Rectifying tower operation optimization method based on big data

Abstract

The invention provides a rectifying tower operation optimization method based on big data. According to the method, sensors are installed in a rectifying tower to obtain pressure, temperature and liquid level data in real time; constructing three data-driven sub-models, namely a temperature-flow sub-model, a liquid level-pressure sub-model and a heating power-temperature-liquid level comprehensive sub-model, respectively analyzing the relationship between the temperature and the flow, the relationship between the liquid level and the pressure and the relationship between the heating power and the temperature and the liquid level, and outputting a sensitivity coefficient and a predicted value; based on the output of the sub-models, constructing a reinforcement learning model, defining a state space, a reward function and an action space, identifying key influence factors through comprehensive performance indexes, and proposing a control method; after control is implemented, new operation data are collected in real time and fed back to the model, parameters are automatically updated, self-learning and adaptation are achieved, and the control method is continuously optimized so that the optimal operation state of the rectifying tower can be achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial automation and control, and specifically to an optimization method for the operation of a rectification column based on big data. Background Art

[0002] In a rectification column, it usually operates relying on fixed process parameters and preset control logics. These systems obtain basic data inside the column through sensors and then adjust according to preset thresholds. However, this control method based on fixed parameters has obvious limitations. It cannot fully consider the complex interactions among various variables during the rectification process. In addition, when facing dynamic changes in the operating conditions of the rectification column, such as fluctuations in raw material composition and changes in environmental temperature, traditional control systems often have difficulty making rapid and accurate adjustments, resulting in unstable product quality, increased energy consumption, and low operating efficiency.

[0003] The prior art is not refined and intelligent enough in monitoring and controlling the operating state of the rectification column. Traditional control systems lack the ability to deeply analyze and mine a large amount of historical data and cannot predict the dynamic change relationships among variables. This makes it that in actual operation, operators often need to rely on experience and manual adjustment to cope with complex working condition changes, which is not only inefficient but also difficult to achieve the optimal operating state of the rectification column. Summary of the Invention

[0004] (I) Technical Problems to be Solved Aiming at the deficiencies of the prior art, the present invention provides an optimization method for the operation of a rectification column based on big data, and solves the problem of how to realize the real-time monitoring and optimal control of the operating state of the rectification column by constructing a sub-model and a reinforcement learning model based on big data.

[0005] (II) Technical Solutions To achieve the above objectives, the present invention is realized through the following technical solutions: An optimization method for the operation of a rectification column based on big data, including: S1. Install pressure sensors, temperature sensors, and liquid level sensors on the condenser, reboiler, and each tray inside the rectification column respectively to obtain the pressure, temperature, and liquid level data of the rectification column in real time; S2. Build three data-driven sub-models, namely the temperature-flow sub-model, which analyzes the historical data of the bottom temperature and the material flow rate, outputs the sensitivity coefficient of the temperature to the flow rate change, and the temperature prediction values under different flow rate conditions; the liquid level-pressure sub-model, which analyzes the interaction between the liquid level and the tower pressure, outputs the sensitivity coefficient of the liquid level to the pressure change, and the liquid level prediction values under different pressure conditions; the heating power-temperature-liquid level comprehensive sub-model, which considers the comprehensive influence of the heating power on the bottom temperature and the liquid level, outputs the sensitivity coefficients of the heating power change to the temperature and the liquid level, and the joint prediction values of the temperature and the liquid level under different heating power conditions; S3. Based on the outputs of the sub-models, build a reinforcement learning model, comprehensively consider the operating objectives of the distillation column, output the comprehensive performance indicators during the operation of the distillation column, and identify the operating parameters that have the greatest impact on the comprehensive performance indicators among the key influencing factors; propose a control method during the operation according to the comprehensive performance indicators to achieve the optimal operating state of the distillation column; S4. After implementing the control method, collect new data in real time, feedback it to the reinforcement learning model and the sub-models, and automatically update the parameters of all models, enabling them to self-learn and adapt according to the new data, and continuously optimize the control method.

[0006] Preferably, pressure sensors, temperature sensors and liquid level sensors are installed at the inlet and outlet of the condenser, in the heating area of the reboiler, and at appropriate positions on each tray in the tower; the installation positions of these sensors can accurately capture the pressure, temperature and liquid level changes at these key positions; for example, sensors are installed at the inlet and outlet of the condenser to monitor the temperature and pressure changes of the material during the condensation process in real time, so as to accurately control the condensation effect; sensors are installed in the heating area of the reboiler to timely feedback the influence of the heating power on the bottom temperature and the liquid level, ensuring the stability and efficiency of the heating process; all the sensors used have high precision, high sensitivity and good anti-interference ability, enabling the sensors to adapt to the complex operating environment of the distillation column; high sensitivity ensures that the sensors can quickly respond to small changes in physical quantities and timely capture the subtle changes during the operation of the distillation column, providing data support for real-time control.

[0007] Preferably, build a temperature-flow sub-model; collect a large amount of historical data of the bottom temperature and the material flow rate. Let the bottom temperature be T and the material flow rate be F, and their relationship is expressed as , where α and β are known constants used to adjust the weights of the linear and logarithmic terms. The constants α and β are determined by regression analysis. Collect the data of the bottom temperature and the material flow rate, calculate ln(F), construct the design matrix and the response vector, and use the least squares method to solve for α and β; obtain the sensitivity coefficient of the temperature to the flow rate change by taking the derivative , the sensitivity coefficient represents the response degree of temperature to flow rate changes under different flow rate conditions. For example, when the flow rate is small, the influence of the logarithmic term is large, and the sensitivity of temperature to flow rate changes is high; while when the flow rate is large, the influence of the linear term gradually dominates, and the sensitivity decreases relatively. Sensitivity analysis provides a basis for optimizing the control method; as well as the predicted temperature values under different flow rate conditions , where F new is the new flow rate input value, capturing the dynamic change law between temperature and flow rate. For example, in the rectification process, if it is necessary to increase the material flow rate to increase the output, the traditional control system may not be able to accurately predict the impact of this adjustment on the bottom temperature of the tower, which may lead to too high or too low temperature and affect the product quality. However, the temperature-flow sub-model can accurately predict the temperature change, enabling the operator to adjust the heating power or other relevant parameters in advance.

[0008] Preferably, a liquid level-pressure sub-model is constructed; deeply analyze the interaction between the liquid level and the pressure inside the tower, considering the change of the hydrostatic pressure of the fluid inside the tower caused by the liquid level change, which affects the pressure distribution, and thus construct the liquid level-pressure sub-model; let the liquid level be L and the pressure inside the tower be P, and establish the coupling equation between the two , where ρ is the density of the fluid, g is the acceleration due to gravity, and P 0 is the initial pressure when the liquid level is zero, so taking the derivative of it obtains the sensitivity coefficient of the liquid level to the pressure change , this sensitivity coefficient indicates that a small change in the liquid level will cause a linear change in the pressure inside the tower. For example, when the liquid level increases, the pressure inside the tower will increase accordingly, and vice versa; as well as the predicted liquid level values under different pressure conditions , where Pnew is the new pressure input value; for example, when it is necessary to regulate the pressure inside the tower to optimize the separation effect, the operator uses this sub-model to quickly predict the required liquid level adjustment amount, thereby achieving more accurate and efficient control.

[0009] Preferably, a heating power-temperature-liquid level comprehensive sub-model is constructed, which is established through the influence of the heating power Q on the bottom temperature T and the liquid level L of the tower. The heating power not only directly affects the bottom temperature of the tower, but also affects the liquid level by changing the thermal balance state inside the tower; assume that the relationship between the heating power Q and the bottom temperature T and the liquid level L is described by a linear combination, that is, Q = aT + bL + c, where a, b, and c are known constants, representing the sensitivity of the heating power to temperature and liquid level and a constant term respectively; where a is determined by measuring the change of the bottom temperature under different heating powers, b is determined by measuring the change of the liquid level under different heating powers, and c is determined by fitting the measurement data; taking the partial derivative of Q obtains the sensitivity coefficient and , representing the response degree of the heating power to the temperature and liquid level changes; at a given new temperature Tnew and liquid level L new When, Q pred =aT new +bL new +c The heating power required under different temperature and liquid level conditions is obtained through this formula, so as to optimize the operation of the distillation column; for example, during the operation of the distillation column, when the bottom temperature needs to be adjusted, the operator uses this model to quickly predict the required adjustment amount of the heating power and then makes the adjustment.

[0010] Preferably, the reinforcement learning model defines its state space S, reward function R, and action space A. The state space consists of the outputs of the temperature-flow sub-model, liquid level-pressure sub-model, and heating power-temperature-liquid level comprehensive sub-model; the state space includes the sensitivity coefficient of temperature to flow rate change , the sensitivity coefficient of liquid level to pressure change , the sensitivity coefficients of heating power change to temperature and liquid level and , as well as the corresponding predicted values T pred (F), L pred (P), T pred (Q), and L pred (Q); the action space consists of the operating parameters of the distillation column, including the heating power Q and the material flow rate F; the reward function is designed according to the operating objectives of the distillation column, such as product purity, equipment life, and energy consumption, and is used to evaluate the operating state under the current operating parameters; the core of the reinforcement learning model is a comprehensive performance index J, and the calculation formula of the comprehensive performance index J is , where w1 and w2 are weight coefficients, representing the importance of the sensitivity coefficient and the predicted value respectively, and are allocated according to the impact on the operation of the distillation column. Those with a direct impact on the distillation column have a high weight; α, β, γ, and δ are the weight coefficients of the sub-model outputs, used to balance the importance of the outputs of different sub-models, and are allocated according to the operating optimization objectives of the distillation column, where the objectives include the service life of the distillation column equipment and the operating efficiency of the distillation column; at each time step, the model selects an action according to the current state. After executing this action, the current state transfers to a new state and obtains the corresponding reward. The reinforcement learning model updates the comprehensive performance index according to the reward; for example, during the operation of the distillation column, if the current state shows that the bottom temperature is lower than the target value, the reinforcement learning model increases the heating power to increase the temperature; after executing this adjustment action, the effect is evaluated according to the new state and the reward value. If the temperature increases and the energy consumption is reasonable, the reinforcement learning model will continue to optimize this control method; if the effect is not good, the control method will be adjusted; in this way, the reinforcement learning model can dynamically adjust the operating parameters to adapt to different operating conditions.

[0011] Preferably, the comprehensive performance index is used to evaluate the operation effect of the distillation column. It quantifies the influence of each operating parameter on the operating state of the distillation column. The comprehensive performance index changes dynamically with the change of operating parameters, and the change range directly reflects the importance of the corresponding operating parameter. For example, if adjusting the heating power causes a significant increase in the comprehensive performance index, it indicates that the heating power is a key factor affecting the operating state of the distillation column. On the contrary, if changing the material flow rate has little effect on the comprehensive performance index, then the material flow rate is not a key influencing factor. The evaluation method based on the comprehensive performance index follows the logical path from evaluation to adjustment. By monitoring the performance of the comprehensive performance index under different operating parameters, the key factors that have the greatest impact on the operating state of the distillation column are identified. According to the positive or negative impact of these key factors on the comprehensive performance index, corresponding control methods are formulated. If the increase of a certain key factor can improve the comprehensive performance index, then the control method will tend to increase the set value of this parameter. On the contrary, if increasing this factor will cause the comprehensive performance index to decrease, the control method will be adjusted to reduce its value. For example, assume that during the operation of the distillation column, when the heating power increases from 100 kW to 120 kW, the comprehensive performance index increases from 0.7 to 0.85, which indicates that the increase in heating power significantly improves the operation effect of the distillation column. Therefore, the control method will tend to appropriately increase the set value of the heating power during operation. On the contrary, if the material flow rate increases from 100 m³ / h to 120 m³ / h and the comprehensive performance index decreases from 0.7 to 0.65, it indicates that the increase in the material flow rate has a negative impact on the operation effect of the distillation column. Therefore, the control method will be adjusted to reduce the set value of the material flow rate.

[0012] Preferably, after implementing the control method, new data will be generated during the operation of the distillation column. These new data are collected in real time and input as feedback into the reinforcement learning model and each sub-model. The model is not only optimized based on the initial data, but also self-updated and adjusted according to the real-time operation data. The reinforcement learning model uses these new data to automatically update its internal weights and sensitivity coefficients. For example, if the new data shows that the influence of the change in heating power on the bottom temperature of the column is different from the previous model prediction, the reinforcement learning model will adjust the relevant sensitivity coefficients to more accurately reflect this change. At the same time, each sub-model will also learn and adapt according to the new data. For example, the temperature-flow sub-model updates its prediction model based on the new flow rate and temperature data to improve the prediction accuracy. Through continuous feedback and learning, the parameters of all models are continuously updated, enabling the reinforcement learning model to more accurately reflect the actual operating state of the distillation column. Based on the updated model parameters, more accurate and effective control methods are proposed. For example, if the new data shows that under a certain operating condition, the product purity can be further improved by fine-tuning the heating power, the set value of the heating power will be automatically adjusted to achieve this optimization goal.

[0013] (III) Beneficial effects The present invention provides an optimization method for the operation of a rectification column based on big data, having the following beneficial effects: 1. By constructing a big-data-driven sub-model and a reinforcement learning model, the present invention accurately captures the dynamic change relationships among key variables such as temperature, flow rate, liquid level, and pressure, enabling the rectification column to predict potential operation risks in advance, adjust operation parameters in a timely manner, and significantly improve the stability and safety of the rectification column operation.

[0014] 2. The present invention realizes the intelligent control of the rectification column operation. The reinforcement learning model automatically identifies key influencing factors according to comprehensive performance indicators and formulates control methods accordingly, effectively reducing energy consumption and improving production efficiency.

[0015] 3. The continuous feedback learning mechanism of the present invention enables the control method to continuously adapt to these changes, maintain accurate monitoring and effective control of the rectification column operation state, contribute to extending the service life of the rectification column, reducing maintenance costs, and at the same time ensuring the continuity of the production process and the consistency of product quality. Description of the drawings

[0016] Figure 1 It is a schematic flow chart of the present invention. Detailed implementation manners

[0017] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0018] At the condenser, reboiler of the distillation column and each tray inside the column, select installation points and install pressure sensors, temperature sensors and liquid level sensors with high precision, high sensitivity and good anti-interference ability respectively; these sensors can output accurate data in real time and stably, accurately capturing the pressure, temperature and liquid level changes at key positions of the distillation column; for example, install pressure sensors and temperature sensors at the inlet and outlet of the condenser to monitor the operating state of the condenser in real time, providing basic data for subsequent data analysis and model construction; install corresponding sensors in the heating area of the reboiler to timely grasp the temperature and pressure changes during the heating process, thereby providing a basis for optimizing the heating power; and install liquid level sensors and temperature sensors at appropriate positions on each tray inside the column to monitor in detail the distribution and flow of materials inside the column, providing data for the precise control of the distillation process; collect the pressure, temperature and liquid level data of the distillation column in real time through the sensors, and use the data as the basis for subsequent data analysis and model construction, providing data resources for realizing the operation optimization of the distillation column.

[0019] Construct a temperature-flow sub-model; collect a large amount of historical data of the bottom temperature and material flow rate. Let the bottom temperature be T and the material flow rate be F, and their relationship is expressed as , where α and β are known constants used to adjust the weights of the linear and logarithmic terms. The constants α and β are determined through regression analysis. Collect the data of the bottom temperature and material flow rate, calculate ln(F), construct the design matrix and response vector, and use the least squares method to solve for α and β; obtain the sensitivity coefficient of the temperature to the flow rate change by taking the derivative , the sensitivity coefficient represents the response degree of the temperature to the flow rate change under different flow rate conditions. For example, when the flow rate is small, the influence of the logarithmic term is large, and the sensitivity of the temperature to the flow rate change is high; while when the flow rate is large, the influence of the linear term gradually dominates, and the sensitivity decreases relatively. The sensitivity analysis provides a basis for optimizing the control method; as well as the temperature prediction values under different flow rate conditions , where F new is the new flow rate input value, capturing the dynamic change law between the temperature and the flow rate; for example, during the distillation process, if it is necessary to increase the material flow rate to increase the output, the traditional control system may not be able to accurately predict the impact of this adjustment on the bottom temperature, which may lead to too high or too low temperature and affect the product quality, while the temperature-flow sub-model can accurately predict the temperature change, enabling the operator to adjust the heating power or other relevant parameters in advance.

[0020] Construct a liquid level-pressure sub-model; deeply analyze the interaction between the liquid level and the pressure inside the column, considering that the change of the liquid level causes the change of the hydrostatic pressure of the fluid inside the column, thereby affecting the pressure distribution, and thus construct a liquid level-pressure sub-model; let the liquid level be L and the pressure inside the column be P, and establish a coupling equation between the two , where ρ is the density of the fluid, g is the acceleration due to gravity, and P 0 is the initial pressure when the liquid level is zero. Therefore, taking the derivative of it gives the sensitivity coefficient of the liquid level to pressure changes . This sensitivity coefficient indicates that a small change in the liquid level will result in a linear change in the pressure inside the tower. For example, when the liquid level increases, the pressure inside the tower will increase accordingly, and vice versa; as well as the predicted values of the liquid level under different pressure conditions , where Pnew is the new pressure input value; for example, when it is necessary to adjust the pressure inside the tower to optimize the separation effect, the operator uses this sub-model to quickly predict the required liquid level adjustment amount, thereby achieving more precise and efficient control.

[0021] Construct a comprehensive sub-model of heating power - temperature - liquid level, which is established based on the influence of heating power Q on the bottom temperature T and liquid level L of the tower. The heating power not only directly affects the bottom temperature of the tower but also affects the liquid level by changing the thermal equilibrium state inside the tower; assume that the relationship between heating power Q and bottom temperature T and liquid level L is described by a linear combination, that is, Q = aT + bL + c. Among them, a, b, and c are known constants, which respectively represent the sensitivity of heating power to temperature and liquid level and a constant term; where a is determined by measuring the change in the bottom temperature of the tower under different heating powers, b is determined by measuring the change in the liquid level under different heating powers, and c is determined by fitting the measured data; taking the partial derivative of Q gives the sensitivity coefficient and , indicating the response degree of heating power to temperature and liquid level changes; at a given new temperature T new and liquid level L new , Q pred = aT new + bL new + c. The required heating power under different temperature and liquid level conditions is obtained through this formula, thereby optimizing the operation of the distillation column; for example, during the operation of the distillation column, when it is necessary to adjust the bottom temperature of the tower, the operator uses this model to quickly predict the required heating power adjustment amount and then makes the adjustment.

[0022] The reinforcement learning model defines its state space S, reward function R, and action space A. This state space is composed of the outputs of the temperature - flow sub-model, liquid level - pressure sub-model, and heating power - temperature - liquid level comprehensive sub-model; the state space includes the sensitivity coefficient of temperature to flow changes , the sensitivity coefficient of liquid level to pressure changes , the sensitivity coefficients of heating power changes to temperature and liquid level and , as well as the corresponding predicted values T pred (F), L pred (P), Tpred (Q) and L pred (Q); the action space consists of the operating parameters of the distillation column, including the heating power Q and the material flow rate F; the reward function is designed according to the operating objectives of the distillation column, including product purity and energy consumption, and is used to evaluate the operating state under the current operating parameters; the core of the reinforcement learning model is a comprehensive performance index J, and the calculation formula of the comprehensive performance index J is , where w1 and w2 are weight coefficients, representing the sensitivity coefficient and the importance of the predicted value respectively, and are allocated according to the influence on the operation of the distillation column; α, β, γ, and δ are weight coefficients of the sub-model outputs, used to balance the importance of different sub-model outputs, and are allocated according to the operating optimization objectives of the distillation column, where the objectives include the service life of the distillation column equipment and the operating efficiency of the distillation column; at each time step, the model selects an action according to the current state. After executing this action, the current state transfers to a new state and a corresponding reward is obtained. The reinforcement learning model updates the comprehensive performance index according to the reward; for example, during the operation of the distillation column, if the current state shows that the bottom temperature is lower than the target value, the reinforcement learning model increases the heating power to increase the temperature; after executing this adjustment action, the effect is evaluated according to the new state and the reward value. If the temperature increases and the energy consumption is reasonable, the reinforcement learning model will continue to optimize this control method; if the effect is not good, the control method is adjusted; in this way, the reinforcement learning model can dynamically adjust the operating parameters to adapt to different operating conditions.

[0023] Comprehensive performance indicators are used to evaluate the operation effect of the distillation column, which quantify the influence of various operating parameters on the operating state of the distillation column; the comprehensive performance indicators change dynamically with the change of operating parameters, and the change range directly reflects the importance of the corresponding operating parameters. For example, if adjusting the heating power results in a significant increase in the comprehensive performance indicators, this indicates that the heating power is a key factor affecting the operating state of the distillation column; on the contrary, if changing the material flow rate has little effect on the comprehensive performance indicators, then the material flow rate is not a key influencing factor; the evaluation method based on the comprehensive performance indicators follows the logical path from evaluation to adjustment. By monitoring the performance of the comprehensive performance indicators under different operating parameters, the key factors that have the greatest impact on the operating state of the distillation column are identified; according to the positive or negative impact of these key factors on the comprehensive performance indicators, corresponding control methods are formulated. If the increase of a certain key factor can improve the comprehensive performance indicators, then the control method will tend to increase the set value of this parameter. On the contrary, if increasing this factor will cause the comprehensive performance indicators to decline, the control method will be adjusted to reduce its value; for example, assume that during the operation of the distillation column, when the heating power increases from 100 kW to 120 kW, the comprehensive performance indicators increase from 0.7 to 0.85, which indicates that the increase in heating power significantly improves the operation effect of the distillation column. Therefore, the control method will tend to appropriately increase the set value of the heating power during operation; on the contrary, if the material flow rate increases from 100 m³ / h to 120 m³ / h and the comprehensive performance indicators decrease from 0.7 to 0.65, this indicates that the increase in the material flow rate has a negative impact on the operation effect of the distillation column; therefore, the control method will be adjusted to reduce the set value of the material flow rate.

[0024] After implementing the control method, new data is generated during the operation of the distillation column. The new data is collected in real time and fed back as input into the reinforcement learning model and each sub-model. The reinforcement learning model automatically updates itself using this data, which involves adjusting the internal weights and sensitivity coefficients of the reinforcement learning model. The sub-models also learn and adapt based on the new data. Through continuous feedback and learning, the parameters of all models are continuously updated, enabling the models to more accurately reflect the actual operating state of the distillation column. Based on the updated model parameters, a more precise and effective optimization control method is proposed. Suppose an enterprise wishes to improve the operating efficiency of the distillation column, reduce energy consumption, and simultaneously increase the purity of the product. By implementing the optimization method for the operation of the distillation column based on big data, high-precision sensors are first installed at key positions of the distillation column to collect operation data in real time. Then, three data-driven sub-models are constructed based on the collected data to model and analyze the relationships between temperature and flow rate, liquid level and pressure, and heating power and temperature and liquid level respectively. Next, based on the outputs of the sub-models, a reinforcement learning model is constructed. By defining the state space, reward function, and action space, intelligent control of the operating state of the distillation column is achieved. After implementing the control method, new data is collected in real time and fed back to the reinforcement learning model and sub-models to continuously update the model parameters and further optimize the control method. After a period of operation, the operating efficiency of the distillation column of this enterprise has been significantly improved, energy consumption has been reduced, and the product purity has also been increased.

[0025] In summary, the optimization method for the operation of the distillation column based on big data realizes real-time monitoring, accurate prediction, and intelligent optimization control of the operating state of the distillation column by constructing data-driven sub-models and a reinforcement learning model. Through effective control, the operating efficiency and product quality of the distillation column are improved.

[0026] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A distillation tower operation optimization method based on big data, characterized in that: include: S1. Install pressure sensors, temperature sensors and liquid level sensors on the condenser, reboiler and each layer of the tower in the distillation tower to obtain the pressure, temperature and liquid level data of the distillation tower in real time; S2. Construct three data-driven sub-models, namely, the temperature-flow sub-model, which analyzes the historical data of tower bottom temperature and material flow, outputs the sensitivity coefficient of temperature to flow change, and the temperature prediction value under different flow conditions; the liquid level-pressure sub-model, which analyzes the interaction between the liquid level and the pressure in the tower, outputs the sensitivity coefficient of the liquid level to pressure change, and the liquid level prediction value under different pressure conditions; the heating power-temperature-liquid level comprehensive sub-model, which considers the comprehensive influence of heating power on tower bottom temperature and liquid level, outputs the sensitivity coefficient of heating power change to temperature and liquid level, and the temperature and liquid level joint prediction value under different heating power conditions; S3. Based on the output of the sub-model, a reinforcement learning model is constructed, the operation objectives of the distillation tower are comprehensively considered, the comprehensive performance indicators of the distillation tower operation are output, and the operating parameters with the greatest impact on the comprehensive performance indicators by the key influencing factors are identified; Propose control methods in operation based on comprehensive performance indicators to achieve the optimal operating state of the distillation tower; S4. After the control method is implemented, new data is collected in real time and fed back to the reinforcement learning model and sub-models to automatically update the parameters of all models so that they can self-learn and adapt according to the new data and continuously optimize the control method.

2. The method for optimizing distillation tower operation based on big data according to claim 1, characterized in that: The installation points are selected at the inlet and outlet of the condenser, the heating area of ​​the reboiler, and the appropriate positions of each tower plate in the tower so that the sensors can accurately capture the pressure, temperature and liquid level changes at these key positions; the selected pressure sensors, temperature sensors and liquid level sensors all have high precision, high sensitivity and good anti-interference capabilities, adapt to the operating environment of the distillation tower, and output accurate data in real time and stably.

3. The method for optimizing distillation tower operation based on big data according to claim 1, characterized in that: The temperature-flow sub-model is constructed; a large amount of historical data of tower bottom temperature and material flow is collected, and the tower bottom temperature is T, and the material flow is F, and the relationship is expressed as follows: , where α and β are known constants used to adjust the weights of linear and logarithmic terms; the sensitivity coefficient of temperature to flow change is obtained by derivation , and the predicted temperature values ​​under different flow conditions , where F new Enter values ​​for new flow rates to capture the dynamic relationship between temperature and flow.

4. The method for optimizing distillation tower operation based on big data according to claim 1, characterized in that: The liquid level-pressure sub-model is constructed; the interaction between the liquid level and the pressure in the tower is deeply analyzed, and the change in the static pressure of the fluid in the tower caused by the change in the liquid level is considered to affect the pressure distribution, thereby constructing the liquid level-pressure sub-model; assuming that the liquid level is L and the pressure in the tower is P, a coupling equation between the two is established , where ρ is the density of the fluid, g is the acceleration of gravity, and P0 is the initial pressure when the liquid level is zero. Therefore, the sensitivity coefficient of the liquid level to the pressure change is obtained by taking the derivative , and the predicted liquid level under different pressure conditions , where P new Enter the new pressure value; this sub-model can accurately reflect the dynamic equilibrium relationship between liquid level and pressure, and provide support for the stable operation of the distillation tower.

5. The method for optimizing distillation tower operation based on big data according to claim 1, characterized in that: The construction of the heating power-temperature-liquid level comprehensive sub-model is established through the influence of the heating power Q on the tower bottom temperature T and the liquid level L. The heating power not only directly affects the tower bottom temperature, but also affects the liquid level by changing the thermal equilibrium state in the tower; the relationship between the heating power Q and the tower bottom temperature T and the liquid level L is described by a linear combination, that is, Q=aT+bL+c, wherein a, b and c are constants, representing the sensitivity of the heating power to the temperature and the liquid level and a constant term respectively; wherein a is determined by measuring the change of the tower bottom temperature under different heating powers, b is determined by measuring the change of the liquid level under different heating powers, and c is determined by fitting the measured data; the sensitivity coefficient is obtained by taking the partial derivative of Q and , which indicates the response of heating power to temperature and liquid level changes; given a new temperature T new and liquid level L new When Q pred =aT new +bL new +c This formula is used to obtain the required heating power under different temperature and liquid level conditions, thereby optimizing the operation of the distillation column.

6. The method for optimizing distillation tower operation based on big data according to claim 1, characterized in that: The reinforcement learning model defines its state space S, reward function R and action space A. The state space is composed of the outputs of the temperature-flow sub-model, the liquid level-pressure sub-model and the heating power-temperature-liquid level comprehensive sub-model; the state space includes the sensitivity coefficient of temperature to flow change , Liquid level sensitivity coefficient to pressure change , Sensitivity coefficient of heating power change to temperature and liquid level and , and the corresponding predicted value T pred (F), L pred (P), T pred (Q) and L pred (Q); The action space consists of the operating parameters of the distillation tower, including heating power Q and material flow F; the reward function is designed according to the operating objectives of the distillation tower, including product purity and energy consumption, and is used to evaluate the operating status under the current operating parameters; The core of the reinforcement learning model is a comprehensive performance index J. The calculation formula of the comprehensive performance index J is: , where w1 and w2 are weight coefficients, representing the importance of the sensitivity coefficient and the predicted value, respectively, and are allocated according to the impact on the operation of the distillation tower; α, β, γ, and δ are weight coefficients of the sub-model outputs, which are used to balance the importance of different sub-model outputs and are allocated according to the operation optimization target of the distillation tower; in each time step, the model selects an action according to the current state. After executing the action, the current state is transferred to the new state and obtains the corresponding reward. The reinforcement learning model updates the comprehensive performance index according to the reward.

7. The method for optimizing distillation tower operation based on big data according to claim 6, characterized in that: The comprehensive performance index quantifies the influence of each operating parameter on the operating effect of the distillation tower. During the operation, the comprehensive performance index changes with the change of the operating parameters, and the magnitude of the change reflects the importance of the corresponding operating parameters. When a certain operating parameter is adjusted, if the comprehensive performance index changes significantly, it means that the parameter has a greater influence on the operating state of the distillation tower, and is thus identified as a key influencing factor. The control method proposed for optimization during operation based on comprehensive performance indicators follows a logical path from evaluation to adjustment. By monitoring the performance of comprehensive performance indicators under different operating parameters, key influencing factors are determined. According to the positive or negative impact of key influencing factors on the comprehensive performance indicators, corresponding control methods are formulated. If the increase of a key influencing factor can improve the comprehensive performance indicators, the control method will tend to increase the set value of the parameter. On the contrary, if the increase of the key influencing factor causes the comprehensive performance indicators to decrease, the control method will be adjusted to reduce its value.

8. The method for optimizing distillation tower operation based on big data according to claim 1, characterized in that: After the control method is implemented, the operation of the distillation column generates new data, which is collected in real time and input into the reinforcement learning model and each sub-model as feedback; The reinforcement learning model uses these data to automatically update itself, which involves adjusting the internal weights and sensitivity coefficients of the reinforcement learning model; the sub-models will also learn and adapt based on the new data; through continuous feedback and learning, the parameters of all models are constantly updated, so that the model can more accurately reflect the actual operating status of the distillation tower; based on the updated model parameters, a more accurate and effective optimization control method is proposed.

Images