Disaster chain dynamic risk assessment method and system based on logistic regression

Abstract

The invention discloses a disaster chain dynamic risk assessment method and system based on logistic regression, and the method comprises the steps: defining a novel extreme disaster, constructing a four-dimensional disaster systematicness scoring frame, scoring indexes based on the four-dimensional disaster systematicness scoring frame, and calculating the weight of each index and a disaster comprehensive score based on an information entropy theory, screening disaster chain trigger factors; based on disaster chain triggering, designing a disaster chain topological structure, and training a logistic regression model to dynamically update a failure probability; and calculating a disaster chain risk degree, and evaluating a disaster chain risk based on the disaster chain risk degree. According to the method, a four-dimensional disaster systematic scoring framework is constructed, the sensitivity to a single index score is reduced, and the inapplicability of a traditional risk matrix to non-traditional risks and low-frequency high-loss risks exceeding original design references is overcome. Moreover, a disaster chain topological structure and a training logistic regression model are provided, so that the accuracy and reliability of disaster chain risk prediction are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of disaster assessment, and in particular to a disaster chain dynamic risk assessment method and system based on logistic regression. Background Art

[0002] Recent research on "compound disasters" has found that the destructive power of the combined effects of multiple "single controllable risks" can result in a nonlinear amplification effect. Furthermore, the propagation paths of some disasters are relatively hidden, making the dynamic evolution of risk unpredictable. For facilities like nuclear power plants, which possess highly complex systems and pose a threat to overall societal safety, the triggers of emergencies are diverse, making static models difficult to cover all scenarios. Real-time prediction of evolving disaster risks by inputting scenario-specific parameters is crucial for emergency decision-making, but traditional methods are unable to dynamically adjust the evolutionary logic. Furthermore, with environmental changes such as climate change and technological advancement, several new disaster scenarios have demonstrated that traditional baseline protection levels lag behind the rate of disaster evolution, creating a significant risk of escalating traditional disasters into "new extreme disasters." Traditional risk matrices, which rely on a "probability-impact" matrix to identify and stratify disaster risks, are ill-suited to emerging challenges such as the coupling of multiple hazards and low-frequency, high-loss events at nuclear power plants.

[0003] Taking into account the different temporal characteristics of disaster intensity, which affect the probability of occurrence of each edge of the disaster chain, the existing quantitative research on the evolutionary effects of disaster chains is mainly based on static coupling parameters and expert experience, and has the following significant defects: (1) The coupling parameters are fixed. For example, the probability of event A triggering event B depends on historical statistics or expert experience, and cannot be adjusted in real time with environmental conditions, lacking dynamic adaptability. (2) There is strong subjective dependence. For example, the setting of the coupling coefficient is highly dependent on expert scoring or limited historical cases, which is easy to introduce human bias. For new emergencies, there is a lack of data support and the reliability of parameters is insufficient. (3) There is a lack of real-time response capability. Once the model parameters are set, they cannot be dynamically updated through real-time data, and thus cannot warn of sudden risks, resulting in delayed emergency decision-making and missing the golden time for disposal.

[0004] Therefore, the prior art still has defects. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a disaster chain dynamic risk assessment method and system based on logistic regression to address the above-mentioned defects of the prior art. The technical solution adopted by the present invention is as follows:

[0006] In a first aspect, the present invention provides a method for dynamic risk assessment of disaster chains based on logistic regression, wherein the method comprises:

[0007] Define new extreme disasters, construct a four-dimensional disaster system scoring framework, score indicators based on the four-dimensional disaster system scoring framework, and calculate the weight of each indicator and the comprehensive disaster score based on information entropy theory to screen the triggering factors of the disaster chain;

[0008] Based on the disaster chain trigger, a disaster chain topology is designed, and a logistic regression model is trained to dynamically update the failure probability;

[0009] A disaster chain risk degree is calculated, and the disaster chain risk is assessed based on the disaster chain risk degree.

[0010] In one implementation, the indicators of the four-dimensional disaster system scoring framework include: disaster impact severity, occurrence frequency, cascade effect potential, and data support. Among them, the disaster impact severity is used to measure the level of consequences caused by the disaster, the occurrence frequency refers to the number of disasters occurring within a fixed period, the cascade effect potential is used to evaluate the coupling ability of a single disaster with other disasters and the ability to trigger multi-system chain failures, and the data support is used to take into account the support level and authority of public data.

[0011] In one implementation, the scoring rules of the four-dimensional disaster system scoring framework are as follows: the scoring range of each indicator is 1 to 5 points, breakpoint scoring is performed within the disaster to be studied, the top 30% of the quantitative indicator results are assigned 5 points, the ranking between 30% and 70% is assigned 3 to 4 points, and the bottom 30% is assigned 1 to 2 points.

[0012] In one implementation, the method of calculating the weight of each indicator and the comprehensive disaster score based on information entropy theory and screening the triggering factors of the disaster chain includes:

[0013] Calculate the information entropy of each indicator to correct the indicator weight;

[0014] Using the revised indicator weights, the weighted total score formula is used to obtain the comprehensive disaster score of each disaster sample;

[0015] The disaster with the highest comprehensive score is selected as the triggering factor of the disaster chain.

[0016] In one implementation, each node in the disaster chain topology represents a disaster event, directed edges represent the evolutionary relationship between events, and edge weights are output by a logistic regression model trained based on historical cases and hypothetical scenario sets.

[0017] In one implementation, the training process of the logistic regression model includes:

[0018] Obtain the initial disaster data and secondary disaster data in the same time series, define the time window of initial disaster-secondary disaster, and for each initial disaster event, find out whether there is a secondary disaster event in the subsequent time window. Count the number of co-occurrences and independent occurrences of the two to obtain sample data;

[0019] Redundant features were eliminated based on the Spearman correlation coefficient, and candidate features with an absolute correlation coefficient with the coupling probability greater than a threshold with a medium effect were selected;

[0020] The optimal feature subset is constructed through forward stepwise regression. The feature with the largest AUC improvement is selected and added to the model in each iteration until all available indicators are iterated.

[0021] The logistic regression model was evaluated using the following evaluation indicators: accuracy, precision, recall, F1 score, and ROC-AUC curve.

[0022] In one implementation, the indicators of the disaster chain risk include: dynamic failure probability, coupling function, and loss index.

[0023] In a second aspect, an embodiment of the present invention further provides a disaster chain dynamic risk assessment system based on logistic regression, wherein the system is used to implement the steps of the disaster chain dynamic risk assessment method based on logistic regression described in any of the above solutions, and the system includes:

[0024] The trigger factor determination module is used to define new extreme disasters, build a four-dimensional disaster system scoring framework, score indicators based on the four-dimensional disaster system scoring framework, and calculate the weight of each indicator and the comprehensive disaster score based on information entropy theory to screen the trigger factors of the disaster chain;

[0025] A logistic regression model training module is used to design a disaster chain topology based on the disaster chain trigger and train a logistic regression model to dynamically update the failure probability;

[0026] The disaster chain risk assessment module is used to calculate the disaster chain risk degree and assess the disaster chain risk based on the disaster chain risk degree.

[0027] In the third aspect, an embodiment of the present invention also provides a terminal, wherein the terminal includes a memory, a processor, and a disaster chain dynamic risk assessment program based on logistic regression stored in the memory and runnable on the processor. When the processor executes the disaster chain dynamic risk assessment program based on logistic regression, the steps of the disaster chain dynamic risk assessment method based on logistic regression of any one of the above-mentioned schemes are implemented.

[0028] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, wherein a dynamic risk assessment program for disaster chains based on logistic regression is stored on the computer-readable storage medium, and the dynamic risk assessment program for disaster chains based on logistic regression implements the steps of the dynamic risk assessment method for disaster chains based on logistic regression described in any one of the above-mentioned schemes on the computer-readable storage medium.

[0029] Beneficial effects: Compared with the existing technology, the present invention provides a dynamic risk assessment method for disaster chains based on logistic regression. The present invention first defines a new type of extreme disaster, constructs a four-dimensional disaster system scoring framework, scores indicators based on the four-dimensional disaster system scoring framework, and calculates the weight of each indicator and the comprehensive disaster score based on the information entropy theory to screen the disaster chain triggering factors. Then, based on the disaster chain trigger, the disaster chain topology structure is designed, and the logistic regression model is trained to dynamically update the failure probability. Finally, the disaster chain risk degree is calculated, and the disaster chain risk is evaluated based on the disaster chain risk degree. The present invention constructs a four-dimensional disaster system scoring framework, reduces the sensitivity to the score of a single indicator, and overcomes the inapplicability of the traditional risk matrix to non-traditional risks and low-frequency high-loss risks that exceed the original design benchmark. In addition, the present invention proposes a disaster chain topology structure and a trained logistic regression model, which significantly improves the accuracy and reliability of disaster chain risk prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 The present invention provides a flowchart of a preferred embodiment of a method for dynamic risk assessment of disaster chains based on logistic regression.

[0031] Figure 2 A schematic diagram of the technical route of a disaster chain dynamic risk assessment method based on logistic regression provided in an embodiment of the present invention.

[0032] Figure 3 This is a flow chart of the principles of the dynamic weight generation module in the disaster chain dynamic risk assessment method based on logistic regression provided by an embodiment of the present invention.

[0033] Figure 4 A schematic diagram of the architecture of a disaster chain dynamic risk assessment system based on logistic regression provided in an embodiment of the present invention.

[0034] Figure 5 This is a functional block diagram of a terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0036] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents, operations, or steps, nor must they be executed in the order described. For example, some operations or steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0037] It should be understood that the terms used in this specification are only for the purpose of describing particular embodiments and are not intended to limit the present invention. As used in the specification and appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0038] It should be understood that, to facilitate a clear description of the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, terms such as "first" and "second" are used to distinguish between identical or similar items with substantially the same functions and effects. For example, the first control information and the second control information are merely used to distinguish different control information and do not limit their order.

[0039] Those skilled in the art can understand that words such as "first" and "second" do not limit the quantity and execution order, and words such as "first" and "second" do not necessarily limit them to be different.

[0040] It should be further understood that the term "and / or" used in the present description and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0041] The disaster chain dynamic risk assessment method based on logistic regression of the present invention can be applied to terminals, including intelligent product terminals such as computers, televisions and mobile phones. Specifically, Figure 1 As shown in , the disaster chain dynamic risk assessment method based on logistic regression of this embodiment includes the following steps:

[0042] Step S100: defining a new type of extreme disaster, constructing a four-dimensional disaster systematic scoring framework, scoring indicators based on the four-dimensional disaster systematic scoring framework, and calculating the weight of each indicator and the comprehensive disaster score based on information entropy theory to screen the disaster chain triggering factors;

[0043] Step S200: Based on the disaster chain trigger, a disaster chain topology structure is designed, and a logistic regression model is trained to dynamically update the failure probability;

[0044] Step S300: Calculate the disaster chain risk degree, and evaluate the disaster chain risk based on the disaster chain risk degree.

[0045] Combine Figure 2As shown in the figure, the technical solution of the present invention mainly includes two aspects: the construction of a new extreme disaster factor evaluation system and an intelligent adjustment mechanism for the edge weights of the disaster chain network. First, in response to the new disaster threats emerging with climate change and technological advancement, the present invention proposes a new definition of extreme disasters. To address the lack of effective assessment methods for new extreme disasters, a four-dimensional disaster system scoring framework is established. This framework collects multi-source data, calculates quantifiable indicator values, compares the indicators within the disaster factor group, and fine-tunes and assigns scores based on expert knowledge and actual disaster performance. Then, based on information entropy theory, the weights of each indicator and the comprehensive disaster score are calculated, and the high-scoring indicators are selected as the triggering factors of the subsequent disaster chain.

[0046] Specifically, numerous studies and disaster records indicate that climate change is significantly impacting the intensity and frequency of natural disasters by increasing extreme precipitation, strengthening tropical cyclones, and increasing the frequency and intensity of compound extreme events. Emerging disasters, whether natural or man-made, share the following common characteristics: 1. They are caused by unconventional factors; 2. They exceed established regulatory thresholds; and 3. They have cross-system cascading effects. Therefore, compound disasters that exceed design benchmarks and can trigger exponentially destructive secondary impacts are defined as new extreme disasters.

[0047] The indicators and their meanings included in the four-dimensional disaster system scoring framework constructed by this invention are as follows:

[0048] (1) Safety Consequence (S): This measures the severity of the consequences of a disaster, prioritizing the prevention and control of high-severity disasters. Direct economic losses typically include economic losses caused by infrastructure such as electricity and transportation, and can therefore effectively measure the severity of the consequences of a disaster.

[0049] (2) Frequency (F): refers to the number of occurrences within a fixed period. Combined with the disaster severity index, it can distinguish between high-frequency low-loss events and low-frequency high-loss events.

[0050] (3) Cascading Effect Potential (C): This assesses the ability of a single disaster to couple with other disasters and trigger multi-system chain failures. The number of secondary disasters that can induce other disasters (e.g., typhoons causing floods, drone attacks causing explosions) and cause physical damage (e.g., cooling system failure) can be used to directly quantify the cascading effect potential of a disaster.

[0051] (4) Data Support (D): This considers both the support and authority of public data. Its sub-indicators include key field completeness, numerical precision level, and data source authority. Since key field completeness directly determines whether the scoring basis exists, this embodiment assigns it the highest weight (50%). Data source authority can avoid the decrease in the credibility of network nodes caused by ambiguous data sources, with a serious weight of 30%. Numerical precision level can deepen the quantitative ability of the model (precision value > interval value > qualitative description), but its impact on the construction of the basic framework is relatively weak, and its weight is relatively low (20%). The segmentation standards of the sub-indicators are detailed in Table 1.

[0052] Table 1 Segmentation standards of sub-indicators of data support

[0053]

[0054]

[0055] The secondary disaster statistics process must adhere to the MECE principle (Mutually Exclusive, Collectively Exhaustive), meaning it must be exhaustive yet mutually independent, with clear hazard-bearing entities. First, the affected entities are stratified into priority protection targets (such as buildings, equipment, and personnel), support systems (such as power and water sources), the surrounding environment, and specific threats. Then, along the hazard chain, the primary disaster is the source, inducing secondary disasters. The primary and secondary disasters form a composite disaster, ultimately leading to consequences. Within each broad category, a clear distinction is made between physical structural damage and system functional impairment, and a new dimension of human attack is added to ensure coverage of human threats. A schematic diagram of the secondary disaster classification is shown in Table 2. Once the classification is complete, the corresponding secondary disaster type number is assigned to the disaster being evaluated, drawing on historical cases, professional knowledge, and expert experience. Based on the total number and a scoring principle, a cascading effect potential score is assigned.

[0056] Table 2 Schematic results of secondary disaster classification

[0057]

[0058]

[0059] The scoring rules for the four-dimensional disaster system scoring framework in this embodiment are as follows: each indicator is scored between 1 and 5 points, with breakpoint scoring applied within the disaster being studied. The top 30% of the quantitative indicator results are assigned 5 points, those between 30% and 70% are assigned 3-4 points, and the bottom 30% are assigned 1-2 points. Geographical, climatic, and social characteristics vary between regions, and the actual manifestations of disasters also vary. Therefore, the objective scoring method requires subjective fine-tuning based on the actual manifestations of the disaster. For example, although the average annual economic losses caused by drone attacks on energy facilities are currently far less than those caused by disasters such as floods and earthquakes, the evolution of the threat is inherently uncertain. With the iteration of attack technologies, the accumulation of vulnerabilities in energy network systems, and the rise of coordinated swarm attack models, single-point physical damage could trigger exponentially escalating, compound crises such as power grid topology collapse and radioactive material breaching containment. Therefore, the disaster severity (S) score for drone attacks should be appropriately adjusted upwards. The same approach applies to fine-tuning the scores for other disasters.

[0060] Next, this embodiment calculates the weight of each indicator and the comprehensive disaster score based on the information entropy theory to screen the triggering factors of the disaster chain. The Entropy Weight Method (EWM) is an objective weighting method commonly used in multi-index decision analysis. The basic idea is to measure the degree of uncertainty in the decision by calculating the information entropy of each indicator. The larger the information entropy of the indicator, the more dispersed the information of the indicator, the smaller the difference, and the lower the impact on the decision; on the contrary, the smaller the information entropy, the greater the difference, and the higher the impact on the decision. Its advantage is that it relies entirely on the objective distribution of the data itself, avoiding the interference of subjective weighting, and the operation process is simple and efficient.

[0061] Assuming there are n samples and m indicators, the indicator matrix is: R = (x ij ) n×m The information entropy of the jth indicator is defined as:

[0062]

[0063] Among them, e j is the information entropy of the jth indicator; n is the number of samples; is a constant used to standardize the entropy value; when p ij =0, define p ij ln(p ij )=0;p ij It represents the proportion of the i-th sample under the j-th indicator, also known as the proportional coefficient, which is defined as:

[0064]

[0065] Among them, rij It is the index value of the i-th sample on the j-th index after normalization. The commonly used normalization methods are minimization and maximization. The four indicators of this evaluation system are all positive indicators, that is, the larger the score, the more likely it is to be rated as a core disaster factor. The normalization formula is:

[0066]

[0067] Among them, x ij is the value of the i-th sample on the j-th index; x min (j) is the minimum value of the jth index; x nax (j) is the maximum value of the j-th index.

[0068] The entropy weight of the jth indicator is defined as:

[0069]

[0070] Among them, w j is the weight of the jth indicator; m is the number of indicators.

[0071] Using the calculated indicator weights, the weighted total score formula is used to modify the comprehensive disaster score of each disaster sample:

[0072]

[0073] The weighted correction scoring results are shown in Table 3.

[0074] Table 3 Weighted revised scoring results

[0075]

[0076]

[0077] The entropy weight method captures the information density of indicators and automatically focuses decision-making on key influencing factors with significant differences in each disaster. The weight distribution in Table 3 confirms that the traditional risk matrix overestimates the influence of disaster frequency (weight value is 19.5%) and underestimates the severity of disaster impact (weight value is 27.7%). Low-frequency, high-loss events are easily overlooked in the allocation of disaster prevention resources, which may lead to serious adverse consequences.

[0078] Furthermore, this embodiment designs a disaster chain topology. The GERT (Graphical Evaluation Review Technique) network model is a type of random network that can describe the logical relationship between various activities and pass parameters in the form of probability. It uses graphical deduction to vividly and intuitively reflect the topological relationship between various activities, and can be used to solve various real-world problems such as engineering, production, operation, and management. New extreme disasters with great research significance for disaster-bearing bodies are screened out through the four-dimensional disaster system scoring framework, covering natural disasters and man-made threats. Starting from this, as the core trigger factor of the disaster chain, the "induction-conduction-amplification" interaction path under the synergistic effect of natural disasters and man-made threats is designed, and the GERT network is used to describe the sudden disaster nodes and coupling relationships to form a disaster chain topology that can be dynamically modified.

[0079] In the disaster chain topology, each node represents a disaster event (such as earthquake, fire, power system collapse), and the state variable is marked Represents event S i The development degree of the event is in the range of [0,1]. If it is >1, it means that the event is triggered. Different types of coupling event sets are marked. There are three types of coupling:

[0080] OR coupling: Meeting any one condition triggers subsequent events, helping to identify the most dangerous single events and prioritize prevention and control. The mathematical model uses a maximum function to describe the strongest impact among multiple events.

[0081] AND coupling: All conditions must be met simultaneously for subsequent events to be triggered, which helps design "redundant safety measures." The minimum function is used in the model to ensure multiple constraints.

[0082] CO coupling: Two events reinforce each other, creating a destructive force greater than 1+1, reminding managers to avoid dangerous combinations. The combined impact is quantified using a weighted summation formula.

[0083] Next, define the edges of the disaster chain topology. In this embodiment, the directed edges represent the evolutionary relationship between events, and the coupling probability P is assigned. ij and coupling coefficient ε ij Two types of properties. Coupling probability P ij Represents event S i Trigger S j The probability of is calculated in real time by the logistic regression model; the coupling coefficient ε ij The correlation strength between events was quantified and calculated using the Jaccard index. The calculation formula is shown in Equation (6).

[0084]

[0085] Where: Ci 、C j 、C ij Represents event S respectively i 、S j and event S i Trigger event S j frequency of occurrence.

[0086] In this embodiment, the dynamic weight generation module in the disaster chain topology structure is as follows: Figure 2 Specifically, this embodiment extracts disaster intensity characteristic parameters from real-time disaster scenarios, such as earthquake peak acceleration, precipitation, and typhoon maximum wind speed, and inputs them into a logistic regression model trained based on historical cases and hypothetical scenario sets to output edge weights P ij .

[0087] Logistic regression (LR), one of the ten most popular machine learning algorithms, is a generalized linear regression analysis model that belongs to supervised learning in machine learning. By establishing a mapping relationship between features and event probability, it dynamically adjusts the coupling weights in the disaster chain network to achieve data-driven evolutionary prediction. LR can find an accurate fitting function to define the nonlinear relationship between the probability of disaster occurrence and a set of feature data. This dependency can be represented by the logistic function (also known as the sigmoid function), as shown in Equation (7). The reason for choosing the logistic regression model here is that, first, the sigmoid function maps the linear output z to the interval [0, 1], which conforms to the definition of probability; second, it is interpretable, and the output can be directly interpreted as "the confidence level of belonging to the positive class (i.e., 1, indicating occurrence)."

[0088]

[0089] Among them, P ij Represents event S i Trigger S j The probability of; z is a linear fitting function, which is expressed by equation (8) consisting of a set of disaster characteristic variables.

[0090] z=ω1x1+ω2x2+…+ω n x n +b (8)

[0092] Where ω=[ω1,ω2,…,ω n ] is the feature weight; x=[x1,x2,…,x n ] are characteristic variables, including seismic peak acceleration, magnitude, focal depth, precipitation, typhoon maximum wind speed, drone attack frequency, etc.; b is the bias term.

[0093] During sample data preparation, to identify natural disaster chains, we downloaded data on initial and secondary disasters within the same time series from a pre-set database. We then defined a time window for the edge "initial disaster-secondary disaster." For each initial event, we searched for any secondary disasters that occurred within the subsequent time window, counting their co-occurrences and independent occurrences. Regarding facility and man-made disaster chains, there is currently no official statistical database on energy facility operational accidents; only operational reports or other publicly available data are available, such as those published by the International Atomic Energy Agency. Therefore, we needed to collect disaster variables and external energy facility event cases through multiple channels, such as online and literature searches. These, combined with scenario assumptions, served as sample data. Each sample data item was labeled 1 (failure) / 0 (non-failure), and a logistic regression model was trained for each edge.

[0094] To address the problem of scarcity of cases and insufficient historical data for some real disaster events, or the problem of insufficient training of logistic regression models due to some extremely rare disaster combinations, this embodiment uses the Wasserstein Generative Adversarial Networks (WGAN) to learn the distribution of real data and generate a large number of realistic "virtual disaster events" as scenario hypothesis supplementary training samples, solving the problem of data scarcity, generating diverse extreme scenarios, and ultimately improving the robustness of the model's predictions in small sample scenarios. Its core is to introduce the Wasserstein distance as a new loss function. By constructing a discriminator that satisfies the K-Lipschitz condition and maximizing its estimate of the Wasserstein distance between real data and generated data, it can ensure that the generator training converges to the global optimal solution. This fundamentally solves the problems of gradient vanishing and mode collapse in traditional GANs, making WGAN more stable during training and able to generate higher quality samples. The main architecture of WGAN is similar to that of traditional GANs. It generates data through a "dynamic game process" of two networks: a generative model and a discriminative model. The generative model is designed to randomly generate realistic disaster scenario data given some implicit information. The discriminative model can be compared to a binary classifier, which uses a certain model to try to distinguish the data generated by the generative model from the real data. The discriminative loss function of WGAN is:

[0095]

[0096] Here, G is the generative model, D is the discriminative model, D(x) represents the discriminator's score on the real data x, and D(G(Z)) represents the discriminator's score on the generated data G(Z). When training a WGAN, the goal is to maximize this loss to increase the Wasserstein distance between the real data and the generated data.

[0097] The feature screening mechanism consists of two steps. First, redundant features are eliminated based on the Spearman correlation coefficient, and candidate features with an absolute correlation coefficient with the coupling probability greater than the medium effect threshold are selected. Second, the optimal feature subset is constructed through forward stepwise regression. Each iteration, the feature with the largest improvement in AUC (Area Under Curve) is selected and added to the model until all available indicators are iterated. The Spearman's Rank Correlation Coefficient is a non-parametric statistical method used to measure the strength of the monotonic relationship between two variables. Its calculation is based on the rank of the variable rather than the original data, and it can effectively handle nonlinear correlations and unconventional distribution data. The formula is defined as:

[0098]

[0099] Among them, d i The rank difference between the two variables corresponding to the data point; n is the number of samples.

[0100] Since disaster characteristics (such as peak seismic acceleration and rainfall) often have a skewed distribution, non-parametric test methods need to be adopted. The Spearman correlation coefficient can effectively eliminate features that have no significant correlation with the target variable (disaster coupling probability), avoid noise interference, identify dynamic characteristics that have a substantial impact on disaster evolution, and retain key driving factors.

[0101] The logistic regression model was evaluated. The evaluation indicators of the logistic regression model include accuracy, precision, recall, F1 score, and ROC-AUC curve. The meaning of the indicators and the calculation formula are shown in Table 4 below.

[0102]

[0103]

[0104]

[0105] Use sample data to fit the logistic regression model, learn the weight ω, and the loss function Expressed as cross entropy, where y k ∈{0,1} is the label of whether the event occurs.

[0106]

[0107] The training phase is followed by the prediction phase, which inputs the feature x in real time and outputs the probability P ij , replacing the fixed P in the original GERTS network ij .

[0108] Furthermore, this embodiment calculates the disaster chain risk degree, and evaluates the disaster chain risk based on the disaster chain risk degree.

[0109] The disaster chain risk index is a comprehensive indicator used to quantify and assess disaster chain risks. It comprehensively considers the interactions between disaster events in the disaster chain, the chain reactions, and the impact on the disaster-bearing body. Calculating the disaster chain risk index helps to comprehensively and accurately assess disaster chain risks, providing a scientific basis for disaster early warning, emergency management, and resource allocation, thereby effectively reducing the losses caused by disaster chains. The measurement methods of the disaster chain risk index usually include the following aspects:

[0110] ① Probability of disaster event: refers to the possibility of a disaster event occurring within a certain period of time;

[0111] ② Losses caused by disaster events: refers to casualties, property losses, etc. that may be caused by disaster events;

[0112] ③Edge vulnerability: refers to the fragility of the triggering relationship between disaster events in the disaster chain, reflecting the intensity of the mutual influence between disaster events.

[0113] Therefore, the present invention defines the three elements of disaster chain risk as dynamic failure probability, coupling function, and loss index, and the calculation method is as shown in formula (12). The dynamic failure probability comes from the node condition failure probability value P output by the logistic regression model in real time. ij The coupling function M is composed of a constant coefficient, a conditional probability, and a coupling coefficient, and is used to quantify the strength of the correlation between events. The function form is shown in formula (13); the loss index S j The relative number that reflects the overall change direction and degree of the disaster loss caused by event j under different spatiotemporal conditions, while the degree centrality indicates the number of other nodes directly connected to the node, including the in-degree value (the number of nodes that affect this node) and the out-degree value (the number of nodes affected by this node). It can reflect the change direction and importance of the evolution of the disaster node. If the degree centrality of event j is large, it indicates that there are more disaster events associated with it, which has a significant driving effect on amplifying the disaster. Therefore, the degree centrality of event j is used here to represent the loss index. The sum of the in-degree and out-degree of the GERT network can be calculated using the NetworkX library.

[0114] Risk (R ij )=Dynamic failure probability (P ij )×coupling function (M)×loss index (S j ) (12)

[0116] M=δ ji ·P ij ·ε ij (13)

[0118] R v =∑R ij (14)

[0120] Among them, R ij Refers to the risk of the edge connecting event v and event j; constant coefficient δ ji The failure probability P ij Decision, if P ij >0, then δ ji =1, if P ij =0, then δ ji =0; coupling coefficient ε ij The calculation formula is shown in formula (6), which is the same as the Jaccard index; R v is the systematic risk value of the disaster chain v, that is, the cumulative value of the risk of all edges in the chain.

[0121] This paper quantitatively assesses the risk differences between single and complex chains, revealing the amplification effect of combined threats composed of natural disasters and man-made threats, achieving the technical requirement of "quantifiable multi-hazard coupling." This paper proposes the concept of "new extreme disasters" facing energy facilities in the current context of climate change and intensifying border conflicts, and constructs a corresponding four-dimensional indicator evaluation system that integrates objective weight optimization and resilience risk management. This system objectively corrects subjective weight bias in scoring, reduces sensitivity to individual indicator scores, and overcomes the inapplicability of traditional risk matrices for non-traditional risks and low-frequency, high-loss risks that exceed the original design benchmark. This system is used to scientifically screen for hazards that are more likely to act as triggers in disaster chain evolution studies, addressing the "visible but ignored" dilemma of new extreme disaster prevention and control. To address the chain-like evolution of "new extreme disasters," this paper proposes an intelligent edge weight adjustment mechanism for disaster chain networks, providing an innovative solution to the shortcomings of traditional static parameter methods in terms of adaptability, accuracy, and real-time performance. This technology significantly improves the accuracy and reliability of disaster chain risk prediction through a data-driven dynamic weight allocation mechanism. In the future, with economic development and advancements in artificial intelligence (AI), this invention, based on multi-source data sensing technologies like the Internet of Things, is expected to be widely applied in fields such as industrial risk assessment and public emergency management, providing intelligent, real-time decision-making support for disaster prevention and control in complex systems. Furthermore, this technology can be easily integrated with existing monitoring systems, enabling low-cost and highly efficient deployment. Therefore, this invention not only overcomes the theoretical limitations of traditional methods but also opens up new avenues for engineering-based disaster chain research.

[0122] Based on the above embodiment, the present invention also provides a disaster chain dynamic risk assessment system based on logistic regression, which is used to implement the steps in the above method embodiment, such as Figure 4As shown, the system of this embodiment includes: a trigger factor determination module 10, a logistic regression model training module 20, and a disaster chain risk assessment module 30. Specifically, the trigger factor determination module 10 is used to define new extreme disasters, construct a four-dimensional disaster system scoring framework, score indicators based on the four-dimensional disaster system scoring framework, and calculate the weight of each indicator and the comprehensive disaster score based on the information entropy theory to screen the disaster chain triggering factors. The logistic regression model training module 20 is used to design the disaster chain topology structure based on the disaster chain trigger, and train the logistic regression model to dynamically update the failure probability. The disaster chain risk assessment module 30 is used to calculate the disaster chain risk degree, and assess the disaster chain risk based on the disaster chain risk degree.

[0123] The working principles of each module in the disaster chain dynamic risk assessment system based on logistic regression in this embodiment are the same as the principles of each step in the above method embodiment, and will not be repeated here.

[0124] Each module in the above-mentioned logistic regression-based dynamic risk assessment system for disaster chains can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a terminal in the form of hardware, or can be stored in a memory in the terminal in the form of software, so that the processor can call and execute the corresponding operations of each module.

[0125] Based on the above embodiment, the present invention further provides a terminal, the principle block diagram of the terminal can be as follows: Figure 5 The terminal may include one or more processors 100 ( Figure 5 Only one is shown), a memory 101, and a computer program 102 stored in the memory 101 and executable on one or more processors 100. For example, a program for dynamic risk assessment of disaster chains based on logistic regression. When one or more processors 100 execute the computer program 102, each step in the embodiment of the method for dynamic risk assessment of disaster chains based on logistic regression can be implemented. Alternatively, when one or more processors 100 execute the computer program 102, the functions of each module / unit in the embodiment of the system for dynamic risk assessment of disaster chains based on logistic regression can be implemented, which is not limited here.

[0126] In one embodiment, the processor 100 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0127] In one embodiment, the memory 101 may be an internal storage unit of an electronic device, such as a hard disk or memory of the electronic device. The memory 101 may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device. Furthermore, the memory 101 may include both an internal storage unit of the electronic device and an external storage device. The memory 101 is used to store computer programs and other programs and data required by the terminal. The memory 101 may also be used to temporarily store data that has been output or is about to be output.

[0128] Those skilled in the art will understand that Figure 5 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the solution of the present invention, and does not constitute a limitation on the terminal to which the solution of the present invention is applied. The specific terminal may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0129] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, operating database or other media used in the embodiments provided by the present invention may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0130] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A disaster chain dynamic risk assessment method based on logistic regression, characterized by: The method comprises: Define new extreme disasters, construct a four-dimensional disaster system scoring framework, score indicators based on the four-dimensional disaster system scoring framework, and calculate the weight of each indicator and the comprehensive disaster score based on information entropy theory to screen the triggering factors of the disaster chain; Based on the disaster chain trigger, a disaster chain topology is designed, and a logistic regression model is trained to dynamically update the failure probability; A disaster chain risk degree is calculated, and the disaster chain risk is assessed based on the disaster chain risk degree.

2. The disaster chain dynamic risk assessment method based on logistic regression according to claim 1 is characterized in that: The indicators of the four-dimensional disaster system scoring framework include: disaster impact severity, frequency of occurrence, cascading effect potential and data support. Among them, the severity of disaster impact is used to measure the level of consequences caused by the disaster, the frequency of occurrence refers to the number of disasters occurring within a fixed period, the cascading effect potential is used to assess the coupling ability of a single disaster with other disasters and the ability to trigger multi-system chain failures, and the data support is used to take into account the support level and authority of public data.

3. The disaster chain dynamic risk assessment method based on logistic regression according to claim 1 is characterized in that: The scoring rules of the four-dimensional disaster system scoring framework are as follows: the scoring range of each indicator is 1 to 5 points, and breakpoint scoring is performed within the disaster to be studied. The top 30% of the quantitative indicator results are assigned 5 points, the 30%-70% are assigned 3 to 4 points, and the bottom 30% are assigned 1 to 2 points.

4. The method for dynamic risk assessment of disaster chains based on logistic regression according to claim 1, characterized in that: Based on the information entropy theory, the weight of each indicator and the comprehensive disaster score are calculated to screen the triggering factors of the disaster chain, including: Calculate the information entropy of each indicator to correct the indicator weight; Using the revised indicator weights, the weighted total score formula is used to obtain the comprehensive disaster score of each disaster sample; The disaster with the highest comprehensive score is selected as the triggering factor of the disaster chain.

5. The method for dynamic risk assessment of disaster chains based on logistic regression according to claim 1 is characterized in that: Each node in the disaster chain topology structure represents a disaster event, the directed edges represent the evolutionary relationship between events, and the edge weights are output by a logistic regression model trained based on historical cases and hypothetical scenario sets.

6. The method for dynamic risk assessment of disaster chains based on logistic regression according to claim 1, characterized in that: The training process of the logistic regression model includes: Obtain initial disaster data and secondary disaster data within the same time series, define the time window of initial disaster-secondary disaster, and for each initial disaster event, find out whether there is a secondary disaster event in its subsequent time window, count the number of co-occurrences and independent occurrences of the two, and obtain sample data; Redundant features were eliminated based on the Spearman correlation coefficient, and candidate features with an absolute correlation coefficient with the coupling probability greater than a threshold with a medium effect were selected; The optimal feature subset is constructed through forward stepwise regression. The feature with the largest AUC improvement is selected and added to the model in each iteration until all available indicators are iterated. The logistic regression model was evaluated using the following evaluation indicators: accuracy, precision, recall, F1 score, and ROC-AUC curve.

7. The method for dynamic risk assessment of disaster chains based on logistic regression according to claim 1, characterized in that: The indicators of the disaster chain risk include: dynamic failure probability, coupling function, and loss index.

8. A disaster chain dynamic risk assessment system based on logistic regression, characterized by: The system is used to implement the steps of the disaster chain dynamic risk assessment method based on logistic regression according to any one of claims 1 to 7, and the system includes: The trigger factor determination module is used to define new extreme disasters, build a four-dimensional disaster system scoring framework, score indicators based on the four-dimensional disaster system scoring framework, and calculate the weight of each indicator and the comprehensive disaster score based on information entropy theory to screen the trigger factors of the disaster chain; A logistic regression model training module is used to design a disaster chain topology based on the disaster chain trigger and train a logistic regression model to dynamically update the failure probability; The disaster chain risk assessment module is used to calculate the disaster chain risk degree and assess the disaster chain risk based on the disaster chain risk degree.

9. A terminal, characterized in that: The terminal includes a memory, a processor, and a disaster chain dynamic risk assessment program based on logistic regression stored in the memory and runnable on the processor. When the processor executes the disaster chain dynamic risk assessment program based on logistic regression, the steps of the disaster chain dynamic risk assessment method based on logistic regression as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a disaster chain dynamic risk assessment program based on logistic regression, and the disaster chain dynamic risk assessment program based on logistic regression implements the steps of the disaster chain dynamic risk assessment method based on logistic regression as described in any one of claims 1 to 7 on the computer-readable storage medium.

Images