Water feed pump system performance monitoring and fault diagnosis method and system based on digital twinning

Abstract

The invention discloses a performance monitoring and fault diagnosis method and system for a water feed pump system based on digital twinning, and the method comprises the steps: constructing a high-precision digital twinning model through a fusion mechanism drive simulation model and a data drive simulation model, and carrying out the thermal parameter simulation of the water feed pump system; and further establishing a water feed pump system fault diagnosis model by means of high-precision simulation parameters, and realizing early fault early warning and fault root judgment based on comprehensive deviation degree and contribution degree analysis of online measurement data. According to the method, the operation reliability of the thermal power plant feed pump system can be remarkably improved, and real-time guidance is provided for closed-loop control and operation and maintenance management of the feed pump system.

Description

Technical Field

[0001] The present invention belongs to the technical field of operation optimization of thermal power systems, and particularly relates to a method and system for performance monitoring and fault diagnosis of a feed water pump system based on digital twin. Background Technique

[0002] Since the 21st century, the world's energy structure has been continuously developing and transforming towards the direction of being clean, low-carbon, efficient, and diversified. However, due to the strong time-varying characteristics of renewable energy sources such as wind energy and solar energy, it has caused difficulties in the consumption of renewable energy power generation in China, and the problems of wind curtailment and photovoltaic curtailment are serious. Improving the operation flexibility of conventional thermal power systems to provide consumption services for renewable energy power generation is an important technical direction. Thermal power will transform from the main energy source to the basic energy source, and the load change range and load change frequency of thermal power units will gradually increase. Therefore, the thermal power system will be in transient operating conditions with frequent load changes for a long time.

[0003] As an important part of thermal power units, subcritical units will undertake more and more important peak shaving tasks, and the requirements for their rapid load change ability are getting higher and higher. As the core subsystem for maintaining the steam power cycle, the operation reliability and stability of the feed water pump system are becoming increasingly important. The feed water pump systems adopted by a large number of in-service coal-fired generating units generally include three core devices: a booster pump, a feed water pump, and a feed water pump steam turbine. Therefore, the operation stability and safety of these three core devices are the core content of the operation reliability of the feed water pump system.

[0004] In order to ensure the operation safety of the feed water pump system during the operation of the unit, it is necessary to establish a reliable simulation model of the feed water pump system to realize its online performance monitoring, and at the same time establish a fault diagnosis model of the feed water pump system to diagnose the key parameters in the feed water pump system, so as to realize the early warning of faults and the determination of the root causes of faults, and further protect each device of the feed water pump system and give operation and maintenance suggestions. Due to the complexity of feed water pump system modeling and the diversity of faults, the existing technology has insufficient real-time simulation accuracy for the feed water pump system, and there are defects in the early warning of system faults and the determination of the root causes of faults. Summary of the Invention

[0005] In order to solve the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a method and system for performance monitoring and fault diagnosis of a feed water pump system based on digital twin. By means of digital twin technology, with the booster pump, the feed water pump, and the feed water pump steam turbine as the modeling objects, integrating the mechanism-driven simulation model of the feed water pump system and the data-driven simulation model of the feed water pump system, realizing the real-time monitoring of various key thermal parameters, and further establishing a fault diagnosis model of the feed water pump system to realize the early warning of system faults and the determination of the root causes of faults. The present invention can realize the real-time performance monitoring of the feed water pump system and greatly improve the reliability of the feed water pump system during the operation of the unit.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A feedwater pump system performance monitoring and fault diagnosis method based on digital twin, the feedwater pump system includes a pre-pump, a feedwater pump and a feedwater pump steam turbine, the feedwater pump system performance monitoring and fault diagnosis method based on digital twin includes the following steps:

[0008] Step 1: Data Collection

[0009] Real-time acquisition of historical operating data of the feedwater pump system from the distributed control system DCS of the thermal power plant, including feedwater mass flow m fw , the outlet pressure of the pre-pump is the inlet pressure of the water pump p obp , water pump outlet pressure p fw , feedwater pump speed N, feedwater pump turbine steam consumption m spt and feedwater pump turbine exhaust pressure p ept ;

[0010] Step 2: Data preprocessing

[0011] Preprocess the historical operation data obtained in step 1, remove outliers and noise data, and reduce the dimension of the high-dimensional data set through principal component analysis (PCA), retain the principal components whose cumulative contribution rate exceeds the preset threshold, and generate a reduced-dimensional data set;

[0012] Step 3: Construction of water supply pump system mechanism driven simulation model

[0013] Based on the physical mechanism of each device in the feedwater pump system, a mechanism-driven simulation model of the feedwater pump system is constructed. The mechanism-driven simulation model includes the following calculation methods for the key thermal parameters of the feedwater pump system. The specific parameters include the boiler feed water mass flow rate m flowing through the feedwater pump fw , water pump outlet pressure p fw , feed water pump speed N, feed water pump efficiency η fp , water supply pump power W p , Steam consumption of feedwater pump turbine m spt , Feedwater pump turbine exhaust pressure p ept and feedwater pump inlet pressure p obp ;

[0014] Among them, the boiler feed water mass flow rate flowing through the feed water pump is m fw The calculation formula is:

[0015] m fw =k1W e +k2

[0016] Where: W eis the output power of the unit, MW; the coefficients k1 and k2 are determined by fitting the historical operation data of the feed pump system;

[0017] The outlet pressure p of the feed pump fw The calculation formula is:

[0018] p fw = k3W e + k4

[0019] In the formula: the coefficients k3 and k4 are determined by fitting the historical operation data of the feed pump system;

[0020] The calculation formula for the rotational speed N of the feed pump is:

[0021]

[0022] In the formula: H fp is the head of the feed pump, MPa; k5 is the correction coefficient for the performance degradation of the feed pump; c1, c2, and c3 are all fitting coefficients of the performance characteristic curve provided by the feed pump manufacturer;

[0023] Based on the performance curve of the feed pump and the similarity law of pumps, combined with the characteristics of the feed pump tapping, the efficiency η of the feed pump is obtained fp The calculation formula:

[0024] η fp = k ie η fp1 +(1 - k ie )η fp2

[0025] In the formula: η fp1 and η fp2 are the efficiencies of the front and rear parts of the feed pump tapping, respectively, %; k ie is the feed pump tapping coefficient;

[0026] Based on the energy balance of the feed pump system, the power W of the feed pump is obtained p The calculation formula:

[0027] W p = W p1 + W p2

[0028] In the formula: W p1 and W p2 are the powers of the front and rear parts of the feed pump tapping, respectively, kW;

[0029] The steam consumption m of the feed pump turbine spt The calculation formula is:

[0030]

[0031] where: h spt and h ept are the steam enthalpy values at the inlet and outlet of the boiler feed pump turbine, respectively; η fptm is the mechanical efficiency of the boiler feed pump turbine, %;

[0032] According to the relationship between the flow rate and pressure loss in the pipeline, the calculation formula for the exhaust pressure p ept of the boiler feed pump turbine is obtained:

[0033] p ept = k6m spt + k7 + p c

[0034] where: p ept is the exhaust pressure of the boiler feed pump turbine, kPa; the coefficients k6 and k7 are both determined by fitting the historical data of the boiler feed pump system; p c is the condenser pressure, kPa;

[0035] The calculation formula for the outlet pressure of the booster pump, i.e., the inlet pressure of the boiler feed pump, is:

[0036] p obp = p de + H bp + H g

[0037] where: p obp is the outlet pressure of the booster pump, i.e., the inlet pressure of the boiler feed pump, MPa; p de is the operating pressure of the deaerator, MPa; H bp is the head of the booster pump, MPa; H g is the pressure difference caused by the elevation difference between the deaerator and the boiler feed pump, MPa;

[0038] All the above calculation formulas for the thermal parameters together constitute the mechanism-driven simulation model of the boiler feed pump system;

[0039] Step 4: Establishment of the data-driven simulation model of the boiler feed pump system

[0040] Using the digital twin method to establish a data-driven simulation model of the boiler feed pump system, this model includes a BP neural network training method for the rotational speed N of the boiler feed pump, the steam consumption m spt of the boiler feed pump turbine and the exhaust pressure m spt of the boiler feed pump turbine. The specific method for model establishment is: using a BP neural network to train the data set after PCA dimensionality reduction. The input parameters include the unit power, the flow rate and pressure of the boiler feed pump tap, the measured value of the rotational speed of the boiler feed pump, the steam enthalpy values at the inlet and outlet of the boiler feed pump turbine, and the condenser pressure. The output parameters include the rotational speed of the boiler feed pump, the steam consumption of the boiler feed pump turbine, and the exhaust pressure of the boiler feed pump turbine;

[0041] Among them, the input parameters of the feed pump speed BP neural network are selected as three parameters: unit power, extraction flow rate, and extraction pressure; the input parameters of the steam consumption of the feed pump turbine BP neural network are selected as six parameters: unit power, extraction flow rate, extraction pressure, measured value of the feed pump speed, and enthalpy values at the inlet and outlet of the feed pump turbine; the input parameters of the exhaust pressure of the feed pump turbine BP neural network are selected as seven parameters: unit power, extraction flow rate, extraction pressure, measured value of the feed pump speed, enthalpy values at the inlet and outlet of the feed pump turbine, and condenser pressure;

[0042] Step Five: Simulation Model Fusion

[0043] Fuse the output parameters of the mechanism-driven simulation model and the data-driven simulation model of the feed pump system to generate high-precision digital twin parameters, including feed water mass flow rate m fw , feed pump outlet pressure p fw , feed pump speed N, feed pump efficiency η fp , feed pump power W p , steam consumption of the feed pump turbine m spt , exhaust pressure of the feed pump turbine p ept and feed pump inlet pressure p obp ;

[0044] By comparing the mean absolute error MAE, root mean square error RMSE, and coefficient of determination R 2 of the output parameters of the mechanism-driven simulation model and the data-driven simulation model of the feed pump system, select the model with higher comprehensive accuracy for simulation for the feed pump speed N, steam consumption of the feed pump turbine m spt and exhaust pressure of the feed pump turbine p ept ;

[0045] Step Six: Establishment of the Feed Pump System Fault Diagnosis Model

[0046] Based on the output parameters of the mechanism-driven simulation model and the data-driven simulation model of the feed pump system, establish a feed pump system fault diagnosis model. This fault diagnosis model includes the calculation methods of comprehensive deviation degree, average deviation degree of the validation data set, deviation degree alarm threshold, and contribution degree of each parameter deviation to the comprehensive deviation degree; the following is the specific establishment method of the feed pump system fault diagnosis model: First, calculate the comprehensive deviation degree l of the measured value and the model simulation value of the feed pump system parameters:

[0047]

[0048] In the formula: w i is the deviation contribution weight artificially assigned to each parameter according to the importance degree and abnormal frequency of the parameters in the feed pump system; r iis the relative error between the measured data and the simulation data of each parameter, %; a sim and a mea are the simulation value and the measured value of the parameter respectively;

[0049] Subsequently, calculate the average deviation degree l of the verification data set ave :

[0050]

[0051] In the formula: n vd is the amount of data contained in the verification data set;

[0052] Calculate the alarm threshold of the fault diagnosis model based on the average deviation degree of the verification data set and the set alarm threshold coefficient:

[0053]

[0054] In the formula: k al is the alarm threshold coefficient of the fault diagnosis model;

[0055] Finally, perform fault early warning and fault root cause determination. When the real-time comprehensive deviation degree is lower than the alarm threshold, trigger a fault alarm and start fault root cause determination; the contribution degree c of each parameter deviation to the comprehensive deviation degree i is calculated as:

[0056]

[0057] By monitoring the magnitude of the contribution degree of each parameter deviation, the parameter where the fault occurs can be located;

[0058] Step 7: Closed-loop control feedback

[0059] Feed back the fault diagnosis result to the distributed control system DCS of the thermal power plant, actively regulate the control system corresponding to the abnormal parameter, and then realize the fault diagnosis and closed-loop control of the feed water pump system; at the same time, transmit the fault information to the unit operation management personnel, give the fault cause analysis result, and provide the recommended equipment operation and maintenance plan for the unit operation management personnel.

[0060] Preferably, the specific method of data preprocessing in the second step includes:

[0061] (1) Analyze the measurement data set, screen out the missing values, abnormal values and error values in the measurement data, fill in the data missing values, correct the data abnormal values, and eliminate the data error values;

[0062] (2) The sliding window method is used to detect outliers in the data, and the missing values are filled by linear interpolation or the mean of adjacent data; the time series data is smoothed, and the Savitzky-Golay filter is used to eliminate high-frequency noise.

[0063] Preferably, the calculation of the feed pump efficiency and the feed pump power in the mechanism modeling in step three further includes:

[0064] Calculate the feed water flow rates of the two parts before and after the feed pump tap according to the feed pump tap flow rate, and at the same time, combine the feed pump speed to calculate the respective efficiencies of the two parts before and after the feed pump:

[0065] m fw1 = m fw + m ie

[0066]

[0067]

[0068] In the formula: m fw1 and m fw are the feed water flow rates before and after the feed pump tap respectively, t / h; m ie is the feed pump tap flow rate, t / h; c4, c5 and c6 are all fitting coefficients of the performance characteristic curve provided by the feed pump manufacturer;

[0069] The total power of the feed pump is the sum of the powers of the two parts before and after the feed pump tap, and the powers of the parts before and after the feed pump tap are calculated as follows:

[0070]

[0071] In the formula: p ie is the feed pump tap pressure, MPa; ρ is the average density of the working medium in the feed pump, kg / m 3 .

[0072] Preferably, the BP neural network structure in the data-driven simulation model of the feed pump system in step four is:

[0073] The BP neural network structure includes an input layer, a hidden layer and an output layer, among which the hidden layer contains 10 neurons, and the data set is divided into a training set, a validation set and a test set according to the ratio of 70%, 15% and 15%.

[0074] Preferably, the specific training method of the BP neural network for data-driven modeling in step four further includes:

[0075] The Levenberg-Marquardt algorithm is used to optimize the network weights, and the activation function is the hyperbolic tangent function;

[0076] At the same time, early stopping is used to prevent overfitting, and the training is terminated when the validation set error does not decrease for 10 consecutive times.

[0077] Preferably, the weight assignment rule of each parameter deviation to the comprehensive deviation degree in step six is as follows:

[0078] Considering the abnormal frequencies of each parameter in the actual feed pump system, higher weights should be set for parameters with high abnormal frequencies. According to the actual number of abnormalities of each parameter, the weights of each parameter are calculated using the following formula:

[0079]

[0080] In the formula: k weight is the weight assignment coefficient; n f,i is the number of abnormalities of a certain parameter within a period of historical time; n f,sum is the sum of the number of abnormalities of all parameters within a period of historical time.

[0081] Preferably, the determination of the root cause of the fault in step six further includes:

[0082] When the comprehensive deviation degree is lower than the alarm threshold, the fault diagnosis system will issue an alarm. At the same time, it will start to track the abnormal thermal parameters. The determination of the root cause of the fault is mainly carried out through two dimensions, namely the absolute value of the contribution degree of each parameter deviation to the comprehensive deviation degree and the growth rate of the contribution degree. When the contribution degree of the deviation of a certain parameter to the comprehensive deviation degree increases rapidly and exceeds 0.95 when the alarm is issued, it can be determined that the parameter is an abnormal parameter.

[0083] A system for implementing the above method includes:

[0084] A data acquisition module, used to obtain operation data from the DCS system in real time;

[0085] A digital twin modeling module, integrating a mechanism-driven simulation model of the feed pump system and a data-driven simulation model of the feed pump system;

[0086] A fault diagnosis module, based on the feed pump system fault diagnosis model, calculates the comprehensive deviation degree and the contribution degree;

[0087] A control execution module, which feeds back the diagnosis result to the DCS and executes a closed-loop control strategy.

[0088] Compared with the prior art, the present invention has the following advantages:

[0089] The beneficial effects of the present invention are as follows: The present invention provides a method and system for performance monitoring and fault diagnosis of a feed pump system based on digital twin. First, relevant thermal parameters of the feed pump system in the distributed control system (DCS) of the power plant are collected, and the data is further preprocessed using the sliding window method, linear interpolation method, Savitzky-Golay filter, and principal component analysis (PCA) method. Then, taking the booster pump, feed pump, and feed pump steam turbine as the modeling objects, the physical mechanisms of each device in the feed pump system are analyzed, and a mechanism-driven simulation model of the feed pump system is established. At the same time, with the help of digital twin technology and based on the BP neural network, a data-driven simulation model of the feed pump system is established. By fusing the above two simulation models, high-precision simulation of the key thermal parameters of the feed pump system is achieved, and real-time monitoring of various key thermal parameters is realized. Based on the high-precision digital twin parameters, a fault diagnosis model of the feed pump system is further established, which can realize fault early warning and determination of the root cause of faults for the key parameters in the feed pump system. The present invention establishes a high-precision simulation model, which can realize high-precision online real-time monitoring of the performance of each thermal parameter of the feed pump system, and greatly improve the reliability of the feed pump system during the operation of the unit through early warning and determination of the source of system faults, providing real-time guidance for the closed-loop control and operation and maintenance management of the feed pump system.

[0090] Advantages of the present invention

[0091] 1) The present invention provides a method and system for performance monitoring and fault diagnosis of a feed pump system based on digital twin, and its advantages are reflected in the establishment method and effect of the feed pump system simulation model: The feed pump system of the present invention includes three devices: a booster pump, a feed pump, and a feed pump steam turbine. First, a mechanism-driven simulation model of the feed pump system is established by analyzing the physical mechanisms in each device, and simulation data of each key thermal parameter is obtained from the perspective of mechanism. At the same time, using digital twin technology, a data-driven simulation model of the feed pump system is established based on the BP neural network, and simulation data of 3 key parameters is obtained from the perspective of data. Further, the two models are fused to finally obtain high-precision simulation results of each thermal parameter. It realizes online real-time monitoring of the performance during the operation of the feed pump system and provides a data basis for the establishment of the subsequent feed pump system fault diagnosis model;

[0092] 2) The present invention provides a method and system for performance monitoring and fault diagnosis of a feed pump system based on digital twin. Its advantages are reflected in the establishment method and effect of the fault diagnosis model of the feed pump system: By means of the comprehensive deviation degree parameter between the measured value and the simulation value of the feed pump system parameters, the average deviation degree of the verification data set is calculated, and the alarm threshold of the fault diagnosis model is obtained by combining the alarm threshold coefficient. When the real-time comprehensive deviation degree is lower than the alarm threshold, a fault alarm is triggered, and the determination of the fault root cause begins with the contribution degree of each parameter deviation to the comprehensive deviation degree. Finally, the fault diagnosis result is fed back to the DCS to regulate the fault parameters, and at the same time, the fault information is transmitted to the unit operation management personnel, and the fault cause analysis and operation and maintenance plan are given. With the fault diagnosis model of the feed pump system established by the present invention, fault early warning and fault root cause determination can be realized during the operation of the feed pump system, thereby greatly improving the reliability of the feed pump system during the operation of the unit, and providing real-time guidance for the closed-loop control and operation and maintenance management of the feed pump system. Description of the Drawings

[0093] Figure 1 is the deviation degree of the diagnosis data set of the feed pump system.

[0094] Figure 2 is the contribution degree of the deviation degree of the diagnosis data set of the feed pump system. Detailed Embodiments

[0095] In order to make the objectives, technical solutions and advantages of the present invention clearer, the following further details the method of the present invention in combination with specific implementation cases. It should be understood that the specific implementation cases described herein are only used to explain the present invention and are not used to limit the present invention.

[0096] The present invention discloses a method and system for performance monitoring and fault diagnosis of a feed pump system based on digital twin. By integrating a mechanism-driven simulation model and a data-driven simulation model, a high-precision digital twin model is constructed to simulate the thermal parameters of the feed pump system. Further, a fault diagnosis model of the feed pump system is established based on the high-precision simulation parameters, and early fault warning and fault root cause determination are realized based on the comprehensive deviation degree and contribution degree analysis of the on-line measurement data. This method can significantly improve the operation reliability of the feed pump system in thermal power plants and provide real-time guidance for the closed-loop control and operation and maintenance management of the feed pump system. The feed pump system includes a booster pump, a feed pump and a feed pump steam turbine, and is characterized in that the method for performance monitoring and fault diagnosis of the feed pump system based on digital twin includes the following steps:

[0097] Step 1: Data acquisition

[0098] Real-time obtain the historical operation data of the feed pump system from the distributed control system (DCS) of the thermal power plant, including the feed water mass flow rate m fw, the outlet pressure of the pre-pump is the inlet pressure of the water pump p obp , water pump outlet pressure p fw , feedwater pump speed N, feedwater pump turbine steam consumption m spt and feedwater pump turbine exhaust pressure p ept ;

[0099] Step 2: Data preprocessing

[0100] Preprocess the historical operation data obtained in step 1, remove outliers and noise data, and reduce the dimension of the high-dimensional data set through principal component analysis (PCA), retain the principal components whose cumulative contribution rate exceeds the preset threshold, and generate a reduced-dimensional data set;

[0101] Step 3: Construction of water supply pump system mechanism driven simulation model

[0102] Based on the physical mechanism of each device in the feedwater pump system, a mechanism-driven simulation model of the feedwater pump system is constructed. The mechanism-driven simulation model includes the following calculation methods for the key thermal parameters of the feedwater pump system. The specific parameters include the boiler feed water mass flow rate m flowing through the feedwater pump fw , water pump outlet pressure p fw , feed water pump speed N, feed water pump efficiency η fp , water supply pump power W p , Steam consumption of feedwater pump turbine m spt , Feedwater pump turbine exhaust pressure p ept and feedwater pump inlet pressure p obp ;

[0103] Among them, the boiler feed water mass flow rate flowing through the feed water pump is m fw The calculation formula is:

[0104] m fw =k1W e +k2

[0105] Where: W e is the unit output power, MW; coefficients k1 and k2 are determined by fitting the historical operating data of the feedwater pump system;

[0106] Feedwater pump outlet pressure p fw The calculation formula is:

[0107] p fw =k3W e +k4

[0108] Where: coefficients k3 and k4 are determined by fitting the historical operating data of the water supply pump system;

[0109] The calculation formula for the water supply pump speed N is:

[0110]

[0111] Where: H fp is the head of the feed water pump, MPa; k5 is the correction coefficient for the performance degradation of the feed water pump; c1, c2, and c3 are all fitting coefficients of the performance characteristic curve provided by the feed water pump manufacturer;

[0112] Based on the performance curve of the feed water pump and the similarity law of pumps, combined with the characteristics of the feed water pump tapping, the efficiency η of the feed water pump is obtained fp Calculation formula:

[0113] η fp = k ie η fp1 +(1 - k ie )η fp2

[0114] Where: η fp1 and η fp2 are the efficiencies of the front and rear parts of the feed water pump tapping respectively, %; k ie is the feed water pump tapping coefficient;

[0115] Based on the energy balance of the feed water pump system, the calculation formula for the power W of the feed water pump is obtained p Calculation formula:

[0116] W p = W p1 + W p2

[0117] Where: W p1 and W p2 are the powers of the front and rear parts of the feed water pump tapping respectively, kW;

[0118] The calculation formula for the steam consumption m of the feed water pump turbine is: spt Calculation formula:

[0119]

[0120] Where: h spt and h ept are the steam enthalpy values at the inlet and outlet of the feed water pump turbine respectively; η fptm is the mechanical efficiency of the feed water pump turbine, %;

[0121] According to the relationship between the flow rate and pressure loss in the pipeline, the calculation formula for the exhaust pressure p of the feed water pump turbine is obtained ept Calculation formula:

[0122] p ept = k6m spt + k7 + p c

[0123] Where: pept is the exhaust pressure of the feed water pump steam turbine, kPa; the coefficients k6 and k7 are both determined by fitting the historical data of the feed water pump system; p c is the condenser pressure, kPa;

[0124] The calculation formula for the outlet pressure of the booster pump, i.e., the inlet pressure of the feed water pump, is:

[0125] p obp = p de + H bp + H g

[0126] In the formula: p obp is the outlet pressure of the booster pump, i.e., the inlet pressure of the feed water pump, MPa; p de is the operating pressure of the deaerator, MPa; H bp is the head of the booster pump, MPa; H g is the pressure difference caused by the elevation difference between the deaerator and the feed water pump arrangement, MPa;

[0127] All the calculation formulas of the above thermal parameters together constitute the mechanism-driven simulation model of the feed water pump system;

[0128] Step Four: Establishment of the data-driven simulation model of the feed water pump system

[0129] Use the digital twin method to establish a data-driven simulation model of the feed water pump system. This model includes the BP neural network training method for the feed water pump speed N, the steam consumption m of the feed water pump steam turbine spt and the exhaust pressure m of the feed water pump steam turbine spt The specific method for model establishment is: Use the BP neural network to train the dataset after PCA dimensionality reduction. The input parameters include the unit power, the booster pump extraction flow rate and pressure, the measured value of the feed water pump speed, the inlet and outlet steam enthalpy values of the feed water pump steam turbine, and the condenser pressure. The output parameters include the feed water pump speed, the steam consumption of the feed water pump steam turbine, and the exhaust pressure of the feed water pump steam turbine;

[0130] Among them, the input parameters of the BP neural network for the feed water pump speed select three parameters: unit power, extraction flow rate, and extraction pressure; the input parameters of the BP neural network for the steam consumption of the feed water pump steam turbine select six parameters: unit power, extraction flow rate, extraction pressure, measured value of the feed water pump speed, and inlet and outlet enthalpy values of the feed water pump steam turbine; the input parameters of the BP neural network for the exhaust pressure of the feed water pump steam turbine select seven parameters: unit power, extraction flow rate, extraction pressure, measured value of the feed water pump speed, inlet and outlet enthalpy values of the feed water pump steam turbine, and condenser pressure;

[0131] Step Five: Fusion of the simulation models

[0132] Fuse the output parameters of the mechanism-driven simulation model and the data-driven simulation model of the feed pump system to generate high-precision digital twin parameters, including the feed water mass flow rate m fw , the outlet pressure p of the feed pump fw , the rotational speed N of the feed pump, the efficiency η of the feed pump fp , the power W of the feed pump p , the steam consumption m of the feed pump turbine spt , the exhaust pressure p of the feed pump turbine ept and the inlet pressure p of the feed pump obp ;

[0133] By comparing the mean absolute error MAE, root mean square error RMSE and coefficient of determination R of the output parameters of the mechanism-driven simulation model and the data-driven simulation model of the feed pump system 2 , select the model with higher comprehensive accuracy for simulation for the rotational speed N of the feed pump, the steam consumption m of the feed pump turbine spt and the exhaust pressure p of the feed pump turbine ept ;

[0134] Step 6: Establish the fault diagnosis model of the feed pump system

[0135] Based on the output parameters of the mechanism-driven simulation model and the data-driven simulation model of the feed pump system, establish a fault diagnosis model for the feed pump system. This fault diagnosis model includes the calculation methods of the comprehensive deviation degree, the average deviation degree of the validation data set, the deviation degree alarm threshold, and the contribution degree of each parameter deviation to the comprehensive deviation degree; the following is the specific method for establishing the fault diagnosis model of the feed pump system: First, calculate the comprehensive deviation degree l between the measured value and the model simulation value of the feed pump system parameters:

[0136]

[0137] In the formula: w i is the deviation contribution weight artificially assigned to each parameter according to the importance degree and abnormal frequency of the parameters in the feed pump system; r i is the relative error between the measured data and the simulation data of each parameter, %; a sim and a mea are the simulation value and the measured value of the parameter respectively;

[0138] Subsequently, calculate the average deviation degree l of the validation data set ave :

[0139]

[0140] In the formula: n vd is the amount of data contained in the validation data set;

[0141] Calculate the alarm threshold of the fault diagnosis model based on the average deviation degree of the verification data set and the set alarm threshold coefficient:

[0142]

[0143] In the formula: k al is the alarm threshold coefficient of the fault diagnosis model;

[0144] Finally, perform fault early warning and fault root cause determination. When the real-time comprehensive deviation degree is lower than the alarm threshold, trigger a fault alarm and start fault root cause determination; the contribution degree c of each parameter deviation to the comprehensive deviation degree i is calculated as:

[0145]

[0146] By monitoring the magnitude of the contribution degree of each parameter deviation, the parameter where the fault occurs can be located;

[0147] Step Seven: Closed-loop control feedback

[0148] Feed back the fault diagnosis result to the distributed control system DCS of the thermal power plant, actively regulate the control system corresponding to the abnormal parameter, and then realize the fault diagnosis and closed-loop control of the feed water pump system; at the same time, transmit the fault information to the unit operation management personnel, give the analysis result of the fault cause, and provide the recommended equipment operation and maintenance plan for the unit operation management personnel.

[0149] Preferably, the specific method of data preprocessing in the second step includes:

[0150] (1) Analyze the measurement data set, screen out the missing values, outliers and error values in the measurement data, fill in the missing data values, correct the outlier data values, and eliminate the error data values;

[0151] (2) Use the sliding window method to detect the outliers in the data, and fill in the missing values by linear interpolation or the mean value of adjacent data; smooth the time series data, and use the Savitzky-Golay filter to eliminate high-frequency noise.

[0152] Through data preprocessing, the availability of the original measurement data can be greatly improved, and then the accuracy of subsequent modeling and simulation and fault diagnosis can be improved.

[0153] Preferably, the calculation of the feed water pump efficiency and the feed water pump power in the mechanism modeling in the third step further includes:

[0154] Calculate the feed water flow rates of the two parts before and after the feed water pump tap according to the feed water pump tap flow rate, and at the same time combine the feed water pump speed to calculate the respective efficiencies of the two parts of the feed water pump:

[0155] mfw1 = m fw + m ie

[0156]

[0157] where: m fw1 and m fw are the feed water flow rates before and after the extraction tap of the feed water pump, respectively, in t / h; m ie is the extraction flow rate of the feed water pump, in t / h; c4, c5 and c6 are all fitting coefficients of the performance characteristic curve provided by the feed water pump manufacturer;

[0158] The total power of the feed water pump is the sum of the powers of the two parts before and after the extraction tap of the feed water pump. The power calculations of the parts before and after the extraction tap of the feed water pump are as follows:

[0159]

[0160] where: p ie is the extraction pressure of the feed water pump, in MPa; ρ is the average density of the working fluid in the feed water pump, in kg / m 3 .

[0161] By separately processing the feed water pump with an intermediate extraction tap into two parts before and after, the accuracy of the feed water pump modeling can be improved, providing a basis for subsequent fault diagnosis.

[0162] Preferably, the BP neural network structure in the data-driven simulation model of the feed water pump system in step four is:

[0163] The BP neural network structure includes an input layer, a hidden layer and an output layer. Among them, the hidden layer contains 10 neurons, and the data set is divided into a training set, a validation set and a test set according to the ratio of 70%, 15% and 15%.

[0164] Preferably, the specific training method of the BP neural network for data-driven modeling in step four further includes:

[0165] The Levenberg-Marquardt algorithm is used to optimize the network weights, where the activation function is the hyperbolic tangent function;

[0166] At the same time, overfitting is prevented by the early stopping method. When the validation set error does not decrease for 10 consecutive times, the training is terminated.

[0167] With the help of a scientific and reasonable BP neural network structure division and training method, the accuracy of the data-driven simulation model of the feed water pump system can be improved.

[0168] Preferably, the weight distribution rule of each parameter deviation to the comprehensive deviation degree in step six is:

[0169] Considering the abnormal frequencies of various parameters of the actual feed water pump system, higher weights should be set for parameters with high abnormal frequencies. According to the actual number of abnormalities of each parameter, the weight of each parameter is calculated using the following formula:

[0170]

[0171] In the formula: k weight is the weight distribution coefficient; n f,i is the number of abnormalities of a certain parameter within a period of historical time; n f,sum is the sum of the number of abnormalities of all parameters within a period of historical time.

[0172] Determining the weights of each thermal parameter of the feed water pump according to the abnormal frequency can focus on the parameters prone to failure and improve the accuracy of fault warning.

[0173] Preferably, the determination of the root cause of the fault in step six further includes:

[0174] When the comprehensive deviation degree is lower than the alarm threshold, the fault diagnosis system will issue an alarm and start to track the thermal parameters with abnormalities. The root cause of the fault is mainly determined through two dimensions, namely the absolute value of the contribution degree of the deviation of each parameter to the comprehensive deviation degree and the growth rate of the contribution degree. When the contribution degree of the deviation of a certain parameter to the comprehensive deviation degree increases rapidly and exceeds 0.95 when the alarm is issued, it can be determined that the parameter is an abnormal parameter.

[0175] Preferably, a system for implementing the above method is characterized by including:

[0176] A data acquisition module for obtaining operation data from the DCS system in real time;

[0177] A digital twin modeling module integrating a mechanism-driven simulation model of the feed water pump system and a data-driven simulation model of the feed water pump system;

[0178] A fault diagnosis module for calculating the comprehensive deviation degree and the contribution degree based on the fault diagnosis model of the feed water pump system;

[0179] A control execution module for feeding back the diagnosis result to the DCS and executing a closed-loop control strategy.

[0180] The method for monitoring the performance and diagnosing faults of the feed water pump system is verified by simulation. Figure 1 is the curve of the change in the deviation degree of a diagnostic data set containing the fault data of the feed water pump system. It can be seen from the figure that the comprehensive deviation degree starts to decline rapidly from 1751 time steps, and the deviation degree is lower than the alarm threshold at 2660 time steps. The system issues a fault warning and starts to determine the root cause of the fault.

[0181] Figure 2 It is the change curve of the contribution degree of each thermal parameter constituting the fault diagnosis model to the deviation degree. It can be found from the figure that before 2000 time steps, the contribution degree of the exhaust steam pressure parameter of the boiler feed pump steam turbine to the deviation degree suddenly soars from below 0.1 to nearly 1, indicating that the root cause of the fault warning issued at 2660 time steps is the abnormal exhaust steam pressure of the boiler feed pump steam turbine.

Claims

1. A performance monitoring and fault diagnosis method for a feed water pump system based on digital twin. The feed water pump system includes a booster pump, a feed water pump, and a feed water pump steam turbine, characterized in that, The water supply pump system performance monitoring and fault diagnosis method based on digital twins includes the following steps: Step 1: Data Collection Real-time obtain the historical operation data of the feed water pump system from the distributed control system (DCS) of the thermal power plant, including the feed water mass flow rate m fw , the outlet pressure of the booster pump, i.e., the inlet pressure of the feed water pump p obp , the outlet pressure of the feed water pump p fw , the rotational speed N of the feed water pump, the steam consumption m of the feed water pump turbine spt , and the exhaust pressure p of the feed water pump turbine ept ; Step 2: Data preprocessing Preprocess the historical operation data obtained in step 1, remove outliers and noise data, and reduce the dimension of the high-dimensional data set through principal component analysis (PCA), retain the principal components whose cumulative contribution rate exceeds the preset threshold, and generate a reduced-dimensional data set; Step 3: Construction of water supply pump system mechanism driven simulation model Based on the physical mechanisms of the various equipment in the feed pump system, a mechanism-driven simulation model of the feed pump system is constructed. The mechanism-driven simulation model includes calculation methods for the following key thermal parameters of the feed pump system. The specific parameters include the mass flow rate m of the boiler feed water flowing through the feed pump fw , the outlet pressure p of the feed pump fw , the rotational speed N of the feed pump, the efficiency η of the feed pump fp , the power W of the feed pump p , the steam consumption m of the feed pump turbine spt , the exhaust pressure p of the feed pump turbine ept , and the inlet pressure p of the feed pump obp ; wherein, the mass flow rate m of the boiler feed water flowing through the feed water pump fw is calculated by the following formula: m fw = k1W e + k2 Where: W e is the output power of the unit, MW; the coefficients k1 and k2 are determined by fitting the historical operation data of the feed pump system; The outlet pressure p of the feed water pump fw is calculated by the following formula: p fw = k3W e + k4 Where: coefficients k3 and k4 are determined by fitting the historical operating data of the water supply pump system; The calculation formula for the water supply pump speed N is: Where: H fp is the head of the feed pump, MPa; k5 is the correction coefficient for the performance degradation of the feed pump; c1, c2, and c3 are all fitting coefficients of the performance characteristic curve provided by the feed pump manufacturer; Based on the performance curve of the feed water pump and the similarity law of pumps, combined with the characteristics of the feed water pump tapping, the calculation formula for the feed water pump efficiency η fp is obtained as follows: η fp = k ie η fp1 +(1 - k ie )η fp2 where: η fp1 and η fp2 are the efficiencies of the front and rear parts of the feed pump before the tapping, respectively, %; k ie is the tapping coefficient of the feed pump; Based on the energy balance of the feed water pump system, the calculation formula for the power W of the feed water pump is obtained p as follows: W p = W p1 + W p2 Where: W p1 and W p2 are the powers of the front part and the rear part of the feed water pump tapping, respectively, in kW; Steam consumption m of the boiler feed pump steam turbine spt The calculation formula is as follows: Where: h spt and h ept are the steam enthalpy values at the inlet and outlet of the boiler feed pump turbine respectively; η fptm is the mechanical efficiency of the boiler feed pump turbine, %; According to the relationship between the flow rate and pressure loss in the pipeline, the calculation formula for the exhaust steam pressure p of the boiler feed pump steam turbine is obtained: ept ​ p ept = k6m spt + k7 + p c Where: p ept is the exhaust steam pressure of the boiler feed pump turbine, kPa; coefficients k6 and k7 are both determined by fitting the historical data of the boiler feed pump system; p c is the condenser pressure, kPa; The calculation formula for the outlet pressure of the fore pump, i.e. the inlet pressure of the water pump, is: p obp = p de + H bp + H g Where: p obp is the outlet pressure of the booster pump, i.e., the inlet pressure of the feed water pump, in MPa; p de is the operating pressure of the deaerator, in MPa; H bp is the head of the booster pump, in MPa; H g is the pressure difference caused by the elevation difference between the deaerator and the feed water pump arrangement, in MPa; All the calculation formulas of the above thermal parameters together constitute the feedwater pump system mechanism-driven simulation model; Step 4: Build a data-driven simulation model for the water pump system A data-driven simulation model of the feed pump system is established using the digital twin method. The model includes the rotational speed N of the feed pump and the steam consumption m of the feed pump turbine spt and the exhaust steam pressure m of the feed pump turbine spt The specific method for model construction is as follows: The BP neural network is used to train the dataset after PCA dimensionality reduction. The input parameters include unit power, feed pump tapping flow and pressure, measured value of feed pump rotational speed, inlet and outlet steam enthalpies of the feed pump turbine, and condenser pressure. The output parameters include feed pump rotational speed, steam consumption of the feed pump turbine, and exhaust steam pressure of the feed pump turbine; Among them, the input parameters of the feedwater pump speed BP neural network are selected from three parameters: unit power, tap flow, and tap pressure; the input parameters of the feedwater pump turbine steam consumption BP neural network are selected from six parameters: unit power, tap flow, tap pressure, feedwater pump speed measurement value, and feedwater pump turbine inlet and outlet enthalpy value; the input parameters of the feedwater pump turbine exhaust pressure BP neural network are selected from seven parameters: unit power, tap flow, tap pressure, feedwater pump speed measurement value, feedwater pump turbine inlet and outlet enthalpy value, and condenser pressure; Step 5: Simulation model fusion Fuse the output parameters of the mechanism-driven simulation model and the data-driven simulation model of the feed pump system to generate high-precision digital twin parameters, including the feed water mass flow rate m fw , the outlet pressure p of the feed pump fw , the rotational speed N of the feed pump, the efficiency η of the feed pump fp , the power W of the feed pump p , the steam consumption m of the feed pump steam turbine spt , the exhaust pressure p of the feed pump steam turbine ept , and the inlet pressure p of the feed pump obp ; By comparing the mean absolute error MAE, root mean square error RMSE, and coefficient of determination R of the output parameters between the mechanism-driven simulation model and the data-driven simulation model of the feed pump system 2 , for the feed pump speed N, the steam consumption m of the feed pump turbine spt and the exhaust steam pressure p of the feed pump turbine ept select a model with higher comprehensive accuracy for simulation; Step 6: Construction of water pump system fault diagnosis model Based on the output parameters of the water pump system mechanism-driven simulation model and the water pump system data-driven simulation model, a water pump system fault diagnosis model is established. The fault diagnosis model includes the calculation method of the comprehensive deviation, the average deviation of the verification data set, the deviation alarm threshold, and the contribution of each parameter deviation to the comprehensive deviation. The following is the specific construction method of the water pump system fault diagnosis model: First, calculate the comprehensive deviation l between the measured value of the water pump system parameter and the model simulation value: where: w i is the deviation contribution weight artificially assigned to each parameter according to the importance degree and abnormal frequency of the parameters in the feed water pump system; r i is the relative error between the measured data and the simulation data of each parameter, %; a sim and a mea are the simulation value and the measured value of the parameter, respectively; Subsequently, calculate the average deviation degree l of the verification dataset ave : where: n vd is the amount of data contained in the validation dataset; The alarm threshold of the fault diagnosis model is calculated based on the average deviation of the validation data set and the set alarm threshold coefficient: Where: k al is the alarm threshold coefficient of the fault diagnosis model; Finally, fault warning and determination of the root cause of the fault are carried out. When the real-time comprehensive deviation degree is lower than the alarm threshold, a fault alarm is triggered, and the determination of the root cause of the fault begins; the contribution degree c of each parameter deviation to the comprehensive deviation degree i is calculated as: By monitoring the contribution of each parameter deviation, the faulty parameter can be located; Step 7: Closed-loop control feedback The fault diagnosis results are fed back to the distributed control system DCS of the thermal power plant, and the control system corresponding to the abnormal parameters is actively adjusted to achieve fault diagnosis and closed-loop control of the water feed pump system; at the same time, the fault information is transmitted to the unit operation management personnel, the fault cause analysis results are given, and the equipment operation and maintenance plan recommended to the unit operation management personnel.

2. The performance monitoring and fault diagnosis method for the feed water pump system according to claim 1, characterized in that, The specific method of data preprocessing in step 2 includes: (1) Analyze the measurement data set, screen the missing values, abnormal values and error values in the measurement data, fill the data missing values, correct the data abnormal values, and eliminate the data error values; (2) The sliding window method is used to detect outliers in the data, and missing values are filled by linear interpolation or the mean of adjacent data; the time series data is smoothed and the Savitzky-Golay filter is used to eliminate high-frequency noise.

3. The method for monitoring the performance and diagnosing faults of a feed water pump system according to claim 1, wherein The calculation of the feed pump efficiency and the feed pump power in the mechanism modeling in Step 3 further includes: Calculating the feed water flow rates of the two parts before and after the feed pump tapping according to the feed pump tapping flow rate, and at the same time, combining with the feed pump speed, calculating the respective efficiencies of the two parts before and after the feed pump: m fw1 = m fw + m ie Where: m fw1 and m fw are the feed water flow rates before and after the tapping of the feed water pump, respectively, in t / h; m ie is the tapping flow rate of the feed water pump, in t / h; c4, c5, and c6 are all fitting coefficients of the performance characteristic curve provided by the feed water pump manufacturer; The total power of the feed pump is the sum of the powers of the two parts before and after the feed pump tapping. The power calculations of the parts before and after the feed pump tapping are as follows: Where: p ie is the extraction pressure of the feed water pump, in MPa; ρ is the average density of the working fluid in the feed water pump, in kg / m 3 .

4. The performance monitoring and fault diagnosis method for the feed water pump system according to claim 1, wherein The BP neural network structure in the data-driven simulation model of the feed pump system in Step 4 is: The BP neural network structure includes an input layer, a hidden layer, and an output layer. Among them, the hidden layer contains 10 neurons, and the data set is divided into a training set, a validation set, and a test set in the proportions of 70%, 15%, and 15%.

5. The method for performance monitoring and fault diagnosis of the feed water pump system according to claim 1, characterized in that, The specific training method of the BP neural network for data-driven modeling in Step 4 further includes: Using the Levenberg-Marquardt algorithm to optimize the network weights, where the activation function is the hyperbolic tangent function; At the same time, early stopping is used to prevent overfitting, and training is terminated when the validation set error does not decrease for 10 consecutive times.

6. The performance monitoring and fault diagnosis method for the feed water pump system according to claim 1, characterized in that, The weight assignment rule of each parameter deviation to the comprehensive deviation degree in Step 6 is: Considering the abnormal frequencies of the parameters in the actual feed pump system, higher weights should be set for the parameters with high abnormal frequencies. According to the actual abnormal times of each parameter, the weights of each parameter are calculated using the following formula: where: k weight is the weight distribution coefficient; n f,i is the number of anomalies of a certain parameter within a period of historical time; n f,sum is the sum of the number of anomalies of all parameters within a period of historical time.

7. The method for monitoring the performance and diagnosing faults of a feed water pump system according to claim 1, characterized in that, The determination of the root cause of the fault in Step 6 further includes: When the comprehensive deviation degree is lower than the alarm threshold, the fault diagnosis system will issue an alarm, and at the same time, start to track the abnormal thermal parameters. The root cause of the fault is mainly determined through two dimensions, namely the absolute value of the contribution degree of each parameter deviation to the comprehensive deviation degree and the growth rate of the contribution degree. When the contribution degree of the deviation of a certain parameter to the comprehensive deviation degree increases rapidly and exceeds 0.95 when the alarm is issued, it can be determined that the parameter is an abnormal parameter.

8. A system for implementing the method according to any one of claims 1-7, characterized in that, Including: A data acquisition module for obtaining operation data from the DCS system in real time; A digital twin modeling module integrating the mechanism-driven simulation model of the feed pump system and the data-driven simulation model of the feed pump system; A fault diagnosis module for calculating the comprehensive deviation degree and the contribution degree based on the feed pump system fault diagnosis model; A control execution module for feeding back the diagnosis result to the DCS and implementing a closed-loop control strategy.

Images