Multi-effect rectification variable working condition intelligent card edge optimization system

Abstract

The invention discloses a multi-effect rectification variable working condition intelligent card edge optimization system, which comprises a card edge optimization module used for card edge optimization control, and comprises the following steps: obtaining a stability judgment result of a current multi-effect multi-stage rectification tower system at the beginning of production; if the system is judged to be stable, acquiring a control variable and an observation variable; performing parameter search on the control variables by using a particle swarm optimization algorithm, acquiring predicted product quality by using a quality inspection module, and iteratively updating the control parameters through an optimization function in the particle swarm optimization algorithm to finally obtain the optimal control variable of the current system; the stability monitoring module is used for dynamically calculating the stability value of each monitoring point position in real time and obtaining the stability judgment result of the multi-effect multi-stage rectifying tower system; the time sequence analysis module is used for obtaining variables influencing the product quality and the influence time of the variables; and the quality inspection module comprises a soft measurement model and is used for inputting a control variable and an observation variable and outputting predicted product quality.

Description

[0001] This invention relates to the field of industrial data detection and processing technology, and in particular to an intelligent edge-finding system for multi-effect distillation under varying operating conditions. Background Technology

[0002] In multi-effect distillation, "margin control," as an advanced process optimization strategy, essentially involves dynamically adjusting key process parameters to precisely approximate product quality indicators to the standard threshold boundaries. This achieves the dual goals of maximizing product yield and minimizing equipment energy consumption while ensuring a high pass rate. However, current industry practice is caught in a "margin paradox": on the one hand, excessive pursuit of product quality leads to "quality redundancy." Taking ethanol distillation as an example, if the purchase standard requires a purity of 95%, but the actual production purity is 96%, then every 100 tons of product produced will result in an excess loss of approximately 1 ton of effective component, accompanied by unnecessary energy and material consumption. Moreover, this quality premium is not compensated by market prices, forming a typical internal cost drain. On the other hand, if product quality falls below the standard requirements, it directly weakens market competitiveness and may even affect product sales, significantly reducing corporate profits. Therefore, the industry currently exhibits a tendency towards "safety margin" management, that is, setting quality indicators higher than the standard to mitigate risks. This extensive control model inevitably leads to long-term overcapacity in product quality, accompanied by decreased output and increased energy consumption, forming a triple contradiction of "overcapacity - rising energy consumption - limited capacity." Currently, multi-effect distillation edge optimization faces significant talent bottlenecks and technological constraints. Due to differences in products, distillation processes, and multi-effect energy-saving solutions, modeling is difficult and reusable. Traditional optimization models heavily rely on the professional experience of senior automation engineers, requiring specialized teams to deeply analyze process mechanisms, build material-energy balance models, and repeatedly debug to construct an optimized model. However, this static optimization model is not only time-consuming and labor-intensive but also difficult to adapt to the dynamic evolution of the equipment throughout its entire lifecycle. When the equipment ages, degrades, malfunctions, or undergoes technical upgrades, the original model becomes invalid due to parameter inaccuracies, forcing companies to repeatedly invest professional resources in systematic reconstruction. Therefore, the multi-effect distillation industry urgently needs a transferable and adaptive intelligent energy management system to assist in the long-term operation of equipment and optimize energy management. Summary of the Invention

[0003] Technical objective: To address the shortcomings of existing technologies, this invention discloses an intelligent edge-finding system for multi-effect distillation under varying operating conditions. This system modifies control parameters in real time based on changes in the current environment, thereby reducing energy consumption, increasing production capacity, and maintaining long-term system stability.

[0004] Technical solution: To achieve the above technical objectives, the present invention adopts the following technical solution.

[0005] A smart edge-finding optimization system for multi-effect distillation under varying operating conditions includes an edge-finding optimization module and a stability monitoring module, a time-series analysis module, and a quality inspection module connected to the edge-finding optimization module; wherein the time-series analysis module is connected to the stability monitoring module and the quality inspection module. The edge-finding optimization module is used for edge-finding control, including: at the start of production, obtaining the stability judgment result of the current multi-effect multi-stage distillation column system from the stability monitoring module; if the system is judged to be stable, obtaining the control variables and observed variables of the current multi-effect multi-stage distillation column system from the time series analysis module; using the particle swarm optimization algorithm to search for parameters of the control variables, and using the quality inspection module to obtain the predicted product quality; using the particle swarm optimization algorithm to iteratively update the control parameters through the optimization function to obtain the control variable parameters, and obtaining all the control variable parameters of the current system as several first control variables; and obtaining the optimal control variables of the current system according to the preset parameter update formula and the first control variables. The stability monitoring module is used to determine the stability points required for edge optimization control based on the process analysis and time sequence analysis of the multi-effect multi-stage distillation column system, and set them as monitoring points; it dynamically calculates the stability value of each monitoring point in real time, and obtains the stability judgment result of the multi-effect multi-stage distillation column system based on the stability value of each monitoring point. The time series analysis module is used to obtain variables affecting product quality using time series correlation analysis methods, and to determine whether the process analysis variables of the multi-effect multi-stage distillation column system are control variables or observed variables. Observed variables refer to relevant variables that are highly correlated with product quality but cannot be directly controlled, while control variables refer to relevant variables that are highly correlated with product quality and whose processes can be modified. The module also works with the quality inspection module to obtain the influence time of control variables affecting product quality. The quality control module includes a soft measurement model, which takes input control variables and observed variables and outputs a predicted product quality.

[0006] Beneficial effects: This invention can modify control parameters in real time according to changes in the current environment, thereby reducing energy consumption, increasing production capacity, and maintaining long-term system stability. In addition, this invention introduces a dual mechanism of historical analysis and real-time analysis in stability analysis. Historical analysis determines the system stability boundary by analyzing historical operating data, while real-time analysis dynamically evaluates the current stability state, realizing an instantaneous stability monitoring system. Attached Figure Description

[0007] Figure 1 This is a system block diagram of Embodiment 1 of the present invention. Detailed Implementation

[0008] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application. Example 1

[0009] As attached Figure 1 As shown, an intelligent edge-finding optimization system for multi-effect distillation under varying operating conditions includes an edge-finding optimization module and a stability monitoring module, a time-series analysis module, and a quality inspection module connected to the edge-finding optimization module. The time-series analysis module is connected to the stability monitoring module and the quality inspection module. The edge-finding optimization module is used for edge-finding control, including: at the start of production, obtaining the stability judgment result of the current multi-effect multi-stage distillation column system from the stability monitoring module; if the system is judged to be stable, obtaining the control variables and observed variables of the current multi-effect multi-stage distillation column system from the time series analysis module; using the particle swarm optimization algorithm to search for parameters of the control variables, and using the quality inspection module to obtain the predicted product quality; using the particle swarm optimization algorithm to iteratively update the control parameters through the optimization function to obtain the control variable parameters, and obtaining all the control variable parameters of the current system as several first control variables; and obtaining the optimal control variables of the current system according to the preset parameter update formula and the first control variables. After obtaining the first control variable, to maintain long-term system stability, if the change in the first control variable is less than the set upper limit, the parameter is updated directly; if it is greater than or equal to the adjustment upper limit, the upper limit is updated according to the adjustment direction of the first control variable, with the set upper limit as the step size for parameter updates. The parameter update formula is: , in, For the current control variable, For the control variables of the previous time step, This represents the maximum step size.

[0010] Through long-term system operation and continuous optimization and iteration, the product quality is constantly approaching the set quality, but will not fall below the set quality. In this process, energy consumption is reduced, production capacity is increased, and system stability is maintained.

[0011] The implementation process of the particle swarm optimization algorithm includes: After initializing several populations, the position of each population is the control variable parameter. The velocity is randomly initialized as its initial search direction. Then, the control variable is combined with the observed variable and put into the quality inspection module to obtain the predicted product quality. Then, the population points are iterated according to the optimization function to obtain the optimal control variable parameter. The formula for calculating the optimization function is as follows: , in, For the quality inspection module to predict product quality, To set a quality threshold, The penalty coefficient is set according to the actual situation. The parameter x with the highest score is the global optimum gBest. Then, a local optimum pBest is searched in the neighborhood of each population point. The update speed for each population point is based on its local optimum and the global optimum.

[0012] in, For individual learning factors, As a population learning factor, it determines whether the update of population points approaches a local optimum or a global optimum. r is a random number between 0 and 1, increasing the randomness of the system and helping to expand the search space. Let be the velocity of particle i at time t+1. Let be the velocity of particle i at time t. This is the inertia factor, set according to the actual situation. Let g be the position of particle i at time t, gBest be the global optimal solution at time t, and pBest be the position of particle i at time t. i This is the local optimum for particle i. Then, the position is updated:

[0013] in, Let i be the position of particle i at time t+1; after updating the position, recalculate the global and local optima, and update the velocity and position again.

[0014] During optimization, different control variables are fed into the quality inspection module to obtain the predicted product quality. If the predicted product quality is lower than the set quality threshold, a large loss will occur. If it is higher than the set quality threshold, the optimization result will converge to the control parameter that is closest to the set quality. That is, the control target is always kept above the set quality threshold, and being close to the target quality will not produce unnecessary losses.

[0015] The stability monitoring module is used to determine the stability points required for edge optimization control based on the process analysis and time series analysis of the multi-effect multi-stage distillation column system, and set them as monitoring points. It dynamically calculates the stability value of each monitoring point in real time, and obtains the stability judgment result of the multi-effect multi-stage distillation column system based on the stability value of each monitoring point. Monitoring points include: top temperature related points, top product related points, bottom temperature related points, side stream product flow rate related points, feed flow rate related points, bottom temperature related points, heat source related points, heat exchanger related points, product reflux related points, and tray temperature related points, etc. In actual production, the decision to perform edge optimization is based on real-time stability. If the stability does not meet the requirements, the process is paused and waited until the system stabilizes or the timeout period is reached. Taking triple-effect distillation as an example, modifications to the steam valve and product reflux during distillation can significantly affect the reflux tank level. Specifically, product quality is controlled by adjusting the steam valve and product reflux ratio. During the distillation stage, the steam valve affects the top temperature of the column. The temperature affects how much water is carried out during alcohol distillation, thus affecting the alcohol content of the product. The reflux ratio affects how much intermediate material participates in distillation and how much participates in reflux, thus affecting the degree of distillation of the product. Therefore, the top reflux tank level and the top temperature difference level are used as the system stability standard. For other distillation systems, the monitoring points need to be determined in conjunction with the process and timing analysis modules. Real-time dynamic calculation of the stability value of each monitoring point includes: calculating the stability value of each monitoring point based on the sigma criterion; for each monitoring point, according to the preset monitoring duration and data sampling frequency (e.g., extracting historical data from the previous 6 hours every 6 hours), calculating the mean and standard deviation of the historical data for the monitoring point; calculating the upper and lower limits of the monitoring point using the set sigma coefficient; and updating the point attributes in real time. Point attributes include: point category, point time series data, point mean, point standard deviation, and point upper and lower limits; where point category refers to whether the monitoring point is a control variable or an observed variable; the formulas for calculating the upper and lower limits of the monitoring point include: , Where Max is the upper limit of the number of monitoring points, Min is the lower limit of the number of monitoring points, mean is the mean, and std is the standard deviation.

[0016] In actual control, based on the preset monitoring duration and data sampling frequency, data for the corresponding time period of the monitoring point is extracted, and the percentage of data within the upper and lower limits during that period is calculated. This percentage is used as the stability of the monitoring point. The numerical calculation formula for the stability of the monitoring point includes: , Where N is the length of the data within the time period. For counting, that is, the proportion within the upper and lower limits during this period, the stability value ranges from [0,1].

[0017] For example, if the monitoring duration is 10 minutes and the monitoring frequency is 1 minute, then the data of the monitoring point over the past 10 minutes will be extracted every minute, and the proportion within the upper and lower limits within 10 minutes will be calculated as the stability of the loop.

[0018] System stability is determined by the average stability of all monitored loops.

[0019] The stability assessment result of the multi-effect multi-stage distillation column system is obtained by calculating the stability value at each monitoring point. This includes setting a stability threshold for each monitoring point. If the stability values ​​at all monitoring points are greater than their respective stability thresholds, the stability assessment result of the multi-effect multi-stage distillation column system is considered stable; otherwise, the stability assessment result is considered unstable. The stability threshold can be obtained empirically, and the stability threshold for each monitoring point can be different.

[0020] The time-series analysis module is used to identify variables affecting product quality using time-series correlation analysis. Based on the process analysis of the multi-effect multi-stage distillation column system, the variables are categorized as either control variables or observed variables. Observed variables are those highly correlated with product quality but cannot be directly controlled, while control variables are those highly correlated with product quality and whose processes allow for modification. The module also works with the quality inspection module to obtain the duration of influence of control variables affecting product quality. The time-series correlation analysis methods in the time-series analysis module can be implemented using existing technologies such as regression analysis and Pearson correlation, which will not be elaborated upon here. Observed variables include all variables related to product quality. Quantities include, for example, steam valve location, top temperature difference, top temperature, top pressure, bottom temperature difference, bottom temperature, bottom pressure, tray temperature difference, tray temperature, packing temperature, product outlet valve, steam valve, reflux ratio, feed flow rate, etc.; control variables are the controllable variables among the observed variables; if process conditions are inconsistent, the control variables will also be inconsistent. For example, the most direct way to increase the heat source is to increase the steam valve, that is, to use the steam valve as the control variable. However, not all production processes allow direct adjustment of the steam valve. Some production areas form a cascade loop between the steam valve and the bottom temperature difference, and indirectly control the steam valve through the change of the bottom temperature difference. The test control object is the bottom temperature difference setpoint rather than the steam valve.

[0021] In conjunction with the quality inspection module, the influence time of control variables affecting product quality is obtained, including: obtaining the first influence time of each control variable based on the process experience of the multi-effect multi-stage distillation column system, and a large time interval of the first influence time; iterating the time control window of each variable, calculating the correlation between its mean and quality score, and taking the time window with the highest correlation as the second influence time, i.e. the final influence time.

[0022] Taking triple-effect distillation as an example, the following parameters are used as inputs to the soft sensing model: column top temperature difference, column top temperature, column top pressure, column bottom temperature difference, column bottom temperature, tray temperature, and packing temperature. The time range of each parameter is as follows: based on the current time, the influence range of column top temperature difference is from the first 90 minutes to the first 60 minutes; column top temperature is from the first 90 minutes to the first 60 minutes; column top pressure is from the first 30 minutes to the first 20 minutes; column bottom temperature difference is from the first 85 minutes to the first 55 minutes; column bottom temperature is from the first 85 minutes to the first 55 minutes; packing temperature is from the first 40 minutes to the first 30 minutes; and tray temperature is from the first 100 minutes to the first 90 minutes.

[0023] The quality inspection module includes a soft measurement model, which takes input control variables and observed variables and outputs a predicted product quality. The soft measurement model is a multi-input single-output neural network, implemented using existing technologies such as a BP neural network, and trained using the Adam optimizer with mean squared error.

[0024] Example 2 In this embodiment, the production process of a three-tower, three-effect distillation system is taken as an example. According to the production process, the three towers are pre-tower C2501, primary distillation tower C2503, and secondary distillation tower C2502. C2502 is the product collection tower. The collection valve and reflux ratio are automatically adjusted by PID based on the set value of the temperature difference at the top of the tower. The steam heat source is at the bottom of the secondary distillation tower. The steam valve is automatically adjusted by PID based on the set value of the temperature difference at the bottom of the tower. That is, the controlled objects are the temperature difference at the top and the temperature difference at the bottom of C2502.

[0025] Based on process experience, the temperature at the top of the column affects the amount of water carried away during distillation, thus affecting the alcohol content of the product. The reflux ratio affects how much intermediate material participates in distillation and how much participates in reflux, thus affecting the completeness of distillation of the product. Therefore, monitoring the liquid level in the reflux tank and the temperature at the top of the column is used as a standard for system stability. During production, data from the past 6 hours is taken at this point every 6 hours, the mean and standard deviation are calculated, and the upper and lower limits of the point are calculated according to the following formula and updated in the point attribute. In this embodiment, sigma is set to 6.

[0026]

[0027] Given that the system data sampling frequency is 3 seconds and chemical detection is performed every 2 hours, there are 12 quality data points and 28,800 production data points per day. Through process experience and correlation analysis, the characteristics affecting product quality and their time intervals are determined as follows: Top temperature difference (90-60 minutes); Top temperature (90-60 minutes); Top pressure (30-20 minutes); Bottom temperature difference (85-55 minutes); Bottom temperature (85-55 minutes); Packing bed temperature (40-30 minutes); Plate temperature (100-90 minutes).

[0028] Using quality data as labels and the time of each quality data point as the baseline, the average of the time intervals corresponding to each feature is taken to form the following training data:

[0029] Each feature is used as input to a BP neural network, and the output is product quality. The model is trained using the mean squared error with the Adam optimizer, and the trained model is used as a soft measurement model.

[0030] Based on process experience, the controlled variables are the temperature difference at the top and bottom of the column, while other variables are observed variables. In actual production, this embodiment sets the control frequency to 1 minute. Every minute, a stability analysis is performed. Specifically, the time-series data of the column top temperature and the reflux tank liquid level point over the past 10 minutes are taken, along with the latest upper and lower limits of these points. Then, the stability of all points is calculated using the following formula:

[0031] Where N is the data length, and count is the count, which calculates the proportion of data within the upper and lower limits during this period. The lower limit of stability for both points is set to 0.7. That is, when the stability of both points is greater than 0.7, the system is considered stable and edge optimization can be performed; otherwise, it is skipped, and stability is checked again on the next trigger. The timeout is set to 1 hour. If the system is still unstable after 1 hour, edge optimization is forced.

[0032] Then, the mean value of each feature of the observed object is taken based on the current time. Next, particle swarm optimization is performed on the control objects, with a population size of 10. That is, 10 sets of control object values ​​are randomly generated as the initial population. For example, (2.5,3) represents an average temperature difference of 2.5 at the top of the tower and an average temperature difference of 3 at the bottom of the tower. Each set of control objects represents one population particle. Then, its corresponding speed is randomly generated. This represents the initial search direction for the particle. Then, for each particle... Search for local optima in its vicinity Specifically, in this embodiment, the upper and lower limits of the domain are restricted to 10%, that is, for particle (2.5,3), the average temperature difference at the top of the tower ranges from 2.25 to 2.75, and the average temperature difference at the bottom of the tower ranges from 2.7 to 3.3. Ten domain particles are regenerated within this range. Each domain particle is concatenated with the average of the observed variables and then input into the soft sensor model to obtain the predicted quality. Calculate the loss using the following formula:

[0033] Where y is the set quality threshold, which is set to 92.3 in this embodiment, and a is the penalty coefficient, which will result in a large loss when the predicted quality of the product is lower than the set quality.

[0034] After calculating the loss of the neighborhood particles, the neighborhood particle with the smallest loss is selected as the initial population particle. Local optima Then, the loss is calculated for each initial population particle, and the particle with the smallest loss is the global optimum gBest. Then, the velocity and position of each initial population particle are updated according to the following formula.

[0035]

[0036]

[0037] The superscript t+1 represents the next update, and the superscript t represents the current update. The inertia factor is set to 0.5 in this embodiment. The individual learning factor is set to 0.3 in this embodiment. The population learning factor is set to 0.7 in this embodiment, and r is a random number between 0 and 1 to increase the randomness of the system and help expand the search space. Repeat the above steps until convergence or the maximum number of parameter searches is reached. The initial population particle with the smallest loss at this point is the optimal control parameter.

[0038] After obtaining the optimal control parameters under the current production environment, this embodiment sets the upper limit for parameter adjustment to 0.1, and updates the parameters according to the following formula: , in, These are the control parameters obtained in this optimization process. If the parameter value is greater than the upper limit before optimization, the parameter will be updated in the direction of the optimization parameter adjustment, with the limit as the step size.

[0039] In the embodiments of this application, the terms "first" and "second" (if they exist) are used only as name identifiers and do not represent the order of first and second.

[0040] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A smart edge-finding system for multi-effect distillation under varying operating conditions, characterized in that, It includes an edge-finding optimization module and a stability monitoring module, a time-series analysis module, and a quality inspection module connected to the edge-finding optimization module; wherein, the time-series analysis module is connected to the stability monitoring module and the quality inspection module; The edge-finding optimization module is used for edge-finding control, including: at the start of production, obtaining the stability judgment result of the current multi-effect multi-stage distillation column system from the stability monitoring module; if the system is judged to be stable, obtaining the control variables and observed variables of the current multi-effect multi-stage distillation column system from the time series analysis module; using the particle swarm optimization algorithm to search for parameters of the control variables, and using the quality inspection module to obtain the predicted product quality; using the particle swarm optimization algorithm to iteratively update the control parameters through the optimization function to obtain the control variable parameters, and obtaining all the control variable parameters of the current system as several first control variables; and obtaining the optimal control variables of the current system according to the preset parameter update formula and the first control variables. The stability monitoring module is used to determine the stability points required for edge optimization control based on the process analysis and time sequence analysis of the multi-effect multi-stage distillation column system, and set them as monitoring points; it dynamically calculates the stability value of each monitoring point in real time, and obtains the stability judgment result of the multi-effect multi-stage distillation column system based on the stability value of each monitoring point. The time series analysis module is used to obtain variables affecting product quality using time series correlation analysis methods, and to determine whether the process analysis variables of the multi-effect multi-stage distillation column system are control variables or observed variables. Observed variables refer to relevant variables that are highly correlated with product quality but cannot be directly controlled, while control variables refer to relevant variables that are highly correlated with product quality and whose processes can be modified. The module also works with the quality inspection module to obtain the influence time of control variables affecting product quality. The quality control module includes a soft measurement model, which takes input control variables and observed variables and outputs a predicted product quality.

2. The intelligent edge-finding system for multi-effect distillation under varying operating conditions according to claim 1, characterized in that: The default parameter update formula is: , in, For the current control variable, For the control variables of the previous time step, This represents the maximum step size.

3. The intelligent edge-finding system for multi-effect distillation under varying operating conditions according to claim 1, characterized in that: The implementation process of the particle swarm optimization algorithm includes: After initializing several populations, the position of each population is the search parameter. The velocity is randomly initialized as its initial search direction. Then, the control variables are combined with the observed variables and put into the quality inspection module to obtain the predicted product quality. Then, the population points are iterated according to the optimization function to obtain the optimal control variable parameters.

4. The intelligent edge-finding system for multi-effect distillation under varying operating conditions according to claim 3, characterized in that: The formulas for calculating the optimization function include: ， in, For the quality inspection module to predict product quality, To set a quality threshold, This is the penalty coefficient.

5. The intelligent edge-finding system for multi-effect distillation under varying operating conditions according to claim 1, characterized in that: The monitoring points include: Points related to the temperature at the top of the tower, points related to the temperature at the top of the tower, points related to the temperature at the bottom of the tower, points related to the flow rate at the side stream, points related to the feed flow rate, points related to the temperature at the bottom of the tower, points related to the heat source, points related to the heat exchanger, points related to the product reflux, and points related to the temperature between the trays.

6. The intelligent edge-finding system for multi-effect distillation under varying operating conditions according to claim 1, characterized in that: The real-time dynamic calculation of the stability value of each monitoring point includes: calculating the stability value of each monitoring point based on the sigma criterion; for each monitoring point, calculating the mean and standard deviation of the historical data of the monitoring point according to the preset monitoring duration and data sampling frequency, calculating the upper and lower limits of the monitoring point through the set sigma coefficient, and updating the point attributes in real time.

7. The intelligent edge-finding system for multi-effect distillation under varying operating conditions according to claim 6, characterized in that: The formulas for calculating the upper and lower limits of monitoring points include: ， Where Max is the upper limit of the number of monitoring points, Min is the lower limit of the number of monitoring points, mean is the mean, and std is the standard deviation.

8. The intelligent edge-finding system for multi-effect distillation under varying operating conditions according to claim 6, characterized in that, The formulas for calculating the stability of monitoring points include: ， Where N is the length of the data within the time period. For counting, that is, the proportion within the upper and lower limits during this period, Max is the upper limit of the monitoring points, and Min is the lower limit of the monitoring points.

9. The intelligent edge-finding system for multi-effect distillation under varying operating conditions according to claim 1, characterized in that, In conjunction with the quality inspection module, the influence time of control variables affecting product quality is obtained, including: obtaining the first influence time of each control variable based on the process experience of the multi-effect multi-stage distillation column system, and a large time interval of the first influence time; iterating the time control window of each variable, calculating the correlation between its mean and quality score, and taking the time window with the highest correlation as the second influence time, i.e. the final influence time.

10. The intelligent edge-finding system for multi-effect distillation under varying operating conditions according to claim 1, characterized in that, The stability judgment result of the multi-effect multi-stage distillation column system is obtained by calculating the stability value of each monitoring point. This includes setting a stability threshold for each monitoring point. If the stability value of all monitoring points is greater than its stability threshold, the stability judgment result of the multi-effect multi-stage distillation column system is stable; otherwise, the stability judgment result of the multi-effect multi-stage distillation column system is unstable.

Images