An alternating current machine fault diagnosis method based on ISBOA-Adaboost

Abstract

The application relates to the technical field of artificial intelligence fault diagnosis, and discloses an alternating current machine fault diagnosis method based on ISBOA-Adaboost. The collected data is pretreated through a VIKOR algorithm, the snake heron optimization algorithm SBOA is improved, a Sinusoidal chaotic mapping is introduced in the SBOA population initialization stage to improve the population initialization uniformity, and the snake heron position is updated according to the current iteration number of the algorithm in the SBOA hunting stage; an adaptive weight factor is introduced in the SBOA escape stage to optimize the snake heron position update, the improved snake heron optimization algorithm ISBOA is used to optimize parameters of an iteration algorithm Adaboost, an ISBOA-Adaboost fault diagnosis model is constructed, and the blindness of parameter selection in the training process is compensated.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of alternating current machine fault diagnosis, and particularly relates to an alternating current machine fault diagnosis method based on ISBOA-Adaboost. BACKGROUND

[0002] The alternating current machine is a key component of a military gun control system and is mainly composed of a coaxial motor-generator set and a frequency stabilizer and the like. The main task of the alternating current machine is to perform power conversion to provide appropriate three-phase alternating current for the power transformer of the gun control box, so that the alternating current machine plays a crucial role in the gun control system. Through fault diagnosis of the tank alternating current machine, the battlefield survival capability of the tank armor can be significantly improved, the stability of the overall system can be effectively improved, the maneuverability and firepower output on the battlefield can be enhanced to a certain extent, and the overall combat effectiveness can be further improved.

[0003] In the field of alternating current machine fault diagnosis, time-frequency analysis and feature extraction methods are usually used for processing. Through short-time Fourier transform (STFT) or wavelet packet decomposition, the frequency domain features of alternating current machine vibration and sound signals are extracted, and threshold determination rules are combined to predict mechanical faults (such as bearing wear and rotor imbalance). However, this method requires a pre-defined fault feature library, and new types of faults are difficult to identify. In terms of model selection, support vector machines (SVM) are usually selected to improve classification accuracy, but the kernel function selection mechanism lacks adaptability when processing multi-channel sensor data of the power module, such as voltage, current, temperature and other multi-dimensional time series signals, resulting in inaccurate feature space mapping. The construction of the knowledge base of the expert system is limited by the scarcity of fault samples. When analyzing the multi-level topology structure of the power module, the combinatorial explosion problem of the fault tree analysis method causes the computational complexity to increase exponentially.

[0004] The AdaBoost algorithm overcomes the limitations of single classifiers by integrating multiple weak classifiers, and performs outstandingly in handling noise and sample imbalance. Compared with support vector machines (SVM) and decision trees, AdaBoost has higher training efficiency and accuracy, and can effectively improve the generalization ability of the model. Through iterative weighting of error samples, AdaBoost can improve the diagnosis capability for complex and dynamic fault conditions. Compared with expert systems and other traditional algorithms, AdaBoost has obvious advantages in real-time performance and accuracy, and has become a powerful tool in fault diagnosis. SUMMARY

[0005] To address the aforementioned shortcomings and deficiencies in existing technologies, this invention provides an AC fault diagnosis method based on ISBOA-Adaboost. It employs an improved ISBOA optimization algorithm to optimize the parameters of the iterative Adaboost algorithm, constructing an ISBOA-Adaboost fault diagnosis model. This overcomes the shortcomings of blind parameter selection during training, and includes the following steps:

[0006] S1, Acquire signal data from the AC unit module;

[0007] S2. The collected data is preprocessed using the VIKOR multi-criteria compromise solution ranking method to select datasets that can be used as model input; the model input dataset is divided into test dataset and training dataset.

[0008] S3. Improvements to the Snake-Heron Optimization Algorithm (SBOA), including introducing a Sinusoidal chaotic mapping to improve population initialization uniformity during the SBOA population initialization phase; updating the snake-heron positions based on the current iteration number during the SBOA hunting phase; and introducing an adaptive weighting factor during the SBOA escape phase. An improved snake-heron optimization algorithm, ISBOA, is obtained by optimizing the snake-heron location update.

[0009] S4. The improved snake-heron optimization algorithm ISBOA is used to optimize the key parameters of the iterative algorithm Adaboost, and the ISBOA-Adaboost fault diagnosis model is constructed. The fault diagnosis model is trained using the training dataset in S2.

[0010] S5. Input the test dataset from S2 into the trained ISBOA-Adaboost model to perform fault diagnosis on the AC unit and output the prediction results.

[0011] In step S3, a sinusoidal chaotic mapping is introduced during the SBOA population initialization phase. The optimized position update formula is as follows:

[0012] ;

[0013] In the formula: For the first i +1 snake-heron position; For the first i Location of the snake-heron; a For control parameters;

[0014] Introducing an adaptive weighting factor during the SBOA escape phase The formula for updating the location of the snake-heron is:

[0015] ;

[0016] ;

[0017] ;

[0018] wherein, is an adaptive weight factor; is the position of the snake eagle in the escape phase; is the optimal position in the current phase; is the random position in the current iteration; and RB are random numbers between 0 and 1; is the initial position of the snake eagle; t is the current iteration number; T is the maximum iteration number; K is a random value 1 or 2; r is a random parameter.

[0019] The improved snake eagle optimization algorithm ISBOA of the step S4 optimizes parameters of the iterative algorithm Adaboost, and the optimized parameters include a learning rate, a number of weak classifiers, a decision tree depth, and an updating mode of sample weights.

[0020] Compared with the prior art, the application has the following beneficial technical effects and advantages:

[0021] The improved snake eagle optimization algorithm ISBOA is used to optimize parameters of the iterative algorithm Adaboost, an ISBOA-Adaboost fault diagnosis model is constructed, the blindness of parameter selection in the training process is compensated, the prediction accuracy of the regression prediction model is improved, the uniformity of population initialization in the SBOA population initialization stage is improved by introducing a Sinusoidal chaotic mapping, and the convergence rate of the algorithm is improved; the position of the snake eagle is updated according to the current iteration number of the algorithm in the SBOA hunting phase; and an adaptive weight factor is introduced in the SBOA escape phase The position of the snake eagle is optimized, and local optimization is avoided. DETAILED DESCRIPTION

[0022] The application will be described in detail below.

[0023] The application provides an alternating current machine fault diagnosis method based on ISBOA-Adaboost, which comprises the following steps:

[0024] S1, signal data values of alternating current machine modules are collected by an equipment test bench as initial data for experiments.

[0025] S2, the collected data are pretreated by a multi-criteria compromise solution ordering method VIKOR, and a data set that can be used as model input is screened out; the model input data set is divided into a test data set and a training data set.

[0026] Specifically, in step S2, the collected data is evaluated, sorted, and normalized using the VIKOR algorithm.

[0027] S201. Constructing the initial decision matrix: Let there be *n* evaluation objects with *n* parameter indicators, and determine the object set.

[0028] and indicator set Construct a decision matrix ;

[0029] S202, Regarding the decision matrix Normalization: Using standard 0-1 transformation, the constructed decision matrix is ​​normalized. Standardization is achieved through normalization. , The specific calculation formula is as follows:

[0030] When the indicator When the indicator is a benefit-type indicator,

[0031] (1-1)

[0032] When the indicator When it is a cost-type indicator,

[0033] (1-2)

[0034] In the formula The normalized decision matrix, The maximum value in the matrix. It is the minimum value in the matrix;

[0035] S203, Based on the normalized matrix Determine the ideal solution and negative ideal solution :

[0036] (1-3)

[0037] in, (1-4)

[0038] In the formula The maximum value in the normalized matrix. This represents the minimum value of the normalized matrix.

[0039] S204, Calculate the weight vector Determining indicators using the method of constructing an optimization model weight The weight vector is calculated as follows:

[0040] Using the positive ideal solution of the decision matrix and negative ideal solution The initial weight vector is calculated. :

[0041] (1-5)

[0042] For the initial vector After normalization, we get:

[0043] (1-6)

[0044] Then the indicator The weight is ;

[0045] In the formula To provide the solution after normalization to the ideal solution. The normalized matrix, For the positive ideal solution, It is a negative ideal solution.

[0046] S205, Calculate the evaluation objects. Group benefit value Individual Regret Value compromise value :

[0047] (1-7)

[0048] (1-8)

[0049] (1-9)

[0050] in, , which represents the maximum group benefit value; , which is the minimum group benefit value;

[0051] , which is the maximum individual regret value; , is the minimum individual regret value; The normalized weights of each indicator; For the positive ideal solution, The negative ideal solution is represented by v, where v is the decision mechanism coefficient.

[0052] For the evaluation object , , These three values ​​are sorted in descending order, resulting in three sorting methods: Sort, Sort and Sorting, the final sorting result is based on value determination.

[0053] S206, determining the optimal scheme: the evaluation objects are sorted according to the value to obtain a scheme , and the scheme with the minimum value is selected as the optimal scheme;

[0054] If the value of the evaluation object in the scheme satisfies , the compromise scheme is obtained by verifying the sorting condition 1 and the sorting condition 2: Sorting condition 1:

[0055] The value of the scheme with the minimum value and the second minimum value satisfies: , (1-10)

[0056] Sorting condition 2: The value of the scheme with the minimum value and the second minimum value satisfies: , (1-10) , and

[0057] If the sorting condition 1 and the sorting condition 2 are satisfied at the same time, the scheme is the final compromise scheme in the decision-making process; if only the sorting condition 1 is satisfied or only the sorting condition 2 is satisfied, the scheme and the scheme are both compromise schemes, and the scheme is selected as the final compromise scheme; if the sorting condition 1 and the sorting condition 2 are not satisfied at the same time, the maximum J is obtained by , and at this time the scheme is the final compromise scheme.

[0058] S3, Secretary bird optimization algorithm (SBOA) is a swarm intelligence optimization algorithm, and the algorithm is inspired by the survival behavior of secretary bird in natural environment. The survival of secretary bird needs to constantly find prey and avoid the pursuit of predators. In order to solve the problems of poor convergence, low exploration efficiency and easy to fall into local optimum, the secretary bird optimization algorithm SBOA is improved, including introducing sinusoidal chaotic mapping in the population initialization stage of SBOA to improve the uniformity of population initialization and improve the convergence rate of the algorithm; in the hunting stage of SBOA, the position of secretary bird is updated according to the current iteration number of the algorithm; in the escape stage of SBOA, the adaptive weight factor is introduced to optimize the position update of secretary bird, avoid falling into local optimum, and get the improved secretary bird optimization algorithm ISBOA.​​​​​​​

[0059] (1) Initialization phase: In the initial stage of the algorithm, each snake eagle is randomly distributed in the search space, which can be expressed as: , (2-1)

[0060] In the formula: is the initial position of the snake eagle; and are the upper and lower boundaries of the algorithm search range, respectively; is a random number between [0, 1]; i is the snake eagle individual; j is the dimension of the search problem.

[0061] In the initialization phase, there is still the problem of uneven initialization, so the Sinusoidal chaotic mapping strategy is introduced to improve the snake eagle optimization algorithm. Sinusoidal chaotic mapping has the advantages of uniform traversal and fast convergence speed, which can make the position of the population initialization uniformly distributed, thereby improving the convergence rate of the algorithm. The updated formula of the improved initialization position is as follows: , (2-2)

[0062] In the formula: is the position of the i+1th snake eagle; is the position of the ith snake eagle; a is a control parameter, generally taken as 2.3.

[0063] (2) Hunting phase: The snake eagle in the hunting phase is divided into three small stages, respectively: finding, consuming, and attacking prey, each small stage accounting for 1 / 3 of the algorithm iteration number. Set t as the current iteration number; T is the maximum iteration number.

[0064] When , the current is the finding stage, and the snake eagle position is updated by the following formula:

[0065] , (2-3)

[0066] When , the current is the consumption stage, and the snake eagle position is updated by the following formula:

[0067] , (2-4)

[0068] When , the current is the attack prey stage, and the snake eagle position is updated by the following formula:

[0069] , (2-5)

[0070] In the formula, is the position of the snake eagle in the hunting stage; and are the random positions in the iteration process of this stage; and RB are random numbers between [0, 1]; RL is a random number between [0, 0.5]; is the initial position of snake eagle; is the optimal position of the current stage.

[0071] (3) Escape stage: when the snake eagle encounters a threat, the escape strategy is started, which is divided into fast escape or camouflage. Set r is a random value;

[0072] When 0 < RL < 0.5, r <0.5, the fast escape strategy is executed, and the snake eagle position is updated by the following formula: , (2-6)

[0073] When 0.5 ≤ RL < 1, r <1, the camouflage strategy is executed, and the snake eagle position is updated by the following formula: , (2-7)

[0074] In the formula, is the position of the snake eagle in the escape stage; is the optimal position of the current stage; is a random position in the current iteration; is a random number between [0, 1]; is the initial position of the snake eagle; K is a random value of 1 or 2.

[0075] The position update of the snake eagle in the escape stage will approach the local optimal solution, but at this time it can only approach the optimal solution, so it cannot perform better local optimization. In order to solve the problems of the algorithm being easy to fall into local optimum and slow convergence speed, an adaptive weight factor not only can affect the ability and accuracy of search, but also can speed up the convergence speed of the algorithm, effectively reduce the situation of falling into local optimum, and the improved position update formula is: adaptive weight factor , (2-8)

[0076] When 0 < RL < 0.5, r <0.5, the fast escape strategy is executed, and the snake eagle position is updated by the following formula: , (2-9)

[0077] When 0.5 ≤ RL < 1, r <1, the camouflage strategy is executed, and the snake eagle position is updated by the following formula: , (2-10).

[0078] S4, the improved snake eagle optimization algorithm ISBOA is used to optimize the parameters of the iterative algorithm Adaboost, and an ISBOA-Adaboost fault diagnosis model is constructed to train the training data set.

[0079] AdaBoost is a classic boosting algorithm that combines multiple weak classifiers to improve the overall diagnostic ability of the model. The core idea of AdaBoost is to gradually correct the errors of the previous classifier by combining multiple simple classifiers through weighting, thereby improving the overall performance of the classifier. In the training process, AdaBoost calculates the diagnostic error of each sample and adjusts the weight of the sample according to the error rate. The weight of the misclassified sample will increase, so that the subsequent classifier will pay more attention to these difficult samples. In each iteration, AdaBoost adjusts the weight of the sample according to the performance of the current classifier and combines the results of the previous round of classifiers to gradually form a strong classifier. In the classification task, the final model is obtained by weighted average of multiple weak classifiers, and the key steps of the algorithm are as follows:

[0080] Step1: Initialize the weight of all training samples, set it as uniform distribution, that is, the weight of each sample is the same: , (3-1)

[0081] In the formula: is the weight of sample i in the first round of training, initially, all sample weights are equal, that is, all samples are treated equally.

[0082] Step2: Iterative training of multiple weak classifiers:

[0083] Train weak classifier under current weight distribution: use training set D and current sample weight Train a weak classifier .

[0084] Calculate classification error rate : Calculate the weighted error rate of weak classifier under the current weight distribution, that is, the total weight of the misclassified sample under the current weight distribution:

[0085] , (3-2)

[0086] In the formula: is the classification error rate; is the weak classifier; represents the sample weight of the current iteration; is the indicator function, if is 1, otherwise is 0.

[0087] Calculate the weight coefficient of weak classifier : Use the classification error rate to determine the weight of the current weak classifier, the smaller the error rate, the larger the weight coefficient :

[0088] , (3-3)

[0089] where: is the weight coefficient of the weak classifier; is the classification error rate; if , then , which means that the classifier has certain discrimination ability; if , then , the effect of the classifier is equivalent to random guessing.

[0090] The sample weights are updated so that the next round of weak classifiers pays more attention to the samples that are currently misclassified. The new weight is defined as: , (3-4)

[0091] where: is the new weight of the t+1th iteration; is the weight of the tth iteration; is the normalization factor, so that the sum of all sample weights is 1; is the weak classifier; is the weight coefficient of the weak classifier; is the label of the training set.

[0092] Step 3: Combine weak classifiers

[0093] After T rounds of training, all weak classifiers are combined into a strong classifier: , (3-5)

[0094] where: is the strong classifier; is the weak classifier; is the weight coefficient of the weak classifier; t is the current iteration number, and T is the maximum iteration number.

[0095] The final classification result is: , (3-6)

[0096] where: the sign function represents the sign; is the output of the final classifier; is the weight coefficient of the weak classifier; t is the current iteration number, and T is the maximum iteration number.

[0097] Optimize the parameters of the improved algorithm;

[0098] The AdaBoost algorithm also involves multiple key parameters in the modeling process, including the learning rate (learning_rate), the number of weak classifiers (n_estimators), the decision tree depth (max_depth), and the updating method of sample weights. The learning rate controls the adjustment amplitude of the weak classifier's weight in each iteration. If the learning rate is set too large, it may lead to overfitting of the model, and if it is set too small, it may slow down the model training and even fail to fully improve the performance of the classifier. The number of weak classifiers controls the number of iterations. Generally, increasing the number of iterations helps improve the accuracy of the model, but it also easily causes overfitting, so it needs to be reasonably set according to the performance of the training set. The decision tree depth (if a decision tree is used as a weak classifier) controls the complexity of each weak classifier. A tree that is too deep may cause overfitting, while a tree that is too shallow may not be able to capture the complex patterns in the data. The updating method of sample weights determines the degree of attention that AdaBoost pays to misclassified samples. The weights of misclassified samples are increased, so that subsequent classifiers pay more attention to these difficult-to-classify samples. In order to ensure the performance of the model, reasonable adjustment of these parameter combinations is crucial for the training effect and diagnostic accuracy of AdaBoost.

[0099] S5, input the test data set into the ISBOA-Adaboost prediction model, perform fault diagnosis on the AC machine, and output the prediction result.

[0100] The AC machine fault prediction process of the application is as follows:

[0101] Step 1), input the data set, pre-process it using the VIKOR algorithm, and divide the training sample and test sample data;

[0102] Step 2), initialize the control parameters in ISBOA, including population size and maximum number of iterations, etc.

[0103] Step 3), use Sinusoidal chaotic mapping to generate initial population positions;

[0104] Step 4), update the positions of the hunting stage snake heron: judge the current number of iterations, when , it is in the search stage and the position of the snake heron is updated according to formula (2-3); when , it is in the consumption stage and the position of the snake heron is updated according to formula (2-4); when , it is in the attack stage and the position of the snake heron is updated according to formula (2-5);

[0105] Step 5), update the position of the escape stage snake heron, when 0 r , the snake heron executes a fast escape strategy and updates the position according to formula (2-9); when 0.5r When <1, the snake heron executes a camouflage strategy and updates the position according to formula (2-10);

[0106] Step 6), update the current best fitness and the corresponding snake heron position;

[0107] Step 7), determine whether the termination condition has been reached, if not, return to step 4);

[0108] Step 8), assign the obtained optimal parameter combination to the Adaboost model;

[0109] Step 9), construct the ISBOA-Adaboost fault diagnosis model with the training data set, and finally verify the accuracy of the model by using the test data set.

[0110] Although the embodiments of the present application have been shown and described above, it is to be understood that the above-described embodiments are exemplary only, and are not to be construed as limiting the present application, and any modification, substitution, replacement and variation of the above-described embodiments made by those skilled in the art shall fall within the scope of the present application.

[0111] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An ISBOA-Adaboost-based alternating current machine fault diagnosis method, characterized in that, The method comprises the following steps: S1, collecting alternating current machine module signal data; S2, preprocessing the collected data by VIKOR to filter out the data set that can be used as model input; dividing the model input data set into test data set and training data set; S3, improving the snake hawk optimization algorithm SBOA, including introducing Sinusoidal chaotic mapping in the SBOA population initialization stage to improve the uniformity of population initialization; In the SBOA hunting stage, the snake hawk position is updated according to the current iteration number of the algorithm; in the SBOA escape stage, an adaptive weight factor w(t) is introduced to optimize the snake hawk position update, and an improved snake hawk optimization algorithm ISBOA is obtained; Specifically, (1) initialization stage: introducing Sinusoidal chaotic mapping in the SBOA population initialization stage, and the optimized position update formula is: X i+1 = aX i 2 in(πX i ); wherein: X i+1 is the (i+1)th snake position; X i is the ith snake position; a is a control parameter; (2) Hunting stage: the snake hawk is divided into three small stages in the hunting stage, which are: searching, consuming and attacking prey; (3) Escape stage: when the snake hawk encounters a threat, it will start the escape strategy, which is divided into fast escape or camouflage; In the SBOA escape stage, an adaptive weight factor w(t) is introduced to update the snake hawk position formula: When 0 < r < 0.5, the fast escape strategy is executed, and the snake hawk position is updated by the following formula: When 0.5 ≤ r < 1, the camouflage strategy is executed, and the snake hawk position is updated by the following formula: where w(t) is an adaptive weight factor; is the position of the snake head during the escape phase. X best is the optimal position for the current stage; X random is a random position in the current iteration; R2 and RB are random numbers between 0 and 1; X i is the initial position of the snake eagle; t is the current iteration number; T is the maximum iteration number; K is a random value of 1 or 2; r is a random parameter; S4, using the improved snake hawk optimization algorithm ISBOA to optimize the key parameters of the iteration algorithm Adaboost, constructing an ISBOA-Adaboost fault diagnosis model, and training the fault diagnosis model through the training data set in S2; S5, inputting the test data set in S2 into the trained ISBOA-Adaboost model to diagnose the faults of the alternating current machine, and outputting the prediction result; The step S4 improves the snake hawk optimization algorithm ISBOA to optimize the parameters of the iteration algorithm Adaboost, and the optimized parameters include learning rate, number of weak classifiers, decision tree depth and sample weight update mode.

Images