A risk assessment method and system based on dynamic upgrading and enhanced environmental evolution of domino effect
The domino effect risk assessment method enhanced by dynamic upgrading and environmental evolution solves the problem that existing technologies cannot reflect dynamic evolution and multi-scale coupling, realizes dynamic risk assessment of domino effects, can accurately identify key risk nodes and support safety design optimization.
Patent Information
- Application Number
- CN202611181110.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-05
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2046-08-05
AI Technical Summary
Existing technologies cannot effectively capture the dynamic evolution characteristics of the domino effect, resulting in large deviations between risk assessment results and actual accidents, and making it difficult to achieve multi-scale coupling, self-triggered secondary accidents, and uncertainty quantification.
A domino effect risk assessment method based on dynamic upgrading and environmental evolution enhancement is adopted. Through initial propagation probability calculation, environmental evolution enhancement, self-triggering mechanism, multi-scale causal network construction and uncertainty quantification, the temporal evolution simulation of accident propagation and system-level risk quantification are realized.
It enables dynamic, multi-scale, interpretable, and verifiable risk assessment of the domino effect, accurately characterizing the temporal evolution of accident propagation, identifying key risk nodes, and supporting safety design optimization.
Smart Images

Figure CN122693301B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the interdisciplinary fields of safety engineering for industrial facilities (such as offshore energy systems), dynamic risk assessment of domino effects, and digital twin safety simulation, and in particular to a method and system for risk assessment of domino effects based on dynamic upgrades and enhanced environmental evolution. Background Technology
[0002] The dynamic evolution of the domino effect is a key challenge in the safety assessment of offshore energy systems. In actual accidents, equipment resilience degrades with continuous loading, the intensity of the accident source amplifies over time, and the synergistic effects of multiple sources accelerate propagation, causing the accident scale and propagation path to change dynamically over time. Static risk assessments cannot capture these dynamic characteristics, leading to significant discrepancies between assessment results and actual accidents. The 1988 Piper Alpha platform accident and the 2010 Deepwater Horizon (DWH) blowout accident both resulted in losses far exceeding static assessment expectations due to dynamic evolution effects.
[0003] Current mainstream domino effect risk assessment techniques have significant shortcomings in terms of dynamic evolution: The first category is static threshold and probabilistic methods, represented by Cozzani and Landucci, which use fixed thresholds or static Probit functions to calculate propagation probabilities. These methods have four major drawbacks: First, they cannot reflect the dynamic characteristics of equipment immunity degradation over time; in actual accidents, steel structures lose 50% of their yield strength at 600°C during continuous fires. Second, they cannot reflect the dynamic characteristics of the amplification of the release intensity of the accident source over time; well blowout flow increases with reservoir pressure release. Third, they cannot reflect the synergistic effect of multiple sources; when multiple sources act on the same target simultaneously, the propagation probability increases nonlinearly. Fourth, they use single-scale assessment, making it impossible to achieve multi-scale coupled analysis at the equipment, site, and regional levels.
[0004] The second category is high-fidelity numerical simulation methods, represented by CFD and finite element methods. These methods can simulate the time evolution of physical processes, but they have three major drawbacks: First, single-condition simulation takes hours to days, which cannot meet the needs of real-time emergency response. Second, it is difficult to achieve system-level dynamic risk assessment with multi-device and multi-scale coupling. Third, the simulation results have poor interpretability and it is difficult to directly extract causal propagation paths and key risk nodes.
[0005] The third category is network analysis methods based on graph theory, which abstract devices as nodes and propagation paths as edges. However, it has three major drawbacks: First, the network construction lacks physical constraints, and the edge weights depend on empirical settings; second, it lacks a time dimension and cannot reflect the temporal characteristics of accident propagation; and third, it lacks the ability to quantify uncertainty, and the evaluation results cannot provide confidence intervals.
[0006] In summary, existing technologies cannot simultaneously address dynamic time evolution, multi-scale coupling, self-triggered secondary accidents, interpretability, and uncertainty quantification. The industry urgently needs a dynamic risk assessment technology solution with dynamic upgrade capabilities, multi-scale collaboration, interpretability, and verifiability for domino effects.
[0007] It should be noted that the information disclosed in the background section above is only for understanding the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention
[0008] The main objective of this invention is to overcome the deficiencies in the aforementioned background technology and provide a method and system for risk assessment of the domino effect based on dynamic upgrading and enhanced environmental evolution.
[0009] To achieve the above objectives, the present invention adopts the following technical solution: A risk assessment method based on the domino effect enhanced by dynamic upgrading and environmental evolution includes the following steps: S1. Initial propagation probability calculation: For equipment networks of industrial facilities such as offshore energy systems, based on the causal strength quantification method, the initial coupling strength and initial propagation probability of each propagation path of the domino effect, as well as the initial cumulative propagation probability of each node, are calculated to provide an initial state benchmark for dynamic risk assessment. S2. Enhanced Environmental Evolution: For continuous accident scenarios, the coupling strength and propagation probability of each propagation path are dynamically updated through three sub-mechanisms: degradation of anti-disturbance strength, continuous source amplification, and enhanced cooperative coupling, so as to realize the temporal evolution simulation of accident propagation. S3. Self-triggering mechanism and dynamic evolution of secondary accidents: When an accident occurs at a node, a secondary accident of the node is triggered based on the self-triggering probability and the delay time window function, realizing the dynamic evolution of accident types and the time-sequence triggering of secondary accidents. S4. Construction of multi-scale causal network: Construct a three-layer causal network including equipment level, station level and regional level to realize cross-scale domino effect coupling analysis; S5. Causal Network Filtering: Perform three-step filtering on the constructed multi-scale causal network: edge probability filtering, breadth-first search unreachable node removal, and isolated initial event node removal to generate a concise and effective causal topology. S6. System-level risk quantification: Based on node-level risk and the importance weight of nodes in the causal network, calculate the system-level comprehensive risk and output the accident propagation time window; S7. Uncertainty Quantification: Input parameters are repeatedly sampled through Monte Carlo simulation to calculate the probability distribution and confidence interval of system-level risk, and key influencing parameters are identified through sensitivity analysis. S8. Benchmark Validation and Interpretability Output: Compare and validate with classic benchmark methods, and output causal network topology visualization, risk spatiotemporal distribution, ranking of key risk nodes, scenario risk decomposition, and accident propagation time window.
[0010] A risk assessment system based on dynamic upgrading and enhanced environmental evolution of the domino effect includes: The hardware acquisition layer includes: Sensor arrays are used to collect temperature, pressure and gas concentration sensing signals from various monitoring points in industrial facilities such as offshore energy systems in real time. The data acquisition unit is communicatively connected to the sensor array and is used to receive and convert the sensing signals into digital process parameter data. A communication interface, connected to the data acquisition unit, is used to transmit the process parameter data to an external network via an industrial communication network. In addition, a computing processing layer is communicatively connected to the hardware acquisition layer. The computing processing layer includes one or more processors and a memory storing computer program instructions. When the processor executes the computer program instructions, it implements the domino effect risk assessment method based on dynamic upgrades and environmental evolution enhancement.
[0011] The present invention has the following beneficial effects: This invention proposes a domino effect risk assessment method and system based on dynamic upgrading and environmental evolution enhancement. It overcomes the technical limitations of existing static methods, such as their inability to reflect dynamic evolution, the inefficiency of simulation methods, and the lack of physical constraints and uncertainty quantification in network methods. It innovatively employs a dynamic upgrading and interpretable nonlinear (DEFEND) technical framework, constructing a complete technical system integrating environmental evolution enhancement, self-triggering mechanisms, multi-scale causal networks, system-level risk quantification, and uncertainty quantification. This internalizes the dynamic propagation process of the domino effect into causal network topology evolution, achieving dynamic, multi-scale, interpretable, and verifiable risk assessment of the domino effect. Based on an integrated hardware and software architecture, this invention can be directly deployed in offshore wind farm monitoring terminals, oil and gas platform safety control systems, and digital twin platforms.
[0012] Regarding dynamic evolution, this invention achieves dynamic simulation of disturbance rejection degradation, continuous source amplification, and enhanced synergistic coupling through an environmental evolution enhancement module. This accurately characterizes the temporal evolution of accident propagation, overcoming the fundamental deficiency of existing static methods such as Cozzani and Landucci, which cannot reflect equipment weakening, source amplification, and multi-source synergistic dynamic effects. Regarding secondary accident evolution, this invention achieves dynamic evolution and temporal triggering among various accident types, such as fires, structural collapses, and secondary explosions from gas leaks, through a self-triggering mechanism. This realistically simulates the chain reaction of accident types, overcoming the limitation of existing methods that assume fixed accident types. Regarding multi-scale coupling analysis, this invention constructs a three-layer causal network at the equipment, station, and regional levels. Through a three-step method of strict adjacent-layer rules and causal network filtering, it achieves cross-scale domino effect coupling analysis, overcoming the limitations of single-scale assessment. Regarding interpretability, this invention achieves transparent analysis of the domino effect propagation path through causal network topology visualization. The probability of each propagation edge can be traced back to the specific physical mechanism and coupling strength. The risk of each node can be decomposed into the product of the initial probability, propagation probability, and consequence severity. The causal network propagates strictly from top to bottom, with the cause always preceding the effect, effectively supporting safety design optimization and the identification of key risk nodes. Regarding uncertainty quantification, this invention outputs the mean, standard deviation, and confidence interval of risk probabilities through Monte Carlo simulation. Sensitivity analysis identifies key influencing parameters and their main and total effect indices, providing a quantitative basis for parameter optimization and prioritizing risk mitigation measures, overcoming the deficiency of existing deterministic assessments that cannot provide confidence levels for results.
[0013] In terms of experimental verification, this invention was systematically validated using two typical scenarios: the Deepwater Horizon blowout accident and an offshore floating wind farm. The Deepwater Horizon accident reproduction results show that this invention successfully identified all eight key propagation nodes, achieving a 100% recall rate. The standard deviation of 10 random seed replication experiments was 0%, demonstrating highly stable results. In contrast, the Cozzani method identified only three nodes, and the Landucci method only identified one node. The propagation time predicted by this invention deviated from the actual propagation time reported by the CSB investigation by ±1 to 3 minutes, proving the physical reliability of the method. The offshore floating wind farm application results show that this invention successfully constructed a complete physical network of 31 nodes and 40 edges, successfully identifying propagation paths of non-thermal mechanisms such as anchor chain dragging, mechanical overload, and electrical faults. These paths all yielded zero probability in the comparative method that only handled thermal radiation mechanisms. The system's peak risk increased by 191% compared to the traditional simplified model, verifying the applicability of this invention in cross-domain, multi-scale scenarios.
[0014] Furthermore, the method of this invention employs an analytical computational framework, with a single dynamic assessment taking only seconds, representing a three to five-order-of-magnitude improvement over traditional CFD numerical simulations. This meets the engineering requirements for batch risk assessment, real-time emergency response, and online digital twin simulations. This invention can be widely applied in offshore oil and gas platforms, floating offshore wind farms, offshore substations, liquefied natural gas receiving terminals, and port hazardous chemical storage facilities. It can automatically complete the dynamic risk quantification of domino effects, propagation path identification, ranking of key risk nodes, prediction of accident time windows, and optimization of emergency plans. It possesses significant application value and promising prospects for industrialization, demonstrating remarkable industrial applicability.
[0015] Other beneficial effects of the embodiments of the present invention will be further described below. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating the overall process of the DEFEND dynamic risk assessment method according to an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the dynamic update of the three-component mechanism of Environmental Evolution Enhancement (EEA) in an embodiment of the present invention. Figure 3 This is a timing diagram showing the dynamic evolution of the self-triggering mechanism and secondary accidents in an embodiment of the present invention. Figure 4 This is a schematic diagram of the multi-scale causal network architecture and strict adjacency layer rules in an embodiment of the present invention; Figure 5 This is a block diagram of the overall architecture of the DEFEND dynamic risk assessment system according to an embodiment of the present invention. Detailed Implementation
[0017] The embodiments of the present invention will be described in detail below. It should be emphasized that the following description is merely exemplary and not intended to limit the scope and application of the present invention.
[0018] This invention aims to overcome the technical pain points of existing static methods that cannot reflect dynamic evolution, simulation methods that are inefficient, and network methods that lack physical constraints and uncertainty quantification. It proposes a dynamic risk assessment method and system for domino effects based on a dynamic upgrade and interpretable nonlinear theoretical framework. By enhancing environmental evolution, using self-triggering mechanisms and multi-scale causal networks, the dynamic propagation process is internalized and topologically evolved, thereby achieving dynamic, multi-scale, interpretable, and verifiable risk assessment of domino effects.
[0019] See Figures 1 to 4 This invention provides a domino effect risk assessment method based on dynamic upgrading and enhanced environmental evolution, comprising the following steps: Step S1, Initial Propagation Probability Calculation: For equipment networks of industrial facilities such as offshore energy systems, the initial coupling strength and initial propagation probability of each propagation path of the domino effect are calculated based on the causal strength quantification method, as well as the initial cumulative propagation probability of each node, to provide an initial state benchmark for dynamic risk assessment.
[0020] In some embodiments, the calculation of the initial coupling strength in step S1 is based on the power function form of the physical threshold ratio, and the propagation path is screened and corrected by the interaction matrix. The influence of distance and medium on propagation is quantified by the spatiotemporal attenuation factor, and the initial propagation probability of each propagation path is determined based on the coupling strength value. The initial cumulative propagation probability of each node is calculated based on the independence assumption of each incoming edge propagation path to characterize the comprehensive probability that the node is triggered by any upstream accident at the initial time.
[0021] Step S2, Enhanced Environmental Evolution: For continuous accident scenarios, the coupling strength and propagation probability of each propagation path are dynamically updated through three sub-mechanisms: degradation of anti-disturbance strength, continuous source amplification, and enhanced cooperative coupling, so as to realize the temporal evolution simulation of accident propagation.
[0022] like Figure 2 As shown, in some embodiments, the environmental evolution enhancement in step S2 includes: a disturbance resistance degradation sub-mechanism: based on the law that the disturbance resistance of the target equipment under continuous load degrades exponentially with the duration of the load, the cumulative effects of material aging, thermal fatigue, and structural damage are simulated to weaken the disturbance resistance, thereby increasing the coupling strength with the duration of the load; a continuous source amplification sub-mechanism: based on the law that the release intensity of the continuous accident source amplifies linearly with the duration of the accident, the amplification effect of the increase in well flow and the boiling expansion process of the storage tank on the intensity of the accident source is simulated, thereby further increasing the coupling strength with time; a synergistic coupling enhancement sub-mechanism: when multiple accident sources act on the same target node simultaneously, a synergistic coupling index is introduced to correct the independent superposition calculation method of the multi-source propagation probability, so that the cumulative propagation probability of the target node changes nonlinearly with the value of the synergistic coupling index, in order to characterize physical effects such as the superposition of multi-source thermal radiation and the synergistic acceleration of multi-path gas diffusion.
[0023] In some embodiments, in the cooperative coupling enhancement sub-mechanism, the non-triggering probability of the target node under multi-source action is calculated based on the propagation probability of each propagation path as the product of the non-triggering probabilities of each path. The cooperative coupling index acts on this product in the form of a power exponent, making the cumulative propagation probability of the target node equal to one minus the product after the power exponent correction. When the cooperative coupling index is greater than 1, the triggering probability of the target node increases compared to the independent propagation case, representing the cooperative enhancement effect. When the cooperative coupling index is less than 1, the triggering probability of the target node decreases compared to the independent propagation case, representing the competitive inhibition effect.
[0024] Step S3, Self-Triggering Mechanism and Dynamic Evolution of Secondary Accidents: When an accident occurs at a node, a secondary accident of that node is triggered based on the self-triggering probability and the delay time window function, thereby realizing the dynamic evolution of accident types and the time-sequence triggering of secondary accidents.
[0025] like Figure 3 As shown, in some embodiments, the self-triggering mechanism and dynamic evolution of secondary accidents in step S3 include: calculating the self-triggering probability based on the product of the probability that the current accident has occurred and the conditional probability that the current accident triggers the target secondary accident, combined with a delay time window function; the delay time window function takes a value of zero before the delay time arrives and a value of one after the delay time arrives, so as to control the triggering time of the secondary accident; incorporating the calculated self-triggering probability into the Bernoulli sampling process, and modifying the load vector according to the delay time when the event is dequeued, so as to realize the dynamic evolution of the accident type and the timing triggering of secondary accidents.
[0026] Step S4: Construction of multi-scale causal network: Construct a three-layer causal network including equipment level, station level and regional level to realize cross-scale domino effect coupling analysis.
[0027] In some embodiments, the multi-scale causal network construction in step S4 includes: device-level scale: characterizing the causal propagation path between components within a single device; site-level scale: characterizing the coupling propagation between multiple devices and components at the single platform or single wind turbine level, with nodes organized according to a causal hierarchy, the causal hierarchy including an initial event layer, an intermediate propagation layer, and a terminal consequence layer; and region-level scale: characterizing the cross-structure propagation path between multiple platforms and multiple wind turbines.
[0028] Step S5, Causal Network Filtering: Perform three-step filtering on the constructed multi-scale causal network: edge probability filtering, breadth-first search unreachable node removal, and isolated initial event node removal, to generate a concise and effective causal topology.
[0029] like Figure 4 As shown, in some embodiments, the causal network filtering in step S5 includes a strict adjacency layer rule and a three-step filtering method: the strict adjacency layer rule is that causal edges are only allowed to exist between adjacent layers, cross-layer causal jumps are prohibited, and the causal network is ensured to propagate strictly from top to bottom; the three-step filtering method is as follows: first, remove weak propagation edges with a propagation probability lower than a preset threshold; second, perform a breadth-first search starting from nodes with no incoming edges or initial event layer nodes to remove unreachable nodes; third, remove initial event layer nodes with no outgoing edges.
[0030] Step S6, System-level risk quantification: Based on node-level risk and the importance weight of nodes in the causal network, calculate the system-level comprehensive risk and output the accident propagation time window.
[0031] In some embodiments, the system-level risk quantification in step S6 includes: calculating the node-level risk of each node based on the product of the cumulative probability of each node being triggered and the severity of the node's consequences; quantifying the importance weight of each node in the causal network based on graph centrality indices; calculating the system-level comprehensive risk based on the weighted sum of each node-level risk and its corresponding importance weight, wherein the system-level comprehensive risk is dynamically updated over time to reflect the dynamic evolutionary effect of enhanced environmental evolution; and calculating the accident propagation time based on a physical model, wherein the physical model comprehensively considers the gas buoyancy rise time, the accumulation time of the lower limit of flammability concentration, and the ignition delay time, and outputs the accident propagation time range and expected value.
[0032] Step S7, Uncertainty Quantification: The input parameters are repeatedly sampled through Monte Carlo simulation to calculate the probability distribution and confidence interval of system-level risk, and key influencing parameters are identified through sensitivity analysis.
[0033] In some embodiments, the uncertainty quantification in step S7 includes: Monte Carlo simulation: using multiple random seeds for repeated sampling, independently running the complete dynamic evolution process for each group of sampled parameters and calculating the system-level risk, outputting the mean, standard deviation and confidence interval of the system-level risk; Sensitivity analysis: using quasi-Monte Carlo sequences for sampling, calculating the main effect index and total effect index of each input parameter, and identifying key influencing parameters and their interaction effects.
[0034] Step S8, Benchmark Validation and Interpretability Output: Compare and validate with classic benchmark methods, and output causal network topology visualization, risk spatiotemporal distribution, ranking of key risk nodes, scenario risk decomposition, and accident propagation time window.
[0035] In some embodiments, the benchmark verification and interpretability output in step S8 includes: comparing and verifying with classical benchmark methods, calculating recall, precision, Spearman rank correlation coefficient, and Kendall rank correlation coefficient; outputting multidimensional interpretable results: visualizing the causal network topology to display the propagation path, node probability, and edge weight; displaying the evolution curve of risk over time for each node in the spatiotemporal distribution of risk; identifying key risk nodes based on importance weight ranking; decomposing the system risk contribution ratio according to the initial event type; and outputting the accident propagation time range and expected value for emergency decision-making.
[0036] See Figure 5 This invention also provides a domino effect risk assessment system based on dynamic upgrades and enhanced environmental evolution. The system includes a hardware acquisition layer and a computational processing layer that is communicatively connected to the hardware acquisition layer.
[0037] The hardware acquisition layer includes a sensor array, a data acquisition unit, and a communication interface. The sensor array is used to acquire real-time temperature, pressure, and gas concentration sensing signals from various monitoring points in industrial facilities, such as offshore energy systems. The data acquisition unit is communicatively connected to the sensor array and is used to receive and convert the sensing signals into digitized process parameter data. The communication interface is connected to the data acquisition unit and is used to transmit the process parameter data externally via an industrial communication network.
[0038] The computational processing layer includes one or more processors and a memory storing computer program instructions. When the processor executes the computer program instructions, it implements the domino effect risk assessment method based on dynamic upgrades and environmental evolution enhancement as described in any of the above embodiments.
[0039] In some embodiments, the hardware acquisition layer further includes a digital twin platform interface. The digital twin platform interface is connected to the communication interface and is used to push the process parameter data and the risk assessment results output by the computational processing layer to an external digital twin platform, and to receive scenario configuration parameters issued by the digital twin platform to support real-time security simulations and dynamic optimization of emergency plans.
[0040] In some embodiments, when the computational processing layer executes the environmental evolution enhancement, the disturbance immunity degradation sub-mechanism is configured as follows: based on the exponential degradation of the disturbance immunity of the target device under continuous load with the duration of the load, the weakening effect of material aging, thermal fatigue, and structural damage accumulation is simulated to increase the coupling strength with the duration of the load. The continuous source amplification sub-mechanism is configured as follows: based on the linear amplification of the release intensity of the continuous accident source with the duration of the accident, the time amplification effect of the release intensity of the accident source is simulated to further increase the coupling strength with time. The cooperative coupling enhancement sub-mechanism is configured as follows: when multiple accident sources act on the same target node simultaneously, a cooperative coupling index is introduced to correct the independent superposition calculation method of the multi-source propagation probability in a power-law form, so that the cumulative propagation probability of the target node changes nonlinearly with the value of the cooperative coupling index.
[0041] In some embodiments, when the computational processing layer performs system-level risk quantification, the physical model of the accident propagation time includes: for a gas leak to explosion scenario, calculating the accident propagation time based on a combination of gas buoyancy rise time, flammable lower limit concentration accumulation time, and ignition delay time; for a fire to structural collapse scenario, calculating the accident propagation time based on the heating time required for the steel structure to rise from the initial temperature to the critical temperature; each physical model outputs the expected value, lower bound, and upper bound of the accident propagation time.
[0042] The present invention provides a dynamic risk assessment method and system based on dynamic upgrading and interpretable nonlinear domino effect, which effectively overcomes the technical pain points of existing static methods that cannot reflect dynamic evolution, simulation methods that are inefficient, and network methods that lack physical constraints and uncertainty quantification. It solves the technical problem of the difficulty in dynamically, multi-scale, interpretable, and verifiable quantitative assessment of domino effects in marine energy systems.
[0043] The following further describes specific embodiments of the present invention, algorithm examples, and experimental verification.
[0044] This invention presents a dynamic risk assessment method and system for domino effects based on a dynamic escalation and interpretable nonlinear theory framework. It innovatively establishes a dynamic escalation and interpretable nonlinear (DEFEND) framework, constructing a complete technical system encompassing environmental evolution enhancement (EEA), self-triggering mechanisms, multi-scale causal networks, system-level risk quantification, and uncertainty quantification. This internalizes the dynamic propagation process of the domino effect as a causal network topology evolution, achieving dynamic, multi-scale, interpretable, and verifiable risk assessment of the domino effect. Based on causal strength quantification and propagation probability calculation (using the coupling strength C and propagation probability P_{ij} provided by the HADES causal quantification method proposed in the applicant's concurrent application), this invention focuses on solving the problems of dynamic evolution, multi-scale coupling, and system-level risk quantification.
[0045] A risk assessment method based on dynamic upgrading and environmental evolution-enhanced domino effect.
[0046] The method of the present invention includes the following steps: S1. Initial Propagation Probability Calculation: Based on a causal strength quantification method (such as the HADES method), calculate the initial coupling strength C and initial propagation probability P_{ij} for each propagation path, as well as the initial cumulative propagation probability P_{node} for each node. This step provides an initial state baseline for the dynamic evaluation of this invention.
[0047] S2, Environmental Evolution Enhancement (EEA): For long-term continuous accident scenarios (such as blowouts, persistent fires, and the continuous effects of typhoons), an environmental evolution enhancement module is introduced to dynamically update coupling strength and propagation probability, including three sub-mechanisms: (1) Sub-mechanism of disturbance immunity strength degradation: The disturbance immunity strength I_{t} of the target equipment degrades over time under continuous loading, simulating the cumulative effects of material aging, thermal fatigue, and structural damage: I t (t) = I t0 × exp(-δ × t) (1) In the formula, I_{t0} is the initial disturbance rejection strength, δ is the degradation rate (based on Eurocode 3 Part 1.2, approximately 50% yield strength loss of steel structures at 600°C corresponds to δ≈0.4 h^{-1}), and t is the duration of loading. This sub-mechanism causes the coupling strength C=(I_{a} / I_{t}(t))^{n} to increase with time (because I_{t} decreases), and the propagation probability to increase with time, reflecting the physical characteristic of the equipment gradually "weakening" under continuous loading.
[0048] (2) Continuous source amplification mechanism: The release intensity of the continuous accident source amplifies over time, simulating physical processes such as increased well blowout flow and tank boiling expansion. I a (t) = I a0 × (1 + γ × t) (2) In the formula, I_{a0} represents the initial intensity of the effect, and γ is the source amplification factor (determined by the physical characteristics of the accident type; for blowouts, γ is determined by the reservoir pressure decrease curve, and for fires, γ is determined by the fuel evaporation rate). This sub-mechanism causes the coupling strength C to increase further over time, and the propagation probability to rise rapidly, reflecting the physical characteristics of the accident source "gradually getting out of control".
[0049] (3) Synergistic Coupling Enhancement Mechanism: When multiple sources act on the same target simultaneously, the synergistic effect increases the propagation probability nonlinearly. A synergistic coupling exponent α is introduced: P node [target] = 1 - (1 - P init ) × ∏(1 - P node [source] × P ij ) α (3) In the formula, α is the cooperative coupling index. When α>1, it indicates cooperative enhancement; when α<1, it indicates competitive inhibition; and when α=1, it indicates independent propagation.
[0050] Mathematical proof of synergistic enhancement: Since ∏(1-P_{ij})∈[0,1], when α>1, [∏(1-P_{ij})]^{α}< ∏(1-P_{ij}), therefore 1-[∏(1-P_{ij})]^{α} > 1-∏(1-P_{ij}), meaning the propagation probability of the target node increases. Physically, this means that synergistic effects such as multi-source thermal radiation and multi-path gas diffusion accelerate equipment failure. A typical synergistic enhancement mode α=1.15 is only activated in extreme accident scenarios (such as well blowouts and typhoons).
[0051] The value of the cooperative coupling index α can be determined by combining theoretical analysis, accident inversion, and conservative selection. The theoretical analysis is based on the mathematical property that the probability of a target node not triggering during multi-source independent propagation is the product of the probabilities of each path not triggering. Introducing the cooperative coupling index corrects this product in a power-law manner, ensuring the correctness of the cooperative enhancement direction. The accident inversion uses the Deepwater Horizon blowout accident as a calibration scenario. The actual triggering time sequence of the accident is compared with the model prediction results, and the inversion determines that the value range of α is 1.12 to 1.18, with a conservative median of 1.15. The discrepancy between the predicted propagation time and the CSB investigation report proves the physical reliability of the calibration. The conservative value is a safe lower bound based on the inversion median, with 1.15 as the conservative lower bound. For extreme vapor cloud explosion or blowout scenarios, α can be increased to 1.2~1.3; for conventional pool fire scenarios, α can be 1.05~1.10. In the code implementation, α is a globally configurable parameter, and its value varies depending on the collaborative mode: for thermal radiation superposition scenarios, α ranges from 1.05 to 1.10; for shock wave focusing scenarios, α ranges from 1.15 to 1.25; for multi-toxic gas mixing scenarios, α ranges from 1.10 to 1.15; and for multi-path gas diffusion scenarios, α ranges from 1.15 to 1.20. α is only enabled in extreme accident scenarios, and its on / off state is controlled by configuration parameters. In normal single-source accident scenarios, α is 1, meaning collaborative enhancement is not enabled.
[0052] The EEA module outputs dynamically updated coupling strength C(t) and propagation probability P_{ij}(t), realizing the temporal evolution simulation of accident propagation.
[0053] S3. Self-triggering mechanism and dynamic evolution of secondary accidents: When an accident A occurs at a node, it can trigger its own secondary accident B (such as fire → structural collapse, gas leak → secondary explosion, electrical fault → fire, etc.) after a delay time Δt, thus realizing the dynamic evolution of accident types.
[0054] P self (A→B, Δt) = P A × P trigger (A→B) × H(Δt) (4)
[0055] In the formula, P_{A} represents the probability that event A has occurred, P_{trigger}(A→B) represents the conditional probability that A triggers B, and H(Δt) is the delay time window function (H=0 before time Δt, H=1 after time Δt). The self-trigger probability P_{self} is added to the Bernoulli sampling, and the delay time modifies the load vector when the event is dequeued, realizing the dynamic evolution of event types and the time-sequential triggering of secondary events.
[0056] The calculation of the delay time Δt uses an independent physical model for different propagation mechanisms. Each model outputs a triplet of expected value, lower bound, and upper bound to represent the uncertainty interval.
[0057] For scenarios where gas leaks and rises to the superstructure, causing an explosion, a gas buoyancy rise model is used to calculate Δt. This model calculates the buoyancy rise time based on the Britter-McQuaid turbulent plume model, calculates the accumulation time of the lower limit of flammability based on the assumption of uniform mixing, and superimposes the ignition delay time based on statistical data.
[0058] For scenarios where gas migrates downwards through floor openings to lower compartments and causes an explosion, a downward gas migration model is used to calculate Δt. This model calculates the migration time based on pressure-driven flow and uses twice the accumulation time of the lower flammable limit concentration to characterize the physical characteristics of low downward mixing efficiency.
[0059] For scenarios where smoke or toxic gases drift into the accommodation cabins through the HVAC system, a smoke drift model is used to calculate Δt. This model calculates the drift time based on the HVAC airflow and natural convection, calculates the toxicity accumulation time based on the immediate life-threatening or health-threatening concentration threshold, and superimposes the HVAC system response lag time.
[0060] For scenarios where direct flame or thermal radiation causes steel structures to heat up to the critical temperature and subsequently collapse, a steel structure heating and ignition model is used to calculate Δt. This model is based on the Eurocode 3 Part 1-2 steel structure heating formula to calculate the heating time required for the steel plate to rise from the initial temperature to the critical temperature. A distance correction factor is used to quantify the effect of reduced viewing angle caused by thermal diffusion. This correction factor increases linearly with the source-target distance.
[0061] Each of the physical models is constructed based on explicit standards and accident statistics, and their uncertainty ranges are shown in Table 1 below: Table 1
[0062] Where d is the source-target distance in meters; the time unit is seconds, which are converted to hours when outputting.
[0063] S4. Construction of Multi-Scale Causal Networks: Constructing a three-layer causal network at the device level (Small), station level (Meso), and regional level (Large) to achieve cross-scale domino effect coupling analysis: (1) Small scale: Single equipment component level causal chain, which describes the causal propagation path between components inside the equipment (such as wind turbine blade failure → gearbox overload → hub failure → tower structure damage; drilling platform moon pool gas diffusion → mud treatment module → drilling platform → superstructure). (2) Meso scale: a single platform or single wind turbine level causal network, which describes the coupling propagation between multiple devices and components. The nodes are organized in layers according to the causal hierarchy L0 (initial event), L1 (intermediate propagation), and L2 (terminal consequence); (3) Large scale: Wind field or regional causal network, characterizing cross-structure propagation between multiple platforms and multiple wind turbines (such as wind turbine anchor chain breakage → floating body drift → adjacent wind turbine collision → submarine cable failure → offshore substation power outage).
[0064] S5. Strict Adjacency Layer Rules and the Three-Step Method for Causal Network Filtering: Multi-scale network construction follows strict adjacency layer rules to ensure unidirectional causality and rational network topology. Strict Adjacent Layer Rule: Causal edges are only allowed between adjacent layers (l_{t}=l_{s}+1), and cross-layer causal jumps are prohibited (e.g., L0→L2, L1→L3). This rule ensures that the causal network propagates strictly from top to bottom, with causes always preceding effects, and there is no reverse causality or cross-layer jumps.
[0065] The three-step method of causal network filtering ensures a streamlined and effective network: Step 1 – Edge Probability Filtering: Remove weak propagation edges with propagation probability P_{ij} < 0.001 to eliminate numerical noise; Step 2 – BFS Unreachable Node Removal: Starting from the trigger source (node with no incoming edges or L0 initial event node), perform a breadth-first search to remove unreachable nodes and preserve valid propagation paths; Step 3 – Removal of isolated L0 nodes: Initial event nodes with no outgoing edges are removed from the graph because they do not cause any propagation.
[0066] In some embodiments, the strict adjacency rule allows no exceptions; that is, causal edges are only allowed to exist between adjacent levels, and cross-level causal jumps are prohibited. This is the fundamental guarantee of the interpretability of causal networks. The reason is that cross-level causal jumps would make the causal network topology uninterpretable, the existence of intermediate layer nodes would become meaningless, the causal chain would be broken, the propagation path would be untraceable, and cross-level jumps might lead to a reversal of causal timing, violating the basic logic that cause always precedes effect.
[0067] For cross-layer propagation scenarios where an initial event directly leads to distant consequences in physical reality, a hierarchical reorganization approach is adopted instead of allowing cross-layer edges. This means that nodes that physically may directly cause distant consequences are remodeled to a level adjacent to the consequence node. Specifically: if a node, although its consequences are severe, is itself caused by an upstream event, it is placed in the intermediate propagation layer (L1) instead of the initial event layer (L0), maintaining a strictly adjacent-layer propagation structure in the causal chain: gas leak / ignition (L0) → explosion (L1) → structural collapse (L2); or, distant consequence nodes are modeled as intermediate propagation layers (L1), and their consequences are in the terminal consequence layer (L2). The node's level is determined by its position in the entire causal chain, not by the severity of the consequences.
[0068] In offshore floating wind farm applications, the initial version placed the mooring system in the terminal consequence layer (L2). However, its initial failure rate, based on statistical frequency, belongs to the initial event, leading to a hierarchical contradiction where the L2 node has no incoming edges. The solution is to move the mooring system back to the initial event layer (L0) and simultaneously move the floating body from the terminal consequence layer (L3) to the intermediate propagation layer (L1), so that the mooring system (L0) → floating body (L1) → tower (L2) forms a complete adjacent layer propagation chain, with all edges satisfying the strict adjacent layer rules and no cross-layer jumps.
[0069] S6. System-level risk quantification: Based on node-level risk, calculate the overall system-level risk. The core formula is as follows: R node = P node × Consequence node (5) R system = ∑(R node × Importance node (6) In the formula, Consequence_{node} represents the severity of node consequences (multi-dimensional consequences indicators such as risk of casualties, economic losses, and environmental impact), Importance_{node} represents the importance weight of a node in the causal network (quantifying the node's contribution to the overall system risk based on graph theory indicators such as PageRank or betweenness centrality), and ∑ represents the summation over all nodes. The system-level risk R_{system} updates dynamically over time, reflecting the dynamic evolution of the EEA module.
[0070] In some embodiments, the Importance_node is a composite weight used to measure the risk amplification capability of a node in a causal network, i.e., the contribution of the node's failure to the domino effect in the entire system. It and the node consequence severity (Consequence_node) are two mutually orthogonal dimensions, and the node consequence severity has been independently modeled in the node-level risk R_node.
[0071] The Importance_node can be defined from the following three levels depending on the application scenario: The first level is the importance of risk elimination, and its expression is:
[0072] This metric is defined as the ratio of node-level risk to system-level risk, that is, the proportion by which system risk decreases after removing the node. The severity of consequences is implicitly included through node-level risk.
[0073] The second level is topological bottleneck centrality, which is calculated using Brandes' algorithm with weighted betweenness centrality:
[0074] The causal strength of the propagation path is used as the weight to quantify how many shortest causal propagation paths a node is located on, thus identifying the bottleneck nodes that the domino propagation must pass through.
[0075] The third level is the composite criticality, which is composed of a weighted sum of three indicators:
[0076] in, The sum of the outgoing edge causal strengths represents the domino driving force of the node; The sum of the causal strengths of the incoming edges represents the triggerability of a node; Weighted betweenness centrality is used to characterize the critical path contribution of a node; the three weights can be adjusted and configured by the user according to the specific application scenario.
[0077] For nodes in densely populated areas or other locations with weak graph connectivity but extremely severe consequences of failure, their high severity is directly assigned to Consequence_node and incorporated into the node-level risk calculation. Even if the node's topological centrality is low, its contribution to system-level risk remains significant. In dynamic risk assessment, Importance_node can also be represented by a time integral.
[0078] This approach considers both the temporal evolution of node risk and the total system risk, to more comprehensively reflect the dynamic risk contribution of nodes.
[0079] Accident propagation time is accurately calculated using a physical model: taking into account the gas buoyancy rise time, the lower flammability limit (LFL) concentration accumulation time, and the ignition delay time, the accident propagation time range [t_{min}, t_{max}] and the expected value E[t] are output, providing a precise time window for emergency response decision-making.
[0080] S7. Uncertainty Quantification: Quantifying the uncertainty of results through Monte Carlo simulation and Sobol sensitivity analysis. (1) Monte Carlo simulation: Multiple random seed repeated sampling is used (typical configuration 10,000 samplings). The system-level risk R_{system} is calculated independently for each sampling, and the mean μ, standard deviation σ and 95% confidence interval of the risk probability are output. [μ-1.96σ, μ+1.96σ]; (2) Sobol sensitivity analysis: Sobol quasi-Monte Carlo sequence (typical configuration 100,000 samples) was used to calculate the main effect index S_i and total effect index S_{Ti} of each input parameter, identify key influencing parameters and their interaction effects, and provide a basis for parameter optimization and priority ranking of risk mitigation measures.
[0081] In some embodiments, the Monte Carlo simulation employs a two-layer sampling protocol: the outer layer samples the uncertainty of the input parameters, and the inner layer performs Bernoulli sampling on probabilistic events during the dynamic evolution process, in order to simultaneously capture parameter uncertainty and path uncertainty.
[0082] The outer sampling objects include the initial failure rate of each device node, the degradation rate of material immunity strength, the amplification factor of accident source release intensity, the cooperative coupling index, and mechanism-specific parameters. The distribution type is selected based on the physical characteristics of the parameters during sampling, as detailed in Table 2 below: Table 2
[0083] The complete uncertainty quantification process has a nested structure: For (Outer layer MC sampling): 1. Extract parameter samples
[0084] 2. Initialize the causal network using this set of parameters.
[0085] 3. Dynamic Evolution During Operation (Inner Bernoulli Sampling): For each target node to be judged: a. Calculation (Including EEA synergistic enhancement) b. Trigger determination: Deterministic / Probabilistic Bernoulli / No trigger c. If triggered, handle self-triggered events (Bernoulli sampling). ) d. Update exposure time and cumulative damage 4. Output
[0086] calculate , , .
[0087] The outer sampling process is as follows: Each sampling extracts a set of parameter samples, inputs this set of parameters into the dynamic evolution model, runs a complete causal network evolution, and outputs the system-level risk. After repeated sampling a preset number of times (e.g., 10,000 times), the mean, standard deviation, and confidence interval of the system-level risk are calculated.
[0088] During the dynamic evolution of each outer layer sampling, the inner layer calculates the cumulative propagation probability (adjusted by environmental evolution) at each node to be judged and performs trigger determination: when the cumulative propagation probability is greater than or equal to a first preset threshold (e.g., 0.95), it is a deterministic trigger; when the cumulative propagation probability is less than the first preset threshold but greater than or equal to a second preset threshold (e.g., 0.5), Bernoulli sampling is performed to determine whether to trigger; when the cumulative propagation probability is less than the second preset threshold, it is not triggered. If triggered, Bernoulli sampling is performed based on the self-trigger probability to determine whether to trigger a secondary accident. The delay time is calculated by the physical model, and the load vector is modified when the event is dequeued; at the same time, the exposure time and cumulative damage are updated, and fatigue failure is triggered when the cumulative damage exponent reaches a threshold.
[0089] To verify the stability of the results, the experiment was repeated multiple times with different random seeds. For strong causal chain scenarios, the trigger probability far exceeds the deterministic threshold, and the Bernoulli sampling results are almost unaffected by the seed. For weak causal chain scenarios, the path sensitivity increases, and the path uncertainty can be fully captured by multiple samplings in the outer layer.
[0090] S8. Benchmarking and Interpretability Output: The method is compared and validated against classic benchmark methods (Cozzani, Landucci), calculating metrics such as recall, precision, Spearman rank correlation ρ, and Kendall τ to ensure its reliability and superiority. Multidimensional interpretability results are output. (1) Visualization of causal network topology: displaying propagation paths, node probabilities, and edge weights; (2) Spatiotemporal distribution of risk: Displaying the evolution curve of risk at each node over time; (3) Ranking of key risk nodes: Identifying system bottlenecks based on the ranking of Importance_{node}; (4) Scenario risk decomposition: Decompose the system risk contribution ratio according to the initial event type; (5) Accident propagation time window: Output [t_{min}, t_{max}] and E[t] supply emergency decision.
[0091] A risk assessment system for the domino effect enhanced by dynamic upgrading and environmental evolution.
[0092] The system of this invention is used to implement the above-mentioned dynamic risk assessment method, and includes a hardware acquisition unit and a software algorithm unit connected in sequence, specifically comprising: 1. Initial probability calculation unit: Configure the causal strength quantification module interface, receive the initial coupling strength C and initial propagation probability P_{ij} of each propagation path, calculate the initial cumulative propagation probability P_{node} of each node, and provide an initial state benchmark for dynamic evaluation; 2. Environmental Evolution Enhancement (EEA) Unit: Connected to the initial probability calculation unit, it dynamically updates the coupling strength C(t) and propagation probability P_{ij}(t) through three sub-mechanisms: disturbance resistance strength degradation, continuous source amplification, and cooperative coupling enhancement, thereby realizing the temporal evolution simulation of accident propagation; 3. Self-triggered evolution unit: Connected to the environment evolution enhancement unit, it realizes the dynamic evolution of accident types and the time-series triggering of secondary accidents based on the self-triggered probability P_{self} and the delay time window function H(Δt); 4. Multi-scale causal network construction unit: Connects self-triggered evolution units to construct a three-layer causal network at the device level, station level, and region level. It executes a three-step method of strict adjacent layer rules and causal network filtering to generate a concise and effective multi-scale causal topology. 5. System Risk Quantification Unit: Connects the multi-scale causal network construction unit, calculates the system-level comprehensive risk R_{system} based on the node risk and network importance weight, and outputs the spatiotemporal distribution of risk, the ranking of key risk nodes, and the accident propagation time window; 6. Uncertainty Quantification Unit: Connects to the system risk quantification unit, and quantifies the uncertainty and parameter sensitivity of the assessment results through Monte Carlo simulation and Sobol sensitivity analysis, outputting confidence intervals and parameter sensitivity indices; 7. Verification and Interpretability Output Unit: Connects to the uncertainty quantification unit, compares and verifies with classic benchmark methods, and outputs causal network topology visualization, risk spatiotemporal distribution, ranking of key risk nodes, scenario risk decomposition, and accident propagation time window; 8. Digital Twin Interface Unit: Connects to the verification and interpretability output unit, provides real-time data interfaces and APIs, supports docking with the marine energy system digital twin platform, and enables real-time safety simulation and dynamic optimization of emergency plans.
[0093] Experimental verification
[0094] Example 1: Dynamic Risk Assessment of Deepwater Horizon Well Blowout
[0095] Example 1 uses the Deepwater Horizon blowout accident as the application object to verify the dynamic propagation recognition capability of the present invention in real complex accident scenarios; Example 2 uses an offshore floating wind farm as the application object to verify the applicability of the present invention in cross-domain and multi-scale scenarios.
[0096] 1. Implementation Data Preparation
[0097] Based on the joint investigation report of the U.S. Chemical Safety and Hazards Investigation Board (CSB), the Marine Energy Management and Regulatory Enforcement Bureau (BSEE), and the Coast Guard (USCG), the IChemE Loss Prevention Bulletin (LPB285), and the Transocean platform layout drawings, a DWH accident scenario dataset was constructed. Platform equipment nodes include the drilling rig, moon pool, mud treatment module, superstructure, accommodation cabins, helicopter deck, and blowout preventer (BOP). The physical layout and connectivity between equipment (moon pool - ventilation duct - mud treatment - drilling rig) are strictly modeled according to the platform drawings. The initial event is a wellhead gas leak, and the propagation mechanisms include gas flow (moon pool → ventilation system), smoke toxicity (ventilation → accommodation cabins), direct flame (structural connectivity propagation), and thermal radiation (long-distance radiation).
[0098] 2. Enhanced environmental evolution and dynamic propagation
[0099] Enabling the EEA module: Continuous blowout leakage leads to I_{t} degradation (δ=0.4 h^{-1}, based on Eurocode 3 high-temperature degradation of steel structures), I_{a} amplification (γ determined by the blowout flow rate decline curve), and enhanced synergistic coupling (α=1.15, multi-path gas diffusion synergistically accelerates equipment failure). The dynamically updated coupling strength C(t) and propagation probability P_{ij}(t) increase significantly over time, reflecting the dynamic deterioration characteristics of the accident.
[0100] 3. Self-triggered and multi-scale networks
[0101] The self-triggering mechanism enables a chain reaction of secondary accidents: gas leak → ignition → fire → structural collapse. The delay time Δt is calculated by the physical model (gas buoyancy rise + LFL accumulation + ignition delay). Multi-scale causal network construction: Small scale represents a single-device component chain (moon pool → ventilation → mud treatment); Meso scale represents a platform-level network (multi-module coupled propagation); and Large scale represents a regional network (platform → marine environment).
[0102] 4. System risk quantification and uncertainty
[0103] System-level dynamic risk was calculated using R_{system}=∑(R_{node}×Importance_{node}). Monte Carlo simulations with 10,000 samples and 10 randomized seed replicates showed a standard deviation of 0%, demonstrating highly stable results. Sobol sensitivity analysis identified blowout flow rate, ventilation conditions, and equipment fire resistance limit as key influencing parameters.
[0104] 5. Implementation Results
[0105] This embodiment successfully reproduced all eight key propagation nodes of the DWH accident, with a Recall of 100% (TP = 8 / 8). The propagation time predicted by this invention deviates from the actual propagation time reported by the CSB (±1-3 minutes), demonstrating the physical reliability of the method. Comparison with benchmarks: the Cozzani method only identified 3 / 8 (Recall = 38%, only the thermal radiation mechanism is effective, gas flow path P = 0); the Landucci method only identified 1 / 8 (Recall = 12%, limited by thermal radiation and TTF).
[0106] Experimental Example 2: Multi-scale Dynamic Risk Assessment of Offshore Floating Wind Farms
[0107] 1. Implementation Data Preparation
[0108] This study uses a 200 MW offshore floating wind farm in the South China Sea as the application example. The wind farm is equipped with six NREL 5 MW semi-submersible floating wind turbines (FOWT) and one offshore substation (OSS), with a turbine spacing of 800 m. Each turbine consists of five components: blades, nacelle (including gearbox and hydraulic system), tower, floating body, and mooring system. Environmental conditions were based on measured data from Typhoon "Mangkut" (2024). Initial events included blade breakage (frequency 5 × 10^-3), gearbox failure (2 × 10^-3), hydraulic system failure (1.5 × 10^-3), and anchor chain breakage (3 × 10^-4, based on DNV-OS-E301 statistical frequencies).
[0109] 2. Construction of multi-scale causal networks
[0110] Small scale: Single wind turbine 5-component causal chain (blade → nacelle → tower → float → mooring), terminal node is blade-nacelle combination failure; Meso scale: Single wind turbine 7-node 6-edge network (L0: blade / gearbox / hydraulic system / anchor chain → L1: nacelle / floating body → L2: tower terminal), following strict adjacent layer rules; Large scale: Wind farm level 31-node 40-edge network (6 wind turbines × 5 components + OSS, causal edges are only allowed between adjacent layers), cross-structure propagation path of anchor chain breakage → float drift → collision with adjacent wind turbines. 800 m blade debris propagation is physically determined to be impossible due to P≈0 and removed from the network.
[0111] 3. Physical calculation and dynamic evolution of anchor chain fracture
[0112] Based on the DNV-OS-E301 specification, the anchor chain tension physical model calculates a peak tension T_{peak} = 4520 kN, a breaking load ratio T_{peak} / T_{break} = 0.565, and an anchor chain breakage probability P_{break} = 0.63. After anchor chain breakage, the float drift induces tower bending (DNV-ST-0119), with a conditional probability override P = 0.95 (kinematically inevitable). The foundation failure rate uses the DNV typical value of 3 × 10^{-4}, with no human intervention. The EEA module simulates the dynamic fluctuations and fatigue accumulation of anchor chain tension under continuous typhoon conditions.
[0113] 4. System risk quantification and scenario decomposition
[0114] The system peak risk R_{peak} = 7.98 × 10^{8}, which is 191% higher than the traditional simplified 10-node model (2.74 × 10^{8}). Risk scenario decomposition: typhoon scenario contributes 68.2%, fire scenario contributes 30.1%, aging scenario contributes 1.5%, and ship collision scenario contributes 0.14%. Component risk ranking: mooring system (45.1%) > floating body (20.6%) > engine room (15.6%) > tower (13.0%) > blade (4.7%) > offshore substation (1.1%).
[0115] 5. Implementation Results
[0116] This embodiment verifies the applicability of the invention in cross-domain applications (marine oil and gas → offshore wind power), successfully identifying non-thermal mechanism propagation paths such as anchor chain dragging, mechanical overload, and electrical faults—pathways that would all result in P=0 in the Cozzani / Landucci method because it only handles thermal radiation mechanisms. This invention provides quantitative risk support for the optimization of offshore wind farm safety design, anchoring system maintenance strategies, and the development of typhoon emergency response plans.
[0117] In summary, this invention proposes a domino effect risk assessment method and system based on dynamic upgrading and environmental evolution enhancement. By enhancing environmental evolution, using self-triggering mechanisms and multi-scale causal networks, the dynamic propagation process is internalized and topologically evolved, achieving dynamic, multi-scale, interpretable, and verifiable risk assessment of the domino effect. This effectively overcomes the technical pain points of existing static methods that cannot reflect dynamic evolution, simulation methods that are inefficient, and network methods that lack physical constraints and uncertainty quantification.
[0118] Compared with the prior art, the significant technical advantages of the present invention are reflected in the following aspects: 1. Dynamic temporal evolution, overcoming the limitations of static assessment. This invention, through an environmental evolution enhancement module, achieves dynamic simulation of disturbance immunity degradation, continuous source amplification, and enhanced synergistic coupling, accurately characterizing the temporal evolution characteristics of incident propagation. Existing Cozzani and Landucci methods are static assessments and cannot reflect the dynamic effects of equipment weakening, source amplification, and multi-source synergy.
[0119] 2. Self-triggered secondary accident evolution, breaking through the limitations of single accident types. This invention realizes the dynamic evolution and sequential triggering of secondary accidents such as fire → structural collapse and gas leakage → secondary explosion through a self-triggered mechanism, realistically simulating the chain transformation process of accident types. Existing methods all assume fixed accident types and cannot simulate the evolution of secondary accidents.
[0120] 3. Multi-scale coupling analysis overcomes the limitations of single-scale assessment. This invention constructs a three-layer causal network at the equipment, site, and regional levels. Through a three-step method of strict adjacent-layer rules and causal network filtering, it achieves cross-scale domino effect coupling analysis. In offshore wind farm applications, the system successfully constructed a complete physical network of 31 nodes and 40 edges, enabling wind farm-level risk scenario decomposition and component risk ranking.
[0121] 4. Excellent interpretability, overcoming the limitations of a black box. This invention achieves transparent analysis of the domino effect propagation path through causal network topology visualization. The probability of each propagation edge can be traced back to the specific physical mechanism and coupling strength, and the risk of each node can be decomposed into the product of the initial probability, propagation probability, and consequence severity. The causal network propagates strictly from top to bottom, with the cause always preceding the effect, supporting security design optimization and identification of key risk nodes.
[0122] 5. Quantifying Uncertainty, Overcoming the Limitations of Deterministic Assessment. This invention outputs the mean, standard deviation, and confidence interval of risk probabilities through Monte Carlo simulation, and identifies key influencing parameters and their main and total effect indices through Sobol sensitivity analysis, providing a quantitative basis for parameter optimization and prioritizing risk mitigation measures. Existing methods are all deterministic assessments and cannot provide the confidence level of the results.
[0123] 6. The accuracy and real-world incident reproducibility were verified. This invention achieved full identification of all 8 / 8 propagation nodes in the Deepwater Horizon real-world incident reproduction, with a recall rate of 100% and a standard deviation of 0% in 10 random seed replication experiments, demonstrating highly stable results. In comparison, the Cozzani method identified only 3 / 8, and the Landucci method identified only 1 / 8. The propagation time predicted by this invention deviated from the actual propagation time reported by the CSB investigation, proving the physical reliability of the method.
[0124] 7. High computational efficiency, adaptable to real-time applications. The method of this invention uses an analytical computational framework, with a single dynamic evaluation taking only seconds, which is three to five orders of magnitude faster than traditional CFD numerical simulation. It can meet the engineering needs of batch risk assessment, real-time emergency response, and online digital twin simulation.
[0125] Based on an integrated hardware and software architecture, this invention can automatically complete the dynamic risk quantification of domino effects, propagation path identification, ranking of key risk nodes, prediction of accident time windows, and optimization of emergency plans. It can be widely applied in offshore oil and gas platforms, floating offshore wind farms, offshore substations, liquefied natural gas receiving stations, and port hazardous chemical storage, etc. It has significant industrial applicability and good application value and industrialization prospects.
[0126] This invention also provides a storage medium for storing a computer program, which, when executed, performs at least the methods described above.
[0127] This invention also provides a control device, including a processor and a storage medium for storing a computer program; wherein the processor executes the computer program by performing at least the method described above.
[0128] This invention also provides a processor that executes a computer program, at least performing the methods described above.
[0129] The storage medium can be implemented by any type of non-volatile storage device, or a combination thereof. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), magnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc or CD-ROM; magnetic surface memory can be disk storage or magnetic tape storage. The storage media described in the embodiments of this invention are intended to include, but are not limited to, these and any other suitable types of memory.
[0130] In the several embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.
[0131] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.
[0132] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.
[0133] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0134] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.
[0135] The methods disclosed in the several method embodiments provided by this invention can be arbitrarily combined without conflict to obtain new method embodiments.
[0136] The features disclosed in the several product embodiments provided by this invention can be arbitrarily combined without conflict to obtain new product embodiments.
[0137] The features disclosed in the several method or device embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method or device embodiments.
[0138] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various equivalent substitutions or obvious modifications can be made without departing from the concept of the present invention, and all such modifications, achieving the same performance or application, should be considered within the scope of protection of the present invention.
Claims
1. A risk assessment method based on the domino effect enhanced by dynamic upgrading and environmental evolution, characterized in that, Includes the following steps: S1. Initial propagation probability calculation: For the equipment network of industrial facilities, based on the causal strength quantification method, the initial coupling strength and initial propagation probability of each propagation path of the domino effect, as well as the initial cumulative propagation probability of each node, are calculated to provide an initial state benchmark for dynamic risk assessment. S2. Enhanced Environmental Evolution: For continuous accident scenarios, the coupling strength and propagation probability of each propagation path are dynamically updated through three sub-mechanisms: degradation of anti-disturbance strength, continuous source amplification, and enhanced cooperative coupling, so as to realize the temporal evolution simulation of accident propagation. S3. Self-triggering mechanism and dynamic evolution of secondary accidents: When an accident occurs at a node, a secondary accident of the node is triggered based on the self-triggering probability and the delay time window function, realizing the dynamic evolution of accident types and the time-sequence triggering of secondary accidents. S4. Construction of multi-scale causal network: Construct a three-layer causal network including equipment level, station level and regional level to realize cross-scale domino effect coupling analysis; S5. Causal Network Filtering: Perform three-step filtering on the constructed multi-scale causal network: edge probability filtering, breadth-first search unreachable node removal, and isolated initial event node removal to generate a concise and effective causal topology. S6. System-level risk quantification: Based on node-level risk and the importance weight of nodes in the causal network, calculate the system-level comprehensive risk and output the accident propagation time window; S7. Uncertainty Quantification: Input parameters are repeatedly sampled through Monte Carlo simulation to calculate the probability distribution and confidence interval of system-level risk, and key influencing parameters are identified through sensitivity analysis. S8. Benchmark Validation and Interpretability Output: Compare and validate with classic benchmark methods, and output causal network topology visualization, risk spatiotemporal distribution, ranking of key risk nodes, scenario risk decomposition, and accident propagation time window.
2. The method according to claim 1, characterized in that, The environmental evolution enhancement mentioned in step S2 includes: Disturbance immunity strength degradation sub-mechanism: Based on the law that the disturbance immunity strength of the target equipment under continuous load degrades exponentially with the duration of the load, the cumulative effects of material aging, thermal fatigue and structural damage are simulated to weaken the disturbance immunity strength, so that the coupling strength increases with the duration of the load. The continuous source amplification mechanism: Based on the linear amplification of the release intensity of the continuous accident source with the duration of the accident, the amplification effect of the increase in well flow rate and the boiling expansion process of the storage tank on the intensity of the accident source is simulated, so that the coupling strength further increases with time. Synergistic Coupling Enhancement Sub-Mechanism: When multiple accident sources act on the same target node simultaneously, a synergistic coupling index is introduced to correct the independent superposition calculation method of multi-source propagation probability, so that the cumulative propagation probability of the target node changes nonlinearly with the value of the synergistic coupling index, in order to characterize the physical effect of multi-source thermal radiation superposition and multi-path gas diffusion synergistic acceleration.
3. The method according to claim 2, characterized in that, In the aforementioned cooperative coupling enhancement sub-mechanism, the non-trigger probability of the target node under multi-source action is calculated based on the propagation probability of each propagation path as the product of the non-trigger probabilities of each path. The cooperative coupling index acts on this product in the form of a power exponent, making the cumulative propagation probability of the target node equal to one minus the product after the power exponent correction. When the cooperative coupling index is greater than 1, the trigger probability of the target node increases compared to the independent propagation case, which represents the cooperative enhancement effect; When the cooperative coupling index is less than 1, the trigger probability of the target node decreases compared to the independent propagation case, which represents the competitive inhibition effect.
4. The method according to claim 1, characterized in that, The self-triggering mechanism and the dynamic evolution of secondary accidents mentioned in step S3 include: The self-trigger probability is calculated by multiplying the probability that the current accident has occurred by the conditional probability that the current accident triggers a secondary accident, and combining this with a delay time window function. The delay time window function takes a value of zero before the delay time arrives and a value of one after the delay time arrives, in order to control the triggering time of secondary accidents; The calculated self-triggering probability is incorporated into the Bernoulli sampling process, and the load vector is modified according to the delay time when the event is dequeued, so as to realize the dynamic evolution of the accident type and the timing triggering of secondary accidents.
5. The method according to claim 1, characterized in that, The multi-scale causal network construction described in step S4 includes: Device-level scale: Depicting the causal propagation path between components within a single device; Site-level scale: Characterizes the coupling propagation between multiple devices and components on a single platform or single wind turbine. Nodes are organized according to a causal hierarchy, which includes an initial event layer, an intermediate propagation layer and a terminal consequence layer. Regional scale: Depicting cross-structure propagation paths between multiple platforms and multiple wind turbines.
6. The method according to claim 5, characterized in that, The causal network filtering described in step S5 includes strict adjacent layer rules and a three-step filtering method: The strict adjacency layer rule is as follows: causal edges are only allowed to exist between adjacent layers, cross-layer causal jumps are prohibited, and the causal network is strictly propagated from top to bottom. The three-step filtering method is as follows: First, remove weak propagation edges whose propagation probability is lower than a preset threshold; The second step is to perform a breadth-first search starting from nodes with no incoming edges or initial event layer nodes, and remove unreachable nodes; the third step is to remove initial event layer nodes with no outgoing edges.
7. The method according to claim 1, characterized in that, The system-level risk quantification described in step S6 includes: The node-level risk of each node is calculated by multiplying the cumulative probability of each node being triggered by the severity of the node's consequences. Quantify the importance weight of each node in a causal network based on graph theory centrality index; The system-level comprehensive risk is calculated by weighted summation of the risks at each node level and their corresponding importance weights. This system-level comprehensive risk is dynamically updated over time to reflect the dynamic evolutionary effect of environmental evolution. Accident propagation time is calculated based on a physical model that comprehensively considers gas buoyancy rise time, lower limit of flammability concentration accumulation time, and ignition delay time, and outputs the accident propagation time range and expected value.
8. The method according to claim 1, characterized in that, The uncertainty quantification mentioned in step S7 includes: Monte Carlo simulation: It uses multiple random seeds for repeated sampling, independently runs the complete dynamic evolution process for each group of sampled parameters, calculates the system-level risk, and outputs the mean, standard deviation and confidence interval of the system-level risk; Sensitivity analysis: Quasi-Monte Carlo sequences were used for sampling, and the main effect index and total effect index of each input parameter were calculated to identify key influencing parameters and their interaction effects.
9. The method according to claim 1, characterized in that, The benchmark verification and interpretability output in step S8 includes: The results were compared and validated with classic benchmark methods, and recall, precision, Spearman rank correlation coefficient and Kendall rank correlation coefficient were calculated. Output multidimensional interpretable results: causal network topology visualization showing propagation paths, node probabilities, and edge weights; risk spatiotemporal distribution showing the evolution curve of risk for each node over time; identification of key risk nodes based on importance weight ranking; decomposition of system risk contribution ratio by initial event type; output of accident propagation time range and expected value for emergency decision-making.
10. A risk assessment system based on dynamic upgrading and enhanced environmental evolution of the domino effect, characterized in that, include: The hardware acquisition layer includes: Sensor arrays are used to collect temperature, pressure and gas concentration sensing signals from various monitoring points in industrial facilities in real time. The data acquisition unit is communicatively connected to the sensor array and is used to receive and convert the sensing signals into digital process parameter data. A communication interface, connected to the data acquisition unit, is used to transmit the process parameter data to an external network via an industrial communication network. And a computing processing layer communicatively connected to the hardware acquisition layer, the computing processing layer including one or more processors and a memory storing computer program instructions, wherein when the processor executes the computer program instructions, it implements the domino effect risk assessment method based on dynamic upgrade and environmental evolution enhancement as described in any one of claims 1 to 9.
Citation Information
Patent Citations
Petrochemical port area accident risk area quantitative assessment method
CN105787677A
Hazardous chemical substance fire domino accident dynamic risk analysis method under multi-disaster coupling
CN112529412A
