Dew point prediction method, device and equipment for fusing agent model with multi-source data

Abstract

The invention relates to the technical field of process industrial production and process optimization, and provides a dew point prediction method, device and equipment for agent model fusion multi-source data, and the method comprises the steps: completing the construction of a normal top volatilization line dew point corrosion risk prediction process mechanism model through Aspen Plus; key influence variables of the process are selected, and historical data and sample data generated by the mechanism model are combined to form a modeling sample data set; training the modeling sample data set by adopting a machine learning algorithm, and constructing and optimizing a normal top volatilization line dew point corrosion risk prediction process agent model; performing volatilization line dew point corrosion based on the proxy model, and evaluating a prediction effect; and fusing multi-source data by adopting an incremental learning technology, and updating the proxy model in real time. The method can dynamically optimize the key operation parameters of the normal top, effectively reduces the dew point corrosion risk, and improves the operation intelligence level of chemical equipment.

Description

Technical Field

[0001] Embodiments of the present invention generally relate to the technical fields of process industry production and process optimization, and more particularly to a dew point prediction method, apparatus, and device integrating a proxy model with multi-source data. Background Art

[0002] As the global refining industry continues to increase the proportion of low-quality crude oil processed, the atmospheric tower top (atmospheric top) system faces increasingly severe flow corrosion challenges. The acidic gases such as H2S, HCl, and CO2 rich in the gas phase of this system form low-pH corrosive media (such as hydrochloric acid and hydrosulfuric acid solutions) during the condensation process, triggering dew point corrosion, resulting in uniform corrosion and pitting of the tower top pipeline, which has become a core bottleneck restricting the long-term safe operation of the device. The severity of dew point corrosion is determined by the synergistic effect of multiple parameters: the temperature field distribution directly affects the amount of acid gas condensation and the concentration gradient of the corrosive medium; when the liquid phase pH value is less than 4, the oxide film on the metal surface dissolves, and the corrosion rate increases exponentially with the increase in acidity; the composition of the acid gas affects the stability of the corrosion product film and the electrochemical reaction path.

[0003] In the existing technology, although domestic and foreign scholars have constructed a thermodynamic model of the top volatility line and a corrosion kinetic equation based on Aspen Plus, and studied the dew point temperature and pH value distribution of parameters such as top oil flow and container pressure in a selected range through sensitivity analysis, single-factor experiments cannot reveal the nonlinear coupling effects of parameters such as oil flow, water injection flow, and gas-liquid phase flow rate (for example, when the water injection flow is insufficient, high-flow acid aggravates erosion corrosion, and excessive water injection dilutes the corrosion inhibitor concentration). It is difficult to extract sensitive features such as corrosion rate mutations and product film impedance anomalies in real time by relying on manual analysis. Empirical prevention and control strategies (such as ammonia injection to adjust pH) have a response delay of 15-30 minutes, which may miss the optimal intervention window.

[0004] The atmospheric tower overhead vaporization line generates dozens of sets of real-time data (such as temperature, pressure, and component concentrations) every minute, resulting in a massive amount of data. Traditional proxy models are unable to dynamically optimize based on the real-time data distribution. When the training data deviates from field conditions by more than 15%, the prediction error can increase to 20%-30%. Summary of the Invention

[0005] To address the above issues, the present invention constructs a multi-source corrosion characteristics database that integrates historical case data, Aspen Plus process simulation data, and DCS real-time production data, establishes an agent model with dynamic learning capabilities, and achieves high-precision real-time monitoring of dew point temperature and corrosion rate during the production process. This forms a closed-loop control system of "real-time prediction-process optimization-optimization model", which can dynamically optimize key operating parameters of the constant top, effectively reduce the risk of dew point corrosion, and enhance the intelligent operation level of chemical equipment, providing scientific support for refining and chemical enterprises to achieve long-term safe operation and improve economic benefits.

[0006] According to an embodiment of the present invention, a method, apparatus, and device for dew point prediction using a proxy model fusing multi-source data are provided.

[0007] In a first aspect of the present invention, a method for dew point prediction using a proxy model fusing multi-source data is provided. The method comprises:

[0008] Step S01: Use Aspen Plus to complete the construction of the mechanism model of the dew point corrosion risk prediction process of the constant top volatile line;

[0009] Step S02: Select key influencing variables of the process, and combine historical data with sample data generated by the mechanism model of the top volatile line dew point corrosion risk prediction process to form a modeling sample data set;

[0010] Step S03: using a machine learning algorithm to train the modeling sample data set, constructing and optimizing a proxy model for the top volatile line dew point corrosion risk prediction process;

[0011] Step S04: performing volatile line dew point corrosion prediction based on the proxy model of the constant top volatile line dew point corrosion risk prediction process, and evaluating the prediction effect;

[0012] Step S05: Adopt incremental learning technology to fuse multi-source data and update the agent model of the top volatile line dew point corrosion risk prediction process in real time.

[0013] Furthermore, the specific steps of step S01 are:

[0014] Step S011: According to the actual operation of the industrial device, the input process parameters of all distributed control systems DCS are counted, and the output data is the dew point temperature: y T And dew point pH: y pH ;

[0015] Step S012: Complete the construction of the mechanism model of the top volatile line dew point corrosion risk prediction process in the Aspen Plus platform;

[0016] Step S013: Collect and obtain industrial field data including process operation data and test data, perform reliability verification and key parameter correction on the mechanism model, and obtain a corrected mechanism model.

[0017] Furthermore, the specific content of the sample data generated by the mechanism model in step S02 is:

[0018] Step S021: Select the key influencing variables of the process, including determining the input variables X=[x1, x2, x3, ..., x n ], the output value Y=[y pH ,y T], where x1,x2,x3,……,x n Adjustable operating parameters input for the topside volatilization line include: topside oil flow, water injection flow, H2S flow, HCl flow, NH3 flow, stripping steam flow, and equivalent flow of NH3 in the neutralizer;

[0019] Step S022: Perform data enhancement on the historical data to generate new samples with similar distribution but different content to the historical data;

[0020] Step S023: Building a Python-Aspen Plus interface to simulate and calculate new samples, as well as input and output simulation data;

[0021] Step S024: Writing a pyautogui program to implement Aspen Plus automated operations;

[0022] Step S025: Start Aspen Plus to perform random simulation experiments and generate a modeling sample data set.

[0023] Furthermore, the generation of new samples with similar distribution but different content to the original data described in step S022 adopts the method of noise injection and feature combination to add a certain degree of noise to the original data to simulate the measurement errors and environmental interference factors in the actual production process.

[0024] Furthermore, the specific steps of step S03 are:

[0025] Step S031: preprocessing the modeling sample data set to eliminate dimensional differences between features;

[0026] Step S032: using the K-fold cross validation method to divide the data set to fully evaluate the performance of the proxy model for the top volatile line dew point corrosion risk prediction process;

[0027] Step S033: using the PSO algorithm to determine the optimal parameters of the proxy model, and optimizing the proxy model for the top volatile line dew point corrosion risk prediction process.

[0028] Furthermore, the specific steps of step S05 are:

[0029] Step S051: Develop a model calling interface and build a real-time communication link that seamlessly connects the model with the industrial data acquisition system;

[0030] Step S052: collecting online data from sensors on the atmospheric tower volatilization line in real time and transmitting the online data to a database;

[0031] Step S053: without retraining the entire model, update the model by combining the online data with the modeling sample data set.

[0032] In a second aspect of the present invention, a device for dew point prediction using a proxy model fusing multi-source data is provided. The device comprises:

[0033] Mechanism model construction module: used to construct the mechanism model of the dew point corrosion risk prediction process using Aspen Plus;

[0034] Dataset generation module: used to select key influencing variables of the process and combine historical data with sample data generated by the mechanism model of the dew point corrosion risk prediction process at the top volatile line to form a modeling sample dataset;

[0035] Surrogate model optimization module: used to train the modeling sample data set using machine learning algorithms, and to build and optimize the surrogate model for the top volatile line dew point corrosion risk prediction process;

[0036] Proxy model evaluation module: used to evaluate the prediction effect of volatile line dew point corrosion based on the proxy model of the constant top volatile line dew point corrosion risk prediction process;

[0037] Agent model update module: used to fuse multi-source data using incremental learning technology and update the agent model of the top volatile line dew point corrosion risk prediction process in real time.

[0038] In a third aspect of the present invention, an electronic device is provided, comprising: a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the program, the method according to the first aspect of the present invention is implemented.

[0039] In a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the method according to the first aspect of the present invention is implemented.

[0040] The present invention constructs a multi-source corrosion characteristic database that integrates historical case data, Aspen Plus process simulation data, and DCS real-time production data, and establishes an agent model with dynamic learning capabilities. This allows high-precision real-time monitoring of dew point temperature and corrosion rate during the production process, forming a closed-loop control system of "real-time prediction-process optimization-optimization model". This system can dynamically optimize key operating parameters of the top, effectively reduce the risk of dew point corrosion, and enhance the intelligent operation level of chemical equipment, providing scientific support for refining and chemical enterprises to achieve long-term safe operation and improve economic benefits.

[0041] It should be understood that the contents described in the summary of the invention are not intended to limit the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description.

[0042] Beneficial effects of the present invention:

[0043] 1. Integrate mechanistic models with data-driven algorithms, combined with PSO parameter tuning, to accurately predict dew point temperature and pH, covering 21 input process parameters and complex crude oil components;

[0044] 2. Dynamic data enhancement techniques such as noise injection and feature combination, combined with K-fold cross-validation, effectively address the scarcity of industrial data and improve the model's adaptability and robustness to complex working conditions and unknown data.

[0045] 3. Realize simulation automation based on Python-Aspen interface and pyautogui program, connect with sensors in real time through RESTful API, combine playback buffer or sliding window technology, support online model update, integrate multi-source data, and ensure long-term prediction stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The above and other features, advantages and aspects of various embodiments of the present invention will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings, in which:

[0047] Figure 1 A flow chart of a method for dew point prediction using an agent model fusing multi-source data according to an embodiment of the present invention is shown;

[0048] Figure 2 A modeling diagram of the atmospheric tower top volatility line according to an embodiment of the present invention is shown;

[0049] Figure 3 A data deviation distribution diagram of predicted values ​​and true values ​​according to an embodiment of the present invention is shown;

[0050] Figure 4 A flowchart of an incremental learning update model according to an embodiment of the present invention is shown;

[0051] Figure 5 A diagram showing online and offline learning methods according to an embodiment of the present invention is shown;

[0052] Figure 6 A block diagram of a device for predicting dew point by fusing multi-source data using an agent model according to an embodiment of the present invention is shown;

[0053] Figure 7 A schematic diagram of a device for dew point prediction based on the fusion of multi-source data using an agent model according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0055] According to the embodiments of the present invention, a method, device, and equipment for dew point prediction by integrating a proxy model with multi-source data are proposed. By constructing a multi-source corrosion characteristic database integrating historical case data, Aspen Plus process simulation data, and DCS real-time production data, an proxy model with dynamic learning capabilities is established. This enables high-precision real-time monitoring of dew point temperature and corrosion rate in the production process, forming a closed-loop control system of "real-time prediction-process optimization-optimization model". This system can dynamically optimize key operating parameters of the constant top, effectively reduce the risk of dew point corrosion, and enhance the intelligent operation level of chemical equipment, providing scientific support for refining and chemical enterprises to achieve long-term safe operation and improve economic benefits.

[0056] The principles and spirit of the present invention are explained in detail below with reference to several representative embodiments of the present invention.

[0057] Figure 1 This is a flow chart of a method for dew point prediction using a proxy model integrating multi-source data according to an embodiment of the present invention. The method includes:

[0058] Step S01: Use Aspen Plus to complete the construction of the mechanism model of the dew point corrosion risk prediction process of the constant top volatile line;

[0059] Step S02: Select key influencing variables of the process, and combine historical data with sample data generated by the mechanism model of the top volatile line dew point corrosion risk prediction process to form a modeling sample data set;

[0060] Step S03: using a machine learning algorithm to train the modeling sample data set, constructing and optimizing a proxy model for the top volatile line dew point corrosion risk prediction process;

[0061] Step S04: performing volatile line dew point corrosion prediction based on the proxy model of the constant top volatile line dew point corrosion risk prediction process, and evaluating the prediction effect;

[0062] Step S05: Adopt incremental learning technology to fuse multi-source data and update the agent model of the top volatile line dew point corrosion risk prediction process in real time.

[0063] It should be noted that although the operations of the method of the present invention are described in a specific order in the above embodiments and drawings, this does not require or imply that these operations must be performed in this specific order, or that all illustrated operations must be performed to achieve the desired results. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0064] In order to more clearly explain the above-mentioned method of dew point prediction by fusing multi-source data with the proxy model, a specific embodiment is described below. However, it should be noted that this embodiment is only for better illustrating the present invention and does not constitute an improper limitation to the present invention.

[0065] The following is a specific example to further illustrate the method of dew point prediction by fusion of multi-source data using the proxy model:

[0066] Step S01: Use Aspen Plus to complete the construction of the mechanism model of the dew point corrosion risk prediction process of the constant top volatile line.

[0067] Step S011: According to the actual operation of the industrial device, the input process parameters of all distributed control systems DCS are counted, and the output data is the dew point temperature: y T And dew point pH: y pH .

[0068] The main components of top-of-the-barrel crude oil are various hydrocarbons and dissolved gases, with light hydrocarbons comprising the majority. H2S, HCl, and NH3 are corrosive components dissolved in the crude oil. During the distillation process in the atmospheric tower, as the temperature rises and the pressure decreases, these dissolved gases and light hydrocarbons evaporate and become part of the non-condensable gas. The sum of the mole fractions of the non-condensable gas components is 1.

[0069] In this embodiment, the specific process of obtaining output data is as follows: Figure 2 As shown, after the top oil is processed in the first separator, vapor and liquid phases are obtained. The vapor phase oil is mixed with non-condensable gas, stripping steam, H2S, HCl, and NH3 in the first mixer and then condensed in the first heat exchanger. An ammonia-containing neutralizer is then injected and mixed with process water in the second mixer. The mixture is then heated and maintained at a temperature through a second heat exchanger before being processed in the second separator. The gaseous outlet of the second separator is the key volatility line for dew point research. A separator module is used to separate the mixed stream into streams of different compositions. A heat exchanger module is then used to simulate the condensation process, defining H2S, HCl, CO2, H2O, and C2-C6 hydrocarbons as key components.

[0070] The gas phase dew point temperature (unit: °C) calculated by the Aspen Plus thermodynamic model is defined as the temperature at which the acid gas mixture begins to condense at the current pressure; the pH value is the acid strength of the corrosive medium at the corresponding dew point temperature.

[0071] The Peng-Robinson equation (applicable to acidic gas systems) is used in the gas phase, and the ElectrolyteNRTL model (describing ion dissociation equilibrium) is used in the liquid phase. The equation HCl→H + +Cl-, Dissociation reaction to obtain the pH value and ion concentration of the condensate.

[0072] Continuously collect DCS data, covering normal operating conditions (70%), variable load conditions (20%), and abnormal fluctuation conditions (10%), and record timestamps and parameter values; regularly collect condensate samples and measure pH values.

[0073] Step S012: Complete the construction of the proxy model for the top volatile line dew point corrosion risk prediction process in the Aspen Plus platform.

[0074] Step S013: Collect and obtain industrial field data including process operation data and test data, perform reliability verification and key parameter correction on the proxy model, and obtain a corrected proxy model.

[0075] The simulated and measured condensate pH values ​​were compared to verify the accuracy of the thermodynamic model.

[0076] For working conditions with temperature difference > 1°C or pH difference > 0.3, error analysis shall be conducted; a "Model Calibration Record" shall be established to archive the calibrated parameters and calibration basis.

[0077] Step S02: Select key influencing variables of the process and generate a modeling sample data set through historical data and a large number of simulation methods.

[0078] Step S021: Select the key influencing variables of the process, including determining the input variables X=[x1, x2, x3, ..., x 20 ], the output value Y=[y pH ,y T ], where x1,x2,x3,……,x 20 The adjustable operating parameters input for the top volatile line include: top oil flow, water injection flow, H2S flow, HCl flow, NH3 flow, stripping steam flow and equivalent flow of NH3 in the neutralizer. The crude oil components are H2, CH4, C2H6, C3H8, C4 hydrocarbons (I-C4H 10 、N-C4H 10 ), C5 hydrocarbons (Y-C5H 12 、Z-C5H12 ), C6 + It is composed of hydrocarbons, O2, N2, CO and CO2, and the sum of the molar fractions of each component is 1.

[0079] In this embodiment, the adjustable process parameters include oil flow rate, water injection flow rate, H2S flow rate, HCl flow rate, NH3 flow rate, stripping steam flow rate, and the equivalent ammonia flow rate in the neutralizer. The operating value range is shown in Table 1:

[0080] Table 1

[0081] Serial number Process parameters Operating value range unit 1 Constant top oil flow 20-100 t / h 2 Water injection flow 2-20 t / h 3 H2S flow 0.10-20.00 kg / h 4 HCl flow rate 0.35-20.00 kg / h 5 NH3 flow 0.50-20.00 kg / h 6 Non-condensable gas flow 100-500 kg / h 7 Stripping steam flow 20-80 t / h 8 NH3 flow in neutralizer 1-20 kg / h

[0082] The stripping steam flow rate has no significant effect on the dew point pH and is therefore used only as an input parameter for the dew point temperature calculation. Furthermore, the neutralizer flow rate is low, its adjustment process is random, and it has little effect on the dew point temperature. Therefore, this parameter is used only as an input parameter for the dew point pH calculation.

[0083] In this embodiment, crude oil is composed of hydrogen, methane, ethane, propane and other three-carbon hydrocarbons, n-butane and other four-carbon hydrocarbons, five-carbon hydrocarbons, six-carbon or higher hydrocarbons, oxygen, nitrogen, carbon monoxide and carbon dioxide. The sum of the crude oil molar fractions is 1, as shown in Table 2.

[0084] Table 2

[0085]

[0086]

[0087] Step S022: Perform data enhancement on the existing data to generate new samples with similar distribution but different content to the original data, thereby expanding the size of the data set and improving data diversity. This step enables the model to learn more data features in different situations and improve the model's generalization ability.

[0088] In this embodiment, the sample is expanded by noise injection and feature combination. A certain degree of noise is added to the original data to simulate the measurement error, environmental interference and other factors in the actual production process. For the top oil flow data, a normal distribution (N(0,σ 2 )), where σ is the standard deviation determined according to the actual measurement error. By adjusting the intensity and distribution of the noise, a series of different new samples can be generated.

[0089] Combining the top oil flow rate and the water injection flow rate yields a new feature. By combining multiple features, a large number of new samples can be generated, enriching the data set.

[0090] Step S023: Build the Python-Aspen Plus interface to launch Aspen Plus simulations and input and output simulation data within the Python environment. Use Python's win32com.client to interact with Aspen Plus. Import this library into your Python code and launch the Aspen Plus software by creating an Aspen Plus COM object. Next, use the object's methods to set the required simulation input parameters, such as the value of the input variable x, and execute the simulation. Finally, read the Aspen Plus output to obtain the desired output value, Y.

[0091] Step S024: To further improve simulation efficiency and reduce manual operation errors, a pyautogui program is written to automate Aspen Plus operations. This program saves the icons and buttons that need to be opened and clicked, and automates the process of resetting parameters, running the simulation, and saving the results.

[0092] Based on the input variable range determined in step S021, a series of different input value combinations are randomly generated. For each input value combination, it is input into Aspen Plus through the Python-Aspen Plus interface for simulation calculation, and the corresponding output value Y=[y pH ,y T ]. Record the input value and output value as a sample, repeat this process multiple times, and build a modeling sample data set.

[0093] It's worth noting that, in practical applications, the adjustable operating parameters and crude oil composition mentioned in step S021 require detailed analysis and determination based on the specific production process and equipment conditions. Furthermore, during data augmentation and simulation experiments, parameters and methods must be adjusted based on actual conditions to ensure that the generated dataset accurately reflects the actual operation of the top volatility curve.

[0094] Step S025: Start Aspen Plus to perform random simulation experiments and generate a modeling sample data set.

[0095] Step S03: Using a machine learning algorithm to train the modeling sample data set, and optimizing the proxy model of the top volatile line dew point corrosion risk prediction process.

[0096] Step S031: Use data preprocessing tools to preprocess the modeling sample data set, eliminate dimensional differences between features, unify the flow unit to kg / h and the component unit to molar fraction, so that the contribution weight of each feature to the loss function during model training is balanced, thereby improving convergence efficiency and generalization ability.

[0097] In this embodiment, a linear scaling method is used to convert the original data range [x min ,x max ] is mapped to [-1,1], the formula is as follows:

[0098]

[0099] Among them, x represents the original data value, x min and x max are the minimum and maximum values ​​of the feature in the sample set, respectively, norm The original data value x including the top oil flow rate determined in step S021 is traversed through the sample statistics to obtain the maximum and minimum values ​​x of each parameter. min 、x max , get the normalized data value x norm .

[0100] It is worth noting that if there are significant outliers in the sample (such as out-of-limit values ​​caused by transient failure of the pressure sensor), the outliers are first eliminated using the 3σ rule, and then the remaining data are normalized to avoid extreme values ​​distorting the scaling interval.

[0101] After completing the prediction task, in order to obtain the original scale results with practical significance, the dew point temperature y T and dew point pHy pH The normalized predicted value y norm Perform restoration. Perform denormalization. Denormalization can restore the prediction results to the scale range of the original data for subsequent analysis and decision-making. The denormalization formula is as follows:

[0102]

[0103] Taking a single sample as an example, y T The normalized predicted value of is 0.6, and its original range is [60℃, 100℃]. The restored value is:

[0104] Step S032: Use K-fold cross validation method to divide the data set to fully evaluate the model performance.

[0105] In this example, the samples were stratified based on the distribution of the output variable Y, with the dew point pH divided into three intervals: pH < 3, 3 ≤ pH < 5, and pH ≥ 5. This ensured that each subset contained samples from each corrosion risk level. Using K = 5, the dataset was randomly divided into five subsets, and a 5-fold cross-validation was performed.

[0106] Step S033: Use the particle swarm optimization (PSO) algorithm to determine the optimal parameters of the agent model. A group of particles are randomly initialized. The position and velocity of each particle are randomly generated in the search space. At the same time, the initial position of each particle is recorded as its individual optimal position, and the position with the best fitness value among all particles is found as the global optimal position. The specific steps are as follows:

[0107] Step S0331: Calculate the fitness value: Calculate the fitness value of each particle according to the objective function and evaluate its quality.

[0108] Step S0332: Update individual optimum and global optimum: For each particle, if the fitness value of its current position is better than the fitness value of its individual optimal position, then update the individual optimal position; if the fitness value of a particle's current position is better than the fitness value of the global optimal position, then update the global optimal position.

[0109] Step S0333: Update the speed and position of the particles: Update the speed and position of each particle according to the following formula:

[0110]

[0111] x i (t+1)=x i (t)+v i (t+1)

[0112] Among them, v i (t) is the velocity of particle i in generation t, x i (t) is the position of particle i in the tth generation, w is the inertia weight, which is used to balance the global search and local search capabilities of the particle; c1 and c2 are acceleration constants, which are usually positive; r1 and r2 are random numbers in the range [0,1]; is the individual optimal position of particle i, g best is the global optimal position.

[0113] Step S0334: Termination condition judgment: Repeat steps S0332 to S0333 until the termination condition is met, such as reaching the maximum number of iterations or finding a solution that meets the accuracy requirements.

[0114] In this embodiment, the XGBoost algorithm is selected to train the modeling sample data set, with a population size of 40, an evolutionary generation of 80, and a crossover probability of 0.8, to obtain a proxy model for the dew point corrosion risk prediction process of the top volatile line: the maximum depth of each decision tree is max_depth = 5, the shrinkage factor learning_rate = 0.017, the subsampling parameter sub_sample = 0.51, and the feature sampling rate colsample_bytree = 1.00.

[0115] Step S04: performing volatile line dew point corrosion prediction based on the proxy model of the constant top volatile line dew point corrosion risk prediction process, and evaluating the prediction effect.

[0116] Specifically, the surrogate model is used to predict the dew point temperature and pH in the offline stage, and a model evaluation index system is established, including mean square error, root mean square error, mean absolute error and determination coefficient.

[0117] In the present invention, when the evaluation mean square error, root mean square error, and mean absolute error of the prediction model are smaller, and the determination coefficient is closer to 1, the prediction accuracy of the model is higher. It is worth noting that the evaluation absolute error MAE, root mean square error RMSE and square correlation coefficient R 2 The expressions are as follows:

[0118]

[0119] Where n is the number of samples, y i is the true value, is the predicted value, is the mean of the true values.

[0120] In this embodiment, the MSE, RMSE, and MAE of the proxy model based on the dew point corrosion risk prediction process of the top volatile line are the smallest on both the training set and the test set, and the correlation R 2 Closest to 1. Training set: Test set: Validation set = 60% (6000 groups): 20% (2000 groups): 20% (2000 groups). When predicting dew point temperature, MSE = 0.0105, RMSE = 0.1025, MAE = 0.0759, R 2 =0.9982. When predicting dew point pH, MSE = 0.1707, RMSE = 0.4132, MAE = 0.3153, R 2 =0.9596.

[0121] Figure 3 a and Figure 3 b are the prediction results and relative errors of the proxy model for the dew point corrosion risk prediction process of the constant top volatile line. From the analysis of the figures, we can see that the vast majority of the prediction errors are controlled between [-2%, 2%], which shows that the constructed proxy model has good fitting accuracy and generalization ability.

[0122] Step S05: Incremental learning technology is used to fuse multi-source data, and online data is used to update the agent model of the top volatile line dew point corrosion risk prediction process in real time.

[0123] Step S051: Develop a model calling interface and build a real-time communication link that seamlessly connects the model with the industrial data acquisition system to ensure low-latency transmission of process parameters and prediction results.

[0124] In this embodiment, a RESTful API architecture is adopted, based on the HTTP / HTTPS protocol, and the following core interfaces are defined:

[0125] POST / data / upload: Receives real-time process parameters; GET / prediction: Returns dew point temperature, predicted pH value, and risk level; PUT / model / update: Triggers an online model update request. API key authentication is used, verifying the caller's identity through the request header X-API-Key. The interface service runs on an industrial-grade edge computing server and supports integration with DCS systems via the OPCUA protocol.

[0126] Step S052: Collect online data in real time from various sensors (such as temperature sensor, pressure sensor, pH sensor, etc.) on the atmospheric tower volatilization line, and transmit the online data to the database.

[0127] In this embodiment, temperature sensors are deployed, and the field control cabinet transmits data to the edge server via industrial Ethernet using the Modbus TCP protocol. The edge server performs online denoising on the real-time data: median filtering is used to remove pulse interference, and trend verification is performed to eliminate logically contradictory values.

[0128] Step S053: without retraining the entire model, update the model by combining the online data with the modeling sample data set.

[0129] Under the time series condition, the model update process is as follows Figure 4 As shown. Preferably, a time window is defined as the last seven days of data, and only the data within that window is used to train the model. The window gradually slides over time, automatically removing outdated data and adapting to process condition drift. This focuses on the current operating condition distribution, improving the model's responsiveness to real-time data. In this example, dew point temperature and dew point pH are learned in five incremental steps.

[0130] In this embodiment, the construction and application process of the atmospheric tower volatile line dew point corrosion risk prediction model is as follows: Figure 5 As shown in the figure, the method is divided into two parts: offline and online. In the offline stage, the historical data of dew point temperature and dew point pH and Aspen simulation data are integrated. After preprocessing, the particle swarm algorithm (PSO) is used to determine the model parameters, complete the model design, training and testing, and optimize the effect. In the online stage, after preprocessing the enterprise field data, the model is updated through incremental learning, the prediction results are output, and the forgetting rate is calculated. The calculation formula of the forgetting rate is as follows:

[0131]

[0132] Refers to the performance index value of the model on the old task dataset before learning the new task; Refers to the performance index value of the model on the same dataset of the old task after learning the new task.

[0133] Preferably, the forgetting rate of dew point temperature of the fused online data source is -0.342, and the forgetting rate of dew point pH is -0.748. The forgetting rate is negative, which ensures that the learning of new data distribution is achieved while reducing the forgetting of old knowledge.

[0134] Based on the same inventive concept, the present invention also proposes a device for dew point prediction using a proxy model that fuses multi-source data. The implementation of the device can refer to the implementation of the above method, and the repeated parts will not be repeated. Figure 6 As shown, the device 100 includes:

[0135] Mechanism model construction module 101: used to complete the construction of the mechanism model of the dew point corrosion risk prediction process of the constant top volatile line using Aspen Plus;

[0136] Dataset generation module 102: used to select key influencing variables of the process, and combine historical data with sample data generated by the mechanism model of the top volatile line dew point corrosion risk prediction process to form a modeling sample dataset;

[0137] Agent model optimization module 103: used to train the modeling sample data set using a machine learning algorithm to build and optimize the agent model for the top volatile line dew point corrosion risk prediction process;

[0138] The proxy model evaluation module 104 is used to evaluate the prediction effect of volatile line dew point corrosion based on the proxy model of the constant top volatile line dew point corrosion risk prediction process;

[0139] The agent model updating module 105 is used to fuse multi-source data using incremental learning technology and update the agent model of the top volatile line dew point corrosion risk prediction process in real time.

[0140] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0141] like Figure 7As shown, the device includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for the operation of the device can also be stored. The CPU, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0142] Many components in a device are connected to the I / O interface, including: input units, such as a keyboard and mouse; output units, such as various types of displays and speakers; storage units, such as magnetic disks and optical disks; and communication units, such as network cards, modems, and wireless communication transceivers. The communication unit allows the device to exchange information / data with other devices via computer networks such as the Internet and / or various telecommunication networks.

[0143] The processing unit performs the various methods and processes described above, such as method steps S01 to S05. For example, in some embodiments, method steps S01 to S05 can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program can be loaded and / or installed on the device via a ROM and / or a communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more of the method steps S01 to S05 described above can be executed. Alternatively, in other embodiments, the CPU can be configured to execute method steps S01 to S05 by any other appropriate means (for example, by means of firmware).

[0144] The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), and the like.

[0145] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. Such program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0146] In the context of the present invention, machine-readable medium can be a tangible medium that can contain or store a program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0147] In addition, although adopting specific order to describe each operation, this should be understood as requiring such operation to be carried out in the specific order shown or in sequential order, or requiring all illustrated operations to be carried out to obtain desired result.Under certain environment, multitasking and parallel processing may be advantageous.Similarly, although comprising some specific implementation details in the above discussion, these should not be construed as limiting the scope of the present invention.Some features described in the context of independent embodiment can also be realized in single realization in combination.On the contrary, the various features described in the context of independent realization also can be realized in multiple realizations individually or in the mode of any suitable subcombination.

[0148] Although the subject matter has been described in language specific to structural features and / or methodological logical acts, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are merely example forms of implementing the claims.

Claims

1. A method for dew point prediction using a proxy model integrating multi-source data, characterized in that: The method includes: Step S01: Use Aspen Plus to complete the construction of the mechanism model of the dew point corrosion risk prediction process of the constant top volatile line; Step S02: Select key influencing variables of the process, and combine historical data with sample data generated by the mechanism model of the dew point corrosion risk prediction process at the top volatile line to form a modeling sample data set; Step S03: using a machine learning algorithm to train the modeling sample data set, constructing and optimizing a proxy model for the top volatile line dew point corrosion risk prediction process; Step S04: performing volatile line dew point corrosion prediction based on the proxy model of the constant top volatile line dew point corrosion risk prediction process, and evaluating the prediction effect; Step S05: Adopt incremental learning technology to fuse multi-source data and update the agent model of the top volatile line dew point corrosion risk prediction process in real time.

2. The method for dew point prediction using a proxy model integrating multi-source data according to claim 1, characterized in that: The specific steps of step S01 are: Step S011: According to the actual operation of the industrial device, the input process parameters of all distributed control systems DCS are counted, and the output data is the dew point temperature: y T And dew point pH: y pH ; Step S012: Complete the construction of the mechanism model of the top volatile line dew point corrosion risk prediction process in the Aspen Plus platform; Step S013: Collect and obtain industrial field data including process operation data and test data, perform reliability verification and key parameter correction on the mechanism model, and obtain a corrected mechanism model.

3. The method for dew point prediction using a proxy model integrating multi-source data according to claim 1, characterized in that: The specific content of the sample data generated by the mechanism model in step S02 is: Step S021: Select the key influencing variables of the process, including determining the input variables X=[x1, x2, x3, ..., x n ] range, the output value Y=[y pH ,y T ], where x1,x2,x3,……,x n Adjustable operating parameters input for the topside volatilization line include: topside oil flow, water injection flow, H2S flow, HCl flow, NH3 flow, stripping steam flow, and equivalent flow of NH3 in the neutralizer; Step S022: Perform data enhancement on the historical data to generate new samples with similar distribution but different content to the historical data; Step S023: Building a Python-Aspen Plus interface to simulate and calculate new samples, as well as input and output simulation data; Step S024: Writing a pyautogui program to implement Aspen Plus automated operations; Step S025: Start Aspen Plus to perform random simulation experiments and generate a modeling sample data set.

4. The method for dew point prediction using a proxy model integrating multi-source data according to claim 3, characterized in that: The generation of new samples with similar distribution but different content to the original data described in step S022 adopts the method of noise injection and feature combination to add a certain degree of noise to the original data to simulate the measurement errors and environmental interference factors in the actual production process.

5. The method for dew point prediction by integrating multi-source data with an agent model according to claim 1, characterized in that: The specific steps of step S03 are: Step S031: preprocessing the modeling sample data set to eliminate dimensional differences between features; Step S032: using the K-fold cross validation method to divide the data set to fully evaluate the performance of the proxy model for the top volatile line dew point corrosion risk prediction process; Step S033: using the PSO algorithm to determine the optimal parameters of the proxy model, and optimizing the proxy model for the dew point corrosion risk prediction process of the constant top volatile line.

6. The method for dew point prediction using a proxy model integrating multi-source data according to claim 1, characterized in that: The specific steps of step S05 are: Step S051: Develop a model calling interface and build a real-time communication link that seamlessly connects the model with the industrial data acquisition system; Step S052: collecting online data from sensors on the atmospheric tower volatilization line in real time and transmitting the online data to a database; Step S053: without retraining the entire model, update the model by combining the online data with the modeling sample data set.

7. A device for predicting dew point by integrating multi-source data with a proxy model, characterized in that: The device implements the method according to any one of claims 1 to 6, comprising: Mechanism model construction module: used to construct the mechanism model of the dew point corrosion risk prediction process using Aspen Plus; Dataset generation module: used to select key influencing variables of the process and combine historical data with sample data generated by the mechanism model of the dew point corrosion risk prediction process at the top volatile line to form a modeling sample dataset; Surrogate model optimization module: used to train the modeling sample data set using machine learning algorithms, and to build and optimize the surrogate model for the top volatile line dew point corrosion risk prediction process; Proxy model evaluation module: used to evaluate the prediction effect of volatile line dew point corrosion based on the proxy model of the constant top volatile line dew point corrosion risk prediction process; Agent model update module: used to fuse multi-source data using incremental learning technology and update the agent model of the top volatile line dew point corrosion risk prediction process in real time.

8. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images