by the population and the migratory behavior determined by the population leader;

[0140] like p < r , then update the individual's location information according to formula (11); if p ≥ r , then update the individual location information according to formula (12);

[0141] like F k < F r , then according to formula (16), the leader is replaced, the migration behavior is abandoned, the individual position information is updated, and a new leader is selected; if F k ≥ F r , then continue to execute the migration behavior instructions according to formula (17) and update the leader's position information;

[0142] Determine whether the maximum number of iterations has been reached. If t < T， Output the optimal parameter combination; if t ≥ T， Update the positions of population individuals and population leaders again.

[0143] S5. Input the test data set into the MBKA-GBDT prediction model, perform fault diagnosis on the gyroscope group, and output the prediction results.

[0144] The present invention uses a data set processed by grey correlation analysis and a gradient boosting decision tree fault prediction model optimized by an improved Black Kite algorithm to predict faults of a gyroscope group. Compared with the original gradient boosting decision tree model, the unimproved Black Kite algorithm optimized gradient boosting decision tree model and other optimization algorithm models, this model has obvious advantages in prediction accuracy, and can effectively predict faults of a gyroscope group.

[0145] 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. A gyroscope group fault prediction method based on MBKA-GBDT, characterized in that The following steps are involved: S1, collect data of gyroscope group pin signal; S2. Preprocess the collected data by grey correlation analysis method, and select pin signal data with high correlation as the model input data set; Divide the model input dataset into a test dataset and a training dataset; S3. Improvements to the black kite optimization algorithm BKA, including the introduction of Tent chaos mapping in the BKA population initialization phase to calculate individual fitness; the introduction of Levy flight strategy in the BKA attack phase; and the introduction of spiral search strategy in the BKA migration phase; The specific steps include: S301, population initialization: introduce Tent chaos mapping, initialize the population, calculate individual fitness, and the mathematical model is: （8） Where τ is the number of mappings; X (τ) is the value of the τth mapping function in the range [0-1]; S302, updating the positions of individuals in the population; （9） （10） In the formula, For the i A black kite t The iteration j The location of the dimension; r A random number between 0 and 1; p A constant of 0.9 is used as the probability threshold to determine the behavior selection; η Iteration attenuation coefficient is used to adjust the individual position update step size; t is the current iteration number; T Total number of iterations; S303. Introduce the Levy flight strategy in the BKA attack phase to enhance the global search capability in the solution space; the improved new formula is: （11） （12） （13） In the formula, L ( s ) is the Levy flight function; is a number in (1, 3); s is the flight step length; is the Gamma function; S304, updating the position of the population leader; （14） （15） In the formula, The position of the leader of the black-winged kite in the jth dimension at the tth iteration is the current optimal solution position; is the fitness value of the current position; is the fitness value of the random position; r is a random number between 0 and 1; m is a scaling factor used to adjust the attraction or repulsion of the solution toward the optimal solution or other positions; is a random number generated from the Cauchy distribution; S305. Introducing a spiral search strategy in the BKA migration phase to increase the possibility of finding a global optimal solution; The improved new formula is: （16） （17） In the formula, is an exponential function used to adjust the expansion or contraction of the spiral. c and l Control the tightness and periodicity of the spiral respectively; S4. Use the improved black kite optimization algorithm MBKA to optimize the parameters of the gradient boosting decision tree GBDT, build the MBKA-GBDT fault diagnosis model, and train the training data set; S5. Input the test data set into the MBKA-GBDT prediction model, perform fault diagnosis on the gyroscope group, and output the prediction results.

2. The gyroscope group fault prediction method based on MBKA-GBDT according to claim 1, characterized in that: In step S2, the data collected in step S1 is subjected to grey correlation analysis to filter out the pin data that has a greater impact on the gyroscope group and remove the data that has a smaller impact on the system, thereby reducing the redundancy between the data and simplifying the state feature information as the model input data set, specifically: S201, determine the analysis sequence: Define the main behavior or target sequence of the research object as the reference sequence Y ; （1） Define the sequence of factors that affect the reference sequence Y as the comparison sequence X i ; （2） In the formula, i =1,2,…n; n is an unknown integer; k is an integer between 1 and n; S202, data normalization processing: solve the problem that the original data units and orders of magnitude may be different: （3） （4） In the formula, For reference sequence Y In the k The value of a point; To compare the sequence X i In the k The value of a point; is the normalized reference sequence Y In the k The value of a point; To compare the sequences after normalization X i In the k The value of a point; S203, calculate the correlation coefficient: calculate the reference sequence Y Compare each sequence X i The difference sequence between （5） In the formula, For reference sequence Y Compare with sequence X i In between k The difference of points; Determine the grey relational coefficient: （6） In the formula, For the i Comparison sequence X i With reference sequence Y In the k The correlation coefficient of the points is the resolution coefficient; S204, calculate the correlation: calculate the correlation between each subsequence and the reference sequence Y The average correlation of: （7） Where, is the comparison sequence X i With reference sequence Y The overall correlation between S205, analysis and decision: sorting is performed according to the calculated correlation degree. The greater the correlation degree, the greater the impact. Based on the result of the correlation degree, corresponding decisions or further analysis are made.

3. The gyroscope group fault prediction method based on MBKA-GBDT according to claim 1, characterized in that: In step S4, the improved black kite optimization algorithm MBKA is used to optimize the parameters of the gradient boosting decision tree GBDT. Specifically, the attack behavior determined by the population and the migration behavior determined by the population leader are updated simultaneously; like p < r , then update the individual's location information according to formula (11); like p ≥ r , then update the individual location information according to formula (12); like F k < F r , then according to formula (16), the leader is replaced, the migration behavior is abandoned, the individual position information is updated, and a new leader is selected; like F k ≥ F r , then continue to execute the migration behavior instructions according to formula (17) and update the leader's position information; Determine whether the maximum number of iterations has been reached. If t < T， Output the optimal parameter combination; if t ≥ T， Re-execute S4 and update the positions of the population individuals and the population leader again.

Images