Verification cascade failure early warning method and device for complex simulation system

Abstract

The application relates to a complex simulation system-oriented verification cascade failure early warning method and device. The method comprises the following steps: constructing a multilayer verification network according to multiple verification tasks of a complex simulation system. According to the multilayer verification network, each verification task is divided into a plurality of continuous time slices, each time slice corresponds to a complete complex simulation system running and index collection process. The state transition probability between nodes of the multilayer verification network is calculated, so that the failure probability of the verification node state in the multilayer verification network is obtained by adopting a Bayesian network. The joint state between the verification nodes on the key propagation path is evaluated for failure, and the joint failure probability is obtained. According to the failure probability and the joint failure probability, a three-level early warning mechanism is constructed, so that the early warning state sample iterative learning is carried out through an optimization evaluation algorithm, and a cascade failure early warning scheme is generated. The method can improve the verification efficiency and adaptability of the complex simulation system.

Description

Technical Field

[0001] This invention relates to the field of simulation system verification technology, and in particular to a method and apparatus for early warning of cascaded verification failures in complex simulation systems. Background Technology

[0002] In digital engineering practices for complex systems with high security levels, model credibility is a fundamental prerequisite for supporting design optimization, status monitoring, and emergency decision-making. Existing model verification and validation frameworks cover multiple stages, including code implementation verification, algorithm convergence analysis, numerical error control, physical consistency verification, and uncertainty quantification. With the development of automated testing and continuous integration / CD technologies, some systems are now able to streamline the execution of verification tasks and output structured metrics (such as the mesh convergence index (GCI), fidelity index, and posterior coverage).

[0003] However, existing verification systems still have significant limitations: each verification step is treated as a static, isolated checkpoint, lacking systematic monitoring and management of the dynamic characteristics of the verification process itself. In fact, the verification process is essentially a strongly coupled, nonlinear, and delayed dynamic process. Small errors in upstream steps can be propagated and amplified through multiple steps, ultimately leading to the complete failure of downstream verification, a phenomenon known as "cascading failure." For example, boundary condition errors in the code implementation may not be immediately apparent, but they can cause numerical oscillations during subsequent algorithm convergence verification, leading to overall model inaccuracy during the physical verification stage, forming a typical "cascading failure of the verification process."

[0004] In recent years, some studies have attempted to introduce graph neural networks or rule engines to monitor the verification status, but these methods struggle to effectively handle observation noise, model uncertainty, and temporal causal relationships. Although existing V&V technologies are becoming increasingly mature in their local verification capabilities, they still suffer from the following key shortcomings in addressing the systemic risks inherent in the verification process itself: 1. Isolated operation of verification stages, lacking overall process monitoring: Each verification task executes independently, failing to detect "upstream anomalies propagating downstream," leading to delayed problem detection. Backtracking analysis is often only conducted after the final model verification fails, resulting in low efficiency. 2. Lack of dynamic risk modeling: The dependencies and impact strength between verification stages are not quantified, making it difficult to assess the cascading impact of a single stage failure on the overall verification credibility. 3. Insufficient early warning capabilities: Existing systems typically only issue alerts after verification failure, failing to provide risk propagation path prediction and tiered early warning. 4. Lack of proactive blocking capabilities: Even when problems are detected, there are no automated strategies to guide the allocation of limited verification resources to prioritize the repair of high-risk stages, resulting in high intervention costs and poor effectiveness. Summary of the Invention

[0005] Therefore, it is necessary to provide a method and device for early warning of cascaded failures in complex simulation systems that can improve the verification efficiency and adaptability of complex simulation systems, in order to address the above-mentioned technical problems.

[0006] A method for early warning of cascaded failures in complex simulation systems, the method comprising:

[0007] A multi-layered verification network is constructed based on the various verification tasks of a complex simulation system.

[0008] Each verification task is divided into several consecutive time slices based on the multi-layer verification network. Each time slice corresponds to a complete operation and index collection process of a complex simulation system.

[0009] Calculate the state transition probabilities between nodes in a multilayer check network so that a Bayesian network can be used to obtain the failure probability of the check node state in the multilayer check network.

[0010] A failure assessment is performed on the joint state of the verification nodes on the critical propagation path to obtain the joint failure probability.

[0011] A three-level early warning mechanism is constructed based on the failure probability and the joint failure probability. The early warning state samples are iteratively learned through optimization evaluation algorithm to generate cascading failure early warning schemes.

[0012] A verification cascade failure early warning device for complex simulation systems, the device comprising:

[0013] The verification network construction module is used to build multi-layer verification networks based on various verification tasks of complex simulation systems.

[0014] The time slice partitioning module is used to divide each verification task into several consecutive time slices according to the multi-layer verification network. Each time slice corresponds to a complete operation and index acquisition process of a complex simulation system.

[0015] The node failure verification module is used to calculate the state transition probability between nodes in a multi-layer verification network, so as to obtain the failure probability of the verification node state in the multi-layer verification network using a Bayesian network.

[0016] The path failure verification module is used to perform failure assessment on the joint state of verification nodes on the critical propagation path and obtain the joint failure probability.

[0017] The early warning assessment module is used to construct a three-level early warning mechanism based on the failure probability and the joint failure probability. It generates a cascading failure early warning scheme by iteratively learning the early warning state samples through the optimization assessment algorithm.

[0018] The aforementioned method and device for cascaded failure early warning in complex simulation systems first construct a multi-layered verification network comprising various verification tasks (such as code implementation and algorithm convergence). By quantifying the dependencies between stages using directed edges with propagation weights and delays, isolated verification tasks are integrated into a strongly correlated dynamic system, making upstream anomaly propagation perceptible and traceable, thus preventing delayed problem discovery. Based on Bayesian networks, the verification process is divided into continuous time slices to characterize cross-cycle state transition patterns, and real-time inference of node failure probabilities and critical path joint failure probabilities is performed. Then, a three-tiered early warning mechanism (node, path, system) is used to hierarchically identify local anomalies, chain propagation, and global instability, achieving a leap from post-event correction to in-event early warning. Furthermore, each time slice corresponds to a complete operation and indicator collection process of the complex simulation system. The collected early warning information clearly identifies high-risk nodes, propagation paths, and core driving indicators, providing accurate basis for resource allocation. Simultaneously, a closed-loop optimization mechanism is used, employing Bayesian optimization to dynamically adjust model parameters and early warning thresholds, continuously optimizing the intervention strategy. It reduces rework costs caused by isolated processes and shortens problem handling time through precise early warning and intelligent intervention, significantly improving verification efficiency. The closed-loop learning mechanism enables the system to adapt to the verification characteristics of different complex simulation scenarios, realize cross-task knowledge transfer, greatly enhance adaptability, and provide highly reliable and efficient process assurance for the verification of complex system models. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a verification cascade failure early warning method for a complex simulation system in one embodiment.

[0020] Figure 2 This is a flowchart illustrating a cascaded failure warning and blocking method for model verification based on dynamic Bayesian networks in one embodiment.

[0021] Figure 3 This is a structural block diagram of a verification cascade failure early warning device for a complex simulation system in one embodiment. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0023] In one embodiment, such as Figure 1 As shown, a method for early warning of cascaded failures in complex simulation systems is provided, including the following steps:

[0024] Step 102: Construct a multi-layer verification network based on the various verification tasks of the complex simulation system.

[0025] Step 104: Divide each verification task into several consecutive time slices according to the multi-layer verification network. Each time slice corresponds to a complete operation and index acquisition process of the complex simulation system.

[0026] Step 106: Calculate the state transition probabilities between nodes in the multilayer check network so that a Bayesian network can be used to obtain the failure probability of the check node state in the multilayer check network.

[0027] Step 108: Perform a failure assessment on the joint state of the verification nodes on the critical propagation path to obtain the joint failure probability.

[0028] Step 110: Construct a three-level early warning mechanism based on the failure probability and the joint failure probability, and generate a cascading failure early warning scheme by iteratively learning the early warning state samples through an optimized evaluation algorithm.

[0029] In the aforementioned method for cascading failure early warning in complex simulation systems, firstly, a multi-layered verification network is constructed, incorporating various verification tasks (such as code implementation and algorithm convergence). Dependencies between stages are quantified using directed edges with propagation weights and delays, integrating isolated verification tasks into a strongly correlated dynamic system. This makes upstream anomaly propagation perceptible and traceable, preventing delayed problem discovery. Based on Bayesian networks, the verification process is divided into continuous time slices, characterizing cross-cycle state transition patterns and inferring the failure probability of nodes and the joint failure probability of critical paths in real time. Then, a three-tiered early warning mechanism (nodes, paths, and the system) is used to hierarchically identify local anomalies, chain propagation, and global instability, achieving a leap from post-event correction to in-event early warning. Furthermore, each time slice corresponds to a complete operation and indicator collection process of the complex simulation system. The collected early warning information clearly identifies high-risk nodes, propagation paths, and core driving indicators, providing accurate basis for resource allocation. Simultaneously, a closed-loop optimization mechanism is used, employing Bayesian optimization to dynamically adjust model parameters and early warning thresholds, continuously optimizing the intervention strategy. It reduces rework costs caused by isolated processes and shortens problem handling time through precise early warning and intelligent intervention, significantly improving verification efficiency. The closed-loop learning mechanism enables the system to adapt to the verification characteristics of different complex simulation scenarios, realize cross-task knowledge transfer, greatly enhance adaptability, and provide highly reliable and efficient process assurance for the verification of complex system models.

[0030] In one embodiment, a multi-layer verification network consisting of multiple verification nodes and directed edges is designed based on various verification tasks of a complex simulation system. Each verification node includes a code implementation verification node, an algorithm convergence verification node, a numerical error control verification node, a physical consistency verification node, and an uncertainty quantification verification node corresponding to the verification task. Directed edges use the verification period corresponding to the verification task as the propagation unit, represent the error propagation direction, and are assigned propagation strength weights and propagation delay parameters.

[0031] In one embodiment, each verification task is divided into several consecutive time slices based on the state variables of the verification nodes in the multi-layer verification network:

[0032]

[0033]

[0034] in, For the first Each verification node in the time slice State variables, For the current time slice The verification process indicators meet the preset thresholds and there are no significant problems. For the current time slice If the verification indicators slightly exceed the preset threshold, recovery is allowed through local adjustments. For the current time slice If the verification indicators deviate significantly from the preset threshold, or if there are two consecutive abnormal periods, or if key criteria are not met, the verification function has been lost. Each time slice corresponds to a complete operation and indicator acquisition process of the complex simulation system, and the state of the complex simulation system is updated at the end of each time slice.

[0035] In one embodiment, a two-time-slice transition network is used to calculate the state transition probabilities between nodes in the multi-layer verification network:

[0036]

[0037] in, Let be the state transition probability. For the first Each verification node in the time slice State variables, The target state variable for the current verification node. The set of parent nodes for the state variables. For multi-class probability normalization function, For the first The inherent bias term of each verification node, For the first The verification node and the first Propagation strength weights among verification nodes For state mapping function, For the first The verification node and the first Propagation delay between verification nodes For the first Each verification node in the time slice State variables, Let be the error intensity normalization function. For the first Verification process metrics for each verification node. For the first One verification node.

[0038] In one embodiment, a Bayesian network is used to calculate the failure probability of the verification node state in the multi-layer verification network after normalizing the multi-source observation indicators of each verification node collected at the end of each verification period.

[0039]

[0040] in, This represents the probability of failure. For the first Each verification node in the time slice State variables, From the first verification cycle to the current verification cycle The observation sequence for each verification period.

[0041] In one embodiment, a Bayesian network is used to evaluate the critical path where the error propagates along a specific link. A joint distribution decomposition and message passing algorithm is then used to calculate the joint failure probability of all verification nodes on the critical path.

[0042]

[0043] in, For the first verification node in the time slice State variables, For the second verification node in the time slice State variables, For the third verification node in the time slice State variables, For the 4th verification node in the time slice The state variable.

[0044] In one embodiment, a three-level early warning mechanism is constructed based on the failure probability and the joint failure probability; the three-level early warning mechanism includes: node-level early warning, path-level early warning, and system-level early warning. When the failure probability of a certain verification node exceeds a preset threshold, and the error index corresponding to the verification node shows an upward trend, the node-level early warning is triggered.

[0045]

[0046]

[0047] in, Let be the error intensity normalization function. A threshold is preset for the failure probability. For the first The current verification node is the [number]th. Verification process indicators for each verification cycle For the first The current verification node is the [number]th. Verification process indicators for each verification cycle This represents the minimum change in the error index. Based on the propagation strength weights in the multi-layer verification network and the failure probability of each verification node, the cumulative risk value of the critical path is calculated:

[0048]

[0049] in, This represents the cumulative risk value of the critical path. For the first The verification node and the first Propagation strength weights among verification nodes For the k-th verification node in time slice The state variables; when the accumulated risk value exceeds the path-level threshold, it is determined that the current critical path has a high-risk failure situation, triggering the path-level warning and marking it as a high-risk error propagation channel. A system instability index is generated by weighted aggregation of the failure probabilities of all verification nodes:

[0050]

[0051] in, The system instability index. For the first The propagation strength weights of each verification node are determined; when the system instability index exceeds the system-level threshold, a system-level early warning is triggered, and an emergency intervention process is initiated. The state variables of all verification nodes corresponding to the node-level early warning, the path-level early warning, and the system-level early warning are used as early warning state samples. The early warning state samples are iteratively learned through an optimization evaluation algorithm to generate a cascading failure early warning scheme.

[0052] In one embodiment, such as Figure 2 As shown, a flowchart of a cascaded failure early warning and blocking method for model validation based on dynamic Bayesian networks is provided. By modeling the model validation process as a dynamic system with temporal evolution characteristics, DBN is used to probabilistically represent the state transitions, error propagation paths, and failure risks of the validation stages, and on this basis, a three-level early warning and intelligent blocking mechanism is implemented. Specifically, the following steps are included:

[0053] The first step is to construct a multi-layer verification network topology.

[0054] This invention first constructs a multi-layered verification network topology to formally express the dependencies and error propagation paths between different stages in the model verification process. The network consists of multiple verification nodes and directed edges, where nodes represent different types of verification tasks, including code implementation verification nodes (C1), algorithm convergence verification nodes (C2), numerical error control verification nodes (C3), physical consistency verification nodes (C4), and uncertainty quantification verification nodes (C5). Nodes are connected by directed edges, representing the propagation direction of errors or defects between different verification stages. For example, code implementation errors may cause algorithm non-convergence, leading to amplified numerical errors. Each directed edge is assigned a propagation strength weight. This is used to quantify the impact of upstream nodes on downstream nodes, and optionally configures the propagation delay parameter. (in units of the verification cycle) to reflect the time lag effect of error propagation.

[0055] The second step is to build a dynamic Bayesian network model.

[0056] This invention constructs a Dynamic Bayesian Network (DBN) based on the verification network topology to probabilistically model the state evolution and risk propagation during the verification process. The DBN employs a discrete-time slice design, and its implementation is as follows:

[0057] 1. Definition of Time Slice

[0058] The continuous model validation process is divided into a series of discrete validation cycles, each cycle corresponding to a complete simulation run and index acquisition process. Each validation cycle is abstracted as a time slice. The system status is updated once at the end of each time slice.

[0059] 2. Node State Space Definition

[0060] For the Verification nodes ( These correspond to the code implementation layer, algorithm convergence layer, numerical error layer, physical consistency layer, and uncertainty quantization verification, respectively, and are located in time slices. The state variable is denoted as :

[0061]

[0062] "Normal" indicates that the indicator in the verification process meets the preset threshold and there are no significant problems; "abnormal" indicates that the indicator slightly exceeds the limit or fluctuates, but can be recovered through local adjustments; "failed" indicates that the indicator deviates seriously, is abnormal for two consecutive periods, or the key criteria are not met, and the verification function has been lost.

[0063] 3. Construction of Two-Time-Slice Transfer Network (2TBN)

[0064] The dynamic behavior of DBN is defined by a 2-Time-Slice Temporal Network (2TBN), which characterizes the transition from time slices... arrive The state transition rules for any node. Its state probability depends on its set of parent nodes. The state and observation error at the previous moment (considering propagation delay). Specifically, the conditional probability distribution is defined as:

[0065]

[0066] in, Indicates the target state of the current node. This represents the inherent bias term of the node. Indicates the propagation strength weight. Indicates delayed transmission. For state mapping function, Let be the error intensity normalization function. This represents a multi-class probability normalization function.

[0067] 4. Initialization and Update of Conditional Probability Table (CPT)

[0068] The construction of CPT comprehensively considers propagation weight, statistical regularity of historical verification data and expert prior knowledge. It can also map actual observed indicators (such as GCI value and fidelity index) into driving factors of state transition through soft evidence fusion mechanism, thereby realizing probabilistic modeling of the dynamic behavior of the verification process.

[0069] Initial time ( =0) Prior state distribution of each node The initial failure probability of code validation is set based on historical project statistics or domain expert experience (e.g., 5% for code validation).

[0070] With the input of new validation period data, the CPT parameters can be optimized in the following ways: Maximum likelihood estimation (MLE) or Bayesian estimation can be used to update them. and A sliding window mechanism is introduced, which updates the data using only the most recent N periods of data to adapt to process drift.

[0071] Through the above modeling, DBN can not only express the static dependencies between verification stages, but also capture the temporal, delayed and uncertain aspects of error propagation.

[0072] The third step is real-time status monitoring and probabilistic reasoning.

[0073] This invention collects multi-source observation indicators of each verification node at the end of each verification cycle and transforms them into soft evidence acceptable to dynamic Bayesian networks (DBNs), thereby performing online probabilistic inference to achieve real-time assessment of the health status of the verification process.

[0074] 1. Multi-source indicator collection and normalization

[0075] For each verification node, the system collects multi-dimensional metrics related to it, including but not limited to unit test failure rate, mesh convergence order, GCI error band width, physics matching degree, posterior coverage, etc., all raw metrics. All are normalized using a preset error intensity function. Mapping to the [0,1] interval, for example:

[0076]

[0077] in, , This is the empirical threshold for this indicator.

[0078] 2. Modeling and Updating

[0079] After normalization Transform into state variables The likelihood function, which can be pre-configured or learned from historical data.

[0080] Based on observation sequence The system employs a forward algorithm for precise inference, or a particle filter for approximate inference when the state space is large, to update the posterior state distribution of each node in real time, particularly calculating its failure probability.

[0081]

[0082] This formula represents the observation from the first verification period to the current period. Under the condition of all evidence (verification index data) in the verification period, the first Each verification node at time... The posterior probability of being in a "failed" state.

[0083] 3. Path Joint Failure Probability Reasoning

[0084] Reasoning about the joint state of nodes on the critical propagation path can assess the likelihood of error propagation along a specific link, and the system can further calculate its joint failure probability:

[0085]

[0086] This value is obtained through the joint distribution decomposition and message passing algorithm of DBN. It serves as the core basis for path-level early warning, providing accurate probabilistic basis for subsequent hierarchical early warning and ensuring timely and reliable risk perception.

[0087] The fourth step is to establish a three-tiered early warning mechanism.

[0088] To achieve risk classification and control from local to global levels, this invention designs a three-level early warning mechanism of "node-path-system".

[0089] 1. Node-level early warning

[0090] At the node level, when the failure probability of a certain verification node exceeds a preset threshold:

[0091]

[0092] Furthermore, its error index shows a continuous upward trend:

[0093]

[0094] When these two conditions are met, a node-level alert will be triggered. The alert information includes, but is not limited to, the node ID, the current state probability, the main driving indicators, and suggested check items, indicating that there is a local anomaly in this process.

[0095] 2. Path-level early warning

[0096] At the path level, based on the propagation weights in the verification network and the failure probabilities of each node, the cumulative risk value of the critical propagation path is calculated:

[0097]

[0098] in The edge weight reflects the propagation strength. When Exceeding path-level threshold When the path is deemed to have a high risk of cascading failure, a path-level early warning is triggered, identifying high-risk error propagation channels.

[0099] 3. System-level early warning

[0100] At the system level, the system instability index (SSI) is generated by weighted aggregation of the failure probabilities of all nodes:

[0101]

[0102] Among them, weight This reflects the importance of nodes to the final model's reliability. When the SSI exceeds the system-level threshold... When this occurs, it indicates that the overall verification process is on the verge of collapse, triggering a system-level warning and initiating an emergency intervention process.

[0103] Step 5: Effectiveness Evaluation and Learning Optimization

[0104] To ensure the long-term effectiveness and adaptability of the method, this invention establishes a closed-loop effect evaluation and learning optimization mechanism. The system continuously records the actual results of each warning and blocking operation, constructing a multi-dimensional evaluation index system, including blocking success rate, system recovery time, resource consumption cost, and warning accuracy (including false alarm rate and false negative rate). Based on the evaluation results, key parameters in the dynamic Bayesian network, such as propagation weights, warning thresholds at each level, and node importance calculation weights, are dynamically adjusted using methods such as Bayesian optimization, reinforcement learning, or online parameter estimation. This mechanism enables the entire verification risk governance system to have the evolutionary capability of "becoming more accurate and better with use," adapting to the verification characteristics of different projects and model types, and achieving cross-task knowledge transfer and continuous optimization.

[0105] It is worth noting that the model validation process itself is modeled as a dynamic risk system, constructing a validation process state evolution and error propagation characterization mechanism based on a dynamic Bayesian network (DBN). Validation tasks such as code implementation, algorithm convergence, numerical error control, physical consistency, and uncertainty quantification are abstracted as nodes with ternary states (normal / abnormal / failure), and a directed topology with weights and propagation delays is constructed based on their engineering dependencies. On this basis, a two-time-slice transition network (2TBN) is established, and the node state transition rules are probabilistically defined using conditional probability tables (CPT) or parameterized softmax functions. This allows for the simultaneous characterization of the self-sustaining characteristics of the validation state, the cross-period propagation effect of upstream errors, and the fusion and updating capability of multi-source observation evidence within a unified framework.

[0106] A proactive blocking closed-loop mechanism integrating "perception-early warning-optimization" is proposed to achieve a leap from passive response to intelligent governance. By using forward algorithms or particle filtering to calculate the posterior failure probability of each node in real time, a three-level early warning mechanism at the node, path, and system levels is established to achieve layered and precise risk perception. Finally, multi-dimensional evaluation indicators (such as blocking success rate, early warning accuracy, and resource consumption) drive Bayesian optimization or reinforcement learning to dynamically adjust core parameters such as propagation weights and early warning thresholds, forming a continuous optimization closed loop of "operation-evaluation-learning-evolution."

[0107] It should be understood that, although Figures 1-2 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figures 1-2 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0108] In one embodiment, such as Figure 3 As shown, a verification cascade failure early warning device for complex simulation systems is provided, comprising: a verification network construction module 302, a time slice partitioning module 304, a node failure verification module 306, a path failure verification module 308, and an early warning evaluation module 310, wherein:

[0109] The verification network construction module 302 is used to construct a multi-layer verification network based on various verification tasks of a complex simulation system.

[0110] The time slice partitioning module 304 is used to divide each verification task into several consecutive time slices according to the multi-layer verification network. Each time slice corresponds to a complete operation and index acquisition process of a complex simulation system.

[0111] The node failure verification module 306 is used to calculate the state transition probability between nodes in the multi-layer verification network, so as to obtain the failure probability of the verification node state in the multi-layer verification network using a Bayesian network.

[0112] The path failure verification module 308 is used to perform failure assessment on the joint state of verification nodes on the critical propagation path and obtain the joint failure probability.

[0113] The early warning assessment module 310 is used to construct a three-level early warning mechanism based on the failure probability and the joint failure probability, and to generate a cascading failure early warning scheme by iteratively learning the early warning state samples through the optimization assessment algorithm.

[0114] Specific limitations regarding the verification cascade failure early warning device for complex simulation systems can be found in the limitations of the verification cascade failure early warning method for complex simulation systems described above, and will not be repeated here. Each module in the aforementioned verification cascade failure early warning device for complex simulation systems can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0115] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0116] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchlink, DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0117] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0118] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A method for early warning of cascaded failures in complex simulation systems, characterized in that, The method includes: Construct a multi-layered verification network based on the various verification tasks of a complex simulation system; According to the multi-layer verification network, each verification task is divided into several consecutive time slices, and each time slice corresponds to a complete operation and index collection process of the complex simulation system. Calculate the state transition probabilities between nodes in the multi-layer verification network so that a Bayesian network can be used to obtain the failure probability of the verification node state in the multi-layer verification network. A failure assessment is performed on the joint state of the verification nodes along the critical propagation path to obtain the joint failure probability; A three-level early warning mechanism is constructed based on the failure probability and the joint failure probability, and a cascading failure early warning scheme is generated by iterative learning of early warning state samples through an optimized evaluation algorithm. The three-level early warning mechanism includes: node-level early warning, path-level early warning, and system-level early warning; When the failure probability of a certain verification node exceeds a preset threshold, and the error index corresponding to the verification node shows an upward trend, the node-level early warning is triggered: in, Let be the error intensity normalization function. A threshold is preset for the failure probability. For the first The current verification node is the [number]th. Verification process indicators for each verification cycle For the first The current verification node is the [number]th. Verification process indicators for each verification cycle This represents the minimum change in the error index. Based on the propagation strength weights in the multi-layer verification network and the failure probability of each verification node, the cumulative risk value of the critical propagation path is calculated: in, This represents the cumulative risk value along the key transmission path. For the first The verification node and the first Propagation strength weights among verification nodes For the k-th verification node in time slice The state variable; when the accumulated risk value exceeds the path-level threshold, it is determined that there is a high-risk failure in the current critical propagation path, triggering the path-level early warning and marking it as a high-risk error propagation channel; The system instability index is generated by weighted aggregation of the failure probabilities of all verification nodes. in, The system instability index. For the first The propagation strength weight of each verification node; when the system instability index exceeds the system-level threshold, the system-level early warning is triggered, and the emergency intervention process is initiated; The state variables of all verification nodes that trigger the node-level warning, the path-level warning, and the system-level warning are used as warning state samples. The warning state samples are iteratively learned through an optimization evaluation algorithm to generate a cascading failure warning scheme.

2. The method according to claim 1, characterized in that, A multi-layered verification network is constructed based on the various verification tasks of a complex simulation system, including: Based on the various verification tasks of a complex simulation system, a multi-layer verification network composed of multiple verification nodes and directed edges is designed. Each verification node includes a code implementation verification node, an algorithm convergence verification node, a numerical error control verification node, a physical consistency verification node, and an uncertainty quantification verification node corresponding to the verification task. The directed edge uses the verification period corresponding to the verification task as the propagation unit to represent the error propagation direction, and is assigned a propagation strength weight and a propagation delay parameter.

3. The method according to claim 1, characterized in that, Based on the multi-layer verification network, each verification task is divided into several consecutive time slices. Each time slice corresponds to a complete operation and index acquisition process of the complex simulation system, including: Based on the state variables of the verification nodes in the multi-layer verification network, each verification task is divided into several consecutive time slices: in, For the first Each verification node in the time slice State variables, For the current time slice The verification process indicators meet the preset thresholds and there are no significant problems. For the current time slice If the verification indicators slightly exceed the preset threshold, recovery is allowed through local adjustments. For the current time slice If the verification indicators deviate significantly from the preset threshold, or if there are two consecutive abnormal periods or if the key criteria are not met, the verification function has been lost. Each time slice corresponds to a complete operation and index collection process of the complex simulation system, and the state of the complex simulation system is updated at the end of each time slice.

4. The method according to claim 3, characterized in that, Calculating the state transition probabilities between nodes in the multilayer verification network includes: The state transition probabilities between nodes in the multilayer verification network are calculated using a two-time-slice transition network: in, Let be the state transition probability. For the first Each verification node in the time slice State variables, The target state variable for the current verification node. The set of parent nodes for the state variables. For multi-class probability normalization function, For the first The inherent bias term of each verification node, For the first The verification node and the first Propagation strength weights among verification nodes For state mapping function, For the first The verification node and the first Propagation delay between verification nodes for, Let be the error intensity normalization function. For the first Verification process metrics for each verification node. For the first One verification node.

5. The method according to claim 4, characterized in that, The failure probability of the check node state in the multi-layer check network is obtained using a Bayesian network, including: Using a Bayesian network, the failure probability of the verification node state in the multi-layer verification network is calculated after normalizing the multi-source observation indicators of each verification node collected at the end of each verification period. in, This represents the probability of failure. For the first Each verification node in the time slice State variables, From the first verification cycle to the current verification cycle The observation sequence for each verification period.

6. The method according to claim 5, characterized in that, The joint state of the verification nodes on the critical propagation path is evaluated for failure to obtain the joint failure probability, and the evaluation also includes: Using a Bayesian network to evaluate the critical propagation path of error along a specific link, a joint distribution decomposition and message passing algorithm is used to calculate the joint failure probability of all verification nodes on the critical propagation path: in, For the first verification node in the time slice State variables, For the second verification node in the time slice State variables, For the third verification node in the time slice State variables, For the 4th verification node in the time slice The state variable.

7. A verification cascade failure early warning device for complex simulation systems, characterized in that, The device includes: The verification network construction module is used to construct multi-layer verification networks based on various verification tasks of complex simulation systems. The time slice division module is used to divide each verification task into several consecutive time slices according to the multi-layer verification network, and each time slice corresponds to a complete operation and index collection process of the complex simulation system. The node failure verification module is used to calculate the state transition probability between nodes in the multi-layer verification network, so as to obtain the failure probability of the verification node state in the multi-layer verification network using a Bayesian network. The path failure verification module is used to perform failure assessment on the joint state of the verification nodes on the critical propagation path and obtain the joint failure probability. The early warning assessment module is used to construct a three-level early warning mechanism based on the failure probability and the joint failure probability, and to generate a cascading failure early warning scheme by iteratively learning early warning state samples through an optimized assessment algorithm. The three-level early warning mechanism includes: node-level early warning, path-level early warning, and system-level early warning. When the failure probability of a certain verification node exceeds a preset threshold, and the error index corresponding to the verification node shows an upward trend, the node-level early warning is triggered. in, Let be the error intensity normalization function. A threshold is preset for the failure probability. For the first The current verification node is the [number]th. Verification process indicators for each verification cycle For the first The current verification node is the [number]th. Verification process indicators for each verification cycle The minimum change in the error index is given; based on the propagation strength weights in the multi-layer verification network and the failure probability of each verification node, the cumulative risk value of the critical propagation path is calculated: in, This represents the cumulative risk value along the key transmission path. For the first The verification node and the first Propagation strength weights among verification nodes For the k-th verification node in time slice The state variables; when the accumulated risk value exceeds the path-level threshold, it is determined that the current critical propagation path has a high-risk failure situation, triggering the path-level early warning and marking it as a high-risk error propagation channel; by weighted aggregation of the failure probabilities of all verification nodes, a system instability index is generated: in, The system instability index. For the first The propagation strength weight of each verification node; when the system instability index exceeds the system-level threshold, the system-level early warning is triggered and the emergency intervention process is initiated; the state variables of all verification nodes corresponding to the node-level early warning, the path-level early warning and the system-level early warning are used as early warning state samples, and the early warning state samples are iteratively learned through an optimization evaluation algorithm to generate a cascading failure early warning scheme.

Images