Accident severity index-based accident cause quantitative risk analysis method

Abstract

The invention discloses an accident cause quantitative risk analysis method based on an accident severity index. The method comprises the following steps: acquiring casualty information, direct economic loss and a plurality of corresponding accident cause indexes of each accident in a historical accident case; based on the casualty information and the direct economic loss of each accident, obtaining a severity index of each accident; taking a plurality of accident cause indexes of all accidents as characteristic variables, taking a severity index of the corresponding accident as a target variable, and training and analyzing through a random forest regression model to obtain a characteristic importance score of each accident cause index; calculating the risk value of each accident cause index based on the feature importance score and the occurrence probability of each accident cause index in the historical accident case; and carrying out risk grade division on each accident cause index according to the risk value. The method provides a new basis for unified quantification of accident consequences, accident cause analysis and accurate prevention and control.

Description

Technical Field

[0001] This invention relates to the field of safety production and accident prevention technology, and more specifically to a method for quantitative risk analysis of accident causes based on an accident severity index. Background Technology

[0002] In accident causation analysis, accurately quantifying accident severity is crucial for connecting accident consequences with causal mechanisms. Existing accident severity assessment systems mainly include the following typical methods.

[0003] 1. The classification method based on the number of casualties or incapacitated working hours, which classifies accidents into levels such as "fatal," "serious," and "minor" based on indicators such as the number of deaths and serious injuries. This method is intuitive and easy to implement, but it can only reflect the static results of the accident's consequences and cannot reflect the cumulative effect of risks during the accident's evolution.

[0004] 2. Economic loss-based assessment measures severity by the direct and indirect economic losses caused by the accident. This method reflects the impact of the accident from an economic perspective, but data acquisition is delayed and is greatly affected by the regional economic level. In recent years, scholars have attempted to establish more complex severity measurement models. Some studies use classification trees within machine learning or data mining frameworks to predict injury severity, while others propose integrating multi-dimensional indicators such as injury severity, duration of disability, and injured body parts for analysis. These methods facilitate risk ranking, but because their classification is relatively coarse, they struggle to capture the continuous changes in accident consequences. This leads to a disconnect between severity analysis results and causal risk research, failing to effectively reveal the contribution of different causal factors to the overall consequences of the accident.

[0005] Therefore, how to establish a unified measurement model for accident consequences, quantify the contribution of risk factors to accident consequences, and thus achieve coupled analysis of accident severity and causal risk, providing new theoretical tools and methodological support for accident severity assessment and risk classification, is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of the above problems, the present invention proposes a method for quantifying the risk analysis of accident causes based on an accident severity index in order to overcome or at least partially solve the above problems.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] This invention provides a method for quantitative risk analysis of accident causes based on an accident severity index, comprising the following steps: Acquire historical accident case data in the target field, including casualty information, direct economic losses, and multiple corresponding accident causation indicators for each accident; Based on the casualty information and direct economic losses of each accident, a severity index is obtained for each accident. Using the multiple accident causation indicators of all accidents as feature variables and the severity index of the corresponding accident as the target variable, a random forest regression model is trained; and based on the trained random forest regression model, the feature importance score of the influence of each accident causation indicator on the accident severity is obtained. Based on the feature importance scores and the probability of occurrence of each accident causation indicator in historical accident cases, the risk value of each accident causation indicator is calculated. Based on the risk values, the risk levels of each accident causative indicator are classified to obtain the quantitative risk classification results of each accident causative indicator.

[0009] Furthermore, the accident causation indicators include human factors, material factors, environmental factors, and management factors.

[0010] Furthermore, the severity index of each accident is expressed as follows:

[0011] in, ASI i Indicates the first i The severity index of the accident; C Represents the set of types of casualties; S i,k Indicates the first i The types of injuries and fatalities in the accident were: k The number of casualties; ω k Indicates the type of casualties. k Severity weighting coefficient; E i Indicates the first i The direct economic losses caused by the accident; α Weighting coefficients representing direct economic losses; β Indicates the scaling factor; ln This represents the natural logarithm function.

[0012] Furthermore, the severity weighting coefficient for death is 1, and the severity weighting coefficient for injury is 0.1.

[0013] Furthermore, the scaling factor β Determined based on the 85th percentile of direct economic losses in historical accident cases; expressed as:

[0014] in, E a This represents the 85th percentile of direct economic losses in historical accident cases. k is a dimensionless constant representing the saturation scale.

[0015] Furthermore, the direct economic loss of each accident is the comprehensive economic cost corresponding to each death, expressed as: D = F + P + L

[0016]

[0017] in, D Indicates the overall economic cost; F Indicates the amount of the predetermined penalty; P Indicates the amount of compensation; L This refers to the economic losses caused by production stoppage due to an accident; T Indicates the duration of compensation and losses due to production stoppage; P t Indicates the first t Compensation for losses incurred during the period; L t Indicates the first t Losses due to production stoppage during the period; r This represents the discount rate.

[0018] Furthermore, when the target field is the field of mine safety production, the severity index of each mine accident is expressed as:

[0019] in, ASI i Indicates the first i The severity index of the accident; C Represents the set of types of casualties; S i,k Indicates the first i The types of injuries and fatalities in the accident were: k The number of casualties; ω k Indicates the type of casualties. k Severity weighting coefficient; E i Indicates the first i The direct economic loss caused by the accident; 2.47 represents the weighting coefficient for direct economic loss; 0.001 represents the scaling factor; ln This represents the natural logarithm function.

[0020] Furthermore, the risk value of each accident causation indicator is the product of the probability of occurrence of each accident causation indicator and the corresponding feature importance score.

[0021] Furthermore, the risk level is divided into four levels: Level 1 Risk: The probability of occurrence is higher than the first preset threshold and the feature importance score is higher than the second preset threshold; Level 2 risk: The probability of occurrence is lower than the first preset threshold but the feature importance score is higher than the third preset threshold, or the probability of occurrence is higher than the first preset threshold and the feature importance score is between the second preset threshold and the third preset threshold; Level 3 risk: The probability of occurrence is higher than the first preset threshold but the feature importance score is lower than the fourth preset threshold, or the probability of occurrence is lower than the first preset threshold but the feature importance score is higher than the fourth preset threshold; Lower risk: The probability of occurrence is lower than the first preset threshold and the feature importance score is lower than the fourth preset threshold.

[0022] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a method for quantitative risk analysis of accident causes based on an accident severity index, which has the following beneficial effects: This invention uses an accident severity index to comprehensively consider both casualties and economic losses, and introduces utility theory and the principle of diminishing marginal utility to achieve unified quantification, making the assessment of accident severity more scientific and comprehensive. It provides new tools and theoretical support for the unified quantification of accident consequences, accident cause analysis, and precise prevention and control.

[0023] Based on random forest feature recognition, this invention constructs a probability-importance dual-dimensional quantitative analysis framework, which defines the product of the probability of occurrence of risk factors in statistical accidents and the importance of their features as the risk value, thereby realizing the classification of accident causes. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0025] Figure 1 This is a schematic diagram of the accident causation quantification risk analysis method based on accident severity index provided in this embodiment of the invention.

[0026] Figure 2 This is a schematic diagram of a random forest process that incorporates an accident severity index, as provided in an embodiment of the present invention.

[0027] Figure 3 This is a schematic diagram illustrating the correlation between the number of casualties, economic losses, and accident severity index provided in this embodiment of the invention.

[0028] Figure 4 This is a schematic diagram of the three-dimensional sensitivity distribution of the accident severity index to key input parameters provided in this embodiment of the invention.

[0029] Figure 5 This is a schematic diagram showing the ranking of the characteristic importance of risk factors in mining accidents provided in this embodiment of the invention.

[0030] Figure 6 This is a schematic diagram illustrating the distribution of risk levels of mine accidents provided in an embodiment of the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] This invention discloses a method for quantifying the risk of accident causes based on an accident severity index, such as... Figure 1 As shown, it includes the following steps: S1. Obtain historical accident case data in the target field. The historical accident case data includes information on casualties, direct economic losses, and multiple corresponding accident causation indicators for each accident. S2. Based on the casualty information and direct economic losses of each accident, obtain the severity index of each accident; S3. Using multiple accident causation indicators of all accidents as feature variables and the severity index of the corresponding accident as the target variable, train a random forest regression model; and based on the trained random forest regression model, obtain the feature importance score of the influence of each accident causation indicator on the severity of the accident. S4. Based on the feature importance score and the probability of occurrence of each accident causation indicator in historical accident cases, calculate the risk value of each accident causation indicator. S5. Based on the risk value, classify the risk level of each accident causative indicator to obtain the quantitative risk classification result of each accident causative indicator.

[0033] To achieve deep coupling between accident severity measurement and causal risk, this method establishes an Accident Severity Indication (ASI) model and combines it with a random forest algorithm. It performs a two-dimensional analysis based on the characteristic importance and occurrence probability of risk factors, and finally classifies different factors into levels. This provides new theoretical tools and methodological support for accident severity assessment and risk classification, contributing to the construction of differentiated accident management and dual prevention mechanisms.

[0034] Next, each of the above steps will be explained in detail.

[0035] In step S1 above, historical accident case data of the target field is obtained. The historical accident case data includes information on casualties, direct economic losses and corresponding multiple accident causation indicators for each accident. Among them, the accident causation indicators include human factors, material factors, environmental factors and management factors.

[0036] In step S2 above, based on the casualty information and direct economic losses of each accident, a severity index for each accident is obtained; specifically: Because casualties and economic losses have different dimensions and different levels of importance in social value and risk management, analyzing either dimension alone cannot comprehensively measure the overall severity of an accident. To establish a unified and comprehensive evaluation standard, this invention constructs a new composite numerical index—the Accident Severity Index (ASI)—to reflect the overall severity of accidents in the target domain. This index comprehensively considers casualties and economic losses, and incorporates utility theory and the principle of diminishing marginal utility from economics, arguing that the "contribution" of an increase in economic loss to the overall severity is diminishing. For example, the increase in severity from a loss of 1 million to 2 million is actually greater than the increase from a loss of 11 million to 12 million. Based on this, the severity index for each accident is expressed as follows:

[0037] in, ASI i Indicates the first i The severity index of the accident; C This represents a set of injury / casualty types, including death and injury; S i,k Indicates the first i The types of injuries and fatalities in the accident were: k The number of casualties; ω k Indicates the type of casualties. k Severity weighting coefficient; E i Indicates the first i The direct economic losses caused by the accident;α The weighting coefficient representing direct economic losses indicates the relative importance of casualties and economic losses. β This represents the scaling factor, used to adjust for economic losses. E i Numerical scale, ensuring scaling to an appropriate order of magnitude; ln This represents the natural logarithm function, used to capture the diminishing marginal utility of economic losses.

[0038] (1) Next, the key parameters in the above severity index calculation formula will be explained.

[0039] 1) Severity weighting coefficient ω k : Determining the weights of different parameters is crucial for constructing an accident severity index, as it directly affects the scientific validity and rationality of the assessment results. First, the types of injuries and fatalities must be defined. C Severity weighting coefficient at time of death ω k The value is set to 1, and this is used as a benchmark to determine the weights for the number of injuries and economic losses. Bird et al., referencing the KABCO injury classification standard, conducted research on the conversion of accident injury levels. They unified the two injury types, death and injury, using "lost life years." Through methods such as life value assessment and disability-adjusted life years (DALYs), they adjusted the weights from the perspective of life loss and health-adjusted years, ultimately determining the severity conversion coefficient for death and serious injury to be 10:1. C The severity weighting coefficient at the time of injury ω k It is 0.1.

[0040] 2) Scaling factor β : In the accident severity index model, the scaling factor β In addition to eliminating direct economic losses E i The numerical scale must be accurate; logarithmic operations must also be ensured. The mathematical requirements. To make β The value of should be appropriate to the scale of economic losses in the dataset itself, and a statistic that can represent the "normal maximum value" of economic losses should be selected as a reference benchmark.

[0041] In this embodiment of the invention, the 85th percentile of economic losses in the sample is selected as the reference value. E a When direct economic losses E i Reaching the reference value E a When, its logarithm A preset "saturation level" should be reached. k This yields the following expression:

[0042] Solving this problem yields the following results:

[0043]

[0044] in, E a This represents the 85th percentile of direct economic losses in historical accident cases. k is a dimensionless constant representing the saturation scale.

[0045] Based on the general assumptions of the model's behavior, saturation scale k The value of needs to meet the following requirement: when the economic loss increases from zero to E a At that time, its contribution exhibits a significant nonlinear growth process. The time function curve satisfies a linear growth process, therefore we set... k The value is 1. k Substituting 1 into the above equation, we get:

[0046] 3) Direct economic losses E i : Direct economic losses from each accident E i The total economic cost corresponding to each death is expressed as: D = F + P + L in, D Indicates the overall economic cost; F Indicates the amount of the predetermined penalty; P It indicates the amount of compensation, including compensation for the families of the victims and compensation for emotional distress, etc. L This refers to the economic losses caused by production stoppage due to an accident; Considering that compensation and production stoppage losses may occur over multiple periods, they need to be discounted. The discount rate is set according to the present value formula in financial economics. r Then the present values ​​of compensation and production stoppage losses are respectively:

[0047]

[0048] in, T Indicates the duration of compensation and losses due to production stoppage; P t Indicates the first t Compensation for losses incurred during the period; L t Indicates the first t Losses due to production stoppage during the period.

[0049] (2) Next, taking mine accidents in the field of mine safety production as an example, we will explain the severity index of each accident.

[0050] Based on the sample data of 100 mining accidents studied in the embodiments of the present invention, the following calculations were performed: E a The value is 1762 (ten thousand yuan). Substituting the scaling factor mentioned above... β After calculating the formula, we get:

[0051] The "Regulations on Reporting and Handling Production Safety Accidents" stipulate the administrative penalty fines corresponding to different levels of accidents. Existing literature, "Valuing mortality risk reductions in a fast-developing society: A meta-analysis of stated preference studies in China from 1998 to 2019," calculates the statistical value of life for different risk scenarios, concluding that the value of life corresponding to environmental health risks is approximately 5 million RMB. Based on the above research and using typical fines for "major accidents" (10 to 30 deaths), combined with accident compensation and production stoppage losses, the comprehensive economic cost D corresponding to each death is ultimately set at 5 million RMB. Using this as a benchmark, we can obtain:

[0052] Substitution ω k =1, E i =5 million yuan, β =0.001, therefore:

[0053]

[0054] Therefore, a weighting coefficient for direct economic losses should be set. α The value is 2.47, thus making the value of life and economic costs comparable within a unified index system.

[0055] The final determination of the weighting coefficients incorporated the opinions of three experts in the field of mine safety. Through multiple rounds of consultation using the Delphi Method, the final accident severity index was determined as follows:

[0056] This final accident severity index reflects the overall severity of a mining accident.

[0057] In step S3 above, a random forest regression model is trained using multiple accident causative indicators of all accidents as feature variables and the severity index of the corresponding accident as the target variable; and based on the trained random forest regression model, the feature importance score of the influence of each accident causative indicator on the accident severity is obtained; specifically: With the rapid development of big data and artificial intelligence technologies, machine learning methods are widely used in mine accident prediction and risk assessment. Common algorithms include logistic regression, neural networks, and ensemble learning methods. Among them, random forest, as a typical ensemble learning algorithm, is widely used in complex system risk identification and multi-dimensional factor analysis due to its strong generalization ability and robustness. This paper implements the algorithm using Python's Scikit-learn library, dividing the dataset into training and test sets in an 8:2 ratio. Model performance is evaluated using the coefficient of determination (R²) and root mean square error (RMSE), and the contribution of each causal indicator is quantified using the importance ranking method, thereby scientifically identifying the key factors affecting the severity of mine accidents.

[0058] However, when constructing a predictive model targeting accident severity, selecting a suitable target variable is crucial. Traditional studies often simply use the number of deaths or direct economic losses as the target variable for model regression. However, single-dimensional severity indices ignore the interaction effect between casualties and economic losses, lack a unified metric, are not conducive to cross-accident comparisons, and can easily lead to model training distortion, reducing the scientific rigor of feature recognition results. Therefore, the Accident Severity Index (ASI) proposed in this embodiment is used as the target variable, and a random forest algorithm is employed for target feature recognition, as follows: Figure 2 As shown.

[0059] In the model construction process, the Accident Severity Index (ASI) is used as a continuous target variable, and accident causation-related indicators are used as input features. A random forest regression model is established to fit the accident severity and identify key risk factors. The specific steps include the following: (1) Determining input-output variables and constructing datasets.

[0060] First, considering both casualties and economic losses, we map them together as a dimensionless Accident Severity Index (ASI). Then, using the various accident causation indicators for all accidents as feature variables and the corresponding accident severity index as the target variable, we train a random forest regression model.

[0061] Taking mine accidents as an example, secondary risk factors related to the causes of accidents (such as violations of operating procedures, explosive substances / flammable gases, roof or sidewall defects, and emergency management deficiencies) are selected from the mine accident database as the independent variable feature vector X, and ASI is used as the dependent variable. y Secondary risk factors are represented by 0-1 variables to indicate whether they occur, while categorical variables such as production stage and mineral type are encoded using dummy variables (one-hot), thus forming a structured sample set suitable for training random forests.

[0062] (2) Dividing the training set and the test set.

[0063] To ensure the objectivity of the model evaluation, all accident samples are randomly divided into training and test sets according to a certain ratio (e.g., 80%:20%). The training set is used to fit the parameters of the random forest regression model, while the test set is used to test the model's prediction accuracy and generalization ability for ASI, avoiding obtaining an "overfit" severity fit effect only on the training samples.

[0064] (3) Random forest parameter settings and model training.

[0065] During the training phase, multiple subsets of samples are generated using the Bootstrap method based on the training set, and a separate regression decision tree is trained for each subset. During tree generation, candidate features are randomly selected from all features when splitting nodes, and the mean squared error (MSE) of the ASI is minimized as the splitting criterion, generating multiple decision trees with different structures layer by layer. By setting hyperparameters such as the number of decision trees, maximum depth, and minimum number of leaf node samples, the model's fitting ability and stability are balanced.

[0066] (4) Model performance evaluation and ASI-based feature importance identification.

[0067] After the random forest is trained, the model is used to make predictions using test set samples, and the coefficient of determination is calculated. The model's fit to the Accident Severity Index (ASI) was evaluated using metrics such as mean squared error (MSE) and mean absolute error (MAE). Assuming the model performance met requirements, the feature importance metric built into random forests was used to calculate the reduction in ASI prediction error resulting from each input feature's participation in node partitioning across all decision trees. This reduction was then normalized at the forest level to obtain the feature importance score for each risk factor. A higher score indicates a more significant impact of the factor on the ASI. Finally, the risk factors were ranked from highest to lowest feature importance, providing a quantitative basis for subsequent risk value calculation and risk level classification.

[0068] pass Figure 2 It is clear that by using the Severity Index (ASI) as the target variable, the Random Forest algorithm brings crucial optimizations to the process. The added "ASI calculation and standardization" step separates the original, multi-dimensional consequence data (number of deaths, number of injuries, and economic losses) into data dimensions, determines the weights of casualties and economic losses through different parameters, and finally fuses the multi-dimensional accident data into a unified, continuous, and dimensionless index using the ASI model. This overcomes the inherent shortcomings of traditional methods, where the target variable (such as the number of deaths) is discrete, discontinuous, and fails to reflect economic losses.

[0069] In step S4 above, the risk value of each accident causation indicator is calculated based on the feature importance score and the probability of occurrence of each accident causation indicator in historical accident cases; wherein, the risk value of each accident causation indicator is the product of the probability of occurrence of each accident causation indicator and the corresponding feature importance score.

[0070] In step S5 above, based on the risk value, each accident causative indicator is classified into risk levels to obtain a quantitative risk classification result for each accident causative indicator. This risk classification is divided into four levels: (1) Level 1 risk: The probability of occurrence is higher than the first threshold and the feature importance score is higher than the second threshold; (2) Secondary risk: The probability of occurrence is lower than the first threshold but the feature importance score is higher than the third threshold, or the probability of occurrence is higher than the first threshold and the feature importance score is between the second threshold and the third threshold; (3) Level 3 risk: The probability of occurrence is higher than the first threshold but the feature importance score is lower than the fourth threshold, or the probability of occurrence is lower than the first threshold but the feature importance score is higher than the fourth threshold; (4) Lower risk: The probability of occurrence is lower than the first threshold and the feature importance score is lower than the fourth threshold.

[0071] In summary, this invention provides a method for quantifying accident causation risk analysis based on an accident severity index, focusing on a novel Accident Severity Index (ASI) calculation model. Addressing the issue of a single quantitative indicator for accident severity, this model innovatively incorporates two core consequence variables—personnel injury and economic loss—into a unified quantitative framework based on utility theory and the principle of diminishing marginal utility. A scaling factor is introduced through a nonlinear function to achieve dynamic representation of severity. This invention combines this index with a random forest algorithm to construct an ASI-based accident risk identification model. Using mining accidents as a typical case, it quantifies the impact of different risk factors on accident severity, achieving an organic integration from accident consequence representation to causal risk identification.

[0072] Next, we will explain the collection and analysis of accident cases.

[0073] 1. Accident case collection: Mining is a high-risk industry. Complex geological conditions and mining processes, along with frequent interactions between humans, machines, and the environment, lead to frequent major mining accidents resulting in mass casualties. Modernizing and refining safety governance capabilities continues to present challenges. Studying the patterns of mine accidents, identifying key hazard factors, and constructing a unified accident severity assessment system have become hot topics and difficult areas in safety research. Therefore, this invention selects mine accidents as typical cases for analysis. This invention collects and organizes 280 mine safety accident cases with publicly reported news coverage, analyzes and verifies the reliability and sensitivity of the Accident Severity Index (ASI) model, and conducts feature importance analysis and accident risk factor classification.

[0074] 2. Results of accident risk factor analysis: 2.1 Classification of Accident Risk Factors: Referring to the standard GB / T13861 "Classification and Code of Hazardous and Harmful Factors in Production Processes", the risk factors of mine safety accidents are classified from the dimensions of people, materials, environment and management according to the potential risk factors in the mining industry. The risk factors are shown in Table 1.

[0075] Table 1: Risk Factors of Mine Accidents

[0076] 2.2 Reliability and sensitivity analysis of the Accident Severity Index (ASI) model: Figure 3A correlation analysis was conducted on the number of casualties, economic losses, and calculated accident severity based on some mine accident cases published by the Mine Safety Administration. The results show that the Accident Severity Index (ASI) has a high correlation of 0.975 with the number of casualties and a high correlation of 0.899 with economic losses. This result indicates that ASI can effectively integrate the dominance of the traditional core indicator of the number of casualties with the diminishing marginal effect of economic losses into a unified quantitative framework. Furthermore, while the correlation between the number of casualties and economic losses is relatively weak (0.806), ASI maintains a strong correlation with both, indicating that it can serve as a comprehensive and robust evaluation indicator to provide a reliable target variable for subsequent risk characteristic identification, fundamentally avoiding the risk of model distortion due to biased target variables.

[0077] Figure 4 This is a schematic diagram illustrating the three-dimensional sensitivity distribution of the Accident Severity Index (ASI) to key input parameters; where, Figure 4 Figure (a) shows the distribution of ASI's sensitivity to the number of injured and dead; Figure 4 Figure (b) shows the distribution of ASI sensitivity to the number of deaths and economic losses; Figure 4 Figure (c) shows the sensitivity distribution of ASI to the number of injured and economic losses. The analysis results indicate significant differences in the degree of influence and interaction effects of different variables on ASI. Comprehensive analysis shows that the number of deaths is the core factor determining ASI, followed by economic losses, while the impact of the number of injured is relatively limited. This sensitivity structure verifies that the ASI model can effectively capture the diminishing marginal returns and nonlinear characteristics of accident consequences, consistent with the design intent of incorporating utility theory during model construction, and confirms its rationality and robustness as a comprehensive evaluation indicator.

[0078] 2.3 Feature Importance Analysis Based on Accident Severity Index: Based on the random forest analysis process, after learning from the training set data and building a prediction model, different risk factors of the accident are analyzed according to the Accident Severity Index (ASI), resulting in a schematic diagram of the feature importance results of the risk factors, as shown below. Figure 5 As shown. (Through) Figure 5 It can be seen that the top five most important features are B5 explosive substances / flammable gases, A6 illegal operations, C7 groundwater, C2 mine working face defects, and D2 inadequate or unimplemented responsibility system. Among them, environmental factors account for the largest proportion, indicating that environmental factors are most closely related to the severity of the accident.

[0079] 3. Priority Analysis of Mine Accident Risk Prevention and Control: To more comprehensively assess the risk factors of mining accidents, a probability-importance dual-dimensional analysis framework was further introduced based on the random forest model to quantify the impact of each accident risk factor on the severity of the accident. The occurrence probability of the risk factors is shown in Table 2.

[0080] Table 2: Probability of Occurrence of Different Risk Factors

[0081] To move from a single-dimensional "importance" ranking to a two-dimensional "risk priority" assessment, this paper draws on the basic ideas of the LS risk analysis method. The risk value is defined as the product of the probability of occurrence of a risk factor and its characteristic importance. Based on this, various risk factors are classified into different levels (see...). Figure 6 This effectively enhances the scientific rigor and relevance of risk identification.

[0082] Figure 6 The results show the risk level distribution of mine accident risk factors. The risk value of A6 (violation of operating procedures) is significantly higher than other factors, reaching 0.069. This is because it has an extremely high probability of occurrence in mine accidents and is prone to causing serious consequences, easily leading to mass casualties; therefore, it is classified as a Level 1 risk factor.

[0083] In contrast, the risk values ​​of six factors, including B5 (explosive materials / flammable gases) and C1 (mine roof or sidewall defects), range from 0.022 to 0.009. Although lower than A6, they are still at a relatively high level overall, and are therefore classified as Level 2 risk factors. Furthermore, the risk values ​​of eight factors, including B2 (protection deficiencies) and D4 (emergency management deficiencies), are relatively low, characterized by either "high probability but low importance" or "high importance but low probability." The former occurs frequently but has a limited direct impact on accidents; the latter, once it occurs, has a serious impact, but its triggering probability is low, classifying it as a Level 3 risk factor. Finally, D1 (inadequate personnel) and C5 (underground fire) have both low probability of occurrence and low characteristic importance, resulting in the lowest risk values. These are classified as low-risk factors, posing a relatively limited overall threat to mine safety.

[0084] In summary, risk-value-based hierarchical assessment can clearly identify key risk factors and their underlying causes in mine accidents, providing a quantitative basis for the formulation of differentiated management strategies.

[0085] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0086] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for quantitative risk analysis of accident causes based on an accident severity index, characterized in that, Includes the following steps: Acquire historical accident case data in the target field, including casualty information, direct economic losses, and multiple corresponding accident causation indicators for each accident; Based on the casualty information and direct economic losses of each accident, a severity index is obtained for each accident. Using the multiple accident causation indicators of all accidents as feature variables and the severity index of the corresponding accident as the target variable, a random forest regression model is trained; and based on the trained random forest regression model, the feature importance score of the influence of each accident causation indicator on the accident severity is obtained. Based on the feature importance scores and the probability of occurrence of each accident causation indicator in historical accident cases, the risk value of each accident causation indicator is calculated. Based on the risk values, the risk levels of each accident causative indicator are classified to obtain the quantitative risk classification results of each accident causative indicator.

2. The method for quantifying accident causation risk analysis based on accident severity index as described in claim 1, characterized in that, The severity index of each accident is expressed as follows: in, ASI i Indicates the first i The severity index of the accident; C Represents the set of types of casualties; S i,k Indicates the first i The types of injuries and fatalities in the accident were: k The number of casualties; ω k Indicates the type of casualties. k Severity weighting coefficient; E i Indicates the first i The direct economic losses caused by the accident; α Weighting coefficients representing direct economic losses; β Indicates the scaling factor; ln This represents the natural logarithm function.

3. The accident causation quantification risk analysis method based on accident severity index as described in claim 2, characterized in that, The severity weighting coefficient for fatal injuries is 1; the severity weighting coefficient for injury injuries is 0.

1.

4. The method for quantifying accident causation risk analysis based on accident severity index as described in claim 2, characterized in that, The scaling factor β Determined based on the 85th percentile of direct economic losses in historical accident cases; expressed as: in, E a This represents the 85th percentile of direct economic losses in historical accident cases. k is a dimensionless constant representing the saturation scale.

5. The accident causation quantification risk analysis method based on accident severity index as described in claim 2, characterized in that, The direct economic loss of each accident is the comprehensive economic cost corresponding to each death, expressed as: D = F + P + L in, D Indicates the overall economic cost; F Indicates the amount of the predetermined penalty; P Indicates the amount of compensation; L This refers to the economic losses caused by production stoppage due to an accident; T Indicates the duration of compensation and production stoppage losses; P t Indicates the first t Compensation for losses incurred during the period; L t Indicates the first t Losses due to production stoppage during the period; r This represents the discount rate.

6. The method for quantifying accident causation risk analysis based on accident severity index as described in claim 1, characterized in that, When the target area is the field of mine safety production, the severity index of each mine accident is expressed as: in, ASI i Indicates the first i The severity index of the accident; C Represents the set of types of casualties; S i,k Indicates the first i The types of injuries and fatalities in the accident were: k The number of casualties; ω k Indicates the type of casualties. k Severity weighting coefficient; E i Indicates the first i The direct economic loss caused by the accident; 2.47 represents the weighting coefficient for direct economic loss; 0.001 represents the scaling factor; ln This represents the natural logarithm function.

7. The method for quantifying accident causation risk analysis based on accident severity index as described in claim 1, characterized in that, The risk value of each accident causation indicator is the product of the probability of occurrence of each accident causation indicator and the corresponding feature importance score.

8. The accident causation quantification risk analysis method based on accident severity index as described in claim 1, characterized in that, The risk levels are divided into four levels: Level 1 Risk: The probability of occurrence is higher than the first preset threshold and the feature importance score is higher than the second preset threshold; Level 2 risk: The probability of occurrence is lower than the first preset threshold but the feature importance score is higher than the third preset threshold, or the probability of occurrence is higher than the first preset threshold and the feature importance score is between the second preset threshold and the third preset threshold; Level 3 risk: The probability of occurrence is higher than the first preset threshold but the feature importance score is lower than the fourth preset threshold, or the probability of occurrence is lower than the first preset threshold but the feature importance score is higher than the fourth preset threshold; Lower risk: The probability of occurrence is lower than the first preset threshold and the feature importance score is lower than the fourth preset threshold.

Images