Information processing device, information processing method, and program

Abstract

To provide an information processing device that can further efficiently analyze, visualize, and control resilience of complex systems.SOLUTION: An information processing device includes a processing unit. The processing unit extracts a plurality of feature quantities representing features of a plurality of pieces of second sampling data including a plurality of pieces of first sampling data generated using a multidimensional first probability distribution representing a distribution of each of a plurality of pieces of input data related to an analysis target, and a plurality of pieces of output data representing a physical quantity of the analysis target, the output data being acquired using the plurality of pieces of first sampling data as an input, and generates a network model that represents relations between a plurality of nodes corresponding to physical quantities, including a node corresponding to the extracted feature quantities.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] An embodiment of the present invention relates to an information processing device, an information processing method, and a program. [Background technology]

[0002] Resilience management based on the Safety-II concept, which consists of prediction, monitoring, response and learning, is expected to reduce downtime and improve availability of infrastructure systems, which are an example of complex systems. Based on the Safety-II concept, methods are required to improve the resilience of complex systems against disturbances (such as reducing system downtime) and to realize analysis, visualization and control to perform appropriate resilience management. In addition, to realize a circular economy, methods are also required to realize analysis, visualization and control to balance multiple values ​​such as environmental friendliness, economic efficiency and resilience. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2022-180873 A [Patent Document 2] JP 2022-049067 A [Non-patent literature]

[0004] [Non-Patent Document 1] Auto-Encoding Variational Bayes, arXiv:1312.6114 [Non-Patent Document 2] FFJORD: FREE-FORM CONTINUOUS DYNAMICS FOR SCALABLE REVERSIBLE GENERATIVE MODELS, arXiv:1810.01367 [Non-Patent Document 3] NICE: NON-LINEAR INDEPENDENT COMPONENTS ESTIMATION, arXiv:1410.8516 [Non-Patent Document 4] A variational eigenvalue solver on a quantum processor, arXiv:1304.3061 [Non-Patent Document 5] A Variational Quantum Linear Solver Application to Discrete Finite-Element Methods, Entropy 2023, 25(4), 580 [Non-Patent Document 6] A Variational Quantum Algorithm for Generalized Eigenvalue Problems and Its Application to Finite Element Method, arXiv:2302.12602 [Non-Patent Document 7] Graph neural networks: A review of methods and applications, arXiv:1812.08434 [Non-Patent Document 8] Benjamin Zanger, Christian B. Mendl, Martin Schulz, and Martin Schreiber, “Quantum Algorithms for Solving Ordinary Differential Equations via Classical Integration Methods”, Quantum, 5, 502 (2021). [Non-Patent Document 9] A Comprehensive Survey of Graph Embedding: Problems, Techniques and Applications, arXiv:1709.07604 [Non-Patent Document 10] A Comprehensive Survey on Graph Neural Networks, arXiv:1901.00596 Summary of the Invention [Problem to be solved by the invention]

[0005] An object of the embodiments of the present invention is to provide an information processing system, an information processing method, and a program that are capable of more efficiently performing analysis, visualization, and control of the resilience of complex systems, etc. [Means for solving the problem]

[0006] An information processing apparatus according to an embodiment includes a processor that extracts a plurality of feature quantities that represent characteristics of a plurality of second sampling data, the feature quantities including a plurality of first sampling data generated using a multidimensional first probability distribution that represents a distribution of each of a plurality of input data related to an analysis target and a plurality of output data that represents physical quantities of the analysis target and is obtained using the plurality of first sampling data as input, and generates a network model that represents relationships between a plurality of nodes that correspond to the physical quantities, including nodes that correspond to the extracted feature quantities. [Brief description of the drawings]

[0007] [Figure 1] FIG. 1 is a block diagram of an information processing system according to an embodiment. [Diagram 2] Block diagram of the analysis unit. [Diagram 3] Block diagram of the SoS section. [Figure 4] FIG. 13 is a diagram showing an example of a sampling data set. [Diagram 5] FIG. 4 is a diagram for explaining a method for extracting a feature amount. [Figure 6] FIG. 1 shows an example of the configuration of a complex network. [Figure 7] FIG. 1 shows an example of the configuration of a complex network. [Figure 8] A diagram showing an example of a battery storage system and a complex network. [Figure 9] 11 is a flowchart of information processing by the platform of the embodiment. [Figure 10] FIG. 11 is a block diagram of an analysis unit according to the first modified example. [Figure 11] 13 is a flowchart of information processing according to the first modified example. [Figure 12] FIG. 11 is a block diagram of an analysis unit according to a second modified example. [Figure 13] 11 is a flowchart of information processing according to the second modified example. [Figure 14] A diagram showing an example of the flow of quantum-classical hybrid computing. [Figure 15] A diagram showing an example of a one-dimensional model of a string divided into eight finite elements. [Figure 16] FIG. 13 is a diagram showing an example of a quantum circuit corresponding to a matrix. [Figure 17] FIG. 1 shows an example of the configuration of a quantum circuit for a Hadamard test. [Figure 18] FIG. 4 is a diagram showing the correspondence between a mechanical vibration model and an electrical vibration model. [Figure 19] FIG. 1 shows an example of a random network. [Figure 20] FIG. 1 shows an example of a freescale network. [Figure 21] FIG. 13 is a graph showing an example of a change in damage probability over time. [Figure 22] FIG. 4 is a diagram showing an example of an equivalent circuit model representing current-voltage characteristics. [Figure 23] FIG. 2 is a hardware configuration diagram of an information processing device. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, preferred embodiments of an information processing apparatus according to the present invention will be described in detail with reference to the accompanying drawings.

[0009] In the following, a case will be described where the target system, which is a complex system to be analyzed, is an infrastructure system. Examples of infrastructure systems include a storage battery system, a wind power generation plant, power electronics, an elevator system, an infrastructure structure, a distributed energy plant, a power grid, a water grid, a transportation network, a robot system, and a communication network. The target system is not limited to an infrastructure system.

[0010] As mentioned above, there is a demand for methods to enable analysis, visualization, and control in order to carry out appropriate resilience management in infrastructure systems.

[0011] Here, we will explain Safety-I, which is contrasted with Safety-II. Safety-I is a concept that improves availability by suppressing the occurrence probability of events that cause the system state to exceed the reliability limit to an acceptable level by taking into account sufficient safety margins in conventional reliability engineering and risk-based engineering that are based on cause analysis and causal relationships.

[0012] In contrast, Safety-II is a new concept that improves the availability (resilience) of a system by acknowledging the existence of risk scenarios that have not been identified or addressed, focusing on normal operation states, and increasing buffer capacity, tolerance, and flexibility.Safety-II does not only focus on eliminating things that might go wrong, but also examines the reasons why things go right and creates a system that can implement a wide variety of measures and adjustments in a timely manner, thereby increasing the possibility of maintaining a normal state.

[0013] Safety-II recognizes that systems are becoming increasingly complex, including nonlinear interactive phenomena, and focuses on the fact that deterioration, damage, failures and troubles that occur in infrastructure systems cannot be avoided by conventional measures alone, such as removing the factors that lead to danger and risk.

[0014] Availability improvement (resilience management) should be understood as the state in which a system works as well as possible. The aim of the approach to availability improvement is not only to prevent a system from failing, but rather the concept of ensuring that a system works well.

[0015] Based on the concept of Safety-II, it is expected that resilience management (analysis, visualization, control) for emergent behavior and phenomena of complex systems will be realized. However, specific methodologies using digital twins, for example, have not yet been systematized.

[0016] On the other hand, quantum-classical hybrid computation methods of FEM (Finite Element Method) based on variational quantum algorithms, quantum-inspired computation methods, and quantum computing methods have been proposed and their effectiveness has been confirmed. However, specific applications other than large-scale and high-speed computation and specific methods of use in resilience management have not yet been clarified.

[0017] This embodiment provides a mechanism and methodology for specifically realizing the concept of Safety-II. Specifically, the present embodiment provides a mechanism and method for dealing with disturbances, changes, and opportunities so that they fall within a normal range by adjusting and controlling the functions (or modes) of a small number of features using the following units (a generation unit that generates sampling data in a multidimensional variable space, an analysis unit that analyzes complex system phenomena, a modeling unit that models a complex network including features, etc.).

[0018] Traditionally, risk-based safety management and crisis management efforts have been made to implement safety measures based on the concept of probabilistic risk (harm), such as hazard scenario analysis, risk scenario analysis, and resilience analysis. First, we will explain the problems involved in analyzing complex systems such as infrastructure systems. Infrastructure systems have complex interactions between parts, units (components or modules), and subsystems. The causes and effects of physical phenomena in the system are often not fully understood, and unpredictable events may occur. Uncertainty, ambiguity, and uncertainty are inherent in modeling complex systems.

[0019] Interactions are likely to occur when the following conditions occur: Common mode or common cause associations · Interrelated Subsystems Feedback Loop ·Indirect information Multiple related controls Limited understanding Limited ability to isolate faulty equipment Limited alternative supply sources

[0020] The combination of multiple hazards, design flaws, equipment failures, and operator errors may result in complex interactions. In complex systems, exogenous and endogenous factors may cause nonlinear phenomena (emergence and resonance phenomena) such as the Bullwhip Effect, self-similarity, phase transitions, metastability, and unexpected behavior, leading to a crisis.

[0021] It is becoming increasingly recognized that conventional technology cannot address safety issues related to highly complex social systems. In complex systems, the following phenomena can lead to accidents: Emergent phenomena in which order emerges in whole or in part from local interactions between elements (temporal and spatial pattern formation) Chaos, where complex and random behavior emerges from a nonlinear but simple deterministic mechanism, or pattern formation (e.g. dissipative structures) Fat tails, where low-frequency events occur more frequently than expected based on the law of large numbers, a fundamental theorem of probability statistics

[0022] In this way, nonlinear interactions between the elements that make up a complex system (system elements) and between the system and external loads can cause the system to become unstable, leading to the formation of a new overall order and resulting in accidents.

[0023] The state of a system or system elements cannot be distinguished as normal or abnormal, and these states are constantly changing. This change is an intrinsic part of the system and cannot be stopped or removed. In complex technological and social systems, small functional changes in many system elements can give rise to emergent phenomena and induce huge functional changes in the system as a whole. When such phenomena occur, multiple safety protection barriers can fail, leading to accidents that exceed expectations.

[0024] When improving the resilience of such complex infrastructure systems, there may be multiple measures that result in a desirable state (stable state) (multistable states).

[0025] Furthermore, in the emergent phenomenon of complex infrastructure systems, competing modes may exist in the early stages (premonition stage), and in the transition process the dominant mode (critical instability hypothesis) may gradually disappear and a stable mode may appear in the steady state, or instability may increase and a crisis may occur. For this reason, it is important to consider how to find premonitions before a crisis occurs, what data should be monitored to capture the premonitions, and how to reconstruct the infrastructure system to prevent instability from increasing.

[0026] New resilience concepts such as Safety-II have been proposed to deal with the above-mentioned situations. The new resilience concept is an extension of risk-based safety and crisis management, which takes safety measures based on the concept of probabilistic risk. In the new resilience concept, it is important to carry out "prediction," "monitoring," "response," and "learning" under an appropriate process approach and a systems approach based on hypothesis verification, prioritization, and optimization of resilience improvement measures using simulation. However, a specific methodology or framework for resilience analysis based on the process approach and systems approach has not yet been established.

[0027] Although resilience analysis that takes into account the interdependence of critical infrastructure has been conducted, it does not involve monitoring, physical simulation, chemical simulation, surrogate model, optimization, quantum-inspired computing, quantum-classical hybrid computing, or quantum computing. In addition, there is a large degree of uncertainty in the hazard model, and hazard scenarios, risk scenarios, and countermeasure scenarios are conducted in a limited range. For this reason, it is insufficient to verify hypotheses, prioritize, and visualize the effects of measures to improve resilience in complex infrastructure systems.

[0028] In this embodiment, measures to improve infrastructure resilience, such as decentralization, modularization, and reconfiguration, are taken from the perspective of infrastructure complex systems and hierarchical networks. This makes it possible to increase the resilience potential (flexibility, robustness) against disturbances. In this embodiment, a framework (methodology) is provided for hypothesis verification, prioritization, and visualization of the effects of measures to improve resilience.

[0029] The framework and components of the resilience analysis by the information processing system of this embodiment are shown below. In the following, resilience analysis and control includes not only the function of analyzing and controlling resilience, but also the function of visualizing the analysis result, the control method, or the control result on a display device or the like.

[0030] To achieve efficient resilience analysis of complex systems, at least one of the following points must be considered: (A1) The number of hazard scenarios (hazard scenarios, risk scenarios) and countermeasure scenarios will be enormous, and it will take a lot of time to prioritize and optimize the countermeasures. (A2) It is often not obvious what items should be monitored to detect signs of anomalies in complex systems and improve their resilience. (A3) There is significant uncertainty in modeling the hazards to which infrastructure systems are exposed.

[0031] The information processing system according to the present embodiment provides a framework for realizing more efficient resilience analysis of complex systems. This framework makes it possible to provide analysis and management functions as a service. Customers can use the service to perform resilience analysis without changing their current customer systems.

[0032] In the framework, it is preferable that analysis and management services such as IIoT (Industrial Internet of Things) are defined (published) as APIs (Application Programming Interfaces), and that the API specifications are industry standards such as OpenAPI and WSDL (Web Services Description Language). The framework of this embodiment may also have a function for synchronizing time information (trigger) and location information, for example. Location information can be synchronized by utilizing GPS (Global Positioning System), GIS (Geographic Information System), and GNSS (Global Navigation Satellite System), for example.

[0033] 1 is a block diagram showing an example of the configuration of an information processing system 41 according to the present embodiment. As shown in FIG. 1, the information processing system 41 includes a platform 100 as an information processing device, an edge terminal 200, a service providing device 300, and a common service 400.

[0034] Here, the edge terminal 200 may be realized in any of the following ways. · The actual sensing / monitoring / control terminal. An acquisition unit on a computer that acquires data such as earthquake observation data, meteorological observation data, traffic information, and satellite image data via a network. A virtual edge terminal (data generating unit or data setting unit) that virtually generates sensing / monitoring data on a computer, or sets assumed data by referring to a database. In addition, the function of the edge terminal 200 to transmit data (such as a first transmitting unit described later) may also be a virtual data transmission on a computer.

[0035] The edge terminal 200 and the platform 100 are connected by an interface that functions as an IoT bus, for example. The platform 100 and the enterprise service are connected via a service bus such as an API, for example. The IoT bus and the service bus may be realized by any method. For example, the IoT bus and the service bus may be realized by a network such as the Internet (which may be either wired or wireless).

[0036] The information processing system 41 may be connected to an external system 42 via a network such as the Internet. The external system 42 is a system that operates in cooperation with the service providing device 300.

[0037] 1, only one each of the edge terminal 200, the platform 100, the service providing device 300, and the external system 42 are shown, but a plurality of each may be provided. The entire information processing system 41 can be interpreted as corresponding to a CPS (Cyber ​​Physical System). Also, each of the service providing device 300 and the external system 42 can be interpreted as corresponding to a CPS system.

[0038] Furthermore, the platform 100 may be physically configured by one device, or may be physically configured by multiple devices. For example, the platform 100 may be constructed on a cloud environment. Furthermore, each unit in the platform 100 may be provided in a distributed manner across multiple devices.

[0039] The edge terminal 200 includes a control unit 210 that controls various processes of the edge terminal 200. For example, the control unit 210 controls the following processes. -Collection of monitoring data and other data Communication with an external device such as the platform 100 (first transmission unit) Data analysis (including data analysis, etc.) Data conversion Control of operations based on control information transmitted from the platform 100, etc.

[0040] For example, the control unit 210 controls a function of transmitting monitoring data (measurement data, monitor data) indicating the state of a target system that is an analysis target (control target) to the platform 100 (first transmission unit).

[0041] When monitoring (measurement, supervision) is not performed, the edge terminal 200 may be included in the platform 100.

[0042] The platform 100 includes a storage unit 110, an analysis unit 120, and an operation unit 130. The storage unit 110 stores data collected by the edge terminal 200 and the like (hereinafter referred to as monitoring data), and master data. The master data is data related to, for example, the specifications of the edge terminal 200, the specifications of the usage environment, design drawings, maintenance history, and the like.

[0043] For resilience analysis, the master data includes, for example, configuration data indicating the configuration of the target system, data indicating measures that affect the target system, data indicating actions that affect the target system, and data indicating a hazard model that models hazards (failures) that occur in the target system.

[0044] The storage unit 110 can be configured with any commonly used storage medium such as a flash memory, a memory card, a RAM (Random Access Memory), a HDD (Hard Disk Drive), and an optical disk. The storage unit 110 may be realized by a plurality of storage units that respectively store the monitoring data and the master data, for example. The plurality of storage units may be physically different storage media, or may be realized as different storage areas of the same physically storage medium.

[0045] The analysis unit 120 analyzes the monitoring data stored in the storage unit 110. For example, the analysis unit 120 executes a resilience analysis using the monitoring data. The function of executing the resilience analysis does not need to be provided in the analysis unit 120, and it is sufficient that the function is provided at least in the platform 100.

[0046] The operation unit 130 generates control information for controlling the edge terminal 200 based on the analysis result by the analysis unit 120, and operates the edge terminal 200 by transmitting the generated control information to the edge terminal 200. The control information may be any information that changes the state of the edge terminal 200. For example, the control information is information for causing the edge terminal 200 (control unit 210) to execute communication, analysis in the edge terminal 200, state change, sensing, operation instructions, and the like.

[0047] At least a part of the above-mentioned units (analysis unit 120 and operation unit 130) may be realized by one processing unit. The above-mentioned units are realized by, for example, one or more processors. For example, the above-mentioned units may be realized by having a processor such as a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit) execute a program, that is, by software. The above-mentioned units may be realized by a processor such as a dedicated IC (Integrated Circuit), that is, by hardware. The above-mentioned units may be realized by a combination of software and hardware. When multiple processors are used, each processor may realize one of the units, or may realize two or more of the units.

[0048] The control unit 210 of the edge terminal 200 transmits monitoring data to the platform 100 and receives control information from the platform 100. The control unit 210 performs control within the edge terminal 200 based on the control information. Based on the changed state of the edge terminal 200, the control unit 210 transmits further monitoring data to the platform 100 and receives control information from the platform 100. By repeating such a loop (CPS loop), control of a large amount of edge terminals 200 is realized based on the monitoring data accumulated in the platform 100 and the analysis results of the monitoring data.

[0049] The service providing device 300 is a device that provides a service that uses at least one of the monitoring data and the analysis result. The service providing device 300 includes a service unit 310, a business unit 320, and an SoS (System of Systems) unit 330.

[0050] The service unit 310 is used for enabling a human (administrator) to check the monitoring data and analysis results accumulated in the platform 100, for a human to inspect the state of the CPS loop based on human wisdom, and for manually changing the state of the CPS loop.

[0051] For example, based on their expert knowledge, the administrator inspects and audits the CPS loop, detects the occurrence of anomalies, and detects errors in AI (Artificial Intelligence) caused by attacks (CPS loop anomalies), etc., to check whether the information processing system 41 is operating properly from a human perspective. If the administrator detects an anomaly in the CPS loop, they may troubleshoot the CPS loop by sending control information via the service unit 310, or by other methods.

[0052] The business department 320 provides services using the monitoring data, analysis results, etc. For example, the business department 320 provides services such as CRM (Customer Relationship Management), ERP (Enterprise Resource Planning), PLM (Product Lifecycle Management), and EAM (Enterprise Asset Management).

[0053] The SoS unit 330 includes a plurality of internal systems and performs cooperation between the plurality of internal systems, etc. For example, the SoS unit 330 orchestrates a CPS loop using information generated within the service providing device 300 and monitoring data and analysis results accumulated in the platform 100.

[0054] When multiple service providing devices 300 are provided, the information processing system 41 includes multiple SoS units 330. Note that one service providing device 300 may include multiple SoS units 330. Each of the multiple SoS units 330 includes multiple internal systems. The multiple SoS units 330 may cooperate with each other via the platform 100 or may cooperate with each other without the platform 100.

[0055] Each internal system may be any system that is managed and operated independently. For example, the internal system may be a CPS system, or a system that provides services using a cloud server or a MEC (Multi Access Edge Computing) server. The internal system may be a system that provides services equivalent to the business department 320.

[0056] The internal system receives information and requests from the edge terminal 200 and the like, and returns the received information and information according to the request. The internal system may accumulate information received from the edge terminal 200 and the like, and construct information (digital twin) that reproduces the real world in the cyber world.

[0057] The SoS unit 330 exchanges information with multiple distributed internal systems, and provides functions that cannot be realized by a single system. For example, in principle, each internal system performs control for optimization (partial optimization) in the area that can be controlled within its own system. In contrast, the SoS unit 330 unifies multiple internal systems, and performs control for optimization (overall optimization) of the entire multiple internal systems that it unifies. The SoS unit 330 is not limited to this, and may be, for example, a system that provides functions different from those of each internal system.

[0058] Each of the internal systems may be a system related to a single industrial domain (service). The SoS unit 330, for example, controls the internal systems that belong to a plurality of different industrial domains.

[0059] In addition, with regard to resilience analysis, the SoS unit 330 is used to confirm the results of resilience analysis targeting a complex system consisting of multiple internal systems it oversees, to specify the analysis conditions for the resilience analysis (configuration data, actions, hazard models, etc.), and to implement measures based on the results of the resilience analysis.

[0060] At least a part of each of the above units (service unit 310, business unit 320, and SoS unit 330) may be realized by one processing unit. Each of the above units is realized by, for example, one or more processors. For example, each of the above units may be realized by having a processor such as a CPU or GPU execute a program, that is, by software. Each of the above units may be realized by a processor such as a dedicated IC, that is, by hardware. Each of the above units may be realized by using a combination of software and hardware. When multiple processors are used, each processor may realize one of the units, or two or more of the units.

[0061] The common service 400 is a device that provides a service that can be commonly used by each unit (edge ​​terminal 200, platform 100, service providing device 300) in the information processing system 41. The common service 400 provides, for example, the following functions. Security features Logging function -Charging function

[0062] Next, a detailed description will be given of the functions of the analysis unit 120 of the platform 100. Fig. 2 is a block diagram showing an example of a functional configuration of the analysis unit 120. As shown in Fig. 2, the analysis unit 120 includes a setting unit 121, a generating unit 122, an analyzing unit 123, a modeling unit 124, and a communication control unit 125.

[0063] The setting unit 121 performs various settings for resilience analysis. For example, the setting unit 121 sets a scenario related to the target system. The setting unit 121 may perform these settings in response to an instruction from the service providing device 300 (SoS unit 330). When the target system is reconfigured, the setting unit 121 sets configuration data of the target system after reconfiguration.

[0064] Here, we will explain the scenario. Analysis, visualization, and control to balance various values ​​such as resilience management and circular economy of the target system, which is a complex system, are carried out after setting up a scenario.

[0065] A scenario is an instance of a use case, and represents the specific flow when the use case is executed. Each use case is expressed as a collection of multiple scenarios, and serves as the basis for extracting necessary classes and objects. Through scenarios and use case examples, the behavior of the target system can be expressed concretely. A scenario contains many variables, some of which are deterministic and some of which are random variables that follow a probabilistic model. Therefore, a scenario is represented as a multivariate probability distribution. For each scenario, there exists a multivariate probability distribution for the variable X=(x,y), which includes the variable x related to the input of the target system and the response variable y of the target system.

[0066] A scenario can be expressed, for example, by a sequence diagram, a class diagram, a collaboration diagram, a statechart diagram, etc. By using these, a multivariate probability distribution (a multivariate probability distribution model, or a multivariate probability model) can be obtained.

[0067] Use cases can describe how a target system should function and its external environment (actors) and can show the behavior of the target system. Use cases can clarify the boundaries between the external and internal parts of the target system and can also clarify customer requirements. An actor is not a part of the target system but represents a role played by the user of the target system. Actors can actively interact with the target system or passively receive information from it. Actors can be humans, hardware, and external systems.

[0068] A use case models an interaction between an actor and the target system, including event flow. A use case is initiated by an actor and performs some function in the target system. A use case is an abstraction of a unit of work that a user of the target system accomplishes with the target system. Collectively, all use cases represent all the ways the target system can be used.

[0069] An object, also called an instance, is something that actually exists in the real world and has state, behavior, and identity. An object's state is a specification of one of the situations in which the object can be. Behavior describes how an object acts and reacts to requests from other objects. Each object has an identity, and must be identified as a separate object even if its state is identical to another object.

[0070] A class is an abstract definition of a collection of objects with a common structure and common behavior. Classes can be described in three areas: First area: Class name Second area: Structure (a collection of attributes) The third area: behavior (set of operations)

[0071] Each of the multiple scenarios that can be set in this embodiment is expressed as a multivariate probability distribution for the variable x=(x,y) as described above. The variable x corresponds to multiple input data related to the target system, and is a variable related to analysis conditions such as load, structure, material properties, boundary conditions, and environmental conditions. Each scenario can include m analysis conditions (m is an integer of 1 or more), that is, m variables x. Setting a scenario includes part or all of setting the type of analysis condition (variable x) and setting the probability distribution of the analysis condition. Hereinafter, a multidimensional probability distribution (multivariate probability distribution) representing the distribution of each variable x (also called analysis condition x) may be referred to as a probability distribution PA (first probability distribution).

[0072] The generating unit 122 generates a plurality of multi-dimensional sampling data SDA (first sampling data) using the probability distribution PA. For example, the generating unit 122 samples the value of each variable x according to the probability indicated by the probability distribution PA, and generates sampling data SDA including the sampled values ​​of the variable x. Any method may be used to generate the sampling data using the probability distribution PA, but for example, a sampling method using the following calculation may be applied. Quantum-inspired computing Quantum computing calculations ·Lagrangian Monte Carlo calculation Hamiltonian Monte Carlo calculations ·Markov Chain Monte Carlo Calculation

[0073] The generating unit 122 generates sampling data of a variable space of variables related to analysis conditions in the multidimensional variable space. The multidimensional variable space means a variable space related to analysis conditions and responses to the analysis conditions. The sampling data of the variable space related to responses in the multidimensional variable space is generated by the analyzing unit 123.

[0074] The generating unit 122 may further calculate the occurrence probability of an emergent phenomenon (including a rare event) in a complex system. The occurrence probability represents, for example, the ratio of the number of sampling data SDA whose corresponding response variable y (output data) satisfies a specific condition to the number of multiple generated sampling data SDA. The specific condition is, for example, a condition in which the response variable y exceeds a predetermined criterion (standard). The criterion is represented by a threshold for the occurrence of the emergent phenomenon, or a curved surface and a function representing the criterion in a multidimensional variable space. The occurrence probability can be used for controlling the target system, etc.

[0075] The analysis unit 123 executes an analysis process to obtain a response variable y of the target system using the sampling data SDA. For example, the analysis unit 123 inputs the multiple sampling data SDA generated by the generation unit 122 as input data, and obtains multiple output data (response variables y) representing physical quantities of the target system. Obtaining the response variable y corresponds to generating sampling data of a variable space related to the response.

[0076] In the following, the sampling data in the multidimensional variable space including the sampling data of the variable x related to the analysis conditions and the sampling data of the response variable y is referred to as the sampling data SDB (second sampling data). In addition, in the following, a plurality of sampling data SDBs may be referred to as a sampling data set.

[0077] The analysis unit 123 executes the analysis process, for example, by the following method. The analysis process by the analysis unit 123 will be described in detail later. -Analysis of phenomena using hybrid quantum-classical calculations -Analysis of phenomena using quantum-inspired computation -Analysis of phenomena using quantum computing -Phenomenon analysis using surrogate models

[0078] The modeling unit 124 generates a network model of a complex system (hereinafter, referred to as a complex network) corresponding to the target system by using the sampling data set. The complex network is, for example, a network model that represents the relationship between multiple nodes corresponding to the sampling data set.

[0079] For example, the modeling unit 124 first extracts a plurality of feature quantities that represent the characteristics of the sampling data set. The modeling unit 124 generates a complex network that represents the relationships between a plurality of nodes, including the nodes that correspond to the extracted feature quantities. The model generation process by the modeling unit 124 will be described in detail later.

[0080] The functions of the generating unit 122, the analyzing unit 123, and the modeling unit 124 allow a complex network including a feature of a complex system phenomenon to be generated. The generated complex network can be used for controlling a target system. For example, by controlling the extracted feature as a control variable, it is possible to control a target system including a large number of variables so that it operates normally with only a small number of feature. In this way, in the present embodiment, even in the case where a plurality of phenomena are interdependent, it is possible to suppress a plurality of responses within a desired range by adjusting the feature after unraveling the interdependence of the plurality of responses. The system may be configured to control some variables using the feature as an index. For example, the system may be configured to identify variables that strongly affect the feature and control the identified variables.

[0081] The communication control unit 125 controls communication with an external device such as the service providing device 300. For example, the communication control unit 125 transmits information indicating the analysis result of the analysis process to the external device such as the service providing device 300.

[0082] Next, a detailed description will be given of the functions of the SoS unit 330 of the service providing device 300. Fig. 3 is a block diagram showing an example of a functional configuration of the SoS unit 330. As shown in Fig. 3, the SoS unit 330 includes an output control unit 331 and a management unit 332.

[0083] The output control unit 331 controls the output of various data processed by the service providing device 300. For example, the output control unit 331 outputs information that visualizes the analysis result based on the analysis result of the resilience analysis. Any method for outputting the information may be used, and for example, a method of displaying the information on a display device and a method of transmitting the information to an external device connected via a network can be applied.

[0084] The management unit 332 executes management of the complex system according to the analysis results of the resilience analysis. For example, the management unit 332 manages the execution of a measure determined from the analysis results. As described above, the measure is, for example, reconfiguration of the complex system network.

[0085] Next, the model generation process performed by the modeling unit 124 will be described in detail.

[0086] First, the modeling unit 124 extracts a sampling data set from the sampling data set of the multidimensional variable space in accordance with the setting information. The setting information includes, for example, the following information. -Target areas of analysis in complex systems -Phenomenon response of the target area (response variable y) Scenario or analysis conditions Number of sampling data

[0087] The target area refers to an area that includes nodes that are targets of analysis among the components (hereinafter also referred to as nodes) included in the target system.

[0088] The modeling unit 124 uses the extracted sampling data set as an input and performs modeling of a complex network including feature quantities (hereinafter, complex network modeling).

[0089] FIG. 4 is a diagram showing an example of a sampling data set that is an input for modeling. The scenario SN1 corresponds to a scenario set by, for example, setting information. The scenario SN1 in FIG. 4 is a scenario set by m analysis conditions x1 to x m Therefore, setting the scenario SN1 includes the analysis conditions x1 to x m This can be interpreted as being equivalent to setting the analysis condition x i (i is an integer satisfying 1≦i≦m) Each of these is associated with a parameter that represents a multivariate probability distribution (such as a probability model parameter for x in a multivariate probability model).

[0090] In Figure 4, the response variables y1(t) to y rAmong these, the response variable y set by the setting information is extracted as the response variable data to be included in the sampling data set. For example, as the phenomenon response of the target area, the phenomenon response related to the representative (major) node (representative node) among the r nodes is set. Note that t(t0 to t s1 The sampling data D1, D2, etc. indicate the time at which the sampling data was sampled. A set number of sampling data are extracted from a plurality of sampling data including the sampling data D1, D2.

[0091] Thus, the sampling data set is a data set of temporal and spatial physical quantities related to the phenomenal response of the complex system to various scenarios and analysis conditions. The phenomenal response of the node (response variable y) is, for example, a variable vector (e.g., current, temperature, load, etc.) or tensor (e.g., stress) composed of multiple physical quantities.

[0092] The scenario may be set, for example, as follows: -Whether to include probability models (analysis conditions) for rare events such as strong winds and large earthquakes as hazards in addition to loads during normal operation -Whether to include a probabilistic model of degradation events such as a decline in system cooling performance as a degradation scenario -Whether to include a probability model for the timing of part replacement as a maintenance condition

[0093] The setting information for extracting a sampling data set that is an input for complex network modeling is set in advance, for example. The target area of ​​the analysis (phenomenal response of the target area) may be set by a relationship analysis, causal / correlation analysis, categorization, classification, clustering, hierarchical organization, and ranking method (e.g., Non-Patent Document 9) using AI such as a graph neural network (GNN), machine learning, a probabilistic method, and a statistical method.

[0094] Next, an example of a procedure for extracting physical quantities corresponding to representative nodes from a sampling data set (a data set of temporal and spatial physical quantities related to a phenomenon response) will be described.

[0095] First, a set of analysis conditions to be targeted and a range of sampling data to be targeted, i.e., a range in the (x, y) data set, are set. The modeling unit 124 analyzes the causal correlation structure while spatially or temporally categorizing (clustering) or hierarchizing the physical quantity data set of all nodes included in the set range, and extracts nodes with large response changes and nodes with large associations as representative nodes.

[0096] As a method for analyzing causal correlation structures, for example, the following models and algorithms can be used. Machine learning algorithms such as graph neural networks Statistical probability models such as Bayesian and Markov models

[0097] When representative physical quantities suitable for predicting and controlling a phenomenon are known in advance, modeling section 124 may extract representative nodes based on information (setting information) obtained in advance.

[0098] The modeling unit 124 may use the extracted representative node to generate a sampling data set to be used as an input for complex network modeling. For example, the modeling unit 124 extracts physical quantities of the representative node from a small number of sampling data sets. The modeling unit 124 generates a large number of sampling data sets for the physical quantities of the extracted representative node using the generation unit 122 and the analysis unit 123. The modeling unit 124 uses the sampling data sets generated in this manner as an input for complex network modeling.

[0099] For example, the modeling unit 124 selects a data set related to the physical quantities of the representative node that can express the characteristics of the phenomenon from a small number of sampling data sets (e.g., 1000 samplings), and then generates a large amount of sampling data (e.g., 1 million samplings) targeting the representative node, and uses this as input for complex network modeling.

[0100] Next, the details of the complex network modeling will be described. The modeling unit 124 generates a complex network including features such that the multivariate probability distribution of the input sampling data set is consistent with the multivariate probability distribution generated from the features.

[0101] For example, the modeling unit 124 extracts a plurality of feature quantities (feature quantity vectors, latent variable vectors) that represent the characteristics of the sampling data set. Any method may be used to extract the feature quantity vectors, but for example, a method using a generative AI algorithm such as VAE (Variational Auto-Encoder Bayes) and a flow-based generative model may be applied.

[0102] 5 is a diagram for explaining an example of a method for extracting a feature (feature vector z). The modeling unit 124 inputs a sampling data set (x, y) to an encoder 501, and obtains an output feature z. The encoder 501 is represented by a probability distribution p(z|x, y). The modeling unit 124 inputs the feature z to a decoder 502, and obtains an output sampling data set (x', y'). The decoder 502 is represented by a probability distribution p(x, y|z).

[0103] At this time, the modeling unit 124 extracts the feature vector z so that the probability distribution of the sampling data set (x', y') generated from the feature vector z is consistent with the multivariate probability distribution p(x, y) of the original input sampling data set (x, y).

[0104] That is, the modeling unit 124 generates a sampling data set including multiple sampling data (third sampling data) based on the feature vector z, and extracts the feature vector so as to optimize an index indicating whether a multidimensional probability distribution (second probability distribution) representing the distribution of the sampling data set is consistent with the multivariate probability distribution p(x, y).

[0105] The modeling unit 124 may gradually increase the number of elements of the feature vector z until the probability distributions are consistent, but in many cases, control is easier when the number of elements is as small as possible.

[0106] The modeling unit 124 generates a complex network that represents the relationship between a plurality of nodes including a node corresponding to the extracted feature (feature vector z). Any method may be used to generate a network model (complex network), but for example, a method of generating a network model in the following format may be applied. Node-link (also called edge) model (including the strength of association of links) AI model Machine learning models Statistical Probability Model

[0107] The complex network may be a network that changes over time.

[0108] In this embodiment, the complex network can be interpreted as a network model for expressing the transmission path, association, or pattern formation of physical quantities (force, heat, electric current, voltage, order variable, etc.) related to complex system phenomena. The complex network of this embodiment is a network model including feature quantities that govern complex system phenomena.

[0109] For example, when a differential equation expressing a complex system phenomenon is analyzed by a numerical analysis method based on the finite element method, each element, each node, and each integration point of the finite element can be a node. The phenomenon response of the complex system includes the physical quantity of each element, the physical quantity of each node, and the physical quantity of each integration point. It may be expressed as any of physical quantities, scalars (temperature, charge, etc.), vectors (load, displacement, current, etc.), and tensors (stress, etc.).

[0110] The interactions of physical quantities at each node can be calculated by performing correlation analysis on data showing temporal and spatial changes. By connecting links with a certain correlation value or higher, a complex network (node-link model) between each element (node) can be generated. The degree of a node (number of links) can also be calculated. For example, a node where physical quantities are concentrated will have a high degree, and a node (concentration area) that is affected not only by local causes but also by global influences will have an even higher degree.

[0111] Complex networks can be classified into sub-networks (cluster classification) or hierarchically organized by using community identification and classification methods.

[0112] The modeling unit 124 may have a function of calculating an index related to a complex network. The index can be used, for example, in management using a complex network. The index indicates the characteristics and rules (robustness against disturbances, flexibility in reconfiguration, scalability, etc.) of a complex network in which nodes are connected by links. Examples of calculation indexes for complex networks include the following (see Non-Patent Document 10). Extraction of order and control variables for pattern formation in complex systems - Calculation of performance degradation, deterioration, damage, and probability of breakage Scale-free metrics Dynamic capabilities (e.g., robustness to disturbances and ease of reconfiguration or reconstruction)

[0113] The modeling unit 124 may further include the following functions. A function that uses a portion of the generated sampling data to calculate the probability distribution of criteria functions (surfaces in a multivariable space with phenomena responses as variables) related to cumulative damage, deterioration, and abnormal occurrence in life prediction. - Function to calculate the probability of damage, deterioration and abnormal occurrence using the above probability distribution

[0114] The calculated occurrence probability (including the occurrence probability of rare events) can be used as an indicator for resilience management, such as extending life spans.

[0115] 6 is a diagram showing an example of the configuration of a complex network generated (modeled) by the modeling unit 124. As shown in FIG. 6, the complex network includes nodes corresponding to multiple response variables y included in the input sampling data, and links (lines connecting the nodes) indicating relationships between the multiple nodes. Of the multiple nodes, nodes 601 to 604 are nodes that are connected to more other nodes, and correspond to nodes where physical quantities are concentrated, for example. In the example of FIG. 6, nodes 601 to 604 correspond to response variables y1, y2, y3 and y r Corresponds to.

[0116] In addition, in a complex network, the relationship between the response variable y and a feature quantity z extracted from a sampling data set (x, y) related to a variable x corresponding to an analysis condition and a response variable y is also represented. In FIG. 6, for example, a node 601 (response variable y1) is represented by a feature quantity z extracted from a sampling data set (x, y) related to the response variable y. m , and is shown to be related to feature z1.

[0117] Fig. 7 is a diagram showing an example of the configuration of a complex network generated based on the classification of a sampling data set. Fig. 7 shows an example of classifying a sampling data set according to analysis conditions x1 and x2.

[0118] For example, for analysis condition x1, a response variable y at a representative node is selected from the sampling data set corresponding to analysis condition x1 by categorization using a graph neural network, and a complex network 701 including a node corresponding to the selected response variable y is generated.

[0119] Similarly, for analysis condition x2, a response variable y at a representative node is selected from the sampling data set corresponding to analysis condition x2 by categorization using a graph neural network, and a complex network 702 including a node corresponding to the selected response variable y is generated.

[0120] Next, a specific example of a generated complex network will be described. In the following, an example will be described in which the target system is a battery storage system. In other words, a case in which complex system phenomena (electrical circuits for charging and discharging, and thermal circuit network phenomena for heating and cooling) in a battery storage system are controlled will be described.

[0121] Fig. 8 is a diagram showing an example of the configuration of a storage battery system and an example of a complex network generated for this storage battery system. As shown in Fig. 8, the storage battery system includes a plurality of storage battery modules (storage battery modules 801a, 801b, etc.), a battery control panel 802, a cooling control panel 803, and air-cooling fans 804a to 804c.

[0122] The storage battery modules are electrically connected to each other by connection lines 811 and the like. Furthermore, each storage battery module is electrically connected to a battery control panel 802 by connection lines 812. Each storage battery module includes a large number of battery cells. The cells are also connected in series and parallel. Current flows through each cell when it is charged or discharged, and the cell generates heat. For this reason, the current and temperature of the cells have temporal and spatial distributions within the storage battery module and the storage battery system. Furthermore, the degree of deterioration of each cell differs depending on the charge / discharge waveform and temperature distribution.

[0123] The temperature distribution also depends on the cooling performance and environment within the battery system. When analyzing the charging / discharging behavior and temperature distribution phenomena of all cells in a battery system, the current, voltage, temperature, etc. of each node (each node or element) that has been discretized to numerically analyze the differential equations that describe the phenomena of each cell and each battery module become physical quantities that represent the state.

[0124] Rather than performing complex network modeling using the physical quantities of all nodes as response variables, it is more efficient to perform complex network modeling by extracting the physical quantities of representative nodes that are suitable for predicting and controlling phenomena.

[0125] For example, as shown in Fig. 8, when temperature unevenness 821a-821c occurs in each storage battery module in the storage battery system, only the physical quantities of the nodes where the temperature unevenness occurs and there is a risk of cell deterioration, and the physical quantities related to them are extracted and complex network modeling is performed as response variables. Areas 822a-822c in the complex network shown on the right side of Fig. 8 correspond to areas (node ​​groups in the complex network) including nodes corresponding to multiple response variables corresponding to the temperature unevenness 821a-821c. The response variables corresponding to each node are, for example, temperature, current, and voltage. Analysis conditions x1-x m are connected by links to nodes corresponding to the associated response variables.

[0126] This section explains the procedure for extracting a representative node in a battery storage system. The following scenario and analysis conditions are set, and an example is given of a situation in which three temperature unevenness regions occur during the operation of the battery storage system. · Scenario: The cooling method is air-cooled fan cooling. Analysis conditions: The charge / discharge waveform pattern conditions, the environmental temperature conditions, and the system configuration conditions are analysis conditions x1 that are expressed by a multivariate probability model.

[0127] First, the modeling unit 124 generates a complex network that does not include the feature value z for each target analysis condition by using a classification and causal correlation analysis method such as a graph neural network for the sampling data related to the analysis condition x1.

[0128] The modeling unit 124 extracts representative nodes from the obtained complex network that can express the characteristics of the phenomenon using indicators such as the strength of the association between each node that is equal to or greater than a predetermined value, or the magnitude of the physical quantity of each node (absolute value, fluctuation range, etc.).

[0129] 8 (clustered into three regions 822a to 822c), the modeling unit 124 may extract a representative node for each region. Furthermore, the modeling unit 124 may generate a complex network so as to partially include a hierarchical network in which a plurality of nodes are newly expressed by a representative node.

[0130] For example, the modeling unit 124 may calculate the response (e.g., response variables y2, y5, y 10 ) is selected, and responses with strong correlations are selected by categorizing and analyzing correlations using a graph neural network, etc. This processing is performed for each analysis condition of each scenario in the set scenarios and analysis conditions, and the selected sampling data sets are collected and used as input data for complex network modeling including feature quantities.

[0131] Next, a description will be given of information processing by the platform 100. Fig. 9 is a flowchart showing an example of information processing by the platform 100 of this embodiment.

[0132] The setting unit 121 sets a scenario in response to, for example, a designation from a user or the service providing device 300 (SoS unit 330) (step S101). The generating unit 122 generates a plurality of sampling data SDA corresponding to the sampling data of the analysis conditions by using a probability distribution PA corresponding to the set scenario (step S102). The analyzing unit 123 generates sampling data of a response (response variable y) by an analysis process using the plurality of sampling data SDA (step S103).

[0133] The modeling unit 124 selects a target region and a physical quantity (phenomenon response) of the target region using a data set (sampling data set) of sampling data SDB including the sampling data SDA of the analysis conditions and the sampling data of the response (step S104). The modeling unit 124 generates a complex network using the selected target region and physical quantity (step S105), and ends the information processing.

[0134] (Variation 1) In the first modification, the scenario is changed and the generation of the complex network is repeated until the generated complex network satisfies the predetermined requirements.

[0135] In this modification, the function of the analysis unit is different from that of the above embodiment. Fig. 10 is a block diagram showing an example of the functional configuration of the analysis unit 120-2 of the modification 1. As shown in Fig. 10, the analysis unit 120-2 includes a setting unit 121, a generating unit 122, an analyzing unit 123, a modeling unit 124, a communication control unit 125, and an optimization unit 126-2.

[0136] Analysis unit 120-2 differs from analysis unit 120 of the above embodiment in that analysis unit 120-2 further includes optimization unit 126-2.

[0137] The optimization unit 126-2 changes the scenario so as to optimize the index of the network model generated by the modeling unit 124. Changing the scenario corresponds to changing the multivariate probability distribution (probability distribution PA) indicated by the scenario. As described above, setting the scenario includes setting the type of analysis condition. Furthermore, the analysis condition may include a system configuration condition. Therefore, changing the scenario includes, for example, changing the system configuration.

[0138] The generator 122 generates a plurality of sampling data SDA using the changed probability distribution PA. The subsequent functions of the analyzer 123 and the like are the same as those of the above embodiment.

[0139] Fig. 11 is a flow chart showing an example of information processing in this modified example. Steps S201 to S205 are similar to steps S101 to S105 in Fig. 9 showing the information processing in the above embodiment, and therefore a description thereof will be omitted.

[0140] In this modification, the optimization unit 126-2 judges whether the generated complex network satisfies a requirement (step S206). Whether the requirement is satisfied is judged, for example, as follows. If features can be extracted, it is determined that the requirements are met. If the calculated indicators such as the scale-free indicator meet the standard values, it is judged that the requirements are met. If the multivariate probability distribution of the sampling data set is consistent with the multivariate probability distribution generated from the features extracted during complex network modeling, it is determined that the requirements are met.

[0141] If the requirements are not satisfied (step S206: No), the optimization unit 126-2 changes the scenario (step S207). After that, the process returns to step S201, the changed scenario is set, and the process is repeated. If the requirements are satisfied (step S206: Yes), the information processing ends.

[0142] (Variation 2) Variation 2 is an example having a function of controlling a target system using the generated complex network, and a function of updating a multivariate probability distribution using monitoring data.

[0143] In this modification, the function of the analysis unit is different from that of the above embodiment. Fig. 12 is a block diagram showing an example of the functional configuration of the analysis unit 120-3 of the modification 2. As shown in Fig. 12, the analysis unit 120-3 includes a setting unit 121, a generating unit 122, an analyzing unit 123, a modeling unit 124, a communication control unit 125, a system control unit 127-3, and an updating unit 128-3.

[0144] Analysis section 120-2 differs from analysis section 120 of the above embodiment in that analysis section 120-2 further includes a system control section 127-3 and an update section 128-3.

[0145] The system control unit 127-3 controls one or more responses of the complex system by adjusting the values ​​of the feature quantities extracted by the complex network modeling. For example, the system control unit 127-3 controls the target system using the multiple feature quantities as control variables.

[0146] When the occurrence probability of an emergent phenomenon (including a rare event) is calculated by the generation unit 122 or the like, the system control unit 127-3 may control the target system so that the occurrence probability becomes a target value.

[0147] The update unit 128-3 updates a multivariate probability distribution (such as a probability distribution PA) by using monitoring data measured from the target system. For example, when there is monitoring data of a phenomenon related to the target system, the update unit 128-3 identifies a variable among the multiple variables that corresponds to the monitoring data by inverse analysis, and updates a preset probability distribution to a posterior distribution for the identified variable. The monitoring data does not need to be actual measurement data, and may be, for example, sampling data generated by simulation.

[0148] The function of controlling the target system using the generated complex network (system control unit 127-3) and the function of updating the multivariate probability distribution using the monitoring data (update unit 128-3) can be executed independently. Therefore, the analysis unit 120-3 may be configured to include either the system control unit 127-3 or the update unit 128-3.

[0149] Fig. 13 is a flow chart showing an example of information processing in this modified example. Steps S301 to S305 are similar to steps S101 to S105 in Fig. 9 showing the information processing in the above embodiment, and therefore a description thereof will be omitted.

[0150] In this modification, the system control unit 127-3 controls the target system using the feature quantities included in the generated complex network as control variables (step S306).

[0151] Furthermore, the update unit 128-3 uses the monitoring data to update the probability distribution to, for example, a posterior distribution (step S307).

[0152] Next, a detailed description will be given of the analysis process performed by the analysis unit 123. First, a configuration example of the analysis unit 123 when performing phenomenon analysis using hybrid quantum-classical calculation will be described.

[0153] In this case, the analysis unit 123 is configured to have the following two main functions. (F1) Variational Quantum Algorithm (F2) Optimization algorithms (using classical computation)

[0154] For example, the analysis unit 123 executes a quantum operation including an operation of decomposing a matrix based on a Hamiltonian generated from multiple multidimensional sampling data SDA generated by the generation unit 122 (F1 above), and obtains multiple output data by optimization processing on the results of the quantum operation (F2 above).

[0155] The above two functions will be further explained. In (F1), the analysis unit 123 uses the analysis conditions (load, structure, material properties, boundary conditions, environmental conditions, etc.) to generate a phenomenon analysis model including a mesh model (FEM mesh) by a numerical calculation method of differential equations such as the finite element method. The phenomenon analysis model is, for example, a model for inputting a variable x corresponding to the analysis conditions and outputting a phenomenon response (response variable y). The analysis unit 123 uses the phenomenon analysis model to generate a Hamiltonian and a matrix form of the Hamiltonian for quantum-classical hybrid calculation (Hamiltonian / matrix form generation unit). In general, a physical quantity such as a Hamiltonian can be expressed in a matrix form.

[0156] For example, in the case of a phenomenon expressed by continuum mechanics, the analysis unit 123 generates a phenomenon analysis model that outputs an energy functional by a method such as that described in Patent Document 2. E For example, the energy functional φ for each element of FEM expressed by the following equation (1) is Ee is calculated by taking the sum for all elements of the FEM as shown in equation (2).

number

number

[0157] In addition, u is displacement, σ is equivalent stress, ε is equivalent strain, F is the external force vector acting on the boundary, V is the volume of the object, and S is the surface area. The second term corresponds to the integral of the equivalent stress and the increment of the equivalent strain rate. The third term corresponds to the work done by the object against the external force (the product of the external force vector and the velocity).

[0158] The analysis unit 123 calculates a Lagrangian from the energy functional and generates a Hamiltonian by performing Legendre transformation on the Lagrangian. The method of generating the Hamiltonian is not limited to this, and any other method may be used.

[0159] Furthermore, the analysis unit 123 can generate a matrix form of Hamiltonian (stiffness matrix, displacement vector, load vector) by expressing a differential equation describing a phenomenon in a weak form using the finite element method and the Galerkin method.

[0160] The following equations (3) and (4) show examples of the generated matrix format, where K is the stiffness matrix, M is the mass matrix, u is the displacement vector, f is the load vector, and k is the time step.

number

number

[0161] Similar techniques can be applied to phenomena other than those that can be expressed by continuum mechanics. For example, differential equations that describe phenomena such as electrical circuits and thermal circuits can also be expressed in matrix form.

[0162] The matrix A(=M+Δt) in equation (4) is expressed by the mass matrix M and the stiffness matrix K. 2 If K) is not a symmetric matrix or a Hermitian matrix, it can be made symmetric by adding a transpose of the matrix as shown in the following equation (5).

number

[0163] Here, u' and f' are arbitrary because they do not affect the original matrix format. Although the symmetrized matrix increases in dimension, it is a sparse matrix, so it is possible to reduce it by utilizing a sparse solver.

[0164] In general, it is possible to find the eigenvalues ​​and eigenvectors of a symmetric matrix or a Hermitian matrix by diagonalizing the matrix.

[0165] Next, as a preprocessing step, the analysis unit 123 expresses the matrix in the matrix format as a quantum circuit by decomposing the matrix using a linear sum related to the tensor product of the Pauli operator. The analysis unit 123 tentatively obtains eigenvalues ​​and eigenvectors using the quantum circuit.

[0166] The process up to this point corresponds to the quantum operation (F1 above) including the operation of decomposing the matrix. After this, the analysis unit 123 executes the optimization process (F2 above) for the operation result by the quantum operation.

[0167] That is, the analysis unit 123 updates the eigenvalues ​​and eigenvectors using a classical optimization algorithm based on the principle of variations (an optimization algorithm using classical calculations) with an optimization index (also called an energy function, a value function, or a cost function), which is an index of the principle of variations, as an objective function.

[0168] The analysis unit 123 can obtain a response variable y, which is a phenomenon response, by a combination of the eigenvalue and the eigenvector. When a sampling data set (x, y) including the response variable y is accumulated up to a pre-specified number, the analysis unit 123 executes an inverse analysis algorithm.

[0169] The inverse analysis algorithm is a process for matching the sampling data with the monitoring data. For example, the analysis unit 123 repeatedly performs the analysis process while updating the probability distribution regarding the variable x until the sampling data matches the monitoring data. For example, similar to the process by the update unit 128-3 described above, the probability distribution is updated to a posterior distribution. The sampling data (sampling data including the response variable y) when it matches the monitoring data is output as the result of the analysis process.

[0170] 14 is a diagram showing an example of the flow of quantum-classical hybrid calculation. An index 1401 corresponds to the initial value of the optimization index. Calculation 1402 and calculation 1403 correspond to the variational quantum algorithm (F1 above) and the optimization algorithm (F2), respectively.

[0171] The variation parameter θ is a parameter set for the state of each quantum bit, J is the number of parameters, and L is an optimization index in the variational principle. The state quantity (unknown vector u) for the phenomenon of the complex system is extracted through a classical optimization calculation (calculation 1403) from an exhaustive combination set of quantum bit states.

[0172] Next, an example is shown below in which a matrix associated with a Hamiltonian (which can also be converted to a Lagrangian) relating to a phenomenon in a complex system is expressed as a quantum circuit by decomposing it using a linear sum of the tensor product of the Pauli operator.

[0173] In the following, an example of describing a string vibration problem using the Poisson equation and performing a numerical analysis using FEM will be described. More specifically, a case will be described in which a string is modeled one-dimensionally and divided into eight elements. FIG. 15 is a diagram showing an example in which a string modeled one-dimensionally is divided into eight finite elements. Note that φ1 to φ8 under the elements represent trial functions of the Galerkin method.

[0174] The equation that describes the behavior of a string is expressed as a one-dimensional wave equation, as shown in the following equation (6).

number

[0175] When the boundary conditions and initial conditions are as described in equation (6), the one-dimensional wave equation can be expressed in the weak form shown in equation (7) below using the trial function φ (and shape function) of the Galerkin method.

number

[0176] Furthermore, the weak form defines the displacement at time step Δt, k as u k By discretizing it with respect to time as

[0177] In the example of FIG. 15, the matrix A is an 8×8 dimensional symmetric matrix, and is expressed as M+Δt 2 It is represented by K.

[0178] When the weak form is discretized in the time direction using the Newmark-β method, it is expressed as the following equations (8) to (10). Here, when β=0, it becomes an explicit method using the central difference method, and when β=1 / 6, it becomes an implicit method using the linear acceleration method.

number

number

number

[0179] In the example of Fig. 15, since it is divided into 8 elements, h = 1 / 8, and the stiffness matrix K and mass matrix M are expressed as 9 x 9 dimensional symmetric matrices as shown in the following equations (11) and (12), respectively, where h indicates the unit spatial dimension of the finite element.

number

number

[0180] When the Dirichlet boundary condition u1 = u9 = 0 is applied, the stiffness matrix K and the mass matrix M are expressed as 8 × 8 dimensional symmetric matrices as shown in the following equations (13) and (14), respectively.

number

number

[0181] In general, a matrix A, which is a Hermitian matrix, can be expressed as in the following equations (15) and (16).

number

number

[0182] In equations (15) and (16), σ represents the Pauli matrix (Pauli operator). The symbol with a circle around a cross represents a tensor product. n represents the number of quantum gates. The size of the matrix, N, is N=2. n j i represents the type of gate (e.g. Pauli gates X, Y, Z and I). Tr represents the trace of the matrix.

[0183] Here, the stiffness matrix K and the Hermitian matrix A are divided into diagonal and off-diagonal terms A m If we consider these separately, they can be expressed by the following equations (17) and (18).

number

number

[0184] In this case, the diagonal terms can be expressed as constant multiples of the identity matrix I (in more complicated cases, we use the Pauli gate Z). m can be expressed as a combination of Pauli gates (specifically X or Y).

[0185] As an example, if we consider the case of two quantum bits, the off-diagonal term A2 can be expressed as a combination of Pauli matrices as shown in the following equation (19).

number

[0186] Similarly, the off-diagonal term A3 in the case of three quantum bits can be expressed as a combination of Pauli matrices as shown in the following equation (20).

number

[0187] In this way, by decomposing matrix A, which is a Hermitian matrix related to physical quantities such as a Hamiltonian capable of describing a phenomenon, into a product of Pauli matrices, it is possible to create a quantum circuit corresponding to matrix A. In this embodiment, the method of decomposing matrix A is not unique, and quantum-classical hybrid computation is performed while modifying the method of decomposing matrix A so that it converges through optimization by classical computation.

[0188] Fig. 16 is a diagram showing an example of a quantum circuit corresponding to the matrix A. Note that Fig. 16 shows an example of a quantum circuit corresponding to the off-diagonal term A2 in the case where there are two quantum bits.

[0189] For example, the off-diagonal term A2 can be expressed by the following equation (21) where the two quantum bits are |q0> and |q1>. The first, second, and third terms on the right-hand side of equation (21) correspond to the quantum circuit shown in FIG.

number

[0190] Any quantum gate can be described as a combination of Pauli matrices. Here, we use only the Pauli matrices to express the off-diagonal terms A of the Hermitian matrix A. m However, quantum circuits may be constructed using other quantum gates. In general, the Hadamard gate H and the controlled NOT (controlled X) gate C^ X , and the rotation gates R^ around each axis X , R^ Y , R^ Z etc. are used.

[0191] In order to calculate the cost function (optimization index), it is necessary to find the expectation value of each operator, and a quantum circuit called a Hadamard test is used for this purpose. Figure 17 shows an example of the configuration of a quantum circuit for the Hadamard test. Figure 17 shows an example of a quantum circuit for the Hadamard test that calculates the expectation value <ψ|U|ψ> for an arbitrary state |ψ> and unitary matrix U. The gate G in Figure 17 is a Hadamard gate H or a rotation gate R^ X (π / 2).

[0192] In the Hadamard test, in addition to an arbitrary state |ψ>, a test bit |0> is prepared to measure the expected value. The test bit is gated by a gate G (a Hadamard gate H when measuring the real part of the expected value, and a rotation gate R^ when measuring the imaginary part of the expected value). X (π / 2)) is applied, and then a controlled U gate with the test bit as the control bit is applied. Finally, a Hadamard gate H is applied to the test bit, and then measurement is performed. If the probability that the test bit is |0> is P0 and the probability that the test bit is |1> is P1, the expected value can be calculated using the following equation (22).

number

[0193] The analysis unit 123 uses the above-mentioned method to calculate the cost function and change the unknown vector u (displacement vector u) so as to minimize the value of the cost function, thereby obtaining a solution to the differential equation.

[0194] Next, an example of applying analytical processing to resilience management will be explained. In the following, resilience management of infrastructure systems (e.g., battery storage systems) will be explained as an example. The application of analytical processing is not limited to resilience management, and may be applied to any other processing. For example, it can be applied to analysis, visualization, and control for balancing multiple values ​​such as environmental friendliness, economic efficiency, and resilience in order to realize a circular economy.

[0195] In a large-scale storage battery system, tens of thousands of battery cells are connected in series and in parallel to form a circuit (network) of current and voltage (resistors, coils, capacitors), and also a circuit (network) of heat.

[0196] The time and space waveform patterns of charging and discharging vary depending on the conditions of use and cooling performance. Temperature unevenness may occur between cells depending on the cooling performance. Battery cell deterioration is highly dependent on the number of charge / discharge cycles, SOC (State Of Charge), C rate, and temperature.

[0197] Potential configurations and scenarios for the target battery storage system include combinations of the following items. ·Structure / composition / layout conditions Cooling method and cooling performance conditions Various current or voltage waveform patterns -Various temperature waveform patterns depending on cooling performance - Correlation between charge / discharge conditions and temperature conditions Percentage of charge remaining when charging Conditions for stopping charging before the battery is fully charged - Conditions for the period of time between charging and discharging Timing of maintenance and renewal, such as reuse / refurbishment / repurposing

[0198] In this embodiment, a huge number of possible scenarios are categorized, and for each scenario group, the target system is controlled with a small number of variables to suppress temperature unevenness and cell deterioration of the battery, thereby extending its life. The scenario groups can be categorized into, for example, scenarios in which regular maintenance and updates are performed and those in which they are not performed, or scenarios in which the cooling method is air-cooled and scenarios in which the cooling method is water-cooled, etc.

[0199] Fig. 18 is a diagram showing the correspondence between a mechanical vibration model and an electrical vibration model. The mechanical vibration model is a model to which differential equations based on continuum mechanics can be applied, for example, a mass-spring-damper vibration model. As shown in Fig. 18, since there is a correspondence between the mechanical vibration model and the electrical vibration model, the above-mentioned numerical solution method for differential equations related to continuum mechanics can also be applied to the analysis of electrical circuits.

[0200] For example, the differential equations of each circuit network (current and voltage circuit network, thermal circuit network, etc.) included in the battery storage system can be expressed in a matrix format as shown in the above formula (4). In the case of a battery storage system, each element of the matrix format corresponds, for example, as follows: Matrix A: Resistor, coil, capacitor (capacity) Vector u: charge or current Vector f: voltage

[0201] Equations expressed in matrix form can be condensed using orthogonal eigenexpansion. The following is an example of applying orthogonal eigenexpansion to a second-order differential equation for a multi-mass system, and speeding up the process by condensing it.

[0202] The equation of motion for damped forced vibration with multiple degrees of freedom is expressed as the following equation (23). Solving equation (23) requires calculation of the inverse matrix, which takes a long time. Here, u(t) is the displacement, [M] is the mass matrix, [C] is the damping matrix, [K] is the stiffness matrix, and F(t) is the time-dependent excitation load.

number

[0203] The actual displacement u(t) is transformed using a transformation matrix [φ] as shown in the following equation (24), where ξ(t) is the condensed displacement.

number

[0204] The transformation matrix [φ] is defined as the following equation (25). Here, {φ j}(j=1, ,N) is a feature vector. The equation to be solved is expressed as the following equation (26).

number

number

[0205] Here, a case will be described in which an eigenvector obtained by eigenvalue analysis as shown in the following formula (27) is used as the feature vector, where ω is the natural frequency.

number

[0206] {φ j} is the eigenvector corresponding to the j-th eigenvalue obtained by eigenvalue analysis. Using this eigenvector, the stiffness matrix and mass matrix are diagonalized. The damping matrix is ​​generally not diagonalized, but if we assume that it is diagonalized, the following equation (28) holds.

number

[0207] Using these results, the matrix of the equation of motion is reduced to only diagonal components, and the equation of motion corresponding to the j-th eigenmode is expressed by the ordinary differential equation shown in the following equation (29): j , ω j represent the modal damping ratio and natural frequency obtained from the following equation (30), respectively.

number

number

[0208] When the initial condition is expressed by the following equation (31), equation (29) can be expressed as the following equation (32) using Duhamel integral (convolution integral). ω d is the damped natural frequency defined by the following equation (33).

number

number

number

[0209] Equation (32) can be expanded as shown in the following equations (34) to (37).

number

number

number

number

[0210] The analysis unit 123 calculates the obtained ξ j The displacement u(t) can be calculated by substituting into equation (24). In the absence of damping, ζ j = 0, and integration is easy. By simplifying using the feature vector in this way, it becomes possible to solve differential equations more quickly.

[0211] The Duhamel integral may not be computed efficiently. In such cases, Nigam's method may be used instead of the Duhamel integral.

[0212] If the damping matrix is ​​not diagonalized, the error will be large. In circuit simulation, if inductance can be ignored compared to resistance and capacitance, the equivalent circuit can be described by a first-order differential equation such as the following equation (38).

number

[0213] (38) In equation (38), C -1 Multiplying by , we obtain the following equation (39): The formal solution of equation (39) is expressed by the Duhamel integral shown in the following equation (40).

number

number

[0214] The eigenvalue λ calculated from the following equation (41) i , eigenvector φ i Using (1≦i≦g), the following equation (42) holds.

number

number

[0215] Using g eigenvalues, equation (42) can be expressed as equation (43), where Λ=diag(λ1, , λ g ), Φ=[φ1, ,φ g ].

number

[0216] Using eigenvectors, the part related to the exponential function in equation (40) can be calculated as shown in equation (44) below.

number

[0217] Φ i = [0, ,φ i , , 0]Φ -1 Then, equation (40) can be simplified as shown in the following equation (45), and can be solved more quickly.

number

[0218] In the second-order differential equation, when the effect of the off-diagonal components of the damping matrix is ​​large and the stiffness matrix and mass matrix cannot be ignored, the state equations are defined as shown in the following equations (46) to (48).

number

number

number

[0219] These state equations can also be simplified using Duhamel integrals, making it possible to solve them more quickly.

[0220] In addition, the thermal network method allows the heat transfer path to be replaced with an equation similar to that of an electric circuit, so the above method can also be applied to heat transfer calculations.

[0221] Next, an example of controlling a battery storage system will be described using a complex network as shown in Fig. 6. In this case, in the complex network of Fig. 6, for example, the input variable x (analysis condition variable x) is the usage condition (cooling performance condition, etc.) and the load waveform of the battery storage system, and the response variable y is the current, voltage, and temperature.

[0222] At this time, the feature value z 1、 If z2 can be extracted, the feature z1、 By focusing on nodes (nodes 601 to 604 in FIG. 6) that have a high correlation with z2, it is possible to control the input variables x related to these nodes to suppress temperature unevenness. This makes it easier to manage the battery to suppress deterioration and extend its lifespan.

[0223] From the relationship between the input variable x and the feature quantity z, for example, the feature quantity z1 can be interpreted as a feature quantity for suppressing temperature unevenness, and the feature quantity z2 as a feature quantity related to the power consumption due to the operation of the cooling method device. This makes it possible to efficiently implement management that achieves both a longer lifespan by suppressing temperature unevenness and a minimized carbon footprint by suppressing power consumption.

[0224] The relationship between the extracted feature value z, the input variable x, and the response variable y may be expressed, for example, by a causal correlation model or by a probability distribution only. Below, we will explain examples of how to control the response variable y for each of these two cases.

[0225] When the relationship between the feature amount z, the input variable x, and the response variable y is expressed by a causal correlation model, a differential equation relating to the feature amount z and the input variable x can be obtained. Therefore, for example, the system control unit 127-3 can efficiently control the response variable y by using the input variable x by controlling the feature amount z by using the input variable x.

[0226] The control methods used in this case include PID (Proportional-Integral-Differential) control, model predictive control, and consensus control, depending on the number of response variables y to be controlled and the number of input variables x that can be used as control variables.

[0227] For example, simple PID control is used when controlling one response variable y with one input variable x as a control variable. In this case, more efficient control is possible by selecting the node of the response variable y to be controlled based on the feature value z extracted in advance.

[0228] Model predictive control is also used when multiple response variables y are controlled by a small number of input variables x. Using model predictive control, it is possible not only to control complex dynamics consisting of multiple response variables y, but also to directly handle restrictions such as operating the response variable y within a certain range. It is also possible to use the cell's lifespan as an objective function. In this case, too, efficient control is possible by selecting the node of the response variable y to be controlled based on the feature value z.

[0229] Consensus control is used when multiple response variables y are controlled by multiple input variables x. Consensus control is particularly effective when controlling the entire battery system by controlling a large number of cells individually. In this case, efficient control is possible by using the relationship between the feature value z and the response variable y to design the control rules for each cell.

[0230] Furthermore, in the case of control aimed at maximizing the lifespan of the entire battery system based on the waveforms of each cell, it is possible to use model predictive control to maximize the lifespan. In this case, by using the relationship between the extracted feature value z and the response variable y, the model used in model predictive control can be described in terms of feature value z instead of response variable y. This reduces the dimensionality of the model, enabling efficient control.

[0231] Next, an example will be described in which the relationship between the feature quantity z, the input variable x, and the response variable y is expressed only by a probability distribution model. In this case, the operating range of the response variable y can be set using the feature quantity z as an index. By determining the desired range of the feature quantity z and sampling the response variable y, a desired operating range for the response variable y can be obtained. Furthermore, by controlling the response variable y using the input variable x so that it falls within this operating range, it is also possible to indirectly control the feature quantity z.

[0232] In this case, too, PID control, model predictive control, consensus control, and the like can be applied depending on the number of response variables y and the number of input variables x to be controlled. For example, when controlling temperature unevenness, the operating range for the temperature distribution of the response variable y can be obtained by sampling using the feature value z related to the temperature unevenness. Then, the temperature unevenness can be controlled by controlling the temperature distribution of the response variable y to be within the obtained operating range using the input variable x, such as the fan rotation amount.

[0233] In this way, efficient control of the battery storage system is possible by using the relationship between the extracted feature value z and the response variable y.

[0234] Next, an example of applying this embodiment to resilience management of the drive train structure of a wind power plant will be shown.

[0235] For example, when a wind power plant is subjected to wind action and seismic loads, the combination of the load waveform and the axial misalignment due to the installation tolerance may induce an emergent excessive load waveform. The complex network showing the transmission path of the physical quantity (such as force) in the drive train may be a random network or a free-scale network depending on the scenario of the combination of the system configuration, structure, load condition, and installation tolerance. The complex network may also be other small networks. The physical quantities may be pre-processed by assuming a pseudo-period in the waveforms of the wind and seismic loads and representing them by the physical quantities at each phase in the pseudo-period.

[0236] Fig. 19 is a diagram showing an example of a random network. Fig. 20 is a diagram showing an example of a free scale network. Node 1901 in Fig. 19 corresponds to a bearing which is a location where damage is suspected. Nodes 2001 to 2003 in Fig. 20 correspond to representative nodes having a high correlation with feature amounts, for example. Node 2011 corresponds to a bearing which is a location where damage is suspected.

[0237] The change in damage probability over time varies greatly depending on the form of the complex network. Figure 21 shows an example of the change in damage probability over time when the complex network is a random network and when it is a free-scale network. The horizontal axis of the graph at the top of Figure 21 represents the elapsed time (years, etc.), and the vertical axis represents the predicted damage probability.

[0238] The graph at the bottom of Fig. 21 shows a graph in which the horizontal axis is cumulative damage value and the vertical axis is probability density. As shown in the graph at the bottom, in the case of a random network, the fat tail phenomenon may occur.

[0239] This embodiment can be applied to the drive train structure of a wind power plant having such characteristics when performing resilience management to reduce downtime.

[0240] For example, when controlling a drivetrain structure using a complex network as shown in Figure 6, the input variable x (analysis condition variable x) is the load waveform of wind action and earthquake load, and the response variable y is the stress generated in the drivetrain structure.

[0241] At this time, the feature value z 1、 If z2 can be extracted, the feature z 1、 By focusing on nodes that have a high correlation with z2 (nodes 601 to 604 in FIG. 6) and adjusting these nodes (for example, by installing a buffer structure to adjust the excessive load), it is possible to buffer the excessive load and easily improve availability.

[0242] Next, a configuration example of the analysis unit 123 when performing phenomenon analysis using quantum-inspired calculation or quantum computing will be described. In the following, a case where the target system is a large-scale battery storage system will be described as an example. In addition, in the following, an example where quantum-inspired calculation or quantum computing is used for the current-voltage characteristics of a single battery cell that constitutes the battery storage system will be described.

[0243] 22 is a diagram showing an example of an equivalent circuit model representing the current-voltage characteristics of each battery cell. A storage battery system 2210 includes a plurality of battery cells 2211. Each battery cell 2211 is represented by an equivalent circuit 2201.

[0244] The voltage v(t) of the battery cell 2211 at a certain time t is given by the circuit equation shown in the following equation (49).

number

[0245] To calculate the time evolution of the battery voltage behavior, a classical computer is used to perform numerical integration in discrete time as shown in the following equation (50). To calculate the time evolution of the voltages v1, v2, and v3 in each RC parallel element, the change rate f(i, v j ) is required.

number

[0246] By implicit method, f(i,v j ) can be calculated using any optimization method, and f(i,v j ) (real optimal solution).

[0247] Here, quantum-inspired computing or quantum computers can basically only handle binary variables. Therefore, in this embodiment, a real number optimization problem is treated as a binary variable optimization problem by introducing an approximate expression of real numbers using binary variables. The binary variable optimization problem is, for example, a quadratic unconstrained binary variable optimization problem (Quadratic unconstrained binary optimization: QUBO).

[0248] This makes it possible to use quantum-inspired calculations or a quantum computer to analyze an equivalent circuit showing current-voltage characteristics such as those shown in Fig. 22. For example, the analysis unit 123 can convert a real-number optimization problem for determining the rate of change of multiple response variables y into a binary variable optimization problem, and obtain multiple response variables y (output data) by solving the binary variable optimization problem through quantum-inspired calculations or calculations using a quantum computer.

[0249] In particular, by using quantum-inspired calculations specialized for solving combinatorial optimization problems such as QUBO, or quantum computers using the quantum annealing method, it becomes possible to analyze the behavior of each battery cell 2211 that makes up the storage battery system more quickly than with classical calculations.

[0250] When performing a phenomenon analysis using a surrogate model, the analysis unit 123 can execute the analysis process using the method described in Patent Document 1, for example.

[0251] As described above, according to the embodiment, analysis, visualization, and control of the resilience of a complex system can be more efficiently performed.

[0252] Next, a hardware configuration of the information processing apparatus according to the embodiment will be described with reference to Fig. 23. Fig. 23 is an explanatory diagram illustrating an example of a hardware configuration of the information processing apparatus according to the embodiment.

[0253] The information processing device of the embodiment includes a control device such as a CPU 51, a storage device such as a ROM (Read Only Memory) 52 and a RAM 53, a communication I / F 54 that connects to a network and performs communication, and a bus 61 that connects each unit.

[0254] The programs executed by the information processing apparatus according to the embodiment are provided in a state that they are pre-installed in the ROM 52 or the like.

[0255] The program executed by the information processing device of the embodiment may be configured to be provided as a computer program product by being recorded in an installable or executable format on a computer-readable recording medium such as a CD-ROM (Compact Disk Read Only Memory), a flexible disk (FD), a CD-R (Compact Disk Recordable), or a DVD (Digital Versatile Disk).

[0256] Furthermore, the program executed by the information processing apparatus of the embodiment may be configured to be stored on a computer connected to a network such as the Internet and provided by being downloaded via the network. Also, the program executed by the information processing apparatus of the embodiment may be configured to be provided or distributed via a network such as the Internet.

[0257] The programs executed by the information processing device of the embodiment can cause the computer to function as each unit of the information processing device described above. In this computer, the CPU 51 can read the programs from a computer-readable storage medium onto a main storage device and execute the programs.

[0258] A configuration example of the embodiment will be described below. (Configuration example 1) extracting a plurality of feature amounts representing characteristics of a plurality of second sampling data including a plurality of first sampling data generated using a multidimensional first probability distribution representing a distribution of each of a plurality of input data related to an analysis target and a plurality of output data representing physical quantities of the analysis target, the output data being obtained using the plurality of first sampling data as an input; and generating a network model representing relationships between a plurality of nodes corresponding to the physical quantities, including nodes corresponding to the extracted feature amounts; Processing section An information processing device comprising: (Configuration example 2) The processing unit includes: controlling the analysis target using the plurality of feature amounts as control variables; The information processing device according to configuration example 1. (Configuration example 3) The processing unit includes: updating the first probability distribution using monitoring data, which is a plurality of the input data measured from the analysis target; The information processing device according to configuration example 1 or 2. (Configuration Example 4) The processing unit includes: modifying the first probability distribution to optimize an index of the generated network model; generating a plurality of the first sampling data using the modified first probability distribution; The information processing device according to any one of configuration examples 1 to 3. (Configuration Example 5) The processing unit includes: generating a plurality of third sampling data based on the plurality of feature amounts, and extracting the plurality of feature amounts so as to optimize an index indicating whether a multidimensional second probability distribution representing a distribution of the plurality of third sampling data is consistent with the first probability distribution; The information processing device according to any one of configuration examples 1 to 4. (Configuration Example 6) The processing unit includes: Executing a quantum operation including an operation of decomposing a matrix based on a Hamiltonian generated from a plurality of the first sampling data, and obtaining a plurality of the output data by an optimization process for a result of the quantum operation. 6. The information processing device according to any one of configuration examples 1 to 5. (Configuration Example 7) The processing unit includes: A real number optimization problem for determining a rate of change of the plurality of first sampling data is converted into a binary variable optimization problem, and the plurality of output data is obtained by solving the binary variable optimization problem through quantum-inspired calculation or calculation using a quantum computer. 6. The information processing device according to any one of configuration examples 1 to 5. (Configuration Example 8) The processing unit includes: generating the first plurality of sampling data from the first probability distribution based on quantum-inspired calculation, quantum computing calculation, Lagrangian Monte Carlo calculation, Hamiltonian Monte Carlo calculation, or Markov chain Monte Carlo calculation; The information processing device according to any one of configuration examples 1 to 7. (Configuration Example 9) The processing unit includes: Calculating an occurrence probability representing a ratio of the number of sampling data pieces whose corresponding output data satisfy a specific condition among the plurality of first sampling data pieces to the number of the plurality of first sampling data pieces generated; The information processing device according to any one of configuration examples 1 to 8. (Configuration Example 10) The processing unit includes: controlling the analysis object so that the occurrence probability becomes a target value; The information processing device according to configuration example 9. (Configuration Example 11) The processing unit includes: generating a plurality of the first sampling data using the first probability distribution; The information processing device according to any one of configuration examples 1 to 10. (Configuration Example 12) The processing unit includes: A plurality of the first sampling data are input, and a plurality of the output data are obtained; The information processing device according to any one of configuration examples 1 to 11. (Configuration Example 13) The processing unit includes: A generating unit that generates a plurality of the first sampling data; An analysis unit for determining a plurality of said output data; a modeling unit for generating the network model; Equipped with The information processing device according to any one of configuration examples 1 to 10. (Configuration Example 14) An information processing method executed by an information processing device, a modeling step of extracting a plurality of feature amounts representing characteristics of a plurality of second sampling data including a plurality of first sampling data generated using a multidimensional first probability distribution representing a distribution of each of a plurality of input data related to the analysis target and a plurality of output data representing physical quantities of the analysis target, the output data being obtained using the plurality of first sampling data as an input, and generating a network model representing relationships between a plurality of nodes corresponding to the physical quantities, the network model including nodes corresponding to the extracted feature amounts; An information processing method comprising: (Configuration Example 15) On the computer, a modeling step of extracting a plurality of feature amounts representing characteristics of a plurality of second sampling data including a plurality of first sampling data generated using a multidimensional first probability distribution representing a distribution of each of a plurality of input data related to the analysis target and a plurality of output data representing physical quantities of the analysis target, the output data being obtained using the plurality of first sampling data as an input, and generating a network model representing relationships between a plurality of nodes corresponding to the physical quantities, the network model including nodes corresponding to the extracted feature amounts; A program for executing.

[0259] Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope and spirit of the invention, and are included in the scope of the invention and its equivalents described in the claims. [Explanation of symbols]

[0260] 41 Information Processing Systems 100 Platform 110 Storage section 120 Analysis Department 121 Setting section 122 Generation part 123 Analysis Department 124 Modeling Department 125 Communication control section 126-2 Optimization section 127-3 System control section 128-3 Update section 130 Operation section 200 Edge Devices 210 Control section 300 Service Providing Equipment 310 Service Department 320 Business Department 330 SoS Department 331 Output control section 332 Management Department 400 Common Services

Claims

1. extracting a plurality of feature quantities representing characteristics of a plurality of second sampling data including a plurality of first sampling data generated using a multidimensional first probability distribution representing a distribution of each of a plurality of input data related to the analysis target and a plurality of output data representing physical quantities of the analysis target, the output data being obtained using the plurality of first sampling data as input; and generating a network model representing a relationship between a plurality of nodes corresponding to the physical quantities, including nodes corresponding to the extracted feature quantities; Processing section An information processing device comprising:

2. The processing unit includes: controlling the analysis target using the plurality of feature amounts as control variables; The information processing device according to claim 1 .

3. The processing unit includes: updating the first probability distribution using monitoring data, which is the plurality of input data measured from the analysis target; The information processing device according to claim 1 .

4. The processing unit includes: modifying the first probability distribution to optimize an index of the generated network model; generating a plurality of the first sampling data using the modified first probability distribution; The information processing device according to claim 1 .

5. The processing unit includes: generating a plurality of third sampling data based on the plurality of feature amounts, and extracting the plurality of feature amounts so as to optimize an index indicating whether a multidimensional second probability distribution representing a distribution of the plurality of third sampling data matches the first probability distribution; The information processing device according to claim 1 .

6. The processing unit includes: executing a quantum operation including an operation of decomposing a matrix based on a Hamiltonian generated from the plurality of first sampling data, and obtaining the plurality of output data by an optimization process for a result of the quantum operation; The information processing device according to claim 1 .

7. The processing unit includes: A real number optimization problem for determining a rate of change of the plurality of first sampling data is converted into a binary variable optimization problem, and the plurality of output data is obtained by solving the binary variable optimization problem through quantum-inspired calculation or calculation using a quantum computer. The information processing device according to claim 1 .

8. The processing unit includes: generating the first plurality of sampling data from the first probability distribution based on quantum-inspired calculation, quantum computing calculation, Lagrangian Monte Carlo calculation, Hamiltonian Monte Carlo calculation, or Markov chain Monte Carlo calculation; The information processing device according to claim 1 .

9. The processing unit includes: Calculating an occurrence probability representing a ratio of the number of sampling data pieces whose corresponding output data satisfy a specific condition among the plurality of first sampling data pieces to the number of the plurality of first sampling data pieces generated; The information processing device according to claim 1 .

10. The processing unit includes: controlling the analysis object so that the occurrence probability becomes a target value; The information processing device according to claim 9.

11. The processing unit includes: generating a plurality of the first sampling data using the first probability distribution; The information processing device according to claim 1 .

12. The processing unit includes: A plurality of the first sampling data are input, and a plurality of the output data are obtained; The information processing device according to claim 1 .

13. The processing unit includes: A generating unit that generates a plurality of the first sampling data; An analysis unit for determining a plurality of said output data; a modeling unit for generating the network model; Equipped with The information processing device according to claim 1 .

14. An information processing method executed by an information processing device, a modeling step of extracting a plurality of feature amounts representing characteristics of a plurality of second sampling data including a plurality of first sampling data generated using a multidimensional first probability distribution representing a distribution of each of a plurality of input data related to the analysis target and a plurality of output data representing physical quantities of the analysis target, the output data being obtained using the plurality of first sampling data as input, and generating a network model representing relationships between a plurality of nodes corresponding to the physical quantities, the network model including nodes corresponding to the extracted feature amounts; An information processing method comprising:

15. On the computer, a modeling step of extracting a plurality of feature amounts representing characteristics of a plurality of second sampling data including a plurality of first sampling data generated using a multidimensional first probability distribution representing a distribution of each of a plurality of input data related to the analysis target and a plurality of output data representing physical quantities of the analysis target, the output data being obtained using the plurality of first sampling data as input, and generating a network model representing relationships between a plurality of nodes corresponding to the physical quantities, the network model including nodes corresponding to the extracted feature amounts; A program for executing the above.

Images