Assessment method and system for safety risks of confined space operations

Abstract

The present invention relates to the fields of safety engineering and artificial intelligence technology, and in particular to a method and system for assessing safety risks of operations in confined spaces. The present invention uses machine learning technology and data analysis methods to effectively improve the core deficiencies of the prior art. The real-time monitoring and assessment capabilities enable the present invention to respond to environmental changes in a timely manner; the high adaptability enables the model to identify and adapt to new risk factors; the accuracy and objectivity of the assessment are improved through a data-driven approach; the iterative optimization mechanism ensures that the model can continue to evolve and enhances the effectiveness of risk management. The present invention provides a more efficient, accurate and reliable safety risk assessment method and system for operations in confined spaces, provides solid technical support for ensuring operational safety, and marks an important step in the development of the field of safety risk assessment towards intelligence and data.

Description

Technical Field

[0001] The present invention relates to the fields of safety engineering and artificial intelligence technology, and in particular to a method and system for assessing safety risks of confined space operations. Background Art

[0002] In confined space operations, safety risk assessment is a key step to ensure the safety of operators and prevent accidents. Confined spaces, such as storage tanks, shafts, narrow pipes, etc., are often accompanied by high safety risks due to their special spatial structure and operating environment, including but not limited to flammable gas leakage, lack of oxygen, exposure to hazardous substances, etc. If these risks are not identified and controlled in a timely manner, they may lead to serious safety accidents such as fire, explosion, poisoning, suffocation, etc., causing casualties and economic losses. Therefore, conducting accurate safety risk assessments and timely discovering and preventing potential risks are crucial to ensuring the safety of confined space operations.

[0003] Safety risk assessment faces multiple problems in practice. First, the identification of risk sources and the assessment of risk levels often rely on the experience and subjective judgment of professionals, which may lead to inconsistency and inaccuracy in the assessment results. Second, the confined space working environment is complex and changeable, involving many risk factors. Traditional risk assessment methods are difficult to capture all potential risk points in real time. In addition, as the working environment and conditions continue to change, the old risk assessment model may not be able to adapt to the new risk situation and needs to be constantly updated and optimized.

[0004] Existing safety risk assessment technologies (Chinese invention patent, publication number: CN111325455B) mainly rely on static checklists, qualitative risk matrix analysis, and statistical methods based on historical data. These methods may be effective enough in dealing with simple, highly repetitive risk assessment scenarios, but they have obvious deficiencies and defects when faced with complex confined space working environments and their dynamically changing characteristics:

[0005] Lack of real-time performance: It is difficult to monitor and evaluate risk changes in the operating environment in real time;

[0006] Poor adaptability: Difficulty in quickly adapting to changes in the working environment and emerging risk factors;

[0007] Low accuracy: Static evaluation methods and subjective judgments can easily lead to deviations in evaluation results. Summary of the invention

[0008] In view of the many problems existing in the above-mentioned prior art, the present invention provides a method and system for assessing the safety risks of confined space operations. The present invention utilizes environmental parameter (such as gas concentration, temperature, air pressure, etc.) data continuously collected from the confined space working environment, performs real-time monitoring and analysis through advanced data processing technology, and then combines a machine learning model to predict and evaluate potential safety risks. The machine learning model learns to identify risk patterns and factors by training historical accident data and real-time environmental data, thereby realizing dynamic identification and assessment of potential risks in the working environment. The present invention can provide real-time and accurate risk assessment results to help prevent possible safety accidents. It overcomes the shortcomings of traditional risk assessment methods in real-time, adaptability and accuracy, and ensures that the safety management of confined space operations is more scientific, systematic and effective. By real-time monitoring of environmental changes and dynamically adjusting the risk assessment model, the present invention can respond to changes in the environment and working conditions in a timely manner, effectively reduce the safety risks of confined space operations, and improve the safety of operators.

[0009] The method for assessing the safety risks of confined space operations includes the following steps:

[0010] By deploying a sensor network, the environmental data inside the limited space is collected in real time and pre-processed;

[0011] Extract key features from pre-processed environmental data, obtain key features related to risk events through preliminary screening using the Pearson correlation coefficient, and obtain key risk features that have the greatest impact on risk assessment through in-depth screening;

[0012] After selecting the type of machine learning algorithm to be applied based on the key risk characteristics, the model is trained and verified to obtain a trained risk assessment model; abnormality detection and external factor elimination are performed on the air pressure data, and a comprehensive analysis is performed with the key risk characteristics to identify potential risks not covered by the risk assessment model and obtain potential risk identification results;

[0013] The risk assessment model is applied to predict the risk of the current limited space to obtain the risk prediction result, and the risk assessment result is constituted by the risk assessment result together with the potential risk identification result; based on the difference between the actual feedback and the risk prediction result, the risk assessment model is iteratively optimized.

[0014] Preferably, the preprocessing includes: performing data cleaning and normalization processing on the environmental data;

[0015] The environmental data include: gas concentration, temperature, humidity and air pressure.

[0016] Preferably, by using data analysis technology, the changing characteristics in the environmental data are identified and key features are extracted;

[0017] Among them, the key characteristics include: gas concentration characteristics, temperature characteristics, humidity characteristics and air pressure characteristics.

[0018] Preferably, a statistical analysis method is applied to calculate the Pearson correlation coefficient between each of the key features and the known risk events, and based on a preset correlation threshold, the key features corresponding to which the absolute value of the Pearson correlation coefficient exceeds the correlation threshold are selected to complete the preliminary screening, wherein the Pearson correlation coefficient calculation expression is:

[0019] ,in, is the Pearson correlation coefficient, which is used to measure the and The degree of linear correlation between is the number of observations; and The variables are and No. observations; and The variables are and The average value of

[0020] The value of Pearson correlation coefficient is :

[0021] when =1, indicating and There is a completely positive correlation between them;

[0022] when =-1, indicating and There is a completely negative correlation between them;

[0023] when =1, indicating and There is no linear correlation between them.

[0024] Preferably, the in-depth screening includes:

[0025] Partial correlation analysis: select one key feature as the analysis object and other key features as control variables, and calculate the partial correlation coefficient between the analysis object and the risk event after controlling the influence of other control variables;

[0026] Multiple regression analysis: Take the risk event as the dependent variable and all the key risk features obtained through preliminary screening as independent variables to build a multiple regression model. Use historical data to complete the training of the multiple regression model and analyze the influence coefficient and influence value of each key feature in the multiple regression model.

[0027] Based on the results of partial correlation analysis and multiple regression analysis, the key risk features that have the greatest impact on risk assessment are screened out.

[0028] Preferably, the historical accident records and environmental parameters corresponding to the key risk features are used as training samples. When constructing the training samples, the historical accident records are used as labels, and the environmental parameters at the corresponding time are used as input features; the training samples are divided into a training set and a validation set, which are used for training and validation of the selected model respectively;

[0029] Train the selected machine learning model through the training set, adjust the model parameters and complete the construction of the risk assessment model;

[0030] The performance of the risk assessment model is evaluated using a validation set, and the model parameters and structure are adjusted based on the evaluation results. After the performance of the risk assessment model is verified, it is used to take the preprocessed environmental data and key risk features as input to output risk prediction results, which include: possibility parameters and severity parameters of risk events.

[0031] Preferably, the process of obtaining the potential risk identification result includes:

[0032] Based on the air pressure data collected by the continuous monitoring sensor, the abnormal changes that do not conform to the normal air pressure fluctuation pattern are identified through a preset air pressure change threshold, and the air pressure abnormality data is generated, wherein the air pressure abnormality data includes all air pressure change information identified as abnormal;

[0033] Collect the external environment data of the limited space at the same time as the abnormal air pressure data, and compare the external environment data with the abnormal air pressure data at the same time. If the correlation coefficient between the external environment data and the abnormal air pressure exceeds the preset correlation threshold, it is determined that the abnormal air pressure is caused by external factors; if not, conduct a comprehensive analysis of the abnormal air pressure data and the determined key risk characteristics to determine whether the abnormal air pressure changes at the same time as the key risk characteristics. If so, ignore the abnormal air pressure data in this time period; if not, it is determined that there is a potential risk that is not covered by the risk assessment model, and the judgment result is used as the potential risk identification result.

[0034] Preferably, the iterative optimization of the risk assessment model includes:

[0035] Verify the predictions of the risk assessment model and collect actual feedback related to the predictions;

[0036] Compare the differences between risk prediction results and actual risk events to evaluate the performance of the risk assessment model;

[0037] Based on the assessment results, the risk assessment model is iteratively optimized.

[0038] A system for implementing the method for assessing safety risks of confined space operations, comprising:

[0039] A sensor network, the sensor network includes: different types of sensors such as gas concentration sensors, temperature sensors, humidity sensors and air pressure sensors, the sensor network is used to monitor the environmental conditions in a limited space in real time and collect raw environmental data;

[0040] Data processing and storage hardware, including: a data collector, which is connected to the sensor network and is responsible for collecting sensor data; a central processing unit, which is responsible for further data preprocessing, feature extraction, and computing tasks of running machine learning models; a data storage system, which is used to store raw data, processed data, and risk assessment results;

[0041] A risk assessment module, which is deployed on a central server or cloud platform and is responsible for executing a risk assessment model, analyzing the relationship between environmental data and risks, and generating risk assessment results;

[0042] A model iteration optimization module, wherein the model iteration optimization module includes an input device for collecting operator feedback and accident reports, and is used to perform iterative optimization and training of the risk assessment model.

[0043] Compared with the prior art, the advantages and beneficial effects of the present invention are:

[0044] (1) Real-time monitoring and assessment: The present invention continuously collects environmental parameters (such as gas concentration, temperature, air pressure, etc.) in a limited space to monitor changes in the working environment in real time and conducts instant risk assessment based on a machine learning model;

[0045] (2) High adaptability: By learning historical data and real-time data, the model can adapt to changes in the operating environment and conditions and identify new risk factors in a timely manner;

[0046] (3) Accuracy and objectivity: The present invention reduces the impact of human error and subjective judgment through a data-driven approach, thereby improving the accuracy and objectivity of risk assessment;

[0047] (4) Iterative optimization capability: The present invention supports continuous iterative optimization of the risk assessment model based on actual feedback, ensuring that the model is always kept up to date and effectively responds to emerging risk challenges. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of the process of the present invention;

[0049] Figure 2It is a data flow diagram of the present invention;

[0050] Figure 3 It is a structural block diagram of the system of the present invention. DETAILED DESCRIPTION

[0051] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is obvious that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0052] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0053] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0054] In the case of using expressions such as "at least one of A, B, and C, etc.", it should generally be interpreted in accordance with the meaning of the expression generally understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.). In the case of using expressions such as "at least one of A, B, or C, etc.", it should generally be interpreted in accordance with the meaning of the expression generally understood by those skilled in the art (for example, "a system having at least one of A, B, or C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0055] Some block diagrams and / or flow charts are shown in the accompanying drawings. It should be understood that some boxes or combinations thereof in the block diagrams and / or flow charts can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that these instructions can create a device for implementing the functions / operations described in these block diagrams and / or flow charts when executed by the processor. The technology of the present disclosure can be implemented in the form of hardware and / or software (including firmware, microcode, etc.). In addition, the technology of the present disclosure can take the form of a computer program product on a computer-readable storage medium storing instructions, which can be used by an instruction execution system or in combination with an instruction execution system.

[0056] like Figure 1-Figure 2 As shown in the figure, the method for assessing the safety risk of confined space operations includes the following steps:

[0057] By deploying a sensor network, the environmental data inside the limited space is collected in real time and pre-processed;

[0058] Confined space work involves work performed in enclosed or partially enclosed spaces that may have limited entry / exit and are not designed for continuous occupancy. Such work may be subject to a variety of risks, including but not limited to the accumulation of flammable / toxic gases, insufficient or excessive oxygen levels, physical hazards, etc.

[0059] Sensor networks play a vital role in the safety risk assessment of confined space operations. These sensors are designed to monitor and record various environmental parameters in confined spaces, such as gas concentration, temperature, humidity, and air pressure. These parameters are the basic data for risk assessment and are directly related to the basis for judging safe operations. The sensor network sends the collected data to the central processing system in real time through wireless or wired means to achieve real-time monitoring of the environmental conditions of confined spaces.

[0060] The collected environmental data usually contains noise, missing values ​​or outliers. Directly using these raw data for risk assessment may affect the accuracy and reliability of the assessment results. Therefore, it is necessary to preprocess the data, including data cleaning (removing noise, repairing or deleting outliers and missing values), data normalization or standardization (scaling the data to a uniform range or distribution) to ensure data quality and improve the performance of the risk assessment model.

[0061] By deploying sensor networks to collect environmental data in confined spaces in real time and preprocessing it, a real-time, accurate and reliable data foundation can be provided for risk assessment. This enables the risk assessment model to make accurate predictions based on current environmental conditions, identify potential risks in a timely manner, and provide a scientific basis for taking preventive measures and emergency responses, thereby significantly improving the safety of confined space operations.

[0062] For example, in a chemical storage warehouse, sensors are used to monitor the concentration of combustible gas, temperature and humidity in the warehouse. If the sensor detects that the concentration of combustible gas exceeds the safety threshold, the information will be transmitted to the central processing system in real time, and after the accuracy of the signal is confirmed through pre-processing, the alarm system will be triggered immediately, and safety measures such as emergency ventilation or evacuation of personnel will be taken to avoid potential fire or explosion accidents.

[0063] Preferably, the preprocessing includes: performing data cleaning and normalization processing on the environmental data;

[0064] The environmental data include: gas concentration, temperature, humidity and air pressure.

[0065] Data cleaning is used to remove noise and outliers from collected environmental data to ensure the accuracy of subsequent analysis. This is done by identifying and processing missing values, erroneous data (such as readings that are not within the normal range), duplicate records, etc. For example, if the sensor reading of gas concentration is far below the normal range, it may be erroneous data caused by sensor failure and should be removed from the analysis data set. Removing abnormal and erroneous data through data cleaning can reduce the interference of noise on the risk assessment model and improve the accuracy of the model.

[0066] Normalization is used to convert data of different scales and dimensions to a unified scale or range for comprehensive analysis. Common methods include minimum-maximum normalization (adjusting the data range to [0, 1] or [-1, 1]) and Z-score normalization (adjusting the data mean to 0 and the standard deviation to 1). For example, temperature and gas concentration have different dimensions and ranges, and after normalization, they can be compared and calculated on the same scale. Normalized data has a unified scale, which allows faster data processing and analysis. In particular, when applying machine learning algorithms, normalized data helps the algorithm converge faster. Normalization allows different environmental parameters to be compared and analyzed under the same standards, which helps to discover the relationships and patterns between different parameters.

[0067] The "Z" in Z-score normalization (also known as standardization) refers to the standard score, which represents the deviation of a value from the mean of the entire data set and is used to measure the distance between a single data point and the group mean. The specific calculation method is to subtract the mean of the entire data set from each observation and then divide it by the standard deviation of the data set. The calculation formula is as follows:

[0068] ,in, is a single observation; is the mean value of the data set; is the standard deviation of the data set; This is the calculated Z score.

[0069] Through this transformation, the new value (Z score) obtained represents the relative distance of the original data from the group mean in units of standard deviation. This process does not change the distribution of the data, but makes the transformed data set have the characteristics of a mean of 0 and a standard deviation of 1. Z-score normalization allows data to be effectively compared between different scales or dimensions, and is particularly suitable for machine learning algorithms involving multiple parameters or features, which can accelerate the learning process and improve algorithm performance.

[0070] In one embodiment, in a confined space operation scenario, the risk of gas leakage needs to be assessed. Environmental data shows that the gas concentration at a certain moment suddenly increases, while the temperature, humidity, and air pressure do not change significantly. After eliminating abnormal conditions such as sensor failure through data cleaning, these parameters are converted to the same scale for analysis through normalization processing, and it is found that the increase in gas concentration is a significant abnormal pattern, which may indicate the risk of gas leakage.

[0071] Extract key features from pre-processed environmental data, obtain key features related to risk events through preliminary screening using the Pearson correlation coefficient, and obtain key risk features that have the greatest impact on risk assessment through in-depth screening;

[0072] Feature extraction is the process of identifying information useful for risk assessment from a large amount of environmental data. These features can reflect the safety status inside a confined space, such as gas concentration, temperature, humidity, and air pressure. Using data analysis and machine learning techniques, we can identify trends, fluctuation patterns, or outliers from environmental data, which may be important bases for risk assessment.

[0073] In the preliminary screening, the Pearson correlation coefficient is a method to measure the degree of linear correlation between two variables. Its value ranges from -1 (complete negative correlation) to 1 (complete positive correlation), and 0 means no linear correlation. By calculating the Pearson correlation coefficient between each feature and known risk events (such as historical accident records), the features that have a significant correlation with risk events are preliminarily screened out.

[0074] Further identification of the features that contribute most to risk assessment while controlling for other factors. In-depth screening includes:

[0075] Partial correlation analysis: Calculate the correlation between features and risk events after controlling for the influence of other features. Multiple regression analysis: Analyze the impact of each feature on risk events when considering multiple features, and identify features with significant influence.

[0076] Through preliminary screening and in-depth screening, the key risk features that are most critical to the safety risk assessment of confined space operations can be accurately identified. Such features are not only statistically significantly correlated with risk events, but also play a decisive role in the actual risk prediction model.

[0077] In one embodiment, in a confined space operation at a chemical plant, historical data showed that in past accidents, the exceeding of a specific gas concentration threshold was a key factor leading to safety accidents. By calculating the Pearson correlation coefficient between gas concentration and historical accident records, it was found that there was a strong positive correlation between the two. Further partial correlation analysis and multiple regression analysis also confirmed that after controlling other factors such as temperature, humidity, and air pressure, abnormal gas concentration was still an important predictor in risk assessment. Therefore, gas concentration was identified as one of the most influential key risk features for risk assessment.

[0078] Preferably, by using data analysis technology, the changing characteristics in the environmental data are identified and key features are extracted;

[0079] Among them, the key characteristics include: gas concentration characteristics, temperature characteristics, humidity characteristics and air pressure characteristics.

[0080] In the context of confined space operation safety risk assessment, it is crucial to identify the changing characteristics in environmental data and extract key features through data analysis technology. This process involves the use of advanced data processing and analysis methods to ensure that the most critical information for risk assessment can be effectively identified.

[0081] The application of the data analysis technology in the present invention includes:

[0082] Time series analysis: Given that environmental data (such as gas concentrations, temperature, humidity, and air pressure) often change over time, time series analysis techniques can be used to identify trends, seasonal changes, and anomalies in the data. For example, using an autoregressive moving average (ARMA) model can help predict future gas concentration levels.

[0083] Anomaly detection: Apply statistical methods (such as box plots, Z scores, IQR, etc.) to perform anomaly detection to identify abnormal changes in the data, which may indicate potential safety risks.

[0084] Association rule learning: Exploring potential associations between different environmental parameters, such as whether an increase in temperature is associated with an increase in the concentration of a specific gas, can help identify compound conditions for risk occurrence.

[0085] Perform feature engineering on the original environmental parameters to extract meaningful features, such as calculating the rate of change of gas concentration, the daily change of temperature, etc. These derived features may reflect the risk status more directly than the original data. In order to compare different features on the same scale, data normalization is performed to ensure that the model can fairly evaluate the influence of each feature.

[0086] By accurately extracting key risk features, risk assessment models can more accurately predict potential safety risks, thus providing a reliable basis for preventive measures. Effective feature extraction and analysis can enhance the model's ability to predict future risk events, especially when dealing with complex environmental conditions.

[0087] In one example, in a confined space operation at a chemical plant, environmental data within the space is monitored in real time through sensors. Through time series analysis, it is found that whenever the temperature exceeds a certain threshold, the concentration of a certain gas increases significantly within a few hours. The recognition of this pattern provides the security team with the ability to predict potential gas leaks. Furthermore, a sudden drop in gas pressure identified by anomaly detection technology may indicate equipment failure, which is another potential risk that requires immediate response.

[0088] Associating the extracted key features with specific risk events is the core of risk assessment, which requires not only technical analysis of the data but also a deep understanding of the relationship between environmental parameters and potential risks.

[0089] The present invention uses statistical analysis and historical data to directly associate changes in specific characteristics with the occurrence of risk events. This analysis is usually based on past event records and environmental parameter data at the corresponding time.

[0090] In a specific embodiment, in the gas concentration and leakage events, by analyzing historical data, it can be found that the event of a sudden increase in gas concentration is usually closely related to chemical leakage. A gas concentration threshold is set, and when the current concentration exceeds this threshold, the model identifies it as a potential leakage risk.

[0091] In another embodiment, in the temperature and humidity and equipment failure events, statistical analysis shows that under a specific temperature and humidity combination, the failure rate of certain equipment increases significantly. This may be because extreme temperature and humidity conditions affect the normal operation of the equipment. By setting a specific temperature and humidity combination threshold, the possible failure risk of the equipment can be warned.

[0092] In the present invention, the complex relationship between environmental parameter changes and risk events is identified by using machine learning algorithms. This method not only relies on a single parameter, but can also reveal the complex patterns of interaction between multiple parameters. Specifically:

[0093] A machine learning model is built with gas concentration, temperature, humidity, and air pressure as input features and the occurrence of risk events as output. During the training process, the model learns that not only a single feature, but also a specific combination of features (such as temperature increase and humidity decrease) is highly correlated with the occurrence of equipment failure or safety accidents.

[0094] Based on a deep understanding of the correlation between key features and risk events, specific monitoring thresholds are set. When the monitored data exceeds these thresholds, risk warnings are issued.

[0095] In one embodiment, in addition to the static threshold value, the threshold value of the gas concentration can also be dynamically adjusted according to other environmental parameters. For example, in the case of high temperature, the safety threshold value of the gas concentration needs to be lowered because high temperature may accelerate chemical reactions.

[0096] Through the above-mentioned method of the present invention, the transformation from data to insight can be achieved, and the changes in environmental parameters can be effectively associated with potential risk events, providing strong support for the safety risk assessment of confined space operations. The successful implementation of this method requires interdisciplinary knowledge integration, including data analysis, domain expertise, and machine learning technology to ensure the accuracy and real-time nature of risk assessment.

[0097] Preferably, a statistical analysis method is applied to calculate the Pearson correlation coefficient between each of the key features and the known risk events, and based on a preset correlation threshold, the key features corresponding to which the absolute value of the Pearson correlation coefficient exceeds the correlation threshold are selected to complete the preliminary screening, wherein the Pearson correlation coefficient calculation expression is:

[0098] ,in, is the Pearson correlation coefficient, which is used to measure the and The degree of linear correlation between is the number of observations; and The variables are and No. observations; and The variables are and The average value of

[0099] The value of Pearson correlation coefficient is :

[0100] when =1, indicating and There is a completely positive correlation between them;

[0101] when =-1, indicating and There is a completely negative correlation between them;

[0102] when =1, indicating and There is no linear correlation between them.

[0103] By calculating the Pearson correlation coefficient between key features (such as gas concentration, temperature, humidity, and air pressure) and known risk events (such as accident records), features with significant linear correlation with risk events can be preliminarily screened out. A correlation threshold (such as |r| > 0.5) is set, and only those key features whose absolute value of the correlation coefficient exceeds this threshold are retained to complete the preliminary screening.

[0104] Preliminary screening can effectively identify environmental parameters that are strongly correlated with safety risk events, providing a scientific basis for subsequent risk assessment and preventive measures; by focusing on features that are strongly correlated with risk events, the predictive accuracy and efficiency of the risk assessment model can be improved.

[0105] In one embodiment, in a confined space operation, historical data analysis shows that when the gas concentration exceeds a certain threshold, the probability of a safety accident increases significantly. By calculating the Pearson correlation coefficient between the gas concentration characteristics and the occurrence of safety accidents, it is found that r=0.8, indicating a strong positive correlation between the two. According to the preset correlation threshold (e.g., |r|> 0.5), the gas concentration characteristics are preliminarily screened as key characteristics that are correlated with risk events. Further analysis may also consider other environmental parameters such as temperature, humidity, and air pressure to comprehensively assess the risks and formulate corresponding preventive measures.

[0106] The statistical analysis method, especially the application of the Pearson correlation coefficient in the present invention, provides an objective and scientific analysis method for the safety risk assessment of confined space operations. By identifying and screening out key features that are significantly correlated with risk events, it is possible to more accurately guide the formulation and implementation of safety management measures, thereby effectively preventing and reducing the occurrence of safety accidents.

[0107] It is an important consideration in risk assessment to deal with key features with Pearson correlation coefficients below a preset threshold, that is, those that show weak or no linear correlation with known risk events. Although these features are not directly related to risk events when viewed alone, they may reveal important risk signals under certain conditions or when combined with other features.

[0108] Some features may not be significant individually, but when combined with other features, their combination may be highly correlated with risk events. This phenomenon is called combination effect or interaction.

[0109] To this end, the present invention also provides a corresponding solution:

[0110] Exploring the combined effects of features by building multivariate statistical models, such as using logistic regression models, and including interaction terms in the model can help identify which combinations of features are important for risk prediction.

[0111] The Pearson correlation coefficient can only measure the strength of a linear relationship. For nonlinear relationships, other statistical methods are needed for analysis. For example: Spearman rank correlation coefficient: used to measure the correlation between the ranks (or orders) of two variables, suitable for the analysis of nonlinear relationships; Kendall rank correlation coefficient: another nonparametric correlation coefficient, suitable for evaluating the consistency or rank correlation of two variables, and can also be used for nonlinear relationship analysis.

[0112] Machine learning and deep learning techniques can capture complex patterns and relationships in data, including nonlinear relationships and higher-order interactions between features. Ensemble learning methods such as random forests and gradient boosting trees can not only handle nonlinear relationships, but also reveal which features (including those with weak linear correlations) contribute more to model predictions through feature importance analysis. Deep learning models, such as neural networks, can capture complex nonlinear relationships between features because they can learn complex functions through multi-layer representations.

[0113] For example, suppose that in the safety risk assessment of confined space operations, the temperature feature alone has a weak direct correlation with safety incidents (low Pearson correlation coefficient). However, through the above analysis, it may be found that when temperature is combined with a certain level of chemical concentration, its predictive ability for safety incidents is significantly improved. This finding emphasizes the importance of considering feature combinations and nonlinear relationships in risk assessment.

[0114] Through the application of these strategies and methods, the potential information of all features can be fully explored and utilized. Even those features that initially appear to have a weak or no linear correlation with risk events may, under certain conditions or when combined with other features, show important contributions to risk assessment.

[0115] Preferably, the in-depth screening includes:

[0116] Partial correlation analysis: select one key feature as the analysis object and other key features as control variables, and calculate the partial correlation coefficient between the analysis object and the risk event after controlling the influence of other control variables;

[0117] Partial correlation analysis is used to measure the degree of correlation between two variables after controlling for the influence of other variables. In risk assessment, it allows the relationship between a key feature and a risk event to be examined independently without being affected by other features. Select one key feature as the analysis object and other key features as control variables, and calculate the partial correlation coefficient between the analysis object and the risk event. This helps to identify features that are significantly correlated with risk events even under the combined effects of complex environmental factors.

[0118] Partial correlation analysis is a statistical method used to calculate the correlation between two variables after controlling for one or more variables. In risk assessment, it can help determine the direct relationship between a specific risk factor and a safety event when other conditions are fixed. The analysis steps include:

[0119] (1) Criteria for selecting control variables

[0120] First, perform a simple correlation analysis (such as Pearson correlation coefficient calculation) to identify all variables that may be associated with the target risk event. Use professional knowledge and existing research to determine which variables may be common causes, that is, variables that may affect other variables and the target risk event at the same time. Considering the integrity and reliability of the data, select variables with stable records in the entire data set as control variables.

[0121] (2) Calculation process

[0122] Make sure the data for the selected variables have been cleaned and normalized. Use a statistical software package (such as R, SPSS, or the SciPy and Pandas libraries in Python) to calculate partial correlation coefficients. Most software packages provide straightforward functions to calculate partial correlation coefficients between two variables after controlling for one or more variables.

[0123] (3) Interpretation of partial correlation coefficient results

[0124] Positive correlation: A partial correlation coefficient close to 1 indicates that there is a positive correlation between the two variables after controlling for the influence of other variables.

[0125] Negative correlation: A partial correlation coefficient close to -1 indicates a negative correlation.

[0126] No correlation: A partial correlation coefficient close to 0 indicates no significant linear correlation.

[0127] Significance test: A significance test (usually using p-value) is used to determine whether the partial correlation coefficient is statistically significant to determine whether the correlation after controlling the variables is reliable.

[0128] Multiple regression analysis: Take the risk event as the dependent variable and all the key risk features obtained through preliminary screening as independent variables to build a multiple regression model. Use historical data to complete the training of the multiple regression model and analyze the influence coefficient and influence value of each key feature in the multiple regression model.

[0129] Multiple regression analysis allows the impact of multiple independent variables (key risk features) on the dependent variable (risk event) to be considered simultaneously. By building a mathematical model, the contribution of each key feature to the probability of a risk event can be analyzed. A multiple regression model is constructed, with all the key risk features obtained from the initial screening as independent variables and risk events as dependent variables. After the model training is completed, the influence coefficient and significance of each feature can be analyzed to determine which features have the greatest impact on risk assessment.

[0130] In multiple regression analysis, each key feature corresponds to an influence coefficient (also called regression coefficient), which represents the expected change in the target variable (such as the probability of a risk event) when the feature changes by one unit. In addition, each coefficient has a significance value (usually a p-value) to determine whether the coefficient is statistically significant, that is, whether the feature has a substantial impact on the prediction of the target variable.

[0131] A positive coefficient indicates that the feature is positively correlated with the target variable, that is, when the feature value increases, the value of the target variable will also increase. A negative coefficient indicates that they are negatively correlated. The absolute value of the coefficient indicates the strength of the influence. The larger the absolute value, the stronger the influence.

[0132] The p-value is used to test the hypothesis that the feature has no effect on the target variable (null hypothesis). When the p-value is lower than the commonly used significance level (such as 0.05 or 0.01), the null hypothesis can be rejected and the feature is considered to have a significant effect on the target variable. A high p-value indicates that the feature has an insignificant effect on the target variable and may not be a good predictor.

[0133] Based on the results of partial correlation analysis and multiple regression analysis, the key risk features that have the greatest impact on risk assessment are screened out.

[0134] Multiple regression analysis can not only analyze the impact of a single feature on the target variable, but also analyze the combined effect between features by introducing interaction terms. The interaction term is the product of two or more features and is used to explore whether there is interaction between features, that is, whether the impact of one feature depends on the value of another feature.

[0135] The analysis of partial correlation analysis results and multiple regression analysis results includes:

[0136] (1) Introduce the product of two features into the model as a new feature. For example, if there are features A and B, introduce an interaction term A*B.

[0137] (2) After model training, interpret the model coefficients including the interaction term and the corresponding significance test results. If the interaction term coefficient is significant, it indicates that the combined effect of the two features has a significant impact on the target variable.

[0138] (3) By comparing the goodness of fit (such as R² value) and prediction accuracy of the model before and after the introduction of the interaction term, we can evaluate whether the interaction term actually improves the model performance.

[0139] In one embodiment, in the safety risk assessment, it is found that high temperature (feature A) and high humidity (feature B) have a certain impact on the risk of safety accidents, but when high temperature and high humidity occur at the same time (interaction term A*B), the increase in the risk of accidents exceeds the simple addition of their individual effects. In this case, by introducing and analyzing the interaction term, the safety risk under specific environmental conditions can be more accurately assessed.

[0140] By deeply analyzing the interaction terms in the multivariate regression model, the interactions of risk factors hidden beneath the surface can be revealed, providing a more detailed and comprehensive perspective for the safety risk assessment of confined space operations.

[0141] Through partial correlation analysis and multiple regression analysis, key features that have a significant impact on risk events, namely core risk factors, can be effectively identified. The key risk feature set after in-depth screening can provide more accurate input for the risk assessment model, thereby improving the model's prediction accuracy.

[0142] In one example, in a confined space operation at a chemical plant, initial screening found that gas concentration, temperature, humidity, and air pressure are potential key risk features. Through partial correlation analysis, it may be found that the relationship between gas concentration and the occurrence of safety accidents is more significant after controlling for humidity and air pressure. And multiple regression analysis may further reveal that small changes in temperature can significantly increase the risk of accidents at specific gas concentration levels. These two analytical methods together help identify that under specific conditions, gas concentration and temperature are the most influential key features for chemical plant safety risk assessment.

[0143] Through this in-depth screening process, the risk assessment model can more accurately predict and prevent potential safety accidents, providing chemical plants with strong data support to formulate effective safety management measures.

[0144] After selecting the type of machine learning algorithm to be applied based on the key risk characteristics, the model is trained and verified to obtain a trained risk assessment model; abnormality detection and external factor elimination are performed on the air pressure data, and a comprehensive analysis is performed with the key risk characteristics to identify potential risks not covered by the risk assessment model and obtain potential risk identification results;

[0145] Based on the nature of key risk features (such as continuity, classification, data distribution, etc.) and problem types (such as classification, regression), select appropriate machine learning algorithms. Common algorithms include decision trees, random forests, support vector machines (SVM), neural networks, etc.

[0146] The rationale for choosing a machine learning algorithm includes:

[0147] Continuous features (such as temperature, humidity) and categorical features (such as chemical species) have a direct impact on the choice of algorithm. For example, support vector machines (SVM) and neural networks can handle continuous data well, while decision trees and random forests can directly handle categorical data.

[0148] The distribution characteristics of the data (e.g., normal distribution, skewed distribution) will also affect the choice of algorithm. Some algorithms, such as Naive Bayes, are more sensitive to the distribution assumptions of the data, while tree-based models are less strict about the data distribution.

[0149] If the goal of risk assessment is to determine whether a specific security incident will occur (yes / no), then it is a classification problem. Decision trees, random forests, support vector machines, and neural networks are all commonly used algorithms to solve classification problems.

[0150] If the goal is to predict a continuous value such as the likelihood or severity of a security incident, the problem is considered a regression problem. Linear regression, support vector regression (SVR), neural networks, etc. are suitable for regression problems.

[0151] Choosing the algorithm that best matches the data characteristics and problem type can significantly improve the model's prediction accuracy. The right algorithm can learn patterns in the data more effectively and reduce prediction errors. Different algorithms vary greatly in training time and resource consumption. Choosing the appropriate algorithm based on key risk characteristics and data volume can improve the efficiency of model training and prediction while ensuring accuracy. In security risk assessment, the interpretability of the model is equally important. For example, the decision paths provided by decision trees and random forests help understand the model's prediction logic and facilitate the formulation and implementation of safety measures. The correct choice of algorithm helps improve the model's generalization ability, that is, its performance on unseen data. By avoiding overfitting, ensure that the model has good predictive ability for new and unknown risk situations.

[0152] In practical application:

[0153] Use decision trees for risk classification: If the key risk features include numerical data such as temperature and gas concentration, and the goal is to classify the work environment as "high risk" or "low risk", a decision tree algorithm can be selected. Decision trees classify data by constructing a series of decision rules, which is easy to understand and explain and suitable for preliminary risk classification.

[0154] Using random forests to handle complex relationships: When there are complex nonlinear relationships between key risk features and the amount of data is large, the random forest algorithm is a better choice. Random forests improve accuracy by building multiple decision trees and combining their prediction results, and are suitable for processing complex and high-dimensional data.

[0155] Applying neural networks for deep learning: When there is a large amount of historical data and the relationship between key risk features and accident occurrence is extremely complex, neural networks can be used for deep learning. Neural networks can automatically extract and learn the deep relationships between features, especially when the interactions between features are complex and the patterns are not obvious.

[0156] By rationally selecting machine learning algorithms based on key risk characteristics, the accuracy and efficiency of safety risk assessment for confined space operations can be effectively improved, providing strong data support for preventive safety management.

[0157] Classification problem: If the risk assessment goal is to determine whether a specific security incident will occur (yes / no), it is a classification problem. Decision trees, random forests, support vector machines, and neural networks are all commonly used algorithms to solve classification problems.

[0158] Regression problem: If the goal is to predict a continuous value such as the likelihood or severity of a security incident, the problem is considered a regression problem. Linear regression, support vector regression (SVR), neural networks, etc. are suitable for regression problems.

[0159] The model is trained using a machine learning algorithm using key risk features in historical data as input and known risk events as output (labels). During the training process, the algorithm learns the relationship between features and risk events to predict the risk level of unknown data.

[0160] Evaluate model performance through cross-validation or independent test data sets, including accuracy, recall, etc. Adjust model parameters or algorithm selection based on performance results to optimize model performance.

[0161] By training and validating the risk assessment model, key risk features are effectively used to predict security risks, improving the accuracy and reliability of risk assessment. Through verification and tuning, it is ensured that the model has good generalization ability in practical applications.

[0162] In the air pressure anomaly detection and comprehensive analysis of the present invention, anomaly detection is performed on the air pressure data to identify data points that do not conform to the normal fluctuation pattern. Anomaly detection methods include statistical tests, threshold settings, isolated forests, etc. Analyze whether the abnormal air pressure changes are caused by external factors (such as climate change, equipment failure). By comparing the external environmental data of the corresponding time period, the real cause of the air pressure change is determined. Comprehensively analyze the abnormal air pressure data with other key risk features to identify potential risks that may not be covered by the model. This step requires the comprehensive application of data analysis, professional knowledge and logical reasoning. By identifying and analyzing abnormal air pressure changes, potential safety risks caused by external factors can be discovered. Comprehensive analysis helps reveal risk areas that may not be covered by risk assessment models, and provides a basis for further improving models and safety measures.

[0163] In one example, assume that in a chemical plant, the risk assessment model is trained based on features such as temperature and chemical concentration. Through anomaly detection, it is found that the air pressure drops abnormally during a period of time, while there is no obvious change in the external environment. After comprehensive analysis, it is found that the drop in air pressure is related to the increase in chemical reactions in a closed container, which is a potential risk factor that the original model did not fully consider.

[0164] Preferably, the historical accident records and environmental parameters corresponding to the key risk features are used as training samples. When constructing the training samples, the historical accident records are used as labels, and the environmental parameters at the corresponding time are used as input features; the training samples are divided into a training set and a validation set, which are used for training and validation of the selected model respectively;

[0165] The principle of constructing training samples is as follows: historical accident records are used as labels, which means that each sample is marked as a specific type of accident (such as chemical leaks, fires, etc.) or no accident. These labels are used during the training process to guide the model to learn how to distinguish different risk levels. Environmental parameters directly related to the occurrence of accidents are selected as input features, such as gas concentration, temperature, humidity, and air pressure. These features are used in model training to predict the possibility and severity of accidents. The constructed training samples are divided into a training set and a validation set. The training set is used for the initial training of the model, while the validation set is used to test the performance of the model on unknown data to evaluate its generalization ability.

[0166] Train the selected machine learning model through the training set, adjust the model parameters and complete the construction of the risk assessment model;

[0167] The performance of the risk assessment model is evaluated using a validation set, and the model parameters and structure are adjusted based on the evaluation results. After the performance of the risk assessment model is verified, it is used to take the preprocessed environmental data and key risk features as input to output risk prediction results, which include: possibility parameters and severity parameters of risk events.

[0168] The selected machine learning model is trained using the training set. During the training process, the algorithm attempts to find the best parameter settings so that the model can accurately predict the occurrence of accidents based on the input features. The performance of the model is evaluated using an independent validation set. Performance indicators include accuracy, recall, F1 score, etc., which reflect the accuracy and reliability of the model when predicting unknown data. Based on the performance evaluation results on the validation set, the parameters and structure of the model are adjusted. This includes changing the algorithm, adjusting parameters, or reselecting features to achieve better prediction results.

[0169] In the risk assessment model, "likelihood parameter" and "severity parameter" are two key indicators used to describe and quantify security risks. They provide a quantitative basis for risk management, making risk decisions more data-driven and precise.

[0170] The likelihood parameter indicates the probability or likelihood of a risk event occurring. It is a value between 0 and 1, where 0 means that the risk event is unlikely to occur and 1 means that the risk event is bound to occur.

[0171] Likelihood parameters are usually calculated by analyzing historical data, statistical models, or machine learning algorithms. For example, in a machine learning model, the likelihood parameter can be represented by a probability score output by the model, which is based on the model's analysis of input features (such as environmental parameters).

[0172] In actual risk management, likelihood parameters are used for prioritization, risk assessment, and resource allocation. Higher likelihood parameters indicate risk areas that require more urgent attention, helping decision makers prioritize preventive measures.

[0173] The severity parameter measures the extent of the impact or loss that a risk event may cause if it occurs. It can be a qualitative description (such as low, medium, high) or a quantitative value, reflecting the potential negative consequences ranging from financial losses, health impacts to environmental damage.

[0174] The calculation of severity parameters is based on expert assessment, historical accident analysis or quantitative models. If there is sufficient data support, the severity can be quantitatively calculated by analyzing the impact and consequences of historical accidents, combining factors such as cost, recovery time and impact scope.

[0175] Severity parameters are essential for understanding the potential impact of risks. In risk assessment, combining likelihood parameters with severity parameters allows for a more comprehensive evaluation of risks and the development of appropriate response strategies. For example, even if the likelihood of a risk event occurring is low, it may need to be prioritized and addressed if its severity is very high.

[0176] In the risk management framework, the likelihood parameter and the severity parameter are usually used together to create a risk matrix or risk map. This approach can not only identify high-risk areas that need to be focused on, but also provide a basis for formulating risk mitigation measures and resource allocation. Through this quantitative analysis, organizations can more effectively manage and reduce safety risks, ensure personnel safety and operational continuity.

[0177] Through the above process, a well-performing risk assessment model can be constructed, which can:

[0178] Predict the possibility of safety accidents based on real-time monitored environmental parameters to help prevent potential safety risks. Not only can it predict the possibility of accidents, but it can also classify risks based on the severity parameters of the predicted results, providing a basis for taking corresponding preventive measures.

[0179] In one example, in a petrochemical plant, historical data showed that the probability of fire increased significantly under certain gas concentration and temperature combinations. By constructing training samples and using these historical accident records and environmental parameters to train a random forest model, it is possible to predict the risk of fire and issue timely warnings when similar environmental parameters are monitored in real time, so that preventive measures can be taken to avoid accidents. After testing and tuning the model with a validation set, the model can accurately predict fire risks under new environmental conditions, significantly improving the safety management level of the plant.

[0180] Preferably, the process of obtaining the potential risk identification result includes:

[0181] Based on the air pressure data collected by the continuous monitoring sensor, the abnormal changes that do not conform to the normal air pressure fluctuation pattern are identified through a preset air pressure change threshold, and the air pressure abnormality data is generated, wherein the air pressure abnormality data includes all air pressure change information identified as abnormal;

[0182] Collect the external environment data of the limited space at the same time as the abnormal air pressure data, and compare the external environment data with the abnormal air pressure data at the same time. If the correlation coefficient between the external environment data and the abnormal air pressure exceeds the preset correlation threshold, it is determined that the abnormal air pressure is caused by external factors; if not, conduct a comprehensive analysis of the abnormal air pressure data and the determined key risk characteristics to determine whether the abnormal air pressure changes at the same time as the key risk characteristics. If so, ignore the abnormal air pressure data in this time period; if not, it is determined that there is a potential risk that is not covered by the risk assessment model, and the judgment result is used as the potential risk identification result.

[0183] In the present invention, the air pressure changes are tracked using the continuously monitored sensor data. By comparing with the preset air pressure change threshold, abnormal changes that do not conform to the normal air pressure fluctuation pattern are identified. The key to this step is to accurately set the threshold to distinguish between normal fluctuations and true abnormal changes.

[0184] Collect external environmental data at the same time as the abnormal air pressure data, such as temperature, humidity or other factors that may affect the air pressure. By analyzing the correlation between these external environmental data and the abnormal air pressure, determine whether the abnormal air pressure is caused by external factors. This requires setting a correlation threshold to quantify the strength of the relationship between the two.

[0185] The process of comprehensive analysis of potential risks is a key step in identifying and assessing safety risks, especially when it involves abnormal changes that are not directly related to conventional monitoring parameters. The following are specific steps for comprehensive analysis of abnormal air pressure data with identified key risk features, as well as the method steps used by the present invention in this process:

[0186] (1) Ensure that the air pressure anomaly data and key risk characteristic data are accurately aligned in time. This means that for each identified air pressure anomaly event, it is necessary to find the risk characteristic data of the corresponding period for further comparative analysis.

[0187] (2) Define a time window to examine the changes in key risk characteristics before and after the abnormal air pressure event. The length of the time window can be set according to the specific situation and past experience. For example, 30 minutes before and after the abnormal air pressure event can be selected as the analysis window.

[0188] (3) For the key risk features in each time window, use statistical methods (such as linear regression analysis, moving average or exponential smoothing, etc.) to analyze their changing trends. The purpose is to determine whether these key risk features also show abnormal changes when the abnormal air pressure event occurs.

[0189] Among them, time series analysis methods are used to analyze the changing trends of key risk features over time, especially before and after the occurrence of abnormal pressure events, which can help identify whether the feature changes are temporally correlated with the pressure anomaly. Correlation analysis (such as Pearson correlation coefficient, Spearman rank correlation coefficient, etc.) is used to quantitatively evaluate the correlation between the pressure anomaly data and key risk features, which helps to determine whether there is a statistical association between them. Anomaly detection algorithms (such as threshold-based detection, isolated forests, clustering-based methods, etc.) are used to identify abnormal changes in key risk features, especially in the context of abnormal pressure events. For features determined to be correlated, causal analysis methods, such as Granger causality test, are further explored to evaluate whether the predictive ability or causal influence of one feature on the change of another feature can be statistically inferred.

[0190] For those key risk features that also show abnormal change trends during the abnormal air pressure events, further analysis is needed to determine whether these changes may point to new potential risks that are not covered by the current risk assessment model. If the analysis determines that certain air pressure anomalies are indeed related to changes in key risk features, and this relationship is not considered by the existing model, then this will become an important basis for optimizing and improving the risk assessment model.

[0191] For those pressure anomalies that cannot be explained by external factors, they are analyzed in combination with the identified key risk features. This step involves evaluating the temporal correlation between the pressure anomalies and the key risk features to determine whether the anomalies indicate potential risks that are not covered by the assessment model.

[0192] This process can reveal potential risks that may not be directly considered by existing risk assessment models. For example, if a specific pattern of air pressure changes is associated with past accidents, and this pattern is not explicitly considered in the current model, this process can help identify and remedy this shortcoming. By identifying new potential risk factors, risk assessment models can be further refined and optimized to make them more comprehensive and accurate.

[0193] In one embodiment, in a chemical plant, historical data shows that during a specific chemical reaction, an abnormal drop in air pressure usually indicates a potential chemical leak risk. However, this air pressure change pattern may not be fully considered in existing risk assessment models. Through the above process, after identifying this air pressure anomaly and confirming its correlation with chemical leaks, this finding can be integrated into the risk assessment model to improve the accuracy of chemical leak prediction.

[0194] Through meticulous data analysis and intelligent risk identification processes, the deficiencies of existing risk assessment models can be effectively revealed and supplemented, providing stronger protection for confined space operation safety. This approach emphasizes the importance of continuous monitoring and data-driven analysis in modern safety management.

[0195] The risk assessment model is applied to predict the risk of the current limited space to obtain the risk prediction result, and the risk assessment result is constituted by the risk assessment result together with the potential risk identification result; based on the difference between the actual feedback and the risk prediction result, the risk assessment model is iteratively optimized.

[0196] In the process of safety risk assessment for confined space operations, applying risk assessment models to predict risks and iteratively optimizing the models based on actual feedback is a dynamic and continuous process. This process aims to improve the accuracy of risk prediction and ensure the effectiveness of safety measures.

[0197] Using the risk assessment model, the safety risks of confined space operations are predicted based on the current environmental parameters and the identified key risk characteristics. This step involves inputting real-time monitoring data into the model and obtaining predictions about the likelihood and severity of potential risk events.

[0198] The risk prediction results are combined with the potential risk identification results to form a comprehensive risk assessment result. This step ensures that the assessment results include both risks predicted based on historical data and models, as well as new or uncovered potential risks identified through real-time monitoring and analysis.

[0199] Based on the difference between actual feedback and risk prediction results, the risk assessment model is iteratively optimized. This includes adjusting model parameters, updating model algorithms, or adding new risk features to improve the accuracy of predictions and the generalization ability of the model.

[0200] Through continuous model optimization and updating, the accuracy of risk prediction can be gradually improved, and potential safety risks can be more effectively identified and predicted. Applying risk assessment models to real-time data and combining them with the process of model optimization makes risk management more dynamic and adaptable, and can respond to environmental changes and emerging risks in a timely manner.

[0201] In one embodiment, in a mine operation scenario, the risk assessment model predicts risks based on features such as temperature, humidity, and harmful gas concentrations, and identifies high-risk areas and time periods. Through actual applications, it was found that the model failed to accurately predict local poor ventilation caused by equipment failure. The actual feedback data collected showed that this situation would cause a sharp increase in the concentration of harmful gases in specific areas. Therefore, the development team iteratively optimized the model, added equipment status monitoring data as a new input feature, and adjusted the model parameters to better reflect this risk factor. The optimized model can more accurately identify and warn of potential poor ventilation risks, effectively improving the safety of mine operations.

[0202] Through this continuous risk assessment and model optimization process, the effectiveness of safety management measures can be ensured, and emerging safety risks can be identified and responded to in a timely manner, thereby ensuring the safety of confined space operations.

[0203] Preferably, the iterative optimization of the risk assessment model includes:

[0204] Verify the predictions of the risk assessment model and collect actual feedback related to the predictions;

[0205] Compare the predictions of the risk assessment model with actual events. This process usually involves collecting actual feedback, including whether the events predicted by the model actually occurred and whether the predicted risk level is consistent with the actual situation. Data and accident reports provided by real-time monitoring systems can serve as a source of verification information to help identify the accuracy and reliability of the model.

[0206] Compare the differences between risk prediction results and actual risk events to evaluate the performance of the risk assessment model;

[0207] Analyze the differences between the predicted results and the actual events and evaluate the performance of the model. This includes calculating indicators such as accuracy, recall, precision, and analyzing the performance of the model under specific conditions. Use statistical and data analysis methods to evaluate the performance of the model in predicting different types of risk events in detail and identify the strengths and weaknesses of the model.

[0208] Based on the assessment results, the risk assessment model is iteratively optimized.

[0209] Based on the performance evaluation results, the model is adjusted as necessary. This includes adjusting model parameters, improving feature selection, updating data processing procedures, and even replacing algorithms. The model is continuously improved in combination with the latest research results and technological advances to improve prediction accuracy and the ability to respond to emerging risks.

[0210] The present invention can significantly improve the accuracy and reliability of risk prediction and better prevent and manage potential security risks by continuously verifying and optimizing the model. Iterative optimization ensures that the model can adapt to environmental changes and new risk factors, maintaining its effectiveness and relevance in a dynamic environment.

[0211] In one example, on a construction site, a preliminary risk assessment model was established to predict safety risks based on factors such as weather conditions, working hours, and frequency of equipment use. After a period of application, by comparing the safety accidents predicted by the model with the actual accidents, it was found that the model was not accurate enough in predicting risks under extreme weather conditions. Therefore, the research team collected more data on accidents under extreme weather conditions and iteratively optimized the model, including introducing new meteorological factor features, adjusting algorithm parameters, and retraining the model. The optimized model showed higher prediction accuracy in subsequent verifications and successfully reduced the incidence of safety accidents on construction sites under severe weather conditions.

[0212] Through such an iterative optimization process, the risk assessment model can continue to improve and provide more accurate and reliable safety risk predictions for confined space operations, thereby effectively guiding the formulation and implementation of safety management measures.

[0213] like Figure 3 As shown, a system for implementing the method for assessing safety risks of confined space operations includes:

[0214] A sensor network, the sensor network includes: different types of sensors such as gas concentration sensors, temperature sensors, humidity sensors and air pressure sensors, the sensor network is used to monitor the environmental conditions in a limited space in real time and collect raw environmental data;

[0215] Data processing and storage hardware, including: a data collector, which is connected to the sensor network and is responsible for collecting sensor data; a central processing unit, which is responsible for further data preprocessing, feature extraction, and computing tasks of running machine learning models; a data storage system, which is used to store raw data, processed data, and risk assessment results;

[0216] A risk assessment module, which is deployed on a central server or cloud platform and is responsible for executing a risk assessment model, analyzing the relationship between environmental data and risks, and generating risk assessment results;

[0217] A model iteration optimization module, wherein the model iteration optimization module includes an input device for collecting operator feedback and accident reports, and is used to perform iterative optimization and training of the risk assessment model.

[0218] The system of the present invention uses a sensor network to monitor key environmental parameters in a confined space in real time, including gas concentration, temperature, humidity and air pressure. These parameters are directly related to operational safety, and their changes can indicate potential safety risks. Sensor data is collected by a data collector and then preprocessed in a central processing unit, including data cleaning and normalization, and key risk features are extracted from the processed data. These features provide the machine learning model with the information needed to assess risks. Based on the extracted key features, the appropriate machine learning algorithm type is selected to train and verify the model, thereby building a model that can accurately assess safety risks. The risk assessment module analyzes the relationship between real-time environmental data and risk, and generates risk assessment results, including identifying high-risk areas and times. At the same time, the system iteratively optimizes the model by collecting feedback and accident reports from operators to improve the accuracy and adaptability of the assessment.

[0219] In one embodiment, in a confined space operation scenario in a chemical plant, the sensor network detected a sudden increase in the concentration of combustible gas in a certain area. After data processing and feature extraction, it was found that the risk pattern was similar to that in historical accident data. The machine learning model then assessed the risk level of the situation as high, and the risk assessment module generated early warning information to guide the operators to take emergency risk avoidance measures. Afterwards, the collected feedback information found that it was caused by equipment leakage, and the model iteration optimization module adjusted the model accordingly to enhance the prediction ability of similar risks.

[0220] The system of the present invention realizes real-time safety monitoring and risk assessment of confined space working environment, and provides timely safety warnings for operators. Through continuous data analysis and model optimization, the system can adapt to environmental changes and new risk factors, improve the accuracy and reliability of the assessment model, support operational safety management decisions, and reduce the probability of accidents by identifying and preventing potential risks.

[0221] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0222] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0223] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0224] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0225] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0226] Memory includes non-permanent memory in a computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0227] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0228] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0229] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for assessing safety risks of confined space operations, characterized in that: The following steps are involved: By deploying a sensor network, the environmental data inside the limited space is collected in real time and pre-processed; Extract key features from pre-processed environmental data, obtain key features related to risk events through preliminary screening using the Pearson correlation coefficient, and obtain key risk features that have the greatest impact on risk assessment through in-depth screening; By using data analysis technology, the change characteristics in the environmental data are identified and key features are extracted; wherein the key features include: gas concentration characteristics, temperature characteristics, humidity characteristics and air pressure characteristics; Apply statistical analysis methods to calculate the Pearson correlation coefficient between each of the key features and known risk events, and based on a preset correlation threshold, select the key features whose absolute value of the Pearson correlation coefficient exceeds the correlation threshold to complete the preliminary screening; The in-depth screening includes: Partial correlation analysis: select one key feature as the analysis object and other key features as control variables, and calculate the partial correlation coefficient between the analysis object and the risk event after controlling the influence of other control variables; Multiple regression analysis: Take the risk event as the dependent variable and all the key risk features obtained through preliminary screening as independent variables to build a multiple regression model. Use historical data to complete the training of the multiple regression model and analyze the influence coefficient and influence value of each key feature in the multiple regression model. Based on the results of partial correlation analysis and multivariate regression analysis, the key risk features that have the greatest impact on risk assessment are screened out; After selecting the type of machine learning algorithm to be applied based on the key risk characteristics, the model is trained and verified to obtain a trained risk assessment model; abnormality detection and external factor elimination are performed on the air pressure data, and a comprehensive analysis is performed with the key risk characteristics to identify potential risks not covered by the risk assessment model and obtain potential risk identification results; The historical accident records and environmental parameters corresponding to the key risk features are used as training samples. When constructing training samples, the historical accident records are used as labels, and the environmental parameters at the corresponding time are used as input features. The training samples are divided into a training set and a validation set, which are used for training and validation of the selected model respectively. Based on the air pressure data collected by the continuous monitoring sensor, the abnormal changes that do not conform to the normal air pressure fluctuation pattern are identified through a preset air pressure change threshold, and the air pressure abnormality data is generated, wherein the air pressure abnormality data includes all air pressure change information identified as abnormal; Collect the external environment data of the limited space at the same time as the abnormal air pressure data, and compare the external environment data with the abnormal air pressure data at the same time. If the correlation coefficient between the external environment data and the abnormal air pressure exceeds the preset correlation threshold, it is determined that the abnormal air pressure is caused by external factors; if not, conduct a comprehensive analysis of the abnormal air pressure data and the determined key risk characteristics to determine whether the abnormal air pressure changes at the same time as the key risk characteristics. If so, ignore the abnormal air pressure data in the corresponding time period; if not, it is determined that there is a potential risk that is not covered by the risk assessment model, and the determination result is used as the potential risk identification result; The risk assessment model is applied to predict the risk of the current limited space to obtain the risk prediction result, and the risk assessment result is constituted by the risk assessment result together with the potential risk identification result; based on the difference between the actual feedback and the risk prediction result, the risk assessment model is iteratively optimized.

2. The method for assessing safety risks of confined space operations according to claim 1, characterized in that: The preprocessing includes: performing data cleaning and normalization processing on the environmental data; The environmental data include: gas concentration, temperature, humidity and air pressure.

3. The method for assessing safety risks of confined space operations according to claim 1, characterized in that: The Pearson correlation coefficient calculation expression is: ,in, is the Pearson correlation coefficient, which is used to measure the and The degree of linear correlation between is the number of observations; and The variables are and No. observations; and The variables are and The average value of The value of Pearson correlation coefficient is : when = 1 o'clock, indicating and There is a completely positive correlation between them; when =- 1 o'clock, indicating and There is a completely negative correlation between them; when = 1 o'clock, indicating and There is no linear correlation between them.

4. The method for assessing safety risks of confined space operations according to claim 1, characterized in that: Train the selected machine learning model through the training set, adjust the model parameters and complete the construction of the risk assessment model; The performance of the risk assessment model is evaluated using a validation set, and the model parameters and structure are adjusted based on the evaluation results. After the performance of the risk assessment model is verified, it is used to take the preprocessed environmental data and key risk features as input to output risk prediction results, which include: possibility parameters and severity parameters of risk events.

5. The method for assessing safety risks of confined space operations according to claim 1, characterized in that: The iterative optimization of the risk assessment model includes: Verify the predictions of the risk assessment model and collect actual feedback related to the predictions; Compare the differences between risk prediction results and actual risk events to evaluate the performance of the risk assessment model; Based on the assessment results, the risk assessment model is iteratively optimized.

6. A system for implementing the method for assessing safety risks of confined space operations as described in any one of claims 1 to 5, characterized in that: include: A sensor network, the sensor network includes: different types of sensors such as gas concentration sensors, temperature sensors, humidity sensors and air pressure sensors, the sensor network is used to monitor the environmental conditions in a limited space in real time and collect raw environmental data; Data processing and storage hardware, including: a data collector, which is connected to the sensor network and is responsible for collecting sensor data; a central processing unit, which is responsible for further data preprocessing, feature extraction, and computing tasks of running machine learning models; a data storage system, which is used to store raw data, processed data, and risk assessment results; A risk assessment module, which is deployed on a central server or cloud platform and is responsible for executing a risk assessment model, analyzing the relationship between environmental data and risks, and generating risk assessment results; A model iteration optimization module, wherein the model iteration optimization module includes an input device for collecting operator feedback and accident reports, and is used to perform iterative optimization and training of the risk assessment model.

Images