Feature selection method and system for screening chemical accident influence factors

Abstract

The invention discloses a feature selection method and system for screening chemical accident influence factors, and the method comprises the steps: determining all features causing a chemical accident according to the historical chemical data of a target chemical device region, and screening a plurality of features meeting a preset importance requirement for the chemical accident from the features as the pre-selected features of the chemical accident in the current time period; inputting the pre-selected features into a preset feature subset optimization model to obtain an optimal feature subset; for each feature in the pre-selected features, obtaining evaluation results which are made based on expert knowledge and respectively represent the feature importance of the current time period and the feature importance of the historical time period, evaluating the reliability degree of each evaluation result according to the source of each evaluation result, and further combining with the optimal feature subset to obtain the feature importance of the current time period and the feature importance of the historical time period. And generating a comprehensive score value representing the actual importance of each feature to the chemical accident in the current time period, thereby obtaining the optimal feature adaptive to the chemical accident in the current time period. The characteristics causing chemical accidents can be scientifically selected.

Description

Technical Field

[0001] This invention belongs to the field of chemical safety technology, and in particular relates to a feature selection method and system for screening influencing factors of chemical accidents. Background Technology

[0002] Choosing appropriate descriptors or features is one of the core issues in using machine learning models to explore chemical plant parameters and chemical accidents. For complex real-world chemical accident processes, the mapping relationship between independent and dependent variables is difficult to represent uniformly using a single linear or nonlinear approach.

[0003] Existing technology discloses a soft measurement method for dioxin (DXN) emission concentration in solid waste incineration (MSWI) processes based on multi-level feature selection, belonging to the field of soft measurement. First, from the perspective of the correlation between a single feature and DXN, a comprehensive evaluation index is constructed by combining correlation coefficients and mutual information to achieve the first-level feature selection of process variables in multiple subsystems of MSWI. Next, from the perspective of multi-feature redundancy and feature selection robustness, a GA-PLS-based feature selection algorithm is run multiple times to achieve the second-level feature selection. Finally, combining the statistical frequency of the selected features from the previous level, model prediction performance, and mechanistic knowledge, a third-level feature selection is performed to construct a soft measurement model for DXN emission concentration, achieving effective measurement of DXN emission concentration.

[0004] The prior art also discloses a cancer gene classification method, device, and storage medium based on two-stage deep feature selection, including: training a cancer gene classification model and cancer gene classification. In the process of training the cancer gene classification model: (1) training data is acquired; in the first stage, three feature selection algorithms are integrated to perform comprehensive feature selection, resulting in a feature subset; in the second stage, an unsupervised neural network is used to obtain the optimal representation of the feature subset; (2) the optimal representation of the feature subset is divided into a training set and a test set, and input into the neural network for training. In the cancer gene classification process, the preprocessed cancer gene data to be detected is input into the trained cancer gene classification model to achieve cancer gene classification. This cancer gene classification method uses an integrated feature selection method to achieve feature selection considering various aspects, and uses an unsupervised neural network to extract the optimal representation of features, obtaining cleaner gene features and improving classification accuracy.

[0005] In addition, existing technologies also disclose a site suitability evaluation method for bottom-seeded scallop aquaculture areas based on the analytic hierarchy process (AHP), including the following steps: constructing a site selection evaluation index system; using multiple normalization methods to quantify all secondary indicators in a targeted manner, establishing a quantitative evaluation standard corresponding to each secondary indicator; adding a classification and quantification process for the secondary indicator system; calculating the normalized values ​​of all secondary indicators step by step by establishing quantitative evaluation standards, and then using the normalized values ​​of the secondary indicators to calculate the normalized values ​​of the primary indicators; using the normalized values ​​of the primary indicators and their weights, calculating the evaluation result numerical value using a weighted linear combination method, and giving the suitability level evaluation result, thereby realizing the suitability evaluation of the target sea area site selection. This evaluation method makes the marine ranch site selection evaluation process more objective, targeted, scientific, and comprehensive, and can achieve efficient calculation and intuitive presentation of the evaluation results.

[0006] Feature selection (FS) mainly includes three types of methods: stability selection, recursive feature elimination, and univariate feature selection based on mutual information. Although these methods yield models with good generalization performance, they require a series of parameter adjustments and model selections.

[0007] In developing this invention, the inventors discovered that current feature selection algorithms typically require tedious hyperparameter tuning and neglect expert prior knowledge about the relevance of certain features, potentially leading to the deletion of some critical features. However, in the face of complex and variable chemical accident data and processes, complex and ambiguous correlations exist not only between features and target attributes but also between features themselves. Identifying and eliminating irrelevant and redundant features, retaining only representative features from the original feature set, is crucial. Therefore, there is an urgent need for a feature selection scheme for chemical accidents that combines the advantages of expert knowledge and multi-layered feature selection, eliminates the risk of critical feature deletion, ensures that selected features are highly correlated with target attributes, successfully replaces the time-consuming model hyperparameter trial-and-error process, and provides the same or better predictive performance with a smaller scale and interpretable feature set. Summary of the Invention

[0008] To address the aforementioned problems, this invention provides a feature selection method for screening influencing factors of chemical accidents, comprising: determining all features that cause chemical accidents based on historical chemical data of a target chemical plant area, and selecting multiple features that meet preset importance requirements for chemical accidents as pre-selected features for the current period; inputting the pre-selected features into a preset feature subset optimization model to obtain an optimal feature subset; for each feature in the pre-selected features, obtaining several first evaluation results representing the importance of features in the current period based on expert knowledge, and several second evaluation results representing the importance of features in historical periods, and evaluating the reliability of each evaluation result based on its source, and then combining the optimal feature subset to generate a comprehensive score value representing the actual importance of each feature to the current period's chemical accident, thereby obtaining the best features adapted to the current period's chemical accident.

[0009] Preferably, the step of selecting multiple features that meet the preset importance requirements for chemical accidents as pre-selected features of chemical accidents in the current period includes: performing sparsity assessment, correlation assessment and redundancy assessment on all the features respectively, so as to select multiple features that simultaneously meet the preset sparsity requirements, preset correlation requirements and preset redundancy requirements as the multiple features that meet the preset importance requirements for chemical accidents, thereby obtaining the pre-selected features of chemical accidents in the current period.

[0010] Preferably, the preset feature subset selection model adopts an ML model.

[0011] Preferably, the first evaluation result and the second evaluation result are important, unimportant, or their importance cannot be determined. The step of generating a comprehensive score representing the actual importance of each feature to the current period's chemical accident includes: assigning a score value to each evaluation result according to its importance; for each feature in the pre-selected features, recording the score value of each first evaluation result as a first score value; for each feature in the pre-selected features, extracting second evaluation results made by experts with experience in evaluating the influencing factors of chemical accidents from among the plurality of second evaluation results, recording the score value of each extracted second evaluation result as a second score value, then determining the reliability of each extracted second evaluation result based on the professional field of the expert who made each extracted second evaluation result, and assigning a weight to the corresponding second evaluation result based on the reliability; based on this, combining the first score value and the second score value, calculating the total score value representing the comprehensive importance of each feature to the current period's chemical accident; and correcting the total score value based on the feature overlap state between the pre-selected features and the optimal feature subset, thereby obtaining the comprehensive score value.

[0012] Preferably, the total score is calculated using the following expression:

[0013]

[0014] Where f represents the total score, f d b1 and b2 represent coefficients, m represents the total number of first and second evaluation results, n represents the number of evaluation results that are the same as the actual evaluation result, and f represents the first rating value. j denoted by , w represents the weight, N represents the total number of first evaluation results, and x represents the sequence number of the first evaluation result.

[0015] Preferably, in determining the reliability of each extracted second evaluation result based on the professional field to which the expert who made each extracted second evaluation result belongs, the process includes: if the professional field is chemical engineering, the reliability of the extracted second evaluation result is determined to be the highest; if the professional field is emergency response, the reliability of the extracted second evaluation result is determined to be medium; if the professional field is neither chemical engineering nor emergency response, the reliability of the extracted second evaluation result is determined to be the lowest.

[0016] Preferably, the process of assigning weights to the corresponding second evaluation results includes: assigning a weight of 2 to the second evaluation result with the highest reliability; assigning a weight of 1.5 to the second evaluation result with medium reliability; and assigning a weight of 1 to the second evaluation result with the lowest reliability.

[0017] Preferably, in the process of configuring a score value for each evaluation result according to its importance, the following steps are taken: for evaluation results that are important, unimportant, and whose importance cannot be determined, a score value of 1 point, 0.5 points, and 0 points are configured respectively.

[0018] Preferably, the process of obtaining the comprehensive score includes: comparing the pre-selected features and the optimal feature subset, assigning different score values ​​to the overlapping and non-overlapping features, and then using the assigned score values ​​to correct the corresponding total score value to obtain the corresponding comprehensive score value.

[0019] Preferably, each overlapping feature is assigned a score of 1, and each non-overlapping feature is assigned a score of 0.

[0020] The present invention also provides a computer-readable storage medium comprising a series of instructions for performing a feature selection method step for screening factors influencing chemical accidents.

[0021] On the other hand, the present invention also provides a feature selection system for screening influencing factors of chemical accidents. The feature selection system includes the following modules: a feature pre-selection module, which is used to determine all features that cause chemical accidents based on historical chemical data of the target chemical plant area, and select multiple features that meet the preset importance requirements for chemical accidents as pre-selected features for chemical accidents in the current period; a feature subset optimization module, which is used to input the pre-selected features into a preset feature subset optimization model to obtain the optimal feature subset; and an optimal feature selection module, which is used to obtain several first evaluation results representing the importance of features in the current period and several second evaluation results representing the importance of features in historical periods for each feature in the pre-selected features, and evaluate the reliability of each evaluation result according to the source of each evaluation result, and then combine the optimal feature subset to generate a comprehensive score value representing the actual importance of each feature to the chemical accident in the current period, thereby obtaining the optimal feature that is suitable for the chemical accident in the current period.

[0022] Compared with the prior art, one or more embodiments of the above solutions may have the following advantages or beneficial effects:

[0023] This invention proposes a feature selection method and system for screening influencing factors of chemical accidents. The method first acquires all features that cause chemical accidents and then selects multiple features that meet preset importance requirements for the chemical accident as pre-selected features for the current period. Next, a preset feature subset optimization model is constructed to obtain the optimal feature subset for the current period from the pre-selected features. Finally, for each feature in the pre-selected features, evaluation results based on expert knowledge are obtained, representing the importance of features in the current period and the importance of features in historical periods, respectively. The reliability of each evaluation result is evaluated according to its source. Then, a total score representing the overall importance of each feature to the current period is generated, and the total score is corrected based on the optimal feature subset to generate a comprehensive score representing the actual importance of each feature to the current period, thereby obtaining the optimal features suitable for the current period's chemical accident. This invention takes into account expert experience and data patterns, and combines the advantages of multi-layer feature selection and expert knowledge. It eliminates the risk of key features being deleted and avoids time-consuming model hyperparameter trial and error processes. The final selected features are a small-scale and interpretable feature set that is highly correlated with the target attribute (i.e., the occurrence of chemical accidents). This enables the scientific selection of features that cause chemical accidents and provides technical support for accurate and efficient prediction of accident development.

[0024] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the description, claims, and drawings. Attached Figure Description

[0025] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0026] Figure 1 This is a flowchart illustrating the steps of a feature selection method for screening influencing factors of chemical accidents, as described in an embodiment of this application.

[0027] Figure 2 This is a block diagram of a feature selection system for screening influencing factors of chemical accidents, as described in an embodiment of this application. Detailed Implementation

[0028] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings and examples, so that the process of how the present invention uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly. It should be noted that, as long as there is no conflict, the various embodiments and features in the various embodiments of the present invention can be combined with each other, and the resulting technical solutions are all within the protection scope of the present invention.

[0029] Furthermore, the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0030] Choosing appropriate descriptors or features is one of the core issues in using machine learning models to explore chemical plant parameters and chemical accidents. For complex real-world chemical accident processes, the mapping relationship between independent and dependent variables is difficult to represent uniformly using a single linear or nonlinear approach.

[0031] Existing technology discloses a soft measurement method for dioxin (DXN) emission concentration in solid waste incineration (MSWI) processes based on multi-level feature selection, belonging to the field of soft measurement. First, from the perspective of the correlation between a single feature and DXN, a comprehensive evaluation index is constructed by combining correlation coefficients and mutual information to achieve the first-level feature selection of process variables in multiple subsystems of MSWI. Next, from the perspective of multi-feature redundancy and feature selection robustness, a GA-PLS-based feature selection algorithm is run multiple times to achieve the second-level feature selection. Finally, combining the statistical frequency of the selected features from the previous level, model prediction performance, and mechanistic knowledge, a third-level feature selection is performed to construct a soft measurement model for DXN emission concentration, achieving effective measurement of DXN emission concentration.

[0032] The prior art also discloses a cancer gene classification method, device, and storage medium based on two-stage deep feature selection, including: training a cancer gene classification model and cancer gene classification. In the process of training the cancer gene classification model: (1) training data is acquired; in the first stage, three feature selection algorithms are integrated to perform comprehensive feature selection, resulting in a feature subset; in the second stage, an unsupervised neural network is used to obtain the optimal representation of the feature subset; (2) the optimal representation of the feature subset is divided into a training set and a test set, and input into the neural network for training. In the cancer gene classification process, the preprocessed cancer gene data to be detected is input into the trained cancer gene classification model to achieve cancer gene classification. This cancer gene classification method uses an integrated feature selection method to achieve feature selection considering various aspects, and uses an unsupervised neural network to extract the optimal representation of features, obtaining cleaner gene features and improving classification accuracy.

[0033] In addition, existing technologies also disclose a site suitability evaluation method for bottom-seeded scallop aquaculture areas based on the analytic hierarchy process (AHP), including the following steps: constructing a site selection evaluation index system; using multiple normalization methods to quantify all secondary indicators in a targeted manner, establishing a quantitative evaluation standard corresponding to each secondary indicator; adding a classification and quantification process for the secondary indicator system; calculating the normalized values ​​of all secondary indicators step by step by establishing quantitative evaluation standards, and then using the normalized values ​​of the secondary indicators to calculate the normalized values ​​of the primary indicators; using the normalized values ​​of the primary indicators and their weights, calculating the evaluation result numerical value using a weighted linear combination method, and giving the suitability level evaluation result, thereby realizing the suitability evaluation of the target sea area site selection. This evaluation method makes the marine ranch site selection evaluation process more objective, targeted, scientific, and comprehensive, and can achieve efficient calculation and intuitive presentation of the evaluation results.

[0034] Feature selection (FS) mainly includes three types of methods: stability selection, recursive feature elimination, and univariate feature selection based on mutual information. Although these methods yield models with good generalization performance, they require a series of parameter adjustments and model selections.

[0035] In developing this invention, the inventors discovered that current feature selection algorithms typically require tedious hyperparameter tuning and neglect expert prior knowledge about the relevance of certain features, potentially leading to the deletion of some critical features. However, in the face of complex and variable chemical accident data and processes, complex and ambiguous correlations exist not only between features and target attributes but also between features themselves. Identifying and eliminating irrelevant and redundant features, retaining only representative features from the original feature set, is crucial. Therefore, there is an urgent need for a feature selection scheme for chemical accidents that combines the advantages of expert knowledge and multi-layered feature selection, eliminates the risk of critical feature deletion, ensures that selected features are highly correlated with target attributes, successfully replaces the time-consuming model hyperparameter trial-and-error process, and provides the same or better predictive performance with a smaller scale and interpretable feature set.

[0036] Therefore, to address the aforementioned problems, this invention proposes a feature selection method and system for screening influencing factors of chemical accidents. This feature selection method first acquires all features that cause chemical accidents and then selects multiple features that meet preset importance requirements for the chemical accident as pre-selected features for the current period. Next, a preset feature subset optimization model is constructed to obtain the optimal feature subset for the current period from the pre-selected features. Finally, for each feature in the pre-selected features, evaluation results based on expert knowledge are obtained, representing the importance of features in the current period and the importance of features in historical periods, respectively. The reliability of each evaluation result is evaluated based on its source. Then, a total score representing the overall importance of each feature to the current period's chemical accident is generated, and the total score is corrected based on the optimal feature subset to generate a comprehensive score representing the actual importance of each feature to the current period's chemical accident, thereby obtaining the optimal features suitable for the current period's chemical accident. This invention takes into account expert experience and data patterns, and combines the advantages of multi-layer feature selection and expert knowledge. It eliminates the risk of key features being deleted and avoids time-consuming model hyperparameter trial and error processes. The final selected features are a small-scale and interpretable feature set that is highly correlated with the target attribute (i.e., the occurrence of chemical accidents). This enables the scientific selection of features that cause chemical accidents and provides technical support for accurate and efficient prediction of accident development.

[0037] Example 1

[0038] Figure 1 This diagram illustrates the steps of a feature selection method for screening influencing factors of chemical accidents, as described in an embodiment of this application. See below for reference. Figure 1 The various steps of the present invention will be explained below.

[0039] like Figure 1 As shown, in step S110, based on historical chemical data of the target chemical plant area, all characteristics that cause chemical accidents are determined, and multiple characteristics that meet preset importance requirements for chemical accidents are selected as pre-selected characteristics for chemical accidents in the current period. Specifically, this embodiment first extracts chemical data at the time of each historical chemical accident from the historical chemical data of the target chemical plant area. Then, by analyzing the data changes of each characteristic at the time of the historical chemical accidents, all characteristics that cause historical chemical accidents are obtained. After determining all characteristics that cause chemical accidents, based on a preset multi-layer data evaluation structure, the importance of each characteristic that causes historical chemical accidents to the occurrence (probability) of chemical accidents is analyzed, and then multiple characteristics that meet preset importance requirements for chemical accidents are selected from all characteristics that cause historical chemical accidents as pre-selected characteristics for chemical accidents in the current period.

[0040] In one specific embodiment of this application, the preset multi-layer data evaluation structure is represented by the following expression:

[0041]

[0042] Where Multi_Layer(X) represents all input features of the multi-layer data evaluation structure, Layer1(X) represents all input features of the first layer, Layer2(X) represents all input features of the second layer, and Layer3(X) represents all input features of the third layer, x i Let represent the i-th feature among all input features, ... γ represents the relevance threshold, redundancy represents redundancy, and γ represents the redundancy threshold.

[0043] In the step of selecting multiple features that meet the preset importance requirements for chemical accidents as pre-selected features for chemical accidents in the current period, sparsity evaluation, correlation evaluation, and redundancy evaluation are performed on all features respectively. This results in multiple features that simultaneously meet the preset sparsity requirements, preset correlation requirements, and preset redundancy requirements being selected as features that meet the preset importance requirements for chemical accidents, thus obtaining the pre-selected features for chemical accidents in the current period. In this embodiment, based on data-driven multi-layer feature selection, the three processing layers in the aforementioned preset multi-layer data evaluation structure are used. All input features of the multi-layer data evaluation structure are used as the input features of the first layer, the second layer, and the third layer, respectively. Simultaneously, features with sparsity less than or equal to a preset sparsity threshold are used as output features of the first layer; features with correlation less than or equal to a preset correlation threshold are used as output features of the second layer; and features with redundancy greater than or equal to a preset redundancy threshold are used as output features of the third layer. Subsequently, features that exist simultaneously in the output features of the first, second, and third layers are extracted. At this point, the extracted features are highly differentiated from and highly correlated with the occurrence of chemical accidents. Therefore, in this embodiment, the extracted features are used as multiple features that meet the preset importance requirements for chemical accidents. In other words, the extracted multiple features are the pre-selected features of chemical accidents in the current time period.

[0044] Furthermore, in step S120, the pre-selected features are input into a preset feature subset optimization model to obtain the optimal feature subset. In this embodiment, feature subset evaluation is performed based on a machine learning model. The pre-selected features are input into the preset feature subset optimization model for automatic filtering and quality assessment, iterating until the feature attributes are represented by a smaller, more interpretable feature set. The optimal feature subset can be obtained without manual tuning of hyperparameters. That is, the pre-selected features from step S110 are input into the preset feature subset optimization model, and the output information of the preset feature subset optimization model is the optimal feature subset. In this way, the number of features contained in the optimal feature subset is much smaller than that of the pre-selected features, which is beneficial for establishing a closer and more accurate structure-property relationship between the selected features and target attributes of the chemical accident in the current period.

[0045] In one specific embodiment of this application, the preset feature subset optimization model adopts an ML model. Specifically, each layer in the ML model of this embodiment is configured with a preset filtering threshold of different values. In the process of obtaining the optimal feature subset, the pre-selected features of the chemical accident in the current time period are first filtered through the first layer of the ML model according to the first layer filtering threshold. By retaining a number of features that are appropriate to the filtering threshold, an initial feature subset (candidate feature subset) is formed. At this time, the quality of the initial feature subset is evaluated to determine whether the initial feature subset can be passed to the next layer. If the initial feature subset can be passed to the next layer, the initial feature subset is passed to the next layer, and filtering and quality evaluation are continued according to the next layer filtering threshold. Then, the corresponding filtering and quality evaluation are performed in each subsequent layer until the filtered feature subset is not as good as the feature subset of the previous layer. At this time, the ML model outputs the feature subset of the previous layer as the optimal feature subset, or until the number of features contained in the filtered feature subset reaches the preset minimum number of features. At this time, the ML model outputs the filtered feature subset that has reached the preset minimum number of features as the optimal feature subset. This invention fully considers the characteristics of large volume and long processing time of chemical data. Based on a combination of sparsity assessment, correlation assessment, redundancy assessment and machine learning, it automatically selects the features most relevant to causing chemical accidents and establishes a closer and more accurate structure-activity relationship between the selected features and target attributes of chemical accidents in the current period, laying the foundation for the scientific selection of features that cause chemical accidents.

[0046] It should be noted that the present invention does not specifically limit the number of layers in the preset feature subset optimization model. Those skilled in the art can set the number of pre-selected features of the chemical accident in the current period and the feature selection accuracy.

[0047] In one specific embodiment of this application, in determining whether a feature subset can be passed to the next layer, a pre-trained learning model based on Support Vector Machine (SVM), Neural Network (NN), or Decision Tree is used to filter the feature subsets at each layer. Then, evaluation parameters such as Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) are used to determine whether the initial feature subset can be passed to the next layer. For example, if the RMSE and MAPE of the initial feature subset are smaller than the pre-selected features of the chemical accident in the current time period, then the initial feature subset can be passed to the next layer.

[0048] Further, in step S130, for each feature in the pre-selected features, several first evaluation results representing the importance of features in the current time period and several second evaluation results representing the importance of features in historical time periods are obtained based on expert knowledge. The reliability of each evaluation result is evaluated according to its source. Then, by combining the optimal feature subset, a comprehensive score value representing the actual importance of each feature to the chemical accident in the current time period is generated, thereby obtaining the best feature adapted to the chemical accident in the current time period. In practical applications, when data evaluation is applied to machine learning, the importance of features depends on the domain knowledge of experts. Therefore, this embodiment quantifies expert knowledge through weighted scoring and integrates the quantification results into the feature selection process. Statistical analysis methods and information theory are used to analyze the relationship between features and target attributes, thereby eliminating the risk of key features being deleted by removing the most redundant or least relevant features associated with the target attributes. For each feature in the pre-selected features, several first evaluation results representing the importance of features in the current time period and several second evaluation results representing the importance of features in historical time periods are obtained from different experts. Because the experts who make each evaluation result typically have differences in experience and expertise, the reliability of each evaluation result also varies. Therefore, this embodiment continues to evaluate the reliability of each evaluation result based on the experience and expertise of the experts who made each evaluation result (i.e., the source of each evaluation result). Then, by assigning values ​​to the importance and reliability represented by each evaluation result, and combining this with the correction effect of the optimal feature subset (described below), a comprehensive score value representing the actual importance of each feature to the current period's chemical accident is generated. Finally, based on the comprehensive score value, the best features (i.e., a specified number of features most relevant to causing the current period's chemical accident) are selected.

[0049] In the step of generating a comprehensive score representing the actual importance of each feature to the current period's chemical accident, firstly, a score value is assigned to each evaluation result according to its importance; then, for each feature in the pre-selected features, the score value of each first evaluation result is recorded as the first score value; next, for each feature in the pre-selected features, several second evaluation results made by experts with experience in evaluating the influencing factors of chemical accidents are extracted, and the score value of each extracted second evaluation result is recorded as the second score value; then, based on the professional field of the expert who made each extracted second evaluation result, the reliability of each extracted second evaluation result is determined, and a weight is assigned to the corresponding second evaluation result according to the reliability; based on this, the total score representing the comprehensive importance of each feature to the current period's chemical accident is calculated by combining the first score value and the second score value; finally, the total score value is corrected according to the feature overlap state of the pre-selected features and the optimal feature subset, thereby obtaining the comprehensive score value.

[0050] In one specific embodiment of this application, the first evaluation result and the second evaluation result are considered important, unimportant, or their importance cannot be determined. First, for each feature in the pre-selected features, a score is assigned to each first and second evaluation result, considering only the importance represented by the evaluation result itself, without considering the source. Next, for each feature in the pre-selected features, a score representing the current importance of that feature to the chemical accident is obtained based on the aforementioned score assigned to each evaluation result, and is denoted as the first score. Then, for each feature in the pre-selected features, several second evaluation results made by experts with experience in evaluating the influencing factors of chemical accidents are extracted. Based on this, and according to the aforementioned score assigned to each evaluation result, the score of each extracted second evaluation result is used as a score representing the historical importance of that feature to the chemical accident, and is denoted as the second score. Further, based on the professional field of the expert who made each extracted second evaluation result, the reliability of each extracted second evaluation result is determined, and based on the reliability, a score that considers only the source, without considering the importance represented by the evaluation result itself, is assigned as a weight for the corresponding second evaluation result, thereby completing the weight assignment. Next, the total score is calculated using the weighted values, the first score, and the second score. Finally, the total score of the corresponding feature is corrected based on the feature overlap between the pre-selected features and the optimal feature subset, and the corrected total score is used as the corresponding comprehensive score.

[0051] In one specific embodiment of this application, during the process of configuring a score value for each evaluation result according to its importance, without considering the source but only the importance represented by the evaluation result itself, a score value of 1 point, 0.5 points, and 0 points are respectively configured for important, unimportant, and uncertain whether it is important.

[0052] In one specific embodiment of this application, in the process of determining the reliability of each extracted second evaluation result based on the professional field of the expert who made each extracted second evaluation result, if the professional field of the expert who made the second evaluation result is chemical engineering, the reliability of the extracted second evaluation result is determined to be the highest; if the professional field of the expert who made the second evaluation result is emergency response, the reliability of the extracted second evaluation result is determined to be medium; if the professional field of the expert who made the second evaluation result is neither chemical engineering nor emergency response, the reliability of the extracted second evaluation result is determined to be the lowest.

[0053] In one specific embodiment of this application, during the process of assigning weights to the corresponding second evaluation results, without considering the importance represented by the evaluation results themselves but only considering the source, according to the principle that the larger the weight value, the more important the second evaluation result, the weight of the second evaluation result with the highest reliability is assigned to 2; the weight of the second evaluation result with medium reliability is assigned to 1.5; and the weight of the second evaluation result with the lowest reliability is assigned to 1.

[0054] In this embodiment, without considering the evaluation result itself but only the data source, the weight is assigned as follows: the larger the weight, the more important the source of the second evaluation result. The weights are 2 for the highest reliability, 1.5 for the medium reliability, and 1 for the lowest reliability. That is, if the second evaluation result has the highest reliability, the weight of the second evaluation result is assigned to 2; if the second evaluation result has the medium reliability, the weight of the second evaluation result is assigned to 1.5; and if the second evaluation result has the lowest reliability, the weight of the second evaluation result is assigned to 1.

[0055] In this embodiment of the application, the total score is calculated using the following expression:

[0056]

[0057] Where f represents the total score, f d b1 and b2 represent coefficients, m represents the total number of first and second evaluation results, n represents the number of evaluation results that are the same as the actual evaluation result, and f represents the first rating value. j denoted by , w represents the weight, N represents the total number of first evaluation results, and x represents the sequence number of the first evaluation result.

[0058] In the embodiments of this application, The score representing the comprehensive importance of historical period characteristics, as determined by experts with experience in evaluating the influencing factors of chemical accidents, is a weighted average of the second score values ​​corresponding to several second evaluation results for the respective characteristics. In other words, this embodiment essentially assigns a first weight to each first evaluation result. And assigning a second weight to historical evaluation results that represent the overall importance of the aforementioned single characteristics of a historical period to chemical accidents. The calculation of the total score is implemented, where the first weight (less than 1 / 2) is always less than the second weight (greater than 1 / 2), and the total score should range from 0 to 1.

[0059] In one specific embodiment of this application, b1 and b2 are constants, with b1 less than or equal to b2. The specific values ​​are determined based on the richness of experimental data and expert experience, and are used to distinguish the credibility of actual evaluation results from historical evaluation results. Since the first weight is obtained based on a scoring method that only considers the importance represented by the evaluation result itself without considering the source, while the second weight is obtained based on a scoring method that considers both the importance represented by the evaluation result itself and the source, the first evaluation result has lower credibility than the second evaluation result. Therefore, the preferred value for b1 is 9, and the preferred value for b2 is 8.

[0060] In the process of obtaining the comprehensive score, the pre-selected features and the optimal feature subset are compared, and different score values ​​are assigned to the overlapping and non-overlapping features. Then, the assigned score values ​​are used to correct the corresponding total score to obtain the corresponding comprehensive score.

[0061] Specifically, this embodiment further embeds expert knowledge into the multi-layer feature selection process, comparing pre-selected features and the optimal feature subset to obtain overlapping and non-overlapping features, and then assigning different score values ​​to the overlapping and non-overlapping features respectively. Finally, by summing each assigned score value with the total score value of the corresponding pre-selected feature, the total score values ​​are corrected, and the obtained corrected total score values ​​are the comprehensive score values ​​of the corresponding features. Finally, by comparing the preset score threshold used to select the best feature from the pre-selected features with each comprehensive score value, the best feature adapted to the current period's chemical accident is selected from the pre-selected features. Therefore, this invention effectively ensures the comprehensiveness and reliability of the selected best features, avoiding the omission of key features.

[0062] In one specific embodiment of this application, a score value of 1 is assigned to each overlapping feature, and a score value of 0 is assigned to each non-overlapping feature.

[0063] In one specific embodiment of this application, the preset scoring threshold is preferably 0.5.

[0064] Example 2

[0065] This invention also provides a computer-readable storage medium storing at least one instruction that is loaded and executed by a processor to perform the feature selection operation for screening influencing factors of chemical accidents as performed in the method of the above embodiments. For example, the computer-readable storage medium may be ROM (Read Only Memory), RAM (Random Access Memory), CD-ROM (Compact Disc-Read Only Memory), magnetic tape, floppy disk, and optical data storage device, etc.

[0066] Example 3

[0067] Based on the feature selection method for screening influencing factors of chemical accidents described in Embodiment 1 above, this embodiment of the invention also provides a feature selection system for screening influencing factors of chemical accidents (hereinafter referred to as "feature selection system").

[0068] Figure 2 This is a block diagram of a feature selection system for screening influencing factors of chemical accidents, as described in an embodiment of this application. Figure 2 As shown, the feature selection system in this embodiment of the invention includes: a feature pre-selection module 21, a feature subset optimization module 22, and an optimal feature selection module 23. Specifically, the feature pre-selection module 21 is implemented according to the method described in step S110 above, configured to determine all features that cause chemical accidents based on historical chemical data of the target chemical plant area, and select multiple features that meet the preset importance requirements for chemical accidents as pre-selected features for chemical accidents in the current period; the feature subset optimization module 22 is implemented according to the method described in step S120 above, configured to input the pre-selected features into a preset feature subset optimization model to obtain the optimal feature subset; the optimal feature selection module 23 is implemented according to the method described in step S130 above, configured to obtain several first evaluation results representing the importance of features in the current period and several second evaluation results representing the importance of features in historical periods for each feature in the pre-selected features, and evaluate the reliability of each evaluation result according to the source of each evaluation result, and then combine the optimal feature subset to generate a comprehensive score value representing the actual importance of each feature to the chemical accident in the current period, thereby obtaining the optimal feature that is suitable for the chemical accident in the current period.

[0069] This invention discloses a feature selection method and system for screening influencing factors of chemical accidents. The method first obtains all features that cause chemical accidents and then selects multiple features that meet preset importance requirements for the chemical accident as pre-selected features for the current period. Next, a preset feature subset optimization model is constructed to obtain the optimal feature subset for the current period from the pre-selected features. Finally, for each feature in the pre-selected features, evaluation results based on expert knowledge are obtained, representing the importance of features in the current period and the importance of features in historical periods, respectively. The reliability of each evaluation result is evaluated according to its source. Then, a total score representing the overall importance of each feature to the current period's chemical accident is generated, and the total score is corrected based on the optimal feature subset to generate a comprehensive score representing the actual importance of each feature to the current period's chemical accident, thereby obtaining the optimal features suitable for the current period's chemical accident. This invention comprehensively considers expert experience and data patterns, taking into account the advantages of multi-layered feature selection and expert knowledge. It eliminates the risk of key features being deleted and avoids time-consuming model hyperparameter trial-and-error processes. The final selected features are a small-scale and interpretable feature set highly correlated with the target attribute (i.e., the occurrence of chemical accidents). This achieves the scientific selection of features that cause chemical accidents, providing technical support for accurate and efficient prediction of accident development. This invention is applicable to accident emergency data analysis in the integrated emergency response platform of refining and chemical enterprises, providing data support for subsequent accident emergency trigger identification.

[0070] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0071] Of course, the present invention may have other various embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art can make various corresponding changes and modifications according to the present invention, but these corresponding changes and modifications should all fall within the protection scope of the claims of the present invention.

[0072] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps as a single integrated circuit module. Thus, the present invention is not limited to any particular hardware and software combination.

[0073] While the embodiments disclosed in this invention are as described above, the content is merely for the purpose of facilitating understanding of the invention and is not intended to limit the invention. Any person skilled in the art to which this invention pertains may make any modifications and variations in form and detail of the implementation without departing from the spirit and scope disclosed herein; however, the scope of patent protection for this invention shall still be determined by the scope defined in the appended claims.

Claims

1. A feature selection method for screening influencing factors of chemical accidents, characterized in that, include: Based on historical chemical data of the target chemical plant area, all characteristics that could cause a chemical accident are identified, and multiple characteristics that meet the preset importance requirements for a chemical accident are selected as pre-selected characteristics for chemical accidents in the current period. The pre-selected features are input into a preset feature subset optimization model to obtain the optimal feature subset; For each of the pre-selected features, several first evaluation results representing the importance of features in the current period and several second evaluation results representing the importance of features in historical periods are obtained based on expert knowledge. The reliability of each evaluation result is evaluated according to its source. Then, the optimal feature subset is combined to generate a comprehensive score value representing the actual importance of each feature to the chemical accident in the current period, thereby obtaining the best feature that is appropriate for the chemical accident in the current period.

2. The feature selection method according to claim 1, characterized in that, The step of selecting multiple features that meet preset importance requirements for chemical accidents as pre-selected features for chemical accidents in the current period includes: Sparsity, correlation, and redundancy are evaluated on all the features respectively. Multiple features that simultaneously meet the preset sparsity requirements, preset correlation requirements, and preset redundancy requirements are selected as the multiple features that meet the preset importance requirements for chemical accidents, thereby obtaining the pre-selected features of chemical accidents in the current period.

3. The feature selection method according to claim 1 or 2, characterized in that, The preset feature subset selection model adopts an ML model.

4. The feature selection method according to any one of claims 1 to 3, characterized in that, The first evaluation result and the second evaluation result are classified as important, unimportant, or of uncertain importance. The step of generating a comprehensive score representing the actual importance of each feature to the current period's chemical accident includes: Assign a score to each evaluation result according to its importance; For each of the pre-selected features, the score value of each first evaluation result is recorded as the first score value; For each of the pre-selected features, extract the second evaluation results made by experts with experience in evaluating the influencing factors of chemical accidents from among the several second evaluation results. Record the score value of each extracted second evaluation result as the second score value. Then, determine the reliability of each extracted second evaluation result according to the professional field of the expert who made each extracted second evaluation result, and assign a weight to the corresponding second evaluation result according to the reliability value. Based on this, combine the first score value and the second score value to calculate the total score value representing the comprehensive importance of each feature to the chemical accident in the current period. The total score is corrected based on the feature overlap between the pre-selected features and the optimal feature subset to obtain the comprehensive score.

5. The feature selection method according to claim 4, characterized in that, The total score is calculated using the following expression: Where f represents the total score, f d b1 and b2 represent coefficients, m represents the total number of first and second evaluation results, n represents the number of evaluation results that are the same as the actual evaluation result, and f represents the first rating value. j denoted by , w represents the weight, N represents the total number of first evaluation results, and x represents the sequence number of the first evaluation result.

6. The feature selection method according to claim 4 or 5, characterized in that, The process of determining the reliability of each extracted second evaluation result based on the professional field of the expert who made each extracted second evaluation result includes: If the professional field is chemical engineering, then the reliability of the extracted second evaluation result is determined to be the highest; If the field of expertise is emergency response, the reliability of the extracted second evaluation result is determined to be moderate. If the field of expertise is neither chemical engineering nor emergency response, then the reliability of the extracted second evaluation result is considered to be the lowest.

7. The feature selection method according to claim 6, characterized in that, The process of assigning weights to the corresponding second evaluation results includes: The second evaluation result, which has the highest reliability, is assigned a weight of 2. The second evaluation result with moderate reliability is assigned a weight of 1.5; The second evaluation result with the lowest reliability is assigned a weight of 1.

8. The feature selection method according to any one of claims 4 to 7, characterized in that, The process of assigning score values ​​to each evaluation result according to its importance includes: For evaluation results that are important, unimportant, and uncertain in terms of importance, a score of 1, 0.5, and 0 points will be assigned respectively.

9. The feature selection method according to any one of claims 4 to 8, characterized in that, The process of obtaining the comprehensive score includes: The pre-selected features and the optimal feature subset are compared, and different score values ​​are assigned to the overlapping and non-overlapping features. The assigned score values ​​are then used to correct the corresponding total score value to obtain the corresponding comprehensive score value.

10. The feature selection method according to claim 9, characterized in that, Each overlapping feature was assigned a score of 1. Each non-overlapping feature was assigned a score of 0.

11. A computer-readable storage medium, characterized in that, It includes a series of instructions for performing the feature selection method steps for screening influencing factors of chemical accidents as described in any one of claims 1 to 10.

12. A feature selection system for screening influencing factors of chemical accidents, characterized in that, The feature selection system includes the following modules: The feature pre-selection module is used to determine all the features that cause chemical accidents based on historical chemical data of the target chemical plant area, and to select multiple features that meet the preset importance requirements for chemical accidents as pre-selected features for chemical accidents in the current period. The feature subset selection module is used to input the pre-selected features into a preset feature subset selection model to obtain the optimal feature subset; The optimal feature selection module is used to obtain, for each feature in the pre-selected features, several first evaluation results representing the importance of features in the current time period based on expert knowledge, and several second evaluation results representing the importance of features in historical time periods. It evaluates the reliability of each evaluation result according to its source, and then combines the optimal feature subset to generate a comprehensive score value representing the actual importance of each feature to the chemical accident in the current time period. Based on this, the optimal feature that is appropriate for the chemical accident in the current time period is obtained.

Images