Adaptive process control method and system based on thermodynamic prediction and machine learning

Abstract

The application discloses a self-adaptive process control method and system based on thermodynamic prediction and machine learning, wherein thermodynamic data of a multi-component mixture is input into a thermodynamic prediction model to calculate a theoretical optimal state of reaction; a future behavior trend of the system is obtained through a machine learning prediction model in combination with historical production and online sensor data; and an APC control module fuses the two results to automatically optimize control variables, so that the system operation approaches a thermodynamic optimal point. The application takes into account both physical limits and actual dynamics, improves control precision in complex working conditions, and is suitable for continuous production processes such as chemical industry and oil refining, and helps to optimize operation and save energy and reduce emissions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of process industry control, and in particular to an adaptive process control method and system based on thermodynamic prediction and machine learning. BACKGROUND

[0002] Industrial control system (ICS) is the core technology foundation in modern industrial production, which can realize automatic alarm and emergency shutdown function through real-time monitoring of key parameters (such as temperature, pressure, flow, etc.), prevent overload, overheating, leakage and other dangerous situations, reduce human error operation, and ensure the safety of production; through precise control of production process, ensure that the reaction conditions and operating parameters are stable in the optimal interval, which can reduce the batch product difference, improve product consistency and traceability; the advanced control system can realize optimal feeding ratio and energy utilization maximization, thereby improving the unit time output.

[0003] At present, most industrial control systems still mainly use traditional DCS and PID regulation, which can realize stable control to a certain extent, but the control accuracy and economy are limited when dealing with multivariable coupling, strong nonlinear process, large time delay system and complex constraint conditions. Although the APC system has better control ability, it depends on the experience to set the target value or heuristic model, and lacks the understanding of the physical limit of the process. Although pure thermodynamic modeling can predict the reaction limit state, it is difficult to capture the disturbance factors in actual operation.

[0004] Therefore, how to invent an adaptive process control method based on thermodynamic prediction and machine learning to realize intelligent process control driven by physical mechanism and actual behavior has become a problem to be solved. SUMMARY

[0005] Therefore, the present application provides an adaptive process control method and system based on thermodynamic prediction and machine learning, which provides a theoretical optimal benchmark through a thermodynamic model, captures actual dynamic disturbance through a machine learning model, and realizes real-time optimization by fusing the two through adaptive APC control, solves the problems of insufficient control accuracy of traditional DCS and PID in complex conditions such as multivariable coupling, lack of physical limit cognition of existing APC depending on experience, and difficulty of pure thermodynamic modeling in resisting actual disturbance, and realizes precise optimization of industrial process.

[0006] In order to achieve the above purpose, the present application provides the following technical scheme: an adaptive process control method based on thermodynamic prediction and machine learning, comprising:

[0007] The thermodynamic data of a plurality of groups of multi-component mixtures is input into the thermodynamic prediction model in the form of Excel;

[0008] Based on the thermodynamic data, data preprocessing and analysis are performed on each group of multi-component mixture by the thermodynamic prediction model, and the theoretical optimal state of the reaction under the current working condition is calculated;

[0009] Based on historical production data and online sensor data, prediction is performed by a machine learning prediction model to obtain the future behavior trend of the control system;

[0010] Based on the theoretical optimal state and the future behavior trend of the control system, the current control variables are automatically optimized by the APC control module, so that the control system operates close to the thermodynamic optimal point.

[0011] As a preferred scheme of the adaptive process control method based on thermodynamic prediction and machine learning, the control system periodically learns and updates the model based on new data generated during the operation of the control system to adapt to the changes of aging and contaminated equipment.

[0012] As a preferred scheme of the adaptive process control method based on thermodynamic prediction and machine learning, the thermodynamic data of the multi-component mixture includes temperature, pressure, name of each component, mole percentage of each component in the mixture, NASA polynomial coefficient of each component, standard enthalpy of formation of each component, standard entropy of each component, average heat capacity of each component, and equilibrium constant of each component.

[0013] As a preferred scheme of the adaptive process control method based on thermodynamic prediction and machine learning, in the process of data preprocessing and analysis of each group of multi-component mixture by the thermodynamic prediction model, the thermodynamic formula of the thermodynamic prediction model includes ideal gas state equation, Van der Waals equation, Van't Hoff equation, Clapeyron equation and Gibbs free energy formula.

[0014] The expression of the ideal gas state equation is:

[0015] PV=nRT

[0016] In the formula, P is the gas pressure; V is the gas volume; n is the amount of substance; R is the gas constant; T is the thermodynamic temperature;

[0017] The expression of the Van der Waals equation is:

[0018]

[0019] In the formula, a is a constant for correcting intermolecular force; b is a constant for correcting molecular volume;

[0020] The expression of the Van't Hoff equation is:

[0021]

[0022] wherein K1, K2 are the reaction equilibrium constants at temperatures T1, T2 respectively; ΔH is the enthalpy change of the reaction; R is the gas constant; T1, T2 are the thermodynamic temperatures;

[0023] The expression of the Clapeuron equation is:

[0024]

[0025] wherein, is the rate of change of the equilibrium pressure with temperature; ΔH m is the molar enthalpy of phase transition; T is the phase transition temperature; ΔV m is the molar volume change;

[0026] The expression of the Gibbs free energy formula is:

[0027] ΔG = -RTlnK

[0028] wherein ΔG is the Gibbs free energy change of the reaction; R is the gas constant; T is the thermodynamic temperature; K is the equilibrium constant of the reaction.

[0029] As a preferred solution of the adaptive process control method based on thermodynamic prediction and machine learning, the machine learning prediction model comprises: a supervised regression model and a time series model;

[0030] The expression of the supervised regression model is:

[0031]

[0032] wherein, is the predicted process index; x i is the operation variable; f k (x) is the kth decision tree; K is the number of decision trees;

[0033] The expression of the time series model is:

[0034] f t = σ(W f · [h t-1 , x t ] + b f )

[0035] i t = σ(W i · [h t-1 , x t ] + b i )

[0036] C t = f t · C t-1 + it • tanh(W C • [h t-1 , x t ] + b C )

[0037] h t = O t • tanh(C t )

[0038] where f t is the output of the forget gate; σ is the sigmoid function; W f is the weight matrix of the forget gate; b f is the corresponding bias term; h t is the hidden state of the current time step; h t-1 is the hidden state of the previous time step; x t is the input vector of the current time step; i t is the output of the input gate; W i is the weight matrix of the input gate; b i is the corresponding bias term; C t is the cell state of the current time step; C t-1 is the cell state of the previous time step; W C is the weight matrix of the candidate state update; b C is the corresponding bias term; O t is the output of the output gate.

[0039] As a preferred solution of the adaptive process control method based on thermodynamic prediction and machine learning, in the process of automatically optimizing the current control variable through the APC control module, the current control variable optimization problem is formalized through a multivariable predictive control architecture;

[0040] The formalized formula is:

[0041]

[0042] where u t is the decision variable that needs to be optimized; N is the prediction horizon; y t+k|t is the system output predicted at time t for the future time t+k; is the desired output at the future time t+k; λ is the regularization parameter; Δu t+k-1 is the change of control input;

[0043] The constraint condition is:

[0044] x t+k+1 = Ax t+k + Bu t+k

[0045] y t+k = Cx t+k + Du t+k

[0046] u min ≤ u t+k ≤ u max

[0047] y min ≤ y t+k ≤ y max

[0048] wherein x t+k is a state vector at time t+k; x t+k+1 is a state vector at time t+k+1; u t+k is a control variable; y t+k is a predicted output variable; A, B, C, D are respectively a state transition matrix, a control input matrix, an output matrix, and a direct transfer matrix; u min is a minimum value of the control input; u max is a maximum value of the control input; y min is a minimum value of the system output; y max is a maximum value of the system output.

[0049] The application also provides an adaptive process control system based on thermodynamic prediction and machine learning, based on the above adaptive process control method based on thermodynamic prediction and machine learning, comprising:

[0050] A thermodynamic data input unit for inputting thermodynamic data of a plurality of groups of multi-component mixtures in the form of Excel into a thermodynamic prediction model;

[0051] A thermodynamic data preprocessing and analysis unit for performing data preprocessing and analysis on each group of multi-component mixtures based on the thermodynamic data through the thermodynamic prediction model, and calculating to obtain a theoretical optimal state of the reaction under the current working condition;

[0052] A control system future behavior trend prediction unit for predicting through a machine learning prediction model based on historical production data and online sensor data to obtain a control system future behavior trend;

[0053] An APC control module automatic optimization unit for automatically optimizing a current control variable through an APC control module based on the theoretical optimal state and the control system future behavior trend, so that the control system runs close to a thermodynamic optimal point.

[0054] As a preferred solution of the adaptive process control system based on thermodynamic prediction and machine learning, the control system periodically learns and updates the model to adapt to the changes of aging and contaminated equipment based on new data generated during the operation of the control system.

[0055] As a preferred solution of the adaptive process control system based on thermodynamic prediction and machine learning, in the thermodynamic data input unit, the thermodynamic data of the multi-component mixture includes: temperature, pressure, name of each component substance, mole percentage of each component substance in the mixture, NASA polynomial coefficient of each component substance, standard enthalpy of formation of each component, standard entropy of each component substance, average heat capacity of each component substance and equilibrium constant of each component substance.

[0056] As a preferred solution of the adaptive process control system based on thermodynamic prediction and machine learning, in the thermodynamic data preprocessing and analysis unit, in the process of data preprocessing and analysis of each multi-component mixture by the thermodynamic prediction model, the thermodynamic formula of the thermodynamic prediction model includes: ideal gas equation of state, van der Waals equation, Van't Hoff equation, Clapearon equation and Gibbs free energy formula.

[0057] The expression of the ideal gas equation of state is:

[0058] PV = nRT

[0059] In the formula, P is the gas pressure; V is the gas volume; n is the amount of substance; R is the gas constant; T is the thermodynamic temperature;

[0060] The expression of the van der Waals equation is:

[0061]

[0062] In the formula, a is a constant for correcting intermolecular force; b is a constant for correcting molecular volume;

[0063] The expression of the Van't Hoff equation is:

[0064]

[0065] In the formula, K1, K2 are the reaction equilibrium constants at temperatures T1, T2 respectively; ΔH is the enthalpy change of the reaction; R is the gas constant; T1, T2 are the thermodynamic temperatures;

[0066] The expression of the Clapearon equation is:

[0067]

[0068] In the formula, is the rate of change of equilibrium pressure with temperature; ΔHm is the molar enthalpy of phase transition; T is the phase transition temperature; ΔV m is the molar volume change;

[0069] The expression of the Gibbs free energy formula is:

[0070] ΔG = -RTlnK

[0071] In the formula, ΔG is the Gibbs free energy change of the reaction; R is the gas constant; T is the thermodynamic temperature; K is the equilibrium constant of the reaction.

[0072] As a preferred scheme of the adaptive process control system based on thermodynamic prediction and machine learning, in the future behavior trend prediction unit of the control system, the machine learning prediction model comprises: a supervised regression model and a time series model;

[0073] The expression of the supervised regression model is:

[0074]

[0075] In the formula, is the predicted process index; x i is the operation variable; f k (x) is the kth decision tree; K is the number of decision trees;

[0076] The expression of the time series model is:

[0077] f t = σ(W f · [h t-1 , x t ] + b f )

[0078] i t = σ(W i · [h t-1 , x t ] + b i )

[0079] C t = f t · C t-1 + i t · tanh(W C · [h t-1 , x t ] + b C )

[0080] h t = O t · tanh(C t )

[0081] In the formula, f tis the output of the forget gate; σ is the sigmoid function; W f is the weight matrix of the forget gate; b f is the corresponding bias term; h t is the hidden state of the current time step; h t-1 is the hidden state of the previous time step; x t is the input vector of the current time step; i t is the output of the input gate; W i is the weight matrix of the input gate; b i is the corresponding bias term; C t is the cell state of the current time step; C t-1 is the cell state of the previous time step; W C is the weight matrix of the candidate state update; b C is the corresponding bias term; O t is the output of the output gate.

[0082] As a preferred solution of the adaptive process control system based on thermodynamic prediction and machine learning, in the automatic optimization unit of the APC control module, in the process of automatic optimization of the current control variable by the APC control module, the current control variable optimization problem is formalized by a multivariable predictive control architecture;

[0083] The formalized formula is:

[0084]

[0085] In the formula, u t is the decision variable to be optimized; N is the prediction time domain; y t+k|t is the system output predicted at time t+k in the future at time t; is the expected output at time t+k in the future; λ is a regularization parameter; Δu t+k-1 is the change of control input;

[0086] The constraint condition is:

[0087] x t+k+1 = Ax t+k + Bu t+k

[0088] y t+k = Cx t+k + Du t+k

[0089] u min ≤ u t+k ≤ u max

[0090] y min ≤ y t+k ≤ ymax

[0091] In the formula, x t+k is a state vector at time t+k; x t+k+1 is a state vector at time t+k+1; u t+k is a control variable; y t+k is a predicted output variable; A, B, C, and D are state transition matrix, control input matrix, output matrix, and direct transfer matrix, respectively; u min is a minimum value of the control input; u max is a maximum value of the control input; y min is a minimum value of the system output; y max is a maximum value of the system output.

[0092] The present application has the following advantages: the present application inputs the thermodynamic data of a plurality of groups of multi-component mixtures into a thermodynamic prediction model in the form of Excel; based on the thermodynamic data, the present application performs data preprocessing and analysis on each group of multi-component mixtures by the thermodynamic prediction model, and calculates to obtain a theoretically optimal state of a reaction under a current working condition; based on historical production data and online sensor data, the present application performs prediction by a machine learning prediction model, and obtains a future behavior trend of a control system; based on the theoretically optimal state and the future behavior trend of the control system, the present application performs automatic optimization on a current control variable by an APC control module, so that the control system operates close to a thermodynamic optimal point. The present application can be applicable to reactions, separations, heat exchanges and other multi-unit operations, and can also be extended to a reinforcement learning control, a self-learning digital twin system and other higher intelligent platforms. The present application takes thermodynamics as a scientific basis, takes machine learning as a feedback correction means, and takes APC as a real-time execution platform, realizes a leap from experience control to intelligent optimal control of an industrial control system, and has wide industrial application prospects and promotion value. BRIEF DESCRIPTION OF DRAWINGS

[0093] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only exemplary, and for those skilled in the art, other drawings can also be derived from the provided drawings without creative labor.

[0094] The structures, proportions, sizes, etc. shown in the present specification are only used to cooperate with the content disclosed in the specification, to be understood and read by those skilled in the art, and do not define the limiting conditions for the implementation of the present application, so they do not have technical substantial significance. Any modification of structure, change of proportion relationship or adjustment of size, without affecting the effects and purposes that can be achieved by the present application, should still fall within the scope covered by the disclosed technical content.

[0095] Figure 1 A flowchart of the adaptive process control method based on thermodynamic prediction and machine learning provided in Embodiment 1 of the present application;

[0096] Figure 2 A schematic diagram of the thermodynamic model prediction model preprocessing and analysis results in the adaptive process control method based on thermodynamic prediction and machine learning provided in Embodiment 1 of the present application;

[0097] Figure 3 A schematic diagram of the control system future behavior trend in the adaptive process control method based on thermodynamic prediction and machine learning provided in Embodiment 1 of the present application;

[0098] Figure 4 A schematic diagram of the APC control module automatic optimization in the adaptive process control method based on thermodynamic prediction and machine learning provided in Embodiment 1 of the present application;

[0099] Figure 5 A schematic diagram of the architecture of the adaptive process control system based on thermodynamic prediction and machine learning provided in Embodiment 2 of the present application. DETAILED DESCRIPTION

[0100] The embodiments of the present application will be described in detail by specific embodiments, and those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in the specification. Obviously, the described embodiments are part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0101] Embodiment 1

[0102] Reference Figure 1 Embodiment 1 of the present application provides an adaptive process control method based on thermodynamic prediction and machine learning, comprising the following steps:

[0103] S1, input the thermodynamic data of a plurality of groups of multi-component mixtures into the thermodynamic prediction model in the form of Excel;

[0104] S2, based on the thermodynamic data, data preprocessing and analysis of each group of multi-component mixture are carried out by the thermodynamic prediction model, and the theoretical optimal state of the reaction under the current working condition is calculated and obtained;

[0105] S3, based on the historical production data and online sensor data, prediction is carried out by the machine learning prediction model, and the future behavior trend of the control system is obtained;

[0106] S4, based on the theoretical optimal state and the future behavior trend of the control system, automatically optimizing the current control variable through an APC control module, so that the control system runs close to the thermodynamic optimal point.

[0107] In this embodiment, in step S1, the thermodynamic data of a plurality of groups of multi-component mixtures are input into the thermodynamic prediction model in the form of Excel;

[0108] Specifically, when constructing the thermodynamic data set of the multi-component mixture, the thermodynamic data set of the multi-component mixture includes temperature, pressure, name of each component substance, mole percentage of each component substance in the mixture, NASA polynomial coefficient of each component substance, standard enthalpy of formation of each component, standard entropy of each component substance, average heat capacity of each component substance, equilibrium constant of each component substance, etc. The data set is input into the thermodynamic prediction model of the adaptive advanced process control system (APC) based on thermodynamic prediction and machine learning enhancement in the form of Excel. Each row represents a group of mixtures, including temperature, pressure and other information. The system will automatically convert temperature to Kelvin and percentage to mole fraction.

[0109] In this embodiment, in step S2, based on the thermodynamic data, the thermodynamic prediction model is used to pre-process and analyze the data of each group of multi-component mixtures, and the theoretical optimal state of the reaction under the current working condition is calculated and obtained;

[0110] In the process of pre-processing and analyzing the data of each group of multi-component mixtures by the thermodynamic prediction model, the thermodynamic formula of the thermodynamic prediction model includes: ideal gas state equation, van der Waals equation, Van't Hoff equation, Clapearon equation and Gibbs free energy formula;

[0111] The expression of the ideal gas state equation is:

[0112] PV = nRT

[0113] In the formula, P is the gas pressure; V is the gas volume; n is the amount of substance; R is the gas constant; T is the thermodynamic temperature;

[0114] The expression of the van der Waals equation is:

[0115]

[0116] In the formula, a is a constant for correcting intermolecular force; b is a constant for correcting molecular volume;

[0117] The expression of the Van't Hoff equation is:

[0118]

[0119] In the formula, K1, K2 are reaction equilibrium constants at temperatures T1, T2 respectively; ΔH is the enthalpy change of the reaction; R is the gas constant; T1, T2 are thermodynamic temperatures;

[0120] The expression of the Clapeuron equation is:

[0121]

[0122] In the formula, is the rate of change of the equilibrium pressure with temperature; ΔH m is the molar phase transition enthalpy; T is the phase transition temperature; ΔV m is the molar volume change;

[0123] The expression of the Gibbs free energy formula is:

[0124] ΔG = -RTlnK

[0125] In the formula, ΔG is the Gibbs free energy change of the reaction; R is the gas constant; T is the thermodynamic temperature; K is the equilibrium constant of the reaction.

[0126] In the process control for methane combustion in the embodiment, when the thermodynamic formula is used to calculate the theoretical optimal state of the reaction of each group of multi-component mixtures under the current working condition, first, the input data is read, when the system automatically identifies that the input information contains reaction entropy change and enthalpy change, the Gibbs free energy calculation formula is used to judge the direction and reversibility of the reaction. When the system automatically identifies that the input information contains Gibbs free energy and temperature, the equilibrium constant calculation formula is used to judge the equilibrium constant of the reaction under the current state, and the formula is specifically:

[0127] K = e -ΔG / RT

[0128] In the formula, e is the base of natural logarithm.

[0129] When the system automatically identifies that the input information contains NASA polynomial coefficients, the theoretical flame temperature (about 2200K) is calculated and predicted according to the NASA polynomial approximation, which is used as the upper limit of the safety control of the APC for reference, and the calculation formula is:

[0130]

[0131] In the formula, ΔT ad is the adiabatic temperature change; ΔH rxn is the enthalpy change of the chemical reaction; n i is the amount of substance of the i-th component in the system; C p,i (T) is the molar heat capacity at constant pressure of the i-th component at temperature T; T is the temperature.

[0132] When the system automatically identifies that the input information contains NASA polynomial coefficients, the enthalpy and entropy values of the reaction at the current state are calculated according to the NASA polynomial, and the calculation formula is:

[0133]

[0134] In the formula, H is the molar enthalpy of the system; R is the universal gas constant; T is the temperature; S is the molar entropy of the system; a1, a2, a3, a4, a5, a6, and a7 are empirical coefficients.

[0135] In one possible embodiment, when the thermodynamic formula is used to calculate the theoretical optimal state of the reaction at the current working condition for each group of multi-component mixtures in the continuous liquid phase industrial reaction of epichlorohydrin, the input data is first read, and when the system automatically identifies that the input information contains reaction entropy and enthalpy, the Gibbs free energy calculation formula is preferentially selected to judge the direction and reversibility of the reaction. When the system automatically identifies that the input information contains Gibbs free energy and temperature, the equilibrium constant calculation formula is preferentially selected to judge the equilibrium constant of the reaction at the current state, and the specific formula is:

[0136] K = e -ΔG / RT

[0137] In the formula, e is the base of natural logarithm.

[0138] When the system automatically identifies that the input information contains reaction enthalpy and temperature, the Van't Hoff equation is preferentially selected to predict the influence of temperature on the reaction, and the calculation formula is:

[0139]

[0140] In the formula, is the derivative of the logarithm of the equilibrium constant with respect to temperature; ΔH is the enthalpy change of the chemical reaction; R is the universal gas constant; and T is the temperature.

[0141] When the system automatically identifies that the input information contains NASA polynomial coefficients, the enthalpy and entropy values of the reaction at the current state are calculated according to the NASA polynomial, and the calculation formula is:

[0142]

[0143] In the formula, H is the molar enthalpy of the system; R is the universal gas constant; T is the temperature; S is the molar entropy of the system; a1, a2, a3, a4, a5, a6, and a7 are empirical coefficients.

[0144] In a possible embodiment, in the production of ammonium nitrate thermal decomposition process, when calculating the theoretical optimal state of each group of multi-component mixture under the current working condition, first, the input data is read, and the system automatically identifies that the input information contains decomposition enthalpy, and the decomposition enthalpy is preferentially used to predict the heat release rate of the reaction, and the formula is as follows:

[0145] Q = n * DeltaH rxn

[0146] In the formula, Q is the total reaction heat; n is the amount of substance of the reactant; DeltaH rxn is the molar enthalpy change of the reaction.

[0147] When the system automatically identifies that the input information contains NASA polynomial coefficients, the specific heat capacity calculated by the NASA polynomial is preferentially selected for temperature rise evaluation, and the formula is as follows:

[0148]

[0149] In the formula, T1 and T2 are the lower and upper limits of the temperature; C p (t) is the specific heat capacity of the substance at constant pressure.

[0150] When the system automatically identifies that the input information contains activation energy and temperature, the Arrhenius equation and Gibbs free energy change are preferentially selected to calculate the reaction rate to evaluate temperature sensitivity, and the calculation formula is as follows:

[0151]

[0152] In the formula, k(T) is the reaction rate constant when the temperature is T; A is the pre-exponential factor; E a is the activation energy; R is the universal gas constant; and T is the temperature.

[0153] In a possible embodiment, in the production of ammonium nitrate thermal decomposition process, when calculating the theoretical optimal state of each group of multi-component mixture under the current working condition, first, the input data is read, and the system automatically identifies that the input information contains decomposition enthalpy, and the decomposition enthalpy is preferentially used to predict the heat release rate of the reaction, and the formula is as follows:

[0154] DeltaG = -nFE cell

[0155] In the formula, DeltaG is the Gibbs free energy change of the electrode reaction; n is the number of electrons transferred in the electrode reaction; F is the Faraday constant; and E cell is the cell electromotive force.

[0156] When the system automatically identifies that the input information contains NASA polynomial coefficients, the entropy and enthalpy values of the reaction at the current state are calculated according to the NASA polynomial, and the calculation formula is as follows:

[0157]

[0158] wherein H is the molar enthalpy of the system; R is the universal gas constant; T is the temperature; S is the molar entropy of the system; a1, a2, a3, a4, a5, a6, a7 are empirical coefficients.

[0159] Meanwhile, the system can automatically identify the numbers of each substance, and calculate the mixing enthalpy of the multi-component mixture according to the molar percentage of different substances and the enthalpy values of different substances. The calculation formula is:

[0160]

[0161] wherein H mix is the mixing enthalpy; x i is the molar fraction of component i; H i (T) is the enthalpy value of pure component i at temperature T.

[0162] The system can combine the thermodynamic entropy-enthalpy estimation theory to calculate the theoretical minimum electrolysis voltage. The calculation formula is:

[0163]

[0164] wherein E min is the theoretical minimum electrolysis voltage; AH is the reaction enthalpy change; AS is the reaction entropy change; T is the temperature; n is the number of electrons transferred in the electrolysis reaction; F is the Faraday constant.

[0165] In this embodiment, the thermodynamic formula is used to pre-process and analyze the data of each multi-component mixture to obtain the theoretical optimal state of the reaction under the current working condition. The pre-processing and analysis results of the thermodynamic model prediction model in the adaptive process control method based on thermodynamic prediction and machine learning are shown in Figure 2 After inputting the reaction conditions and initial components into the system, the system is initialized and the gas model is loaded. The chemical equilibrium constant is calculated under the condition of maintaining temperature and pressure. The product distribution at equilibrium is output and the trace components are filtered out. The results are saved and displayed through images. Through the images, the content percentage of the products generated when each reaction reaches equilibrium under different initial conditions can be clearly seen.

[0166] In this embodiment, in step S3, the future behavior trend of the control system is obtained by predicting through the machine learning prediction model based on historical production data and online sensor data.

[0167] Specifically, the mapping relationship of the machine learning prediction model constructed based on historical production data and online sensor data mainly includes the mapping relationship between product properties such as product yield, energy consumption, and purity, and control variables such as temperature, pressure, and flow.

[0168] In the process control for methane combustion, the machine learning prediction model uses the XGBoost regression model to predict the conversion rate, the CO emission modeling variable contribution, the LSTM time series model to predict the yield trend in the next few hours considering the system delay and dynamic changes, and the Bayesian optimization reverse search to search the optimal setting conditions to achieve the goal of thermodynamic prediction.

[0169] In a possible embodiment, in the continuous liquid phase industrial reaction for epichlorohydrin, the machine learning prediction model uses the XGBoost regression model to predict the conversion rate, the relationship between the ECH concentration and the operating conditions, the LSTM time series model to predict the yield trend in the next few hours considering the system delay and dynamic changes, and the Bayesian optimization reverse search to search the optimal setting conditions to achieve the goal of thermodynamic prediction.

[0170] In a possible embodiment, in the production of ammonium nitrate thermal decomposition process, the machine learning prediction model mainly uses the LSTM neural network to monitor the abnormal pressure rise trend based on the mapping relationship constructed by the historical production data and the online sensor data.

[0171] In a possible embodiment, in the chlor-alkali electrolysis reaction, the machine learning prediction model mainly uses the XGBoost, LSTM, PLS regression, neural network and other algorithms based on the mapping relationship constructed by the historical production data and the online sensor data.

[0172] In this embodiment, the machine learning prediction model includes a supervised regression model and a time series model.

[0173] The expression of the supervised regression model is:

[0174]

[0175] In the formula, is the predicted process index; x i is the operating variable; f k (x) is the kth decision tree; K is the number of decision trees;

[0176] The expression of the time series model is:

[0177] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0178] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0179] C t = f t · C t-1 + i t · tanh(W C · [h t-1 , x t ] + b C )

[0180] h t = O t · tanh(C t )

[0181] where f t is the output of the forget gate; σ is the sigmoid function; W f is the weight matrix of the forget gate; b f is the corresponding bias term; h t is the hidden state of the current time step; h t-1 is the hidden state of the previous time step; x t is the input vector of the current time step; i t is the output of the input gate; W i is the weight matrix of the input gate; b i is the corresponding bias term; C t is the cell state of the current time step; C t-1 is the cell state of the previous time step; W C is the weight matrix of the candidate state update; b C is the corresponding bias term; O t is the output of the output gate.

[0182] In this embodiment, based on historical production data and online sensor data, prediction is carried out through a machine learning prediction model to obtain the future behavior trend of the control system. The schematic diagram of the future behavior trend of the control system in the adaptive process control method based on thermodynamic prediction and machine learning is as shown in Figure 3 The data is loaded into the system, the temperature and pressure are selected as input variables, a random forest is created and trained for each product, the molar fraction of each product under future conditions is predicted by simulating future experimental conditions, the prediction results are converted into DataFrame and saved to a file, and the prediction curves of all products are drawn. Through the image, the percentage content of each product component under the set temperature and pressure can be clearly seen.

[0183] In this embodiment, in step S4, based on the theoretical optimal state and the future behavior trend of the control system, the current control variable is automatically optimized through the APC control module, so that the control system runs close to the thermodynamic optimal point.

[0184] Specifically, asFigure 4 As shown, the constructed adaptive APC control module mainly adopts a multivariable predictive control (MPC) framework, which can adaptively adjust the prediction time domain, model weight and target interval. The APC control module selects the thermodynamic model output as the "target window", and the machine learning model is used for controller constraint and target update. Through dynamic adjustment of the controller on the operating variables such as feed, temperature and reflux ratio, the maximum yield, minimum energy consumption or optimal comprehensive index is realized.

[0185] In this embodiment, in the process of automatically optimizing the current control variable by the APC control module, the current control variable optimization problem is formalized through a multivariable predictive control framework;

[0186] The formalized formula is:

[0187]

[0188] In the formula, u t is the decision variable to be optimized; N is the prediction time domain; y t+k|t is the system output at time t+k predicted at time t; is the expected output at time t+k; λ is a regularization parameter; Δu t+k-1 is the change of control input;

[0189] The constraint condition is:

[0190] x t+k+1 = Ax t+k + Bu t+k

[0191] y t+k = Cx t+k + Du t+k

[0192] u min ≤ u t+k ≤ u max

[0193] y min ≤ y t+k ≤ y max

[0194] In the formula, x t+k is the state vector at time t+k; x t+k+1 is the state vector at time t+k+1; u t+k is the control variable; y t+k is the predicted output variable; A, B, C, D are respectively the state transition matrix, the control input matrix, the output matrix and the direct transfer matrix; u min is the minimum value of the control input; u max is the maximum value of the control input; ymin is the minimum value of the system output; y max is the maximum value of the system output.

[0195] In this embodiment, based on the new data generated in the operation process of the control system, the control system performs periodic self-learning and model updating to adapt to the changes of the aging and polluted equipment.

[0196] Specifically, the adaptive advanced process control system (APC) based on thermodynamic prediction and machine learning enhancement adapts to equipment changes through the combination of data-driven modeling, model updating mechanism and abnormal monitoring algorithm. When the system identifies that the process behavior and model prediction deviation increase, it means that the system or equipment characteristics have changed. The system will periodically update the machine learning model to maintain the fitting accuracy of the actual process, and update the APC controller target in real time to ensure that the control strategy matches the real working condition.

[0197] The application scenarios of the present application are as follows:

[0198] The present application is applicable to reaction units, such as parameter optimization and safety control of chemical reaction processes such as methane combustion, liquid phase reaction of epichlorohydrin, thermal decomposition of ammonium nitrate, etc. The present application can balance and control the purity and energy consumption in separation processes such as rectification and extraction; at the same time, it can optimize the efficiency of various heat transfer equipment. The present application can be extended to special processes such as chlor-alkali electrolysis, and can also support the construction of intelligent platforms such as reinforcement learning control, self-learning digital twin system, etc.

[0199] It should be noted that the method of the embodiments of the present disclosure can be executed by a single device, such as a computer or a server, etc. The method of the embodiments can also be applied in a distributed scenario, and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only execute one or more steps in the method of the embodiments of the present disclosure, and the multiple devices can interact with each other to complete the method.

[0200] It should be noted that some embodiments of the present disclosure have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different than that described above and still achieve desirable results. Additionally, the processes depicted in the figures do not necessarily require the particular order shown, or sequential order, to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous.

[0201] Embodiment 2

[0202] Referring to Figure 5 Embodiment 2 of the present application also provides an adaptive process control system based on thermodynamic prediction and machine learning, comprising:

[0203] a thermodynamic data input unit 001, configured to input thermodynamic data of a plurality of multi-component mixtures in the form of Excel into a thermodynamic prediction model;

[0204] a thermodynamic data preprocessing and analysis unit 002, configured to perform data preprocessing and analysis on each multi-component mixture based on the thermodynamic data through the thermodynamic prediction model, and calculate a theoretical optimal state of reaction under a current working condition;

[0205] a control system future behavior trend prediction unit 003, configured to predict a future behavior trend of a control system based on historical production data and online sensor data through a machine learning prediction model;

[0206] an APC control module automatic optimization unit 004, configured to perform automatic optimization on a current control variable through an APC control module based on the theoretical optimal state and the future behavior trend of the control system, so that the control system runs close to a thermodynamic optimal point.

[0207] In the embodiment, the thermodynamic data of the multi-component mixture in the thermodynamic data input unit 001 includes temperature, pressure, names of each component substance, molar percentage of each component substance in the mixture, NASA polynomial coefficients of each component substance, standard enthalpy of formation of each component, standard entropy of each component substance, average heat capacity of each component substance, and equilibrium constant of each component substance.

[0208] In the embodiment, in the process of performing data preprocessing and analysis on each multi-component mixture through the thermodynamic prediction model in the thermodynamic data preprocessing and analysis unit 002, the thermodynamic formula of the thermodynamic prediction model includes an ideal gas state equation, a Van der Waals equation, a Van't Hoff equation, a Clapeyron equation, and a Gibbs free energy formula.

[0209] The expression of the ideal gas state equation is:

[0210] PV = nRT

[0211] In the formula, P represents gas pressure, V represents gas volume, n represents the amount of substance, R represents a gas constant, and T represents thermodynamic temperature.

[0212] The expression of the Van der Waals equation is:

[0213]

[0214] In the formula, a represents a constant for correcting intermolecular force, and b represents a constant for correcting molecular volume.

[0215] The expression of the Van't Hoff equation is:

[0216]

[0217] In the formula, K1, K2 are reaction equilibrium constants at temperatures T1, T2 respectively; ΔH is the enthalpy change of the reaction; R is the gas constant; T1, T2 are thermodynamic temperatures;

[0218] The expression of the Clapeuron equation is:

[0219]

[0220] In the formula, is the rate of change of the equilibrium pressure with temperature; ΔH m is the molar enthalpy of phase transition; T is the phase transition temperature; ΔV m is the molar volume change;

[0221] The expression of the Gibbs free energy formula is:

[0222] ΔG = -RTlnK

[0223] In the formula, ΔG is the Gibbs free energy change of the reaction; R is the gas constant; T is the thermodynamic temperature; K is the equilibrium constant of the reaction.

[0224] In the control system future behavior trend prediction unit 003 in the embodiment, the machine learning prediction model comprises a supervised regression model and a time series model;

[0225] The expression of the supervised regression model is:

[0226]

[0227] In the formula, is the predicted process index; x i is the operation variable; f k (x) is the kth decision tree; K is the number of decision trees;

[0228] The expression of the time series model is:

[0229] f t = σ(W f ·[h t-1 ,x t ]+b f )

[0230] i t = σ(W i ·[h t-1 ,x t ]+b i )

[0231] C t= f t · C t-1 + i t · tanh(W C · [h t-1 , x t ] + b C )

[0232] h t = O t · tanh(C t )

[0233] where f t is the output of the forget gate; σ is the sigmoid function; W f is the weight matrix of the forget gate; b f is the corresponding bias term; h t is the hidden state of the current time step; h t-1 is the hidden state of the previous time step; x t is the input vector of the current time step; i t is the output of the input gate; W i is the weight matrix of the input gate; b i is the corresponding bias term; C t is the cell state of the current time step; C t-1 is the cell state of the previous time step; W C is the weight matrix of the candidate state update; b C is the corresponding bias term; O t is the output of the output gate.

[0234] In the APC control module automatic optimization unit 004 in the present embodiment, in the process of automatically optimizing the current control variable through the APC control module, the current control variable optimization problem is formalized through a multivariable predictive control framework;

[0235] The formalized formula is:

[0236]

[0237] where u t is the decision variable that needs to be optimized; N is the prediction horizon; y t+k|t is the system output predicted at time t+k in the future at time t; is the desired output at time t+k in the future; λ is the regularization parameter; Δu t+k-1 is the change in control input;

[0238] The constraint condition is:

[0239] x t+k+1 = Ax t+k + But+k

[0240] y t+k = Cx t+k + Du t+k

[0241] u min ≤ u t+k ≤ u max

[0242] y min ≤ y t+k ≤ y max

[0243] wherein x t+k is a state vector at time t+k; x t+k+1 is a state vector at time t+k+1; u t+k is a control variable; y t+k is a predicted output variable; A, B, C, D are respectively a state transition matrix, a control input matrix, an output matrix, a direct transfer matrix; u min is a minimum value of the control input; u max is a maximum value of the control input; y min is a minimum value of the system output; y max is a maximum value of the system output.

[0244] The method embodiment in Example 1 is based on the same concept, and the technical effects brought by the method embodiment are the same as those of the present application. For specific content, refer to the description in the foregoing method embodiments of the present application, which will not be repeated here.

[0245] Example 3

[0246] Example 3 of the present application provides a non-transitory computer-readable storage medium, wherein the computer-readable storage medium stores program code of a self-adaptive process control method based on thermodynamic prediction and machine learning, and the program code includes instructions for executing the self-adaptive process control method based on thermodynamic prediction and machine learning of Example 1 or any possible implementation manner thereof.

[0247] The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk (SSD)), etc.

[0248] Example 4

[0249] Example 4 of the present application provides an electronic device, comprising: a memory and a processor;

[0250] The processor and the memory complete the communication between each other through the bus; the memory stores program instructions executable by the processor, and the processor calling the program instructions can execute the adaptive process control method based on thermodynamic prediction and machine learning of embodiment 1 or any possible implementation manner thereof.

[0251] Specifically, the processor can be implemented by hardware or software. When implemented by hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented by software, the processor can be a general-purpose processor, which reads software codes stored in the memory to implement the processor. The memory can be integrated in the processor or exist independently outside the processor.

[0252] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable systems. The computer instructions can be stored in a computer readable storage medium or transferred from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions can be transferred from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) mode.

[0253] Obviously, those skilled in the art should understand that the above-mentioned modules or steps of the present application can be realized by a general computing system, which can be concentrated on a single computing system or distributed on a network composed of multiple computing systems, and optionally, they can be realized by program codes executable by the computing system, so that they can be stored in the storage system and executed by the computing system, and in some cases, the steps shown or described can be executed in different order, or they can be manufactured into individual integrated circuit modules, or multiple modules or steps can be manufactured into a single integrated circuit module. Thus, the present application is not limited to any specific combination of hardware and software.

[0254] Although the present application has been described in detail with general description and specific embodiments above, it is obvious to those skilled in the art that some modifications or improvements can be made on the basis of the present application. Therefore, these modifications or improvements made on the basis of not deviating from the spirit of the present application, all belong to the scope of protection claimed by the present application.

Claims

1. A method of adaptive process control based on thermodynamic predictions and machine learning, characterized in that, The method comprises the following steps: inputting thermodynamic data of a plurality of multi-component mixtures in the form of Excel into a thermodynamic prediction model; based on the thermodynamic data, performing data preprocessing and analysis on each multi-component mixture by the thermodynamic prediction model to calculate a theoretical optimal state of the reaction under the current working condition; based on historical production data and online sensor data, predicting the future behavior trend of the control system by a machine learning prediction model; based on the theoretical optimal state and the future behavior trend of the control system, automatically optimizing the current control variables by an APC control module to make the control system operate close to the thermodynamic optimal point.

2. The method of adaptive process control based on thermodynamic predictions and machine learning of claim 1, wherein, Based on the new data generated during the operation of the control system, the control system periodically learns and updates the model to adapt to the changes of the aging and contaminated equipment.

3. The method of adaptive process control based on thermodynamic predictions and machine learning of claim 2, wherein, The thermodynamic data of the multi-component mixture includes temperature, pressure, the name of each component, the mole percentage of each component in the mixture, the NASA polynomial coefficient of each component, the standard enthalpy of formation of each component, the standard entropy of each component, the average heat capacity of each component, and the equilibrium constant of each component.

4. The method of adaptive process control based on thermodynamic predictions and machine learning of claim 3, wherein, In the process of data preprocessing and analysis on each multi-component mixture by the thermodynamic prediction model, the thermodynamic formula of the thermodynamic prediction model includes the ideal gas state equation, the Van der Waals equation, the Van't Hoff equation, the Clapeyron equation, and the Gibbs free energy formula. The expression of the ideal gas state equation is: PV = nRT where P is the gas pressure, V is the gas volume, n is the amount of substance, R is the gas constant, and T is the thermodynamic temperature. The expression of the Van der Waals equation is: where a is the constant for correcting intermolecular force, and b is the constant for correcting molecular volume. The expression of the Van't Hoff equation is: where K1 and K2 are the reaction equilibrium constants at temperatures T1 and T2, respectively, ΔH is the enthalpy change of the reaction, R is the gas constant, and T1 and T2 are the thermodynamic temperatures. The expression of the Clapeyron equation is: wherein is the rate of change of pressure with temperature; ΔΗ m is the molar enthalpy of phase change; T is the phase change temperature; Δν m is the molar volume change; The expression of the Gibbs free energy formula is: ΔG = -RTlnK where ΔG is the Gibbs free energy change of the reaction, R is the gas constant, T is the thermodynamic temperature, and K is the equilibrium constant of the reaction.

5. The method of adaptive process control based on thermodynamic predictions and machine learning according to claim 4, characterized in that, The machine learning prediction model includes a supervised regression model and a time series model. The expression of the supervised regression model is: In the formula, is a predicted process index; x i is an operating variable; f k (x) is the kth decision tree; K is the number of decision trees; The expression of the time series model is: f t = σ(W f · [h t-1 , x t ]+ b f ) i t = σ(W i · [h t-1 , x t ]+ b i ) C t = f t · C t-1 + i t · tanh(W C · [h t-1 , x t ] + b C ) h t =O t tanh(C t ) where f t is the output of the forget gate; σ is the sigmoid function; W f is the weight matrix of the forget gate; b f is the corresponding bias term; h t is the hidden state of the current time step; h t-1 is the hidden state of the previous time step; x t is the input vector of the current time step; i t is the output of the input gate; W i is the weight matrix of the input gate; b i is the corresponding bias term; C t is the cell state of the current time step; C t-1 is the cell state of the previous time step; W C is the weight matrix of the candidate state update; b C is the corresponding bias term; O t is the output of the output gate.

6. The method of adaptive process control based on thermodynamic predictions and machine learning according to claim 5, wherein, In the process of automatically optimizing the current control variables by the APC control module, the current control variable optimization problem is formalized by a multivariate predictive control framework. The formalized formula is: where u t is the decision variable to be optimized; N is the prediction horizon; y t+k|t is the system output predicted at time t for future time t+k; is the desired output at future time t+k; λ is the regularization parameter; Δu t+k-1 is the change in control input; The constraint condition is: x t+k+1 = Ax t+k + Bu t+k y t+k = Cx t+k + Du t+k u min ≤u t+k ≤u max y min ≤y t+k ≤y max where x t+k is the state vector at time t+k; x t+k+1 is the state vector at time t+k+1; u t+k is the control variable; y t+k is the predicted output variable; A, B, C, D are the state transition matrix, the control input matrix, the output matrix, and the direct transfer matrix, respectively; u min is the minimum value of the control input; u max is the maximum value of the control input; y min is the minimum value of the system output; y max is the maximum value of the system output.

7. An adaptive process control system based on thermodynamic predictions and machine learning, employing the adaptive process control method based on thermodynamic predictions and machine learning according to any one of claims 1 to 6, characterized in that The method comprises the following steps: a thermodynamic data input unit for inputting thermodynamic data of a plurality of multi-component mixtures in the form of Excel into a thermodynamic prediction model; a thermodynamic data preprocessing and analysis unit for performing data preprocessing and analysis on each multi-component mixture by the thermodynamic prediction model based on the thermodynamic data to calculate a theoretical optimal state of the reaction under the current working condition; The control system future behavior trend prediction unit predicts, based on historical production data and online sensor data, through a machine learning prediction model, to obtain a control system future behavior trend; The APC control module automatic optimization unit is configured to automatically optimize, based on the theoretical optimal state and the control system future behavior trend, a current control variable through an APC control module, so that the control system operates close to a thermodynamic optimal point.

8. The thermodynamics prediction and machine learning based adaptive process control system of claim 7, wherein, In the thermodynamic data input unit, the thermodynamic data of the multi-component mixture includes temperature, pressure, name of each component substance, mole percentage of each component substance in the mixture, NASA polynomial coefficient of each component substance, standard enthalpy of formation of each component, standard entropy of each component substance, average heat capacity of each component substance, and equilibrium constant of each component substance.

9. The thermodynamics prediction and machine learning based adaptive process control system of claim 8, wherein, In the thermodynamic data preprocessing and analysis unit, in the process of preprocessing and analyzing data of each multi-component mixture through the thermodynamic prediction model, the thermodynamic formula of the thermodynamic prediction model includes an ideal gas state equation, a Van der Waals equation, a Van't Hoff equation, a Clapeyron equation, and a Gibbs free energy formula. The expression of the ideal gas state equation is: PV = nRT In the expression, P is gas pressure, V is gas volume, n is the amount of substance, R is gas constant, and T is thermodynamic temperature. The expression of the Van der Waals equation is: In the expression, a is a constant for correcting intermolecular force, and b is a constant for correcting molecular volume. The expression of the Van't Hoff equation is: In the expression, K1 and K2 are reaction equilibrium constants at temperatures T1 and T2 respectively, ΔH is enthalpy change of the reaction, R is gas constant, and T1 and T2 are thermodynamic temperatures. The expression of the Clapeyron equation is: wherein is the rate of change of pressure with temperature; ΔΗ m is the molar enthalpy of phase change; T is the phase change temperature; Δν m is the molar volume change; The expression of the Gibbs free energy formula is: ΔG = -RTlnK In the expression, ΔG is Gibbs free energy change of the reaction, R is gas constant, T is thermodynamic temperature, and K is equilibrium constant of the reaction.

10. The thermodynamics prediction and machine learning based adaptive process control system of claim 9, wherein, In the control system future behavior trend prediction unit, the machine learning prediction model includes a supervised regression model and a time series model. The expression of the supervised regression model is: wherein is a predicted process metric; x i is an operating variable; f k (x) is the kth decision tree; K is the number of decision trees; The expression of the time series model is: f t = σ(W f · [h t-1 , x t ]+ b f ) i t = σ(W i · [h t-1 , x t ]+ b i ) C t = f t · C t-1 + i t · tanh(W C · [h t-1 , x t ] + b C ) h t =O t tanh(C t ) where f t is the output of the forget gate; σ is the sigmoid function; W f is the weight matrix of the forget gate; b f is the corresponding bias term; h t is the hidden state of the current time step; h t-1 is the hidden state of the previous time step; x t is the input vector of the current time step; i t is the output of the input gate; W i is the weight matrix of the input gate; b i is the corresponding bias term; C t is the cell state of the current time step; C t-1 is the cell state of the previous time step; W C is the weight matrix of the candidate state update; b C is the corresponding bias term; O t is the output gate.

Images