Method and system for controlling temperature of hydrocracking reactor

Abstract

The invention discloses a hydrocracking reactor temperature control method and system, and solves the problem that temperature runaway accidents cannot be avoided due to the fact that great temperature fluctuation cannot be accurately prevented in the prior art, and the method comprises the following steps: collecting characteristic variable data influencing the bed temperature of a hydrocracking reactor in real time, and performing preprocessing; on the basis of the characteristic variables, a multivariable time sequence prediction framework is adopted, a prediction model is constructed, and the temperature trend is predicted; constructing an intelligent control model, and performing optimization control on the bed temperature based on a prediction value output by the prediction model by taking smooth reaching of a set value as an optimization target; and an abnormity early warning model is constructed, and abnormal condition identification and abnormity early warning are carried out. According to the invention, the control of the bed temperature of the hydrocracking reactor can be optimized, the fluctuation amplitude of the bed temperature is reduced, the actual control online index of the reaction temperature is improved, overtemperature and temperature runaway can be prevented, and the coking and inactivation of the catalyst are avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of petrochemical production, and specifically provides a temperature control method and system for a hydrocracking reactor. Background Art

[0002] Hydrocracking is a highly exothermic reaction. As refining technology shifts toward lower-quality oil and higher conversion rates, higher requirements are placed on catalyst activity and selectivity. However, as activity increases, even subtle temperature changes are amplified by the catalyst reaction and reflected in the form of temperature. When the inlet bed temperature fluctuates significantly and deviates significantly from the set value, a catalyst "temperature runaway" accident can occur. Once a "temperature runaway" accident occurs, it can lead to catalyst deactivation, equipment damage, environmental pollution, personal injury, and social impact.

[0003] Existing temperature control methods primarily employ DCS (distributed control systems) control, which achieves regulation and control of the inlet temperature by connecting the amount of cold hydrogen in the bed with the inlet temperature in series. For example, a "reactor temperature control method and system" disclosed in Chinese patent literature, with publication number CN111408326A, is limited by the limitations of reaction kinetics and process simulation software and cannot accurately simulate the hydrocracking process. Because the reaction process is multivariable, strongly coupled, nonlinear, and has large time lags, it cannot be accurately characterized, predicted in advance, or prevented from large temperature fluctuations using traditional control models, resulting in the inability to avoid temperature runaway accidents. Summary of the Invention

[0004] In order to solve the problem in the prior art that large temperature fluctuations cannot be accurately prevented, resulting in unavoidable temperature runaway accidents, the present invention provides a temperature control method and system for a hydrocracking reactor, which can optimize the control of the bed temperature of the hydrocracking reactor, reduce the fluctuation amplitude of the bed temperature, and improve the actual control upper limit index of the reaction temperature. At the same time, it can prevent overheating and temperature runaway, and avoid catalyst coking and deactivation.

[0005] The specific solutions of the present invention are as follows.

[0006] In a first aspect, a method for controlling the temperature of a hydrocracking reactor comprises: S1: Real-time collection of characteristic variable data affecting the bed temperature of the hydrocracking reactor and preprocessing; S2: Based on the characteristic variables, a multivariate time series prediction framework is used to build a prediction model to predict the temperature trend; S3: constructing an intelligent control model, taking smooth reaching of the set value as the optimization goal, and optimizing the bed temperature based on the predicted value output by the prediction model; S4: Build an abnormal warning model to identify abnormal situations and issue abnormal warnings.

[0007] By integrating machine learning and control theory, we constructed prediction models, intelligent control models, and abnormal warning models, and realized the full closed-loop intelligent optimization of "prediction-control-warning", reducing the fluctuation range of the bed outlet temperature, achieving precise control of the cracking reaction temperature, reducing the by-product yield, and effectively reducing the gap between the final operating temperature of the catalyst and the set value, thereby increasing the service life of the catalyst. At the same time, by avoiding the occurrence of "temperature runaway" accidents, we effectively avoided catalyst coking and the generation of bed hot spots.

[0008] Furthermore, the characteristic variables include raw material temperature and feed amount and furnace feed temperature as well as circulating hydrogen amount and cold hydrogen amount; the data preprocessing includes: using Z-score normalization to eliminate dimension effects and standardize the data.

[0009] Specifically, the DCS system collects characteristic variable data in real time, sampling every 10 seconds, and preprocesses the data as input variables of the prediction model.

[0010] Furthermore, the input variables of the prediction model include the characteristic variables, and the output variables of the prediction model include the predicted value of the bed inlet temperature, the predicted value of the temperature rise rate, and the predicted value of the average temperature within the future time t1.

[0011] Specifically, based on the characteristic variables affecting the bed temperature, such as raw material temperature, feed rate, furnace feed temperature, circulating hydrogen amount, and cold hydrogen amount, a prediction model for the inlet temperature trend, temperature rise trend, and bed average temperature trend is established to predict the inlet temperature trend, temperature rise trend, and bed average temperature trend of each bed of the hydrocracking reactor in advance.

[0012] Furthermore, the prediction model was constructed based on LSTM, and the hidden layer state update formula and the final output layer formula were established. The prediction model was trained using three months of historical data, with a test set error (RMSE) of ≤0.8°C. The prediction model includes a hidden layer and a final output layer, and the hidden layer state update formula and the final output layer formula were established during the construction of the prediction model.

[0013] Furthermore, the intelligent control model introduces a dynamic weight adjustment mechanism and adopts a model predictive control optimization algorithm to roll-optimize the control quantity sequence within the future time t3 every t2 time, and solves the optimal solution through the gradient descent algorithm.

[0014] Through the feedback of the prediction model, an intelligent control model is established. The interdisciplinary fusion multi-driven control model of "machine learning + control theory + industry experience" is adopted. The smooth arrival of the set value is used as the optimization goal. The amount of cold hydrogen is optimized according to the predicted value, and the valve opening is adjusted through the DCS actuator. The appropriate control amount is used at the right time to ensure the stability of the bed temperature and temperature rise. It can accurately characterize the large hysteresis, nonlinearity, and multi-objective control process to reduce damage to the catalyst.

[0015] Furthermore, it also includes constructing an online optimization and learning model, which adopts a reinforcement learning framework, establishes an evaluation mechanism to optimize the intelligent control model, and performs model parameter adaptation.

[0016] Furthermore, the model parameter adaptation specifically includes: Constructing a state space S, wherein the state space S includes historical data of bed temperature and temperature rise rate and cold hydrogen amount; Define the action space A as the control amount adjustment range; Set the reward function based on the control amount penalty factor and stability reward coefficient; Based on the Q-learning algorithm update strategy, the model parameters are optimized online.

[0017] Specifically, the control amount adjustment range can be set to include {-10%, -5%, 0%, +5%, +10%}, and the reward function is activated when the temperature fluctuates ≤1°C for 5 consecutive minutes.

[0018] Furthermore, the abnormal warning model includes a temperature rise rate warning, which triggers a first-level warning when the temperature rise rate exceeds a first threshold, and triggers a second-level warning when the temperature rise rate exceeds a second threshold; the abnormal warning model also includes a temperature deviation warning, which triggers an adjustment instruction when the temperature deviation value exceeds a third threshold for t4 time.

[0019] The abnormal warning model adds warning functions such as temperature rise and temperature velocity on the basis of bed temperature warning. When the abnormal warning is triggered, the system automatically increases the amount of cold hydrogen according to the warning information. For example, when the first-level warning is triggered, the system automatically increases the amount of cold hydrogen by 5%, and when the second-level warning is triggered, the system automatically increases the amount of cold hydrogen by 10%.

[0020] Furthermore, the abnormal warning model also includes a composite warning, which integrates the temperature rise rate, temperature deviation value and historical trend, and uses fuzzy logic to judge the risk level, specifically: Risk_Score = u1(△T)*u2(dT / dt), where u1 and u2 are membership functions, assuming that the membership function is a triangular distribution.

[0021] In a second aspect, a hydrocracking reactor temperature control system implements the above-mentioned hydrocracking reactor temperature control method, comprising: A field data layer for acquiring field data of a hydrocracking reactor, wherein the field data includes the characteristic variable data; The bottom control layer regulates and controls the temperature of the hydrocracking reactor; The intelligent control layer performs feedforward control on the underlying control layer, and includes an intelligent prediction control module and an abnormal warning model. The intelligent prediction control module includes a prediction model, an intelligent control model, and an online optimization and learning model.

[0022] Therefore, the present invention has the following beneficial effects: A hydrocracking reactor temperature control method and system of the present invention adds an intelligent control layer to the existing hydrocracking reaction DCS control system. By constructing a prediction model, an intelligent control model, an online optimization and learning model, and an abnormal warning model, the fluctuation range of the bed outlet temperature is reduced, precise control of the cracking reaction temperature is achieved, and the by-product yield is reduced. By integrating machine learning and control theory, a full closed-loop intelligent optimization of "prediction-control-warning" is achieved, effectively reducing the gap between the final operating temperature of the catalyst and the set value, and improving the service life of the catalyst. At the same time, since the occurrence of "temperature runaway" accidents is avoided, the generation of catalyst coking and bed hot spots is effectively avoided. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0024] Figure 1 Schematic diagram of a temperature control method for a hydrocracking reactor according to the present invention.

[0025] Figure 2 The present invention is a flow chart of a temperature control method for a hydrocracking reactor. DETAILED DESCRIPTION

[0026] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0027] Example 1.

[0028] like Figure 1FIG2 is a schematic diagram of a temperature control method for a hydrocracking reactor according to the present invention. The temperature control method for a hydrocracking reactor according to this embodiment is specifically as follows.

[0029] First, data on characteristic variables influencing bed temperature is collected in real time. These include the hydrocracking reactor's feedstock temperature, feed rate, furnace feed temperature, circulating hydrogen flow, and cold hydrogen flow. In a preferred embodiment, this data is collected in real time via a DCS system, with a sampling interval of every 10 seconds.

[0030] The collected characteristic variable data affecting the bed temperature are pre-processed. In a preferred embodiment, the characteristic variable data are normalized using Z-score to eliminate the dimension effect and standardize the characteristic variable data.

[0031] A prediction model is constructed based on characteristic variables that affect bed temperature. Preprocessed characteristic variable data is used as input variables for the prediction model. The prediction model predicts temperature trends based on the characteristic variable data. This prediction includes predicting the inlet temperature trend, temperature rise trend, and average bed temperature trend of the hydrocracking reactor. Specifically, based on characteristic variables that affect bed temperature, such as feedstock temperature, feed rate, furnace feed temperature, circulating hydrogen rate, and cold hydrogen rate, a prediction model for inlet temperature trend, temperature rise trend, and average bed temperature trend is established. This model predicts the inlet temperature trend, temperature rise trend, and average bed temperature trend of each bed in the hydrocracking reactor in advance, outputting predicted values ​​for the bed inlet temperature, temperature rise rate, and average bed temperature within the future time period t1.

[0032] An intelligent control model is constructed to optimize the control of the bed temperature of the hydrocracking reactor based on the predicted value output by the prediction model. That is, an intelligent control model is established through feedback from the prediction model. The multi-driven intelligent control model with interdisciplinary integration of "machine learning + control theory + industry experience" is adopted to smoothly reach the set value as the optimization goal. The amount of cold hydrogen is optimized in a rolling manner according to the predicted value, and the valve opening is adjusted through the DCS actuator. The appropriate control amount is used at the right time to ensure the stability of the bed temperature and temperature rise. It can accurately characterize the control process with large hysteresis, nonlinearity, and multiple objectives, and reduce damage to the catalyst.

[0033] Construct an online optimization and learning model, and establish an evaluation mechanism to optimize the intelligent control model, that is, perform online optimization on the model parameters of the intelligent control model to achieve model parameter adaptation.

[0034] Construct an abnormal warning model to identify abnormal situations and issue abnormal warnings. On the basis of bed temperature warning, add warning functions such as temperature rise and temperature velocity, and execute abnormal response mechanism according to abnormal warning.

[0035] A hydrocracking reactor temperature control system of this embodiment implements the hydrocracking reactor temperature control method described above, and includes a field data layer, a bottom control layer, and an intelligent control layer. The field data layer acquires field data of the hydrocracking reactor, including temperature data, flow data, pressure data, and weight data, as well as characteristic variable data affecting the bed temperature, such as the hydrocracking reactor's raw material temperature, feed rate, furnace feed temperature, circulating hydrogen volume, and cold hydrogen volume; the bottom control layer can use a DCS control system to regulate and control the hydrocracking reactor's temperature; the intelligent control layer performs feedforward control on the bottom control layer, including an intelligent predictive control module and an abnormality warning model, wherein the intelligent predictive control module includes a prediction model, an intelligent control model, and an online optimization and learning model.

[0036] The present embodiment provides a hydrocracking reactor temperature control method and system, adding an intelligent control layer to the existing hydrocracking reactor DCS control system. By constructing a prediction model, an intelligent control model, an online optimization and learning model, and an anomaly warning model, the fluctuation range of the bed outlet temperature is reduced, achieving precise control of the cracking reaction temperature and reducing by-product yield. Based on characteristic variables affecting bed temperature, such as feedstock temperature, feed rate, furnace feed temperature, circulating hydrogen amount, and cold hydrogen amount, a prediction model for inlet temperature trend, temperature rise trend, and bed average temperature trend is established. This allows for early prediction of the inlet temperature trend, temperature rise trend, and bed average temperature trend of each bed in the hydrocracking reactor. Based on feedback from the prediction model, an intelligent control model is established. This multi-driven intelligent control model, which integrates a multidisciplinary approach combining machine learning, control theory, and industry experience, optimizes the smooth arrival of setpoints. Appropriate control variables are applied at the appropriate time, ensuring stable bed temperature and temperature rise. This model accurately characterizes large-hysteresis, nonlinear, and multi-objective control processes, minimizing catalyst damage. At the same time, an abnormal warning module is established, and on the basis of the bed temperature warning, warning functions such as temperature rise and temperature velocity are added. By integrating machine learning and control theory, a full closed-loop intelligent optimization of "prediction-control-warning" is realized, which effectively reduces the gap between the final operating temperature of the catalyst and the set value, and increases the service life of the catalyst. At the same time, since the "temperature runaway" accident is avoided, the catalyst coking and bed hot spots are effectively avoided. A hydrocracking reactor temperature control method and system of this embodiment can not only optimize the control of the bed temperature of the hydrocracking reactor, reduce the fluctuation amplitude of the bed temperature, and improve the actual control upper limit index of the reaction temperature, but also prevent overheating and temperature runaway, avoid catalyst coking and deactivation, and have significant economic benefits.

[0037] Example 2.

[0038] like Figure 2FIG2 is a flow chart of a temperature control method for a hydrocracking reactor according to the present invention. The temperature control method for a hydrocracking reactor according to this embodiment is specifically as follows.

[0039] First, the DCS system collects the characteristic variable data that affects the bed temperature in real time, sampling once every 10 seconds. Among them, the characteristic variables that affect the bed temperature include the feed temperature T of the hydrocracking reactor. in (t), feed rate F feed (t), furnace feed temperature T furnace (t), circulating hydrogen amount H cycle (t) and the amount of cold hydrogen H cool (t). The collected characteristic variable data are then normalized using Z-score normalization to eliminate the dimension effect.

[0040] Based on the standardized characteristic variable data, prediction models for inlet temperature trends, temperature rise trends, and bed average temperature trends were constructed. These trends were used to predict the inlet temperature, temperature rise trends, and bed average temperature trends of each bed in the hydrocracking reactor in advance. The prediction model was trained using three months of historical data, achieving a test set error (RMSE) of 0.8°C or less. This example employed a multivariate time series prediction framework and constructed a prediction model based on an LSTM (Long Short-Term Memory) network. The mathematical expression for the prediction model is as follows.

[0041] The input variables of the prediction model are the standardized feature variable data, namely X t =[T in (t), F feed (t), T furnace (t), H cycle (t), H cool (t)], where T in (t) is the raw material temperature (℃), which is the inlet temperature. feed (t) is the feed rate (kg / h), T furnace (t) is the furnace feed temperature (℃), H cycle (t) is the amount of circulating hydrogen (Nm 3 / h), H cool (t) is the amount of cold hydrogen (Nm 3 / h).

[0042] The output variables of the prediction model include T^ in (t+△t),△T^ rise (t+△t), T^ avg (t+△t), where T^ in (t+△t) is the predicted inlet temperature at time △t in the future (℃), △T^ rise(t+△t) is the predicted value of the temperature rise rate in the future △t time (℃ / min), T^ avg (t+△t) is the predicted average bed temperature at time △t in the future (℃).

[0043] The prediction model is built based on LSTM, including a hidden layer and a final output layer. When building the prediction model, the hidden layer state update formula and the final output layer formula are established. The LSTM hidden layer state update formula is as follows: f t =σ(W f ·[h t-1 ,X t ]+b f ), i t =σ(W i ·[h t-1 ,X t ]+b i ), ~C t =tanh(W C ·[h t-1 ,X t ]+b C ), C t =f t ⊙C t-1 +i t ⊙~C t , o t =σ(W o ·[h t-1 ,X t ]+b o ), h t =o t ⊙\tanh(C t ); The final output layer formula of LSTM is as follows: ~Y t =W y ·h t +b y ; Where σ is the Sigmoid function, ⊙ represents element-wise multiplication, and W and b are trainable parameters. In a preferred embodiment, the prediction module outputs T^ for the next 5 minutes every 1 minute. in and △T^ rise .

[0044] Then, an intelligent control model is constructed to optimize the bed temperature of the hydrocracking reactor based on the predicted value output by the prediction model. That is, an intelligent control model is established through feedback from the prediction model. The intelligent control model of this embodiment introduces a dynamic weight adjustment mechanism to achieve the set value smoothly as the optimization target. The specific formula of the objective function is as follows: y = min(a1△L1+a2△L2+a3△L3); △L1=‖T current -T set ‖ 2 , △L2=‖u k -u k-1 ‖ 2 , △L3=‖dT / dt‖ max ; a l =1 / {1+e -β·△L1}, a2=1-a1, a3=γ·a1; Where y represents the minimum value of the comprehensive deviation between the current state and the control target, △L1=‖T current -T set ‖ 2 (Square of temperature deviation) is the current bed temperature T current With the set temperature T set Euclidean distance, △L2=‖u k -u k-1 ‖ 2 is the control amount adjustment step, that is, the control amount u at adjacent moments k with u k-1 The difference between ∆L3 and dT is: max is the temperature rise rate constraint, a1, a2, and a3 are weight coefficients, β is the sensitivity coefficient, and its value range is [0.1, 1.0], and γ is the temperature rise penalty factor (the default is 0.2).

[0045] The intelligent control model of this embodiment adopts the model predictive control optimization algorithm (MPC), which optimizes the control quantity sequence within the next t3 time every t2 time, and solves the optimal solution through the gradient descent algorithm. In a preferred embodiment, the intelligent control model optimizes the control quantity sequence u within the next 15 minutes every 5 minutes. k .

[0046] In a preferred embodiment, the intelligent control model performs PID and MPC comparison and parameter adjustment in a simulation environment, selects the optimal parameters, and optimizes the cold hydrogen quantity H according to the predicted value. cool and adjust the valve opening through the DCS actuator.

[0047] This embodiment adopts an intelligent control model that integrates multiple drivers through the interdisciplinary integration of "machine learning + control theory + industry experience" to achieve the smooth arrival of the set value as the optimization goal, uses the appropriate control amount at the right time to ensure the stability of the bed temperature and temperature rise, and can accurately characterize large hysteresis, nonlinearity, and multi-objective control processes to reduce damage to the catalyst.

[0048] Then, an online optimization and learning model is constructed, and an evaluation mechanism is established to optimize the intelligent control model, that is, the model parameters of the intelligent control model are optimized online to achieve model parameter adaptation.

[0049] The online optimization and learning model of this embodiment adopts a reinforcement learning framework and performs model parameter adaptation based on the Q-learning algorithm update strategy.

[0050] The implementation of parameter adaptation specifically includes the following steps.

[0051] First, construct the state space S, which includes the bed temperature T current , temperature rise rate dT / dt and cold hydrogen quantity historical data H cool (t-3), H cool (t-2), H cool (t-1).

[0052] Then define the action space A as the control amount adjustment range △H cool , containing {-10%, -5%, 0%, +5%, +10%}.

[0053] Based on the control penalty factor and stability reward coefficient, the reward function R = -(△L1+λ△L2)+η·Stability_Bonus is set, where λ = 0.5 is the control penalty factor and η = 10 is the stability reward coefficient. It is activated when the temperature fluctuates ≤ 1°C for 5 consecutive minutes.

[0054] Finally, based on the Q-learning algorithm, the update strategy (-Q(s,a)+a[R+γmaxQ(s',a')-Q(s,a)], where the learning rate a=0.1 and the discount factor γ=0.9) is used to achieve online optimization of the model parameters.

[0055] Then establish an abnormal warning module, and add warning functions such as temperature rise and temperature speed on the basis of the bed temperature warning. The abnormal warning module of this embodiment performs temperature rise rate warning specifically including: when the temperature rise rate exceeds the first threshold, that is, when |dT / dt|>2℃ / min, triggering a first-level warning; when the temperature rise rate exceeds the second threshold, that is, when |dT / dt|>5℃ / min, triggering a second-level warning. The abnormal warning module of this embodiment performs temperature deviation warning specifically including: when the temperature deviation value exceeds the third threshold for t4 time, that is, |Tcurrent -T set When the temperature >3℃ lasts for more than 5 minutes, the adjustment command is triggered.

[0056] The abnormal warning module of this embodiment also includes a composite warning, which integrates the temperature rise rate, temperature deviation value and historical trend, and uses fuzzy logic to determine the risk level: Risk_Score=u1(△T)*u2(dT / dt); Among them, u1 and u2 are membership functions, assuming that the membership function is a triangular distribution.

[0057] The hydrocracking reactor temperature control system of this embodiment also includes an abnormality response mechanism. When an abnormality warning is triggered, the system automatically increases the amount of cold hydrogen based on the warning information. For example, in a preferred embodiment, when the abnormality warning module triggers a level one warning, the system automatically increases the amount of cold hydrogen by 5%. When the abnormality warning module triggers a level two warning, the system automatically increases the amount of cold hydrogen by 10%.

[0058] A hydrocracking reactor temperature control method of this embodiment also includes a residence time identification algorithm for aligning data, determining the predicted duration, and the timing of the control means action.

[0059] A hydrocracking reactor temperature control system of this embodiment implements a hydrocracking reactor temperature control method described above in this embodiment, and includes a field data layer, a bottom control layer, and an intelligent control layer. The field data layer obtains field data of the hydrocracking reactor, which includes temperature data, flow data, pressure data, and weight data. It also includes characteristic variable data that affect the bed temperature, such as the raw material temperature, feed rate, furnace feed temperature, circulating hydrogen amount, and cold hydrogen amount of the hydrocracking reactor. The bottom control layer can use a DCS control system to adjust and control the temperature of the hydrocracking reactor. The intelligent control layer performs feedforward control on the bottom control layer, including data application, intelligent control APP, data storage, and data acquisition.

[0060] The data acquisition section features a data access layer to control the DCS. The data storage section includes a cache database (Redis) and a relational database (mariadb). The data application section uses reactor temperature field visualization. The intelligent control app includes an intelligent predictive control module, an abnormality warning module, and a statistical analysis module. The intelligent predictive control module includes a prediction model, an intelligent control model, and an online optimization and learning model, enabling graphical display of reactor bed temperature and temperature rise prediction curves, as well as automated intelligent control functions. The abnormality warning module includes an abnormality warning model for monitoring reaction temperature and equipment, providing early warnings and alarms for abnormal situations. The statistical analysis module provides statistical functions for key variables in the production process and operational stability statistics.

[0061] In a preferred embodiment, the present invention was applied to a 2 million tons / year hydrocracking unit, and 30 days of operating data were collected. The results showed that the bed temperature fluctuation range was reduced from ±5°C in traditional DCS to ±1.2°C; the time required to adjust the temperature to within ±1.5°C of the set value was shortened from 15 minutes to 5 minutes; and the catalyst coking rate was reduced by 42%, verifying the effectiveness of the model.

[0062] The actual temperature fluctuation of the cracking reaction around the set temperature is significantly reduced, significantly increasing the yield of the target product. The addition of an intelligent control layer enables precise control of the bed temperature. The temperature fluctuation range at each measuring point at the bed outlet is ≤3°C within one hour. After the temperature deviates from the set point, it takes ≤8 minutes to stabilize within ±1.5°C of the set point.

[0063] A hydrocracking reactor temperature control method and system of this embodiment adds an intelligent control layer to the existing hydrocracking reaction DCS control system. By constructing a prediction model, an intelligent control model, an online optimization and learning model, and an abnormality warning model, the fluctuation range of the bed outlet temperature is reduced, the cracking reaction temperature is precisely controlled, and the by-product yield is reduced.

[0064] Based on the characteristic variables affecting the bed temperature, such as raw material temperature, feed rate, furnace feed temperature, circulating hydrogen amount, and cold hydrogen amount, a prediction model for the inlet temperature trend, temperature rise trend, and bed average temperature trend is established to predict the inlet temperature trend, temperature rise trend, and bed average temperature trend of each bed of the hydrocracking reactor in advance.

[0065] Through the feedback of the prediction model, an intelligent control model is established. The interdisciplinary fusion and multi-driven intelligent control model of "machine learning + control theory + industry experience" is adopted to smoothly reach the set value as the optimization goal. The appropriate control amount is used at the right time to ensure the stability of bed temperature and temperature rise. It can accurately characterize the large lag, nonlinear, and multi-objective control process, reducing damage to the catalyst.

[0066] At the same time, an abnormal warning module is established, which adds warning functions such as temperature rise and temperature speed on the basis of bed temperature warning.

[0067] By integrating machine learning and control theory, a fully closed-loop intelligent optimization of "prediction-control-early warning" is achieved, which effectively reduces the gap between the final operating temperature of the catalyst and the set value, and increases the service life of the catalyst. At the same time, by avoiding the occurrence of "temperature runaway" accidents, the generation of catalyst coking and bed hot spots is effectively avoided.

[0068] A hydrocracking reactor temperature control method and system of this embodiment can not only optimize the control of the hydrocracking reactor bed temperature, reduce the fluctuation amplitude of the bed temperature, and improve the actual control upper limit index of the reaction temperature, but also prevent overheating and temperature runaway, avoid catalyst coking and deactivation, and have significant economic benefits.

[0069] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any simple modification, change and equivalent structural transformation made to the above embodiment based on the technical essence of the present invention still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for controlling the temperature of a hydrocracking reactor, characterized in that: include: S1: Real-time collection of characteristic variable data affecting the bed temperature of the hydrocracking reactor and preprocessing; S2: Based on the characteristic variables, a multivariate time series prediction framework is used to construct a prediction model to predict the bed temperature trend; S3: constructing an intelligent control model, taking smooth reaching of the set value as the optimization goal, and optimizing the bed temperature based on the predicted value output by the prediction model; S4: Build an abnormal warning model to identify abnormal situations and issue abnormal warnings.

2. A hydrocracking reactor temperature control method according to claim 1, characterized in that: The characteristic variables include raw material temperature and feed amount, furnace feed temperature, circulating hydrogen amount and cold hydrogen amount; the data preprocessing includes: using Z-score normalization to eliminate dimension effects and standardize the data.

3. A hydrocracking reactor temperature control method according to claim 1 or 2, characterized in that: The input variables of the prediction model include the characteristic variables, and the output variables of the prediction model include the predicted value of the inlet temperature of the bed layer, the predicted value of the temperature rise rate, and the predicted value of the average temperature within the future time t1.

4. A hydrocracking reactor temperature control method according to claim 3, characterized in that: The prediction model is constructed based on LSTM, and the hidden layer state update formula and the final output layer formula are established.

5. The method for controlling the temperature of a hydrocracking reactor according to claim 1, wherein: The intelligent control model introduces a dynamic weight adjustment mechanism and adopts a model predictive control optimization algorithm to roll-optimize the control quantity sequence within the future time t3 every t2 time, and solves the optimal solution through the gradient descent algorithm.

6. A hydrocracking reactor temperature control method according to claim 1 or 5, characterized in that: It also includes constructing an online optimization and learning model, which adopts a reinforcement learning framework, establishes an evaluation mechanism to optimize the intelligent control model, and performs model parameter adaptation.

7. A hydrocracking reactor temperature control method according to claim 6, characterized in that: The model parameter adaptation specifically includes: Constructing a state space S, wherein the state space S includes historical data of bed temperature and temperature rise rate and cold hydrogen amount; Define the action space A as the control amount adjustment range; Set the reward function based on the control penalty factor and stability reward coefficient; Based on the Q-learning algorithm update strategy, the model parameters are optimized online.

8. The method for controlling the temperature of a hydrocracking reactor according to claim 1, wherein: The abnormal warning model includes a temperature rise rate warning, which triggers a first-level warning when the temperature rise rate exceeds a first threshold, and a second-level warning when the temperature rise rate exceeds a second threshold; The abnormal warning model also includes a temperature deviation warning, which triggers an adjustment instruction when the temperature deviation value exceeds a third threshold value for a period of time t4.

9. A hydrocracking reactor temperature control method according to claim 8, characterized in that: The abnormal warning model also includes a composite warning, which integrates the temperature rise rate, temperature deviation value and historical trend, and uses fuzzy logic to judge the risk level.

10. A hydrocracking reactor temperature control system, characterized in that: The method for controlling the temperature of a hydrocracking reactor according to any one of claims 1 to 9 comprises: A field data layer for acquiring field data of a hydrocracking reactor, wherein the field data includes the characteristic variable data; The bottom control layer regulates and controls the temperature of the hydrocracking reactor; The intelligent control layer performs feedforward control on the underlying control layer, and includes an intelligent prediction control module and an abnormal warning model. The intelligent prediction control module includes a prediction model, an intelligent control model, and an online optimization and learning model.

Images