A multi-level system model construction method, system, device and medium

Abstract

The application discloses a multilevel system model construction method, system, device and medium, and relates to the technical field of system engineering and system modeling. The method comprises the following steps: acquiring information data of a multilevel system to be constructed; determining a structure undirected graph according to the structure relationship and the construction objects between construction objects; performing regional segmentation according to a set path step based on the level corresponding to the construction objects and the structure undirected graph, and obtaining a plurality of regional network structures; for any regional network structure, determining the dependency relationship between the construction objects by using a data structure learning algorithm, and determining a graphical network according to the dependency relationship; determining a structure network model of the multilevel system to be constructed according to the structure undirected graph and all the graphical networks; and the structure network model is used for representing the mechanism characteristic information of the multilevel system in cooperative interaction. The application can perfect the global structure of the system and realize the construction of a global static model.

Description

Technical Field

[0001] This invention relates to the fields of systems engineering and system modeling technology, and in particular to a method, system, device and medium for constructing multi-level system models. Background Technology

[0002] Understanding the structural relationships within a system is fundamental for building a system model and is a crucial step in comprehensively utilizing system information. Clearly defining the global spatial structural relationships of components and subsystems in a multi-level dynamic system is a basic prerequisite for dynamic system modeling and health status assessment. On the one hand, current methods for understanding structure primarily rely on empirically constructed dynamic fault trees combined with Markov processes for quantification, which has significant limitations in understanding the complex hierarchical relationships of system state structures and cannot provide a comprehensive understanding of the structure. On the other hand, data-based Bayesian network structure learning algorithms can achieve rapid and effective learning of system structure on a small scale, but they have high requirements for system size and data type rules, and are technically immature for large-scale systems. Summary of the Invention

[0003] The purpose of this invention is to provide a method, system, device, and medium for constructing a multi-level system model, which can improve the global structure of the system and realize the construction of a global static model.

[0004] To achieve the above objectives, the present invention provides the following solution:

[0005] A method for constructing a multi-level system model, the method comprising:

[0006] Acquire information data of the multi-level system to be constructed; the information data includes: construction objects, the levels corresponding to the construction objects, and the structural relationships between the construction objects; the construction objects include: multiple component devices;

[0007] Based on the structural relationships between the constructed objects and the constructed objects themselves, a structural undirected graph is determined;

[0008] Based on the level corresponding to the constructed object and the undirected graph of the structure, the region is segmented according to the set path step size to obtain multiple region network structures;

[0009] For any of the aforementioned regional network structures, a data structure learning algorithm is used to determine the dependencies between the building objects, and a graphical network is determined based on the dependencies; the dependencies are the logical relationships between the building objects in the regional network structure.

[0010] Based on the undirected structural graph and all the graphical networks, the structural network model of the multi-level system to be constructed is determined; the structural network model is used to characterize the mechanistic characteristics of the multi-level system during collaborative interaction.

[0011] Optionally, based on the structural relationships between the constructed objects and the constructed objects, a structural undirected graph is determined, specifically including:

[0012] Determine the initial undirected graph based on the constructed object;

[0013] Based on the structural relationship between the initial undirected graph and the constructed objects, a structural undirected graph is determined.

[0014] Optionally, based on the level corresponding to the constructed object and the undirected graph of the structure, region segmentation is performed according to a set path step size to obtain multiple region network structures, specifically including:

[0015] The baseline is determined based on the structural relationships between the constructed objects and the corresponding levels of the constructed objects;

[0016] Construct a node set; the node set is a collection composed of multiple component devices in the constructed object;

[0017] For any component device in the node set, with the reference point as the center, determine whether the path length is not greater than the set path step size according to the undirected graph of the structure; the path length is the distance between the component device and the reference point;

[0018] If so, then construct the first blank undirected graph, and connect the corresponding component device and the reference point in the first blank undirected graph according to the corresponding path length to obtain the connecting edge;

[0019] The first undirected subgraph is determined based on all the edges, and the first undirected subgraph is determined as the first region network structure;

[0020] If not, then construct the i-th empty set and transfer the corresponding component device into the i-th empty set; i≥1;

[0021] The component set is determined based on all the component devices transferred into the i-th blank set;

[0022] Transfer all component devices whose path length is equal to the set path step size to the (i+1)th empty set to obtain the set;

[0023] Determine the target path step size based on the set path step size;

[0024] Using any of the component devices in the set as the central reference point, and based on the undirected graph of the structure, determine whether the length of each node is not greater than the target path step length; the node length is the distance between the central reference point and any component device in the component set;

[0025] If so, construct the (i+1)th blank undirected graph, and connect the corresponding component device and the central reference point in the (i+1)th blank undirected graph according to the corresponding node length to obtain the connecting edge;

[0026] The (i+1)th undirected subgraph is determined based on all the edges, and the (i+1)th undirected subgraph is determined as the (i+1)th region network structure.

[0027] If not, construct the (i+1)th blank set, transfer the corresponding component device into the (i+1)th blank set, and return to the step "determine the component set based on all the component devices transferred into the i-th blank set".

[0028] Optionally, for any of the aforementioned regional network structures, a data structure learning algorithm is used to determine the dependencies between the construction objects, and a graphical network is determined based on the dependencies, specifically including:

[0029] For any of the aforementioned regional network structures, the PC algorithm, based on the characteristics of multivariate Gaussian distribution, is used to calculate the partial correlation coefficient of the constructed object, thereby obtaining the dependency relationship between each component device in the constructed object;

[0030] For any two component devices connected by any edge in any of the aforementioned regional network structures, a directed connection between the two component devices is determined based on the dependency and structural relationships, and added to the regional network structure to obtain a regional update network structure.

[0031] For any three component devices in any of the aforementioned region update network structures, a triplet structure is determined according to a set partition set, and the triplet structure is added to the region update network structure to obtain a graphical network.

[0032] Optionally, based on the undirected structural graph and all the graphical networks, the structural network model of the multi-level system to be constructed is determined, specifically including:

[0033] Based on the undirected graph of the structure, all the graphical networks are connected to obtain the initial structural network model;

[0034] For any two component devices in the defined set, a directed connection between the two component devices is determined based on the dependency relationship and all the graphical networks.

[0035] The structural network model is determined based on all the directed connections corresponding to the set and the initial structural network model.

[0036] A multi-level system model construction system, the system comprising:

[0037] The acquisition module is used to acquire information data of the multi-level system to be built; the information data includes: construction objects, the levels corresponding to the construction objects, and the structural relationships between the construction objects; the construction objects include: components and devices.

[0038] The undirected graph determination module is used to determine the undirected graph based on the construction objects and the structural relationships between them.

[0039] The region segmentation module is used to segment regions based on the level corresponding to the construction object and the undirected graph of the structure, according to a set path step size, to obtain multiple region network structures.

[0040] A graphical network determination module is used to determine the dependencies between construction objects for any of the aforementioned regional network structures using a data structure learning algorithm, and to determine the graphical network based on the dependencies; the dependencies are the logical relationships between construction objects in the regional network structure.

[0041] The determination module is used to determine the structural network model of the multi-level system to be constructed based on the undirected structural graph and all the graphical networks; the structural network model is used to characterize the mechanistic characteristics of the multi-level system during collaborative interaction.

[0042] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor runs the computer program to enable the electronic device to perform the multi-level system model construction method described above.

[0043] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the multi-level system model construction method described above.

[0044] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0045] This invention provides a method, system, device, and medium for constructing a multi-level system model. First, information data of the multi-level system to be constructed is acquired. Based on the structural relationships between the construction objects and the construction objects themselves, a structural undirected graph is determined. Based on the levels corresponding to the construction objects and the structural undirected graph, regions are segmented according to a set path step size, resulting in multiple regional network structures. For any regional network structure, a data structure learning algorithm is used to determine the dependencies between the construction objects, and a graphical network is determined based on these dependencies. Based on the structural undirected graph and all graphical networks, the structural network model of the multi-level system to be constructed is determined. This invention improves the system structure by employing region segmentation and data structure learning methods. Furthermore, by combining the structural undirected graph with a comprehensive overview of the system, a structural network model is ultimately obtained. Therefore, this invention can improve the global structure of the system and realize the construction of a global static model. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 A flowchart of a multi-level system model construction method provided in an embodiment of the present invention;

[0048] Figure 2 A schematic diagram illustrating the construction framework principle of the multi-level system model construction method provided in this invention in practical applications;

[0049] Figure 3 FTA analysis diagram of the servo control device provided in the embodiment of the present invention;

[0050] Figure 4 A STAMP analysis diagram of the servo control device provided in an embodiment of the present invention;

[0051] Figure 5 An undirected structural diagram of the servo control device provided in the embodiments of the present invention;

[0052] Figure 6 A Bayesian network model diagram of a solar-powered unmanned aerial vehicle (UAV) airborne system provided in an embodiment of the present invention;

[0053] Figure 7 This is a diagram showing the results of structure learning using the PC algorithm provided in an embodiment of the present invention.

[0054] Figure 8 The figure shows the analysis results of qualitative analysis of an airborne system using FTA, provided in an embodiment of the present invention.

[0055] Figure 9 The figure shows the analysis results of qualitative analysis of airborne systems using the STAMP method, as provided in an embodiment of the present invention.

[0056] Figure 10 This is a schematic diagram illustrating the preliminary undirected graph model construction of a solar-powered unmanned aerial vehicle (UAV) airborne system, as provided in an embodiment of the present invention.

[0057] Figure 11 This is a diagram showing the result of structure learning using a constraint-based structure learning method provided in an embodiment of the present invention.

[0058] Figure 12 This is a Bayesian network structure diagram for the output of a solar-powered UAV airborne system after region segmentation and reorganization, provided in an embodiment of the present invention.

[0059] Figure 13 Construct a system structure diagram for a multi-level system model.

[0060] Symbol explanation:

[0061] Acquisition Module-1, Structure Undirected Graph Determination Module-2, Region Segmentation Module-3, Graphical Network Determination Module-4, Determination Module-5. Detailed Implementation

[0062] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] According to system reliability modeling methods, when the structure is relatively simple, the design structure and empirical information are sufficient to fully express the internal structural relationships of the system. However, as the design scale increases and the complexity of the relationships between components within the system increases, the limitations and insufficiency of empirical information gradually become apparent.

[0064] ①Although the experiential information provided by knowledge scholars is highly domain-specific, in a multi-type subsystem collaborative context, even if the structural cognition within an independent subsystem is fully developed, the impactful cognition of collaborative interactions cannot be guaranteed.

[0065] ② Although the mechanisms of traditional equipment and components are well understood, the reliability characteristics of emerging devices and equipment lack long-term accumulation, and empirical information cannot guarantee a correct understanding of the reliability mechanisms of such objects.

[0066] Conversely, data-driven Bayesian network structure learning methods, because their construction criteria are based on data relationships of system states, achieve higher accuracy and realism in structure construction compared to modeling methods based on empirical information. However, data-driven structure learning methods also have significant drawbacks in practical applications:

[0067] ① The data-driven algorithm principle makes such methods inherently require a large amount of measured data to ensure correctness, but in engineering applications, it is almost impossible to collect complete system-wide health status data.

[0068] ② Due to limitations in algorithm performance, the computational cost of data-driven methods increases rapidly with the size of the structure. These issues prevent data-driven structure learning methods from being practically applied at present.

[0069] The purpose of this invention is to provide a method, system, device, and medium for constructing a multi-level system model, which can improve the global structure of the system and realize the construction of a global static model.

[0070] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0071] Example 1

[0072] like Figure 1 As shown, this embodiment of the invention provides a method for constructing a multi-level system model, the method comprising:

[0073] Step 100: Obtain information data of the multi-level system to be built; the information data includes: the building objects, the levels corresponding to the building objects, and the structural relationships between the building objects; the building objects include: multiple component devices.

[0074] Figure 2 This diagram illustrates the framework principle of the multi-level system model construction method provided by this invention in practical applications. First, the information data to be analyzed in the system, such as components, devices, subsystems, and the system itself, is defined. This definition remains unchanged throughout the subsequent process and forms the basic structure of the system network. Considering the current design stage of the analyzed objects (i.e., the information data), and leveraging background information such as structural design and expert experience, different analytical perspectives are employed to obtain the structural relationships between the constructed objects. If S... i Let S represent the set of relations given by the i-th method. This gives the final set of node relations, which serves as the structural relationship between the constructed objects. i .

[0075] The methods used to determine structural relationships include, but are not limited to, fault tree analysis, FMEA, STAMP, and other empirical information-based analysis techniques.

[0076] Step 200: Determine the undirected structural graph based on the structural relationships between the constructed objects and the constructed objects themselves.

[0077] Specifically, based on the structural relationships between the constructed objects and the constructed objects themselves, a structural undirected graph is determined, including:

[0078] The initial undirected graph is determined based on the objects being constructed.

[0079] Based on the structural relationships between the initial undirected graph and the constructed objects, the structural undirected graph is determined.

[0080] In practical applications, the specific implementation process is as follows:

[0081] (1) Create a complete undirected graph C containing the set V of all nodes, i.e. component devices, in the building object, and complete the undirected graph initialization.

[0082] (2) For any pair of nodes (x,y) in an undirected graph C, determine whether there is a node pair association in any relation set S; if not, delete the edge between the pair of nodes (x,y) in the undirected graph C, otherwise retain it.

[0083] (3) Update the undirected graph C and repeat (2) until all node pairs have been traversed.

[0084] (4) Output the updated undirected graph C as a structured undirected graph.

[0085] The servo control device is used as a specific implementation example for detailed explanation. Figure 3 and Figure 4 The figures shown are the FTA and STAMP analysis diagrams for the servo control device. Figure 3 and Figure 4 The analysis results yield component relationship sets S1 and S2 provided by the FTA and STAMP methods, respectively; Table 1 shows the node relationship set of the UAV servo control device. Then, the final undirected structural graph obtained from the node sets is shown in Table 1. Figure 5 .

[0086] Table 1. Set of node relationships for UAV servo control devices

[0087] node logo <![CDATA[Set of relationships S1]]> <![CDATA[Set of relationships S2]]> Flight control computer a ab, ac, ad, ae ac, ad, ae, af Servo unit b ba, bf bf, be Inertial navigation equipment c ca ca, cf antenna d da da, df sensor e ea ea, eb FlexRay bus f fb fa, fb, fc, fd

[0088] Step 300: Based on the level and structure undirected graph corresponding to the constructed object, perform region segmentation according to the set path step size to obtain multiple region network structures.

[0089] Specifically, based on the level and structure undirected graph corresponding to the constructed object, region segmentation is performed according to the set path step size to obtain multiple region network structures, including:

[0090] The baseline is determined based on the structural relationships between the constructed objects and the corresponding levels of the constructed objects.

[0091] Build a node collection; a node collection is a collection of multiple component devices in a build object.

[0092] For any component device in the node set, with the reference point as the center, determine whether the path length is not greater than the set path step size based on the undirected graph of the structure; the path length is the distance between the component device and the reference point.

[0093] If so, construct the first blank undirected graph, and connect the corresponding component devices and reference points within the first blank undirected graph according to the corresponding path length to obtain the connecting edges; then, determine the first undirected subgraph based on all the connecting edges, and determine the first undirected subgraph as the first regional network structure.

[0094] If not, construct the i-th blank set and transfer the corresponding component devices into the i-th blank set; i≥1; then determine the component set based on all the component devices transferred into the i-th blank set.

[0095] Transfer all component devices whose path length is equal to the set path step to the (i+1)th empty set to obtain the set; determine the target path step based on the set path step.

[0096] Using any component device within the set as the central reference point, and based on the undirected graph of the structure, determine whether the length of each node is not greater than the target path step length; the node length is the distance between the central reference point and any component device in the component set.

[0097] If so, construct the (i+1)th blank undirected graph, and connect the corresponding component devices and the central reference point in the (i+1)th blank undirected graph according to the corresponding node length to obtain the connecting edges; determine the (i+1)th undirected subgraph based on all the connecting edges, and determine the (i+1)th undirected subgraph as the (i+1)th regional network structure.

[0098] If not, construct the (i+1)th blank set, transfer the corresponding component devices into the (i+1)th blank set, and return the step "determine the component set based on all component devices transferred into the i-th blank set".

[0099] Specifically, the undirected structural graph output above, denoted as E, serves as the input for subsequent graph region segmentation. Before segmentation, it's necessary to first identify which node in the system structure represented by the undirected structural graph E represents the highest-level system node and use it as the initial reference node. Then, all independent nodes connected to the reference node are searched, progressively advancing the system hierarchy to construct a subgraph. Throughout this process, the subgraph expansion level L and the path increment step d need to be adaptively adjusted based on the overall system scale and segmentation accuracy. The specific operation process in practical applications can be as follows:

[0100] (1) If the subgraph expansion level L = 1, output the undirected graph E and end this step; otherwise, determine the system-level node x0 as the reference node x according to the design structure, set the path discrimination standard D = d and go to (2).

[0101] (2) For the base node x, take any node y in the node set V, and determine whether the minimum path length between the node pair (x, y) is not greater than D. If so, add the node pair (x, y) and its connecting edge to the undirected subgraph.

[0102] (3) Repeat operation (2) until all nodes in the node set V have been traversed, and output the undirected subgraph.

[0103] (4) Take a set of nodes W, where the minimum path length between any node y in the set of nodes W and the system-level node x0 is D.

[0104] (5) Increase the step size d by the path discrimination standard D, that is, D = D + d at this time.

[0105] (6) Take any node w in the node set W as the reference node, traverse any node y in the node set V that does not exist in the output undirected subgraph, and determine whether the minimum path length between the node pair (x0, y) is not greater than D in process (5), and whether the minimum path passes through the reference node w. If so, add the node pair (w, y) and its connecting edge to the undirected graph.

[0106] (7) Repeat process (6) until all nodes in the node set W have been traversed, and output the undirected subgraph.

[0107] (8) Repeat process (4) to (6) until every node in the node set V exists in a certain undirected subgraph.

[0108] Step 400: For any regional network structure, use a data structure learning algorithm to determine the dependencies between the construction objects, and determine the graphical network based on the dependencies; the dependencies are the logical relationships between the construction objects in the regional network structure.

[0109] Specifically, for any regional network structure, a data structure learning algorithm is used to determine the dependencies between the building objects, and a graphical network is determined based on these dependencies, including:

[0110] For any regional network structure, the PC algorithm is used to calculate the partial correlation coefficient of the constructed object based on the characteristics of multivariate Gaussian distribution, so as to obtain the dependency relationship between each component device in the constructed object.

[0111] For any two component devices with any connection edge in any regional network structure, the directed connection between the two component devices is determined based on the dependency and structural relationship, and added to the regional network structure to obtain the regional update network structure.

[0112] For any three component devices in any region update network structure, the triplet structure is determined according to the set partition set, and the triplet structure is added to the region update network structure to obtain a graphical network.

[0113] In practical applications, all segmented regional network structures are independently subjected to data-driven structure learning and supplementation to improve the network structure of each region, including the learning of node relationships and the orientation of edges.

[0114] (1) Take any undirected subgraph, i.e., the regional network structure, obtain the observation dataset containing all nodes in the subgraph, and use the PC algorithm to perform structure learning on the undirected subgraph to determine the undirected relationship between nodes.

[0115] (2) Take any pair of nodes (x, y) in the undirected subgraph. If their node relation exists in any relation set S, then... i In this process, the edge connecting node pairs (x, y) is added to the undirected subgraph.

[0116] (3) Take any triple (x, y, z) in any undirected subgraph. If both the node pairs (x, y) and (y, z) exist in the undirected subgraph, and the node pair (x, z) does not exist, determine whether node y belongs to the defined partition set of the node pair (x, z), i.e., the minimum partition set S. xz If it does not belong to the category, the triple (x, y, z) can be defined as a v-structure x→y←z.

[0117] (4) Repeat process (3) until all triples are traversed.

[0118] (5) Repeat process (1) to (4) until all undirected subgraphs have been traversed.

[0119] (6) Determine the direction of edges in non-v-structures by referring to supplementary rules: 1) Determine the direction of any remaining edge so that a new v-structure appears in the directed acyclic graph of that part; 2) For any directed path, if its initial node is connected to its end node, a loop cannot be formed.

[0120] (7) Update the regional network structure that determines all edge directions into a graphical network, also known as a sub-Bayesian network G. l .

[0121] Regarding the minimum partition set S xz In a v-structure, if the state of one node is known, then there will be no probabilistic relationship between the other two nodes. This phenomenon is called d-separation.

[0122] For any node X i and X j The set of all d-separated nodes is called the minimum split set of these two nodes. For a given network structure, the number of split sets is uncertain, but the minimum split set is deterministic.

[0123] Step 500: Based on the undirected structural graph and all graphical networks, determine the structural network model of the multi-level system to be constructed; the structural network model is used to characterize the mechanistic characteristics of the multi-level system during collaborative interaction.

[0124] Specifically, based on the structured undirected graph and all graphical networks, the structural network model of the multi-level system to be constructed is determined, including:

[0125] Based on the undirected structural graph, all the graphical networks are connected to obtain the initial structural network model.

[0126] For any two component devices in the set, determine the directed connection between the two component devices based on dependencies and all graphical networks.

[0127] Based on all the directed connections and the initial structure network model corresponding to the given set, determine the structure network model.

[0128] The completed local structure, i.e., all the graphical networks, is merged with the inter-regional relationships before region segmentation. The relative positions of the reference nodes of each subgraph in the original global undirected graph are reorganized and connected to form a global Bayesian network model G, i.e., a structural network model, thus completing the construction of the system structure. The specific implementation process is as follows:

[0129] (1) For any undirected subgraph During the construction process, there is a baseline node w, and all baseline nodes are combined into a set Z.

[0130] (2) Take any pair of nodes (x, y) in set Z. If the node pair relationship exists in any sub-Bayesian network G, then... l In the middle, the sub-Bayesian network G with nodes x and y as reference nodes will be... l Connect based on the relationship between x and y.

[0131] (3) Repeat process (2) until all sub-Bayesian networks G are completed. l The connections form a global Bayesian network G.

[0132] In practical applications, let's take a simplified solar-powered drone onboard system structure as an example. A drone is a typical multi-level system. Under the influence of various internal and external stresses (such as temperature, humidity, vibration, and load distribution), there are potential interactions and influences between the health states of its components and subsystems. These interactions are not entirely derived from the design structural relationships and therefore cannot be fully identified based solely on design and engineering information.

[0133] This embodiment only involves the airborne system, which includes a flight control system, an energy system, and a power system. To better demonstrate the implementation process of the proposed method, the detailed composition structure of some subsystems was adaptively expanded. In the structural expansion, while considering structural relationships, the underlying component relationships of some high-level equipment were refined according to the working conditions of the UAV system. For example, in the empirical analysis based on design information, there is no direct influence relationship between the energy subsystems because they have independent management systems and peripheral control circuits in their design structure. However, considering that during flight, each energy subsystem jointly supplies power to the electrical equipment and shares the workload, when one energy subsystem fails to supply power normally, the other subsystems must bear a greater load for normal flight, accelerating the decay process.

[0134] Based on the understanding of the design structure and component relationships of airborne systems, a complete Bayesian network model of the solar-powered UAV airborne system is first presented, see [link to relevant documentation]. Figure 6 Based on past engineering implementation experience, marginal and conditional probability tables for the model were pre-defined. Some of the pre-defined probability tables are shown in Tables 2 to 6. Table 2 shows the marginal probability distribution of the power management device.

[0135] Table 2 Edge Probability Distribution of Power Management Devices

[0136]

[0137] Table 3 is the conditional probability table for solar cells.

[0138] Table 3 Conditional Probability Table for Solar Cells

[0139]

[0140] Table 4 Conditional Probability Table for Energy Subsystem A

[0141]

[0142] Table 5 Conditional Probability Table for Energy Subsystem B

[0143]

[0144]

[0145] Table 6 Conditional Probability Table of Energy Systems

[0146]

[0147] Based on the known structure and parameters of the above model, this embodiment simulates 1000 sets of fault data vectors of airborne equipment and systems.

[0148] We chose a constraint-based PC structure learning algorithm to compare its performance with that of region segmentation and reconstruction methods. Figure 7 The structure learning results of the PC algorithm are derived from... Figure 7 As can be seen, the algorithm ignores the edges between the power management device and the energy subsystem, between energy subsystem A and energy subsystem B, and between the left wing and the right wing.

[0149] In region segmentation and reassembly methods, a qualitative analysis of the system is first required. In this case, the most commonly used FTA and STAMP methods are employed to perform the qualitative analysis of the airborne system. The analysis results of FTA and STAMP are as follows: Figure 8 and Figure 9 As shown, this leads to the relation set shown in Table 7. For ease of explanation, nodes are no longer identified by their full names, but by their corresponding uppercase letters. Similarly, nodes in the network are also identified by their identifiers.

[0150] Table 7 Quantitative Analysis of Node Relationships

[0151]

[0152]

[0153] Based on the set of node relationships given in Table 7, a preliminary undirected graph model of the system can be constructed, such as... Figure 10 As shown. Based on the region segmentation process, the entire map is divided into 4 sub-maps. The first sub-map... The system includes airborne system nodes and subsystem nodes of each subsystem; the second subgraph β, the third subgraph γ, and the fourth subgraph δ each include nodes within their respective subsystems. A constraint-based structure learning method is used to perform structure learning on each of the four subgraphs, clarifying the relationships between nodes and the directions of edges. The calculation results are shown below. Figure 11 Ultimately, Figure 11 After the subgraphs in the network are restructured, the final Bayesian network structure diagram output by the region segmentation and restructuring can be obtained, as shown in the figure below. Figure 12 As shown, the edges represented by dashed lines are node relationships that differ from those obtained by the PC algorithm.

[0154] Will Figure 7 , Figure 12 respectively with Figure 6 Comparing the Bayesian network structures obtained by the two methods, the differences in the connections between nodes are clearly visible. Firstly, the most significant difference is... Figure 7 The relationship between energy subsystems A and B, and between the left and right wings, is not included. Due to the core idea of ​​constraint-based algorithms, edges are often not set between two parent nodes in the v-structure. Even if a close fault relationship exists between them in real data, this relationship is weakened in the network structure, reflected as a joint effect of both on the child nodes. In other words, the core idea of ​​constraint-based learning methods is to provide a relatively comprehensive consideration of the global system structure, with minimal negative impact on the overall system evaluation capability; however, it still ignores many local details, making the learning results less suitable for detailed system analysis in systems with strong fault coupling. Secondly, in Figure 7 In this model, a connection exists between the power management device and the solar cells, but the edge between the power management system and the energy subsystem is ignored. This is equivalent to replacing the v-structure in the triplet structure with a unidirectional structure, which obviously introduces some error. Although the probabilistic influence of the power management device on the energy subsystem can be transmitted through the solar cell nodes, these two structures are not equivalently substituted, especially when there is measured state evidence input for either the solar cell node or the power management device node. The inference results of the two network structures will be completely different. Furthermore, there are slight differences in the network structures given by the two methods in the flight control system. Table 8 shows the differences in the Bayesian network structures between the region segmentation and recombination method and the PC algorithm.

[0155] Table 8. Differences in Bayesian network structure between the region segmentation and recombination method and the PC algorithm.

[0156]

[0157] Example 2

[0158] like Figure 13 As shown, this embodiment of the invention provides a multi-level system model construction system, which includes: an acquisition module 1, a structural undirected graph determination module 2, a region segmentation module 3, a graphical network determination module 4, and a determination module 5.

[0159] Module 1 is used to acquire information data of the multi-level system to be built; the information data includes: the building objects, the levels corresponding to the building objects, and the structural relationships between the building objects; the building objects include: components and devices.

[0160] The undirected graph determination module 2 is used to determine the undirected graph based on the constructed objects and the structural relationships between them.

[0161] Region segmentation module 3 is used to segment regions based on the level and structure undirected graph corresponding to the constructed object and according to the set path step size, so as to obtain multiple region network structures.

[0162] The graphical network determination module 4 is used to determine the dependencies between construction objects for any regional network structure using a data structure learning algorithm, and to determine the graphical network based on the dependencies; the dependencies are the logical relationships between construction objects in the regional network structure.

[0163] Module 5 is used to determine the structural network model of the multi-level system to be constructed based on the undirected structural graph and all graphical networks; the structural network model is used to characterize the mechanistic characteristics of the multi-level system during collaborative interaction.

[0164] Example 3

[0165] This invention provides an electronic device, including a memory and a processor. The memory stores a computer program, and the processor runs the computer program to enable the electronic device to execute the multi-level system model construction method in Embodiment 1.

[0166] As an alternative implementation, the electronic device can be a server.

[0167] In one embodiment, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the multi-level system model construction method of embodiment 1.

[0168] Knowledge-based empirical analysis and modeling methods can efficiently construct the main framework using only design information; data-driven structural learning algorithms possess fast and accurate small-scale structural learning capabilities, which can precisely compensate for the shortcomings of empirical information modeling methods. Based on the principle of complementing the advantages of these two methods, this approach combines empirical information analysis methods with constraint methods. By utilizing the regional reorganization and segmentation process based on the global structure and leveraging the structural characteristics of multi-level systems to divide local structures, the input data scale of constraint-based structural learning methods is reduced. A robust static Bayesian network is then established to represent the relationships between components in the system, effectively improving the overall efficiency and accuracy of structural learning algorithms.

[0169] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0170] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for constructing a multi-level system model, characterized in that, The method includes: Acquire information data of the multi-level system to be constructed; the information data includes: construction objects, the levels corresponding to the construction objects, and the structural relationships between the construction objects; the construction objects include: multiple component devices; Based on the structural relationships between the constructed objects and the constructed objects themselves, a structural undirected graph is determined; Based on the level corresponding to the constructed object and the undirected graph of the structure, the region is segmented according to the set path step size to obtain multiple region network structures; For any of the aforementioned regional network structures, a data structure learning algorithm is used to determine the dependencies between the building objects, and a graphical network is determined based on the dependencies; the dependencies are the logical relationships between the building objects in the regional network structure. Based on the undirected structural graph and all the graphical networks, the structural network model of the multi-level system to be constructed is determined; the structural network model is used to characterize the mechanistic characteristics of the multi-level system during collaborative interaction. Based on the level corresponding to the constructed object and the undirected graph of the structure, region segmentation is performed according to the set path step size to obtain multiple region network structures, specifically including: The baseline is determined based on the structural relationships between the constructed objects and the corresponding levels of the constructed objects; Construct a node set; the node set is a collection composed of multiple component devices in the constructed object; For any component device in the node set, with the reference point as the center, determine whether the path length is not greater than the set path step size according to the undirected graph of the structure; the path length is the distance between the component device and the reference point; If so, then construct the first blank undirected graph, and connect the corresponding component device and the reference point in the first blank undirected graph according to the corresponding path length to obtain the connecting edge; The first undirected subgraph is determined based on all the edges, and the first undirected subgraph is determined as the first region network structure; If not, then construct the i-th empty set and transfer the corresponding component device into the i-th empty set; i≥1; The component set is determined based on all the component devices transferred into the i-th blank set; Transfer all component devices whose path length is equal to the set path step size to the (i+1)th empty set to obtain the set; Determine the target path step size based on the set path step size; Using any of the component devices in the set as the central reference point, and based on the undirected graph of the structure, determine whether the length of each node is not greater than the target path step length; the node length is the distance between the central reference point and any component device in the component set; If so, construct the (i+1)th blank undirected graph, and connect the corresponding component device and the central reference point in the (i+1)th blank undirected graph according to the corresponding node length to obtain the connecting edge; The (i+1)th undirected subgraph is determined based on all the edges, and the (i+1)th undirected subgraph is determined as the (i+1)th region network structure. If not, construct the (i+1)th blank set, transfer the corresponding component device into the (i+1)th blank set, and return to the step "determine the component set based on all the component devices transferred into the i-th blank set".

2. The method for constructing a multi-level system model according to claim 1, characterized in that, Based on the structural relationships between the constructed objects and the constructed objects themselves, a structural undirected graph is determined, specifically including: Determine the initial undirected graph based on the constructed object; Based on the structural relationship between the initial undirected graph and the constructed objects, a structural undirected graph is determined.

3. The method for constructing a multi-level system model according to claim 1, characterized in that, For any of the aforementioned regional network structures, a data structure learning algorithm is used to determine the dependencies between the constructed objects, and a graphical network is determined based on the dependencies, specifically including: For any of the aforementioned regional network structures, the PC algorithm, based on the characteristics of multivariate Gaussian distribution, is used to calculate the partial correlation coefficient of the constructed object, thereby obtaining the dependency relationship between each component device in the constructed object; For any two component devices connected by any edge in any of the aforementioned regional network structures, a directed connection between the two component devices is determined based on the dependency and structural relationships, and added to the regional network structure to obtain a regional update network structure. For any three component devices in any of the aforementioned region update network structures, a triplet structure is determined according to a set partition set, and the triplet structure is added to the region update network structure to obtain a graphical network.

4. The method for constructing a multi-level system model according to claim 1, characterized in that, Based on the aforementioned undirected graph and all the aforementioned graphical networks, the structural network model of the multi-level system to be constructed is determined, specifically including: Based on the undirected graph of the structure, all the graphical networks are connected to obtain the initial structural network model; For any two component devices in the defined set, a directed connection between the two component devices is determined based on the dependency relationship and all the graphical networks. The structural network model is determined based on all the directed connections corresponding to the set and the initial structural network model.

5. A multi-level system model construction system, characterized in that, The system includes: The acquisition module is used to acquire information data of the multi-level system to be built; the information data includes: the construction object, the level corresponding to the construction object, and the structural relationship between the construction objects; the construction object includes: multiple component devices; The undirected graph determination module is used to determine the undirected graph based on the construction objects and the structural relationships between them. The region segmentation module is used to segment regions based on the level corresponding to the construction object and the undirected graph of the structure, according to a set path step size, to obtain multiple region network structures. A graphical network determination module is used to determine the dependencies between construction objects for any of the aforementioned regional network structures using a data structure learning algorithm, and to determine the graphical network based on the dependencies; the dependencies are the logical relationships between construction objects in the regional network structure. The determination module is used to determine the structural network model of the multi-level system to be constructed based on the undirected structural graph and all the graphical networks; the structural network model is used to characterize the mechanistic characteristics of the multi-level system during collaborative interaction; Based on the level corresponding to the constructed object and the undirected graph of the structure, region segmentation is performed according to the set path step size to obtain multiple region network structures, specifically including: The baseline is determined based on the structural relationships between the constructed objects and the corresponding levels of the constructed objects; Construct a node set; the node set is a collection composed of multiple component devices in the constructed object; For any component device in the node set, with the reference point as the center, determine whether the path length is not greater than the set path step size according to the undirected graph of the structure; the path length is the distance between the component device and the reference point; If so, then construct the first blank undirected graph, and connect the corresponding component device and the reference point in the first blank undirected graph according to the corresponding path length to obtain the connecting edge; The first undirected subgraph is determined based on all the edges, and the first undirected subgraph is determined as the first region network structure; If not, then construct the i-th empty set and transfer the corresponding component device into the i-th empty set; i≥1; The component set is determined based on all the component devices transferred into the i-th blank set; Transfer all component devices whose path length is equal to the set path step size to the (i+1)th empty set to obtain the set; Determine the target path step size based on the set path step size; Using any of the component devices in the set as the central reference point, and based on the undirected graph of the structure, determine whether the length of each node is not greater than the target path step length; the node length is the distance between the central reference point and any component device in the component set; If so, construct the (i+1)th blank undirected graph, and connect the corresponding component device and the central reference point in the (i+1)th blank undirected graph according to the corresponding node length to obtain the connecting edge; The (i+1)th undirected subgraph is determined based on all the edges, and the (i+1)th undirected subgraph is determined as the (i+1)th region network structure. If not, construct the (i+1)th blank set, transfer the corresponding component device into the (i+1)th blank set, and return to the step "determine the component set based on all the component devices transferred into the i-th blank set".

6. An electronic device, characterized in that, The device includes a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to cause the electronic device to perform the multi-level system model construction method as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the multi-level system model construction method as described in any one of claims 1 to 4.

Images