Steel production risk dynamic early warning method, device, equipment and medium

Abstract

The invention provides a steel production risk dynamic early warning method and device, equipment and a medium, and the method comprises the steps: obtaining original data in a steel production process, carrying out the fluctuation analysis and stability analysis of the original data, carrying out the pairwise comparison of all dimension indexes based on the analyzed data, generating a weight matrix, calculating a comprehensive weight coefficient, and carrying out the dynamic early warning of the steel production risk. Inputting the comprehensive weight coefficient into a preset hybrid model, calculating a comprehensive safety risk value of the to-be-detected area in combination with a predefined membership function, triggering an early warning signal according to a comparison result of the risk value and a preset threshold value, executing an emergency response strategy, and after the response strategy is executed, performing early warning on the to-be-detected area. Model parameters are adjusted based on the verification data set, and full-life-cycle management is achieved through time sequence version comparison; according to the method, through dynamic weight calculation, non-deterministic risk factor characterization and a closed-loop optimization mechanism, risk identification perspectiveness and emergency response efficiency are improved.

Description

Technical Field

[0001] The present application relates to the field of steel smelting technology, and in particular to a method, device, equipment and medium for dynamic early warning of steel production risks. Background Art

[0002] As a pillar industry of the national economy, the steel industry's production processes involve high-risk scenarios such as high-temperature molten metal, clusters of specialized equipment, and gas transmission and distribution systems. These processes are characterized by high complexity and multi-source risk coupling. Risk factors such as human behavior, equipment status, process parameters, and environmental conditions are intertwined and subject to frequent dynamic changes during steelmaking, rolling, and transportation. For example, abnormal blast furnace top pressure can lead to cascading equipment failures, gas leaks can create explosion risks, and operator errors or fatigue can further exacerbate potential accidents. However, current mainstream risk management still relies on traditional methods such as manual inspections, single-dimensional threshold alarms, and offline statistical analysis, which exhibit significant lags and limitations.

[0003] In addition, the traditional "post-event processing" model focuses on accident emergency response and lacks the ability to predict and proactively intervene in the risk evolution process, making it difficult to adapt to the needs of modern steel companies for the transformation of production safety from "passive response to active defense." Summary of the Invention

[0004] In view of the above-mentioned shortcomings of the prior art, the present application provides a steel production risk dynamic early warning method, device, equipment and medium to solve the above-mentioned technical problems.

[0005] The present application provides a dynamic early warning method for steel production risks, which includes: obtaining original data in the steel production process, wherein the original data includes data of multiple different data dimensions; performing volatility analysis and stability analysis on the original data, including: calculating the dynamic fluctuation range of indicators in each dimension, marking abnormal fluctuation points, and identifying the stability of the original data based on the abnormal fluctuation points within a unit time; based on the analyzed data, performing pairwise comparisons on the indicators in each dimension to generate a weight matrix, and calculating a comprehensive weight coefficient; inputting the comprehensive weight coefficient into a preset hybrid model, and calculating the comprehensive safety risk value of the area to be detected in combination with a predefined membership function; triggering an early warning signal and executing an emergency response strategy based on the comparison result of the risk value with a preset threshold; after executing the response strategy, adjusting the model parameters based on the verification data set and realizing full life cycle management through time series version comparison.

[0006] In one embodiment of the present application, the volatility analysis and stability analysis include: determining the standard fluctuation range of the original data based on the standard value and mean of historical data; if any original data exceeds the standard fluctuation range, marking it as an abnormal fluctuation point; identifying abnormal fluctuation points within a preset unit time to obtain the abnormal fluctuation frequency, and if the abnormal fluctuation frequency exceeds the preset frequency threshold, the original data is determined to be unstable; wherein, the result of the stability assessment is used to score the original data, if stable, points are added, if unstable, points are deducted, and the scoring result is used to adjust the parameters of the preset hybrid model when constructing the preset hybrid model.

[0007] In one embodiment of the present application, the method also includes constructing a time series prediction model based on historical risk data, specifically including: multi-source integration of risk values, volatility analysis results and risk assessment results generated in real time to generate a unified time series data set; constructing a prediction model based on the time series data set, dynamically adjusting model parameters according to data characteristics, and improving prediction accuracy through an indicator optimization mechanism; using the prediction model to output risk trend values ​​for future time periods, generating a warning signal when the risk trend value exceeds a preset risk threshold, and linking the execution of a gradient emergency strategy.

[0008] In one embodiment of the present application, a weight matrix is ​​generated and the comprehensive weight coefficient of each dimensional indicator is calculated, including: decomposing the security issues of the area to be detected into multiple independent risk factors; arranging them hierarchically according to the influence and dominance relationship of each risk factor to form a hierarchical structure; comparing the risk factors at each level pairwise, and converting the comparison results into a weight matrix, which is used to reflect the importance of each factor on security; calculating the consistency ratio of the weight matrix, and if the consistency ratio is greater than a preset threshold, readjusting the pairwise comparison scale of the risk factors and iteratively calculating until the consistency ratio does not exceed the preset threshold, and the consistency ratio is determined by the ratio of the consistency index to the random consistency index; combining the weight values ​​at different levels to generate a comprehensive weight coefficient of the impact of each risk factor on security.

[0009] In one embodiment of the present application, the calculation of the comprehensive safety risk value of the area to be inspected includes: constructing a membership function based on historical data and expert knowledge before inputting the model; standardizing the indicator values ​​of each dimension, and inputting the standardized indicator values ​​of each dimension into the membership function to obtain the corresponding fuzzy risk value; multiplying the fuzzy risk value with the corresponding comprehensive weight coefficient to generate a weighted risk value of each dimension; inputting the weighted risk value of each dimension into a preset probability model for fusion calculation, and outputting the comprehensive safety risk value of the area to be inspected.

[0010] In one embodiment of the present application, the method also includes risk response closed-loop management, specifically including: mapping the comprehensive safety risk value to the digital twin model of the production line in real time, and using a thermal rendering engine to create a risk gradient distribution map; when the early warning signal is triggered, automatically associating and activating the one-key traceability interface on the risk gradient distribution map, responding to the user's interactive operations, and realizing the following functions: according to the coordinates of the high-risk area in the risk gradient distribution map, reverse tracing to generate the risk data path of the area; tracing back along the hierarchy of the production process to locate the data source equipment and process parameters of the previous link that caused the anomaly; marking the three-dimensional position of the source of the anomaly in the digital twin model, and generating a traceability report at the same time.

[0011] In one embodiment of the present application, an early warning signal is triggered and an emergency response strategy is executed based on the comparison result between the risk value and the preset threshold, including: judging the current risk level based on the comparison result, and triggering a gradient emergency strategy corresponding to the current risk level, the gradient emergency strategy including at least a first-level emergency disposal strategy, a second-level area control strategy, and a third-level preventive maintenance strategy; if the current risk level is a high-risk level, the first-level emergency disposal strategy is executed, including triggering an emergency shutdown operation of key equipment, starting a personnel evacuation broadcast and alarm mechanism, and activating real-time monitoring and information enhancement display of risk areas; if the current risk level is a medium-risk level, the second-level area control strategy is executed, including implementing operation restrictions in high-risk areas, sending alarm information to relevant personnel, and enhancing monitoring of key areas; if the current risk level is a low-risk level, the third-level preventive maintenance strategy is executed, including generating an equipment maintenance work order, generating an equipment maintenance plan based on risk and status data, and pushing the equipment maintenance work order and the equipment maintenance plan to maintenance personnel.

[0012] In one embodiment of the present application, the full life cycle management includes: retrospectively verifying the warning results based on the verification data set, and generating a verification report including the false alarm rate and the missed alarm rate; dynamically adjusting the risk level threshold and the weight coefficient of each dimension based on the false alarm / missing alarm data analysis in the verification report; inputting the adjusted parameters into the hybrid model to simulate the impact of parameter changes on the risk value output, and identifying key risk factors whose sensitivity to the warning results exceeds the preset value; for the identified key factors, comparing the differences in historical version model parameters, locating the change nodes of weight coefficients, risk thresholds and scoring rules; updating the model configuration based on the difference analysis results, generating a new version model for deployment to the production environment, and collecting the deployed warning data as the verification data set for the next cycle.

[0013] The present application provides a dynamic early warning device for steel production risks, characterized in that the device includes: a data acquisition module for acquiring original data in the steel production process, wherein the original data includes data of multiple different data dimensions; a data preprocessing module for performing volatility analysis and stability analysis on the original data, including: calculating the dynamic fluctuation range of indicators of each dimension, marking abnormal fluctuation points, and identifying the stability of the original data based on abnormal fluctuation points within a unit time; a risk assessment module for performing pairwise comparison of indicators of each dimension based on the analyzed data to generate a weight matrix and calculate a comprehensive weight coefficient; inputting the comprehensive weight coefficient into a preset hybrid model, and calculating the comprehensive safety risk value of the area to be detected in combination with a predefined membership function; an early warning response module for triggering an early warning signal and executing an emergency response strategy based on the comparison result of the risk value with a preset threshold; a risk response closed-loop management module for adjusting model parameters based on a verification data set after executing the response strategy and realizing full life cycle management through time series version comparison.

[0014] The present application provides an electronic device, which includes: one or more processors; a storage device for storing one or more programs, and when the one or more programs are executed by the one or more processors, the electronic device implements the dynamic early warning method for steel production risks as described above.

[0015] The present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor of a computer, the computer is caused to execute the above-mentioned dynamic early warning method for steel production risks.

[0016] Beneficial effects of the present application: The dynamic early warning method for steel production risks proposed in the present application obtains multi-source heterogeneous data (such as equipment status, process parameters, environmental monitoring, personnel behavior, etc.) in the steel production process, compares the indicators of each dimension pairwise based on the hierarchical analysis method, constructs a weight matrix and calculates the comprehensive weight coefficient; the weight coefficient is input into the preset fuzzy logic-probability theory hybrid model, and the indicator value is converted into a fuzzy risk value through the membership function, and the regional comprehensive safety risk value is dynamically calculated in combination with the weight coefficient; according to the comparison result between the risk value and the preset threshold, the graded early warning signal is triggered and the corresponding emergency response strategy is executed. This method significantly improves the globality and foresight of risk identification through dynamic weight adjustment and fuzzy risk characterization, breaking through the limitations of traditional single-dimensional threshold alarms; combined with the closed-loop management link (monitoring-assessment-early warning-disposal-optimization), it realizes the transformation from "post-processing" to "active defense" mode, effectively shortens the emergency response time, and reduces the rate of production safety accidents.

[0017] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings are incorporated into and constitute a part of the specification, illustrating embodiments consistent with the present application and, together with the specification, serving to explain the principles of the present application. It is obvious that the drawings described below are merely some embodiments of the present application, and a person of ordinary skill in the art can derive other drawings based on these drawings without inventive effort. In the drawings:

[0019] Figure 1 1 is a schematic diagram of an implementation environment of a dynamic early warning method for steel production risks according to an exemplary embodiment of the present application;

[0020] Figure 2 This is a flow chart of a method for dynamic early warning of steel production risks shown in an exemplary embodiment of the present application;

[0021] Figure 3 This is a schematic diagram of the steps of risk assessment and weight calculation based on the analytic hierarchy process, shown in an exemplary embodiment of the present application;

[0022] Figure 4 is a block diagram of a steel production risk dynamic early warning device shown in an exemplary embodiment of the present application;

[0023] Figure 5 A schematic diagram of the structure of a computer system suitable for implementing an electronic device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0024] The following will describe the embodiments of the present application with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand the other advantages and effects of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be understood that the preferred embodiments are only for the purpose of illustrating the present application and are not intended to limit the scope of protection of the present application.

[0025] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present application. Therefore, the illustrations only show components related to the present application and are not drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component can be changed at will, and the component layout type may also be more complicated.

[0026] In the following description, a large number of details are discussed to provide a more thorough explanation of the embodiments of the present application. However, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present application difficult to understand.

[0027] Figure 1 It is a schematic diagram of the implementation environment of the steel production risk dynamic early warning method shown in an exemplary embodiment of the present application.

[0028] like Figure 1 As shown, the implementation environment of the dynamic early warning method for steel production risks includes a data acquisition module 101 and computer equipment 102. Data acquisition module 101 is responsible for acquiring multi-dimensional, heterogeneous data from the steel production area in real time to support risk assessment and early warning decisions. Specifically, data acquisition module 101 consists of an industrial sensor network, a video surveillance system, personnel positioning equipment, and an IT business data interface. The industrial sensor network covers key areas such as blast furnaces, hot blast stoves, converters, and gas tanks, collecting equipment status data, process parameters, and environmental monitoring data via protocols such as Modbus, OPC, and MQTT. The video surveillance system uses GBT26875 and GBT28181 protocols to capture real-time images of the work area, which is used to identify personnel violations or abnormal events. The personnel positioning equipment combines Beidou, UWB, and Bluetooth technologies to track the location and behavior of workers in real time, preventing them from remaining in high-risk areas. The IT business data interface connects to the enterprise security management platform through REST API or MQ messages to obtain personnel information (such as occupational health records and training status), hidden danger rectification records, emergency plans and historical accident case libraries, providing business background support for risk assessment.

[0029] Computer equipment 102 is the core of the system, responsible for data processing, risk assessment, early warning strategy generation, and emergency response control. Its functional modules include a data preprocessing unit, a risk assessment model, a dynamic early warning engine, and an emergency response control center. The data preprocessing unit cleans, analyzes volatility, and integrates features of collected raw data, unifying multi-source heterogeneous data into standardized risk assessment inputs. The risk assessment model constructs a multi-level risk weighting system based on the Analytic Hierarchy Process (AHP). It generates comprehensive weight coefficients for each dimensional indicator through pairwise expert comparison and data analysis. Combining fuzzy logic and probabilistic hybrid modeling, it converts indicators such as equipment status, process parameters, and environmental monitoring into fuzzy risk values. The comprehensive safety risk value for the region is calculated based on the comprehensive weights. The dynamic early warning engine uses the ARIMA time series model to predict historical risk scores. Combined with preset four-dimensional risk level thresholds, it triggers graded early warning signals and generates a dynamic risk map. The emergency response control center then implements a three-level gradient emergency response strategy based on the risk level. At the same time, the risk level is projected onto the three-dimensional model in real time through the digital twin visualization platform, and the thermal rendering engine is used to achieve gradient coloring expression of regional risks. The video surveillance images, personnel location information and multimodal alarms (SMS, broadcast, telephone) are linked to form a closed-loop emergency command mechanism.

[0030] Furthermore, computer device 102 includes a model optimization module that continuously iterates the risk assessment model through a closed-loop mechanism involving cross-validation, sensitivity analysis, and expert feedback. For example, based on the alignment of historical alarm records with a database of production safety incident cases, the AHP weight matrix and fuzzy membership function parameters are dynamically adjusted to improve the model's adaptability to complex production scenarios. Through multi-source data fusion, dynamic weight analysis, and intelligent prediction algorithms, this implementation environment achieves a transition from a "post-processing" to a "proactive defense" safety management model, providing steel companies with technical support for risk prevention and control throughout their lifecycle.

[0031] Figure 2 It is a flow chart of a method for dynamic early warning of steel production risks shown in an exemplary embodiment of the present application.

[0032] like Figure 2 As shown, in an exemplary embodiment, the steel production risk dynamic early warning method includes at least steps S210 to S240, which are described in detail as follows:

[0033] Step S210: acquiring original data in the steel production process, where the original data includes data of multiple different data dimensions.

[0034] In one embodiment of the present application, multi-dimensional, multi-source, heterogeneous data related to security in the steel production process, that is, the original data in the steel production process, is obtained in real time through the Internet of Things transmission technology. Among them, OT data (operational technology data) is mainly collected through industrial protocols (such as Modbus, OPC, MQTT, GBT26875, GBT28181), covering the equipment status of key areas such as blast furnaces, hot blast stoves, and converters (such as smoke detector signals, pressure sensor readings, equipment interlocking utilization rate), process operation parameters (such as blast furnace top pressure, converter oxygen blowing volume, cooling water flow), operating environment monitoring data (such as toxic and harmful gas concentration, noise level, dust concentration), etc. IT data (information technology data) obtains business data from third-party information systems (such as safety management platforms, hidden danger rectification record libraries, and dangerous operation plan libraries) through REST API interfaces or MQ message subscription methods, including management dimension data such as personnel location information, education and training records, emergency drill records, and labor protection clothing equipment status.

[0035] Step S220, performing volatility analysis and stability analysis on the original data, including: calculating the dynamic fluctuation range of each dimensional indicator, marking abnormal fluctuation points, and identifying the stability of the original data based on the abnormal fluctuation points within a unit time. The volatility analysis and stability analysis include: determining the standard fluctuation range of the original data based on the standard value and mean of the historical data; if there is any original data that exceeds the standard fluctuation range, it is marked as an abnormal fluctuation point; identifying the abnormal fluctuation points within a preset unit time to obtain the abnormal fluctuation frequency, and if the abnormal fluctuation frequency exceeds the preset frequency threshold, the original data is determined to be unstable. The result of the stability assessment is used to score the original data, adding points if it is stable and deducting points if it is unstable. The scoring result is used to adjust the parameters of the preset hybrid model when constructing the preset hybrid model.

[0036] In addition, when it is identified that the original data is unstable, data verification of multiple source devices in the same dimension can be further initiated. When data anomalies are confirmed, the sliding window mean is used to replace the abnormal values. When a device failure is confirmed, the dimension data is frozen and the backup data stream is enabled.

[0037] In one embodiment of the present application, the stability of multi-dimensional data of steel production is guaranteed by dynamically calculating the standard fluctuation range, marking abnormal points, triggering redundancy check and data repair mechanism. First, the mean (μ) and standard deviation (σ) of each indicator are calculated based on historical data (such as the past 7 days), and the standard fluctuation range [μ-kσ, μ+kσ] (k is a dynamic adjustment coefficient, such as k=3) is dynamically set, and the benchmark value is regularly updated according to the production conditions to adapt to real-time needs. In the real-time processing stage, the system compares the current data with the fluctuation range, marks the points outside the range as abnormal, and counts the abnormal frequency within the unit time window (such as 10 minutes); if the frequency exceeds the threshold (such as ≥3 times), redundancy check is triggered. Redundancy check verifies the consistency by calling multi-source device data of the same dimension (such as multiple sensors): if the multi-source data is normal, the current sensor is determined to be abnormal; if the multi-source data is also abnormal, it is determined to be a systemic fault. The data repair strategy includes: using a sliding window mean replacement for single-point anomalies; freezing the dimension data stream for continuous anomalies and verification failures and enabling a backup data stream (such as historical data predictions or interpolation replacements). Regarding dynamic adjustment of key parameters, the k value is set according to production stability requirements (e.g., k=2 for high-risk areas, k=3 for low-risk areas), and the sliding window size is set according to the data frequency (e.g., a 5-second window for high-frequency data, a 1-minute window for low-frequency data). Through the above steps, the system detects data fluctuation anomalies in real time, combines multi-source redundancy verification with a dynamic repair mechanism, ensures the stability and reliability of production data, and provides high-quality input for subsequent risk assessments.

[0038] In a specific embodiment of the present application, after the original data is collected, it also includes cleaning and volatility analysis of the original data to generate standard data, so as to perform subsequent dimensional comparison and other related data processing based on the standard data.

[0039] In one embodiment of the present application, pre-processing such as data cleaning and fluctuation analysis is performed on the collected raw data, and the specific steps include:

[0040] Linear interpolation is used to fill in missing time series values ​​in OT data, such as the short-term data gap of the blast furnace top pressure sensor when communication is interrupted; missing business records in IT data (such as personnel physical examination records not updated in a timely manner) are supplemented by historical data or marked as "pending confirmation".

[0041] Based on the 3σ principle (mean ± 3 times the standard deviation), anomalies in OT data are identified and filtered, such as a sudden increase in the pressure sensor of a gas tank to 100kPa. Logical contradictions in IT data (such as the same person being registered as "on duty" and "on vacation" at the same time) are corrected through cross-checking.

[0042] Normalize indicators of different dimensions (such as temperature in °C and gas concentration in ppm) to the range [0, 1] to facilitate subsequent multi-dimensional comparisons. For example, convert the blast furnace top pressure (range 0-300 kPa) to a standardized value between 0 and 1 using a formula.

[0043] Next, after cleaning the data, we further analyze the volatility of the OT data's time series characteristics to assess system stability and dynamically adjust the threshold. The specific steps are as follows:

[0044] Calculate the volatility characteristics of the indicator within a sliding time window (such as a 5-minute window), including standard deviation (σ), coefficient of variation (CV = σ / μ), and kurtosis. For example, for the real-time data of the converter oxygen blowing amount, calculate its standard deviation within a 5-minute window. If σ>15m 3 / min is marked as high volatility.

[0045] The threshold range is dynamically adjusted based on historical data distribution and current production conditions. For example, the fluctuation threshold of the blast furnace top pressure is dynamically adjusted to [μ_kσ,μ+kσ] based on historical data, where k is the dynamic coefficient (initial value is 1.5 and automatically adjusted according to changes in production load).

[0046] The volatility analysis results are combined with the real-time parameter values ​​to comprehensively assess the risk through fuzzy logic algorithms. For example, if the top pressure value at a certain moment is 280kPa and its volatility standard deviation is 25kPa (higher than the threshold of 20kPa), it will be mapped to the "high risk" level (membership degree is 0.8) through a membership function (such as a triangular membership function).

[0047] Finally, the cleaned data and the volatility analysis results are combined to form a standard data set as the input for subsequent risk assessment and early warning. For example: output standardized equipment interlock commissioning rate (0.92), inspection plan execution rate (0.95), hidden danger elimination rate (0.88), and volatility analysis results (such as top pressure standard deviation 18kPa); output standardized personnel positioning information (coordinates [x, y]), continuous working time (4.5 hours), occupational health examination rate (0.90), and behavior monitoring data (such as fatigue score 0.75); output standardized toxic gas concentration (CO: 50ppm→0.67), dust concentration (80mg / m 3 →0.80), as well as volatility analysis results (e.g., dust concentration kurtosis value of 2.1). The resulting standard data is then output in JSON format through a real-time processing engine (e.g., Apache Flink) for subsequent Analytic Hierarchy Process (AHP) weight calculation and risk dynamic assessment module invocation.

[0048] It can be understood that the standard data set obtained based on the above embodiment provides a unified and reliable foundation for subsequent multi-dimensional risk assessment, and supports accurate tracing and dynamic early warning from "single indicator anomaly" to "regional-level risk evolution".

[0049] In step S230, based on the analyzed data, each dimension indicator is compared pairwise to generate a weight matrix and calculate the comprehensive weight coefficient.

[0050] In one embodiment of the present application, a weight matrix is ​​generated and the comprehensive weight coefficient of each dimensional indicator is calculated, including: decomposing the security issues of the area to be detected into multiple independent risk factors; arranging them hierarchically according to the influence and dominance relationship of each risk factor to form a hierarchical structure; comparing the risk factors at each level pairwise, and converting the comparison results into a weight matrix, which is used to reflect the importance of each factor on security; calculating the consistency ratio of the weight matrix, if the consistency ratio is greater than a preset threshold, readjusting the pairwise comparison scale of the risk factors and iterating the calculation until the consistency ratio does not exceed the preset threshold, and the consistency ratio is determined by the ratio of the consistency index to the random consistency index; combining the weight values ​​at different levels to generate a comprehensive weight coefficient of the impact of each risk factor on security.

[0051] In one embodiment of the present application, the decomposition and weight calculation of safety risk factors in steel production areas are achieved through the analytic hierarchy process (AHP) to ensure the scientific nature and operability of the assessment results. First, the safety issues in the area to be inspected are decomposed into risk factors in dimensions such as equipment, process, environment, and personnel, and a hierarchical structure (target layer-criterion layer-indicator layer) is constructed. Subsequently, experts are organized to compare the criterion layer factors in pairs, and the 1-9 scaling method is used to determine the relative importance, generate a weight matrix, and calculate the weights of each criterion layer through the eigenvector method. If the consistency ratio (CR=CI / RI) of the weight matrix exceeds the preset threshold (such as 0.1), the pairwise comparison scale is adjusted and iterative calculation is performed until the consistency requirements are met. On this basis, similar pairwise comparisons and weight matrix construction are performed on the indicator layer factors, and the comprehensive weight coefficients of each indicator are calculated in combination with the criterion layer weights (comprehensive weight=criterion layer weight×indicator layer weight). For example, the comprehensive weight of the equipment interlock utilization rate is the product (0.2) of the equipment dimension weight (0.4) and the sub-weight (0.5). The comprehensive weight of the blast furnace top pressure is the product (0.12) of the process dimension weight (0.3) and the sub-weight (0.4). By dynamically adjusting the pairwise comparison scale, regularly updating the weight matrix, and introducing an expert feedback closed-loop mechanism, the system continuously optimizes weight allocation, identifies key risk factors (such as prioritizing monitoring of high-weight items), and ultimately generates a quantified comprehensive weight coefficient, providing a reliable basis for subsequent risk assessment and early warning.

[0052] Figure 3It is a schematic diagram of the steps of risk assessment and weight calculation based on the hierarchical analysis method, shown in an exemplary embodiment of the present application.

[0053] In a specific embodiment of the present application, Figure 3 As shown in the figure, in the process of risk assessment and weight calculation based on the Analytic Hierarchy Process (AHP), the assessment objectives are first clarified and a complete indicator system is constructed. The overall risk is divided into multiple first-level dimensions (such as equipment safety, process safety, and environmental safety). Each dimension is further refined into second-level indicator groups and third-level sub-indicators, forming a hierarchical structure. Subsequently, through expert scoring or historical data analysis, indicators at each level are compared pairwise, and a judgment matrix is ​​constructed to quantify the relative importance of different indicators.

[0054] After constructing the judgment matrix, a normalization method is used to calculate the weight vector of each indicator, obtaining the local weight of each level indicator. To ensure the consistency of the judgment matrix and avoid subjective bias, a consistency test is required, including calculating the maximum eigenvalue, the consistency index CI, and the random consistency ratio CR. If the CR value is less than 0.1, the judgment matrix is ​​considered to have satisfactory consistency; otherwise, the judgment matrix needs to be adjusted until it meets the requirements.

[0055] Once the single-level ranking results pass the consistency check, the hierarchical ranking process continues, aggregating the local weights of each level layer by layer to ultimately obtain the overall comprehensive weight coefficient from the bottom indicator to the top. At this point, the hierarchical ranking results must be re-checked for consistency to ensure the logical rationality of the entire evaluation process and the reliability of the results.

[0056] Final output analysis results

[0057] It's important to note that the method's final output includes risk scores for each indicator and an overall risk score. It automatically labels items with negative scores and provides analysis of possible causes. It also uses visualization to present assessment results and supports a one-click traceability mechanism, allowing managers to quickly locate anomalies and take appropriate measures. The entire process embodies a unified approach of systematicity, scientificity, and operability, making it suitable for dynamic risk assessments in complex industrial scenarios.

[0058] In one embodiment of the present application, first, the safety issues of the area to be inspected (such as blast furnaces, converters, etc.) are decomposed into multiple independent risk factors, covering dimensions such as equipment, process, environment, and personnel. For example: the equipment dimension is decomposed into multiple risk factors such as equipment interlock commissioning rate, hidden danger elimination rate, and maintenance records; the process dimension is decomposed into multiple risk factors such as blast furnace top pressure stability and converter oxygen blowing volume fluctuation; the environment dimension is decomposed into multiple risk factors such as toxic gas concentration and dust concentration; the personnel dimension is decomposed into multiple risk factors such as continuous working hours and occupational health examination rate. Subsequently, a hierarchical structure is constructed based on the influence and dominance relationship of each risk factor. Taking the blast furnace area as an example, the target layer is "overall safety assessment", the criterion layer includes four dimensions: equipment, process, environment, and personnel, and the indicator layer is the above-mentioned specific risk factors (such as top pressure stability, dust concentration, etc.).

[0059] Then, through a combination of expert evaluation and data analysis, the risk factors at the criterion and indicator levels were compared pairwise. For example, five domain experts were invited to rate the importance of the four dimensions of equipment, process, environment, and personnel (using a 1-9 scale). Assuming that the experts generally believe that equipment factors are more important than process factors (a score of 3) and environmental factors are more important than personnel factors (a score of 5), the criterion-level judgment matrix is ​​constructed as: [1, 3, 4, 6] [1 / 3, 1, 2, 4] [1 / 4, 1 / 2, 1, 2] [1 / 6, 1 / 4, 1 / 2, 1]. Furthermore, by calculating the eigenvectors, the criterion-level weight vector is obtained as [0.42, 0.28, 0.18, 0.12] [0.42, 0.28, 0.18, 0.12], and a consistency test is performed (CR = 0.05 < 0.1, passing the test). Furthermore, within the equipment dimension, the three indicators of "interlocking commissioning rate," "hazard elimination rate," and "maintenance record" were compared pairwise to construct a judgment matrix and calculate weights. For example, the weight of interlocking commissioning rate was 0.55, the weight of hazard elimination rate was 0.30, and the weight of maintenance record was 0.15.

[0060] Next, the weight values ​​of different levels are passed layer by layer to calculate the comprehensive weight coefficient of each risk factor's impact on security. For example:

[0061] Equipment dimension indicators: The interlocking utilization rate index value is expressed as the criterion layer equipment weight (0.42) × indicator layer weight (0.55) = 0.231; the hidden danger elimination rate index value is expressed as 0.42 × 0.30 = 0.126; the maintenance record index value is expressed as 0.42 × 0.15 = 0.063.

[0062] Process dimension indicators: The top pressure stability index value is expressed as the criterion layer process weight (0.28) × the index layer weight (0.60) = 0.168; the oxygen blowing amount fluctuation index value is expressed as 0.28 × 0.40 = 0.112.

[0063] Environmental dimension indicators: The dust concentration index value is expressed as the criterion layer environmental weight (0.18) × the index layer weight (0.70) = 0.126; the toxic gas concentration index value is expressed as 0.18 × 0.30 = 0.054.

[0064] Personnel dimension indicators: The continuous working hours index value is expressed as the criterion layer personnel weight (0.12) × the indicator layer weight (0.65) = 0.078; the physical examination rate index value is expressed as 0.12 × 0.35 = 0.042.

[0065] Finally, based on the above calculations, the combined weight coefficients of each risk factor were ranked as follows: interlock utilization rate (0.231) > top pressure stability (0.168) > dust concentration (0.126) > hidden danger elimination rate (0.126) > continuous working hours (0.078) > oxygen blowing rate fluctuation (0.112) > toxic gas concentration (0.054) > physical inspection rate (0.042). It should be noted that this result indicates that the equipment interlock utilization rate and process top pressure stability are the most critical factors affecting blast furnace area safety and require priority monitoring and intervention.

[0066] In addition, in another embodiment of the present application, the weight matrix is ​​dynamically modified through expert feedback and historical accident case library data. For example, if an accident investigation finds that the "toxic gas concentration" is underestimated, the weight of this indicator under the environmental dimension is adjusted and the comprehensive weight coefficient is recalculated. At the same time, a cross-validation method (such as dividing the data set into a training set and a test set) is used to verify the stability of the model and ensure the rationality of the weight distribution.

[0067] It can be understood that through the above steps, the AHP method realizes the quantitative analysis from multidimensional raw data to key risk factors, provides a scientific basis for subsequent dynamic early warning and emergency response, and significantly improves the accuracy and foresight of safety assessment in steel production areas.

[0068] Step S240 , inputting the comprehensive weight coefficient into the preset hybrid model, and combining it with the predefined membership function to calculate the comprehensive security risk value of the area to be detected.

[0069] In one embodiment of the present application, calculating the comprehensive safety risk value of the area to be inspected includes: constructing a membership function based on historical data and expert knowledge before inputting the model; standardizing the indicator values ​​of each dimension, and inputting the standardized indicator values ​​of each dimension into the membership function to obtain the corresponding fuzzy risk value; multiplying the fuzzy risk value with the corresponding comprehensive weight coefficient to generate a weighted risk value of each dimension; inputting the weighted risk value of each dimension into a preset probability model for fusion calculation, and outputting the comprehensive safety risk value of the area to be inspected.

[0070] In one embodiment of the present application, a quantitative assessment of the comprehensive safety risk value of a steel production area is achieved by constructing a membership function, standardizing indicators, weighted fusion of fuzzy risk values, and integrating probability models. First, a membership function is constructed based on historical data and expert knowledge, and the safety risk is divided into four levels: high, relatively high, general, and low. The fuzzy membership of each indicator is quantified by a linear or trapezoidal function (such as when the ambient temperature exceeds 50°C, the membership is 1). Subsequently, multi-dimensional indicators such as equipment status, process operation, and operating environment are standardized, and the membership function is input after eliminating dimensional differences to calculate the corresponding fuzzy risk value (such as the fuzzy value of 85% equipment interlocking utilization rate is 0.6). Combined with the comprehensive weight coefficient calculated by the analytic hierarchy process (AHP) (such as the equipment dimension weight 0.4), the weighted risk value of each dimension is generated (such as the equipment dimension weighted risk value = 0.4×0.6 = 0.24). Finally, the weighted risk values ​​are fed into a probabilistic model, such as a Bayesian network, for fusion calculations. The system outputs a comprehensive regional safety risk value (ranging from 0 to 100) and maps it to a four-dimensional risk level spectrum (red 0-40 for high risk, orange 40-60 for medium risk, yellow 60-80 for low risk, and green 80-100 for safe). By dynamically adjusting membership function parameters, optimizing the weight matrix, and implementing a closed-loop feedback mechanism, the system achieves real-time risk analysis and precise early warning, supporting thermal rendering of digital twin models and the generation of emergency response strategies.

[0071] In one embodiment of the present application, first, based on the method of the aforementioned embodiment, the risk factors of the criterion layer and the indicator layer are compared pairwise (1-9 scaling method), a judgment matrix is ​​constructed, and the weights are calculated. For example, the weight of the equipment dimension is 0.42, the process dimension is 0.28, the environment dimension is 0.18, and the personnel dimension is 0.12. The weights are transferred layer by layer to calculate the comprehensive weight coefficient of each risk factor. For example, the comprehensive weight of the "interlocking utilization rate" under the equipment dimension is 0.231 (criterion layer weight 0.42 × indicator layer weight 0.55), which becomes a key factor affecting safety.

[0072] Then, by combining fuzzy logic with probability theory, a dynamic assessment of non-deterministic risks can be achieved. The specific steps are as follows:

[0073] Step S241: Based on historical data and expert knowledge, a triangular or trapezoidal membership function is designed for each indicator. Assume that the membership function of the ambient temperature is defined as follows:

[0074]

[0075] Furthermore, the original index value of the ambient temperature is mapped to the fuzzy membership degree (0-1 interval) according to formula (1).

[0076] Step S242: Substitute the normalized index value into the membership function and calculate the weighted fuzzy risk value by combining it with the AHP comprehensive weight. For example, the dust concentration fuzzy value is 0.65×weight 0.126=0.082.

[0077] In step S243, all weighted fuzzy risk values ​​are accumulated to obtain the comprehensive safety risk value R of the area to be inspected. For example, if R = 0.686, then according to the four-dimensional risk level classification (red [0, 40], orange [40, 60], yellow [60, 80], green [80, 100]), it is determined to be yellow (low risk).

[0078] Finally, using historical risk score sequences (including real-time values, volatility analysis results, and fuzzy logic assessment results), future risk trends are predicted through autoregression (AR), differencing (I), and moving average (MA). For example, the parameters p = 2, d = 1, and q = 1 are determined using the ACF / PACF graph, and the model is trained and dynamically adjusted using Python's statsmodels library. The comprehensive risk value is then projected onto the digital twin model in real time, using a thermal rendering engine for gradient coloring. For example, the risk value for the blast furnace area is 0.686 → yellow (low risk), and the risk value for the gas tank area is 0.92 → red (high risk). By annotating the three-dimensional model and overlaying the thermal map, managers can intuitively identify high-risk areas.

[0079] It can be understood that the method proposed based on the above embodiment realizes the closed-loop management of the entire process from multi-dimensional data collection to dynamic risk prediction, visualization and emergency response, which significantly improves the foresight, accuracy and response efficiency of risk warning in steel production areas.

[0080] Step S250: triggering an early warning signal and executing an emergency response strategy based on the comparison result between the risk value and the preset threshold.

[0081] In one embodiment of the present application, a warning signal is triggered and an emergency response strategy is executed based on the comparison result between the risk value and the preset threshold, including: judging the current risk level based on the comparison result, and triggering a gradient emergency strategy corresponding to the current risk level, the gradient emergency strategy including at least a first-level emergency disposal strategy, a second-level area control strategy, and a third-level preventive maintenance strategy; if the current risk level is a high-risk level, the first-level emergency disposal strategy is executed, including triggering an emergency shutdown operation of key equipment, starting a personnel evacuation broadcast and alarm mechanism, and activating real-time monitoring of risk areas and enhanced information display; if the current risk level is a medium-risk level, the second-level area control strategy is executed, including implementing operation restrictions in high-risk areas, sending alarm information to relevant personnel, and enhancing monitoring of key areas; if the current risk level is a low-risk level, the third-level preventive maintenance strategy is executed, including generating an equipment maintenance work order, generating an equipment maintenance plan based on risk and status data, and pushing the equipment maintenance work order and the equipment maintenance plan to maintenance personnel.

[0082] In one embodiment of the present application, the results of risk prediction are divided into a four-dimensional risk level spectrum: red corresponds to a high-risk state in the range of [0,40], orange corresponds to a medium-risk state in the range of [40,60], yellow corresponds to a low-risk state in the range of [60,80], and green corresponds to a safe state in the range of [80,100]; an integrated spatial mapping algorithm projects the grading results to a digital twin model in real time, and a thermal rendering engine is used to realize the gradient coloring expression of the risk field.

[0083] In a specific embodiment of the present application, when a comprehensive safety risk value R∈[0,40] (red high risk) of a certain area is detected, a first-level emergency response strategy is immediately executed:

[0084] An emergency shutdown command is sent to the PLC controller via industrial IoT protocols (such as Modbus / TCP), shutting down critical equipment such as the blast furnace's top pressure regulating valve and gas pipelines to prevent the accident from escalating. An industrial sound system (supporting the GBT26875 protocol) is activated to broadcast evacuation instructions (e.g., "Gas leak in the blast furnace area, please evacuate immediately!") around high-risk areas. Three-dimensional annotations (flashing red) on the digital twin model guide personnel on evacuation routes. A UWB positioning system (accuracy ±10cm) is combined with smart helmets / smart bracelets to track the location of personnel within the area in real time. If a person is found stranded, the system automatically sends an escape route to the terminal device (e.g., AR glasses prompt "Evacuate to the north safety passage") and triggers an audible and visual alarm (frequency >3kHz) via a Bluetooth beacon.

[0085] In another specific embodiment of the present application, when a comprehensive security risk value R∈[40,60] (orange medium risk) of a certain area is detected, the secondary area control strategy is immediately executed:

[0086] Through industrial automation systems (such as SCADA), operating permissions in high-risk areas are locked, and non-essential operations (such as adjusting the oxygen blowing rate of the converter) are prohibited. At the same time, an electronic fence (triggered by UWB positioning) is set up in the digital twin model to prevent people from entering the danger radius (such as 30 meters around the gas tank); the alarm method is automatically matched according to the risk level, including but not limited to sending customized alarm information to the on-duty engineer in the form of SMS / phone (such as "The top pressure fluctuation in the converter area is abnormal, please check immediately!"), and playing voice prompts in medium-risk areas (such as "Please wear a gas mask and increase ventilation"); and retrieving multi-angle monitoring images of high-risk points (supporting RTSP protocol) and pushing them to managers in real time through the web or mobile terminal to assist in deciding whether to escalate the response.

[0087] In another specific embodiment of the present application, when a comprehensive safety risk value R∈[60,80] (yellow low risk) of a certain area is detected, a three-level preventive maintenance strategy is immediately executed:

[0088] Push risk factors (such as "dust concentration approaches the threshold") to the ERP system, generate maintenance work orders and assign them to responsible teams (such as "dust removal system maintenance"); optimize maintenance cycles based on historical maintenance records and equipment life data through reinforcement learning algorithms (such as DQN). For example, if the blast furnace cooling water flow volatility increases, shorten the water pump maintenance cycle by 10% in advance; use the ARIMA model to predict the risk score for the next 3 hours (such as R pred =75), if there is a trend towards the medium-risk range (>60), maintenance recommendations will be pushed in advance (e.g., "It is recommended to complete the dust collector filter replacement before 18:00 today").

[0089] Furthermore, in the three-level response strategy proposed in the above example, the alarm system for the steel production risk dynamics employs the NSGA-II multi-objective optimization algorithm to dynamically balance emergency response speed, resource consumption, and production losses. Specifically, the first-level response (high risk) prioritizes personnel safety (weight 0.6), enabling rapid intervention through emergency shutdowns, evacuation broadcasts, and personnel location alerts. The second-level response (medium risk) prioritizes equipment protection (weight 0.3), minimizing equipment damage through restricted access permissions, area management, and video linkage. The third-level response (low risk) prioritizes cost control (weight 0.1), implementing preventive maintenance through the generation of repair work orders and optimized maintenance plans. Based on this, the digital twin model uses a WebGL engine (Three.js) to implement real-time gradient coloring (red → orange → yellow → green) of the risk field, refreshing every 5 seconds. This allows managers to click on highlighted areas to view detailed scores for specific risk factors (e.g., "top pressure stability score 0.82"). If an accident occurs in a low-risk area (yellow) due to untimely maintenance, the system will automatically mark the historical data of this area as a "misjudgment case" and adjust the AHP weight matrix based on the expert correction annotation (for example, reducing the weight of "dust concentration" by 0.05), ultimately triggering a model version update to iteratively optimize the evaluation logic.

[0090] It can be understood that the method proposed based on the above embodiment has achieved closed-loop management of the entire process from risk identification to emergency response, ensuring millisecond-level intervention in high-risk events, precise control of medium-risk events, and active prevention of low-risk events, significantly improving the safety management efficiency and accident avoidance capabilities of steel production areas.

[0091] Step S260: After executing the response strategy, adjust the model parameters based on the verification data set and implement full lifecycle management through time series version comparison.

[0092] In one embodiment of the present application, after executing the corresponding emergency response strategy, the method also includes: verifying the prediction results of the preset hybrid model in different scenarios based on multiple verification data sets to obtain verification results; adjusting the preset risk level threshold and the weight coefficient of each dimensional indicator based on the verification results, and simulating the impact of each parameter change on the model output to identify key risk factors; correcting the annotation of the verification data set based on the key risk factors, and aligning the corrected annotation results with the historical alarm records to optimize the comprehensive weight coefficient; realizing the full life cycle management of the model configuration through time series version comparison analysis, and adjusting the scoring rules to optimize the preset hybrid model.

[0093] Among them, the full life cycle management of model configuration is achieved through time series version comparison analysis, including: comparing the differences in model parameters in different periods and identifying key change points. Model parameters include weight coefficients, risk level thresholds, and scoring rules; updating model configuration according to the results of difference analysis, generating a new model version and deploying it to the production environment to continuously optimize the preset hybrid model.

[0094] In one embodiment of the present application, full life cycle management includes: retrospective verification of the warning results based on the verification data set, and generating a verification report including the false alarm rate and the missed alarm rate; dynamically adjusting the risk level threshold and the weight coefficient of each dimension based on the false alarm / missing alarm data analysis in the verification report; inputting the adjusted parameters into the hybrid model to simulate the impact of parameter changes on the risk value output, and identifying key risk factors whose sensitivity to the warning results exceeds the preset value; for the identified key factors, comparing the differences in historical version model parameters, locating the change nodes of weight coefficients, risk thresholds and scoring rules; updating the model configuration based on the difference analysis results, generating a new version model for deployment to the production environment, and collecting the deployed warning data as the verification data set for the next cycle.

[0095] In a specific embodiment of the present application, the full life cycle management generates false alarm rate and missed alarm rate reports through retrospective verification of the verification data set, dynamically adjusts the risk level threshold and the weight coefficients of each dimension, and uses sensitivity analysis to identify key risk factors whose sensitivity to warning results exceeds the preset value; based on the differences in historical version model parameters, the change nodes of weight coefficients, risk thresholds and scoring rules are located, and the model configuration is updated and deployed to the production environment, and new warning data is collected as the verification data set for the next cycle to form a closed-loop iteration. The specific process includes: calculating the false alarm rate (FAR) and missed alarm rate (MAR) based on the validation data set, and generating a validation report containing typical cases; adjusting the threshold (such as reducing the red alert threshold from 40 to 35) and weight (such as increasing the gas concentration weight from 0.2 to 0.25) according to the report; quantifying the impact of parameter changes on risk values ​​through sensitivity analysis (such as changes in gas concentration weights leading to an 8% increase in the overall risk value) and identifying key factors; comparing historical version model parameter differences (such as Git version control records) to locate change nodes (such as the weight adjustment event in March 2025); finally, updating the model configuration and deploying it to the production environment, collecting new warning data for continuous optimization, and realizing dynamic adaptation of model parameters to production scenarios.

[0096] In one embodiment of the present application, the dynamic early warning method for steel production risks proposed in the present application also includes constructing a time series prediction model based on historical risk data, specifically including: multi-source integration of risk values, volatility analysis results and risk assessment results generated in real time to generate a unified time series data set; constructing a prediction model based on the time series data set, dynamically adjusting model parameters according to data characteristics, and improving prediction accuracy through an indicator optimization mechanism; using the prediction model to output risk trend values ​​for future time periods, generating a warning signal when the risk trend value exceeds a preset risk threshold, and linking the execution of a gradient emergency strategy.

[0097] In a specific embodiment of the present application, the process of implementing the risk situation prediction function includes three stages: multi-source data fusion processing, dynamic modeling and parameter optimization, and advanced warning linkage response. First, the security risk value, volatility analysis results, and risk assessment results generated by the risk dynamic assessment module are integrated in real time. The three types of data are standardized and converted, where the risk value is normalized using the interval compression method, the volatility results are converted using a nonlinear function, and the risk assessment results are mapped according to preset levels. The processed data is weighted and fused based on the dynamic weight coefficient to construct a unified time series input set.

[0098] During the modeling and optimization phase, the system initializes the time series forecasting model, configures a rolling time window, and loads a recent historical dataset. The model continuously monitors changes in data features and automatically adjusts the differencing order to eliminate trend fluctuations when non-stationary trends are identified. The autoregressive order is determined through residual correlation analysis, and the sliding average order is dynamically optimized based on real-time forecast errors. Accuracy assessments are performed regularly. If the forecast error is detected to have increased continuously beyond a set threshold, a parameter reset and re-optimization mechanism is triggered to ensure the model continues to adapt to evolving production conditions.

[0099] After the model outputs the risk trend curve for the future period, it compares it to the preset risk level threshold in real time. When the predicted value exceeds the warning threshold, the advance warning is immediately activated and the gradient emergency strategy is linked: the corresponding regional control measures are implemented, operations in the risk area are automatically restricted, alarm instructions are pushed to the responsible personnel, and the monitoring screen of the target point is retrieved simultaneously. At the same time, the digital twin traceability mechanism is activated, the predicted risk coordinates are marked in the three-dimensional model, and the historical process parameters are linked to locate potential abnormal sources (such as abnormal status of key equipment). The entire process from risk prediction to emergency response takes significantly less time than traditional methods, and the response efficiency is improved by orders of magnitude.

[0100] It is understandable that this embodiment achieves accurate perception and efficient handling of complex risk situations through dynamic fusion of multi-source data, adaptive parameter optimization, and closed-loop coordination of prediction and response. Specifically, by dynamically weighting heterogeneous data, dimensional differences are eliminated, thereby improving the comprehensiveness and accuracy of predictions; by optimizing the differential order and sliding average parameters in real time, the model is ensured to adapt to complex working conditions with high precision over the long term; by combining gradient emergency strategies with digital twin traceability, proactive risk intervention and precise location of abnormal sources are achieved; in addition, through full-link automation of prediction-response-optimization, response speed and resource scheduling efficiency are improved by orders of magnitude.

[0101] In one embodiment of the present application, after executing the emergency response strategy, the preset hybrid model is continuously iterated and upgraded by building a multi-dimensional verification and optimization mechanism to improve its prediction accuracy and stability in complex scenarios. The specific process is as follows:

[0102] Based on multiple validation datasets (covering different time periods, geographical distributions, and risk types), a time series segmentation method was used for cross-validation to evaluate the model's recall and false alarm rates at low, medium, and high risk levels. Based on the validation results, the risk level thresholds (e.g., optimizing the "high risk" threshold from 0.85 to 0.82) and the weighting coefficients of various dimensional indicators (e.g., increasing the weight of the "equipment aging index" to 0.35) were dynamically adjusted. Monte Carlo parameter perturbation experiments were used to simulate the impact of a ±10% change in the weight coefficient on the prediction results, identifying key risk factors (e.g., "cyber attack frequency" and "abnormal ambient temperature and humidity").

[0103] On this basis, the risk score of the verification data set is recalculated based on the adjusted weight coefficient, and multi-dimensional alignment is performed with the historical alarm records. The labeled data of events that are not warned by the model are corrected (such as supplementing the "ventilation system failure" related features). At the same time, the expert experience matrix is ​​introduced, and the contribution of each feature to the risk score is analyzed through the SHAP value, and the weight coefficient is dynamically adjusted (such as increasing the weight of "personnel illegal operation" to 0.25).

[0104] The full lifecycle management of model configuration is achieved through time series version comparison analysis, including: comparing model parameter differences in different periods (such as adjusting the "temperature fluctuation threshold" from ±2°C to ±1.5°C), using differential analysis tools to quantify the impact of parameter changes on model output, and generating difference reports; updating the model parameter configuration file (such as risk_model_v2.1.yaml) based on the analysis results, verifying the performance of the new version in the production environment through A / B testing (such as a 5% increase in prediction accuracy and an 8% decrease in false alarm rate), and deploying it to the online system.

[0105] At the same time, the optimized scoring rules (such as "when 'device vibration abnormality' lasts for 2 hours and 'current fluctuation' exceeds the threshold, a secondary warning is triggered") are solidified into decision tree rules, incorporated into the knowledge base system, and converted into an operation and maintenance manual through natural language processing technology.

[0106] Furthermore, during continuous monitoring, a model monitoring module is deployed to collect new event data (such as sensor alerts and user behavior logs) in real time. If the prediction deviation exceeds a MAPE of 15%, an automatic rollback mechanism is triggered to switch to a stable version. An online learning framework is built based on streaming data, and the ARIMA model is periodically and incrementally trained (for example, adjusting the differencing order d and the moving average term q) to adapt to changes in data distribution. This closed-loop process enables full-process management from emergency response to model optimization, ensuring the risk warning system's ability to rapidly respond to new threats and continuously evolve.

[0107] It is understood that the method proposed based on the above embodiment, through the fusion of multi-source heterogeneous data (such as real-time values, volatility analysis and fuzzy logic results) and the dynamic tuning mechanism of parameters, combined with the modeling ability of the ARIMA model for time series characteristics, significantly improves the accuracy of risk prediction. At the same time, the adaptability and robustness of the model to complex scenarios are enhanced by eliminating non-stationarity through differential elimination, suppressing noise through sliding average, and identifying key parameters through sensitivity analysis. In addition, with the help of cross-validation, annotation correction and expert experience closed-loop feedback, the model can dynamically adjust the weight coefficient and risk threshold to avoid performance degradation due to changes in data distribution or the emergence of new threats. In summary, the method proposed in this embodiment, through data-driven and expert experience fusion, dynamic parameter tuning and full life cycle management, constructs a high-precision and robust risk warning system, which significantly improves the risk management and control capabilities in complex environments.

[0108] In one embodiment of the present application, the dynamic early warning of steel production risks also includes risk response closed-loop management, specifically including: mapping the comprehensive safety risk value to the digital twin model of the production line in real time, and using the thermal rendering engine to create a risk gradient distribution map; when the early warning signal is triggered, automatically associating and activating the one-key traceability interface on the risk gradient distribution map, responding to the user's interactive operations, and realizing the following functions: according to the coordinates of the high-risk area in the risk gradient distribution map, reverse tracing to generate the risk data path of the area; tracing back along the hierarchy of the production process to locate the data source equipment and process parameters of the previous link that caused the abnormality; marking the three-dimensional position of the source of the abnormality in the digital twin model, and generating a traceability report at the same time.

[0109] In one embodiment of the present application, closed-loop management of dynamic early warning of steel production risks is achieved through a digital twin model and a thermal rendering engine. The system maps the comprehensive safety risk value to the digital twin model of the production line in real time, and generates a risk gradient distribution map through thermal rendering (red, orange, yellow, and green represent high risk to safe status). When the early warning signal is triggered, the system automatically associates the coordinates of the high-risk area in the risk gradient distribution map, activates the one-key traceability interface, reversely traces the risk data path generated in the area, and reversely analyzes along the production process level to locate the data source equipment and process parameters of the preceding link that caused the anomaly. The three-dimensional position of the source of the anomaly (such as a gas tank sensor) is marked in the digital twin model, and a traceability report containing the source of the anomaly, the scope of impact, the cause analysis, and the disposal suggestions is generated to support managers to quickly locate the problem and implement emergency response strategies, forming a closed-loop management link of "monitoring-assessment-early warning-disposal-optimization".

[0110] In a specific embodiment of the present application, the scoring of each indicator is quickly grasped through different visualization methods (such as charts, heat maps, trend charts, etc.). When the score of a sub-indicator is lower than the preset threshold, the loss item will be automatically marked and its possible reasons (such as equipment failure, improper operation or environmental abnormality, etc.) will be provided. This mechanism can display the evaluation results layer by layer according to the structure of hierarchical analysis, first displaying the most detailed single indicator (such as temperature, pressure and other data of a specific device), and then gradually summarizing it into indicator groups, dimensions and the final overall risk score. Through hierarchical display, users can clearly understand the details of the loss and quickly and accurately locate the abnormal points.

[0111] Finally, it should be noted that the dynamic early warning of steel production risks proposed in this application significantly improves the foresight of risk identification and the efficiency of emergency response through multi-dimensional data fusion and intelligent analysis. The method first ensures data stability and reliability through real-time collection and volatility analysis of multi-source heterogeneous data such as equipment status, process operation, working environment, and personnel behavior, combined with redundant verification and data repair mechanisms. Secondly, based on the hierarchical analysis method, each risk factor is compared pairwise to generate a weight matrix, and combined with fuzzy logic-probability theory hybrid modeling, the regional comprehensive safety risk value is quantified to achieve accurate traceability from single indicator anomaly to regional risk evolution. In terms of early warning and emergency response, the risk status is intuitively identified through a four-dimensional risk level spectrum (red, orange, yellow, green), triggering hierarchical emergency strategies (emergency disposal, regional control, preventive maintenance), and linking multimodal alarms (SMS, broadcast, video retrieval, personnel positioning), shortening the emergency response time, and forming a closed-loop management link. At the same time, the model dynamically optimizes parameter and weight configuration through cross-validation, sensitivity analysis and expert feedback closed loop, and realizes full life cycle management in combination with time series version comparison, continuously improving adaptability to complex production scenarios. Furthermore, the digital twin model integrates a thermal rendering engine, projecting risk field gradient shading in real time, helping managers quickly locate high-risk areas and transforming safety management from a "post-event" approach to a "proactive defense" approach. In summary, this method significantly reduces the rate of production safety accidents through a data-driven intelligent early warning system, providing technical support for steel companies to build a full-lifecycle risk prevention and control system, and possesses significant industrial application value.

[0112] Figure 4 This is a block diagram of a steel production risk dynamic early warning device shown in an exemplary embodiment of the present application. The device can be applied to Figure 1 The device may also be applicable to other exemplary implementation environments and specifically configured in other devices. This embodiment does not limit the implementation environment to which the device is applicable.

[0113] like Figure 4 As shown, the exemplary steel production risk dynamic early warning device includes: a data acquisition module 410, a weight configuration module 420, a risk assessment module 430, and an early warning response module 440.

[0114] Among them, the data acquisition module 410 is used to obtain the original data in the steel production process, and the original data includes data of multiple different data dimensions; the data preprocessing module 420 is used to perform volatility analysis and stability analysis on the original data, including: calculating the dynamic fluctuation range of indicators in each dimension, marking abnormal fluctuation points, and triggering the sensor redundancy check and data repair mechanism when the abnormal frequency per unit time exceeds the threshold; the risk assessment module 430 is used to compare the indicators in each dimension pairwise based on the analyzed data to generate a weight matrix and calculate the comprehensive weight coefficient; the comprehensive weight coefficient is input into the preset hybrid model, and the comprehensive safety risk value of the area to be detected is calculated in combination with the predefined membership function; the early warning response module 440 is used to trigger the early warning signal and execute the emergency response strategy according to the comparison result between the risk value and the preset threshold; the risk response closed-loop management module 450 is used to adjust the model parameters based on the verification data set after executing the response strategy and realize full life cycle management through time series version comparison.

[0115] It should be noted that the steel production risk dynamic early warning device provided in the above embodiment and the steel production risk dynamic early warning method provided in the above embodiment are based on the same concept. The specific manner in which each module and unit performs operations has been described in detail in the method embodiment and will not be repeated here. In actual applications, the steel production risk dynamic early warning device provided in the above embodiment can distribute the above functions to different functional modules as needed, that is, divide the internal structure of the device into different functional modules to complete all or part of the functions described above, and this is not limited here.

[0116] An embodiment of the present application also provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the electronic device implements the dynamic early warning method for steel production risks provided in the above-mentioned embodiments.

[0117] Figure 5 The following is a schematic diagram showing the structure of a computer system suitable for implementing an electronic device according to an embodiment of the present application. Figure 5 The computer system 500 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0118] like Figure 5As shown, the computer system 500 includes a central processing unit (CPU) 501, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 502 or the program loaded from the storage part 508 to the random access memory (RAM) 503, such as executing the method described in the above embodiment. Various programs and data required for system operation are also stored in the RAM 503. The CPU 501, ROM 502 and RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0119] The following components are connected to the I / O interface 505: an input section 506 including a keyboard, a mouse, and the like; an output section 507 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 508 including a hard disk; and a communication section 509 including a network interface card such as a LAN (Local Area Network) card or a modem. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to the I / O interface 505 as needed. Removable media 511, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 510 as needed, so that computer programs read therefrom can be installed into the storage section 508 as needed.

[0120] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 509, and / or installed from a removable medium 511. When the computer program is executed by the central processing unit (CPU) 501, the various functions defined in the system of the present application are executed.

[0121] It should be noted that the computer-readable medium shown in the embodiments of the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, which carries a computer-readable computer program. This propagated data signal can take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. A computer program embodied on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0122] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. Among them, each box in the flowchart or block diagram can represent a module, program segment, or part of the code, and the above-mentioned module, program segment, or part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0123] The units involved in the embodiments described in this application may be implemented by software or hardware, and the units described may also be set in a processor. In some cases, the names of these units do not constitute limitations on the units themselves.

[0124] Another aspect of the present application provides a computer-readable storage medium storing a computer program. When executed by a computer processor, the computer program causes the computer to execute the aforementioned dynamic early warning method for steel production risks. The computer-readable storage medium may be included in the electronic device described in the above embodiments, or may exist independently and not be incorporated into the electronic device.

[0125] Another aspect of the present application provides a computer program product or computer program, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to implement the steel production risk dynamic early warning method provided in each of the above embodiments.

[0126] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, any equivalent modifications or alterations accomplished by a person of ordinary skill in the art without departing from the spirit and technical concepts disclosed in this application shall be covered by the claims of this application.

Claims

1. A dynamic early warning method for steel production risks, characterized in that: The method comprises: Acquire raw data from a steel production process, wherein the raw data includes data from multiple different data dimensions; Performing volatility analysis and stability analysis on the raw data, including: calculating the dynamic fluctuation range of each dimensional indicator, marking abnormal fluctuation points, and identifying the stability of the raw data based on the abnormal fluctuation points within a unit time; Based on the analyzed data, the indicators of each dimension are compared pairwise to generate a weight matrix and calculate the comprehensive weight coefficient; The comprehensive weight coefficient is input into the preset hybrid model, and the comprehensive safety risk value of the area to be inspected is calculated in combination with the predefined membership function; Trigger early warning signals and execute emergency response strategies based on the comparison results of risk values ​​with preset thresholds; After executing the response strategy, the model parameters are adjusted based on the validation dataset and full life cycle management is achieved through time series version comparison.

2. The steel production risk dynamic early warning method according to claim 1, characterized in that: The volatility analysis and stability analysis include: Determine the standard fluctuation range of the original data based on the standard value and mean of the historical data; If any raw data exceeds the standard fluctuation range, it will be marked as an abnormal fluctuation point; Identifying abnormal fluctuation points within a preset unit time to obtain an abnormal fluctuation frequency, and if the abnormal fluctuation frequency exceeds a preset frequency threshold, determining that the original data is unstable; The result of the stability assessment is used to score the original data, adding points if stable and deducting points if unstable. The scoring result is used to adjust the parameters of the preset hybrid model when constructing the preset hybrid model.

3. The steel production risk dynamic early warning method according to claim 2, characterized in that: The method further includes constructing a time series prediction model based on historical risk data, specifically including: The real-time generated risk value, volatility analysis results and risk assessment results are integrated into a multi-source data set to generate a unified time series data set; Build a prediction model based on the time series data set, dynamically adjust model parameters according to data characteristics, and improve prediction accuracy through an indicator optimization mechanism; The prediction model is used to output the risk trend value of the future period. When the risk trend value exceeds the preset risk threshold, an early warning signal is generated, and a gradient emergency strategy is executed in conjunction.

4. The steel production risk dynamic early warning method according to claim 1, characterized in that: Generate a weight matrix and calculate the comprehensive weight coefficient of each dimension indicator, including: Decompose the safety issues of the area to be inspected into multiple independent risk factors; Arrange risk factors in a hierarchical manner according to their influence and dominance, forming a hierarchical structure; Compare risk factors at each level in pairs and convert the comparison results into a weight matrix, which is used to reflect the importance of each factor to safety impact; Calculating the consistency ratio of the weight matrix, and if the consistency ratio is greater than a preset threshold, re-adjusting the pairwise comparison scale of the risk factors and iteratively calculating until the consistency ratio does not exceed the preset threshold, the consistency ratio being determined by the ratio of the consistency index to the random consistency index; Combining the weight values ​​at different levels, a comprehensive weight coefficient of the impact of each risk factor on safety is generated.

5. The steel production risk dynamic early warning method according to claim 1, characterized in that: Calculating the comprehensive safety risk value of the area to be inspected includes: Construct membership functions based on historical data and expert knowledge before inputting into the model; Standardize the indicator values ​​of each dimension, and input the standardized indicator values ​​of each dimension into the membership function to obtain the corresponding fuzzy risk value; Multiplying the fuzzy risk value by the corresponding comprehensive weight coefficient to generate a weighted risk value for each dimension; The weighted risk values ​​of each dimension are input into the preset probability model for fusion calculation, and the comprehensive safety risk value of the area to be inspected is output.

6. The steel production risk dynamic early warning method according to claim 1, characterized in that: The method also includes risk response closed-loop management, specifically including: Map comprehensive safety risk values ​​to the digital twin model of the production line in real time, and use a thermal rendering engine to create a risk gradient distribution map; When an early warning signal is triggered, the one-key traceability interface is automatically associated and activated on the risk gradient distribution map, responding to user interactions to achieve the following functions: According to the coordinates of the high-risk area in the risk gradient distribution map, reverse tracing is performed to generate the risk data path of the area; Trace back along the production process hierarchy to locate the data source equipment and process parameters of the previous link that caused the anomaly; The three-dimensional location of the anomaly source is marked in the digital twin model, and a traceability report is generated at the same time.

7. The steel production risk dynamic early warning method according to claim 1, characterized in that: Based on the comparison between the risk value and the preset threshold, an early warning signal is triggered and an emergency response strategy is implemented, including: Determine the current risk level based on the comparison result, and trigger a gradient emergency response strategy corresponding to the current risk level, wherein the gradient emergency response strategy includes at least a first-level emergency response strategy, a second-level regional control strategy, and a third-level preventive maintenance strategy; If the current risk level is high, the first-level emergency response strategy is implemented, including triggering emergency shutdown of key equipment, initiating personnel evacuation broadcast and alarm mechanisms, and activating real-time monitoring of risk areas and enhanced information display; If the current risk level is medium, a secondary area control strategy will be implemented, including restricting operations in high-risk areas, sending warnings to relevant personnel, and increasing monitoring of key areas. If the current risk level is a low risk level, a three-level preventive maintenance strategy is implemented, including generating an equipment maintenance work order, generating an equipment maintenance plan based on risk and status data, and pushing the equipment maintenance work order and the equipment maintenance plan to maintenance personnel.

8. The steel production risk dynamic early warning method according to any one of claims 1 to 7, characterized in that: The full life cycle management includes: Back-test the warning results based on the validation data set and generate a validation report including false positive rate and false negative rate; Dynamically adjust the risk level threshold and weight coefficients of each dimension based on the analysis of false positive / missing negative data in the verification report; Input the adjusted parameters into the hybrid model to simulate the impact of parameter changes on the risk value output and identify key risk factors whose sensitivity to early warning results exceeds the preset value; For the identified key factors, compare the differences in model parameters of historical versions to locate the change nodes of weight coefficients, risk thresholds and scoring rules; Update the model configuration based on the results of the difference analysis, generate a new version of the model and deploy it to the production environment, and collect the early warning data after deployment as the verification data set for the next cycle.

9. A dynamic early warning device for steel production risks, characterized in that: The device comprises: A data acquisition module is used to obtain raw data from the steel production process, wherein the raw data includes data of multiple different data dimensions; A data preprocessing module is used to perform volatility analysis and stability analysis on the raw data, including: calculating the dynamic fluctuation range of each dimensional indicator, marking abnormal fluctuation points, and identifying the stability of the raw data based on the abnormal fluctuation points within a unit time; The risk assessment module is used to compare the indicators of each dimension pairwise based on the analyzed data to generate a weight matrix and calculate the comprehensive weight coefficient. The comprehensive weight coefficient is input into the preset hybrid model and combined with the predefined membership function to calculate the comprehensive safety risk value of the area to be inspected. The early warning response module is used to trigger early warning signals and execute emergency response strategies based on the comparison results of risk values ​​and preset thresholds; The risk response closed-loop management module is used to adjust model parameters based on the verification data set after executing the response strategy and to achieve full life cycle management through time series version comparison.

10. An electronic device, characterized in that: The electronic device comprises: one or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, enables the electronic device to implement the dynamic early warning method for steel production risks as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed by a processor of a computer, the computer is caused to execute the dynamic early warning method for steel production risks according to any one of claims 1 to 8.

Images