Method and system for rectifying tower fault identification

Abstract

The invention discloses a method and system for rectifying tower fault identification, and relates to the technical field of rectifying tower fault diagnos.The method comprises the steps that first historical data of a tower top product and second historical data of a tower bottom product are obtained when a rectifying tower has different faults, and changing curves corresponding to product components are constructed and fitted respectively; screening out a first influence factor and a second influence factor, establishing a mapping relation between an influence factor change curve and a fault type, and storing the mapping relation to a first data set and a second data set; then real-time data is obtained through online monitoring, a real-time change curve of the influence factors is constructed, the similarity between the real-time curve and a curve in a data set is calculated through a root-mean-square error to obtain an analysis result, and finally the analysis result is synthesized to output the fault type of the rectifying tower. The method effectively solves the problems that in the prior art, modeling is difficult, the method depends on experience and is easily influenced by instantaneous data fluctuation in rectifying tower fault recognition.

Description

Technical Field

[0001] This invention relates to the field of distillation column fault diagnosis technology, and in particular to a method and system for identifying distillation column faults. Background Technology

[0002] Distillation columns are widely used material separation equipment in the petrochemical industry. Their main function is to achieve gas-liquid separation, thus playing a crucial role in ensuring production continuity and product quality separation. Therefore, the fault detection and diagnosis methods and technologies for distillation columns occupy an important position in the entire intelligent manufacturing system of chemical industry.

[0003] Existing distillation column fault diagnosis technologies suffer from modeling challenges or reliance on experience. On the one hand, if complex mechanistic models are used for fault diagnosis, the complex operation of equipment such as distillation columns and the numerous influencing factors make it difficult to construct such models. On the other hand, methods that rely on online sampling data are susceptible to fluctuations in real-time data if they are not combined with historical data to form a systematic diagnostic standard. This makes it difficult to reliably identify fault sources and further highlights the problem of insufficient fault diagnosis accuracy. Summary of the Invention

[0004] In view of this, this application provides a method and system for identifying faults in distillation columns to address the shortcomings of the prior art.

[0005] The first aspect of this application provides a method for identifying faults in a distillation column, comprising: When different types of failures occur in the distillation column, the historical analysis data of the top product is recorded as the first historical data, and the historical analysis data of the bottom product is recorded as the second historical data. Based on the first historical data, a variation curve of different components in the top product of the tower is constructed within a preset period, which is recorded as the first variation curve and fitted. Analyze all the first change curves after fitting and select the first number of component types as the first influencing factors; Establish a mapping relationship between the first change curves corresponding to all first influencing factors and the fault types, and store them in the first dataset; Based on the second historical data, a variation curve of different components in the bottom product of the tower is constructed within a preset period, which is recorded as the second variation curve and fitted. Analyze all the second change curves after fitting and select the second number of component types as the second influencing factors; Establish a mapping relationship between the second change curves corresponding to all second influencing factors and the fault types, and store them in the second dataset; The top and bottom products of the distillation column were monitored online, and the first real-time change curves of all primary influencing factors and the second real-time change curves of all secondary influencing factors were constructed. Based on the first dataset, all first real-time change curves are analyzed to obtain a first analysis result, and based on the second dataset, all second real-time change curves are analyzed to obtain a second analysis result; Based on the combined results of the first and second analyses, the distillation column failure type is output.

[0006] In one possible implementation of the first aspect, constructing variation curves of different components in the top product of the tower within a preset period, denoted as the first variation curve, and performing fitting processing includes: Using the preset period as the horizontal axis and the purity data of any component in the top product of the tower as the vertical axis, a corresponding first variation curve is constructed; The first variation curve is fitted to obtain a corresponding fitted straight line, and the slope of the fitted straight line is calculated.

[0007] In one possible implementation of the first aspect, all the first change curves after fitting are analyzed, and a first number of component types are selected as the first influencing factors, including: Obtain any first variation curve after fitting and the corresponding fitted line, and denote the corresponding fitted line as the first fitted line. Calculate the residuals between the values ​​corresponding to all original data points in the first change curve and the values ​​predicted by the first fitted straight line; Take the absolute value of all residuals and calculate the arithmetic mean, which is used as the dispersion between the first variation curve and the first fitted line; For each of the remaining first change curves after fitting, the dispersion between the curve and the corresponding fitted line is calculated, and the first change curves with dispersion less than the first set threshold are simultaneously selected to complete one selection. For all the first change curves after the first screening, obtain the slope of the corresponding fitted line, and simultaneously screen the second change curves whose absolute value of the slope of the corresponding fitted line is greater than the second set threshold to complete the second screening. Based on all the first change curves after secondary screening, the corresponding component types are obtained and used as the first influencing factors.

[0008] In one possible implementation of the first aspect, constructing variation curves of different components in the bottom product of the tower within a preset period based on the second historical data, denoted as the second variation curve, and performing fitting processing includes: Using the preset period as the horizontal axis and the purity data of any component in the bottom product as the vertical axis, a corresponding second variation curve is constructed. The second variation curve is fitted to obtain a corresponding fitted straight line, and the slope of the fitted straight line is calculated.

[0009] In one possible implementation of the first aspect, all the second variation curves after fitting are analyzed, and a second number of component types are selected as the second influencing factors, including: Obtain any second variation curve after fitting and the corresponding fitted line, and denote the corresponding fitted line as the second fitted line; Calculate the residuals between the values ​​corresponding to all original data points in the second change curve and the predicted values ​​of the second fitted straight line; Take the absolute value of all residuals and calculate the arithmetic mean, which is used as the dispersion between the second variation curve and the second fitted line; For each of the remaining second change curves after fitting, the dispersion between the curve and the corresponding fitted line is calculated, and the second change curves with dispersion less than the third set threshold are simultaneously selected to complete one selection. For all the second change curves after the first screening, obtain the slope of the corresponding fitted line, and simultaneously screen the second change curves whose absolute value of the slope of the corresponding fitted line is greater than the fourth set threshold to complete the second screening. Based on all the second change curves after secondary screening, the corresponding component types are obtained and used as the second influencing factors.

[0010] In one possible implementation of the first aspect, constructing the first real-time change curves for all first impact factors and the second real-time change curves for all second impact factors includes: The purity data of all components corresponding to the first influencing factor were acquired in real time using an online analyzer to obtain the first acquisition result, and the purity data of all components corresponding to the second influencing factor were acquired in real time to obtain the second acquisition result. Based on the first acquisition results, a first real-time change curve of the corresponding component purity data over time is constructed for each first influencing factor. Based on the second acquisition result, a second real-time change curve of the corresponding component purity data over time is constructed for each second influencing factor.

[0011] In one possible implementation of the first aspect, based on the first dataset, all first real-time change curves are analyzed to obtain first analysis results, including: The number of the first impact factors is set to ; Each first real-time change curve is compared with the first change curve corresponding to each fault type in the first dataset to calculate the root mean square error between each first real-time change curve and the corresponding first change curve for all fault types. Specifically: The root mean square error between each first real-time change curve and the corresponding first change curve for any fault type is defined as follows: The number of data points selected for the first real-time change curve. Select data point values ​​for the first real-time change curve. Select the values ​​of the data points corresponding to the first change curve.

[0012] In one possible implementation of the first aspect, based on the second dataset, the second analysis results obtained for all second real-time change curves include: The number of the second impact factor is set to ; Each second real-time change curve is compared with the second change curve corresponding to each fault type in the second dataset to calculate the root mean square error between each second real-time change curve and the corresponding second change curve for all fault types. Specifically: The root mean square error between each second real-time change curve and the corresponding second change curve for any fault type is given. The number of data points selected for the second real-time change curve Select data point values ​​for the second real-time change curve. Select the values ​​of the data points corresponding to the second change curve.

[0013] In one possible implementation of the first aspect, combining the first analysis results and the second analysis results, the output distillation column failure types include: Based on the first dataset, count the number of cases where the root mean square error between the first change curve and the corresponding first real-time change curve is less than the first threshold for each fault type. and simultaneously acquire all The maximum value of the ratios is denoted as the first ratio. Based on the second dataset, the number of cases where the root mean square error between the second change curve and the corresponding second real-time change curve is less than the second threshold in each fault type is counted. and simultaneously acquire all The maximum value of the ratio is denoted as the second ratio. If the first ratio is 1, then the fault type corresponding to the first ratio is output as the fault type of the distillation column; If the first ratio is not 1 and the second ratio is 1, then the fault type corresponding to the second ratio is output as the fault type of the distillation column; If the first ratio is within a first set range, the corresponding fault type is obtained and recorded as the first fault type; if the second ratio is within a second set range, the corresponding fault type is obtained and recorded as the second fault type; determine whether the first fault type and the second fault type are consistent; if so, the first fault type or the second fault type is used as the fault type of the distillation column. If the first ratio is less than the lower limit of the first set range, or the second ratio is less than the lower limit of the second set range, then the distillation column fault is investigated to obtain the fault type, and the corresponding fault type and the first real-time change curve are synchronously stored in the first dataset, and the corresponding fault type and the second real-time change curve are stored in the second dataset.

[0014] A second aspect of this application provides a system for identifying faults in a distillation column, comprising: The historical data unit is used to obtain historical analysis data of the top product (recorded as the first historical data) and historical analysis data of the bottom product (recorded as the second historical data) when different types of failures occur in the distillation column. The first variation curve unit is used to construct variation curves of different components in the top product of the tower within a preset period based on the first historical data, which are denoted as the first variation curve and fitted. The first screening unit is used to analyze all the first change curves after fitting and screen a first number of component types as the first influencing factors. The first dataset unit is used to establish the mapping relationship between the first change curves corresponding to all first influencing factors and the fault types, and to store them in the first dataset; The second variation curve unit is used to construct variation curves of different components in the bottom product of the tower within a preset period based on the second historical data, which are recorded as the second variation curve and fitted. The second screening unit is used to analyze all the second change curves after fitting and to screen a second number of component types as the second influencing factors. The second dataset unit is used to establish the mapping relationship between the second change curves corresponding to all second influencing factors and the fault types, and to store them in the second dataset; The real-time monitoring unit is used to monitor both the top and bottom products of the distillation column online, and to construct the first real-time change curves of all first influencing factors and the second real-time change curves of all second influencing factors. The analysis unit is configured to analyze all first real-time change curves based on the first dataset to obtain a first analysis result, and to analyze all second real-time change curves based on the second dataset to obtain a second analysis result; The fault output unit is used to combine the first analysis result and the second analysis result to output the fault type of the distillation column.

[0015] The beneficial effects are as follows: This invention discloses a method for identifying faults in distillation columns. First, it acquires first historical data of the top product and second historical data of the bottom product when different faults occur in the distillation column. Then, it constructs and fits the corresponding product component variation curves, identifies the first and second influencing factors, establishes a mapping relationship between the influencing factor variation curves and the fault type, and stores them in the first and second datasets. Next, it acquires real-time data through online monitoring, constructs real-time variation curves of the influencing factors, calculates the similarity between the real-time curve and the curves in the dataset using the root mean square error, and obtains the analysis results. Finally, it outputs the distillation column fault type based on the comprehensive analysis results. This invention effectively solves the problems of difficult modeling, reliance on experience, and susceptibility to instantaneous data fluctuations in existing technologies for distillation column fault identification. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] Figure 1 This is a schematic flowchart of a method for identifying faults in a distillation column provided in an embodiment of this application; Figure 2 This is a schematic diagram of a system composition for identifying faults in a distillation column, provided in an embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0020] Example 1 Existing technologies suffer from modeling challenges or limitations due to reliance on experience. On the one hand, if complex mechanistic models are used for fault diagnosis, the complex operation of equipment such as distillation columns and the numerous influencing factors make it difficult to construct such models. On the other hand, methods that rely on online sampling data are susceptible to fluctuations in real-time data if they are not combined with historical data to form systematic diagnostic criteria. This makes it difficult to reliably identify fault sources and further highlights the problem of insufficient fault diagnosis accuracy.

[0021] Therefore, this application provides a method for identifying faults in distillation columns, such as... Figure 1 As shown, it includes: When different types of failures occur in the distillation column, the historical analysis data of the top product is recorded as the first historical data, and the historical analysis data of the bottom product is recorded as the second historical data. Based on the first historical data, a variation curve of different components in the top product of the tower is constructed within a preset period, which is recorded as the first variation curve and fitted. Analyze all the first change curves after fitting and select the first number of component types as the first influencing factors; Establish a mapping relationship between the first change curves corresponding to all first influencing factors and the fault types, and store them in the first dataset; Based on the second historical data, a variation curve of different components in the bottom product of the tower is constructed within a preset period, which is recorded as the second variation curve and fitted. Analyze all the second change curves after fitting and select the second number of component types as the second influencing factors; Establish a mapping relationship between the second change curves corresponding to all second influencing factors and the fault types, and store them in the second dataset; The top and bottom products of the distillation column were monitored online, and the first real-time change curves of all primary influencing factors and the second real-time change curves of all secondary influencing factors were constructed. Based on the first dataset, all first real-time change curves are analyzed to obtain a first analysis result, and based on the second dataset, all second real-time change curves are analyzed to obtain a second analysis result; Based on the combined results of the first and second analyses, the distillation column failure type is output.

[0022] The core principle of this embodiment for identifying distillation column faults is based on a logical chain of historical fault data modeling, real-time data monitoring, similarity matching, and multi-dimensional verification. By mining the fault response characteristics of the top and bottom product components, a mapping relationship between influencing factor change curves and fault types is established. Then, through comparative analysis of real-time monitoring data and historical models, and multi-dimensional verification, the fault type is accurately output. Specifically: When different types of failures occur in the distillation column, historical analysis data of the top product and the bottom product are obtained separately. Since the core function of the distillation column is to enrich light components at the top and heavy components at the bottom, the impact of failures on the components of the two is fundamentally different. Therefore, by obtaining historical analysis data of the top and bottom products, the specific characteristics of the failure can be preserved.

[0023] Based on two types of historical data, curves representing the change in component purity over a preset period are constructed (the top product is the first curve, and the bottom product is the second curve). These curves are then fitted to eliminate random fluctuations in the historical data and highlight the trend of component purity change over time (e.g., if a fault causes a continuous decrease in the purity of a component, the fitted line will show a significant negative slope). This trend is the core characteristic of the fault. It should be noted that the selection of the preset period needs to consider the sampling frequency of the purity data. For example, if the purity data of the top product is sampled at a high frequency, typically automatically checking component purity every 1-5 minutes (time unit: minutes), then the time unit of the preset period is minutes, and the selected period is 1-2 hours.

[0024] A secondary screening process was used to identify key components (influence factors) that are sensitive to failure from all component purity curves: First screening (dispersion screening): Calculate the average absolute value of the residuals (dispersion) between the first / second change curve and the corresponding fitted line, and screen curves with dispersion less than a set threshold to ensure that the selected components have "high stability" and exclude components with large random fluctuations and no fault indication significance. Secondary screening (slope screening) filters curves whose slope of the fitted line is greater than a set threshold to ensure that the selected components have a "strong trend" of change, that is, the component has a significant response to the fault (e.g., a large absolute value of the slope indicates that the fault causes a rapid change in the purity of the component). Finally, the curves corresponding to the first influencing factor for the top product selection and the second influencing factor for the bottom product selection are stored in the first dataset and the second dataset, respectively, to form a fault feature benchmark library that can be used for comparison.

[0025] An online analyzer is used to collect the purity data of the first influencing factor (key component at the top of the column) and the second influencing factor (key component at the bottom of the column) in real time. It is not necessary to monitor all components, but only to focus on the fault-sensitive components that have been verified in history. This reduces data redundancy and ensures the fault correlation of the collected data. Based on the real-time collected purity data of the influencing factors, a real-time change curve of time-purity is constructed (the first real-time change curve is for the top of the column, and the second real-time change curve is for the bottom of the column). The real-time change curve directly reflects the current component change trend of the distillation column and is a real-time signal source for judging whether there is a fault in the distillation column and the type of fault.

[0026] The similarity between the real-time curve and the historical curve is quantified by the root mean square error (RMSE) to determine whether the current change conforms to the characteristics of a certain type of historical fault. The principle is as follows: The RSME calculation logic calculates the root mean square error of each real-time curve and each fault type in the dataset. The tower top uses... The base of the tower is adopted The root mean square error is essentially the average deviation of the data points of two curves. The smaller the root mean square error, the closer the trend of the real-time curve is to the historical curve, and the higher the matching degree between the current fault and the historical fault type. For tower top products, count the number of influencing factors with root mean square error less than a set threshold (i.e., high matching degree) in a certain fault type. Calculate the ratio ( (This refers to the number of first influencing factors in the top-of-tower products), and this ratio reflects the overall matching degree between the real-time characteristics of the top-of-tower and a certain fault. Similarly, calculate the ratio for the base of the tower. This reflects the overall matching degree between the real-time characteristics of the tower bottom and a certain fault. The closer the ratio is to 1, the higher the degree of matching between the current real-time features and the historical fault, providing a quantitative basis for subsequent fault determination.

[0027] Cross-validation of matching results from both the top and bottom of the tower avoids misjudgment based on a single dimension. The principle is as follows: If the ratio at the top of the tower is 1 (all top influence factors match a certain fault), the fault type is directly output because the top influence factors respond most fully to the corresponding fault, and the result has the highest reliability. If the ratio at the top of the tower is not 1, but the ratio at the bottom of the tower is 1, the fault type matched at the bottom of the tower is also output. Similarly, the sufficiency of response is prioritized, but the sufficiency of response of the bottom influence factors is less than that of the top influence factors. If both the top and bottom ratios are within the set valid range (e.g., 0.8~1), it is determined whether the fault types matched by the two are consistent. If they are consistent, the fault type is output (dual-dimensional verification improves credibility). If they are inconsistent, manual investigation is required (to avoid misjudgment caused by single-dimensional bias). If either the ratio of the top product or the bottom product is lower than the lower limit of the set range (the real-time feature does not match any of the historical faults), then after manually checking and determining the fault type, the fault type and the corresponding real-time curve are updated to the first / second dataset. By iteratively expanding the fault feature library, the ability to identify new faults is improved, forming a closed loop of identification-optimization-re-identification.

[0028] This embodiment avoids misjudgment based on a single dimension by accurately locating the fault source. The core function of a distillation column is to enrich light components at the top and heavy components at the bottom through gas-liquid separation. The material composition and operating conditions at the top and bottom of the column are fundamentally different, and the impact of faults on them often exhibits different characteristics. For example, if the fault is "blockage of the rectification section trays," the purity of the light components at the top of the column will decrease (the slope of the corresponding component change curve in the first historical data is abnormal), but the purity of the heavy components at the bottom of the column may change less. If the fault is "insufficient heating in the stripping section," the purity of the heavy components at the bottom of the column will decrease significantly (the slope of the corresponding component change curve in the second historical data is abnormal), while the purity of the light components at the top of the column will change less. If only a single part is analyzed, fault misjudgment may occur. This embodiment analyzes the top and bottom products separately and performs cross-validation based on the differentiated fault response characteristics of the top and bottom of the column to accurately locate the column section where the fault occurred and the fault type, significantly reducing the rate of misjudgment and missed judgment.

[0029] In this embodiment, the first and second influencing factors are determined at the top and bottom of the tower through a secondary screening process, respectively. Its advantages are: To avoid interference from invalid components, the top product is mainly composed of light components, and the components most sensitive to failure are also light components; the bottom product is mainly composed of heavy components, and the sensitive components are also heavy components. If the top and bottom product data are mixed for screening, components that do not respond to failure (such as heavy components at the top and light components at the bottom) may be included, resulting in redundant influencing factors and interfering with subsequent similarity analysis. To ensure the fault sensitivity of influencing factors, for example, when screening at the top of the tower, the first influencing factor is determined by dispersion and slope. Similarly, when screening at the bottom of the tower, the second influencing factor is determined by dispersion and slope. This ensures that the two types of influencing factors accurately match the fault response patterns at the top and bottom of the tower, respectively, providing a high-quality data foundation for subsequent real-time monitoring and RSME analysis, and avoiding invalid data from lowering the accuracy of fault diagnosis.

[0030] One of the core shortcomings of existing technologies is their susceptibility to instantaneous data fluctuations. However, analyzing the top and bottom products separately can mitigate these fluctuations through two-dimensional cross-validation, thereby improving diagnostic stability. To resist instantaneous fluctuations, if the real-time curve of a certain influencing factor at the top of the tower becomes abnormal due to instantaneous sampling error, but the real-time curve of the corresponding influencing factor at the bottom of the tower is highly matched with the historical curve, the fault type can be verified by the normal result of the second ratio, thus avoiding misjudgment of fault due to instantaneous fluctuations in a single part. To address complex faults, the system calculates a first ratio (top of the tower) and a second ratio (bottom of the tower) separately. If both ratios point to the same fault type (both within the set range and the faults are consistent), the reliability of the fault diagnosis is enhanced. If both ratios point to different fault types, manual investigation and dataset updates are triggered to avoid fault identification errors due to the limitations of single-dimensional analysis.

[0031] In this embodiment, when the first ratio or the second ratio is less than the lower limit of the corresponding set range, the fault type and the corresponding real-time curve need to be updated to the dataset. If the fault only has a significant impact on the top of the tower, only the fault type and the first real-time curve at the top of the tower need to be updated to the first dataset, and the second dataset at the bottom of the tower does not need to be updated redundantly. Conversely, if the fault has a significant impact on the bottom of the tower, only the fault type and the second real-time curve at the bottom of the tower need to be updated to the second dataset, and the first dataset at the top of the tower does not need to be updated redundantly. This refined update can prevent the data from expanding due to the inclusion of curves with no fault response, ensuring that the first and second datasets always focus on high-value fault features, maintaining high efficiency in similarity analysis during long-term use, and improving the ability to identify local faults.

[0032] In some embodiments, constructing variation curves of different components in the top product of the tower within a preset period, denoted as the first variation curve, and performing fitting processing includes: Using the preset period as the horizontal axis and the purity data of any component in the top product of the tower as the vertical axis, a corresponding first variation curve is constructed; The first variation curve is fitted to obtain a corresponding fitted straight line, and the slope of the fitted straight line is calculated.

[0033] In some embodiments, analyzing all the first change curves after fitting and selecting a first number of component types as the first influencing factors includes: Obtain any first variation curve after fitting and the corresponding fitted line, and denote the corresponding fitted line as the first fitted line. Calculate the residuals between the values ​​corresponding to all original data points in the first change curve and the values ​​predicted by the first fitted straight line; Take the absolute value of all residuals and calculate the arithmetic mean, which is used as the dispersion between the first variation curve and the first fitted line; For each of the remaining first change curves after fitting, the dispersion between the curve and the corresponding fitted line is calculated, and the first change curves with dispersion less than the first set threshold are simultaneously selected to complete one selection. For all the first change curves after the first screening, obtain the slope of the corresponding fitted line, and simultaneously screen the second change curves whose absolute value of the slope of the corresponding fitted line is greater than the second set threshold to complete the second screening. Based on all the first change curves after secondary screening, the corresponding component types are obtained and used as the first influencing factors.

[0034] In some embodiments, based on the second historical data, constructing variation curves of different components in the bottom product of the tower within a preset period, denoted as the second variation curve, and performing fitting processing includes: Using the preset period as the horizontal axis and the purity data of any component in the bottom product as the vertical axis, a corresponding second variation curve is constructed. The second variation curve is fitted to obtain a corresponding fitted straight line, and the slope of the fitted straight line is calculated.

[0035] In some embodiments, analyzing all the second variation curves after fitting and selecting a second number of component types as second influencing factors includes: Obtain any second variation curve after fitting and the corresponding fitted line, and denote the corresponding fitted line as the second fitted line; Calculate the residuals between the values ​​corresponding to all original data points in the second change curve and the predicted values ​​of the second fitted straight line; Take the absolute value of all residuals and calculate the arithmetic mean, which is used as the dispersion between the second variation curve and the second fitted line; For each of the remaining second change curves after fitting, the dispersion between the curve and the corresponding fitted line is calculated, and the second change curves with dispersion less than the third set threshold are simultaneously selected to complete one selection. For all the second change curves after the first screening, obtain the slope of the corresponding fitted line, and simultaneously screen the second change curves whose absolute value of the slope of the corresponding fitted line is greater than the fourth set threshold to complete the second screening. Based on all the second change curves after secondary screening, the corresponding component types are obtained and used as the second influencing factors.

[0036] In some embodiments, constructing the first real-time change curves for all first impact factors and the second real-time change curves for all second impact factors includes: The purity data of all components corresponding to the first influencing factor were acquired in real time using an online analyzer to obtain the first acquisition result, and the purity data of all components corresponding to the second influencing factor were acquired in real time to obtain the second acquisition result. Based on the first acquisition results, a first real-time change curve of the corresponding component purity data over time is constructed for each first influencing factor. Based on the second acquisition result, a second real-time change curve of the corresponding component purity data over time is constructed for each second influencing factor.

[0037] In some embodiments, based on the first dataset, all first real-time change curves are analyzed to obtain first analysis results, including: The number of the first impact factors is set to ; Each first real-time change curve is compared with the first change curve corresponding to each fault type in the first dataset to calculate the root mean square error between each first real-time change curve and the corresponding first change curve for all fault types. Specifically: The root mean square error between each first real-time change curve and the corresponding first change curve for any fault type is defined as follows: The number of data points selected for the first real-time change curve. Select data point values ​​for the first real-time change curve. Select the values ​​of the data points corresponding to the first change curve.

[0038] In some embodiments, based on the second dataset, the second analysis results obtained for all second real-time change curves include: The number of the second impact factor is set to ; Each second real-time change curve is compared with the second change curve corresponding to each fault type in the second dataset to calculate the root mean square error between each second real-time change curve and the corresponding second change curve for all fault types. Specifically: The root mean square error between each second real-time change curve and the corresponding second change curve for any fault type is given. The number of data points selected for the second real-time change curve Select data point values ​​for the second real-time change curve. Select the values ​​of the data points corresponding to the second change curve.

[0039] In some embodiments, combining the first analysis result and the second analysis result, the output distillation column failure type includes: Based on the first dataset, count the number of cases where the root mean square error between the first change curve and the corresponding first real-time change curve is less than the first threshold for each fault type. and simultaneously acquire all The maximum value of the ratios is denoted as the first ratio. Based on the second dataset, the number of cases where the root mean square error between the second change curve and the corresponding second real-time change curve is less than the second threshold in each fault type is counted. and simultaneously acquire all The maximum value of the ratio is denoted as the second ratio. If the first ratio is 1, then the fault type corresponding to the first ratio is output as the fault type of the distillation column; If the first ratio is not 1 and the second ratio is 1, then the fault type corresponding to the second ratio is output as the fault type of the distillation column; If the first ratio is within a first set range, the corresponding fault type is obtained and recorded as the first fault type; if the second ratio is within a second set range, the corresponding fault type is obtained and recorded as the second fault type; determine whether the first fault type and the second fault type are consistent; if so, the first fault type or the second fault type is used as the fault type of the distillation column. If the first ratio is less than the lower limit of the first set range, or the second ratio is less than the lower limit of the second set range, then the distillation column fault is investigated to obtain the fault type, and the corresponding fault type and the first real-time change curve are synchronously stored in the first dataset, and the corresponding fault type and the second real-time change curve are stored in the second dataset.

[0040] Example 2 Based on the method for identifying distillation column faults provided in Embodiment 1 of this application, correspondingly, Embodiment 2 of this application also provides a system for identifying distillation column faults, such as... Figure 2 As shown, it includes: The historical data unit is used to obtain historical analysis data of the top product (recorded as the first historical data) and historical analysis data of the bottom product (recorded as the second historical data) when different types of failures occur in the distillation column. The first variation curve unit is used to construct variation curves of different components in the top product of the tower within a preset period based on the first historical data, which are denoted as the first variation curve and fitted. The first screening unit is used to analyze all the first change curves after fitting and screen a first number of component types as the first influencing factors. The first dataset unit is used to establish the mapping relationship between the first change curves corresponding to all first influencing factors and the fault types, and to store them in the first dataset; The second variation curve unit is used to construct variation curves of different components in the bottom product of the tower within a preset period based on the second historical data, which are recorded as the second variation curve and fitted. The second screening unit is used to analyze all the second change curves after fitting and to screen a second number of component types as the second influencing factors. The second dataset unit is used to establish the mapping relationship between the second change curves corresponding to all second influencing factors and the fault types, and to store them in the second dataset; The real-time monitoring unit is used to monitor both the top and bottom products of the distillation column online, and to construct the first real-time change curves of all first influencing factors and the second real-time change curves of all second influencing factors. The analysis unit is configured to analyze all first real-time change curves based on the first dataset to obtain a first analysis result, and to analyze all second real-time change curves based on the second dataset to obtain a second analysis result; The fault output unit is used to combine the first analysis result and the second analysis result to output the fault type of the distillation column.

[0041] The specific principles and execution processes of each unit in the system for identifying distillation column faults disclosed in Embodiment 2 of this application are the same as those in the method for identifying distillation column faults disclosed in Embodiment 1 of this application. Please refer to the corresponding parts of the method for identifying distillation column faults disclosed in Embodiment 1 of this application, which will not be repeated here.

[0042] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computing software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0043] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0044] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for identifying faults in a distillation column, characterized in that, include: When different types of failures occur in the distillation column, the historical analysis data of the top product is recorded as the first historical data, and the historical analysis data of the bottom product is recorded as the second historical data. Based on the first historical data, a variation curve of different components in the top product of the tower is constructed within a preset period, which is recorded as the first variation curve and fitted. Analyze all the first change curves after fitting and select the first number of component types as the first influencing factors; Establish a mapping relationship between the first change curves corresponding to all first influencing factors and the fault types, and store them in the first dataset; Based on the second historical data, a variation curve of different components in the bottom product of the tower is constructed within a preset period, which is recorded as the second variation curve and fitted. Analyze all the second change curves after fitting and select the second number of component types as the second influencing factors; Establish a mapping relationship between the second change curves corresponding to all second influencing factors and the fault types, and store them in the second dataset; The top and bottom products of the distillation column were monitored online, and the first real-time change curves of all primary influencing factors and the second real-time change curves of all secondary influencing factors were constructed. Based on the first dataset, all first real-time change curves are analyzed to obtain a first analysis result, and based on the second dataset, all second real-time change curves are analyzed to obtain a second analysis result; Based on the combined results of the first and second analyses, the distillation column failure type is output.

2. The method for identifying faults in a distillation column according to claim 1, characterized in that, Constructing the variation curves of different components in the top product of the tower within a preset period, denoted as the first variation curve, and performing fitting processing includes: Using the preset period as the horizontal axis and the purity data of any component in the top product of the tower as the vertical axis, a corresponding first variation curve is constructed; The first variation curve is fitted to obtain a corresponding fitted straight line, and the slope of the fitted straight line is calculated.

3. The method for fault identification in a distillation column according to claim 2, characterized in that, Analyzing all the first change curves after fitting, the first number of component types selected as the first influencing factors include: Obtain any first variation curve after fitting and the corresponding fitted line, and denote the corresponding fitted line as the first fitted line. Calculate the residuals between the values ​​corresponding to all original data points in the first change curve and the values ​​predicted by the first fitted straight line; Take the absolute value of all residuals and calculate the arithmetic mean, which is used as the dispersion between the first variation curve and the first fitted line; For each of the remaining first change curves after fitting, the dispersion between the curve and the corresponding fitted line is calculated, and the first change curves with dispersion less than the first set threshold are simultaneously selected to complete one selection. For all the first change curves after the first screening, obtain the slope of the corresponding fitted line, and simultaneously screen the second change curves whose absolute value of the slope of the corresponding fitted line is greater than the second set threshold to complete the second screening. Based on all the first change curves after secondary screening, the corresponding component types are obtained and used as the first influencing factors.

4. The method for fault identification in a distillation column according to claim 1, characterized in that, Based on the second historical data, constructing variation curves of different components in the bottom product of the tower within a preset period, denoted as the second variation curve, and performing fitting processing includes: Using the preset period as the horizontal axis and the purity data of any component in the bottom product as the vertical axis, a corresponding second variation curve is constructed. The second variation curve is fitted to obtain a corresponding fitted straight line, and the slope of the fitted straight line is calculated.

5. The method for fault identification in a distillation column according to claim 4, characterized in that, Analyzing all the second variation curves after fitting, the second number of component types were selected as the second influencing factors, including: Obtain any second variation curve after fitting and the corresponding fitted line, and denote the corresponding fitted line as the second fitted line; Calculate the residuals between the values ​​corresponding to all original data points in the second change curve and the predicted values ​​of the second fitted straight line; Take the absolute value of all residuals and calculate the arithmetic mean, which is used as the dispersion between the second variation curve and the second fitted line; For each of the remaining second change curves after fitting, the dispersion between the curve and the corresponding fitted line is calculated, and the second change curves with dispersion less than the third set threshold are simultaneously selected to complete one selection. For all the second change curves after the first screening, obtain the slope of the corresponding fitted line, and simultaneously screen the second change curves whose absolute value of the slope of the corresponding fitted line is greater than the fourth set threshold to complete the second screening. Based on all the second change curves after secondary screening, the corresponding component types are obtained and used as the second influencing factors.

6. The method for identifying faults in a distillation column according to claim 1, characterized in that, The first real-time change curves for all first-impact factors and the second real-time change curves for all second-impact factors are constructed as follows: The purity data of all components corresponding to the first influencing factor were acquired in real time using an online analyzer to obtain the first acquisition result, and the purity data of all components corresponding to the second influencing factor were acquired in real time to obtain the second acquisition result. Based on the first acquisition results, a first real-time change curve of the corresponding component purity data over time is constructed for each first influencing factor. Based on the second acquisition result, a second real-time change curve of the corresponding component purity data over time is constructed for each second influencing factor.

7. The method for fault identification in a distillation column according to claim 1, characterized in that, Based on the first dataset, all the first real-time change curves are analyzed, and the first analysis results include: The number of the first impact factor is set to ; Each first real-time change curve is compared with the first change curve corresponding to each fault type in the first dataset to calculate the root mean square error between each first real-time change curve and the corresponding first change curve for all fault types. Specifically: The root mean square error between each first real-time change curve and the corresponding first change curve for any fault type is defined as follows: The number of data points selected for the first real-time change curve. Select data point values ​​for the first real-time change curve. Select the values ​​of the data points corresponding to the first change curve.

8. A method for identifying faults in a distillation column according to claim 7, characterized in that, Based on the second dataset, the second analysis results for all second real-time change curves include: The number of the second impact factor is set to ; Each second real-time change curve is compared with the second change curve corresponding to each fault type in the second dataset to calculate the root mean square error between each second real-time change curve and the corresponding second change curve for all fault types. Specifically: The root mean square error between each second real-time change curve and the corresponding second change curve for any fault type is given. The number of data points selected for the second real-time change curve Select data point values ​​for the second real-time change curve. Select the values ​​of the data points corresponding to the second change curve.

9. A method for identifying faults in a distillation column according to claim 8, characterized in that, Based on the combined results of the first and second analyses, the output distillation column failure types include: Based on the first dataset, the number of cases where the root mean square error between the first change curve and the corresponding first real-time change curve is less than the first threshold in each fault type is counted. and simultaneously acquire all The maximum value of the ratios is denoted as the first ratio. Based on the second dataset, the number of cases where the root mean square error between the second change curve and the corresponding second real-time change curve is less than the second threshold in each fault type is counted. and simultaneously acquire all The maximum value of the ratio is denoted as the second ratio. If the first ratio is 1, then the fault type corresponding to the first ratio is output as the fault type of the distillation column; If the first ratio is not 1 and the second ratio is 1, then the fault type corresponding to the second ratio is output as the fault type of the distillation column; If the first ratio is within a first set range, the corresponding fault type is obtained and recorded as the first fault type; if the second ratio is within a second set range, the corresponding fault type is obtained and recorded as the second fault type; determine whether the first fault type and the second fault type are consistent; if so, the first fault type or the second fault type is used as the fault type of the distillation column. If the first ratio is less than the lower limit of the first set range, or the second ratio is less than the lower limit of the second set range, then the distillation column fault is investigated to obtain the fault type, and the corresponding fault type and the first real-time change curve are synchronously stored in the first dataset, and the corresponding fault type and the second real-time change curve are stored in the second dataset.

10. A system for identifying faults in a distillation column, characterized in that, include: The historical data unit is used to obtain historical analysis data of the top product (recorded as the first historical data) and historical analysis data of the bottom product (recorded as the second historical data) when different types of failures occur in the distillation column. The first variation curve unit is used to construct variation curves of different components in the top product of the tower within a preset period based on the first historical data, which are denoted as the first variation curve and fitted. The first screening unit is used to analyze all the first change curves after fitting and screen a first number of component types as the first influencing factors. The first dataset unit is used to establish the mapping relationship between the first change curves corresponding to all first influencing factors and the fault types, and to store them in the first dataset; The second variation curve unit is used to construct variation curves of different components in the bottom product of the tower within a preset period based on the second historical data, which are recorded as the second variation curve and fitted. The second screening unit is used to analyze all the second change curves after fitting and to screen a second number of component types as the second influencing factors. The second dataset unit is used to establish the mapping relationship between the second change curves corresponding to all second influencing factors and the fault types, and to store them in the second dataset; The real-time monitoring unit is used to monitor both the top and bottom products of the distillation column online, and to construct the first real-time change curves of all first influencing factors and the second real-time change curves of all second influencing factors. The analysis unit is configured to analyze all first real-time change curves based on the first dataset to obtain a first analysis result, and to analyze all second real-time change curves based on the second dataset to obtain a second analysis result; The fault output unit is used to combine the first analysis result and the second analysis result to output the fault type of the distillation column.

Images