A dynamic assessment and early warning system for safety risk of chemical process

Abstract

The application belongs to the technical field of chemical safety, and discloses a kind of chemical process safety risk dynamic evaluation and early warning system;Through obtaining real-time process parameter deviation, depth frequency spectrum feature extraction is carried out, quantitative evaluation of deviation is realized through critical deviation calculation, and process parameter anomaly is accurately divided into fluctuation type and trend type based on deviation classification determination module.The application adopts differentiated baseline updating strategy for different deviation types, and the fluctuation type deviation is moderately included in the baseline update to adapt to normal process evolution, and the trend type deviation is excluded from the baseline update and accumulated evaluation is carried out through risk integration mechanism.The application can sensitively capture progressive risk signals, while avoiding false alarms caused by working condition changes, and realizes the unity of high sensitivity and high reliability of safety monitoring.

Description

Technical Field

[0001] This invention relates to the field of chemical safety technology, and more specifically, to a dynamic assessment and early warning system for chemical process safety risks. Background Technology

[0002] Existing chemical process safety risk assessment systems suffer from fundamental technical problems, making it difficult to effectively cope with the complex and dynamic changes in industrial production environments. Traditional fixed threshold models (high-high limit / low-low limit) are too rigid and cannot adapt to natural fluctuations in process parameters and changes in production conditions, leading to frequent false alarms or missed hazard warnings in practical applications. Emerging adaptive threshold algorithms, on the other hand, fall into a "response paradox": their baseline adjustment faces an irreconcilable speed contradiction. Taking polymerization reactor production as an example, when a new batch of catalyst is introduced, the reaction baseline temperature naturally rises but remains within the safe range. The system either becomes oversensitive (stubbornly treating normal process fluctuations as abnormal and continuously triggering alarms, leading to operator alarm fatigue and misjudgments) or falls into a risk assimilation trap (rapidly "learning" the rising temperature as the new baseline, and then continuously assimilating the slowly rising temperature anomaly). This gradual risk accumulation is particularly dangerous because it often approaches the safety threshold before being detected, while the system consistently treats it as a "normal fluctuation" and continuously updates the baseline. Secondly, existing technologies lack a deep understanding and classification ability regarding the nature of deviations, failing to distinguish between normal fluctuations stemming from batch variations and dangerous trends arising from equipment malfunctions. This results in a lack of targeted baseline update strategies. In modern chemical production environments with frequent changes in operating conditions and significant fluctuations in raw material characteristics, this technological deficiency has become a fatal weakness in safety monitoring, harboring significant safety hazards. The system cannot discern when to adhere to baseline standards and when to make flexible adjustments, ultimately leading either to low production efficiency due to over-vigilance or to safety risks due to blind adaptation, making it difficult to achieve a balance between process safety and production efficiency.

[0003] In view of this, the present invention proposes a dynamic assessment and early warning system for chemical process safety risks to solve the above problems. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a dynamic assessment and early warning system for safety risks in chemical processes, comprising:

[0005] The deviation calculation module is used to obtain the process parameter values ​​at each sampling time in the chemical production process, and calculate the difference between the process parameter value at each sampling time and the current dynamic baseline value as the instantaneous deviation.

[0006] The complex domain feature analysis module is used to construct a deviation sequence from the instantaneous deviations at each sampling time within the preset analysis window, and to perform complex domain feature transformation on the deviation sequence to obtain a set of complex feature coefficients.

[0007] The critical deviation calculation module is used to calculate the vertical distance between the distribution position of each complex characteristic coefficient in the complex characteristic coefficient set on the complex plane and the preset critical zone, and to use the weighted average of all vertical distances as the critical deviation at the current sampling time.

[0008] The deviation classification and determination module is used to determine the deviation type of the current deviation sequence based on the critical deviation degree. The deviation types include fluctuation deviation and trend deviation.

[0009] The deviation decomposition module is used to decompose the deviation sequence into mutually orthogonal fluctuation components and trend components.

[0010] The dynamic baseline update module is used to determine the baseline absorption rate of the fluctuation component based on the deviation type, and to include the fluctuation component in the dynamic baseline update according to the baseline absorption rate; while excluding the trend component from the dynamic baseline update.

[0011] The risk integral calculation module is used to accumulate the trend components at each sampling time, determine the cumulative weight coefficient based on the critical deviation, and perform a weighted summation of the accumulated trend components to obtain the risk integral value at the current sampling time.

[0012] The early warning output module is used to compare the risk score with the graded early warning threshold and output the corresponding level of early warning signal based on the comparison result.

[0013] The technical effects and advantages of the dynamic assessment and early warning system for chemical process safety risks of this invention are as follows:

[0014] This invention can accurately distinguish between normal process fluctuations and potential risk trends, effectively eliminating alarm fatigue and improving the reliability and response efficiency of early warnings. In complex production scenarios with frequent changes in operating conditions, this invention can quickly adapt to reasonable process changes while also keenly capturing weak signals of hidden risks, giving operators time to respond. Its early identification capability for the accumulation of gradual risks fills the fatal blind spots of traditional monitoring systems, preventing potential major safety accidents. This invention reduces the frequency of shutdowns for maintenance and losses from unexpected shutdowns, and further realizes a paradigm shift in safety management from passive response to proactive prevention. In modern chemical production with fluctuating raw material batches and complex process conditions, this invention helps operators make more timely and accurate decisions in complex and ever-changing production environments, achieving an optimal balance between safety and efficiency. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of a dynamic assessment and early warning system for chemical process safety risks according to the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] This application provides a dynamic assessment and early warning system for chemical process safety risks. The system's execution entities include, but are not limited to, those mounted on the system: a process safety monitoring platform, a chemical production control center, a real-time early warning analysis system, and a process parameter monitoring platform, which can be considered general computing nodes in this application. The early warning system includes, but is not limited to, at least one of: a cloud-based risk analysis engine, a distributed parameter monitoring system, and an intelligent deviation detector.

[0018] Please see Figure 1 In this embodiment of the invention, a dynamic assessment and early warning system for chemical process safety risks includes:

[0019] The deviation calculation module acquires process parameter values ​​at various sampling times during chemical production and calculates the difference between each sampling time's process parameter value and the current dynamic baseline value as the instantaneous deviation. This module acquires real-time measurements of key production indicators such as temperature, pressure, flow rate, and concentration from chemical process parameters via a multi-channel data acquisition interface. The system first establishes an initial baseline value, typically determined based on statistical analysis of historical stable operating data, and then continuously updates it to a dynamic baseline during operation. For each newly acquired sampling point, the system calculates its difference from the current dynamic baseline to obtain the instantaneous deviation value. These instantaneous deviations constitute the foundational data for subsequent analysis, reflecting the real-time deviation status of the process and providing initial signals for risk assessment.

[0020] The complex-domain feature analysis module is used to construct a deviation sequence from the instantaneous deviations at each sampling time within a preset analysis window, and then perform a complex-domain feature transformation on the deviation sequence to obtain a set of complex feature coefficients. This module first collects continuous instantaneous deviation values ​​within a sliding time window to form a deviation sequence. Subsequently, the system converts these time-domain signals to the complex domain for analysis, extracting deeper frequency characteristics and phase information. Complex-domain analysis can reveal periodic patterns and trend features that are difficult to detect in time-domain analysis, providing a more comprehensive feature description of abnormal behavior of process parameters. The complex feature coefficients reflect the energy distribution and phase relationship of the deviation sequence at different frequency components, providing important input for subsequent critical deviation calculations.

[0021] The critical deviation calculation module calculates the vertical distance between the distribution position of each complex characteristic coefficient in the complex plane and a preset critical band, and uses the weighted average of all vertical distances as the critical deviation at the current sampling time. This module first defines a specific critical band region on the complex plane as a baseline for judging the nature of the deviation. The system calculates the vertical distance of each complex characteristic coefficient to this critical band and performs a weighted average based on the importance of different frequency components to obtain a comprehensive critical deviation index. This index quantifies the severity and nature of the current deviation state, providing a quantitative basis for deviation classification. The larger the critical deviation, the more severe the deviation of the process parameters from normal operating conditions, and the higher the potential safety risk.

[0022] The deviation classification and determination module is used to determine the deviation type of the current deviation sequence based on the critical deviation degree. Deviation types include fluctuation-type deviations and trend-type deviations. Based on the magnitude of the critical deviation degree and the distribution characteristics of the complex characteristic coefficients, this module classifies process parameter deviations into two basic types. Fluctuation-type deviations typically manifest as random fluctuations in parameter values ​​near the baseline, generally caused by system noise, environmental interference, or small-amplitude disturbances, and have limited impact on the production process. Trend-type deviations, on the other hand, manifest as unidirectional changes in parameter values ​​that continuously deviate from the baseline, potentially indicating safety risks such as equipment failure, raw material changes, or abnormal process conditions. Accurate determination of the deviation type is a crucial prerequisite for dynamic baseline updates and risk assessment, directly affecting the system's early warning decisions.

[0023] The deviation decomposition module decomposes the deviation sequence into mutually orthogonal fluctuation and trend components. This module uses mathematical methods to separate the mixed deviation signal into independent fluctuation and trend parts for separate processing. The fluctuation component typically reflects random fluctuations and high-frequency disturbances in the system, while the trend component reflects the continuous direction of change in the system state. The decomposition process is based on least squares fitting and orthogonalization to ensure no correlation between the two components, achieving an accurate characterization of the deviation signal. This decomposition method enables the system to distinguish deviation components of different natures, allowing for targeted baseline updates and risk assessment.

[0024] The dynamic baseline update module determines the baseline absorption rate of fluctuation components based on the deviation type, incorporating these components into the dynamic baseline update according to the absorption rate, while excluding trend components. This module enables intelligent adaptive adjustment of the baseline, crucial for the long-term stable operation of the system. For fluctuation-type deviations, the system partially incorporates them into the baseline update, allowing the baseline to adapt to normal fluctuations and slow changes in the process. For trend-type deviations, the system excludes them from the baseline update, preventing potential risks from being incorrectly "learned" into the baseline. The baseline absorption rate is dynamically adjusted based on the deviation type and operating conditions, achieving intelligent control of the baseline update. This allows the system to adapt to normal process evolution while sensitively detecting abnormal trends.

[0025] The risk integral calculation module accumulates the trend components at each sampling time, determines the cumulative weighting coefficient based on the critical deviation, and then sums the accumulated trend components to obtain the risk integral value at the current sampling time. This module transforms discrete trend signals into continuous risk assessment indicators through time accumulation. The system dynamically adjusts the cumulative weighting coefficient based on the critical deviation, giving higher weight to trends that deviate continuously in the same direction, thus enhancing sensitivity to continuous anomalies. The risk integral mechanism considers the direction, duration, and severity of the deviation, comprehensively assessing the cumulative impact of process parameter deviations and providing a quantitative basis for early warning decisions.

[0026] The early warning output module compares the risk score with the tiered early warning thresholds and outputs an early warning signal of the corresponding level based on the comparison result. This module is the final output stage of the system, transforming complex analysis results into clear decision recommendations. The system sets multiple early warning thresholds, from low to high, corresponding to different levels of severity. When the risk score exceeds the corresponding threshold, the system outputs an early warning signal of the corresponding level, reminding operators to take appropriate intervention measures. The tiered design of the early warning signals avoids over-alarms while ensuring timely response to major risks, improving the availability and effectiveness of the early warning system.

[0027] The modules are connected via wired and / or wireless means to enable data transmission between them.

[0028] In this embodiment of the invention, the detailed implementation steps of complex domain feature transformation include:

[0029] Each instantaneous deviation value in the deviation sequence is treated as a discrete signal point. A preset window function is applied to these discrete signal points for weighting, resulting in a weighted deviation sequence. Window function processing is a fundamental step in improving the accuracy of spectrum analysis by reducing spectral leakage and enhancing analysis quality. The process begins by selecting an appropriate window function type, commonly including Hanning, Hamming, Blackman, or Kaiser windows, optimized based on the characteristics of the chemical parameters and the analysis objectives. The window function is applied to the time-domain deviation sequence, and signal weighting is achieved through dot product operations. The choice of window function balances frequency resolution and sidelobe suppression. For chemical process parameter analysis, window functions with good sidelobe suppression capabilities are typically chosen to improve the detection capability of small-amplitude anomalous frequency components. The weighted deviation sequence retains the temporal characteristics of the original deviation while reducing the impact of boundary discontinuities on subsequent transformations, laying the foundation for accurate spectrum analysis.

[0030] A discrete orthogonal transform is performed on the weighted bias sequence, and the real and imaginary coefficients in the transform result are used as the real and imaginary components of the complex characteristic coefficients, respectively. The discrete orthogonal transform is a crucial step in mapping a time-domain signal to the frequency domain. In this system, the Discrete Fourier Transform (DFT) or its efficient implementation, the Fast Fourier Transform (FFT), is employed. The transform process converts the weighted bias sequence from the time domain to the frequency domain, revealing the spectral characteristics of the signal. For a length of... The formula for calculating the DFT of the biased sequence is:

[0031] , ;

[0032] in, The weighted bias sequence is the first... The value of each sampling point, For the first Complex representation of each frequency component The imaginary unit. Transformation result. The real and imaginary parts of the coefficients characterize the amplitude and phase properties of the corresponding frequency components, respectively, and together constitute the complex characteristic coefficients. These coefficients fully describe the spectral characteristics of the deviation sequence, providing rich feature representations for subsequent complex domain analysis.

[0033] The complex characteristic coefficients are filtered according to the energy proportion of their corresponding frequency components, and the complex characteristic coefficients with an energy proportion greater than a preset energy threshold are retained to form a set of complex characteristic coefficients. Energy filtering is an important step in reducing computational complexity and improving analysis efficiency, filtering out unimportant noise by retaining the main frequency components. The filtering process first calculates the total energy of the entire spectrum, then calculates the energy proportion of each frequency component, and retains the frequency components whose energy proportion exceeds the preset threshold. The formula for calculating the energy of a frequency component is:

[0034] ;in, For the first The real part coefficients of a frequency component reflect the cosine characteristic of that frequency component. For the first The imaginary part coefficient of a frequency component reflects the sinusoidal characteristics of that frequency component.

[0035] The formula for calculating the energy percentage is:

[0036] ;in, For the first The energy percentage of each frequency component ranges from [0,1], and the sum of the percentages of all components is 1.

[0037] The preset energy threshold is typically set between 0.01 and 0.05, with the specific value adjusted according to the application scenario and analysis requirements. The filtered set of complex feature coefficients retains the main energy components in the deviation sequence while significantly reducing the computational burden, making subsequent critical deviation calculations more efficient and focused on important features.

[0038] In this embodiment of the invention, the detailed implementation steps of the method for calculating the critical deviation include:

[0039] A critical zone is preset on the complex plane. The critical zone is a band-shaped region centered on a preset critical line and with a preset bandwidth as its radius; the critical line is a vertical line whose real part equals the preset critical real part value. The critical zone setting is the benchmark definition for judging the nature of deviations. By establishing a discrimination region on the complex plane, it provides a spatial reference for deviation classification. The setting process first determines the critical real part value, which is usually determined through historical data analysis and reflects the boundary characteristics between fluctuating and trend-type deviations; then, the bandwidth is set to consider the allowable range of parameter fluctuations. The critical zone is distributed vertically on the complex plane, extending along the imaginary axis, with a limited width in the real part direction. Different process parameters may require different critical zone settings. The system supports personalized configurations for specific parameters, improving the accuracy of deviation judgment. The reasonable setting of the critical zone is key to the system's accurate differentiation of deviation types and directly affects subsequent baseline update strategies and risk assessments.

[0040] The absolute value of the difference between the real part of each complex characteristic coefficient and the critical real part is calculated as the horizontal distance between that complex characteristic coefficient and the center of the critical band. The horizontal distance is then subtracted from the bandwidth radius, and the result is truncated to obtain the vertical distance of the complex characteristic coefficient. Calculating the vertical distance is the core step in quantifying the degree to which the complex characteristic coefficient deviates from the critical band. By calculating the shortest distance from a point to the band, the deviation characteristics are evaluated. The calculation process first determines the position coordinates of each complex characteristic coefficient on the complex plane; then, the horizontal distance from that point to the critical line is calculated, i.e., the absolute value of the difference between the real part and the critical real part; finally, considering the bandwidth factor, the horizontal distance is subtracted from the bandwidth radius, and the non-negative result is taken as the final vertical distance. The formula for calculating the vertical distance is:

[0041] ;

[0042] in, Complex characteristic coefficients vertical distance, It is its real value. The critical real part value, This represents the critical band width. A vertical distance of zero indicates that the coefficient is within the critical band and is considered within the normal fluctuation range; a larger distance indicates a greater deviation from the critical characteristics and a higher degree of anomaly.

[0043] Using the energy proportion of each complex characteristic coefficient's corresponding frequency component as a weight, a weighted average is calculated on the vertical distances of all complex characteristic coefficients to obtain the critical deviation at the current sampling time. Weighted averaging is a crucial step in integrating the influence of multiple frequency components, highlighting the contribution of the main frequency components through energy weighting. The averaging process first determines the weight of each frequency component, using the previously calculated energy proportion as the natural weight; then, it calculates the weighted sum of the vertical distances of all complex characteristic coefficients; finally, it divides by the total weight to obtain the weighted average. The formula for calculating the critical deviation is:

[0044] ;

[0045] in, This represents the critical deviation. This is the set of filtered complex characteristic coefficients. Frequency components The proportion of energy, This corresponds to the vertical distance. Critical deviation is a comprehensive indicator that quantifies the degree to which a deviation sequence deviates from the normal fluctuation range in the frequency domain, providing an important basis for subsequent deviation classification and risk assessment.

[0046] In this embodiment of the invention, the detailed implementation steps of the deviation type determination method include:

[0047] The critical deviation is compared with a preset type determination threshold. If the critical deviation is less than or equal to the type determination threshold, the current deviation sequence is determined to be a fluctuating deviation. Initial type determination is the first screening step in deviation classification, using a threshold comparison of the critical deviation for rapid identification. The determination process sets an appropriate type determination threshold, typically determined based on historical data analysis and expert knowledge, reflecting the maximum permissible range of fluctuating deviations. When the critical deviation is below this threshold, it indicates that the deviation mainly manifests as random fluctuations near the critical band, consistent with the characteristics of fluctuating deviations. Fluctuating deviations are usually caused by measurement noise, environmental interference, or normal process fluctuations, and have a low risk level. The system will partially incorporate them into baseline updates, allowing the baseline to gradually adapt to normal process changes. This initial determination provides a rapid classification channel for most common deviations, improving the system's response efficiency.

[0048] If the critical deviation exceeds the type determination threshold, the proportion of complex characteristic coefficients located outside the critical band in the complex characteristic coefficient set is further calculated, denoted as the outer proportion. If the outer proportion exceeds a preset proportion threshold, it is determined to be a trend-type deviation; otherwise, it is determined to be a fluctuation-type deviation. The secondary determination is a refined step in handling boundary cases, using spectral distribution characteristics for deeper analysis. When the critical deviation exceeds the threshold, the system enters a more detailed judgment process, analyzing the distribution pattern of the complex characteristic coefficients. Calculating the proportion of coefficients outside the critical band reflects the degree of dispersion of the deviation in the spectrum. The formula for calculating the outer proportion is:

[0049] ;

[0050] in, The proportion of the outer side, This is an indicator function; it takes the value 1 when the condition is met, and 0 otherwise. This serves as the cardinality of the complex feature coefficient set. The proportion threshold is typically set between 0.3 and 0.5, adjusted according to the system's sensitivity to trends. When the proportion of the outer components exceeds the threshold, it indicates that most frequency components deviate from the critical band, conforming to the characteristics of trend-type deviation; otherwise, even if the critical deviation is large, it may be caused by abnormal fluctuations in a few frequency components, still classified as fluctuation-type deviation. This two-stage judgment mechanism effectively reduces misclassification and improves the system's accuracy in identifying different types of deviation.

[0051] In this embodiment of the invention, the detailed implementation steps of the method for decomposing a deviation sequence into fluctuation components and trend components include:

[0052] Least squares fitting is performed on the deviation sequence to obtain a first-order linear fitting line; the projection values ​​of each instantaneous deviation in the deviation sequence onto the fitting line constitute the initial trend component sequence. Linear fitting is the fundamental method for extracting the trend component from the deviation sequence, capturing the overall direction of data change through a mathematical model. The fitting process uses the least squares method to find the best fitting line that minimizes the sum of squared distances from the deviation points to the line. For a length of... The deviation sequence is fitted with the following linear equation: ,in For time sequence number, parameter and The optimal solution is:

[0053] ;

[0054] ;

[0055] in, For the first The deviation value at each sampling time. This corresponds to the time sequence number. The slope of the fitted straight line. The intercept reflects the overall rate of change of the deviation. This reflects the initial offset state. The straight line value corresponding to each time point is used as the initial trend component to form the initial trend component sequence. This sequence captures the main trend of the deviation, but it may be correlated with the fluctuation component and needs further orthogonalization processing.

[0056] Subtracting the initial trend component sequence from the deviation sequence yields the initial volatility component sequence; the inner product of the initial volatility component sequence and the initial trend component sequence is then calculated. Component separation is an intermediate step in deviation decomposition, using subtraction to initially separate the trend and volatility components. The separation process first subtracts the initial trend component from the original deviation sequence to obtain the initial volatility component; then, the inner product of the two component sequences is calculated to test their orthogonality. The formula for calculating the inner product is:

[0057] ;

[0058] in, for Sequence and The inner product of sequences is used to quantify the correlation between two sequences. For the initial trend component sequence, This is the initial wave component sequence. The first trend component sequence The value at each sampling time; The first wave component in the wave component sequence The values ​​at each sampling time point. Ideally, the trend component and the fluctuation component should be orthogonal, with an inner product of zero, indicating that the two components are completely independent. However, due to the limitations of linear fitting, the initial decomposition usually cannot achieve complete orthogonality and requires further adjustment. The magnitude of the inner product directly reflects the correlation between the two components, providing a basis for subsequent orthogonalization.

[0059] If the absolute value of the inner product is greater than a preset orthogonality threshold, the initial trend component sequence is orthogonally corrected. The corrected trend component sequence is used as the final trend component, and the deviation sequence minus the final trend component is used as the final fluctuation component. Orthogonalization correction is a key step to ensure component independence, eliminating correlation between components through projection adjustment. The correction process first sets an appropriate orthogonality threshold, typically 1-5% of the trend component energy; then it checks if the inner product value exceeds the threshold. If it does, orthogonalization adjustment is performed; otherwise, the initial decomposition result is used directly. Orthogonalization uses the Gram-Schmidt orthogonalization method, calculating the projection of the fluctuation component onto the trend direction and subtracting the corresponding proportion of the projection from the trend component. The trend component correction formula is:

[0060] ;

[0061] in, This is the corrected trend component sequence. This is the inner product (energy) of the trend component. The fluctuation component is then recalculated after correction. ,in This is the original deviation sequence. This orthogonalization process ensures the complete independence of the trend component and the volatility component, enabling the system to handle deviation components of different natures separately, thus improving the accuracy and relevance of risk assessment.

[0062] In this embodiment of the invention, the detailed implementation steps of the method for determining the baseline absorption rate include:

[0063] When the deviation type is a fluctuating deviation, the baseline absorption rate is set as the first absorption rate, which is determined by a positive correlation with the stable duration of the operating condition indicator. Fluctuating baseline adjustment is an adaptive mechanism for the system to adapt to normal fluctuations, using a higher absorption rate to allow the baseline to gradually follow process changes. The setting process first determines the stability of the current operating condition by obtaining the stable duration from the operating condition indicator; then, an appropriate absorption rate level is determined based on the stable duration. The longer the stable duration, the more reliable the process condition, and the higher the corresponding absorption rate. The first absorption rate is calculated using a duration mapping function, mapping the duration value to a reasonable absorption rate range (usually 0.2-0.8). The mapping function design considers the saturation effect, ensuring that even after an extremely long stable period, the absorption rate will not exceed the safety limit, maintaining the system's sensitivity to abnormal changes. This adaptive absorption rate mechanism based on operating condition stability allows the system to flexibly adjust the baseline update strategy at different production stages, quickly adapting to normal process changes while maintaining the ability to detect anomalies.

[0064] When the deviation type is trend-based, the baseline absorption rate is set as the second absorption rate. The second absorption rate is determined based on a negative correlation with the critical deviation, and it is lower than the first absorption rate. Trend-based baseline adjustment is a conservative mechanism for the system to cope with abnormal changes, limiting the impact of potential risks on the baseline through a lower absorption rate. The setting process first assesses the severity of the trend deviation based on the critical deviation; then, a negative correlation function is used, where the larger the critical deviation, the lower the corresponding absorption rate, which can be reduced to zero. The formula for calculating the second absorption rate is:

[0065] ;

[0066] in, The second absorption rate, and These are the minimum and maximum allowable absorption rates (typically) Approaching 0 (not exceeding 50% of the first absorption rate) This represents the critical deviation. The attenuation coefficient controls the rate of decrease in the absorptivity. This dynamic absorptivity mechanism based on deviation enables the system to adopt differentiated baseline adjustment strategies for different degrees of trend deviation. The greater the critical deviation, the more conservative the baseline update, effectively preventing abnormal trends from being incorrectly absorbed into the baseline.

[0067] The dynamic baseline update method involves adding the product of the mean of the fluctuation component and the baseline absorptivity to the current dynamic baseline value to obtain the updated dynamic baseline value. Baseline update is the core iterative step of the dynamic evaluation system, achieving intelligent baseline adjustment by controlling the absorptivity. The update process first calculates the mean of the current fluctuation component, reflecting the average level of the deviation; then, it selects the appropriate absorptivity based on the deviation type; finally, it incorporates the mean of the fluctuation component into the baseline update proportionally to the absorptivity. The baseline update formula is:

[0068] ;

[0069] in, For the updated dynamic baseline value, This is the current dynamic baseline value. Baseline absorbance (selected according to the type of deviation) or ), This represents the mean of the current fluctuation components. This classification and update mechanism ensures that the system treats deviations of different natures differently: for fluctuation-type deviations, the baseline can learn and follow appropriately; for trend-type deviations, the baseline remains relatively stable, allowing abnormal trends to be fully accumulated and assessed in the risk score, thereby improving the system's risk identification capability and timely early warning.

[0070] In this embodiment of the invention, the detailed implementation steps of the method for determining the cumulative weighting coefficient include:

[0071] The system calculates whether the sign of the trend component at the current sampling time is consistent with the sign of the trend component at the previous sampling time. If they are consistent, the same-direction continuity count is incremented by one; otherwise, it is reset to one. Same-direction continuity analysis is a fundamental step in assessing trend persistence, tracking the consistency of trend direction through sign comparison. The analysis process first extracts the signs (positive or negative) of the trend components at the current and previous times; then, it compares the signs to determine whether the trend direction remains consistent; finally, it updates the same-direction continuity count based on the comparison results. The continuity count intuitively reflects the duration of the trend; a larger continuity count indicates a more stable and persistent trend, but also a higher potential risk. For trends that have just reversed direction, the continuity count is reset to one, indicating the start of a new trend. This sign-consistency-based counting mechanism enables the system to identify and strengthen risk assessment of persistent same-direction trends, improving sensitivity to gradual anomalies.

[0072] The trend enhancement factor is the product of the natural logarithm of the continuous counts in the same direction and the critical deviation. The cumulative weight coefficient of the trend component at the current sampling time is the sum of the preset base weight and the trend enhancement factor. Calculating the weight coefficient is a crucial step in risk accumulation, reflecting the persistence and severity of the trend through dynamic weights. The calculation process first takes the natural logarithm of the continuous counts in the same direction to ensure that the weight growth exhibits logarithmic characteristics, avoiding excessive expansion of long-term trend weights. Then, the logarithm is multiplied by the critical deviation to obtain the trend enhancement factor, considering both the persistence and degree of deviation of the trend. Finally, the enhancement factor is added to the base weight to ensure that even short-term trends have a minimum weight. The formula for calculating the cumulative weight coefficient is:

[0073] ;

[0074] in, The cumulative weighting coefficient at the current sampling time. The preset base weights are (usually set to 0.1-0.3). For continuous counting in the same direction, This represents the critical deviation at the current moment. This dynamic weighting mechanism allows the system to assign higher risk assessment weights to trends that are consistently in the same direction and have high deviations, thereby strengthening the ability to identify gradual security risks and improving the sensitivity and foresight of the early warning system.

[0075] In this embodiment of the invention, the detailed implementation steps of the risk score calculation method include:

[0076] Within a preset cumulative window, the trend component at each sampling time is multiplied by its corresponding cumulative weight coefficient to obtain the weighted trend component. Time window weighting is the initial step in risk integration, strengthening the contribution of important trends through weight multiplication. The weighting process first determines an appropriate cumulative window size, typically set to 10-60 sampling points, adjusted according to the process change rate and response characteristics; then, it acquires the trend component and its corresponding cumulative weight coefficient at each time point within the window; finally, it obtains the weighted sequence through dot product. The weighted trend component retains the direction information of the trend while incorporating the assessment of trend persistence and severity, providing an optimized input signal for subsequent integration. The window size balances short-term volatility sensitivity and long-term trend stability; a window that is too short may be sensitive to noise, while a window that is too long may delay risk identification. The system supports setting differentiated window parameters based on different parameter characteristics.

[0077] The raw integral value is obtained by summing all weighted trend components; the number of times the trend component sign flips within the accumulation window is counted, recorded as the flip count. Trend accumulation is the core calculation of the risk integral, assessing the degree of risk accumulation through weighted summation over the time dimension. The accumulation process first sums all weighted trend components within the window to obtain the raw integral value, reflecting the cumulative effect of the trend; simultaneously, the number of trend sign flips within the window is counted to assess the stability of the trend. The formula for calculating the raw integral value is:

[0078] ;

[0079] in, The original integral value, To accumulate window length, For the first The cumulative weighting coefficient at each time step. This corresponds to the trend component. The flip count, by scanning the trend symbol sequence within the window, calculates the number of sign changes, reflecting the degree of instability in the trend direction. The magnitude of the original integral value intuitively reflects the cumulative level of risk, while the sign reflects the direction of risk (positive values ​​indicate an increased risk of parameter loss, and negative values ​​indicate a decreased risk of parameter loss).

[0080] The risk integral value is obtained by multiplying the original integral value by the flip decay factor. The flip decay factor is negatively correlated with the flip count; the larger the flip count, the smaller the flip decay factor. Stability correction is a refinement step in the risk integral, adjusting the weight of the influence of unstable trends through the decay factor. The correction process first calculates an appropriate decay factor based on the flip count, using a negative correlation function to reduce the impact of trends with frequent flips; then, the decay factor is multiplied by the original integral value to obtain the final risk integral value. The formula for calculating the decay factor is:

[0081] ;

[0082] in, To reverse the decay factor, For flip counting, This is the attenuation coefficient, controlling the intensity of attenuation. Risk integral value. The calculation formula is:

[0083] ;

[0084] This trend stability-based correction mechanism enables the system to distinguish between stable and persistent risk trends and volatile, random changes, reducing false alarms caused by fluctuations and improving the accuracy and reliability of risk assessment. The final risk score comprehensively considers trend direction, duration, deviation, and stability, providing a comprehensive quantitative indicator of risk for early warning decisions.

[0085] In this embodiment of the invention, the detailed implementation steps of the graded early warning threshold and early warning signal output method include:

[0086] The tiered early warning threshold system includes a concern threshold, a warning threshold, and an alarm threshold, with values ​​increasing sequentially. Setting these tiered thresholds is a fundamental configuration of the early warning system, classifying risk states of varying severity through multiple threshold levels. The setting process is based on historical data analysis, expert experience, and process safety standards to determine three key threshold levels. The concern threshold is the lowest level, used for early risk identification, and is typically set at 1.5-2 times the normal fluctuation range. The warning threshold is the medium level, representing significant risk, and is typically set at 1.5-2 times the concern threshold. The alarm threshold is the highest level, representing severe risk requiring immediate intervention, and is typically set at 1.3-1.8 times the warning threshold. Threshold settings balance timeliness and reliability; thresholds that are too low may lead to frequent false alarms, while thresholds that are too high may delay risk identification. The system supports setting differentiated thresholds for different parameters and operating conditions, improving the targeting and effectiveness of early warnings.

[0087] When the absolute value of the risk score exceeds the attention threshold but not the warning threshold, a Level 1 warning signal is output, and the current deviation type and critical deviation degree are recorded. The Level 1 warning output is the system's initial response mechanism, providing a mild alert to remind operators to pay attention to potential risks. The output process first compares the absolute value of the risk score with the attention threshold to confirm whether the Level 1 warning condition has been triggered; then, it generates warning information, including key information such as parameter identifier, deviation type, critical deviation degree, and risk score value; finally, it conveys the warning information to operators through the user interface or notification system in an appropriate manner (such as status icon color change, information bar prompts, etc.). Level 1 warning signals typically do not require immediate intervention; their main function is to alert operators to potential abnormal trends and provide a reference for preventative maintenance and adjustments.

[0088] A Level 2 warning signal is output when the absolute value of the risk score exceeds the warning threshold but does not exceed the alarm threshold; a Level 3 warning signal is output when the absolute value of the risk score exceeds the alarm threshold. Higher-level warnings are enhanced responses to significant and severe risks, using differentiated signals to indicate different levels of intervention needs. A Level 2 warning indicates that the risk has reached a level requiring attention and preparedness for intervention, drawing operators' attention through more obvious visual and audible cues (such as flashing yellow lights and intermittent audible alarms). A Level 3 warning indicates that the risk has reached a severe level requiring immediate intervention, ensuring immediate response from operators through the highest level of alarm methods (such as flashing red lights, continuous audible alarms, and mandatory confirmation). In addition to basic information, higher-level warning signals include extended content such as risk development trends, estimated evolution time, and suggested intervention measures, providing operators with more comprehensive decision support. This refined classification of risk levels ensures that intervention measures are matched to the severity of the risk, improving the practicality and acceptability of the warning system.

[0089] In this embodiment of the invention, the detailed implementation steps of the method for dynamically adjusting the graded early warning threshold include:

[0090] When a change in operating condition identifier is detected, the attention threshold, warning threshold, and alarm threshold are multiplied by a preset relaxation coefficient, which is greater than one. Operating condition change response is the adaptive mechanism of the early warning system, adjusting thresholds to adapt to fluctuations during process transitions. The response process first monitors changes in operating condition identifiers, such as production formula switching, equipment start-up and shutdown, and process condition adjustments; then, upon detecting a change, it immediately adjusts the warning threshold, using a relaxation coefficient to increase the tolerance range. The relaxation coefficient is typically set between 1.2 and 2.0, differentiated according to the type and magnitude of the operating condition change; the larger the change, the larger the corresponding relaxation coefficient. This immediate response mechanism allows the system to tolerate temporary fluctuations during operating condition transitions, avoiding frequent false alarms during transitions and improving the availability and reliability of the early warning system in dynamic process environments.

[0091] From the moment the operating condition indicator changes, the relaxation coefficient is gradually and linearly reduced to one over time, with the reduction period being a preset transition duration. A smooth transition is crucial for dynamic threshold adjustment; gradual reduction achieves a smooth transition from relaxed to normal. The reduction process first determines an appropriate transition duration, typically based on the process stabilization time; different types of operating condition changes may require different transition durations. Then, within this time period, the relaxation coefficient is gradually reduced according to a linear function, eventually restoring to the normal threshold level. The formula for the relaxation coefficient's reduction over time is:

[0092] ;

[0093] in, For a moment The relaxation factor, This is the initial relaxation factor. When the operating conditions change, To preset the transition duration, The function ensures that the coefficient is fixed at 1 after the transition period. This smooth transition mechanism avoids abrupt adjustments to the threshold, enabling the system to synchronously restore normal monitoring sensitivity as the process gradually stabilizes. It balances the fault tolerance during the transition period with the sensitivity during the stable period, improving the accuracy and continuity of the system's early warning across the entire operating range.

[0094] This invention achieves dynamic assessment and early warning of chemical process safety risks through deviation calculation, complex domain feature analysis, critical deviation calculation, deviation classification and decomposition, dynamic baseline updating, risk integral calculation, and graded early warning output. The feature fingerprint and critical band analysis methods of this invention can accurately distinguish different types of process deviations; the dynamic baseline updating strategy can adapt to process changes while maintaining sensitivity to abnormal trends; and the risk integral and graded early warning mechanism can promptly detect potential safety risks and provide differentiated early warning signals.

[0095] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0096] It should be noted that all formulas in this manual are calculated by removing dimensions and taking their numerical values. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0097] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A dynamic assessment and early warning system for safety risks in chemical processes, characterized in that, include: The deviation calculation module is used to obtain the process parameter values ​​at each sampling time in the chemical production process, and calculate the difference between the process parameter value at each sampling time and the current dynamic baseline value as the instantaneous deviation. The complex domain feature analysis module is used to construct a deviation sequence from the instantaneous deviations at each sampling time within a preset analysis window, and to perform complex domain feature transformation on the deviation sequence to obtain a set of complex feature coefficients. The critical deviation calculation module is used to calculate the vertical distance between the distribution position of each complex feature coefficient in the complex feature coefficient set on the complex plane and the preset critical zone, and to take the weighted average of all vertical distances as the critical deviation at the current sampling time. The deviation classification and determination module is used to determine the deviation type of the current deviation sequence based on the critical deviation degree. The deviation type includes fluctuation deviation and trend deviation. The deviation decomposition module is used to decompose the deviation sequence into mutually orthogonal fluctuation components and trend components. The dynamic baseline update module is used to determine the baseline absorption rate of the fluctuation component according to the deviation type, and to incorporate the fluctuation component into the dynamic baseline update according to the baseline absorption rate. The trend component is excluded from dynamic baseline updates; The risk integral calculation module is used to accumulate the trend components at each sampling time, determine the cumulative weight coefficient based on the critical deviation, and perform a weighted summation of the accumulated trend components to obtain the risk integral value at the current sampling time. The early warning output module is used to compare the risk score with the graded early warning threshold and output an early warning signal of the corresponding level based on the comparison result.

2. The system according to claim 1, characterized in that, The method for complex domain feature transformation includes: Each instantaneous deviation value in the deviation sequence is taken as a discrete signal point, and a preset window function is applied to the discrete signal points for weighting to obtain a weighted deviation sequence. The weighted bias sequence is subjected to a discrete orthogonal transformation, and the real and imaginary coefficients in the transformation result are taken as the real and imaginary components of the complex characteristic coefficients, respectively. Each complex characteristic coefficient is filtered according to the energy proportion of its corresponding frequency component, and the complex characteristic coefficients with an energy proportion greater than a preset energy threshold are retained to form a set of complex characteristic coefficients.

3. The system according to claim 1, characterized in that, The method for calculating the critical deviation includes: A critical zone is preset on the complex plane. The critical zone is a strip-shaped region centered on a preset critical line and with a preset bandwidth as its radius. The critical line is a vertical line whose real part is equal to the preset critical real part value. Calculate the absolute value of the difference between the real part of each complex characteristic coefficient and the critical real part value, and use it as the horizontal distance between the complex characteristic coefficient and the center of the critical band; subtract the bandwidth radius from the horizontal distance and take a non-negative truncation to obtain the vertical distance of the complex characteristic coefficient; Using the energy proportion of the frequency component corresponding to each complex characteristic coefficient as the weight, a weighted average of the vertical distances of all complex characteristic coefficients is performed to obtain the critical deviation at the current sampling time.

4. The system according to claim 3, characterized in that, The method for determining the type of deviation includes: The critical deviation is compared with a preset type determination threshold; if the critical deviation is less than or equal to the type determination threshold, the deviation type of the current deviation sequence is determined to be a fluctuation type deviation. If the critical deviation is greater than the type determination threshold, the proportion of the number of complex feature coefficients located outside the critical band in the complex feature coefficient set is further calculated and denoted as the outer proportion. If the outer proportion is greater than the preset proportion threshold, it is determined to be a trend deviation; otherwise, it is determined to be a fluctuation deviation.

5. The system according to claim 1, characterized in that, The method for decomposing the deviation sequence into fluctuation components and trend components includes: The deviation sequence is fitted with least squares to obtain a first-order linear fitting line; the projection values ​​of each instantaneous deviation in the deviation sequence onto the fitting line constitute an initial trend component sequence. Subtract the initial trend component sequence from the deviation sequence to obtain the initial fluctuation component sequence; calculate the inner product of the initial fluctuation component sequence and the initial trend component sequence; If the absolute value of the inner product is greater than the preset orthogonality threshold, the initial trend component sequence is orthogonally corrected, the corrected trend component sequence is used as the final trend component, and the deviation sequence minus the final trend component is used as the final fluctuation component.

6. The system according to claim 1, characterized in that, The method for determining the baseline absorbance includes: When the deviation type is a fluctuating deviation, the baseline absorption rate is set to the first absorption rate, which is determined based on the stable duration of the operating condition indicator. When the deviation type is trend-type deviation, the baseline absorption rate is set as the second absorption rate, which is determined based on the negative correlation of the critical deviation degree, and the second absorption rate is less than the first absorption rate; The dynamic baseline update method is as follows: add the product of the mean of the fluctuation component and the baseline absorptivity to the current dynamic baseline value to obtain the updated dynamic baseline value.

7. The system according to claim 1, characterized in that, The method for determining the cumulative weighting coefficient includes: Calculate whether the trend component sign at the current sampling time is consistent with the trend component sign at the previous sampling time; if consistent, increment the consecutive count in the same direction by one; if inconsistent, reset the consecutive count in the same direction to one. The product of the natural logarithm of the continuous count in the same direction and the critical deviation is used as the trend enhancement factor; the sum of the preset basic weight and the trend enhancement factor is used as the cumulative weight coefficient of the trend component at the current sampling time.

8. The system according to claim 7, characterized in that, The method for calculating the risk score includes: Within a preset cumulative window, the trend component at each sampling time is multiplied by the corresponding cumulative weight coefficient to obtain the weighted trend component. The original integral value is obtained by summing all weighted trend components; the number of times the trend component sign flips within the cumulative window is calculated and recorded as the flip count. The risk integral value is obtained by multiplying the original integral value by the inversion attenuation factor.

9. The system according to claim 1, characterized in that, The tiered early warning thresholds include a concern threshold, an early warning threshold, and an alarm threshold, with the values ​​of the three increasing sequentially. When the absolute value of the risk score exceeds the attention threshold but does not exceed the warning threshold, a first-level warning signal is output and the current deviation type and critical deviation degree are recorded. When the absolute value of the risk score exceeds the warning threshold but does not exceed the alarm threshold, a level two warning signal is output. When the absolute value of the risk score exceeds the alarm threshold, a level three warning signal is output.

10. The system according to claim 9, characterized in that, The tiered early warning threshold is dynamically adjusted based on the switching status of the operating condition identifier; When a change in the operating condition indicator is detected, the attention threshold, the warning threshold, and the alarm threshold are multiplied by a preset relaxation coefficient, wherein the relaxation coefficient is greater than one. From the moment the operating condition indicator changes, the relaxation coefficient is gradually reduced to one linearly over time, with the reduction period being a preset transition time.

Images