An intelligent optimization control system for the whole process of AES production

Abstract

The application provides a system containing a digital twin real-time simulation layer, a process data acquisition layer (containing a raw material quality online analysis sub-module), an intelligent soft measurement layer, an advanced optimization control layer, a safety interlocking control layer, a falling film reactor health evaluation layer and a human-computer interaction layer. The digital twin layer establishes a one-dimensional distributed parameter mechanism model for the falling film sulfonation reactor and undertakes the control instruction pre-play verification function; the soft measurement layer online predicts key quality indexes based on the RBF neural network and has the recursive least square online correction function; the optimization control layer adopts the strategy of combining MPC with the Pareto multi-objective optimization and is provided with a decision preference dynamic switching unit to adapt to market changes; the safety interlocking layer adopts the double protection of the two-out-of-three voting and the MPC output limiting. The application improves the production safety from the raw material quality perception, the reactor internal state mirroring, the control pre-play verification, the multi-objective compromise optimization to the safety interlocking protection.

Description

Technical Field

[0001] This invention relates to the field of automatic control of chemical production processes, specifically to a full-process intelligent control system for the production process of fatty alcohol polyoxyethylene ether sulfate (AES), which integrates digital twin simulation and multi-objective optimization decision-making. Background Technology

[0002] Sodium fatty alcohol polyoxyethylene ether sulfate (AES) is an anionic surfactant widely used in detergents, personal care products, and textile auxiliaries due to its good solubility, strong resistance to hard water, rich foam, good biodegradability, and low skin irritation. AES production typically involves three steps: condensation of fatty alcohol polyoxyethylene ether (AEO) with ethylene oxide, followed by sulfonation with sulfur trioxide (SO3) to obtain a sulfate ester, neutralization with alkali to form a salt, and finally removal of the byproduct 1,4-dioxane through vacuum neutralization or stripping. Typical operating conditions for the sulfonation reaction are: SO3 inlet temperature approximately 45°C, circulating cooling water temperature 30-35°C, SO3 / AEO molar ratio 1.01-1.02, and SO3 concentration 2.5%-3.5%. The neutralization temperature is generally controlled at 45-50°C to prevent product decomposition and color darkening.

[0003] Most existing AES production lines use a DCS system combined with conventional PID loops for control. The limitations of this approach are: the control loops for each process are independent, lacking overall process-wide coordination; key quality indicators such as sulfonation rate and dioxane content rely on manual sampling and offline testing, making real-time online feedback impossible; the sulfonation reaction itself is highly exothermic, nonlinear, and exhibits significant time-varying characteristics, causing fixed-parameter PID controllers to often deviate from their optimal state when operating conditions change. Furthermore, the hydroxyl value and moisture content of AEO raw materials fluctuate between different batches, and existing control systems lack the ability to adaptively compensate for such disturbances. Summary of the Invention

[0004] This invention addresses the aforementioned shortcomings in existing AES production process control by proposing an intelligent control system that integrates raw material quality perception with multi-objective optimization across the entire process. Specifically, it aims to solve problems such as: fragmented control of various processes, lack of online feedback for quality indicators, inability to adaptively compensate for operational disturbances caused by raw material quality fluctuations, difficulty in achieving economic optimization based solely on experience-based setpoints, and a lack of monitoring methods for the internal state of the falling film sulfonation reactor.

[0005] This invention provides an intelligent optimization control system for the entire AES production process, the system comprising the following functional levels:

[0006] (1.0) Digital Twin Real-Time Simulation Layer: A digital twin based on a one-dimensional distributed parameter mechanism model is established for the falling film sulfonation reactor, including a reaction kinetics submodule, a liquid film heat and proton transfer submodule, and a virtual sensor submodule; the digital twin simulates the entire process response before the operation variable values ​​solved by MPC are sent to the actuator. If the simulation result shows that the key status indicators exceed the alarm threshold, the set of operation variable values ​​is rejected and MPC is re-solved, forming a closed loop of MPC solution → digital twin simulation → safety verification → execution output;

[0007] (2.0) Process data acquisition layer: Deployed at each process stage of AES production, including temperature sensors, pressure sensors, flow meters, pH meters and SO3 concentration analyzers, to collect key process parameters of sulfonation, neutralization and dioxane removal processes in real time, and transmit the collected data to the upper control module via industrial Ethernet or fieldbus; the process data acquisition layer also includes a set of AEO raw material quality online analysis submodules, which infer the hydroxyl value and moisture content of AEO raw materials online through near-infrared or Raman spectroscopy analyzers;

[0008] (3.0) Intelligent Soft Measurement Layer: A soft measurement model for AES key quality indicators is established based on the radial basis function (RBF) neural network, and the easily measurable variables provided by the process data acquisition layer are used as auxiliary variables to realize online continuous prediction of key quality indicators; the soft measurement model has an online recursive least squares correction mechanism based on offline test values;

[0009] (4.0) Advanced Optimization Control Layer: This layer includes a Model Predictive Control (MPC) module and a full-process coordinated optimization module. The MPC module establishes multivariate predictive control models for the sulfonation, neutralization, and dioxane removal processes. Each process control model uses the process variables and operational variables from the previous moment as input to predict the controlled variable responses at multiple future moments. The objective function is to minimize the prediction error and the weighted control increment, and the optimal operational variable value is obtained through rolling optimization. Specifically, the sulfonation process MPC uses the hydroxyl value and moisture content output from the online raw material quality analysis submodule as feedforward perturbation variables; the dioxane removal process MPC uses the dioxane residue output from the intelligent soft measurement layer as the pre-calculated value. The measured values ​​are used as feedforward perturbation variables. The full-process coordination optimization module, based on the output of the MPC module of each process and the quality prediction results of the soft measurement layer, adopts a Pareto optimal multi-objective evolutionary optimization algorithm to simultaneously optimize the three sub-objectives of output, energy consumption, and quality risk, generating a set of non-dominated Pareto optimal solutions. The full-process coordination optimization module includes a decision preference dynamic switching unit, which selects a compromise solution from the Pareto optimal solution set as the operation setpoint according to market environment parameters or operator-specified priority. The full-process coordination optimization module sets the sulfonation reaction temperature and SO3 / AEO molar ratio as hard constraints, where the molar ratio hard constraint is a dynamic value M_ratio ≤ 1.02 + δ(t), and δ(t) is corrected by feedback from the dioxane prediction value.

[0010] (5.0) Safety Interlock Control Module: Connected to the safety instrumented system of DCS, it sets multi-level alarm and interlock protection logic for sulfonation reaction temperature, reactor pressure and SO3 gas concentration; when any safety parameter exceeds the preset threshold, it automatically executes the interlock action after confirmation by the three-out-of-two voting logic; the safety interlock control module executes the sequential logic of "optimization → safety verification → amplitude limiting output" on the optimal operating variable value solved by MPC.

[0011] (6.0) Falling film reactor health status assessment and predictive maintenance module: Extract abnormal membrane distribution health indicators by utilizing the temperature difference characteristics and pressure fluctuation characteristics of multi-point temperature sensors arranged along the axial direction of the reaction tube, establish a prediction model for fouling and aging trends of the reaction tube based on historical operating data, predict the remaining effective operating time and trigger maintenance warnings;

[0012] (7.0) Human-machine interaction layer: including DCS operator station and advanced management interface, used to display the dynamics of the entire process flow, real-time process parameters, digital twin simulation results, soft measurement prediction values, controller operating status and alarm information, and provide an operator manual intervention interface.

[0013] Preferably, in the digital twin real-time simulation layer, the virtual sensor submodule includes at least inference outputs of local temperature peaks, local liquid film thickness minimums, and local sulfonation conversion rate distributions along the reaction tube axis.

[0014] Preferably, the controlled variables of the sulfonation process MPC module include the sulfonation reaction temperature, the sulfate ester content in the sulfonation product, and the color of the sulfonation product; the operating variables include the SO3 inlet flow rate, the AEO feed flow rate, and the sulfonator circulating cooling water flow rate; and the disturbance variables include the SO3 inlet temperature, the AEO feed temperature, and the AEO hydroxyl value and moisture content provided by the raw material quality online analysis submodule.

[0015] Preferably, the controlled variables of the neutralization process MPC module include neutralization temperature, neutralization product pH value, and product active ingredient content; the operating variables include alkali flow rate and neutralization circulating cooling water flow rate; the disturbance variables include sulfonation product feed temperature and flow rate; and the neutralization product pH value is controlled using a Hammerstein-Wiener type nonlinear prediction model.

[0016] Preferably, the controlled variables of the dioxane removal process MPC module include the top and middle temperatures of the vacuum stripping tower and the top vacuum degree; the operating variables include the steam injection flow rate and the opening degree of the vacuum pump regulating valve; and the disturbance variables include the feed temperature and the initial concentration of dioxane in the feed.

[0017] Preferably, the Pareto optimality-based multi-objective evolutionary optimization algorithm of the full-process coordinated optimization module adopts the NSGA-III or MOEA / D algorithm, and the three sub-objectives are:

[0018] Maximize output f1 = Product output × Product qualification rate; Minimize energy consumption f1 = Comprehensive energy consumption per unit product; Minimize quality risk f3 = Dioxane residue + Color index.

[0019] Preferably, the decision preference dynamic switching unit automatically reselects a compromise solution from the Pareto optimal solution set when the fluctuation of raw material purchase price, electricity unit price or product market price exceeds its respective preset threshold.

[0020] Preferably, in the three-out-of-two voting logic of the safety interlock control module, three independent temperature sensors are installed on the outlet manifold of the sulfonation reactor, and the interlock action is triggered when any two sensors simultaneously exceed the set temperature threshold.

[0021] Preferably, in the falling film reactor health status assessment module, the multi-point temperature sensor is arranged with at least three temperature measurement points along the reactor tube axis: inlet section, middle section, and outlet section; the abnormal membrane distribution health indicators include the temperature difference between adjacent temperature measurement points and the ratio of the dominant frequency amplitude of the pressure pulsation in the reactor feed main pipe.

[0022] Preferably, in the intelligent soft measurement layer, the auxiliary variables of the RBF neural network include sulfonation reaction temperature, SO3 / AEO molar ratio, SO3 flow rate, AEO flow rate, cooling water temperature, neutralization temperature, alkali flow rate, stripping tower temperature, and tower top vacuum; the recursive least squares correction mechanism updates the model parameters using periodically acquired offline test values, with the decay factor set to 0.95-0.99.

[0023] The beneficial effects of this invention are as follows:

[0024] First, the introduction of the digital twin simulation layer enables the real-time acquisition of unmeasurable states inside the falling film sulfonation reactor (local hot spots, liquid film thickness, conversion rate distribution along the tube length). At the same time, control commands are pre-verified before being issued, which significantly reduces the risk of misoperation.

[0025] Secondly, the online raw material quality analysis submodule, in conjunction with the MPC feedforward compensation channel, enables the system to adapt to quality fluctuations between AEO batches, reducing product quality fluctuations caused by changes in raw materials.

[0026] Third, by adopting Pareto multi-objective optimization instead of traditional fixed-weighted optimization, the conflict relationship between output, energy consumption, and quality risks can be revealed. Furthermore, by dynamically switching decision preferences, the system adapts to changes in the market environment, thereby improving the economy and flexibility of the entire process.

[0027] Fourth, the dioxane source suppression and constraint mechanism (dynamic molar ratio upper limit) directly couples the soft measurement prediction value with the constraint conditions, suppressing the generation of byproducts from the reaction source, forming "two lines of defense" with the back-end vacuum vaporization removal.

[0028] Fifth, the falling film reactor health status assessment module can provide early warning of scaling and aging trends, providing a basis for planned maintenance and reducing unplanned shutdowns. Detailed Implementation

[0029] To further understand the features, technical means, and specific objectives and functions achieved by the present invention, the present invention will be described in further detail below with reference to specific embodiments.

[0030] Example 1: This example provides a specific implementation of the digital twin simulation layer.

[0031] Taking a production line with an annual output of 50,000 tons of AES as an example, the falling film sulfonation reactor has a total of 36 reaction tubes, each with an inner diameter of 12 mm and a length of 6 m. The digital twin simulation layer is built upon the following physical framework:

[0032] (1) Reaction kinetic model: The sulfonation reaction of AEO with SO3 was simplified as a pseudo-first-order irreversible reaction controlled by intramembrane diffusion. The kinetic parameters of the main reaction (sulfonation) and the two main side reactions—dioxane formation and persulfonation—were calibrated by offline small-scale experiments. The reaction tube was discretized into 200 micro-segments along its length, and the material balance and energy balance equations were solved in each micro-segment.

[0033] (2) Liquid film heat and mass transfer model: The distribution of liquid film thickness along the pipe length is calculated using Nusselt's falling film theory, combined with empirical correlations of the mass transfer coefficients of SO3 in the gas phase and liquid film. The convective heat transfer coefficient on the cooling water side is corrected in real time based on the cooling water flow rate and temperature.

[0034] (3) Virtual sensor: The digital twin completes a full-length simulation calculation every 2 seconds and outputs the following internal state variables that cannot be directly measured by physical sensors: the highest temperature and its location in the temperature profile along the pipe length, the thickness value of the thinnest part of the liquid film (below 0.1 mm is considered a risk of film rupture), and the percentage of pipe length required for the sulfonation conversion rate to reach 99% along the pipe length.

[0035] (4) Control simulation: When the MPC solves for a new set of operating variable values ​​(such as increasing SO3 flow rate by 5% or decreasing cooling water flow rate by 10%), the set of values ​​is first sent to the digital twin to perform forward simulation, which lasts for 10 minutes. If the simulation results show that the local hot spot temperature will exceed 85°C or the thickness of the thinnest part of the liquid film will be less than 0.1 mm within the next 10 minutes, the system rejects the set of operating variable values, and the MPC resolves the problem and reduces the search step size near the original solution.

[0036] Example 2: This example provides a specific implementation of the intelligent soft measurement layer.

[0037] The input layer of the RBF neural network contains 9 nodes, corresponding to 9 auxiliary variables: sulfonation reaction temperature, SO3 / AEO molar ratio, SO3 flow rate, AEO flow rate, cooling water temperature, neutralization temperature, alkali flow rate, stripping tower temperature, and tower top vacuum. The number of hidden layer nodes was determined to be 15 through leave-one-out cross-validation, and the expansion constant σ of the radial basis function was set to 0.5.

[0038] The output layer contains 5 nodes, corresponding to sulfonation reaction conversion rate, free oil content, dioxane concentration, pH value, and product color. The model sampling period is 5 seconds.

[0039] The online correction mechanism is as follows: Samples are taken from the production line every 4 hours and sent to the laboratory for analysis (laboratory analysis results lag by approximately 1 hour). The deviation between the offline test values ​​and their corresponding soft measurement prediction values ​​is used as a correction signal. The connection weights of the output layer of the RBF neural network are updated using the recursive least squares method, with an attenuation factor set to 0.98. Testing showed that this correction mechanism can reduce the root mean square error of the soft measurement prediction values ​​by approximately 40% within 24 hours of implementation.

[0040] Example 3: This example provides an implementation method for multi-objective optimization in the full-process coordination and optimization module.

[0041] The optimization algorithm used is NSGA-III, with a population size of 100 and a maximum number of generations of evolution of 200. The three objective functions are:

[0042] F1 = -(AES output × Product qualification rate)

[0043] F2 = Unit product comprehensive energy consumption (kgce / t)

[0044] F3 = Dioxane residue in the product + Color normalization index

[0045] The constraints include: T_react ≤ 90℃, SO3 concentration ≤ 5%, M_ratio ≤ 1.02 + δ(t) (δ(t) is automatically adjusted based on the dioxane prediction value), and neutralization temperature ≤ 65℃.

[0046] The algorithm performs a full-process optimization every 2 hours (in conjunction with the hourly cycle of the upper-level global optimization), generating a Pareto front solution set under the current operating conditions. The decision preference dynamic switching unit checks market environment parameters every 15 minutes: when raw material price fluctuations exceed 5% or product price fluctuations exceed 3%, it automatically reselects the optimal compromise solution from the solution set. For example, during periods of rising raw material prices, it automatically selects a compromise solution biased towards "low raw material consumption per unit," appropriately reducing the output target to achieve lower unit costs.

[0047] Example 4: This example provides the implementation method of the online raw material quality analysis submodule and MPC feedforward compensation.

[0048] A Fourier transform near-infrared spectrometer with a spectral acquisition range of 4000-10000 cm⁻¹ was installed on the AEO raw material feed pipeline. -1 The resolution is 8 cm. -1 Each spectral scan lasted 30 seconds. The quantitative analysis model employed partial least squares regression to predict the hydroxyl value (reflecting the average molecular weight) and moisture content of AEO using spectral data. The root mean square error of the hydroxyl value prediction on the calibration set was 1.8 mgKOH / g, and the root mean square error of the moisture prediction was 0.05%.

[0049] The hydroxyl value and moisture content are updated every minute and used as feedforward perturbation variables in the prediction model of the MPC module for the sulfonation process. When the hydroxyl value deviates from the nominal value by more than ±5%, the MPC controller automatically adjusts the SO3 / AEO molar ratio setpoint upwards (because a higher hydroxyl value means a larger average molecular weight of AEO, requiring more SO3 to achieve the same degree of sulfonation). When the moisture content is too high, the MPC reduces the SO3 flow rate, as moisture consumes SO3 to generate sulfuric acid, affecting sulfonation efficiency and product quality.

[0050] Example 5: This example provides an implementation method for a safety interlock control module.

[0051] The safety interlock control module is divided into three layers:

[0052] First layer: Instrumentation layer. Three independent temperature sensors are installed on the outlet main pipe of the sulfonation reactor, with a distance of no less than 5 cm between each pair. If any two of the three measuring points exceed 70°C simultaneously, a high-temperature interlock (two out of three votes) is triggered, shutting off the SO3 feed quick-closing valve (full closure time ≤ 1 second) and simultaneously opening the emergency cooling water valve.

[0053] The second layer: MPC output verification layer. Before the SO3 flow rate setpoint and cooling water flow rate setpoint calculated by MPC are output to the actuator via DCS, they are first logically verified by the safety interlock module. For example, if the SO3 flow rate setpoint calculated by MPC is 550 kg / h, and the current safety upper limit maintained by the safety interlock module is 520 kg / h (calculated in real time based on reactor pressure and temperature), the output value will be automatically truncated to 520 kg / h, instead of being directly issued as 550 kg / h.

[0054] The third layer: Digital twin pre-visualization layer (see Example 1). This layer runs before the MPC output verification and is a forward-looking security verification.

[0055] In the three-layer logic described above, any layer that triggers a protection action will generate an alarm record with a timestamp on the human-machine interface and write it into the historical database for later tracing.

[0056] Example 6: This example provides an implementation method for a falling film reactor health status assessment module.

[0057] Along the outer wall of three typical measuring tubes out of the 36 reaction tubes, a surface thermocouple is attached at each of the inlet section (0.5 m from the top), the middle section (3.0 m), and the outlet section (5.5 m), for a total of nine temperature measuring points. Under normal operation, the axial temperature difference does not exceed 8°C. When the temperature difference between any two adjacent measuring points exceeds 15°C, it is determined to be an abnormal film distribution.

[0058] Meanwhile, a pressure transmitter (sampling frequency 50 Hz) was installed on the reactor feed manifold, and the frequency domain characteristics of the pressure signal were extracted using Fourier transform. When the amplitude of the dominant frequency of the pressure pulsation exceeded twice the normal reference value, it was determined that there might be local blockage or scaling.

[0059] Historical operating data indicates that detectable scaling begins to appear in the reaction tubes after approximately 3000 hours of operation. A support vector regression model predicts the remaining effective operating time based on cumulative operating time, feedstock quality parameters (higher iodine values ​​indicate a greater tendency to scale), and cooling water quality parameters. When the predicted remaining time is less than 500 hours, a maintenance suggestion window pops up on the human-machine interface, providing a suggested cleaning time window.

[0060] The embodiments described above merely illustrate the optimal implementation of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. An intelligent optimization control system for the entire AES production process, characterized in that, The system includes the following functional levels: (1.0) Digital Twin Real-Time Simulation Layer: A digital twin based on a one-dimensional distributed parameter mechanism model is established for the falling film sulfonation reactor, including a reaction kinetics submodule, a liquid film heat and proton transfer submodule, and a virtual sensor submodule; the digital twin simulates the entire process response before the operation variable values ​​solved by MPC are sent to the actuator. If the simulation result shows that the key status indicators exceed the alarm threshold, the set of operation variable values ​​is rejected and MPC is re-solved, forming a closed loop of MPC solution → digital twin simulation → safety verification → execution output; (2.0) Process data acquisition layer: Deployed at each process stage of AES production, including temperature sensors, pressure sensors, flow meters, pH meters and SO3 concentration analyzers, to collect key process parameters of sulfonation, neutralization and dioxane removal processes in real time, and transmit the collected data to the upper control module via industrial Ethernet or fieldbus; the process data acquisition layer also includes a set of AEO raw material quality online analysis submodules, which infer the hydroxyl value and moisture content of AEO raw materials online through near-infrared or Raman spectroscopy analyzers; (3.0) Intelligent Soft Measurement Layer: A soft measurement model for AES key quality indicators is established based on the radial basis function (RBF) neural network, and the easily measurable variables provided by the process data acquisition layer are used as auxiliary variables to realize online continuous prediction of key quality indicators; the soft measurement model has an online recursive least squares correction mechanism based on offline test values; (4.0) Advanced Optimization Control Layer: This layer includes a Model Predictive Control (MPC) module and a full-process coordinated optimization module. The MPC module establishes multivariate predictive control models for the sulfonation, neutralization, and dioxane removal processes, respectively. Each process control model uses the process variables and operational variables from the previous moment as inputs to predict the responses of the controlled variables at multiple future moments. The objective function is to minimize the prediction error and the weighted control increment, and the optimal operational variable value is obtained through rolling optimization. Specifically, the sulfonation process MPC uses the hydroxyl value and moisture content output by the online raw material quality analysis submodule as feedforward perturbation variables; the dioxane removal process MPC uses the dioxane residue prediction output by the intelligent soft measurement layer. The value is used as a feedforward perturbation variable; the whole process coordination optimization module, based on the output of the MPC module of each process and the quality prediction results of the soft measurement layer, adopts a Pareto optimal multi-objective evolutionary optimization algorithm to simultaneously optimize the three sub-objectives of output, energy consumption and quality risk, and generates a set of non-dominated Pareto optimal solutions; the whole process coordination optimization module includes a decision preference dynamic switching unit, which selects a compromise solution from the Pareto optimal solution set as the operation setting value according to market environment parameters or operator-specified priority; the whole process coordination optimization module sets the sulfonation reaction temperature and SO3 / AEO molar ratio as hard constraints, where the molar ratio hard constraint is a dynamic value M_ratio≤ 1.02 + δ(t), and δ(t) is corrected by the dioxane prediction value; (5.0) Safety Interlock Control Module: Connected to the safety instrumented system of DCS, it sets multi-level alarm and interlock protection logic for sulfonation reaction temperature, reactor pressure and SO3 gas concentration; when any safety parameter exceeds the preset threshold, it automatically executes the interlock action after confirmation by the three-out-of-two voting logic; the safety interlock control module executes the sequential logic of "optimization → safety verification → amplitude limiting output" on the optimal operating variable value solved by MPC. (6.0) Falling film reactor health status assessment and predictive maintenance module: Extract abnormal membrane distribution health indicators by utilizing the temperature difference characteristics and pressure fluctuation characteristics of multi-point temperature sensors arranged along the axial direction of the reaction tube, establish a prediction model for fouling and aging trends of the reaction tube based on historical operating data, predict the remaining effective operating time and trigger maintenance warnings; (7.0) Human-machine interaction layer: including DCS operator station and advanced management interface, used to display the dynamics of the entire process flow, real-time process parameters, digital twin simulation results, soft measurement prediction values, controller operating status and alarm information, and provide an operator manual intervention interface.

2. The intelligent optimization control system for the entire AES production process according to claim 1, characterized in that, In the digital twin real-time simulation layer, the virtual sensor submodule includes at least inference outputs of local temperature peaks, local liquid film thickness minimums, and local sulfonation conversion rate distributions along the reaction tube axis.

3. The intelligent optimization control system for the entire AES production process according to claim 1, characterized in that, The controlled variables of the sulfonation process MPC module include the sulfonation reaction temperature, the sulfate ester content in the sulfonation product, and the color of the sulfonation product; the operating variables include the SO3 inlet flow rate, the AEO feed flow rate, and the sulfonator circulating cooling water flow rate; the disturbance variables include the SO3 inlet temperature, the AEO feed temperature, and the AEO hydroxyl value and moisture content provided by the raw material quality online analysis submodule.

4. The intelligent optimization control system for the entire AES production process according to claim 1, characterized in that, The controlled variables of the neutralization process MPC module include neutralization temperature, neutralization product pH value, and product active ingredient content; the operating variables include alkali flow rate and neutralization circulating cooling water flow rate; the disturbance variables include sulfonation product feed temperature and flow rate; the neutralization product pH value is controlled using a Hammerstein-Wiener type nonlinear prediction model.

5. The intelligent optimization control system for the entire AES production process according to claim 1, characterized in that, The controlled variables of the MPC module for the dioxane removal process include the top and middle temperatures of the vacuum stripping tower and the vacuum degree at the top of the tower; the operating variables include the steam injection flow rate and the opening degree of the vacuum pump regulating valve; and the disturbance variables include the feed temperature and the initial concentration of dioxane in the feed.

6. The intelligent optimization control system for the entire AES production process according to claim 1, characterized in that, The Pareto optimality-based multi-objective evolutionary optimization algorithm of the full-process coordinated optimization module adopts the NSGA-III or MOEA / D algorithm, and the three sub-objectives are: Maximize output f1 = Product output × Product qualification rate; Minimize energy consumption f1 = Comprehensive energy consumption per unit product; Minimize quality risk f3 = Dioxane residue + Color index.

7. The intelligent optimization control system for the entire AES production process according to claim 1, characterized in that, The decision preference dynamic switching unit automatically selects a compromise solution from the Pareto optimal solution set when the fluctuation of raw material purchase price, electricity unit price or product market price exceeds its respective preset threshold.

8. The intelligent optimization control system for the entire AES production process according to claim 1, characterized in that, In the three-out-of-two voting logic of the safety interlock control module, three independent temperature sensors are installed on the outlet manifold of the sulfonation reactor. When any two sensors exceed the set temperature threshold at the same time, the interlock action is triggered.

9. The intelligent optimization control system for the entire AES production process according to claim 1, characterized in that, In the falling film reactor health status assessment module, the multi-point temperature sensor is arranged with at least three temperature measurement points along the axial direction of the reaction tube: inlet section, middle section, and outlet section; the abnormal membrane distribution health indicators include the temperature difference between adjacent temperature measurement points and the ratio of the main frequency amplitude of the pressure pulsation in the reactor feed main pipe.

10. The intelligent optimization control system for the entire AES production process according to claim 1, characterized in that, In the intelligent soft measurement layer, the auxiliary variables of the RBF neural network include sulfonation reaction temperature, SO3 / AEO molar ratio, SO3 flow rate, AEO flow rate, cooling water temperature, neutralization temperature, alkali flow rate, stripping tower temperature, and tower top vacuum. The recursive least squares correction mechanism updates the model parameters using periodically acquired offline test values, with the decay factor set to 0.95-0.99.