Lightweight method for DBN model of multi-electric aircraft starting generation system based on GBIC

Abstract

The embodiment of the application discloses a DBN model lightweight method of a multiple-electric aircraft starting power generation system based on GBIC, relates to the technical field of reliability design and modeling of complex equipment systems, and comprises the following steps: a DBN model of a starting power generation system is established, nodes in the DBN model correspond to key components and working states of the starting power generation system, directed edges in the DBN model are used for describing mutual relations between nodes, redundant structures in the DBN model are identified and lightweight processing is performed, and a fault state of the starting power generation system is identified by using the DBN model subjected to the lightweight processing.The embodiment of the application introduces a grey system theory and a Bayesian information criterion aiming at the problems of complex and low efficiency of reasoning calculation in the process of a dynamic Bayesian analysis method of a complex polymorphic system, realizes lightweight of the dynamic Bayesian network of the complex polymorphic system, eliminates redundant structures, and guarantees the accuracy of results.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of reliability design and modeling of starting power generation systems, and relates to converting data related to a starting power generation system of a more electric aircraft into a form that can be processed by a computer and subsequent processing, and in particular to a lightweight method for a DBN model of a starting power generation system of a more electric aircraft based on Grey-Bayesian Information Criterion (GBIC) required in a computer processing process. BACKGROUND

[0002] With the continuous rapid development of the electrification and intelligentization of starting power generation systems, the equipment system presents the characteristics of complexity, multiple failure states and dynamic changes. On the one hand, the equipment system is subjected to diversified risk factors and failure modes in a harsh working environment, and presents diversified failure states, which cannot be described by a simple "working / failure". On the other hand, due to the complex connection between components and the redundant design of the functional structure of the equipment system, the reliability analysis method of the traditional static or simple model is difficult to meet the requirements. The complex polymorphic system with these two characteristics not only greatly increases the difficulty and cost of equipment support, but also may shorten the life cycle of the equipment due to the inaccurate reliability analysis method. Therefore, the reliability analysis method for complex polymorphic systems has become a hot issue in the field of reliability.

[0003] In recent years, in view of the difficulty of reliability analysis of complex polymorphic systems, various methods have been widely studied and applied, among which the Bayesian Network (BN) method stands out due to its unique advantages. The BN method can not only intuitively express the complex probability dependency relationship between systems (directed acyclic graph structure) and effectively handle the problem of multiple state changes, but also can fuse static and dynamic information, form a dynamic Bayesian Network (DBN) by introducing a time slice, and realize the time sequence evolution process of the modeling system state. What is particularly key is that the BN method has a unique bidirectional reasoning mechanism, which can not only perform forward prediction reasoning, but also perform reverse diagnosis reasoning, which is very suitable for the reliability analysis of complex polymorphic systems. However, with the continuous improvement of the complexity and redundancy of the starting power generation system, the corresponding model size expands dramatically, and the dimension of the required Conditional Probability Table (CPT) also increases significantly. These all lead to the complexity and inefficiency of the reasoning calculation of the traditional BN / DBN method, which not only occupies more and more computing resources, but also makes it difficult to improve the accuracy of the analysis results.

[0004] Therefore, how to reduce the complexity of the analysis model in the reliability analysis of the starting power generation system to save computing resources while maintaining the accuracy of the calculation has become a research topic. SUMMARY

[0005] The embodiment of the application provides a lightweight method for a DBN model of a GBIC-based more electric aircraft starting power generation system, which can reduce the complexity of an analysis model in the reliability analysis of the starting power generation system to save computing resources while maintaining the accuracy of the calculation.

[0006] To achieve the above object, the embodiment of the application adopts the following technical scheme:

[0007] A lightweight method for a DBN model of a GBIC-based more electric aircraft starting power generation system, as shown in the accompanying drawings, comprises the following steps: Figure 10

[0008] S1, establishing a DBN model of the starting power generation system.

[0009] In the DBN model, nodes correspond to key components and working states of the starting power generation system, and directed edges in the DBN model are used to describe the mutual relationship between nodes.

[0010] In actual application, the starting power generation system can have multiple types, and in the embodiment, the starting power generation system is taken as an example, and the key components include a power supply, a signal feedback sensor, a starting rule calculation sensor, a starting control signal generating component, a computer control system, a starting controller, an exciter, a starting generator, an engine rotor, a rotating rectifier, an engine, a voltage regulating circuit, a rectifier bridge and the starting power generation system.

[0011] Specifically, the working states of the starting power generation system include a starting mode and a power generation mode. A normal working state (Normal) is a non-fault state, a fault state is divided into a mistake (Mistake), a partial failure (Ploss) and a complete failure (Tloss), and the accuracy and the calculation efficiency of the starting power generation system in the fault state before and after the lightweight DBN model are verified in the complete failure state.

[0012] S2, identifying redundant structures in the DBN model and performing lightweight processing.

[0013] S3, identifying the fault state of the starting power generation system by using the DBN model subjected to the lightweight processing.

[0014] In the embodiment, S1 comprises reading a maintenance manual and a fault isolation manual of the starting power generation system and determining the CPT of the nodes in the DBN model containing interval grey numbers.

[0015] ​The DBN model of the starting power generation system comprises: , denotes the state transition probability of node x from time slice t to t+1, x t denotes the node state at t, x t+1 denotes the node state at t+1, t denotes a time slice, i denotes a node corresponding number, x i t denotes the node state of the i th node at time slice t, x i t+1 denotes the node state of the i th node at t+1, N denotes the total number of nodes.

[0016] In the embodiment, S2 comprises: obtaining the fitting degree of the DBN model to data and the parameter dimension of the node in the DBN model; and calculating the GBIC value of the parent node combination in the DBN model, wherein the size of the GBIC value is positively correlated with the structural redundancy of the DBN model.

[0017] The fitting degree of the DBN model to data comprises:

[0018] ,

[0019] ;

[0020] wherein, and respectively denote the upper limit and the lower limit of the gray fuzzy likelihood value in time slice t j , j denotes a time slice corresponding number, and denote two initial nodes in time slice t j , and denote the parent node set corresponding to the two initial nodes at time slice t j , and respectively denote the gray fuzzy conditional probability value corresponding to the two initial nodes in time slice t j , p denotes the total number of root nodes x, and q denotes the total number of intermediate nodes y.

[0021] The parameter dimension of the node in the DBN model comprises: wherein, denotes the parameter dimension of the node, and respectively denote the number of parameters of nodes x i and y i , S denotes the failure mode of the node, p denotes the total number of root nodes x, and q denotes the total number of intermediate nodes y.

[0022] The GBIC value of the parent node combination in the DBN model is calculated, including calculating the upper and lower extreme values of the parent node combination in the DBN model to obtain a GBIC value range, wherein: , ;

[0023] and respectively represent the upper and lower extreme values of the node and at the task time T, and respectively represent the traversal parent node combination corresponding to the GBIC extreme value of the node and in the time slice t j of the task time T, and respectively represent the parent node combination of the node and at the time slice t j . and respectively represent the upper and lower bounds of the maximum likelihood value and , and m represents the total number of time slices t j in the task time T.

[0024] , ,

[0025] , , wherein, represents the conditional probability value of the node containing interval grey numbers in the time slice t j , represents the probability lower limit corresponding to , represents the probability upper limit corresponding to , r represents the number of conditional probability values containing interval grey numbers, and n represents the total number of conditional probability values containing interval grey numbers of the node in the time slice t j .

[0026] The GBIC-based multi-electric aircraft starting power generation system DBN model lightweight method provided by the embodiment of the application is used in the process of the dynamic Bayesian analysis method of the complex multi-state system, and the model scale corresponding to the complex multi-state system is sharply expanded, the dimension of the conditional probability table is greatly increased, the reasoning calculation is complex and inefficient, the grey system theory and the Bayesian information criterion are introduced, the lightweight of the dynamic Bayesian network of the complex multi-state system is realized, and the redundant structure is removed. Moreover, the accuracy of the fault state calculation of the lightweight DBN and the calculation time are improved. BRIEF DESCRIPTION OF DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only constitute some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0028] Figure 1 The comparative schematic diagram of the model structure provided by the embodiments of the present application is shown in the following figure, wherein, Figure 1 the part (a) in the figure is a schematic diagram of BN structure, Figure 1 the part (b) in the figure is a schematic diagram of DBN structure;

[0029] Figure 2 The logic architecture schematic diagram of the dynamic Bayesian network lightweight scheme based on GBIC provided by the embodiments of the present application is shown in the following figure;

[0030] Figure 3 The operation mechanism and the working relationship between key components of the starting generator (SG) system provided by the embodiments of the present application are shown in the following figure, wherein, Figure 3 the part (a) in the figure is the starting mode of the SG system, Figure 3 the part (b) in the figure is the power generation mode of the SG system; Figure 3 the English abbreviations in the figure respectively represent: BR represents a rectifier bridge, CEC represents a generator excitation contactor, EXG represents an exciter, RR represents a rotary rectifier, SG represents a starting generator, SC represents a starting contactor, CMSC represents a general motor starting controller, GCB represents a generator control circuit breaker, ATRU represents an autotransformer rectifier unit, APB represents an auxiliary power circuit breaker, EPC represents an external power contactor, APU represents an auxiliary power unit, GEN represents a generator, and ATU represents an autotransformer unit;

[0031] Figure 4 The working mechanism and the relationship between key components of the SG system provided by the embodiments of the present application are shown in the following figure; Figure 4 the English abbreviations in the figure respectively represent: SFS represents a signal feedback sensor, SLS represents a starting law calculation sensor, SCSG represents a starting control signal generation component, PWR represents a power supply, DCU represents a computer control system, SCU represents a starting controller, APU represents an auxiliary power unit, EXG represents an exciter, SG represents a starting generator, BR represents a rectifier bridge, RR represents a rotary rectifier, VR represents a voltage regulating circuit, ENG represents an engine, and OE represents an on-board equipment;

[0032] Figure 5The SG system DBN model schematic diagram considering common cause failure provided by the embodiment of the present application, wherein, Figure 5 The (a) part in the figure is the DBN model structure applied in the embodiment scenario, Figure 5 The (b) part in the figure is the dynamic process of the DBN model; Figure 5 The English abbreviations in the figure respectively represent: SFS represents a signal feedback sensor, SLS represents a start law calculation sensor, SCSG represents a start control signal generation component, PWR represents a power supply, DCU represents a computer control system, SCU represents a start controller, EXG represents an excitation machine, SG represents a start generator, ER represents an engine rotor, RR represents a rotating rectifier, BR represents a rectifier bridge, VR represents a voltage regulation circuit, ENG represents an engine, SM represents a start mode, GM represents a power generation mode, and SGS represents a start power generation system;

[0033] Figure 6 The node y 17 in the task time T provided by the embodiment of the present application,

[0034] Figure 7 The node y j in each time slice t 17 provided by the embodiment of the present application,

[0035] Figure 8 The SG system LDBN structure schematic diagram provided by the embodiment of the present application; Figure 8 The meaning of the number under the node symbol in the figure is: the node x 1 , x 2 The normal operation (Normal) and the complete failure (Tloss) of two modes are respectively represented by fuzzy numbers 0 and 1; the node x 3 , x 4 , y 1, y 2, the normal operation (Normal), the partial failure (Ploss) and the complete failure (Tloss) of three modes are respectively represented by fuzzy numbers 0, 0.5 and 1; the node x 5, y i = 3, 4, …, 18 , Y The normal operation (Normal), the abnormal operation (Mistake), the partial failure (Ploss) and the complete failure (Tloss) of four modes are respectively represented by fuzzy numbers 0, 0.33, 0.66 and 1; the current possible operation mode of the node can be represented by the number under the node symbol;

[0036] Figure 9 The result verification schematic diagram in a complete failure state based on the light-weight method provided by the embodiment of the application is shown in FIG. 6.

[0037] Figure 10 The flowchart provided by the embodiment of the application is shown in FIG. 7. DETAILED DESCRIPTION

[0038] To enable those skilled in the art to better understand the technical solutions of the present application, the present application will be further described in detail below in conjunction with the drawings and specific embodiments. In the following, the embodiments of the present application will be described in detail, and the examples of the embodiments are shown in the drawings, in which the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are only used to explain the present application, and cannot be interpreted as a limitation on the present application. Those skilled in the art can understand that, unless specifically stated, the singular forms "a", "an" and "the" used herein also include the plural forms. It should be further understood that the phrase "comprising" used in the specification of the present application means that the features, integers, steps, operations, elements and / or components exist, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we say that an element is "connected" or "coupled" to another element, it can be directly connected or coupled to the other element, or there can be intermediate elements. In addition, "connected" or "coupled" used herein can include wireless connection or coupling. The phrase "and / or" used herein includes any one of the associated listed items and all combinations thereof. Those skilled in the art can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as that generally understood by those skilled in the art in the field to which the present application belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have meanings consistent with those in the context of the prior art, and should not be interpreted with idealized or overly formal meanings unless defined as such.

[0039] In this embodiment, a light-weight method based on GBIC for a multi-electric aircraft starting and generating system DBN model is needed to be designed. The main design idea is to introduce the grey system theory, fuse it with the Bayesian information criterion, construct the evaluation basis of model fitting degree and structure complexity, so as to remove the redundant structure of the complex polymorphic system, simplify the connection relationship, and obtain a light-weight DBN (LDBN) model with significantly reduced calculation complexity.

[0040] It should be noted that the dynamic Bayesian network mentioned in the embodiment refers to a probabilistic graphical model for representing the relationship between variables, and is used for prediction, diagnosis, decision making and other tasks through probabilistic reasoning. It is based on the extension of Bayes' theorem, composed of nodes and directed edges, and the nodes represent variables and the edges represent the dependency relationship between variables, which is one of the very effective theoretical models in the field of uncertain knowledge representation and reasoning. The value of a node depends on the value of its parent node and its conditional probability. It has the ability to describe the polymorphism and non-deterministic logic relationship of events. In the reasoning process of Bayesian network, posterior probability, prediction, diagnosis and decision making can be used for analysis. Among them, the posterior probability refers to solving the probability of other unknown events given some observed events; prediction refers to predicting future events given some observed events; diagnosis refers to inferring the cause given some abnormal events; decision making refers to selecting the optimal decision based on some known actions and results. Bayesian network (BN) is a directed acyclic graph composed of nodes and directed edges. The nodes represent random variables, and the directed edges represent the conditional dependency relationship or causal relationship between these variables. Bayesian network analysis calculates the posterior probability of variables based on prior knowledge and observed evidence, thereby realizing probabilistic reasoning.

[0041] In Bayesian network analysis, a directed acyclic graph (DAG) is used to describe the conditional dependency relationship between variables. The basic BN model structure is shown in part (a) of Figure 1 The nodes x1, x2,...x i and y in the figure are random variables, and x1, x2,...x i are the parent nodes of y. The joint distribution of the network is obtained by multiplying the probability distributions attached to each node variable in the network.

[0042] DBN is an extension of BN in temporal logic, composed of an initial BN and a transition network. The static BN model is expanded to a probability distribution model capable of handling temporal information by adding a time dimension. DBN can be constructed by expanding multiple time slices, where each slice corresponds to a static BN model. The structural dependency relationship between adjacent time slices is preserved to ensure temporal consistency, as shown in part (b) of Figure 1 .

[0043] In practical applications, the main construction methods of dynamic Bayesian networks can be divided into two categories: based on domain knowledge and based on data-driven. Among them, the construction method based on domain knowledge, such as: expert knowledge method: experts determine the structure and parameters of Bayesian network through subjective judgment according to their domain knowledge. For example: model structure learning method: according to the characteristics of the domain model, automatically learn the structure of Bayesian network from the perspective of model structure. The construction method based on data-driven, such as: Bayesian learning method, constraint satisfaction method, minimum description length method and Gaussian graph model selection method.

[0044] The establishment method of Bayesian network can be divided into the following steps: 1) Determine the variable, first need to determine the variable involved in modeling, can be continuous variable or discrete variable. 2) Build network structure, in Bayesian network, node represents random variable, edge represents the conditional dependence relationship between two variables. Therefore, the conditional dependence relationship between variables needs to be determined, and the network structure needs to be constructed. 3) Determine the probability distribution of variable, after determining the network structure, the probability distribution of each variable needs to be determined, which can be discrete distribution or continuous distribution. The method of determining the probability distribution can be subjective assignment method, maximum likelihood method, Bayesian method, etc. 4) Model evaluation, after establishing the model, it needs to be evaluated to determine its prediction performance. Common methods include: cross-validation, information criterion, error evaluation, etc. 5) Model application, after establishing Bayesian network, it can be used to infer the conditional dependence relationship between variables, and can be used for probability inference, decision analysis, data mining, etc.

[0045] Bayesian information criterion (BIC) is a model selection criterion derived based on Bayesian theory. It uses Bayesian formula to correct subjective probability estimation, and realizes the comprehensive consideration of model fitting degree and complexity by evaluating the likelihood rate and parameter number of the model at the same time, and then determines the most appropriate model under the specific data set.

[0046] Grey system theory refers to a system composed of interrelated and constrained elements. It is divided into three types according to its known degree: black system, white system and grey system. The black system refers to a complex unknown system whose internal characteristics are completely unknown and no internal structure information can be obtained. The white system refers to a known system whose internal characteristics and action principles are completely clear and an accurate mathematical model can be established. The grey system is between the black system and the white system, and its internal characteristics are partially known and partially unknown. In theoretical analysis, the known information can be mined and analyzed to describe and understand the grey system, and as much internal information as possible can be mastered to establish a quantitative analysis model. However, in the field of practical engineering application, it is often very difficult to establish an absolute white system. Although some factors affecting the system can be qualitatively or quantitatively analyzed by mathematical statistics method, it is difficult to exhaust and clearly all the factors related to the system, and it is also difficult to determine the specific mapping relationship between these factors. In engineering practice, such a system without a determined mapping function relationship can be regarded as a grey system.

[0047] The embodiment is based on the above scheme and algorithm, and a maintenance detection scheme applied to aviation equipment is designed and improved again, which is based on the lightweight processing of dynamic Bayesian network based on Grey-Bayesian Information Criterion (GBIC). Specifically, in order to reduce the redundancy of DBN in a complex multi-state system, the grey system theory is introduced, which is combined with the Bayesian information criterion to construct the evaluation basis of model fitting degree and structure complexity, and the redundancy degree between parent node combinations is compared to realize the transformation of DBN model into lightweight DBN model (Lightweighting DBN, LDBN), so as to improve the calculation efficiency.

[0048] The specific method is as follows:

[0049] 1. Grey maximum likelihood value calculation: calculate the grey maximum likelihood value to evaluate the fitting degree of the model to the data. Wherein, N is the total number of nodes in DBN; represents the root node, and p is the number of root nodes; represents the intermediate node, q is the number of intermediate nodes, and Y is the leaf node. The task time T is divided into m time segments, and the duration of each segment is t j , wherein the subscript i in this paper is used to represent the node number, and j represents the time slice number, such as the i-th node x i , wherein i represents the node number; t jrepresents the jth time slice. Then, based on the assumption of conditional independence, the joint probability is decomposed under the condition that the DBN nodes contain interval grey number conditional probability data have conditional independence. At the same time, the interval grey number probability is determined by calculating the upper and lower bounds of the data, respectively, as follows:

[0050] ,

[0051] ;

[0052] wherein, and represent the upper and lower bounds of the grey fuzzy likelihood value in the time slice t j , j represents the number corresponding to the time slice, and represent two initial nodes in the time slice t j , and represent the parent node set corresponding to the two initial nodes in the time slice t j , and represent the grey fuzzy conditional probability value corresponding to the two initial nodes in the time slice t j , p represents the total number of root nodes x, and q represents the total number of intermediate nodes y.

[0053] 2. DBN parameter dimension calculation: The parameter dimension k needs to be calculated from two levels of the parameter dimension of the node and the total parameter dimension of the network. The parameter dimension of the node is the number of parameters of a single node in the network, and the parameter R is usually determined by the number of its parent nodes and its own state number. The specific calculation method is: , wherein, and represent the number of parameters of nodes x i and y i , represents the failure mode of the current node.

[0054] The total parameter dimension of the network is the sum of all the parameters of the nodes in the entire Bayesian network, which is used to measure the complexity of the model as a whole. Thus, the parameter dimension of the node in the DBN model can be obtained:

[0055]

[0056] wherein, represents the parameter dimension of the node, which is used to evaluate the complexity of the entire model in the GBIC calculation, and is used as a penalty term to affect the final model selection, so as to prevent the model from being too complex while maintaining good fitting degree.

[0057] 3. Grey Bayesian Information Criterion (GBIC) calculation: The GBIC value is calculated by regarding the conditional probability as an interval grey number, whose upper and lower bounds represent two extreme cases. If the conditional probability is an interval grey number, both the upper and lower bounds are used in the calculation. Otherwise, it is assumed that the upper and lower bounds have equal probability in the calculation. Node or The GBIC value is calculated as follows:

[0058]

[0059] where and denote the GBIC values of the parent node combination of node and at time slice t j , respectively, and denote the upper and lower bounds of the maximum likelihood value and , respectively.

[0060] According to the and of node and , the upper and lower maximum values of each parent node combination are taken as the GBIC maximum values and of node and , where: , .

[0061] On this basis, considering the inherent uncertainty and interval characteristics of interval grey numbers, a method that can consider this uncertainty must be used to constrain the GBIC maximum values. For this purpose, a constraint condition is introduced. Through this equation, the value range of the GBIC maximum values and can be reasonably defined, ensuring that they remain reasonable and effective while fully considering the uncertainty of interval grey numbers. The constraint condition is: , .

[0062] where is the conditional probability value of the interval grey number corresponding to node and in time slice t j ; is the lower probability limit of the corresponding interval grey number , is the upper probability limit of the corresponding interval grey number , and are time slice t j middle node and The parent node traverses the corresponding GBIC maximum value.

[0063] node and GBIC value at task time T and Will be obtained, expressed as: , .

[0064] In this embodiment, the node conditional probability represented by interval grey number is determined. This method uses grey maximum likelihood value and parameter dimension to evaluate the fitting degree and complexity of the model. By integrating GBIC and parameter dimension, this method can comprehensively identify the optimal DBN model within the task time T, eliminate redundant node relationships, and thus realize model lightweight.

[0065] Further, the logic architecture of the DBN model lightweight of the multi-electric aircraft starting power generation system based on GBIC is shown in Figure 2 , mainly including:

[0066] Step 1: Constructing a DBN model of a complex polymorphic system

[0067] Firstly, based on the working mechanism and structural composition of the complex polymorphic system, the faults of the key components and their fault characteristics are determined, and the relationship between the key components and the system fault modes is determined. Then, based on the working mechanism and fault mode of the complex polymorphic system, the network structure of the BN model is constructed, which accurately reflects the functional dependence relationship and fault propagation within the system. Finally, the BN model is converted into a DBN model by time slice and task time, and the DBN model of the complex polymorphic system is established.

[0068] Step 2: Constructing a Grey Conditional Probability Table (GCPT)

[0069] According to the fault mode and working mechanism of the complex polymorphic system, interval grey number of grey system theory is introduced as the basis of the DBN model. By combining it with the maintenance manual and fault isolation manual of the complex polymorphic system, the mutual relationship between the nodes in the DBN model can be expressed, especially for the fuzzy and uncertain mutual relationship of the complex polymorphic system, which can effectively represent the content structure relationship of the system.

[0070] Step 3: Establishing a LDBN model based on GBIC

[0071] The GBIC-based dynamic Bayesian network lightweight method introduces the grey system theory and combines with the BIC criterion. Firstly, the grey maximum likelihood value is calculated based on the DBN model to measure the fitting degree between the DBN model and the data; secondly, the parameter dimension of the DBN is calculated as a measure of the complexity of the model to prevent overfitting; finally, according to the grey maximum likelihood value and the parameter dimension, the GBIC of each node corresponding to the parent node combination of the DBN model is obtained, and the results of each parent node combination are obtained by combining the conditional restrictions of interval grey numbers, so as to identify the redundant structure in the DBN model and improve the calculation efficiency of the model.

[0072] The effectiveness of the proposed GBIC-LDBN method is verified by specific application examples as follows:

[0073] The multi-electric aircraft starter-generator (SG) system is taken as a complex multi-state system. The starter-generator system is an important part of the power supply system of the multi-electric aircraft, and there are various failure states caused by electrical faults, mechanical faults, environmental influences and improper maintenance. At the same time, the starter-generator system is affected by load changes and environmental conditions, resulting in a dynamic running state of the system working state, and the key components and working mechanism are shown in Figure 3 The running process of the starter-generator system can be divided into two modes: starting mode and generating mode, as shown in Figure 3 part (a) and Figure 3 part (b) of Figure 4 The failure of the multi-electric aircraft starter-generator system is mainly caused by the failure of key components. Based on the mechanism and main components of the multi-electric aircraft starter-generator system, the risk cause elements can be determined as the failure of key components, including the failure of the main generator, the failure of the exciter, the failure of the excitation contactor, the failure of the starting contactor, the failure of the rectifier bridge, the failure of various sensors, the failure of the power control unit, the failure of the rotary rectifier and the failure of the voltage regulating circuit. Therefore, these main key components are classified, i.e. the function execution components responsible for directly controlling the starting / generating work, the information collection components responsible for collecting the working voltage, power and other information of the generator, engine and other components, and the information processing components responsible for summarizing and processing the collected information, as shown in

[0074] Table 1 Classification of starter-generator system failure modes

[0075]

[0076] Through in-depth analysis of the working mechanism and failure modes of key components of the SG system, its DBN model was constructed, such as... Figure 5 The results are shown in parts (a) and (b). The mission time T is divided into three time slices, and the CPT of the DBN nodes containing the interval gray numbers is determined according to the aircraft maintenance manual and the fault isolation manual. The DBN model considers the common cause failure problem among critical components, characterizes the interaction between components and the fault propagation path within the system, and effectively reflects the dynamic characteristics of the system under different operating states. In the DBN model, nodes represent the critical components of the SG system and their operating states, and directed edges describe the relationships between nodes. Therefore, the DBN model of the SG system clearly shows the fault propagation process between components, as well as the impact of changes in the operating state of each component on system performance and the dynamics between related components. The names of the relevant critical components are shown in Table 2.

[0077] Table 2 Key Components and Operating Modes of the SG System

[0078]

[0079] Taking node SCU1 as an example, when node SCU1 fails, the DBN model can quickly identify two other components, EXG1 and SG1, that may be affected by the common cause failure, as well as the extent of the failure's impact on the entire system. This DBN-based modeling method provides strong support for fault diagnosis and maintenance decisions. Furthermore, the DBN model can describe the dynamic changes of the system, thereby improving its accuracy and reliability in assessing the operating status of the SG system.

[0080] The lightweight method for dynamic Bayesian networks based on GBIC aims to reduce the redundancy of DBNs in SG systems. For example... Figure 4 As shown in Table 2, the failure modes of each node in the DBN model are first clarified. The failure modes and working mechanism of the SG system are analyzed, and the power supplies (PWR1 and PWR2) are defined to have two failure modes, denoted as x. i = [0, 1], i = 1, 2. The Signal Feedback Sensor (SFS), Start-up Law Calculation Sensor (SLS), and Computer Control Unit (DCU) have three fault modes, denoted as x. i y1, y2 = [0, 0.5, 1], i = 3, 4. The remaining nodes exhibit four failure modes, represented as x5, y6, y7, y8, y9, y1, y2 ... i Y = [0, 0.33, 0.66, 1], i = 3, 4, …, 18. Taking the CPT of nodes y1 and Y containing interval gray numbers as an example, as shown in Tables 3 and 4. Each row in the tables represents the time slice t. jWhen the parent node is in different fault states, the conditional probability of the child node failing.

[0081] Table 3. CPT of node y1 containing interval gray numbers at time slice t1

[0082]

[0083] Table 4. CPT of node Y containing interval gray numbers at time slice t1

[0084]

[0085] Due to the complex structure of the SG system and the existence of multiple fault states, it is necessary to eliminate redundant structures in the DBN to improve model accuracy. Therefore, for intermediate nodes composed of two or more parent nodes, the GBIC results of traversing and combining their parent nodes are calculated to identify redundant structures in the DBN model. The CPT of a node is the fundamental support for calculating the redundant structures of each intermediate node. By comparing and analyzing the GBIC results within task time T, it can be found that node y 17 There is redundant structure among the corresponding parent node combinations, as shown in Table 5.

[0086] Table 5 Node y 17 GBIC value of parent node combination

[0087]

[0088] like Figure 6 As shown, the GBIC result of the parent node combination of node y17. The overall trend is upward within the task time T. Among them, the GBIC results of combination 1 are... The lowest range [337.50, 370.77] indicates that this combination can more effectively characterize node y. 17 The state. Generally, when the difference in GBIC results exceeds 4, the model with the lower GBIC result has a better association. Therefore, among all parent node combinations, combination 1 (node ​​y) is the best. 13 and y 14 Combination 11 exhibits the best model fit. In contrast, other combinations show higher redundancy, especially combination 11, with a GBIC result as high as [3650.45, 3689.26]. This value is significantly higher than other combinations, making it difficult to accurately represent the relationships between nodes.

[0089] To verify the effect of parent node combination on node y within task time T. 17 The impact was analyzed, and the GBIC changes of each parent node combination within the three time slices were examined, such as... Figure 7The results show that the GBIC trend of each parent node combination in each time slice is relatively consistent, and the overall trend is consistent with Figure 5 the trend of the whole system, which indicates that the influence of each parent node combination on the node y 17 in different task stages has stability. It is worth noting that the combination 1 (i.e., the nodes y 13 and y 14 ) always remains at a very low level within the task time T. This indicates that this parent node combination can effectively represent the state change of the node y 17 while reducing the complexity of the DBN model, generating a lightweight DBN model (LDBN).

[0090] In summary, by using the lightweight method based on GBIC to analyze the SG system model, the GBIC values of the parent node combinations of the node y 17 present consistent trends in different time slices. Among them, the combination of the nodes y 13 and y 14 always has the lowest GBIC value, which indicates that this parent node combination can effectively represent the state change of the node y 17 . In contrast, the GBIC values of other combinations are relatively high, especially the combination 11, which has a significantly higher GBIC value than other combinations, indicating that there is a certain redundant structure in the association relationship between nodes. Therefore, it is reasonable to select the nodes y 13 and y 14 as the parent nodes of the node y 17 , which reduces the complexity of the DBN model and improves the computational efficiency of the model. The optimized LDBN model is shown in Figure 8 .

[0091] Finally, the validation of the GBIC lightweight method: In order to verify the effectiveness of the lightweight method based on GBIC, the most likely failure state Tloss of the SG system was studied, and a multi-dimensional comparative analysis was carried out, as shown in Figure 9 . On the one hand, compared with the DBN model, the LDBN model using the lightweight method based on GBIC has a significantly improved computational efficiency, with a 38.9% reduction in execution time. On the other hand, the accuracy of the system in identifying the most likely failure state remains unchanged. During the entire task time, the accuracy of the LDBN gray fuzzy possibility always remains at 0.97, and the upper and lower limit deviation rates are not more than 0.02 within the accuracy interval. These results confirm that the lightweight method based on GBIC for dynamic Bayesian networks can significantly reduce the complexity of the model while maintaining the accuracy of the calculation.

[0092] In general, the embodiment is directed to the problem of the rapid expansion of the model size and the significant increase of the dimension of the conditional probability table in the process of using the dynamic Bayesian analysis method for complex polymorphic systems, which leads to complex and inefficient reasoning calculation. The gray system theory and the Bayesian information criterion are introduced to realize the lightweight of the dynamic Bayesian network of complex polymorphic systems, eliminate redundant structures, and ensure the accuracy of the results.

[0093] A lightweight computing method for dynamic Bayesian networks based on GBIC is constructed. By organizing the parent node combinations of each node respectively, the degree of fitting is judged from the time slice and the task time respectively, and the most combination of the corresponding node parent node combination is selected, and the redundant structure of the node is eliminated, so as to realize the lightweight of the dynamic Bayesian network. Among them, by combining the gray system theory with the DBN method, the problem of describing the multiple fault states of the system caused by the many interacting components, complex mutual relationships and diversified operating environment of the complex polymorphic system is effectively solved, and the interval gray number is used to describe the conditional probability in the DBN model, and the fault relationship between the components in the complex polymorphic system is accurately described.

[0094] In combination with the above specific experimental cases, the embodiment takes the starting power generation system of a multi-electric aircraft as a research object, and constructs a starting power generation system DBN model structure through the failure fault and working mechanism of the starting power generation system, which serves as a subsequent model support. For the DBN model of this complex polymorphic system, the GBIC-LDBN method is used to eliminate the redundant structure in the DBN model, construct a lightweight DBN model (LDBN) of the starting power generation system, and verify the advantages of LDBN and DBN model in calculation efficiency and accuracy. On the basis of ensuring the calculation accuracy, the calculation efficiency can be effectively improved.

[0095] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment mainly describes the difference from other embodiments. Especially, the device embodiment is described relatively simply because it is basically similar to the method embodiment, and the related parts can be referred to the part of the method embodiment. The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or replacements within the technical range disclosed by the present application can be easily thought by those skilled in the art, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A DBN model lightweight method for a GBIC-based multi-electric aircraft starting power generation system, characterized in that, The method comprises the following steps: S1, establishing a DBN model of a starting power generation system, wherein nodes in the DBN model correspond to key components and working states of the starting power generation system; S2, identifying redundant structures in the DBN model and performing lightweight processing; S3, identifying a fault state of the starting power generation system by using the DBN model after the lightweight processing; S2 comprises: obtaining a fitting degree of the DBN model to data and a parameter dimension of a node in the DBN model; calculating a GBIC value of a parent node combination in the DBN model, wherein the size of the GBIC value is positively correlated with the structural redundancy of the DBN model; obtaining the fitting degree of the DBN model to data comprises: ， ； wherein, and respectively represent the upper and lower bounds of the gray fuzzy likelihood value in the time slice t j , j represents the number corresponding to the time slice, and represent two initial nodes in the time slice t j , and represent the parent node set corresponding to the two initial nodes in the time slice t j , and respectively represent the gray fuzzy conditional probability values corresponding to the two initial nodes in the time slice t j , p represents the total number of root nodes x, and q represents the total number of intermediate nodes y; obtaining the parameter dimension of the node in the DBN model comprises: wherein, denotes the parameter dimension of a node, and denotes the number of parameters of a node x i and a node y i respectively. the calculation of the GBIC value of the parent node combination in the DBN model comprises: The upper and lower extreme values of the parent node combination in the DBN model are calculated to obtain a GBIC value range, wherein: , ; and denote the nodes and the upper and lower extreme values at task time T, and denote the nodes j at time slice t and of task time T, m denotes the total number of time slices t j . ， ， , where, denotes the time slice t j The node contains the interval grey number in the conditional probability value, denotes the lower limit of the probability corresponding to, denotes the upper limit of the probability corresponding to, r denotes the number of conditional probability values containing interval grey numbers, and n denotes the total number of conditional probability values containing interval grey numbers in the time slice t j .

2. The method of claim 1, wherein, the working state of the starting power generation system comprises: a starting mode and a power generation mode; the fault state of the key component of the starting power generation system comprises: normal working, abnormal working, partial failure and complete failure.

3. The method of claim 1, wherein, the DBN model of the starting power generation system comprises: , denotes the state transition probability of node x from time slice t to time slice t+1, x t denotes the state of a node at time slice t, x t+1 denotes the state of a node at t+1, t denotes a time slice, i denotes a node corresponding number, x i t denotes the state of the i-th node at time slice t, x i t+1 denotes the state of the i-th node at time slice t+1, N denotes the total number of nodes.

Images