An intelligent dynamic split control method for adsorption enhanced water gas shift system

Abstract

The application discloses an intelligent dynamic shunt control method for an adsorption enhanced water-gas shift system, and relates to the technical field of control of the adsorption enhanced water-gas shift system. The method comprises the following steps: S1, parameter acquisition and preprocessing; S2, construction and training of a prediction model; S3, real-time prediction and demand analysis; S4, intelligent shunt ratio decision; and S5, dynamic adjustment and model updating. The method adopts a multi-model fusion and an intelligent optimization algorithm to solve an optimal shunt ratio, combines time series prediction, performance degradation evaluation and system output response simulation, and has higher control precision compared with existing static threshold adjustment or single PID control.

Description

Technical Field

[0001] This invention relates to the field of control technology for adsorption-enhanced water-gas shift systems, and more specifically to an intelligent dynamic diversion control method for adsorption-enhanced water-gas shift systems. Background Technology

[0002] Adsorption-enhanced water-gas shift technology integrates CO2 adsorption within the reactor, enabling simultaneous CO conversion and CO2 separation in a single device. This significantly improves reaction efficiency and reduces subsequent separation energy consumption, making it widely applicable in industrial scenarios such as syngas purification and hydrogen production. The split ratio, a core control parameter of the adsorption-enhanced water-gas shift H2 / CO regulation system, represents the proportion of inlet gas distributed between the reactor and the bypass. Its value directly determines multiple product distribution indicators, including H2 / CO and CO2 content.

[0003] Existing split control methods for adsorption-enhanced gas-water shift systems (SEWGS) mostly employ static regulation based on preset thresholds or traditional PID feedback control. Static regulation sets a split ratio threshold based on a fixed range of process parameters, making step-by-step adjustments when parameters deviate from the threshold. This method cannot adapt to dynamic fluctuations in inlet flow rate and composition (such as CO and H2O content), easily leading to system instability and decreased reaction efficiency. While traditional PID control can achieve a certain degree of dynamic regulation, it relies on precise mathematical models. SEWGS is a complex nonlinear system involving multiple physicochemical processes such as chemical reactions, adsorption / desorption, and mass and heat transfer, making it difficult to establish accurate mechanistic models. This results in significant lag and insufficient control precision in PID control when facing large-scale inlet flow fluctuations or sudden changes in downstream process demands.

[0004] Furthermore, existing technologies do not consider the impact of reactor performance degradation on split control. In adsorption-enhanced gas-water shift systems (SEWGS), adsorbents experience performance degradation such as capacity decay and activity decline during long-term cyclic adsorption / desorption, leading to changes in reaction and adsorption efficiency at the same split ratio. If a fixed control strategy is still used, it will further exacerbate system performance deterioration, reducing operational reliability and economy. Simultaneously, existing methods can only passively adjust based on real-time monitoring data, failing to predict trends such as inlet gas fluctuations and reactor performance changes, making proactive control difficult.

[0005] Therefore, given the shortcomings of existing technologies, such as their inability to accurately adapt to dynamic fluctuations in the intake air, difficulty in dealing with reactor performance degradation, and lack of forward-looking control capabilities, there is an urgent need to develop a dynamic diversion control method with real-time prediction and intelligent decision-making capabilities to improve the operational stability, control accuracy, and economy of the adsorption-enhanced gas-water conversion system SEWGS.

[0006] Therefore, proposing an intelligent dynamic diversion control method for adsorption-enhanced water-gas shift systems to address the difficulties in existing technologies is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] In view of this, the present invention provides an intelligent dynamic split control method for an adsorption-enhanced water-gas shift system. By integrating machine learning and real-time prediction models, it achieves precise and forward-looking adjustment of the split ratio, improves the system's adaptability to fluctuations in intake gas and changes in downstream process requirements, and takes into account the impact of reactor performance degradation, ensuring long-term stable and efficient operation of the system.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A smart dynamic diversion control method for an adsorption-enhanced water-gas shift system, the adsorption-enhanced water-gas shift system comprising: a feed gas inlet pipe, a bypass pipe, a reactor, a diversion valve, a merging valve, a sensor group, and a controller, wherein the diversion valve is located at the branch point of the feed gas inlet pipe and the bypass pipe, and the merging valve is located at the confluence of the reactor outlet and the bypass pipe outlet; the control method includes the following steps: S1. Parameter Acquisition and Preprocessing: The operating parameters of the adsorption-enhanced water-gas shift system are acquired in real time through a sensor array, and historical operating data is also obtained. The acquired operating parameters and historical operating data are preprocessed to obtain a standardized dataset. The operating parameters include inlet parameters, reactor operating parameters, adsorbent state parameters, and downstream demand parameters. S2. Construct and train prediction models: Based on the preprocessed historical operating data, construct an intake fluctuation prediction model, a reactor performance degradation prediction model, and a system output prediction model. Use machine learning algorithms to train and optimize each prediction model to obtain the trained prediction model. S3. Real-time Prediction and Demand Analysis: Input the pre-processed operating parameters into the trained intake fluctuation prediction model to obtain the predicted values ​​of the intake parameters within a preset time period; input the pre-processed operating parameters and historical operating data into the reactor performance degradation prediction model to obtain the predicted values ​​of the reactor performance degradation degree within a preset time period; analyze the downstream process demand parameters to determine the system output target value. S4. Intelligent split ratio decision: Based on the predicted values ​​of inlet parameters, the predicted value of reactor performance degradation, and the target value of system output, the optimal split ratio is solved through the objective function constructed by the controller; S5. Dynamic Adjustment and Model Update: The controller sends adjustment commands to the diversion valve based on the optimal diversion ratio obtained by the solution, so as to realize the dynamic adjustment of the diversion ratio; at the same time, the operating parameters, adjustment results and actual system output parameters are fed back to the prediction model to update and optimize the prediction model online.

[0009] Optionally, the intake parameters in S1 include: feed gas flow rate, temperature, pressure, CO volume fraction, H2O volume fraction, and H2 volume fraction; The reactor operating parameters include: reactor bed temperature, bed pressure, outlet CO volume fraction, outlet CO2 volume fraction, outlet H2 / CO volume fraction ratio, and outlet H2 volume fraction; Adsorbent state parameters include: cumulative adsorbent operating time, cumulative adsorption / desorption cycle count, and outlet CO2 volume fraction; Downstream demand parameters include: downstream product gas flow rate demand, CO conversion rate demand, CO2 capture rate demand, downstream demand for H2 / CO volume fraction ratio, and CO2 volume fraction demand.

[0010] Optionally, preprocessing in S1 includes: data cleaning, data standardization, and data dimensionality reduction; Data cleaning uses the 3σ criterion or interpolation to replace outliers; Data standardization can be achieved using either min-max standardization or z-score standardization. Data dimensionality reduction is achieved using principal component analysis or linear discriminant analysis.

[0011] Optionally, the intake fluctuation prediction model in S2 is constructed using a long short-term memory network, a gated recurrent unit, or a temporal convolutional network; the input of the intake fluctuation prediction model is the time series of historical intake parameters, and the output is the predicted value of the intake parameters within the next 5-30 minutes.

[0012] Optionally, the reactor performance degradation prediction model in S2 is constructed using random forest, gradient boosting tree, or convolutional neural network. The input to the reactor performance degradation prediction model is the historical reactor operating parameters, adsorbent state parameters, and corresponding reactor performance indicators. The output is the degree of reactor performance degradation in the next 1-24 hours, and the degree of reactor performance degradation is expressed as the decay rate of the performance indicators.

[0013] Optionally, the reactor performance degradation prediction model also includes a performance degradation threshold judgment module, which sends an early warning signal to the controller when the predicted performance index decay rate exceeds a preset threshold.

[0014] Optionally, the system output prediction model in S2 is constructed using support vector regression, neural network regression, or ensemble learning regression. The inputs to the system output prediction model are the split ratio, inlet parameters, and reactor performance status parameters, and the outputs are the corresponding system output parameters, including CO conversion rate, CO2 capture rate, outlet gas flow rate, and outlet gas H2 / CO volume fraction ratio.

[0015] Optionally, the objective function in S4 is constrained by CO conversion rate ≥ preset lower limit, CO2 capture rate ≥ preset lower limit, and downstream product gas flow rate ≥ demand value, with the goal of minimizing system energy consumption. The objective function is solved using a genetic algorithm, particle swarm optimization algorithm, or simulated annealing algorithm to obtain the optimal split ratio. The optimal split ratio ranges from 0.3 to 0.9 and needs to be combined with the initial H2 / CO volume fraction ratio of the intake gas, the downstream target H2 / CO volume fraction ratio, and the rated CO conversion rate to comprehensively determine the intake and reaction conditions, ensuring that the split ratio is adapted to real-time operating conditions and downstream demand.

[0016] Optionally, the online update of the prediction model in S5 adopts an incremental learning algorithm. Every preset time interval or after accumulating a certain amount of data, the parameters of each prediction model are fine-tuned using newly collected real-time data. A model verification module is set up so that when the prediction error of the updated model exceeds a preset error threshold, the entire dataset is retrained.

[0017] Optionally, an emergency adjustment mechanism is also included: when the sensor detects that the fluctuation of the intake parameters exceeds the preset safety threshold or the reactor is in abnormal operating condition, the controller pauses the machine learning predictive control mode, switches to the preset emergency diversion ratio, and switches back to the intelligent predictive control mode after the system condition stabilizes.

[0018] As can be seen from the above technical solution, compared with the prior art, the present invention discloses an intelligent dynamic diversion control method for an adsorption-enhanced water-gas shift system, the beneficial effects of which are: 1) This invention introduces a machine learning-driven real-time prediction model, which enables forward prediction of intake fluctuations. Compared with the "lag adjustment" of existing traditional feedback control, it can predict the trend of intake parameter changes 5-30 minutes in advance, and adjust the split ratio in advance accordingly, which greatly improves the system's adaptability to intake dynamic fluctuations and avoids system performance fluctuations caused by adjustment lag. 2) This invention is the first to integrate reactor performance degradation prediction into the split control strategy. By constructing a performance degradation prediction model, it can predict the impact of adsorbent decay and reactor performance decline on system output in real time, and make targeted compensation in the split ratio decision. This solves the problem that the existing technology ignores performance degradation, which leads to a gradual decrease in control accuracy, and ensures the stability and reliability of the system in long-term operation. 3) This invention employs multi-model fusion and intelligent optimization algorithms to solve for the optimal diversion ratio, combining time-series prediction, performance degradation assessment, and system output response simulation. Compared with existing static threshold adjustment or single PID control, it achieves higher control accuracy, reducing the fluctuation range of CO conversion rate to within ±2% and the fluctuation range of CO2 capture rate to within ±3%, while reducing system energy consumption by 5%-15%. 4) This invention sets up an online model update mechanism and an emergency adjustment mechanism, which takes into account both the adaptability and safety of the control: online updates can avoid model drift and ensure long-term control effect; emergency adjustments can cope with extreme working conditions and improve the robustness of system operation. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 The flowchart of an intelligent dynamic diversion control method for an adsorption-enhanced water-gas shift system provided by the present invention is shown. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] See Figure 1 As shown, this invention discloses an intelligent dynamic diversion control method for an adsorption-enhanced water-gas shift system. The adsorption-enhanced water-gas shift system includes: a feed gas inlet pipe, a bypass pipe, a reactor, a diversion valve, a merging valve, a sensor group, and a controller. The diversion valve is located at the branch point of the feed gas inlet pipe and the bypass pipe, and the merging valve is located at the confluence of the reactor outlet and the bypass pipe outlet. The control method includes the following steps: S1. Parameter Acquisition and Preprocessing: The operating parameters of the adsorption-enhanced water-gas shift system are acquired in real time through a sensor array, and historical operating data is also obtained. The acquired operating parameters and historical operating data are preprocessed to obtain a standardized dataset. The operating parameters include inlet parameters, reactor operating parameters, adsorbent state parameters, and downstream demand parameters. S2. Construct and train prediction models: Based on the preprocessed historical operating data, construct an intake fluctuation prediction model, a reactor performance degradation prediction model, and a system output prediction model. Use machine learning algorithms to train and optimize each prediction model to obtain the trained prediction model. S3. Real-time Prediction and Demand Analysis: Input the pre-processed operating parameters into the trained intake fluctuation prediction model to obtain the predicted values ​​of the intake parameters within a preset time period; input the pre-processed operating parameters and historical operating data into the reactor performance degradation prediction model to obtain the predicted values ​​of the reactor performance degradation degree within a preset time period; analyze the downstream process demand parameters to determine the system output target value. S4. Intelligent split ratio decision: Based on the predicted values ​​of inlet parameters, the predicted value of reactor performance degradation, and the target value of system output, the optimal split ratio is solved through the objective function constructed by the controller; S5. Dynamic Adjustment and Model Update: The controller sends adjustment commands to the diversion valve based on the optimal diversion ratio obtained by the solution, so as to realize the dynamic adjustment of the diversion ratio; at the same time, the operating parameters, adjustment results and actual system output parameters are fed back to the prediction model to update and optimize the prediction model online.

[0023] Furthermore, the intake parameters in S1 include: raw material gas flow rate, temperature, pressure, CO volume fraction, H2O volume fraction, and H2 volume fraction; The reactor operating parameters include: reactor bed temperature, bed pressure, outlet CO volume fraction, outlet CO2 volume fraction, outlet H2 / CO volume fraction ratio, and outlet H2 volume fraction; Adsorbent state parameters include: cumulative adsorbent operating time, cumulative adsorption / desorption cycle count, and outlet CO2 volume fraction; Downstream demand parameters include: downstream product gas flow rate demand, CO conversion rate demand, CO2 capture rate demand, downstream demand for H2 / CO volume fraction ratio, and CO2 volume fraction demand.

[0024] Furthermore, preprocessing in S1 includes: data cleaning, data standardization, and data dimensionality reduction; Data cleaning uses the 3σ criterion or interpolation to replace outliers; Data standardization can be achieved using either min-max standardization or z-score standardization. Data dimensionality reduction is achieved using principal component analysis or linear discriminant analysis.

[0025] Furthermore, the intake fluctuation prediction model in S2 is constructed using a long short-term memory network, a gated recurrent unit, or a temporal convolutional network; the input of the intake fluctuation prediction model is the time series sequence of historical intake parameters, and the output is the predicted value of the intake parameters within the next 5-30 minutes.

[0026] Furthermore, the reactor performance degradation prediction model in S2 is constructed using random forest, gradient boosting tree, or convolutional neural network. The input of the reactor performance degradation prediction model is the historical reactor operating parameters, adsorbent state parameters, and corresponding reactor performance indicators. The output is the degree of reactor performance degradation in the next 1-24 hours, and the degree of reactor performance degradation is expressed as the decay rate of performance indicators.

[0027] Furthermore, the reactor performance degradation prediction model also includes a performance degradation threshold judgment module, which sends an early warning signal to the controller when the predicted performance degradation rate exceeds a preset threshold.

[0028] Furthermore, the system output prediction model in S2 is constructed using support vector regression, neural network regression, or ensemble learning regression. The inputs to the system output prediction model are the split ratio, inlet parameters, and reactor performance status parameters, and the outputs are the corresponding system output parameters, including CO conversion rate, CO2 capture rate, outlet gas flow rate, and outlet gas H2 / CO volume fraction ratio.

[0029] Furthermore, in S4, the objective function is constrained by CO conversion rate ≥ preset lower limit, CO2 capture rate ≥ preset lower limit, and downstream product gas flow rate ≥ demand value, with the goal of minimizing system energy consumption. The objective function is solved using a genetic algorithm, particle swarm optimization algorithm, or simulated annealing algorithm to obtain the optimal split ratio. The optimal split ratio ranges from 0.3 to 0.9, and it needs to be combined with the initial H2 / CO volume fraction ratio of the intake gas, the downstream target H2 / CO volume fraction ratio, and the rated CO conversion rate to comprehensively determine the intake and reaction conditions, ensuring that the split ratio is adapted to real-time operating conditions and downstream demand.

[0030] Furthermore, the online update of the prediction model in S5 adopts an incremental learning algorithm. Every preset time interval or after accumulating a certain amount of data, the parameters of each prediction model are fine-tuned using newly collected real-time data. A model verification module is set up so that when the prediction error of the updated model exceeds a preset error threshold, the entire dataset is retrained.

[0031] Furthermore, it also includes an emergency adjustment mechanism: when the sensor detects that the fluctuation of the intake parameters exceeds the preset safety threshold or the reactor is in abnormal operating condition, the controller pauses the machine learning predictive control mode, switches to the preset emergency diversion ratio, and switches back to the intelligent predictive control mode after the system condition stabilizes.

[0032] In a specific embodiment: Parameter Acquisition and Preprocessing: The following parameters are acquired in real time using a sensor array: feed gas flow rate (0-500 Nm³ / h), temperature (200-400℃), pressure (1.0-3.0 MPa), CO volume fraction (10%-30%), H₂ volume fraction (30%-60%), H₂O volume fraction (20%-50%), and inlet CO₂ volume fraction (1%-5%); reactor bed temperature (250-450℃), bed pressure (1.0-3.0 MPa), outlet CO volume fraction (0.5%-5%), outlet CO₂ volume fraction (0%-35%), outlet H₂ volume fraction (40%-70%), outlet H₂ / CO volume fraction ratio (1:1-3:1), and outlet CO volume fraction (0.5%-5%); cumulative adsorbent operating time (0-8000 hours). The parameters include: h), cumulative adsorption / desorption cycle count (0-1000 times), outlet CO2 volume fraction (simultaneously used as an indicator of adsorbent saturation); downstream hydrogen-rich gas flow rate requirement (300-450 Nm³ / h), CO conversion rate requirement (≥95%), CO2 capture rate requirement (≥90%), and downstream target H2 / CO volume fraction ratio (4:1-8:1). Simultaneously, historical operating data from the past year were obtained, outliers were removed using the 3σ criterion, and all parameters were converted to standardized data with a mean of 0 and a variance of 1 using z-score standardization. Dimensionality reduction was performed using PCA, retaining 8 principal components (cumulative contribution rate ≥95%) to obtain the standardized dataset.

[0033] Construct and train the following prediction models: ① Inlet air fluctuation prediction model: Constructed using an LSTM network. The input is the time series sequence of inlet air parameters over the past 60 minutes (time step of 1 minute). The output is the predicted curves of inlet air flow rate, CO volume fraction, and temperature for the next 15 minutes, and also the predicted value of the inlet H2 / CO volume fraction ratio for the next 15 minutes. The model is trained using the Adam optimizer with the mean squared error (MSE) as the loss function, trained until MSE ≤ 0.01. ② Reactor performance degradation prediction model: Constructed using the XGBoost algorithm. The input is the reactor operating parameters over the past 100 sets, adsorbent state parameters, and the corresponding CO conversion rate and CO2 capture rate. The output is the CO conversion rate decay rate and CO2 capture rate decay rate for the next 8 hours. A performance degradation threshold of 2% is set, and an early warning is triggered when the predicted decay rate exceeds 2%. ③ System output prediction model: The SVR algorithm is used to construct the model. The inputs are the split ratio, inlet parameters, and reactor performance status parameters. The outputs are the corresponding CO conversion rate, CO2 capture rate, and downstream hydrogen-rich gas flow rate. The model parameters are optimized by grid search, and the coefficient of determination R² ≥ 0.95.

[0034] Real-time forecasting and demand analysis: The pre-processed real-time inlet parameters are input into the inlet fluctuation prediction model to obtain the predicted values ​​of feed gas flow rate (380-420 Nm³ / h) and CO volume fraction (18%-22%) for the next 15 minutes, as well as the predicted H2 / CO volume fraction ratio for the next 15 minutes. The real-time reactor operating parameters and adsorbent state parameters (cumulative operating time 6500 h, cumulative cycle count 820 times) are input into the reactor performance degradation prediction model to obtain the predicted values ​​of CO conversion rate decay rate (1.2%) and CO2 capture rate decay rate (1.5%) for the next 8 hours, both of which do not exceed the threshold. Downstream demand parameters are analyzed: hydrogen-rich gas flow rate ≥ 400 Nm³ / h, CO conversion rate ≥ 95%, and CO2 capture rate ≥ 90%.

[0035] Intelligent split ratio decision: Constructing an objective function: Constrained by CO conversion rate ≥95%, CO2 capture rate ≥90%, hydrogen-rich gas flow rate ≥400 Nm³ / h, and the outlet H2 / CO volume fraction ratio approaching the downstream target value, with the optimization objective being the lowest system compression energy consumption; The objective function is solved using a particle swarm optimization algorithm. During the solution process, the current CO conversion rate (real-time detection is 96.8%), the current initial inlet H2 / CO volume fraction ratio (real-time detection is 2.1:1), and the downstream target H2 / CO volume fraction ratio (2.0:1) are combined for real-time calculation and adjustment, finally obtaining the optimal split ratio of 0.72.

[0036] Dynamic adjustment and model update: The controller sends adjustment commands to the diversion valve to adjust the opening of the diversion valve to the position corresponding to the optimal diversion ratio; every hour, the parameters of the three prediction models are fine-tuned using the newly collected 100 sets of real-time data through incremental learning algorithm; the model validation error threshold is set to 0.03. When the model prediction error exceeds 0.03 after fine-tuning, the model is retrained using all historical data.

[0037] In this embodiment, when the feed gas flow rate suddenly fluctuates (from 400 Nm³ / h to 450 Nm³ / h), the inlet gas fluctuation prediction model predicts the fluctuation 10 minutes in advance, and the controller adjusts the split ratio to 0.68 in advance to avoid a sudden rise in reactor bed temperature. After 8 hours of operation, the reactor performance degradation prediction model predicts that the CO conversion rate will decrease by 1.8%, which does not trigger an early warning. The system CO conversion rate is maintained at 96.5%-97.2%, the CO2 capture rate is maintained at 92%-93%, and the downstream hydrogen-rich gas flow rate is maintained at 410-430 Nm³ / h, which meets the requirements.

[0038] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0039] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A smart dynamic diversion control method for an adsorption-enhanced water-gas shift system, characterized in that, The adsorption-enhanced water-gas shift system includes: a feed gas inlet pipe, a bypass pipe, a reactor, a diversion valve, a merging valve, a sensor group, and a controller. The diversion valve is located at the branch point of the feed gas inlet pipe and the bypass pipe, and the merging valve is located at the confluence of the reactor outlet and the bypass pipe outlet. The system is characterized by the following control method: S1. Parameter Acquisition and Preprocessing: The operating parameters of the adsorption-enhanced water-gas shift system are acquired in real time through a sensor array, and historical operating data is also obtained. The acquired operating parameters and historical operating data are preprocessed to obtain a standardized dataset. The operating parameters include inlet parameters, reactor operating parameters, adsorbent state parameters, and downstream demand parameters. S2. Construct and train prediction models: Based on the preprocessed historical operating data, construct an intake fluctuation prediction model, a reactor performance degradation prediction model, and a system output prediction model. Use machine learning algorithms to train and optimize each prediction model to obtain the trained prediction model. S3. Real-time Prediction and Demand Analysis: Input the pre-processed operating parameters into the trained intake fluctuation prediction model to obtain the predicted values ​​of the intake parameters within a preset time period; input the pre-processed operating parameters and historical operating data into the reactor performance degradation prediction model to obtain the predicted values ​​of the reactor performance degradation degree within a preset time period; analyze the downstream process demand parameters to determine the system output target value. S4. Intelligent split ratio decision: Based on the predicted values ​​of inlet parameters, the predicted value of reactor performance degradation, and the target value of system output, the optimal split ratio is solved through the objective function constructed by the controller; S5. Dynamic Adjustment and Model Update: The controller sends adjustment commands to the diversion valve based on the optimal diversion ratio obtained by the solution, so as to realize the dynamic adjustment of the diversion ratio; at the same time, the operating parameters, adjustment results and actual system output parameters are fed back to the prediction model to update and optimize the prediction model online.

2. The intelligent dynamic diversion control method for an adsorption-enhanced water-gas shift system according to claim 1, characterized in that, The intake parameters in S1 include: feed gas flow rate, temperature, pressure, CO volume fraction, H2O volume fraction, and H2 volume fraction; The reactor operating parameters include: reactor bed temperature, bed pressure, outlet CO volume fraction, outlet CO2 volume fraction, outlet H2 / CO volume fraction ratio, and outlet H2 volume fraction; Adsorbent state parameters include: cumulative adsorbent operating time, cumulative adsorption / desorption cycle count, and outlet CO2 volume fraction; Downstream demand parameters include: downstream product gas flow rate demand, CO conversion rate demand, CO2 capture rate demand, downstream demand for H2 / CO volume fraction ratio, and CO2 volume fraction demand.

3. The intelligent dynamic diversion control method for an adsorption-enhanced water-gas shift system according to claim 1, characterized in that, Preprocessing in S1 includes: data cleaning, data standardization, and data dimensionality reduction; Data cleaning uses the 3σ criterion or interpolation to replace outliers; Data standardization can be achieved using either min-max standardization or z-score standardization. Data dimensionality reduction is achieved using principal component analysis or linear discriminant analysis.

4. The intelligent dynamic diversion control method for an adsorption-enhanced water-gas shift system according to claim 1, characterized in that, The intake fluctuation prediction model in S2 is constructed using a long short-term memory network, a gated recurrent unit, or a temporal convolutional network. The input of the intake fluctuation prediction model is a time series of historical intake parameters, and the output is a prediction of the intake parameters for the next 5-30 minutes.

5. The intelligent dynamic diversion control method for an adsorption-enhanced water-gas shift system according to claim 1, characterized in that, The reactor performance degradation prediction model in S2 is constructed using random forest, gradient boosting tree, or convolutional neural network. The input of the reactor performance degradation prediction model is the historical reactor operating parameters, adsorbent state parameters, and corresponding reactor performance indicators. The output is the degree of reactor performance degradation in the next 1-24 hours, and the degree of reactor performance degradation is expressed as the decay rate of performance indicators.

6. The intelligent dynamic diversion control method for an adsorption-enhanced water-gas shift system according to claim 5, characterized in that, The reactor performance degradation prediction model also includes a performance degradation threshold judgment module, which sends an early warning signal to the controller when the predicted performance degradation rate exceeds a preset threshold.

7. The intelligent dynamic diversion control method for an adsorption-enhanced water-gas shift system according to claim 1, characterized in that, The system output prediction model in S2 is constructed using support vector regression, neural network regression, or ensemble learning regression. The inputs to the system output prediction model are the split ratio, inlet parameters, and reactor performance status parameters, and the outputs are the corresponding system output parameters, including CO conversion rate, CO2 capture rate, outlet gas flow rate, and outlet gas H2 / CO volume fraction ratio.

8. The intelligent dynamic diversion control method for an adsorption-enhanced water-gas shift system according to claim 1, characterized in that, In S4, the objective function is constrained by the following conditions: CO conversion rate ≥ preset lower limit, CO2 capture rate ≥ preset lower limit, and downstream product gas flow rate ≥ demand. The optimization objective is to minimize system energy consumption. The objective function is solved using a genetic algorithm, particle swarm optimization algorithm, or simulated annealing algorithm to obtain the optimal split ratio. The optimal split ratio ranges from 0.3 to 0.

9. It is necessary to combine the initial H2 / CO volume fraction ratio of the intake gas, the downstream target H2 / CO volume fraction ratio, and the rated CO conversion rate to comprehensively determine the intake and reaction conditions, ensuring that the split ratio is adapted to the real-time operating conditions and downstream demand.

9. The intelligent dynamic diversion control method for an adsorption-enhanced water-gas shift system according to claim 1, characterized in that, In S5, the online update of the prediction model adopts an incremental learning algorithm. Every preset time interval or after accumulating a certain amount of data, the parameters of each prediction model are fine-tuned using newly collected real-time data. The model validation module is configured to trigger a full retraining of the entire dataset when the updated model's prediction error exceeds a preset error threshold.

10. The intelligent dynamic diversion control method for an adsorption-enhanced water-gas shift system according to claim 1, characterized in that, It also includes an emergency adjustment mechanism: when the sensor detects that the fluctuation of the intake parameters exceeds the preset safety threshold or the reactor has an abnormal operating condition, the controller pauses the machine learning predictive control mode, switches to the preset emergency diversion ratio, and switches back to the intelligent predictive control mode after the system operating condition stabilizes.

Images