Feeding system and feeding method for intermediate synthesis

Abstract

The present invention relates to the field of feeding and discharging, and discloses a feeding system and a feeding method for intermediate synthesis, comprising: based on monitoring and analyzing the distribution of a catalyst bed during an alkylation process, establishing a deactivation model in combination with pressure fluctuations to dynamically compensate for catalyst activity, and adjusting the catalyst dispersion and feeding rate through the deactivation model; upon completion of alkylation, obtaining an oxidant concentration during an oxidation process, establishing an objective function related to reaction time based on the oxidant change rate, calculating a reaction time deviation through the objective function, and intelligently adjusting the oxidant feeding frequency in combination with the analysis; obtaining feeding parameters of multiple historical reaction cycles, establishing a three-layer intelligent control model to deeply coordinately optimize the reaction time and feeding parameters, constructing a time-scale-associated deactivation model and a time-scale-coupled free radical prediction model to solve the optimal adjustment amount; performing real-time adaptive control on the feeding parameters, and integrating predictive maintenance for feedback iteration.

Description

Technical Field

[0001] The present invention relates to the technical field of feeding and conveying materials, and in particular to a feeding system and a feeding method for intermediate synthesis. Background Art

[0002] Traditional feeding systems mostly rely on mechanical pumps and metering tanks. In traditional intermediate alkylation reactions, catalyst activity monitoring mainly relies on single-point thermocouple temperature measurement and timed sampling analysis.

[0003] In the existing technology, it is impossible to capture the difference in axial and radial temperature distribution, resulting in difficulty in timely detection of local overheating of the catalyst. The chemical deactivation problem is often not detected until the conversion rate of the reaction product decreases, which is a lag. There is a lack of a pressure-temperature multi-parameter coupling model, and the feed rate is only passively reduced according to the temperature exceeding the threshold value. The mechanical deactivation (bed blockage) indicated by pressure fluctuations is not considered, resulting in the inability to actively adjust the vibration frequency or backflush when the catalyst dispersion deteriorates. The oxidant concentration is detected offline after manual sampling, and the sampling cycle is long. It is impossible to capture the concentration change rate in real time, and the reaction time deviation is large, resulting in excessive addition of oxidant. The feed frequency is fixed, and no hysteresis compensation algorithm is introduced. The delay from the feed port to the reaction zone was ignored, which could easily trigger an "overoxidation" side reaction in the later stages of the reaction. A time-scale alignment mechanism for different reaction cycles was not established, resulting in large deviations in the time it takes for the oxidant to decay to the minimum deactivation standard between batches. The current deactivation trend could not be predicted based on historically similar operating conditions, and the catalyst life fluctuation range was unstable. The free radical kinetic equation was not linked to the feed parameters, resulting in a large deviation between the measured and theoretical free radical concentrations. Uneven initiator consumption extended the reaction induction period. The monitoring dimension was single, making it impossible to achieve multi-physical quantity fusion. The model was static and lacked dynamic adaptive capabilities. Feed control was crude, lacking coordination across time and space scales. Data-driven capabilities were weak, making full-cycle optimization impossible.

[0004] Therefore, it is necessary to provide a feeding system and feeding method for intermediate synthesis. Summary of the Invention

[0005] The object of the present invention is to provide a feeding system and feeding method for intermediate synthesis. In order to solve the above-mentioned problems in the prior art, the present invention is achieved through the following technical solutions:

[0006] In a first aspect, an embodiment of the present invention provides a feeding method for synthesizing an intermediate, comprising the following steps:

[0007] Step 1: Based on the monitoring and analysis of the distribution of the catalyst bed during the alkylation process, a deactivation model is established to dynamically compensate for the catalyst activity in combination with pressure fluctuations. The catalyst dispersion and feed rate are adjusted using the deactivation model.

[0008] Step 2: Based on the completion of alkylation, the oxidant concentration is obtained during the oxidation process, the oxidant change rate is calculated, and an objective function related to the reaction time is established. The reaction time deviation is calculated based on the objective function, and the oxidant feeding frequency is intelligently adjusted based on the analysis.

[0009] Step 3: Obtain the feeding parameters of multiple historical reaction cycles, establish a three-layer intelligent control model, deeply coordinate the reaction time and feeding parameters, and construct a time-scale correlation deactivation model and a time-scale coupled free radical prediction model to solve the optimal adjustment amount;

[0010] Step 4: Based on the optimal adjustment amount, the feeding parameters are adaptively controlled in real time, and predictive maintenance is integrated for feedback iteration.

[0011] In a second aspect, an embodiment of the present invention provides a feeding system for intermediate synthesis, which specifically includes the following modules:

[0012] Alkylation adjustment module: Based on the monitoring and analysis of the distribution of the catalyst bed during the alkylation process, a deactivation model is established in combination with pressure fluctuations to dynamically compensate for catalyst activity. The catalyst dispersion and feed rate are adjusted using the deactivation model.

[0013] Oxidation adjustment module: Based on the completion of alkylation, the oxidant concentration is obtained during the oxidation process, the oxidant change rate is calculated to establish an objective function with the reaction time, the reaction time deviation is calculated based on the objective function, and the oxidant feeding frequency is intelligently adjusted based on the analysis;

[0014] Adjustment and update module: Obtain the feeding parameters of multiple historical reaction cycles, establish a three-layer intelligent control model, deeply coordinate and optimize the reaction time and feeding parameters, and build a time-scale correlation deactivation model and a time-scale coupled free radical prediction model to solve the optimal adjustment amount;

[0015] Adaptive module: Based on the optimal adjustment amount, the feeding parameters are adaptively controlled in real time, and predictive maintenance is integrated for feedback iteration.

[0016] Beneficial effects of the present invention:

[0017] A three-dimensional temperature monitoring network for the catalyst bed is constructed by fusing infrared thermal imaging with thermocouples. A multi-parameter deactivation model is established by combining pressure fluctuations with BP neural networks. Catalyst activity is dynamically compensated and the feeding rate is adjusted. Ultraviolet spectroscopy and dynamic objective functions are used to achieve real-time tracking of the rate of change of oxidant concentration. Through a three-layer intelligent control model: a time-scale-correlated deactivation model, a free radical prediction model, and a multi-objective optimization solver, the reaction cycle feeding parameters are deeply collaboratively optimized, ultimately forming a monitoring-modeling-control-iteration closed loop, thereby reducing catalyst activity prediction errors, improving oxidant utilization, and extending catalyst life, providing a high-precision, low-energy intelligent feeding solution for intermediate synthesis. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] 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 only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 This is a flow chart of the steps of a feeding method for synthesizing an intermediate provided in Example 1 of the present invention;

[0020] Figure 2 This is a schematic structural diagram of a feeding system for intermediate synthesis provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present invention.

[0022] Example 1

[0023] like Figure 1 As shown, an embodiment of the present invention provides a feeding method for synthesizing an intermediate, which specifically includes the following steps:

[0024] Step 1: Based on the monitoring and analysis of the distribution of the catalyst bed during the alkylation process, a deactivation model is established to dynamically compensate for the catalyst activity in combination with pressure fluctuations. The catalyst dispersion and feed rate are adjusted using the deactivation model.

[0025] In a specific embodiment, during the alkylation reaction of the synthetic intermediate, the temperature distribution of the catalyst bed is monitored by an infrared thermal imager;

[0026] Specifically, quartz glass windows are opened at equal height intervals along the bed axis, and nitrogen purge rings are installed on the inner wall to prevent reactants from condensing and contaminating the windows.

[0027] An angled viewing window is set at the head, with the angle preset to 30°, to monitor the coking area at the top of the catalyst bed and the blockage of the distribution plate at the bottom;

[0028] Monitoring is done by infrared thermal imaging cameras that cover the normal operating temperature of the catalyst and have a safety margin greater than 20%;

[0029] Keep the infrared thermal imager 2 meters away from the quartz glass window to ensure that the field of view covers the entire cross-section of the bed. Calculate the installation angle using the triangulation method to coincide the imaging edge with the edge of the bed.

[0030] Specifically, using the center of the bed cross section as a reference, mark points A and B on the edge of the viewing window. Use an infrared thermal imager to photograph the viewing window. Adjust the equipment's pitch angle and horizontal deflection angle so that points A and B in the image coincide with the actual edge of the bed. Use trigonometric functions to calculate the equipment's pitch angle and horizontal deflection angle to ensure that the infrared thermal imager's optical axis is perpendicular to the center of the bed cross section.

[0031] For large reactors with a diameter greater than 1.5m, two infrared thermal imagers are symmetrically installed, fixed with a rigid bracket, and equipped with a three-dimensional adjustable pan / tilt head for fine-tuning of focus and angle. The angle is preset to 90°, and the panoramic thermal image is stitched together using software.

[0032] K-type thermocouples were embedded in the catalyst bed at depths preset to 20%, 50%, and 80% of the bed height. Thermocouple data and thermal imaging data were collected simultaneously to establish a temperature-pixel grayscale correction model.

[0033] The BP neural network model is used, with the input layer being the average pixel grayscale value and the output layer being the temperature value. The number of hidden layers and neurons is adjusted according to the data characteristics and model training results.

[0034] The neural network is trained using the training data set, and the gradient descent method is used to optimize the network weights to reduce the prediction error of the model;

[0035] Based on the analysis of the relationship between temperature and pixel grayscale value in the temperature-pixel grayscale correction model at different depths, it is analyzed whether there is a change in the temperature-pixel grayscale correction model parameters caused by depth differences;

[0036] If it exists, a temperature-pixel grayscale correction model of different depths is established; if it does not exist, a unified correction model is established;

[0037] By acquiring the temperature at different positions of the bed in real time, the temperature state of the catalyst during the reaction is obtained and the effect of temperature on the catalyst activity is analyzed;

[0038] It should be noted that temperature changes directly affect the activity and reaction rate of the catalyst. Different temperature zones lead to different catalyst activity performances. An accurate catalyst activity model is established based on monitoring and analyzing the temperature distribution.

[0039] The thermal image is divided into three axial regions, including the top preheating zone, the middle reaction zone, and the bottom cooling zone;

[0040] A temperature threshold is set in each axial region to identify abnormal temperature characteristics;

[0041] For example, the normal reaction temperature of the middle reaction zone is set to [120°C, 130°C], the warning value temperature is preset to 135°C, and the interlock value temperature is preset to 140°C;

[0042] It should be noted that the warning value temperature serves as an early warning signal for abnormal reaction temperature. When the temperature in the middle reaction zone reaches 135°C, the system immediately issues an alarm. If the reaction temperature approaches the warning value, it indicates that the reaction system is deviating from normal operation and there are potential risks. Potential risks include but are not limited to: changes in catalyst activity, fluctuations in reactant concentrations, and poor heat dissipation.

[0043] Within the normal operating temperature range of the catalyst, set catalytic temperature points, including the minimum operating temperature, rated operating temperature, and maximum operating temperature. The maximum operating temperature is the temperature after considering a 20% safety margin. At each catalytic temperature point, low-load, rated-load, and high-load operating conditions are tested to obtain temperature data and thermal imaging data under different operating conditions.

[0044] It should be noted that temperature data includes the temperature value of each temperature measurement point, the acquisition time, and the corresponding operating parameters. The operating parameters include the temperature setting value, load, and reaction time. Thermal imaging data includes the thermal imaging image file, acquisition time, the corresponding operating parameters, and the focal length, angle, and gain of the thermal imager. Considering the characteristics of the catalyst at different stages of use, typical operating conditions are selected for data acquisition at each stage to reflect the impact of catalyst aging on temperature distribution.

[0045] Under stable operating conditions, thermocouple data and thermal imaging data are collected synchronously at a frequency of once per minute to ensure data continuity and synchronization. During changing operating conditions, such as heating, cooling, and load adjustments, the frequency is increased to once per second to capture the dynamic process of temperature changes.

[0046] Through the BP neural network, the input layer is the average pixel grayscale value, the output layer is the temperature value, and the number of hidden layers and neurons is adjusted according to the data characteristics and model training results. The neural network is trained using the training data set, and the gradient descent method is used to optimize the network weights.

[0047] Specifically, the mean square error of the pixel grayscale average is calculated as the loss function, the initial learning rate is 0.001, and the learning rate is optimized by the Adam optimizer;

[0048] The established temperature-pixel grayscale correction model is integrated into the monitoring system to perform real-time correction of thermal imaging data. The system automatically extracts the grayscale value of each pixel from the bed thermal imaging image acquired by the thermal imager, calculates the corresponding temperature value based on the correction model, generates a temperature distribution cloud map, and displays the temperature of each area of ​​the bed in real time.

[0049] The pressure sensor collects the reaction system pressure data in real time, including the bed inlet / outlet pressure and pressure difference, and inputs them into the system synchronously;

[0050] Combine real-time pressure data with corrected temperature data as input to a catalyst activity calculation model;

[0051] In the initial model, pressure fluctuations were used to compensate for the limitations of the temperature model;

[0052] For example, the temperature in the middle reaction zone is normal and the sudden increase in pressure indicates a local blockage;

[0053] It should be noted that the foundation for multi-parameter monitoring has been established, but the dynamic compensation part of the deactivation model has not yet been completed. Pressure data can be provided for subsequent writing to quantify the mechanical deactivation of the catalyst, while temperature focuses on chemical deactivation, achieving high-precision three-dimensional temperature monitoring and data-driven modeling of the catalyst bed. The logical chain is: installing monitoring equipment, collecting multi-source data, building a correction model, analysis and early warning, and integrating pressure data;

[0054] The integrated learning method is used, and the input layer includes: corrected temperature data, pressure fluctuation data and supplementary working condition parameters; the corrected temperature data includes: regional average value and gradient change; the pressure fluctuation data includes: pressure difference and fluctuation frequency; the supplementary working condition parameters include: feeding speed history;

[0055] The output layer is the catalyst activity coefficient, which is a 0-1 scale, with 1 indicating full activity;

[0056] Proportional adjustment of catalyst dispersion and feed rate based on the difference between the catalyst activity coefficient and the standard activity coefficient;

[0057] The model is updated through an online learning mechanism to fine-tune the weights every 5 minutes based on new data to adapt to the catalyst aging dynamics;

[0058] Specifically, when the model detection activity coefficient drops to the preset activity threshold, dynamic compensation is triggered:

[0059] If the temperature of the middle reaction zone is >135°C, which leads to deactivation, adjust the feed rate first;

[0060] If the pressure fluctuation pressure difference is greater than 10%, it indicates deactivation and blockage. First, adjust the catalyst dispersion and start the bed vibrator to improve the distribution.

[0061] Specifically, the catalyst dispersion is improved by an adjustable distribution plate built into the bed. When the activity coefficient decreases and the temperature distribution is uneven, and the temperature difference in the top preheating zone is >15°C, the vibration frequency is increased to improve the catalyst uniformity;

[0062] Combined with the pressure data, if the pressure difference suddenly increases, reverse purge will be started to prevent blockage;

[0063] The initial setting is to perform real-time data correction through the reaction kinetics model: when the activity predicted decay rate is greater than 5% / min, the feed rate is gradually reduced by 0.5% / time, and vice versa;

[0064] Step 2: Based on the completion of alkylation, the oxidant concentration is obtained during the oxidation process, the oxidant change rate is calculated, and an objective function related to the reaction time is established. The reaction time deviation is calculated based on the objective function, and the oxidant feeding frequency is intelligently adjusted based on the analysis.

[0065] Three sets of UV-visible spectrometer sensors are symmetrically installed at the top, middle and bottom of the oxidation reactor.

[0066] Built-in flow cell sampling probe, set the sampling flow rate to control the sampling speed and avoid the influence of dead volume;

[0067] The oxidant solution with standard concentrations was used, with a concentration gradient of 0.1 mol / L, 0.5 mol / L, 1.0 mol / L, and 2.0 mol / L. Spectral calibration was performed to establish the Lambert-Beer law calibration model:

[0068] ;

[0069] The oxidant monitoring concentration C is calculated. is the absorbance, is the molar absorptivity, is the optical path length, is the offset correction amount;

[0070] The original absorbance is filtered using a 5-point moving average, and the formula Calculate the noise-reduced absorbance ,in, is the time index, is the original absorbance;

[0071] Based on the obtained de-noised absorbance, a model of oxidant concentration change rate was constructed through multi-stage filtering and feature extraction;

[0072] Specifically, based on the 5-point moving average filter, wavelet transform is introduced to reduce noise and separate high-frequency noise from effective signals;

[0073] A dynamic threshold elimination algorithm is used for transient bubble interference. If three consecutive data points deviate from the mean by ±3σ, they are marked as invalid and interpolated, where σ is the standard deviation of the deviation.

[0074] The oxidant concentration change rate is obtained by performing ratio processing on the change amount of the oxidant monitoring concentration per unit time and the change time. ;

[0075] The continuous curve is fitted by piecewise cubic spline interpolation to solve the derivative calculation error caused by discrete sampling;

[0076] Based on the obtained oxidant concentration change rate , through the formula:

[0077] ;

[0078] Constructing a dynamic objective function ,in, is the theoretical value of concentration, which is generated by the power law equation of the reaction kinetic model;

[0079] Based on the obtained dynamic objective function, the formula Calculate the reaction time deviation ;

[0080] Design a multi-modal feeding frequency adjustment mechanism based on reaction time deviation and real-time working conditions;

[0081] Specifically, low deviation area: if <0.1, maintain the current feeding frequency, and control the pre-adjusted parameters through model prediction to reduce subsequent fluctuations;

[0082] Medium deviation area: 0.1≤ <0.3, start the fuzzy logic controller, input parameters include reaction time deviation, oxidant concentration gradient, reactor temperature, and output feed frequency adjustment;

[0083] High deviation area: ≥0.3, triggering the interlocking mechanism, combining the reinforcement learning decision tree to select the optimal strategy, urgently adding oxidant or suspending the feed;

[0084] Develop a dual-channel feeding system: the main channel feeds according to the set feeding frequency f, and the auxiliary channel dynamically adjusts according to the reaction time deviation;

[0085] The introduction of a hysteresis compensation algorithm calculates the feed delay effect in advance based on the residence time distribution of the oxidant in the reactor to reduce the risk of overshoot;

[0086] Specifically, a two-parameter axial diffusion model was used to solve the RTD function, and the cumulative distribution F(t) = 90% corresponding time was taken;

[0087] Dynamically fine-tune the real-time receiving target frequency according to the reaction time deviation, and calculate the dynamic delay by multiplying the ratio of current flow to calibration flow and the original delay;

[0088] Retrieve the frequency instruction of the current moment minus the dynamic time delay moment from the historical database and output it to compensate the oxidant feeding frequency;

[0089] Step 3: Obtain the feeding parameters of multiple historical reaction cycles, establish a three-layer intelligent control model, deeply coordinate the reaction time and feeding parameters, and construct a time-scale correlation deactivation model and a time-scale coupled free radical prediction model to solve the optimal adjustment amount;

[0090] Specifically, the feeding parameters of multiple historical reaction cycles are obtained;

[0091] Feeding parameters include, but are not limited to: catalyst dispersion, feed rate, free radical concentration, initiator concentration, substrate concentration, and oxidant feed frequency;

[0092] Establish a three-layer intelligent control model:

[0093] The first layer is to build a time-scale correlation deactivation model:

[0094] The dynamic time warping algorithm was used to align the time axes of different periods and the reaction stages were divided into: initiation stage, oxidant concentration 0→80%; main reaction stage, concentration 80%→95%; decay stage, concentration 95%→end point;

[0095] The time scale scaling factor is obtained by taking the ratio of the average time for the historical oxidant activity to decay to 80% and the current time for the oxidant activity to decay to 80%. ;

[0096] The historical similar cycle data is weighted by the attention mechanism, and the predicted deactivation curve is output:

[0097] ;

[0098] in, is the original inactivation rate, is the rate of change of oxidant concentration;

[0099] The second layer is to build a time-scale coupled free radical prediction model:

[0100] Based on the oxidation reaction chain mechanism, through the free radical kinetic equation:

[0101] ;

[0102] Get the free radical concentration compensation term ,in, is the initiation rate constant of the initiator, is the initiator concentration, is the termination rate constant of free radicals, is the free radical concentration, is the chain growth rate constant of free radicals, is the substrate concentration;

[0103] Use the Transformer model to learn the mapping relationship between historical free radical concentration and feeding parameters, and output the compensation term;

[0104] The third layer is to build a multi-objective optimization solver:

[0105] Based on the multi-objective optimization solver, the formula is:

[0106] ;

[0107] The constraints are: predict the free radical concentration , , , ;

[0108] in, is the minimum reaction time, is the feeding frequency, is the catalyst dispersion vibration frequency, is the economic weight factor, is the catalyst loss;

[0109] The optimal adjustment amount is calculated by a multi-objective optimization solver;

[0110] Step 4: Based on the optimal adjustment amount, the feeding parameters are adaptively controlled in real time, and predictive maintenance is integrated for feedback iteration;

[0111] Input compensation parameters: oxidant feed frequency, catalyst dispersion vibration frequency and free radical concentration;

[0112] Through the governing equation:

[0113] ;

[0114] Get the adjusted compensation frequency ,in, is the proportional gain, is the inactivated derivative compensation, is the actual value of free radical concentration, is the predicted value of free radical concentration, is the reference value of free radical concentration;

[0115] Input feeding parameters into the knowledge base and write them into the operation log every cycle; generate digital twins through the knowledge base to generate simulation scenarios; use the optimizer to recommend feeding parameter adjustments and update the control feeding parameters;

[0116] Reinforcement learning is performed based on the acceleration of catalyst deactivation, Bayesian parameter optimization is performed based on insufficient oxidant utilization, and fault tree analysis and rule base expansion are performed based on mechanical vibration anomalies.

[0117] Example 2

[0118] like Figure 2 As shown, an embodiment of the present invention provides a feeding system for intermediate synthesis, which specifically includes the following modules:

[0119] Alkylation adjustment module: Based on the monitoring and analysis of the distribution of the catalyst bed during the alkylation process, a deactivation model is established in combination with pressure fluctuations to dynamically compensate for catalyst activity. The catalyst dispersion and feed rate are adjusted using the deactivation model.

[0120] Oxidation adjustment module: Based on the completion of alkylation, the oxidant concentration is obtained during the oxidation process, the oxidant change rate is calculated to establish an objective function with the reaction time, the reaction time deviation is calculated based on the objective function, and the oxidant feeding frequency is intelligently adjusted based on the analysis;

[0121] Adjustment and update module: Obtain the feeding parameters of multiple historical reaction cycles, establish a three-layer intelligent control model, deeply coordinate and optimize the reaction time and feeding parameters, and build a time-scale correlation deactivation model and a time-scale coupled free radical prediction model to solve the optimal adjustment amount;

[0122] Adaptive module: Based on the optimal adjustment amount, the feeding parameters are adaptively controlled in real time, and predictive maintenance is integrated for feedback iteration.

[0123] An embodiment of the present invention is described in detail above, but the content described is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention; the above formulas are all dimensionless and numerical calculations, and the formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field based on actual conditions and historical experience, and can be adjusted according to actual conditions; the above description is only a preferred embodiment of the present invention and is not used to limit the present invention. All equal changes and improvements made according to the scope of application of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A feeding method for intermediate synthesis, characterized in that: The following steps are involved: Based on the monitoring and analysis of the temperature distribution of the catalyst bed during the alkylation process, a deactivation model was established to dynamically compensate for catalyst activity in combination with pressure fluctuations. The catalyst dispersion and feed rate were adjusted using the deactivation model. Based on the completion of alkylation, the oxidant concentration is obtained during the oxidation process, the oxidant concentration change rate is calculated, and an objective function is established with the reaction time. The reaction time deviation is calculated based on the objective function, and the oxidant feeding frequency is intelligently adjusted based on the analysis. Obtain the feeding parameters of multiple historical reaction cycles, establish a three-layer intelligent control model, deeply coordinate and optimize the reaction time and feeding parameters, and build a time-scale correlation deactivation model and a time-scale coupled free radical prediction model to solve the optimal adjustment amount; The method for establishing the three-layer intelligent control model is: Establish a three-layer intelligent control model: The first layer builds a time-scale correlation deactivation model; The dynamic time warping algorithm was used to align the time axes of different periods and the reaction stages were divided into: initiation stage, oxidant concentration 0→80%; main reaction stage, concentration 80%→95%; decay stage, concentration 95%→end point; The time scale scaling factor is obtained by taking the ratio of the average time for the historical oxidant activity to decay to 80% and the current time for the oxidant activity to decay to 80%. ; The historical similar cycle data is weighted through the attention mechanism and the predicted deactivation curve is output; The second layer is to build a time-scale coupled free radical prediction model; Based on the oxidation reaction chain mechanism, the free radical concentration compensation term is obtained through the free radical kinetic equation; Use the Transformer model to learn the mapping relationship between historical free radical concentration and feeding parameters, and output the compensation term; The third layer is to build a multi-objective optimization solver; Based on the optimal adjustment amount, the feeding parameters are adaptively controlled in real time, and predictive maintenance is integrated for feedback iteration.

2. The feeding method for synthesizing an intermediate according to claim 1, characterized in that: The method for compensating the catalyst activity is: The deactivation model uses an online learning mechanism to fine-tune weights every 5 minutes using new data to dynamically adapt to catalyst aging; When the model detection activity coefficient drops to the preset activity threshold, dynamic compensation is triggered: If the temperature of the middle reaction zone is >135°C, which leads to deactivation, adjust the feed rate first; If the pressure fluctuation pressure difference is greater than 10%, it indicates deactivation and blockage. First, adjust the catalyst dispersion and start the bed vibrator to improve the distribution. The catalyst dispersion is adjusted by an adjustable distribution plate built into the bed. If the activity coefficient decreases and the temperature distribution is uneven, and the temperature difference in the top preheating zone is >15°C, the vibration frequency is increased; Combined with the pressure data, if the pressure difference suddenly increases, reverse purge is started; The initial setting is to perform real-time data correction through the reaction kinetics model: when the activity predicted decay rate is greater than 5% / min, the feed rate is gradually reduced by 0.5% / time, and vice versa.

3. The feeding method for synthesizing an intermediate according to claim 1, characterized in that: The method for adjusting the catalyst dispersion and feeding rate is: The pressure sensor collects the reaction system pressure data in real time, including the bed inlet / outlet pressure and pressure difference, and inputs them into the system synchronously; Combine real-time pressure data with corrected temperature data as input to a catalyst activity calculation model; An integrated learning method is used, and the input layer is: corrected temperature data, pressure fluctuation data and supplementary working condition parameters, the supplementary working condition parameters include: feeding speed history records; The output layer is the catalyst activity coefficient, which is a 0-1 scale, with 1 indicating full activity; The catalyst dispersion and feed rate are proportionally adjusted based on the difference between the catalyst activity coefficient and the standard activity coefficient.

4. The feeding method for synthesizing an intermediate according to claim 1, characterized in that: The method for obtaining the objective function is: Based on 5-point moving average filtering, wavelet transform is introduced to reduce noise and separate high-frequency noise from effective signals; A dynamic threshold elimination algorithm is used for transient bubble interference. If three consecutive data points deviate from the mean by ±3σ, they are marked as invalid and interpolated, where σ is the standard deviation of the deviation. The oxidant concentration change rate is obtained by performing ratio processing on the change amount of the oxidant monitoring concentration per unit time and the change time. ; The continuous curve is fitted by piecewise cubic spline interpolation to solve the derivative calculation error caused by discrete sampling; Based on the obtained oxidant concentration change rate , through the formula: ； Constructing a dynamic objective function ,in, is the theoretical value of concentration, which is generated by the power-law equation of the reaction kinetics model.

5. The feeding method for synthesizing an intermediate according to claim 1, characterized in that: The method for calculating the reaction time deviation is: Based on the obtained dynamic objective function, the formula Calculate the reaction time deviation .

6. The feeding method for synthesizing an intermediate according to claim 1, characterized in that: The method for adjusting the oxidant feeding frequency is: Design a multi-modal feeding frequency adjustment mechanism based on reaction time deviation and real-time working conditions; Low deviation area: If <0.1, maintain the current feeding frequency and control the pre-adjusted parameters through model prediction; Medium deviation area: 0.1≤ <0.3, start the fuzzy logic controller, input parameters include reaction time deviation, oxidant concentration gradient, reactor temperature, and output feed frequency adjustment; High deviation area: ≥0.3, triggering the interlocking mechanism, combining the reinforcement learning decision tree to select the optimal strategy, urgently adding oxidant or suspending the feed; Develop a dual-channel feeding system: the main channel feeds according to the set feeding frequency f, and the auxiliary channel dynamically adjusts according to the reaction time deviation; A hysteresis compensation algorithm is introduced to calculate the feed delay effect in advance based on the residence time distribution of the oxidant in the reactor; The two-parameter axial diffusion model was used to solve the RTD function, and the cumulative distribution F(t) = 90% corresponding time was taken; Dynamically fine-tune the real-time receiving target frequency according to the reaction time deviation, and calculate the dynamic delay by multiplying the ratio of current flow to calibration flow and the original delay; The frequency instruction of the current moment minus the dynamic delay moment is retrieved from the historical database and output to compensate the oxidant feeding frequency.

7. The feeding method for synthesizing an intermediate according to claim 1, characterized in that: The method for solving the optimal adjustment amount is: Based on the multi-objective optimization solver, the formula is: ； The constraints are: predict the free radical concentration , , , ; in, is the minimum reaction time, is the feeding frequency, is the catalyst dispersion vibration frequency, is the economic weight factor, is the catalyst loss; The optimal adjustment amount is calculated by a multi-objective optimization solver.

8. The feeding method for synthesizing an intermediate according to claim 1, characterized in that: The method for performing feedback iteration of the integrated predictive maintenance is: Input compensation parameters: oxidant feed frequency, catalyst dispersion vibration frequency and free radical concentration; Through the governing equation: ； Get the adjusted compensation frequency ,in, is the proportional gain, is the inactivated derivative compensation, is the actual value of free radical concentration, is the predicted value of free radical concentration, is the reference value of free radical concentration; Input feeding parameters into the knowledge base and write them into the operation log every cycle; generate digital twins through the knowledge base to generate simulation scenarios; use the optimizer to recommend feeding parameter adjustments and update the control feeding parameters; Reinforcement learning is performed based on the acceleration of catalyst deactivation, Bayesian parameter optimization is performed based on insufficient oxidant utilization, and fault tree analysis and rule base expansion are performed based on mechanical vibration anomalies.

9. A feeding system for intermediate synthesis, the feeding system being used to implement the feeding method according to any one of claims 1 to 8, characterized in that: include: Alkylation Adjustment Module: Based on the monitoring and analysis of the temperature distribution of the catalyst bed during the alkylation process, a deactivation model is established in combination with pressure fluctuations to dynamically compensate for catalyst activity. The catalyst dispersion and feed rate are adjusted using the deactivation model. Oxidation adjustment module: Based on the completion of alkylation, the oxidant concentration is obtained during the oxidation process, the oxidant concentration change rate is calculated, and an objective function is established with the reaction time. The reaction time deviation is calculated based on the objective function, and the oxidant feeding frequency is intelligently adjusted based on the analysis. Adjustment and update module: Obtain the feeding parameters of multiple historical reaction cycles, establish a three-layer intelligent control model, deeply coordinate and optimize the reaction time and feeding parameters, and build a time-scale correlation deactivation model and a time-scale coupled free radical prediction model to solve the optimal adjustment amount; The method for establishing the three-layer intelligent control model is: Establish a three-layer intelligent control model: The first layer builds a time-scale correlation deactivation model; The dynamic time warping algorithm was used to align the time axes of different periods and the reaction stages were divided into: initiation stage, oxidant concentration 0→80%; main reaction stage, concentration 80%→95%; decay stage, concentration 95%→end point; The time scale scaling factor is obtained by taking the ratio of the average time for the historical oxidant activity to decay to 80% and the current time for the oxidant activity to decay to 80%. ; The historical similar cycle data is weighted through the attention mechanism and the predicted deactivation curve is output; The second layer is to build a time-scale coupled free radical prediction model; Based on the oxidation reaction chain mechanism, the free radical concentration compensation term is obtained through the free radical kinetic equation; Use the Transformer model to learn the mapping relationship between historical free radical concentration and feeding parameters, and output the compensation term; The third layer is to build a multi-objective optimization solver; Adaptive module: Based on the optimal adjustment amount, the feeding parameters are adaptively controlled in real time, and predictive maintenance is integrated for feedback iteration.

Images