Thermal power plant coal mill fault sample generation method, system and device

Abstract

The invention belongs to the technical field of coal mill fault simulation, and relates to a thermal power plant coal mill fault sample generation method, system and device. The method comprises the following steps: establishing a dynamic mathematical model of the coal mill; the problem of parameters to be identified in the dynamic mathematical model is converted into a nonlinear optimization problem with constraints; parameter identification is achieved through a differential evolution algorithm by means of historical operation data, and a mechanism model with the result closer to actual operation data is obtained; generating fault data by using the mechanism model, mixing the fault data with actual operation fault data to obtain a virtual fault training set and a test set, training the generative adversarial network through the training set, and verifying through the test set to obtain a generator meeting a precision requirement; and adopting the trained generator to generate fault sample data of a specified type. By adopting the method provided by the invention, the fault data of the specified type can be generated without limitation, the generated data is closer to actual operation data, the data coverage range is wider, and the requirements are well met.

Description

Technical Field

[0001] The invention belongs to the technical field of coal mill fault simulation, and in particular relates to a method, system and device for generating fault samples of a coal mill in a thermal power plant. Background Art

[0002] Coal mills play an important role in the pulverizing system of thermal power plants. However, the process mechanism of coal mills is complex and the working conditions are changeable. Various faults often occur during operation. Once a fault occurs, it will directly affect the stability of boiler combustion, and then affect the operating status of the entire thermal power unit, causing unplanned shutdowns, causing large economic losses to the thermal power plant and even safety accidents. Therefore, the fault monitoring and diagnosis of coal mills is crucial to the normal operation of thermal power units. At present, the methods for monitoring and diagnosing coal mill faults are roughly divided into two directions. One is to use the operating mechanism of coal mills to establish a mechanism model for analysis. Due to the complex process mechanism of coal mills, the model is often simplified in the modeling process, and some parameters are empirical parameters, which will cause deviations between the mechanism model and the actual data; the other is to use machine learning and statistical knowledge to establish a black box or gray box model for diagnosis. However, this method often requires a large amount of fault data in the operation process for analysis and training. In the actual production process, it is difficult to collect enough samples with a wide coverage.

[0003] The existing technology usually directly uses the mechanism model to provide fault warning for the coal mill, but the mechanism model is often simplified during the modeling process, and the mechanism model usually contains some empirical parameters, which makes the output results of the model established for coal mills of different types or specifications have certain errors compared with the actual operation data, affecting the judgment of the fault type; another way is to establish a grayscale model through machine learning related knowledge, which requires the support of a large amount of fault production data. It is difficult to collect enough fault data samples during the actual operation of the equipment. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the prior art and propose a method, system and device for generating fault samples of a coal mill in a thermal power plant.

[0005] In order to achieve the above object, the present invention proposes a method for generating fault samples of a coal mill in a thermal power plant, comprising:

[0006] Establish the dynamic mathematical model of coal mill;

[0007] Transform the problem of identifying parameters in dynamic mathematical models into a nonlinear optimization problem with constraints;

[0008] Using historical operation data to identify parameters through differential evolution algorithm, a mechanism model with results closer to actual operation data is obtained;

[0009] After the fault data is generated by the mechanism model, it is mixed with the actual operation fault data to obtain the virtual fault training set and test set. The generative adversarial network is trained with the training set, and then verified with the test set to obtain a generator that meets the accuracy requirements.

[0010] Use the trained generator to generate fault sample data of the specified type.

[0011] Preferably, the coal mill dynamic mathematical model includes: a coal mill inlet primary air flow and temperature model, a coal mill raw coal and coal powder amount model, and a coal mill outlet temperature model; wherein,

[0012] The primary air flow and temperature model of the coal mill inlet satisfies the following dynamic differential equation:

[0013]

[0014] Among them, T1 and T2 are the flow inertia time and temperature inertia time of primary air from the damper to the coal mill inlet, respectively. Indicates the primary air flow W air Find the first-order derivative, Indicates the primary air temperature T at the coal mill inlet in Find the first-order derivative, T L 、T H are the temperatures of cold and hot primary air, respectively, U L , U H are the openings of the cold air valve and the hot air valve, respectively, C L , C H are the specific heat capacities of cold and hot primary air, C in is the specific heat capacity of primary air at the coal mill inlet, are the maximum flow rates of the cold and hot air valves respectively;

[0015] The model of raw coal and pulverized coal in the coal mill satisfies the following dynamic differential equation:

[0016]

[0017] W c =K c N c

[0018] W pf =K pf M pf ΔP

[0019] I=K1M pf +K2M c +K3

[0020] in, Respectively represent the raw coal mass M in the coal mill c and coal powder mass M pf Find the first-order derivative, W c is the raw coal feed rate, W pf K is the flow rate of pulverized coal blown out, conv K is the conversion coefficient of raw coal grinding to pulverized coal, c is the proportionality coefficient, N c is the speed of the coal feeder, K pf is the pulverized coal flow rate proportional coefficient, ΔP is the primary air differential pressure, I is the coal mill current, K1, K2, K3 are coefficients;

[0021] The coal mill outlet temperature model satisfies the following dynamic differential equation:

[0022]

[0023] Q in =Q air +Q rc +Q mac +Q seal +Q le

[0024] Q out =Q airseal +Q ΔM +Q pf +Q loss

[0025] in, Indicates the temperature of the air-powder mixture at the outlet of the coal mill T out Find the first-order derivative, Q in , Q out are the total heat input and output of the coal mill per unit time, M metal is the metal mass involved in heat exchange in the coal mill, C mix is the average specific heat capacity of coal and metal equipment, Q air , Q rc are the heat transferred by the sealed air and the leaked cold air, Q airseal is the heat output of primary air and seal air from the coal mill, Q ΔM is the heat consumed by evaporation of moisture in raw coal, Q pf The heat consumed by heating the fuel, Q loss Heat lost to dissipate equipment.

[0026] Preferably, the parameters to be identified in the dynamic mathematical model include: the maximum flow of the cold and hot air valves and The flow inertia time T1 and temperature inertia time T2 of the primary air from the damper to the coal mill inlet, the conversion coefficient K of raw coal grinding to coal powder conv , pulverized coal flow rate proportional coefficient K pf , coefficients K1, K2 and K3, and the metal mass M involved in heat exchange in the coal mill metal .

[0027] Preferably, the problem of identifying parameters in the dynamic mathematical model is converted into a nonlinear optimization problem with constraints; comprising:

[0028] Set the opening of the cold air valve and the hot air valve to L and U H And the raw coal feed amount W C As the input parameter of the model, the primary air flow rate W air , the temperature of the air-powder mixture at the coal mill outlet T out and the coal powder flow rate W pf As output parameters, let the given operating condition be C, the condition to be identified be D, the output condition be Y, and the mechanism model be H. The identification problem can be described as:

[0029] Y=H(C,D)

[0030] Transformed into a nonlinear optimization problem with constraints:

[0031]

[0032] d i,min ≤d i ≤d i,max

[0033] Among them, f is the target value, y s Output data for actual operation, y m is the model output data, i is the number of variables to be identified, d i is the i-th variable to be identified, d i,min and d i,max The minimum and maximum values ​​of the i-th variable to be identified are obtained respectively, and the parameters are identified by differential evolution algorithm using the historical operation data of the coal mill.

[0034] Preferably, the method of using historical operation data to identify parameters through a differential evolution algorithm to obtain a mechanism model whose results are closer to the actual operation data includes:

[0035] Step 1: Initialize the population and randomly generate a matrix of size N×D as the initial population, where N is the population size and D is the same as the number of variables to be identified;

[0036] Step 2: Mutation operation, select two different individuals in the population, scale their vectors and synthesize them with the vector of the individual to be mutated;

[0037] Step 3: Crossover operation, perform crossover operation between individuals on the g-th generation population and its corresponding variant intermediates;

[0038] Step 4: Selection operation, using the greedy algorithm to select individuals to enter the next generation population;

[0039] Step 5: Determine whether the end condition is met. If not, go to Step 2; otherwise, go to Step 6;

[0040] Step 6: Output identification parameters: Maximum flow of cold and hot air valves and The flow inertia time T1 and temperature inertia time T2 of the primary air from the damper to the coal mill inlet, the conversion coefficient K of raw coal grinding to coal powder conv , pulverized coal flow rate proportional coefficient K pf , coefficients K1, K2 and K3, and the metal mass M involved in heat exchange in the coal mill metal , and then the mechanism model is obtained.

[0041] Preferably, the generative adversarial network includes a generator G and a discriminator D, wherein:

[0042] The generator G and the discriminator D are both three-layer neural networks with opposite structures, including an input layer, a hidden layer and an output layer; the training process includes:

[0043] The generator G uses a randomly generated seed as input to generate data of the same dimension as the virtual fault training set, and uses the comparison loss of the discriminator D to update the generator G;

[0044] The discriminator D receives a pair of data from the generator G and the data of the virtual fault training set, classifies and compares it with the label, and updates the discriminator D through the loss;

[0045] Repeat the above process until the training requirements are met and a trained generator G is obtained.

[0046] In a second aspect, the present invention provides a system for generating fault samples of coal mills in thermal power plants, the system comprising: a mechanism model and parameter identification module and a fault sample generation module, wherein:

[0047] The mechanism model and parameter identification module are used to establish a dynamic mathematical model of the coal mill; convert the problem of parameters to be identified in the dynamic mathematical model into a nonlinear optimization problem with constraints; use historical operation data to identify parameters through a differential evolution algorithm to obtain a mechanism model whose results are closer to the actual operation data;

[0048] The fault sample generation module is used to generate fault sample data of a specified type using a trained generator;

[0049] The training process of the generator includes: after generating fault data using the mechanism model, mixing it with the actual operation fault data to obtain a virtual fault training set and a test set, training the generative adversarial network through the training set, and then verifying it through the test set to obtain a generator that meets the accuracy requirements.

[0050] In a third aspect, the present invention provides a device for generating fault samples of a coal mill in a thermal power plant, comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0051] Compared with the prior art, the advantages of the present invention are:

[0052] The present invention utilizes a differential evolution algorithm to identify empirical parameters in the mechanism model, so that the results generated by the mechanism model are closer to the actual operating data, and then learns and trains through a generative adversarial network to obtain a fault data generation model. Compared with the mechanism model, the fault data generated by this model can generate fault data of a specified type without limit, and the generated data is closer to the actual operating data and has a wider data coverage range, which can well meet the subsequent learning and training of the fault warning model. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a processing flow chart of the method for generating fault samples of coal mill in thermal power plant of the present invention;

[0054] Figure 2 It is a flowchart of the differential evolution algorithm. DETAILED DESCRIPTION

[0055] The present invention proposes a method for generating coal mill fault samples. This method generates specified types of fault data using a mechanism model according to the coal mill fault mechanism, and then mixes it with a small amount of fault data during operation to obtain virtual fault training samples. Through training and learning with a generative adversarial network, a fault sample generator that can generate specified fault types is obtained. This method can not only generate different types of specified fault data without restriction, but also has higher accuracy, a wider range of generated data, and is closer to actual production data than general mechanism models.

[0056] The technical solution of the present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0057] Example 1

[0058] like Figure 1As shown, Embodiment 1 of the present invention proposes a method for generating fault samples of a coal mill in a thermal power plant, comprising:

[0059] 1) Mechanism model and parameter identification part:

[0060] Establish the dynamic mathematical model of coal mill;

[0061] Transform the problem of identifying parameters in dynamic mathematical models into a nonlinear optimization problem with constraints;

[0062] Using historical operation data to identify parameters through differential evolution algorithm, a mechanism model with results closer to actual operation data is obtained;

[0063] 2) Fault sample generation part:

[0064] After the fault data is generated by the mechanism model, it is mixed with the actual operation fault data to obtain the virtual fault training set and test set. The generative adversarial network is trained with the training set, and then verified with the test set to obtain a generator that meets the accuracy requirements.

[0065] Use the trained generator to generate fault sample data of the specified type.

[0066] The following is a detailed explanation:

[0067] 1) Mechanism model and parameter identification

[0068] A dynamic mathematical model of the coal mill is established, and the model parameters that need to be identified are identified using the actual normal operation data through the chaos optimization algorithm, so as to obtain a mechanism model whose results are closer to the actual operation data. Finally, the dynamic model is used to simulate the typical fault data during the operation of the coal mill to generate typical fault sample data of the coal mill. The specific process is as follows:

[0069] (1) Establishing a mechanism model

[0070] The coal mill mechanism model generates fault data of specified types according to the fault mechanism using the primary air temperature and flow model at the coal mill inlet, the raw coal and pulverized coal quantity model in the coal mill, and the coal mill outlet temperature model. The data is mixed with a small amount of fault data and normal data in the historical operation data to form mixed operation data, which is finally transmitted to the fault sample generation module.

[0071] a) Primary air temperature and flow rate model at coal mill inlet

[0072] The dynamic differential equation of primary air temperature and flow rate at the coal mill inlet is:

[0073]

[0074] Among them, T1 and T2 are the flow inertia time and temperature inertia time of primary air from the damper to the coal mill inlet, respectively. L , T H are the temperatures of cold and hot primary air, T in is the primary air temperature at the coal mill inlet, W air is the primary air flow rate, are the maximum flow rates of the cold and hot air valves, U L , U H are the openings of the cold air valve and the hot air valve, respectively, C L , C H are the specific heat capacities of cold and hot primary air, C in is the specific heat capacity of primary air at the coal mill inlet.

[0075] b) Model of raw coal and pulverized coal quantity in coal mill

[0076] According to the mass balance equation of coal storage in the coal mill, the following dynamic differential equation can be established:

[0077]

[0078] W c =K c N c

[0079] W pf =K pf M pf ΔP

[0080] I=K1M pf +K2M c +K3

[0081] Among them, W c is the raw coal feed rate, W pf M is the flow rate of pulverized coal blown out, c 、M pf The mass of raw coal and pulverized coal in the pulverizer, K conv K is the conversion coefficient of raw coal grinding to pulverized coal, c is the proportionality coefficient, N c is the speed of the coal feeder, K pf is the pulverized coal flow rate proportional coefficient, ΔP is the primary air differential pressure, I is the coal mill current, and K1, K2, and K3 are the parameters that need to be identified.

[0082] c) Coal mill outlet temperature model

[0083] Based on the heat balance at the inlet and outlet of the coal mill, the dynamic differential equation of the temperature of the air-powder mixture at the outlet of the coal mill is established:

[0084]

[0085] Q in =Q air +Q rc +Q mac +Q seal +Q le

[0086] Q out =Q airseal +Q ΔM +Q pf +Q loss

[0087] Among them, Q in , Q out are the total heat input and output of the coal mill per unit time, M metal is the metal mass involved in heat exchange in the coal mill, C mix is the average specific heat capacity of coal and metal equipment, T out is the temperature of the air-powder mixture at the coal mill outlet, Q air , Q rc are the heat transferred by the sealed air and the leaked cold air, Q airseal is the heat output of primary air and seal air from the coal mill, Q ΔM is the heat consumed by evaporation of moisture in raw coal, Q pf The heat consumed by heating the fuel, Q loss Heat lost to dissipate equipment.

[0088] (2) Model parameter identification

[0089] The model parameters that need to be identified are T1, T2, K conv , K pf , K1, K2, K3, M metal , will U L , U H , W C As the input parameter of the model, W air 、T out , W pf As the output parameter. If the given operating condition is C, the identification condition is D, the output condition is Y, and the mechanism model is H, then the problem can be described as:

[0090] Y=H(C,D)

[0091] The identification process can be transformed into a nonlinear optimization problem with constraints:

[0092]

[0093] d i,min ≤di ≤d i,max

[0094] Among them, f is the target value. Here, the minimum error is taken as the target value, which has no physical meaning. s Output data for actual operation, y m is the model output data, i is the number of variables to be identified, d i is the i-th variable to be identified, d i,min is the minimum value of the i-th variable to be identified, d i,max is the maximum value of the i-th variable to be identified. The historical operation data of the coal mill is used to identify the parameters through the differential evolution algorithm.

[0095] Using the historical operation data of the coal mill, the parameters are identified by differential evolution algorithm. The specific steps are as follows: Figure 2 .

[0096] Step 1: Initialize the population and randomly generate a matrix of size N×D as the initial population, where N is the population size and D is the same as the number of variables to be identified.

[0097] Step 2: Mutation operation, select two different individuals in the population, scale their vectors and synthesize them with the vector of the individual to be mutated:

[0098] v i (g+1)=d r1 (g)+F(d r2 (g)-d r3 (g))

[0099] i≠r1≠r2≠r3

[0100] Where F is the scaling factor, d i (g) represents the i-th individual in the g-th generation population.

[0101] Step 3: Crossover operation, for the g-th generation population {d i (g)} and its corresponding variant intermediate {v i (g+1)} performs crossover operation between individuals:

[0102]

[0103] Among them, CR is the crossover rate, j rand is a random integer in [1,2,…,D].

[0104] Step 3: Selection operation, using the greedy algorithm to select individuals to enter the next generation population:

[0105]

[0106] 2) Fault sample generation

[0107] The fault sample generation part is mainly composed of a conditional adversarial network. The fault sample generation part marks the fault type of the mixed operation data generated by the mechanism model and parameter identification module, and then hands it over to the conditional adversarial network for learning and training. Finally, the trained network is used to generate different types of virtual fault data as needed and handed over to the model learning part in the fault diagnosis module for sample training.

[0108] The conditional adversarial network consists of two parts: the generator and the discriminator. The training goal of the discriminator is to maximize the accuracy of classification, while the training goal of the generator is to minimize the accuracy of the discriminator. The function evaluations of the two are opposite, thus forming a zero-sum confrontation. The objective function can be expressed as:

[0109]

[0110] Among them, G is the generator, D is the discriminator, and P data Represents the initial given data distribution, P z represents the randomly initialized noise distribution, G(z) represents the input is the random seed z and the output is the virtual fault data, D(f(x)) represents the input is the mixed operation data generated by the mechanism model and the parameter identification module, and the output is the probability.

[0111] The specific composition is as follows:

[0112] a) Generator

[0113] The generator is a three-layer neural network consisting of an input layer, a hidden layer, and an output layer. In contrast to the discriminator structure, it is actually a reverse discriminator. The generator uses a randomly generated seed as input to generate data of the same dimension as the mixed operation data generated by the wind turbine mechanism model, and calls the discriminator to update the generator using the discriminator's comparison loss (the discriminator is not updated at this time).

[0114] b) Discriminator

[0115] The discriminator is a three-layer neural network consisting of an input layer, a hidden layer, and an output layer. The discriminator receives a pair of data from the generator and mixed operation data generated by the mechanism model and parameter identification module, and marks the data from the generator as 0 and the data from the mixed operation as 1. The discriminator is used to classify and compare it with the label, and the discriminator is updated through the loss.

[0116] Example 2

[0117] Embodiment 2 of the present invention provides a coal mill fault sample generation system for a thermal power plant, comprising two modules: a mechanism model and parameter identification module and a fault sample generation module, which is implemented based on the method of embodiment 1.

[0118] The mechanism model and parameter identification module is used to establish a dynamic mathematical model of the coal mill; the problem of identifying parameters in the dynamic mathematical model is converted into a nonlinear optimization problem with constraints; the historical operation data is used to identify parameters through the differential evolution algorithm to obtain a mechanism model whose results are closer to the actual operation data;

[0119] The fault sample generation module is used to use the trained generator to generate fault sample data of a specified type.

[0120] The training process of the generator includes: after using the mechanism model to generate fault data, it is mixed with the actual operation fault data to obtain a virtual fault training set and a test set, the generative adversarial network is trained with the training set, and then verified with the test set to obtain a generator that meets the accuracy requirements.

[0121] Example 3

[0122] Embodiment 3 of the present invention provides a device for generating fault samples of a coal mill in a thermal power plant, comprising a memory, a processor and a computer program stored in the memory, and the processor executes the computer program to implement the method steps of embodiment 1.

[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention is described in detail with reference to the embodiments, it should be understood by those skilled in the art that any modification or equivalent replacement of the technical solutions of the present invention does not depart from the spirit and scope of the technical solutions of the present invention and should be included in the scope of the claims of the present invention.

Claims

1. A method for generating fault samples of a coal mill in a thermal power plant, comprising: Establish the dynamic mathematical model of coal mill; Transform the problem of identifying parameters in dynamic mathematical models into a nonlinear optimization problem with constraints; Using historical operation data to identify parameters through differential evolution algorithm, a mechanism model with results closer to actual operation data is obtained; After the fault data is generated by the mechanism model, it is mixed with the actual operation fault data to obtain the virtual fault training set and test set. The generative adversarial network is trained with the training set, and then verified with the test set to obtain a generator that meets the accuracy requirements. Use the trained generator to generate fault sample data of the specified type.

2. The method for generating fault samples of coal mill in thermal power plant according to claim 1, characterized in that: The coal mill dynamic mathematical model includes: a coal mill inlet primary air flow and temperature model, a coal mill raw coal and coal powder amount model and a coal mill outlet temperature model; wherein, The primary air flow and temperature model of the coal mill inlet satisfies the following dynamic differential equation: Among them, T1 and T2 are the flow inertia time and temperature inertia time of primary air from the damper to the coal mill inlet, respectively. Indicates the primary air flow W air Find the first-order derivative, Indicates the primary air temperature T at the coal mill inlet in Find the first-order derivative, T L 、T H are the temperatures of cold and hot primary air, respectively, U L , U H are the openings of the cold air valve and the hot air valve, respectively, C L , C H are the specific heat capacities of cold and hot primary air, C in is the specific heat capacity of primary air at the coal mill inlet, are the maximum flow rates of the cold and hot air valves respectively; The model of raw coal and pulverized coal in the coal mill satisfies the following dynamic differential equation: W c =K c N c W pf =K pf M pf ΔP I=K1M pf +K2M c +K3 in, Respectively represent the raw coal mass M in the coal mill c and coal powder mass M pf Find the first-order derivative, W c is the raw coal feed rate, W pf is the flow rate of pulverized coal blown out, K conv K is the conversion coefficient of raw coal grinding to pulverized coal, c is the proportionality coefficient, N c is the speed of the coal feeder, K pf is the pulverized coal flow rate proportional coefficient, ΔP is the primary air differential pressure, I is the coal mill current, K1, K2, K3 are coefficients; The coal mill outlet temperature model satisfies the following dynamic differential equation: Q in =Q air +Q re +Q mac +Q seal +Q le Q out =Q airseal +Q ΔM +Q pf +Q loss in, Indicates the temperature of the air-powder mixture at the outlet of the coal mill T out Find the first-order derivative, Q in , Q out are the total heat input and output of the coal mill per unit time, M metal is the metal mass involved in heat exchange in the coal mill, C mix is the average specific heat capacity of coal and metal equipment, Q air , Q rc are the heat transferred by the sealed air and the leaked cold air, Q airseal is the heat output of primary air and seal air from the coal mill, Q ΔM is the heat consumed by evaporation of moisture in raw coal, Q pf The heat consumed by heating the fuel, Q loss Heat lost to dissipate equipment.

3. The method for generating fault samples of coal mill in thermal power plant according to claim 2, characterized in that: The parameters to be identified in the dynamic mathematical model include: the maximum flow of the cold and hot air valves and The flow inertia time T1 and temperature inertia time T2 of the primary air from the damper to the coal mill inlet, the conversion coefficient K of raw coal grinding to coal powder conv , pulverized coal flow rate proportional coefficient K pf , coefficients K1, K2 and K3, and the metal mass M involved in heat exchange in the coal mill metal .

4. The method for generating fault samples of coal mill in thermal power plant according to claim 3, characterized in that: The problem of converting the problem of identifying parameters in the dynamic mathematical model into a nonlinear optimization problem with constraints includes: Set the opening of the cold air valve and the hot air valve to L and U H And the raw coal feed amount W C As the input parameter of the model, the primary air flow rate W air , the temperature of the air-powder mixture at the coal mill outlet T out and the coal powder flow rate W pf As output parameters, let the given operating condition be C, the condition to be identified be D, the output condition be Y, and the mechanism model be H. The identification problem can be described as: Y=H(C,D) Transformed into a nonlinear optimization problem with constraints: d i，min ≤d i ≤d i，max Among them, f is the target value, y s Output data for actual operation, y m is the model output data, i is the number of variables to be identified, d i is the i-th variable to be identified, d i，min and d i，max The minimum and maximum values ​​of the i-th variable to be identified are obtained respectively, and the parameters are identified by differential evolution algorithm using the historical operation data of the coal mill.

5. The method for generating fault samples of coal mill in thermal power plant according to claim 4, characterized in that: The method of using historical operation data to identify parameters through a differential evolution algorithm to obtain a mechanism model whose results are closer to the actual operation data includes: Step 1: Initialize the population and randomly generate a matrix of size N×D as the initial population, where N is the population size and D is the same as the number of variables to be identified; Step 2: Mutation operation, select two different individuals in the population, scale their vectors and synthesize them with the vector of the individual to be mutated; Step 3: Crossover operation, perform crossover operation between individuals on the g-th generation population and its corresponding variant intermediates; Step 4: Selection operation, using the greedy algorithm to select individuals to enter the next generation population; Step 5: Determine whether the end condition is met. If not, go to Step 2; otherwise, go to Step 6; Step 6: Output identification parameters: Maximum flow of cold and hot air valves and The flow inertia time T1 and temperature inertia time T2 of the primary air from the damper to the coal mill inlet, the conversion coefficient K of raw coal grinding to coal powder conv , pulverized coal flow rate proportional coefficient K pf , coefficients K1, K2 and K3, and the metal mass M involved in heat exchange in the coal mill metal , and then the mechanism model is obtained.

6. The method for generating fault samples of coal mill in thermal power plant according to claim 1, characterized in that: The generative adversarial network includes a generator G and a discriminator D, wherein: The generator G and the discriminator D are both three-layer neural networks with opposite structures, including an input layer, a hidden layer and an output layer; the training process includes: The generator G uses a randomly generated seed as input to generate data of the same dimension as the virtual fault training set, and uses the comparison loss of the discriminator D to update the generator G; The discriminator D receives a pair of data from the generator G and the data of the virtual fault training set, classifies and compares it with the label, and updates the discriminator D through the loss; Repeat the above process until the training requirements are met and a trained generator G is obtained.

7. A coal mill fault sample generation system for a thermal power plant, characterized in that: The system includes: a mechanism model and parameter identification module and a fault sample generation module, wherein: The mechanism model and parameter identification module are used to establish a dynamic mathematical model of the coal mill; convert the problem of parameters to be identified in the dynamic mathematical model into a nonlinear optimization problem with constraints; use historical operation data to identify parameters through a differential evolution algorithm to obtain a mechanism model whose results are closer to the actual operation data; The fault sample generation module is used to generate fault sample data of a specified type using a trained generator; The training process of the generator includes: after generating fault data using the mechanism model, mixing it with the actual operation fault data to obtain a virtual fault training set and a test set, training the generative adversarial network through the training set, and then verifying it through the test set to obtain a generator that meets the accuracy requirements.

8. A device for generating fault samples of coal mills in a thermal power plant, comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method of claim 1.

Images