IPKO-lightgbm-based fire control computer power module failure prediction method

Abstract

The IPKO-LightGBM-based fire control computer power module fault prediction method belongs to the technical field of power module fault diagnosis and comprises the following steps: step S01, collecting data of fire control computer power module pin signals, and dividing training sample data and test sample data; step S02, using an improved kingfisher optimization algorithm IPKO to optimize key parameters of a light gradient boosting machine LightGBM, and constructing a fault prediction model IPKO-LightGBM and the like. The improved kingfisher optimization algorithm IPKO is used to optimize main parameters of the light gradient boosting machine LightGBM, a fault prediction model is constructed, and the defect of blindness in parameter selection in the training process is compensated.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power module fault diagnosis, and particularly relates to a fire control computer power module fault prediction method based on IPKO-LightGBM. BACKGROUND

[0002] The reliability of the fire control computer power module directly affects the fire control solution accuracy, sensor data acquisition stability and response speed of the actuator. With the increasing use requirements, the power module needs to work continuously and stably in complex electromagnetic environments and severe temperature changes, and its failure may cause the overall failure of the fire control system, resulting in zero effectiveness and even safety accidents. Therefore, the fire control computer power module fault prediction has important value in ensuring the integrity of the equipment and reducing maintenance costs.

[0003] The current fire control computer power module fault prediction methods mainly include simulation analysis based on physical models, analysis methods based on vibration signals and traditional time series analysis methods, but the above methods have the following disadvantages: the simulation analysis based on physical models relies on accurate circuit parameter modeling and is difficult to adapt to the nonlinear characteristic changes caused by component aging and environmental disturbances. The analysis method based on vibration signals needs high-precision sensor array support, and the deployment cost is too high. The traditional time series analysis method is sensitive to noise interference, and the signal-to-noise ratio is seriously reduced in complex electromagnetic environments. In the field of intelligent algorithms, the existing methods still have significant limitations: the support vector machine SVM has a lack of adaptability in kernel function selection mechanism 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 expert system is limited by the scarcity of fault samples when building a knowledge base. The fault tree analysis method has an exponential growth of computational complexity when analyzing the multi-level topology structure of the power module due to the combination explosion problem. SUMMARY

[0004] In view of the above disadvantages and deficiencies of the prior art, the application provides a fire control computer power module fault prediction method based on IPKO-LightGBM, which optimizes the main parameters of the LightGBM (Light Gradient Boosting Machine) by using the improved IPKO (Imperial Parrot Optimization) algorithm, constructs a fault prediction model, and makes up for the blindness of parameter selection in the training process.

[0005] In order to achieve the above purpose, the main technical scheme adopted by the application includes:

[0006] The fire control computer power module fault prediction method based on IPKO-LightGBM includes the following steps:

[0007] Step S01, collect the data of the fire control computer power module pin signal, preprocess the collected data using KPCA algorithm, and divide the training sample data and test sample data;

[0008] Step S02, the improved kingfisher optimization algorithm IPKO is used to optimize the key parameters of the light gradient boosting machine LightGBM, and a fault prediction model IPKO-LightGBM is constructed, the improved kingfisher optimization algorithm IPKO includes introducing Bernoulli chaotic mapping in the initialization stage of the kingfisher optimization algorithm PKO, and adopting Gaussian random walk strategy to improve the position update of the kingfisher in the exploration stage;

[0009] Step S03, the training sample data in step S01 is used to train the fault prediction model IPKO-LightGBM constructed in step S02;

[0010] Step S04, the test sample data in step S01 is used to test the fault prediction model IPKO-LightGBM trained in step S03;

[0011] Step S05, the fault prediction model IPKO-LightGBM that passes the test in step S04 is used to predict the fault of the fire control computer power module.

[0012] Further, in step S02, Bernoulli chaotic mapping is introduced in the initialization stage of the kingfisher optimization algorithm PKO, and the optimized position update formula is:

[0013] ;

[0014] In the formula, is the position of the t+1th kingfisher in the dth dimension; i is the position of the t+1th kingfisher in the dth dimension; j is the position of the t+1th kingfisher in the dth dimension; is the position of the t+1th kingfisher in the dth dimension; i is the position of the t+1th kingfisher in the dth dimension; j is the position of the t+1th kingfisher in the dth dimension; is the Bernoulli mapping control parameter;

[0015] In the exploration stage, the position update formula of the kingfisher is:

[0016] ;

[0017] In the formula, is the position of the t+1th kingfisher; is a random number with mean and variance; is a random number with mean and variance; is a random number with mean and variance; is the position of the t+1th kingfisher;t Optimal individual of the next iteration Variance The first t The position of the plover in the n th iteration r 1 and r 2 are random numbers between [0, 1].

[0018] Further, in the step S02, the key parameters of the LightGBM are optimized by using the improved plover optimization algorithm IPKO, and the optimized parameters include the learning rate, the maximum number of iterations, the subsampling ratio, the maximum depth of the decision tree, and the minimum number of samples of the leaf node.

[0019] The present application has the following advantages:

[0020] The present application optimizes the main parameters of the LightGBM by using the improved plover optimization algorithm IPKO, constructs a fault prediction model, and makes up for the defects of blindness in parameter selection in the training process, and improves the prediction accuracy of the regression prediction model. The improved plover optimization algorithm IPKO introduces Bernoulli chaotic mapping in the initialization stage of the plover optimization algorithm PKO, and then improves the diversity of the initial population distribution, improves the convergence speed and convergence accuracy of the algorithm, and improves the exploration ability of the algorithm by using the Gaussian random walk strategy to improve the plover position update in the exploration stage, so as to avoid the algorithm from falling into premature convergence. DETAILED DESCRIPTION

[0021] In order to better explain the present application, the following specific embodiments are used to describe the present application in detail.

[0022] The present application provides a fire control computer power module fault prediction method based on IPKO-LightGBM, which comprises the following steps:

[0023] Step S01, collect the data of the fire control computer power module pin signal, and use the KPCA algorithm to preprocess the collected data, specifically, perform dimension reduction and normalization processing, and divide the training sample data and test sample data, wherein 80% of the data is used as the training sample data, and 20% of the data is used as the test sample data.

[0024] The KPCA algorithm can effectively process nonlinear data in data preprocessing, and through the kernel technology, the data is mapped to a high-dimensional space, so as to capture the complex structure in the data. In addition, the KPCA algorithm can realize dimension reduction, remove redundant features, and thus improve the performance of the subsequent model, and has good performance in noise reduction, which helps to improve the overall quality of the data.

[0025] Step S02, the key parameters of the light gradient boosting machine LightGBM are optimized by the improved pied kingfisher optimization algorithm IPKO, and a fault prediction model IPKO-LightGBM is constructed, the improved pied kingfisher optimization algorithm IPKO includes introducing Bernoulli chaotic mapping in the initialization stage of the pied kingfisher optimization algorithm PKO, and improving the position update of the pied kingfisher in the exploration stage by adopting the Gaussian random walk strategy.

[0026] The pied kingfisher optimization algorithm (PKO) is derived from the hunting behavior and symbiotic relationship of the pied kingfisher in nature. The pied kingfisher has a hunting strategy of aerial habitat. This hovering can allow it to stay in one place for a long time. The pied kingfisher usually circles over the water looking for food, then dives vertically with its beak forward to catch fish. Inspired by the behaviors of perching, hovering, diving and symbiosis of the pied kingfisher, the pied kingfisher optimization algorithm is developed, which has the advantages of strong optimization ability and fast convergence speed, and is suitable for global optimization problems. However, the algorithm still has room for improvement in terms of convergence and local optimal solution. Therefore, Bernoulli chaotic mapping and Gaussian random walk strategy are introduced to improve it, and the key parameters of the light gradient boosting machine LightGBM algorithm are optimized by the improved pied kingfisher optimization algorithm IPKO, so as to improve the prediction accuracy.

[0027] Specifically, the improved pied kingfisher optimization algorithm IPKO includes the following stages:

[0028] Initialization stage: mainly the initialization of the distribution of the population in the optimization space, the initial population formula is:

[0029] (1-1);

[0030] In the formula: is the position of the ith individual in the jth dimension; rand is a random value between 0 and 1; and are the upper and lower boundaries of the optimization problem respectively; j is the spatial dimension.

[0031] In order to further expand the range of the initial population to improve the local search ability, in view of the problem of uneven distribution of the initial population in the basic pied kingfisher optimization algorithm PKO, Bernoulli chaotic mapping is introduced in the initialization stage to improve the diversity of the initial population distribution, thereby improving the convergence speed and accuracy of the algorithm. The optimized position update formula is:

[0032] (1-2);

[0033] In the formula: is the position of the i+1th pied kingfisher in the jth dimension; is the position of the i-th bird in the j-th dimension of the space; is the control parameter of the Bernoulli map.

[0034] Exploration phase: The exploration phase is derived from the perching and hovering behavior of the birds. Observations of the birds in their natural habitat show that they alternate between perching and hovering positions depending on various factors. The update formula is given by:

[0035] (1-3);

[0036] where, is the position of the i-th bird in the j-th dimension of the space; is the position of the i-th bird in the j-th dimension of the space; is a random control parameter, , is a random number from a normal distribution, is the dimension of the problem under consideration; is the position of the i-th bird in the j-th dimension of the space; is the control parameter.

[0037] When 0.5 < rand < 0.8, the perching strategy is executed, and the parameter T is calculated by:

[0038] (1-4);

[0039] where, is the current iteration number; is the maximum iteration number; is the jump factor, which is set to 8; is the control parameter of the bird's crest angle; rand represents a random value.

[0040] When 0 < rand ≤ 0.5, the hovering strategy is executed, and the parameter T is calculated by:

[0041] (1-5);

[0042] where, is the current iteration number; is the maximum iteration number; is the jump factor, which is set to 8; and are the fitness values of the i-th and j-th birds, respectively.

[0043] To improve the problem of insufficient exploration ability in the basic kingfisher optimization algorithm, the position update of the exploration stage of the kingfisher is improved by using the strategy of Gaussian random walk, so as to improve the exploration ability of the algorithm and avoid the algorithm from falling into premature convergence. The formula improved by introducing Gaussian random walk is:

[0044] (1-6);

[0045] where, is the position of the kingfisher at the t+1th iteration; is a random number with mean and variance; is the optimal individual at the tth iteration; is the variance; is the position of the kingfisher at the tth iteration; r1 and r2 are random numbers between 0 and 1. Development stage: The diving behavior of the kingfisher makes it a highly efficient hunter, which is one of the reasons why this species is so successful in its habitat. The kingfisher's fast and accurate diving ability, combined with its sharp beak and excellent vision, make it an effective predator in aquatic environments. The mathematical expression for the development stage is:

[0046] Development stage: The diving behavior of the kingfisher makes it a highly efficient hunter, which is one of the reasons why this species is so successful in its habitat. The kingfisher's fast and accurate diving ability, combined with its sharp beak and excellent vision, make it an effective predator in aquatic environments. The mathematical expression for the development stage is:

[0047] (1-7);

[0048] where, is the position of the kingfisher at the t+1th iteration; is the position of the kingfisher at the tth iteration; represents hunting ability; represents diving ability; is the optimal position of the current kingfisher; is the flapping frequency of the kingfisher's wings; is a random control parameter.

[0049] Symbiotic stage: If the newly generated solution is closer to the optimal solution than the original solution, replace the original solution and execute the symbiotic stage. The symbiotic stage is inspired by the symbiotic relationship between otters and kingfishers, which allows the kingfisher to benefit from the hunting behavior of the otter. This behavior is represented as:

[0050] (1-8);

[0051] where, is the position of the kingfisher at the t+1th iteration; is the position of the kingfisher at the tth iteration; and are the positions of two randomly selected kingfisher individuals.​ for predation efficiency; represents hunting ability; is a random control parameter; rand represents a random value between 0 and 1.

[0052] LightGBM is a high-efficiency gradient boosting algorithm, designed with lightweight and high performance as its core, especially suitable for processing large-scale data and high-dimensional feature datasets. In the task of fault data regression prediction, the model training based on LightGBM realizes step-by-step optimization by constructing a series of decision trees, and each tree is trained based on the prediction residual of the previous model. In each iteration process, LightGBM quickly determines the optimal split point by calculating the gradient of the current prediction error, thereby effectively minimizing the loss function. This gradient-driven splitting mechanism significantly improves the computational efficiency and prediction accuracy of the algorithm. In the regression task, the model gradually optimizes the prediction performance of each decision tree through multiple iterations, and finally generates a strong prediction model that can accurately predict continuous target variables. The core algorithm process mainly includes the following key steps:

[0053] Step 1: Calculate the fitting residual at the tth iteration, the formula is:

[0054] (2-1);

[0055] where: t is the iteration number, m is the mth tree, is the predicted label, is the model prediction value of the mth tree, is the fitting residual of the mth tree.

[0056] Step 2: Calculate the prediction update value, the formula is:

[0057] (2-2);

[0058] where: is the prediction update value of the mth tree, represents the prediction value of the mth tree, represents the learning rate set by the model, represents the input at the tth iteration.

[0059] In the modeling process of the LightGBM algorithm, multiple key parameters have important influence on the model performance and prediction accuracy. Among them, the key parameters include learning rate learning_rate, maximum iteration number n_estimators, subsample ratio subsample, maximum depth of decision tree max_depth and minimum sample number of leaf node min_samples_leaf. Learning rate learning_rate determines the step size of each model update, which directly affects the convergence speed and stability. If the learning rate is set too high, the model training process may be unstable or even divergent. If it is set too low, more iteration times may be needed to achieve the desired convergence effect. Correspondingly, the maximum iteration number n_estimators is inversely proportional to the learning rate. When the learning rate is low, the iteration number needs to be increased to avoid premature convergence of the model and improve the model performance. The subsample ratio subsample is used to control the proportion of samples used for fitting in each iteration. Setting it to less than 1 can effectively suppress overfitting and improve the generalization ability of the model. Regarding the complexity control of the decision tree, the maximum depth max_depth and the minimum sample number of leaf node min_samples_leaf are important adjustment parameters. Reasonable setting of these parameters can balance the complexity of the model, so as to achieve a good balance between overfitting and underfitting. Therefore, finding the appropriate parameter combination is crucial to improve the prediction accuracy and computational performance of the model.

[0060] Step S03, training the fault prediction model IPKO-LightGBM constructed in step S02 by using the training sample data in step S01;

[0061] Step S04, testing the fault prediction model IPKO-LightGBM trained in step S03 by using the test sample data in step S01;

[0062] Step S05, using the fault prediction model IPKO-LightGBM that passes the test in step S04 to predict the fault of the fire control computer power module.

[0063] The fault prediction process of the fire control computer power module of the present application is as follows:

[0064] Step 1), input the data set, and use KPCA algorithm for pretreatment to divide the training sample and test sample data.

[0065] Step 2), initialize the control parameters in IPKO, including population size and maximum iteration number, etc.

[0066] Step 3), use Bernoulli chaotic mapping to generate initial population position.

[0067] Step 4), the classification error of the test sample is taken as the fitness of the individual of the kingfisher.

[0068] Step 5), the value of the rand parameter is judged, when 0.8 < rand < 1, at this time, it is in the development stage and the position of the kingfisher is updated according to formula (1-7).

[0069] Step 6), when 0.5 < rand < 0.8, the kingfisher executes the habitat strategy and updates the position of the kingfisher according to formula (1-6) and (1-4); when 0 < rand ≤ 0.5, the kingfisher executes the hovering strategy and updates the position of the kingfisher according to formula (1-6) and (1-5).

[0070] Step 7), the position of the kingfisher in the symbiotic stage is updated according to formula (1-8).

[0071] Step 8), the current best fitness and the corresponding position of the kingfisher are updated.

[0072] Step 9), it is judged whether the termination condition has been reached, if not, return to step 5).

[0073] Step 10), the optimal parameter combination obtained is assigned to the LightGBM model.

[0074] Step 11), the IPKO-LightGBM classification prediction model is constructed with the training sample data, and finally the accuracy of the model is verified by using the test sample data.

[0075] Although the embodiments of the present application have been shown and described above, it can be understood that the above-mentioned embodiments are exemplary and cannot be understood as limiting the present application, and the modifications, modifications, replacements and variations of the above-mentioned embodiments made by those skilled in the art are within the scope of the present application.

Claims

1. A method for predicting faults in the power supply module of a fire control computer based on IPKO-LightGBM, characterized in that, Includes the following steps: Step S01: Collect data from the pin signals of the fire control computer power module, and preprocess the collected data using the KPCA algorithm to divide the training sample data and test sample data; the KPCA algorithm reduces dimensionality and removes redundant features, thereby improving the performance of subsequent models. Step S02: Optimize the key parameters of the Lightweight Gradient Boosting Machine (LightGBM) using the improved Kingfisher Optimization Algorithm (IPKO) to construct the fault prediction model IPKO-LightGBM. The improved Kingfisher Optimization Algorithm (IPKO) includes introducing Bernoulli chaotic mapping in the initialization stage of the Kingfisher Optimization Algorithm (PKO) to increase the diversity of the initial population distribution, thereby improving the convergence speed and convergence accuracy of the algorithm. In the exploration stage, a Gaussian random walk strategy is used to improve the Kingfisher position update. Step S03: Train the fault prediction model IPKO-LightGBM constructed in step S02 using the training sample data from step S01. Step S04: Test the fault prediction model IPKO-LightGBM trained in step S03 using the test sample data from step S01. Step S05: Use the fault prediction model IPKO-LightGBM, which passed the test in step S04, to predict the faults of the fire control computer power module. In step S02, Bernoulli chaotic mapping is introduced during the initialization phase of the Kingfisher Optimization Algorithm (PKO), and the optimized position update formula is: In the formula: X i+1,j Let X be the position of the (i+1)th kingfisher in the j-th dimension; i,j Let λ be the position of the i-th kingfisher in the j-th dimension; λ is the Bernoulli mapping control parameter. During the exploration phase, a Gaussian random walk strategy is used to update the kingfisher's position using the following formula: In the formula, X i (t+1) represents the position of the kingfisher in the (t+1)th iteration; It is to generate a σ and σ are random numbers representing the mean and variance; Let X be the optimal individual in the t-th iteration; σ is the variance; X i (t) represents the position of the kingfisher in the t-th iteration; r1 and r2 are random numbers between [0,1].

2. The method for predicting the fault of the power supply module of the fire control computer based on IPKO-LightGBM according to claim 1, characterized in that: In step S02, the improved Kingfisher Optimization Algorithm (IPKO) is used to optimize the key parameters of the Lightweight Gradient Boosting Machine (LightGBM). The parameters to be optimized include the learning rate, the maximum number of iterations, the subsampling ratio, the maximum depth of the decision tree, and the minimum number of samples in the leaf nodes.

Images