Fire control computer power supply module health prediction method

Abstract

A fire control computer power supply module health prediction method belongs to the fire control computer power supply module fault diagnosis technical field, and comprises the following steps: S01, collecting a fire control computer power supply module power supply signal as original data; s02, adopting a multi-scale geometric analysis algorithm (MGA) to separate different characteristics of signals from the collected original data; and step S03, optimizing key parameters of the GRU by adopting an improved subtraction average optimization algorithm MSABO, and constructing a health prediction model MSABO-GRU and the like. According to the invention, a multi-scale geometric analysis algorithm MGA is adopted to preprocess the collected power signal data, so that the prediction efficiency of the model is improved; an improved subtraction average optimization algorithm MSABO is adopted to predict key parameters of the GRU, and the accuracy of model prediction is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis of the power supply module of a fire control computer, and particularly relates to a method for predicting the health of the power supply module of a fire control computer. Background Art

[0002] The fire control computer is the core component for controlling the operation of the weapon control firepower system, and its performance is directly related to the combat ability and tactical advantages of the equipment on the battlefield. As the energy supply for the operation of the fire control computer, the state of the power supply system not only affects the stable operation of the equipment, but also is an early warning signal for potential faults. Therefore, it has important practical significance to predict the health of the power signal of the fire control computer power supply module.

[0003] In the field of health prediction, traditional methods such as statistical methods, classical machine learning methods, and other deep learning algorithms such as fully connected neural networks and convolutional neural networks have achieved certain results, but there are limitations in dealing with data with time series characteristics such as power signals. When the time series of the power signal is relatively complex, it is often difficult to establish the model of the statistical method, the model generalization ability of the classical machine learning method is poor, and the convolutional neural network requires a large amount of computing resources. Summary of the Invention

[0004] In view of the above-mentioned disadvantages and deficiencies of the prior art, the present invention provides a method for predicting the health of the power supply module of a fire control computer, which uses the multi-scale geometric analysis algorithm MGA to preprocess the collected power signal data to improve the prediction efficiency of the model; and uses the improved subtraction average optimization algorithm MSABO to optimize the key parameters of the gated recurrent unit network GRU to improve the prediction accuracy of the model.

[0005] In order to achieve the above object, the main technical solutions adopted by the present invention include: A method for predicting the health of the power supply module of a fire control computer, comprising the following steps: Step S01, collecting the power signal of the power supply module of the fire control computer as the original data; Step S02, using the multi-scale geometric analysis algorithm MGA to separate different features of the collected original data, extract valuable geometric information and suppress noise interference, screen out the key information signals to construct a data set, and divide the training set and the test set; Step S03, using the improved subtraction average optimization algorithm MSABO to optimize the key parameters of the gated recurrent unit network GRU, and constructing a health prediction model MSABO-GRU, where the improved subtraction average optimization algorithm MSABO includes introducing a Circle chaotic mapping and a reverse learning strategy in the initialization stage, and introducing a Levy flight strategy in the position update stage; Step S04: Use the training set in Step S02 to train the health prediction model MSABO-GRU in Step S03; Step S05: Use the test set in Step S02 to test the trained health prediction model MSABO-GRU in Step S04; Step S06: Use the qualified health prediction model MSABO-GRU to perform health prediction on the power supply module of the fire control computer and output the prediction result.

[0006] Further, in the said Step S02, separating different features of the signal by using the multi-scale geometric analysis algorithm MGA includes the following steps: Step S201: Signal decomposition: Decompose the original signal f(t) into components at different scales according to the following formula: ; Where: f(t) represents the original signal; and are respectively the low-frequency and high-frequency parts of the multi-scale basis function; and are respectively the low-frequency component and the high-frequency component; j is the scale parameter; k is the position parameter; J represents the maximum scale layer of decomposition; Step S202: Geometric feature extraction: Extract the geometric characteristics of the decomposed high-frequency component including local singularity detection and multi-directional feature extraction; local singularity detection is to analyze the modulus of the high-frequency component to detect signal mutation points; multi-directional feature extraction is to extract signal features in different directions by using directional basis functions; Step S203: Signal reconstruction: Select important components according to the following formula to eliminate noise interference, ; Where: is the reconstructed signal; S is the set of retained components; Step S204: Extract the state characteristics of the fire control system.

[0007] Further, in the said Step S202, the directional basis function is the Curvelet function: ; Where: and are respectively the components of the high-frequency component in the x and y directions.

[0008] Further, in the said Step S03, the improved subtraction average optimization algorithm MSABO includes introducing the Circle chaotic map and the reverse learning strategy in the initialization stage, including: Initialize the population position according to the following formula: ; where, X i is the position of the i-th individual; ub and lb are the lower and upper bounds of the spatial search respectively; rand(0,1) is a random number uniformly distributed between [0,1]; Introduce the Circle chaotic mapping formula in the initialization stage as follows: ; where, X inew is the new individual position after Circle chaotic mapping; mod represents the modulo operation; Introduce the opposition-based learning strategy, and calculate the opposition position Y of the individual according to the following formula i : ; When the fitness value of the opposition-based learning individual is less than that of the original individual, retain the position of the opposition-based learning individual Y i ; otherwise, retain the position of the original individual X inew ; r is a random number between [0,1]; Initialize the population through Circle chaotic mapping and opposition-based learning strategy to form a new initial population position.

[0009] Furthermore, in the step S03, for the improved subtraction average optimization algorithm MSABO, introduce the Levy flight strategy in the position update stage, including: The subtraction average optimization algorithm SABO model is: ; where, v is a multi-dimensional vector randomly generated from the set {1,2}; the operation "*" represents the Hadamard product of two vectors; F(A) and F(B) are the fitness values of the objective functions of the search individuals A and B respectively, and sign is the signum function; Perform position update according to the following formula: ; where, is the updated individual position; is the position of the i-th individual at the t-th iteration; r i is a multi-dimensional pseudo-random number vector subject to normal distribution; is the position of the j-th individual at the t-th iteration; N is the number of individuals; Introduce the Levy flight strategy in the position update stage, and the formula is as follows: ; ; Among them, L(S) is the Levy flight function; λ is a number within (1, 3); is the Gamma function; s is the flight step length; According to the fitness value, the optimal position is retained, and the optimal parameters are output according to the following formula: ; In the formula, X i is the position of the i-th individual; The adaptive fitness value of the i-th individual at the t-th iteration; is the adaptive fitness value of the i-th individual at the (t + 1)-th iteration.

[0010] Furthermore, in the step S03, the improved subtraction average optimization algorithm MSABO is used to optimize the key parameters of the gated recurrent unit network GRU, including using the improved subtraction average optimization algorithm MSABO to optimize the learning rate, weight, bias, number of hidden layers, and regularization parameters of the gated recurrent unit network GRU.

[0011] The beneficial effects of the present invention are as follows: The present invention processes the power supply signal of the fire control computer power supply module collected by using the multi-scale geometric analysis algorithm MGA, separates different features of the signal, extracts valuable geometric information and suppresses noise interference, screens out key information signals, realizes the separation of signal details and overall trends, and is beneficial to improving the prediction efficiency of the model. By introducing the Circle chaotic mapping, reverse learning strategy, and Levy flight strategy into the subtraction average optimization algorithm SABO, problems such as the contingency of the initial population generation and the easy entrapment in local optimal solutions are solved. The improved subtraction average optimization algorithm MSABO is used to optimize the key parameters of the gated recurrent unit network GRU, and a health prediction model MSABO-GRU is constructed to improve the prediction accuracy of the model. Specific Embodiments

[0012] In order to better explain the present invention for easy understanding, the following specific embodiments will describe the present invention in detail.

[0013] The present invention provides a method for predicting the health of a fire control computer power supply module, including the following steps: Step S01: Collect the power supply signal of the fire control computer power supply module as the original data.

[0014] Step S02: Use the multi-scale geometric analysis algorithm MGA to separate different features of the collected original data, extract valuable geometric information and suppress noise interference, screen out key information signals to construct a data set, and divide it into a training set and a test set.

[0015] Specifically, the data collected in the present invention is processed by the multi-scale geometric analysis algorithm (MGA algorithm), decomposing the complex signal into different scale spaces, extracting the geometric features of the signal at different scales, and realizing noise reduction, feature extraction or compression through a reasonable reconstruction method.

[0016] The steps of separating different features of the signal by using the multi-scale geometric analysis algorithm (MGA) include the following: Step S201, signal decomposition: Based on the multi-scale framework, the original signal f(t) is decomposed into components at different scales according to the following formula: ; Where: f(t) represents the original signal; and are the low-frequency and high-frequency parts of the multi-scale basis function respectively; and are the low-frequency component and high-frequency component respectively; j is the scale parameter; k is the position parameter; J represents the maximum scale layer of decomposition; Step S202, geometric feature extraction: For the decomposed high-frequency component Extract the geometric characteristics of the signal, including local singularity detection and multi-directional feature extraction.

[0017] Local singularity detection is to analyze the modulus value of the high-frequency component and detect the signal mutation points.

[0018] Multi-directional feature extraction is to use directional basis functions, such as the Curvelet function, to extract signal features in different directions; The formula of the Curvelet function is as follows: ; Where: and are the components of the high-frequency component in the x and y directions respectively.

[0019] Step S203, signal reconstruction: Select important components according to the following formula, eliminate noise interference, and realize noise reduction or feature enhancement of the signal.

[0020] ; Where: is the reconstructed signal; S is the set of retained components.

[0021] Step S204, extract the state features of the fire control system.

[0022] Step S03, optimize the key parameters of the gated recurrent unit network (GRU) by using the improved mean subtraction and averaging optimization algorithm (MSABO), and construct a health prediction model MSABO-GRU.

[0023] Specifically, the improved subtraction average optimization algorithm MSABO is used to optimize parameters such as the learning rate, the number of hidden layers, and the regularization parameter of the gated recurrent unit network, so as to improve the accuracy of model prediction.

[0024] The improved subtraction average optimization algorithm MSABO includes introducing the Circle chaotic map and the reverse learning strategy in the initialization stage, and introducing the Levy flight strategy in the position update stage. The specific method is as follows: Initialize the population position according to the following formula: ; where, X i is the position of the i-th individual; ub and lb are the lower and upper bounds of the space search respectively; rand(0,1) is a random number uniformly distributed between [0,1]; Introduce the Circle chaotic map in the initialization stage to avoid uneven distribution of randomly generated individuals in the population, and further generate new individual positions. The formula is as follows: ; where, X inew is the new individual position after the Circle chaotic map; mod represents the modulo operation; Subsequently, to improve the diversity of the population position, introduce the reverse learning strategy, and calculate the reverse position Y of the individual according to the following formula i : ; When the fitness value of the reverse learning individual is less than the fitness value of the original individual, retain the position of the reverse learning individual Y i ; otherwise, retain the position of the original individual X inew ; r is a random number between [0,1]. After the Circle chaotic map and the reverse learning strategy are used to initialize the population, a new population initial position is formed.

[0025] The subtraction average optimization algorithm SABO model is based on a special operation "-v", which is called the v-subtraction of search individual B in search individual A. Specifically: ; where, v is a multi-dimensional vector randomly generated from the set {1,2}; the operation "*" represents the Hadamard product of two vectors; F(A) and F(B) are the fitness values of the objective functions of search individuals A and B respectively, and sign is the signum function.

[0026] In the subtraction average optimization algorithm SABO, the displacement of any search individual X i in the search space is through each individual Xj Calculated by the arithmetic mean of the "-v" subtraction for (j = 1, 2...), the position update method is as follows: ; Among them, is the updated individual position; is the position of the i-th individual at the t-th iteration; r i is a multi-dimensional pseudo-random number vector subject to a normal distribution; is the position of the j-th individual at the t-th iteration; N is the number of individuals.

[0027] Since the Subtraction Averaging Optimization Algorithm (SABO) does not utilize the global optimum value during each iteration, but instead uses the positions of all individuals to calculate the subtraction mean to achieve the update purpose, it is extremely prone to falling into local optima. Therefore, the Levy flight strategy is introduced to effectively expand the search space, enabling the algorithm to make occasional long-distance jumps and more short-distance local searches during the iteration process, which can enhance the global search ability of the algorithm in the solution space. The improved formula is: ; ; Among them, L(S) is the Levy flight function; λ is a number within (1, 3); is the Gamma function; s is the flight step size.

[0028] According to the fitness value, the optimal position is retained, and the optimal parameters are output according to the following formula: ; In the formula, X i is the position of the i-th individual; is the adaptive fitness value of the i-th individual at the t-th iteration; is the adaptive fitness value of the i-th individual at the (t + 1)-th iteration. Among them is the updated individual position after introducing the Levy flight strategy.

[0029] Judge the fitness values of the positions before and after the update, and retain the position with the larger fitness value.

[0030] Due to its special network structure, the Gated Recurrent Unit network (GRU) simplifies the calculation and can more easily capture complex time series signals, demonstrating significant advantages in health prediction tasks. The power signal of the fire control computer is essentially a complex time series data, containing time dependence and dynamic characteristics. As an improved type of recurrent neural network, GRU can capture the time correlation in the signal and model the law of signal evolution over time. Compared with traditional machine learning methods, such as support vector machines and decision trees, GRU can utilize gating mechanisms, such as update gates and reset gates, to dynamically select important time-dependent information; it has higher computational efficiency and powerful non-linear modeling capabilities.

[0031] The calculation process of the Gated Recurrent Unit network (GRU) can be divided into the following steps: 1. Calculation of the reset gate: The Gated Recurrent Unit network (GRU) calculates the value of the reset gate based on the current input and the hidden state at the previous moment. The role of the reset gate in the model is to determine which information in the hidden state at the previous moment needs to be discarded or forgotten. The formula for the reset gate r t is: ; where () is the activation function; W r is the weight matrix of the reset gate; h t-1 is the hidden state at the previous moment; x t is the input at the current moment; b r is the bias term of the reset gate.

[0032] 2. Calculation of the update gate: The Gated Recurrent Unit network (GRU) calculates the value of the update gate. The update gate determines what proportion of the hidden state at the current moment comes from the state at the previous moment and what proportion comes from the current input. The formula for the update gate z t is: ; where is the weight matrix of the update gate, is the bias term of the update gate.

[0033] 3. Calculation of the candidate hidden state: After calculating the reset gate and the update gate, the Gated Recurrent Unit network (GRU) uses the reset gate to calculate a candidate hidden state, representing the network state at the current moment. The formula for the candidate hidden state h t is: ; where tanh is the hyperbolic tangent function, W h is the weight matrix of the candidate hidden state, b h is the bias term of the candidate hidden state.

[0034] 4. Calculation of the final hidden state: Finally, the GRU weights the candidate hidden state at the current moment and the hidden state at the previous moment through the value of the update gate to obtain the final hidden state h t , and the formula is: .

[0035] Step S04: Use the training set in step S02 to train the health prediction model MSABO-GRU in step S03; Step S05: Use the test set in step S02 to test the trained health prediction model MSABO-GRU in step S04; Step S06: Use the qualified health prediction model MSABO-GRU to perform health prediction on the power supply module of the fire control computer and output the prediction result.

[0036] The present invention uses the multi-scale geometric analysis algorithm MGA to preprocess the collected power signal data, extracts valuable geometric information, and constructs a data set, which is beneficial to improving the prediction efficiency of the model. Improve the subtraction average optimization algorithm SABO to solve problems such as the contingency of the initial population generation and the easy convergence to local optimal solutions. Use the improved subtraction average optimization algorithm MSABO to optimize parameters such as the learning rate, the number of hidden layers, and the regularization parameter of the gated recurrent unit network GRU, and improve the prediction accuracy of the model.

[0037] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Modifications, alterations, substitutions, and variations made by those of ordinary skill in the art to the above embodiments all fall within the scope of the present invention.

Claims

1. A method for predicting the health of a power supply module of a fire control computer, characterized in that, It includes the following steps: Step S01: Collect the power signal of the fire control computer power module as the original data; Step S02: Use the multi-scale geometric analysis algorithm MGA to separate different features of the signal for the collected original data, extract valuable geometric information and suppress noise interference, screen out key information signals to construct a data set, and divide it into a training set and a test set; Step S03: Use the improved subtraction average optimization algorithm MSABO to optimize the key parameters of the gated recurrent unit network GRU, and construct a health prediction model MSABO-GRU. The improved subtraction average optimization algorithm MSABO includes introducing a Circle chaotic map and a reverse learning strategy in the initialization stage, and introducing a Levy flight strategy in the position update stage; Step S04: Use the training set in Step S02 to train the health prediction model MSABO-GRU in Step S03; Step S05: Use the test set in Step S02 to test the health prediction model MSABO-GRU trained in Step S04; Step S06: Use the health prediction model MSABO-GRU with qualified test results to perform health prediction on the fire control computer power module and output the prediction result.

2. A method for predicting the health of a power supply module of a fire control computer according to claim 1, characterized in that, In Step S02, using the multi-scale geometric analysis algorithm MGA to separate different features of the signal includes the following steps: Step S201: Signal decomposition: Decompose the original signal f(t) into components at different scales according to the following formula: ； where: f(t) represents the original signal; and are the low-frequency and high-frequency parts of the multi-scale basis function, respectively; and are the low-frequency component and high-frequency component, respectively; j is the scale parameter; k is the position parameter; J represents the maximum scale level of decomposition; Step S202: Geometric feature extraction: For the decomposed high-frequency components Extract the geometric features of the signal, including local singularity detection and multi-directional feature extraction; local singularity detection is to analyze the modulus of the high-frequency components to detect signal mutation points; multi-directional feature extraction is to extract signal features in different directions using directional basis functions; Step S203: Signal reconstruction: Select important components according to the following formula to eliminate noise interference, ； Wherein: is the reconstructed signal; S is the set of retained components; Step S204: Extract the state features of the fire control system.

3. A method for predicting the health of a power supply module of a fire control computer according to claim 2, characterized in that: In Step S202, the directional basis function is the Curvelet function: ； Wherein: and are the components of the high-frequency component in the x and y directions, respectively.

4. A method for predicting the health of a power supply module of a fire control computer according to claim 1, characterized in that, In Step S03, the improved subtraction average optimization algorithm MSABO includes introducing a Circle chaotic map and a reverse learning strategy in the initialization stage, including: Initialize the population position according to the following formula: ； where X i is the position of the i-th individual; ub and lb are the lower and upper bounds of the spatial search, respectively; rand(0,1) is a random number uniformly distributed between [0,1]; The Circle chaotic map formula introduced in the initialization stage is as follows: ； Among them, X inew is the position of the new individual after the Circle chaotic mapping; mod represents the modulo operation; Introduce a reverse learning strategy and calculate the reverse position Y of an individual according to the following formula i :[[]]END]] ； When the fitness value of the reverse learning individual is less than that of the original individual, retain the position of the reverse learning individual Y i ; otherwise, retain the position X of the original individual inew ; r is a random number between [0, 1]; initialize the population through the Circle chaotic mapping and the reverse learning strategy to form the initial position of the new population.

5. A method for predicting the health of a power supply module of a fire control computer according to claim 1, characterized in that: In Step S03, for the improved subtraction average optimization algorithm MSABO, a Levy flight strategy is introduced in the position update stage, including: The subtraction average optimization algorithm SABO model is: ； where v is a multi-dimensional vector randomly generated from the set {1, 2}; the operation "*" represents the Hadamard product of two vectors; F(A) and F(B) are the fitness values of the objective functions of search individuals A and B respectively, and sign is the signum function; Perform position update according to the following formula: ； Among them, is the updated individual position; is the position of the i-th individual at the t-th iteration; r i is a multi-dimensional pseudo-random number vector subject to a normal distribution; is the position of the j-th individual at the t-th iteration; N is the number of individuals; Introduce a Levy flight strategy in the position update stage, and the formula is as follows: ； ； where \(L(S)\) is the Levy flight function; \(\lambda\) is a number within \((1, 3)\); \(\varGamma\) is the Gamma function; \(s\) is the flight step length; According to the fitness value, retain the optimal position and output the optimal parameters according to the following formula: ； Where X i is the position of the i-th individual; is the fitness value of the i-th individual at the t-th iteration; is the fitness value of the i-th individual at the (t + 1)-th iteration.

6. A method for predicting the health of a power supply module of a fire control computer according to claim 1, characterized in that: In Step S03, using the improved subtraction average optimization algorithm MSABO to optimize the key parameters of the gated recurrent unit network GRU includes using the improved subtraction average optimization algorithm MSABO to optimize the learning rate, weight, bias, number of hidden layers, and regularization parameters of the gated recurrent unit network GRU.

Images