CN-122021301-A - Method, system, equipment and medium for modeling state of steam feed pump and back calculation of steam inflow flow based on thermodynamic mechanism and machine learning

CN122021301ACN 122021301 ACN122021301 ACN 122021301ACN-122021301-A

Abstract

The application relates to the technical field of intelligent management of a steam-driven water supply pump, in particular to a steam-driven water supply pump state modeling and steam inflow back calculation method, system, equipment and medium based on thermodynamic mechanism and machine learning, which comprises the steps of constructing a thermodynamic mechanism model based on a first law of thermodynamics and an equipment design efficiency curve; the method comprises the steps of obtaining real-time operation parameters of a steam-driven water supply pump set, calculating theoretical parameters through a thermodynamic mechanism model, constructing and inputting the real-time operation parameters and the theoretical parameters into a pre-trained machine learning correction model together, outputting efficiency correction factors, calibrating theoretical isentropic efficiency through the efficiency correction factors to obtain actual isentropic efficiency, and calculating and outputting high-precision values of steam inlet flow of a steam turbine and enthalpy of steam exhaust of the steam turbine through the thermodynamic mechanism model based on the actual isentropic efficiency. The application combines the physical interpretability of the mechanism model and the self-adaptive capacity of the data driving model, and can realize high-precision back calculation of the inlet steam flow and dynamic tracking of the equipment performance state.

Inventors

WANG JIAN
LIU XINJUN
Cui Haodi
TIAN ZHIGANG
LI XUESONG
SUN LEIPENG
SUN HAITAO
LI YAPENG
WU DEJIN
TANG JUN
LI HAO
BAI JIAN

Assignees

青岛华丰伟业电力科技工程有限公司

Dates

Publication Date: 20260512
Application Date: 20260128

Claims (10)

1. A method for modeling the state of a steam feed pump and back calculation of the inflow steam based on thermodynamic mechanism and machine learning is characterized by comprising the following steps: s1, constructing a thermodynamic mechanism model of the water supply pump set based on a thermodynamic first law and an equipment design efficiency curve, wherein the thermodynamic mechanism model defines the relationship among the following parameters: A calculated relationship between feed water flow, feed water pump inlet pressure, feed water pump outlet pressure, feed water density, and feed water pump shaft power; the mapping relation between the water supply pump shaft power and the theoretical isentropic efficiency; Calculating relation among steam inlet enthalpy of the steam turbine, condenser pressure and isentropic enthalpy drop of steam; The calculation relation between the water supply pump shaft power, isentropic enthalpy drop, isentropic efficiency and steam turbine inlet flow and steam turbine exhaust enthalpy; S2, acquiring real-time operation parameters of a steam-driven water supply pump set, wherein the real-time operation parameters comprise water supply flow, water supply pump inlet pressure, water supply pump outlet pressure, water supply density, steam turbine steam inlet enthalpy and condenser pressure; s3, calculating theoretical parameters including theoretical water supply pump shaft power, theoretical isentropic enthalpy drop, theoretical isentropic efficiency, theoretical steam turbine steam inlet flow and theoretical steam turbine steam exhaust enthalpy through a thermodynamic mechanism model based on real-time operation parameters; S4, constructing real-time operation parameters and theoretical parameters together into input feature vectors, inputting the input feature vectors into a pre-trained machine learning correction model, and outputting an efficiency correction factor representing the deviation degree of the current performance of the equipment relative to the design performance of the equipment The machine learning correction model is a regression model constructed based on a gradient lifting tree algorithm; S5, calibrating the theoretical isentropic efficiency by utilizing the efficiency correction factor to obtain the actual isentropic efficiency The expression is: representing theoretical isentropic efficiency; s6, calculating and outputting high-precision values of steam inlet flow of the steam turbine and exhaust enthalpy of the steam turbine through a thermodynamic mechanism model based on actual isentropic efficiency.
2. The method for modeling and back-calculating intake flow rate of a steam feed pump state according to claim 1, wherein in step S1, the calculated relationship among the feed water flow rate, the feed water pump inlet pressure, the feed water pump outlet pressure, the feed water density, and the feed water pump shaft power is defined by the following formula: Wherein, the Representing the water supply pump shaft power; representing the feed water flow rate; representing the feed pump outlet pressure; representing feed pump inlet pressure; representing feed water density; the water feeding pump efficiency is represented as a preset constant value or is obtained by inquiring the performance curve of the water feeding pump according to the operation working condition.
3. The method for modeling and back-calculating steam inflow rate according to claim 1, wherein in step S1, the calculated relationship between the steam turbine inflow enthalpy, the condenser pressure and the isentropic enthalpy drop of the steam is defined by the following formula: Wherein, the Represents the isentropic enthalpy drop of the steam; representing steam turbine admission enthalpy; Representing isentropic exhaust enthalpy value according to condenser pressure And inquiring a water vapor thermodynamic property table to obtain the water vapor thermodynamic property table.
4. The method for modeling and back-calculating the steam feed pump state according to claim 1, wherein in step S1, the calculated relationship between the power of the feed pump shaft, isentropic enthalpy drop, isentropic efficiency, steam turbine steam inlet flow, and steam turbine exhaust enthalpy is defined by the following formula: Wherein, the Representing the steam inlet flow of the steam turbine; representing the water supply pump shaft power; Representing isentropic enthalpy drop; representing isentropic efficiency; Representing the exhaust enthalpy of the steam turbine; representing the steam turbine admission enthalpy.
5. The method for modeling and back calculation of steam feed pump state as defined in claim 1, wherein in step S4, the machine learning correction model performs regression prediction by integrating a plurality of decision trees based on a gradient lifting tree algorithm, wherein each decision tree performs nonlinear transformation on an input feature vector according to a feature splitting rule, and the model outputs a final efficiency correction factor The core formula for the weighted sum of all decision tree outputs is expressed as: Wherein, the Representing an input feature vector; Represent the first For input feature vectors, a decision tree is used Is a predictive output of (2); Represent the first The weight coefficient of the decision tree is obtained by minimizing a loss function in the training process; Representing the total number of decision trees.
6. The method for modeling and back calculation of intake flow rate according to claim 1, wherein in step S4, the training step of the machine learning correction model is as follows: S401, collecting a historical data set, wherein the historical data set comprises historical real-time operation parameters of a plurality of training samples, corresponding actual measured water supply pump shaft power and actual measured steam turbine inlet flow; s402, based on historical real-time operation parameters of each training sample, calculating corresponding theoretical parameters including theoretical water supply pump shaft power, theoretical isentropic enthalpy drop, theoretical isentropic efficiency, theoretical steam turbine steam inlet flow and theoretical steam turbine steam exhaust enthalpy through a thermodynamic mechanism model; S403, based on actual measurement of water supply pump shaft power and actual measurement of steam turbine inlet flow rate of each training sample, calculating actual isentropic efficiency Practical isentropic efficiency of individual training samples The calculation formula of (2) is as follows: Wherein, the Represent the first The actual shaft power corresponding to the training samples; Represent the first The actual steam inflow corresponding to each training sample; Represent the first Isentropic enthalpy drop corresponding to each training sample; S404, calculating the real efficiency correction factor of each training sample, the first Real efficiency correction factor for individual training samples The calculation formula of (2) is as follows: Represent the first Theoretical isentropic efficiency of individual training samples; S405, taking the historical real-time operation parameters and the corresponding theoretical parameters of each training sample as input characteristics, taking the corresponding real efficiency correction factors as training target labels, and performing supervised training on the machine learning model to obtain a pre-trained machine learning correction model.
7. The method for modeling and back calculation of steam feed pump state as defined in claim 6, wherein in step S405, the objective of the supervision training is to minimize an objective function, and the expression of the objective function is: Wherein, the Represent the first Real efficiency correction factors of the individual training samples; representing machine learning correction model pair Predicted output values of the individual training samples; representing the total number of training samples; the method is characterized by comprising the steps of measuring the difference between a predicted value of a training sample and a training target label as a loss function; Is regularization term for controlling the first Sub-model Is prevented from overfitting.
8. A steam feed pump state modeling and steam inflow back calculation system based on thermodynamic mechanism and machine learning, which is characterized by being used for realizing the steam feed pump state modeling and steam inflow back calculation method as claimed in any one of claims 1-7, comprising: the thermodynamic mechanism model construction module is used for constructing a thermodynamic mechanism model of the water supply pump set based on a thermodynamic first law and an equipment design efficiency curve; The real-time operation parameter acquisition module is used for acquiring real-time operation parameters of the steam-driven water supply pump set, including water supply flow, water supply pump inlet pressure, water supply pump outlet pressure, water supply density, steam turbine inlet enthalpy and condenser pressure; the theoretical parameter calculation module is used for calculating theoretical parameters through a thermodynamic mechanism model based on real-time operation parameters; The efficiency correction factor generation module is used for constructing real-time operation parameters and theoretical parameters together into input feature vectors, inputting the input feature vectors into a pre-trained machine learning correction model, and outputting an efficiency correction factor representing the deviation degree of the current performance of the equipment relative to the design performance of the equipment; the theoretical isentropic efficiency calibration module is used for calibrating the theoretical isentropic efficiency by utilizing the efficiency correction factor to obtain the actual isentropic efficiency; the high-precision value calculation module is used for calculating and outputting a high-precision value of steam turbine inlet flow and steam turbine exhaust enthalpy through a thermodynamic mechanism model based on actual isentropic efficiency.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and operable on the processor, wherein the processor is configured to implement the steps of the method for modeling the state of a steam feed pump and back calculation of the inflow rate of steam as claimed in any one of claims 1 to 7 when the computer program is executed.
10. A storage medium having a computer program stored thereon, wherein the computer program when executed by a processor performs the steps of the method for modeling the state of a steam feed pump and back calculation of the inflow steam flow as claimed in any one of claims 1 to 7.

Description

Method, system, equipment and medium for modeling state of steam feed pump and back calculation of steam inflow flow based on thermodynamic mechanism and machine learning Technical Field The application relates to the technical field of intelligent management of a steam-driven water supply pump, in particular to a steam-driven water supply pump state modeling and steam inflow back calculation method, a steam-driven water supply pump state modeling and steam inflow back calculation system, steam inflow back calculation equipment and a steam inflow back calculation medium based on a thermodynamic mechanism and machine learning. Background In some large-scale generator sets, a steam feed pump is a key auxiliary equipment, and the running state of the steam feed pump is directly related to the safety and economy of the whole generator set. Along with the promotion of intelligent power plant construction, the digital twin technology is utilized to realize accurate state sensing, performance evaluation and predictive maintenance of key equipment, and the intelligent power plant has become an important development direction of industry. The digital twin body of the high-fidelity steam feed pump is established, and particularly, reliable back calculation of the steam inlet flow of the steam turbine which is difficult to directly and accurately measure is realized, and the digital twin body has important significance for optimizing operation and early warning faults. At present, modeling methods for a steam feed pump are mainly divided into two types, namely mechanism modeling based on a physical law and data driving modeling based on historical data. According to the pure mechanism modeling method, an equation set with definite physical meaning is constructed according to the first law of thermodynamics, hydrodynamics and other basic principles, and target parameters (such as shaft power, efficiency and steam inlet flow) are calculated by inputting measurable parameters (such as pressure, temperature and flow). The structure and parameters of the model are usually derived from equipment design data, the interpretation is good, and the pure data driving modeling method, such as various neural networks, support vector machines and the like, completely depends on a large amount of historical operation data, learns the complex mapping relation between input and output variables through training, does not need an explicit physical formula, and shows flexibility in processing nonlinear problems. However, the modeling methods of the two types of steam-driven water supply pumps have inherent limitations that the precision of a pure mechanism model is seriously dependent on the accuracy of internal efficiency parameters (such as isentropic efficiency of a steam turbine) of the pure mechanism model, the parameters usually adopt a fixed efficiency curve provided during equipment design, in actual operation, the actual efficiency of the pure mechanism model gradually deviates from the design curve due to factors such as equipment aging, scaling, abrasion and the like, so that the precision of calculation of the mechanism model based on the fixed efficiency curve is reduced after long-term operation and the current performance state of the equipment cannot be truly reflected, and a pure data driving model can learn the change of the efficiency from data, but is completely dependent on the data, lacks physical constraints, and can generate a prediction result against basic thermodynamic rules when the operation working condition exceeds the training data range, so that the generalization capability and the reliability of the model are poor. Disclosure of Invention Aiming at the technical problem that in the modeling method of the existing steam feed pump, the modeling method of a pure mechanism model cannot accurately reflect the equipment performance degradation in long-term operation due to the fact that a fixed equipment design efficiency curve is relied on, and the modeling method of a pure data driving model possibly generates prediction against a physical rule when working conditions exceed a training range due to lack of physical constraint, the application provides a steam feed pump state modeling and steam inlet flow back calculation method, a system, equipment and a medium based on thermodynamic mechanism and machine learning. In a first aspect, the application provides a method for modeling the state of a steam feed pump and back calculation of the inflow steam flow based on thermodynamic mechanism and machine learning, which comprises the following steps: s1, constructing a thermodynamic mechanism model of the water supply pump set based on a thermodynamic first law and an equipment design efficiency curve, wherein the thermodynamic mechanism model defines the relationship among the following parameters: A calculated relationship between feed water flow, feed water pump inlet pressure, feed water pump outlet pressure