CN114021283A - Gas turbine compressor detection and water washing optimization method - Google Patents

Gas turbine compressor detection and water washing optimization method Download PDF

Info

Publication number
CN114021283A
CN114021283A CN202111328290.9A CN202111328290A CN114021283A CN 114021283 A CN114021283 A CN 114021283A CN 202111328290 A CN202111328290 A CN 202111328290A CN 114021283 A CN114021283 A CN 114021283A
Authority
CN
China
Prior art keywords
compressor
steady
state
data
gas turbine
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111328290.9A
Other languages
Chinese (zh)
Inventor
杨建球
吴二涛
李锦峰
黄元平
孙子立
南泽瑞
许峰
耿大斌
何杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong Yuedian Zhongshan Thermal Power Plant Co ltd
Original Assignee
Guangdong Yuedian Zhongshan Thermal Power Plant Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guangdong Yuedian Zhongshan Thermal Power Plant Co ltd filed Critical Guangdong Yuedian Zhongshan Thermal Power Plant Co ltd
Priority to CN202111328290.9A priority Critical patent/CN114021283A/en
Publication of CN114021283A publication Critical patent/CN114021283A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/17—Mechanical parametric or variational design
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/04—Constraint-based CAD
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/08—Thermal analysis or thermal optimisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Geometry (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computer Hardware Design (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

The invention discloses a gas turbine compressor detection and water washing optimization method, which comprises the following specific steps: constructing a gas turbine compressor fault model by utilizing an FMEA (failure mode analysis) and FTA (fiber to the array) analysis tool to obtain parameters influencing compressor scaling and measuring points for monitoring the parameters; acquiring historical operating data of the gas turbine according to parameters influencing scaling of the gas compressor and measuring points for monitoring the parameters; performing data steady-state screening on historical operating data by adopting a sliding window algorithm, and acquiring health references under different working conditions by adopting a DBSCAN clustering algorithm; training and optimizing to obtain a compressor state prediction model according to the acquired steady-state data set and the health benchmark; and testing the compressor state prediction model to obtain the evaluation of the current compressor scaling degree and make a reasonable offline water washing decision. The invention adopts big data and artificial intelligence technology, and makes a reasonable gas turbine off-line water washing strategy according to economy, thereby promoting the energy conservation and consumption reduction of the gas turbine.

Description

Gas turbine compressor detection and water washing optimization method
Technical Field
The invention relates to the field of off-line water washing of a gas-steam combined cycle generator set, in particular to a gas turbine compressor detection and water washing optimization method.
Background
The gas-steam combined cycle unit is developed rapidly in China in recent years by virtue of the advantages of high heat efficiency, less three-waste emission, suitability for peak regulation and the like. The performance of the gas turbine as a core component of the combined cycle generator set directly affects the operation condition of the whole set. The gas turbine needs to continuously suck air during operation, and the sucked air may contain dust, insects and the like. Although most of the pollutants are removed by the air inlet filter before entering the air compressor, a small amount of pollutant particles enter the air compressor and are continuously deposited in the air compressor, so that the efficiency of the air compressor is reduced, severe accidents such as surging and the like can be caused in the serious condition, and the safety, the reliability and the economy of a unit are reduced. In order to ensure the safe and economic operation of the unit, the power plant can regularly arrange off-line water washing to recover the performance of the compressor.
Currently, off-line water washing of a gas turbine of a power station is generally performed periodically according to the number of operating hours of the gas turbine. However, due to the influence of environmental factors and variable working condition operation of the unit, the scaling degree of the compressor and the time are not in a simple linear relationship, the periodic off-line washing mode cannot maintain the high-efficiency operation of the unit, and a large amount of desalted water and station service power are wasted if the washing is frequent; otherwise, the efficiency of the compressor is reduced due to untimely water washing, and the power generation cost is increased. With the help of artificial intelligence technology, state monitoring and fault diagnosis are carried out, the running state of equipment is predicted, and practical situation-based maintenance technology is developed, which becomes a profound task. The gas-steam combined cycle unit is difficult to keep working under a rated working condition for a long time due to the fact that the gas-steam combined cycle unit participates in peak shaving by power grid dispatching. Therefore, the method has practical significance of energy conservation and consumption reduction by identifying the scaling degree of the compressor under variable working conditions and formulating a water washing strategy by combining the water washing cost.
Disclosure of Invention
The invention aims to overcome the problems in the prior art and provides a gas turbine compressor detection and water washing optimization method.
The purpose of the invention is realized by the following technical scheme:
a gas turbine compressor detection and water washing optimization method comprises the following specific steps:
the method comprises the following steps: constructing a gas turbine compressor fault model by utilizing an FMEA (failure mode analysis) and FTA (fiber to the array) analysis tool to obtain parameters influencing compressor scaling and measuring points for monitoring the parameters;
step two: acquiring historical operating data of the gas turbine according to parameters influencing scaling of the gas compressor and measuring points for monitoring the parameters;
step three: performing data steady-state screening on the historical operating data by adopting a sliding window algorithm, and acquiring health benchmarks of the historical operating data under different working conditions by adopting a DBSCAN clustering algorithm;
step four: training and optimizing to obtain a compressor state prediction model according to the acquired steady-state data set and the health benchmark;
step five: inputting the real-time unit operation data of the gas turbine compressor into a compressor state prediction model to obtain the evaluation of the current compressor scaling degree, and providing a reasonable offline water washing decision according to the evaluation of the compressor scaling degree.
The first step is specifically as follows: firstly, analyzing gas compressor equipment of a gas turbine to form an equipment tree; and secondly, forming equipment fault analysis about the gas compressor according to knowledge accumulation and theoretical analysis of operation and maintenance of the gas power plant, and obtaining a fault mode, a generation part, fault parameter symptoms and fault reasons.
The second step is specifically as follows: and (4) combining the step one and actual experience analysis to obtain a plurality of parameter types and measuring points which affect the scaling of the compressor, and storing the acquired data to form a historical database.
The data steady-state screening in the third step comprises the following specific steps: determining the length of a sliding window and a stable data stability index in a sliding window algorithm through linear programming of data;
the objective function of the linear program is:
Figure BDA0003347951370000021
the constraint conditions are as follows:
Figure BDA0003347951370000022
0≤w1≤10，
0≤w1≤10，
wherein epsiloniThe steady state stability index of the ith parameter is obtained; n represents the percentage of the currently screened steady-state data volume to the total data volume; Δ l represents the sliding window length in min; n ═ f (Δ l) represents the sliding window length Δ l as a function of n; w is a1、w2Is a weight coefficient;
given an initial value of iteration epsiloniΔ l and set weight coefficient w1、w2(ii) a Performing steady-state screening on each parameter by adopting a steady-state screening algorithm based on CST, wherein the screening result is a steady state or an unsteady state; the steady state time periods of all the parameters are subjected to union gathering to obtain a total steady state time period, and steady state data are obtained corresponding to the time periods; calculating to obtain a linear programming objective function; changing the variable ε using the PSO algorithmiAnd delta l, and finally solving the related parameters through iteration to obtainA steady state data set.
Steady state stability indicator epsilon for the ith parameteriThe calculation formula is as follows:
Figure BDA0003347951370000031
wherein, XiRepresenting a steady-state data column which is screened by the ith parameter according to the currently given sliding window length and the steady-state data stability index; max refers to the maximum value of the sequence; min refers to the minimum value of the number sequence; average refers to the average of the number series.
The percentage n of the currently screened steady-state data volume to the total data volume is calculated in the following way:
Figure BDA0003347951370000032
wherein N is the number of the steady-state data screened currently, and N is0The total number of data pieces in the historical database.
The specific steps of calculating the health standard in the third step are as follows:
with the unit load as an object, performing DBSCAN clustering on the steady-state data set based on the density to obtain an optimal clustering number S, a clustering interval and a unit load clustering cluster;
for the unit load cluster, with the ambient temperature as an object, performing DBSCAN clustering on the steady-state data set of each unit load cluster based on the density to obtain the optimal cluster number K and a cluster interval;
the clustering interval of the unit load clustering cluster is a working condition dividing unit, and the clustering center of each unit load clustering cluster is a unit health standard of the working condition dividing unit.
The fourth step is specifically as follows:
inputting the steady state data set of the unit into a compressor state prediction model, and performing model training by using a desiccant self-encoder principle;
and performing optimization iterative computation on the compressor state prediction model by adopting an SGD algorithm, and finishing iteration and training the compressor state prediction model when the optimization objective function reaches the minimum.
The fifth step is specifically as follows:
inputting real-time unit operation data of a gas turbine compressor into a compressor state prediction model, calculating a compressor health index, and when the compressor health index is smaller than a preset value, scaling faults occur in the compressor;
and (4) combining the offline washing income of the unit and the actual maintenance plan of the power plant, making an offline washing strategy and performing offline washing on the compressor.
The off-line washing profit calculation process specifically comprises the following steps:
calculating unit additional cost and shutdown washing cost of the gas-steam combined cycle unit by combining historical data of a power plant;
the unit additional cost is calculated by the formula:
Figure BDA0003347951370000041
wherein M isrunRepresents the unit operating parasitic loss, Δ W1Indicating a low power generation during the operating time after a combustion engine fouling failure, WrunIndicates the generating capacity of the unit after the scaling fault occurs, delta F1Indicating that the combustion engine consumes more natural gas fuel after scaling failure, CeIndicating the price of electricity on the Internet, CfRepresents the price of the fuel;
the calculation formula of the shutdown washing cost is as follows:
Mstop＝ΔW2Ce-ΔF2Cf+Mw，
wherein, Δ W2Indicating the amount of power, Δ F, of the combined cycle unit during shutdown2Representing the amount of fuel saved during shutdown, MwRepresents the cost of water washing;
calculating the offline washing profit by combining unit operation additional loss and shutdown washing cost;
the off-line washing yield calculation formula is as follows:
E＝Mrun×We-Mstop，
wherein E represents the net gain of the washing, and WeAnd (4) the expected power generation amount of the combustion engine at the next stop time after the washing.
The invention has the beneficial effects that:
according to the invention, big data and artificial intelligence technology are adopted, the prediction and evaluation of the overall performance state of the gas turbine compressor can be realized by mining the historical data of the unit and monitoring the real-time data of the unit, the scaling degree of the gas turbine under the current variable working condition is identified, and a reasonable gas turbine off-line water washing strategy is formulated according to economy, so that the energy conservation and consumption reduction of the gas turbine are further promoted.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the structures shown in the drawings without creative efforts.
FIG. 1 is a general flow diagram of the compressor condition monitoring and off-line water washing process of the present invention;
FIG. 2 is a flow chart of the data steady state screening of the present invention;
FIG. 3 is a flow chart of a DBSCAN clustering algorithm of the present invention;
FIG. 4 is a flow chart of denoising encoder model construction and health monitoring according to the present invention.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1-4, a gas turbine compressor detection and water wash optimization method includes the following steps:
the method comprises the following steps: constructing a gas turbine compressor fault model by utilizing an FMEA (failure mode analysis) and FTA (fiber to the array) analysis tool to obtain parameters influencing compressor scaling and measuring points for monitoring the parameters;
step two: acquiring historical operating data of the gas turbine according to parameters influencing scaling of the gas compressor and measuring points for monitoring the parameters;
step three: performing data steady-state screening on the historical operating data by adopting a sliding window algorithm, and acquiring health benchmarks of the historical operating data under different working conditions by adopting a DBSCAN clustering algorithm;
step four: training and optimizing to obtain a compressor state prediction model according to the acquired steady-state data set and the health benchmark;
step five: inputting the real-time unit operation data of the gas turbine compressor into a compressor state prediction model to obtain the evaluation of the current compressor scaling degree, and providing a reasonable offline water washing decision according to the evaluation of the compressor scaling degree.
The first step is specifically as follows: firstly, analyzing gas compressor equipment of a gas turbine to form an equipment tree; and secondly, forming equipment fault analysis about the gas compressor according to knowledge accumulation and theoretical analysis of operation and maintenance of the gas power plant, and obtaining a fault mode, a generation part, fault parameter symptoms and fault reasons. Because the off-line water washing of the compressor after the scaling fault of the compressor is only a countermeasure, the frequency of the scaling fault of the compressor is reduced by searching the relevant fault reason, and the cost caused by the off-line water washing is reduced.
The results of the correlation analysis are shown in table 1:
TABLE 1
Figure BDA0003347951370000061
Figure BDA0003347951370000071
In the table, the corresponding decreasing (decreasing), normal, increasing means: and (4) solving the small, normal and high health reference of the unit under a certain working condition relative to the steps.
The compressor pressure ratio and the compressor efficiency need to be calculated in real time, and the calculation mode is as follows:
Figure BDA0003347951370000072
wherein, pi represents the compressor pressure ratio, P2Indicating compressor outlet pressure, P1Compressor inlet pressure.
The calculation formula of the compressor efficiency is as follows:
Figure BDA0003347951370000073
where eta represents the compressor efficiency, t1Indicating compressor inlet temperature, t2Indicating compressor outlet temperature, t2sRepresenting the isentropic temperature at the compressor outlet.
The second step is specifically as follows: and (3) combining the step one and actual experience analysis to obtain a plurality of parameter types and measuring points which influence the scaling of the compressor, as shown in a table 2, and storing the collected data to form a historical database.
TABLE 2
Figure BDA0003347951370000081
The data steady-state screening in the third step comprises the following specific steps: determining the length of a sliding window and a stable data stability index in a sliding window algorithm through linear programming of data;
the objective function of the linear program is:
Figure BDA0003347951370000082
the constraint conditions are as follows:
Figure BDA0003347951370000083
0≤w1≤10，
0≤w1≤10，
wherein epsiloniThe steady state stability index of the ith parameter is obtained; n represents the percentage of the currently screened steady-state data volume to the total data volume;
and delta l represents the length of the sliding window, the unit is min, and the constraint condition that delta l is more than or equal to 25 and less than or equal to 40 is set to prevent the problem that the solution of the optimization problem caused by the overlong length of the sliding window cannot be converged, but the length of the sliding window cannot be too short, because the steady state of the unit data in a very short time is not significant on the thermodynamic operation of the unit, the steady operation of a gas turbine compressor in a longer time is required to reflect that the unit is in the steady operation state currently.
n ═ f (Δ l) represents the functional relationship between the sliding window length Δ l and n, but the functional relationship has no specific expression, and the sliding window length can only be changed in the iterative calculation for solving the linear programming problem, and n is calculated after steady-state screening.
w1、w2Is a weight coefficient; the weighting factor is given empirically by a human being, and w is given when the quality of the steady-state data is considered important1Should be higher than w2On the contrary w2Should be higher than w1。
Given an initial value of iteration epsiloniWhen i is equal to 0.1(i is equal to 1, 2, …, 13) and Δ l is equal to 30, the weighting factor w is set1、w2(ii) a Performing steady-state screening on each parameter by adopting a steady-state screening algorithm based on CST, wherein the screening result is a steady state or an unsteady state; the steady state time periods of all the parameters are subjected to union gathering to obtain a total steady state time period, and steady state data are obtained corresponding to the time periods; calculating an objective function to obtain a linear program(ii) a Changing the variable ε using the PSO algorithmiAnd delta l, and finally solving the relevant parameters through iteration to obtain a steady-state data set.
Steady state stability indicator epsilon for the ith parameteriThe calculation formula is as follows:
Figure BDA0003347951370000091
wherein, XiRepresenting a steady-state data column which is screened by the ith parameter according to the currently given sliding window length and the steady-state data stability index; max refers to the maximum value of the sequence; min refers to the minimum value of the number sequence; average refers to the average of the number series. And the ith parameter corresponds to the parameter in the column numbered in table 2.
The percentage n of the currently screened steady-state data volume to the total data volume is calculated in the following way:
Figure BDA0003347951370000092
wherein N is the number of the steady-state data screened currently, and N is0The total number of data pieces in the historical database.
The specific steps of calculating the health standard in the third step are as follows:
with the unit load as an object, performing DBSCAN clustering on the steady-state data set based on density to obtain an optimal clustering number S and a clustering interval;
the optimal clustering number S is calculated as:
Figure BDA0003347951370000101
where N represents the total number of steady-state samples, biRepresents the minimum distance from the ith sample point to the sample points of other classes, aiRepresenting the average distance of the ith sample point to other sample points within the same class.
And the calculation mode of the clustering interval of each type is to record the maximum value and the minimum value of the unit load in each clustering type.
For the unit load cluster, with the ambient temperature as an object, performing DBSCAN clustering on the steady-state data set of each unit load cluster based on the density to obtain the optimal cluster number K and a cluster interval;
the clustering interval of the unit load clustering cluster is a working condition dividing unit, and the clustering center of each unit load clustering cluster is a unit health standard of the working condition dividing unit.
In order to determine the real-time operation state of the unit, the health reference data of the unit under each working condition needs to be acquired and used as a measurement standard. When the current operating data of the unit deviates significantly from the healthy reference, it is considered that the unit may have some problems. The reference selection standard of the unit is the operation state with the lowest heat consumption rate under the condition of safe operation. Therefore, the invention uses the density-based DBSCAN clustering algorithm for the division of the unit working conditions and the acquisition of the health standard.
Because the operation of the unit is limited by the environment and load requirements of the unit, it is impossible to have a large amount of steady-state data available for extraction for each operating condition. Then a simple method of equally dividing the operating conditions according to the unit boundary parameters will generate a large number of useless operating condition units.
The density-based DBSCAN algorithm is used for obtaining the running health reference of the gas turbine compressor, the method can reflect the running of the unit better based on the clustering degree clustering of the parameters than other clustering methods, meanwhile, the clustering method can cluster high-dimensional data without considering the magnitude of different data, and meanwhile, the sensitivity to abnormal data points is low. This makes it possible to have better applicability than the K-Means clustering algorithm in clustering gas turbine unit operating data that are high-dimensional and prone to outliers. Generally, if the steady-state operation data of the unit is more aggregated in a certain block, the higher the aggregation degree, the more the data can indicate that the block is a common operation state of the unit under the working condition.
Secondly, aiming at the problem of generating a large number of useless working condition units, the invention adopts a multi-step DBSCAN algorithm, namely firstly clustering is carried out by taking the unit load as an object, and then clustering is carried out on each classified type of data by taking the environmental temperature as the object, so that the classification is carried out according to the running state of the existing unit instead of artificial division, and the problem is effectively solved.
The fourth step is specifically as follows:
inputting the steady state data set of the unit into a compressor state prediction model, and performing model training by using a desiccant self-encoder principle;
the steady-state operation parameters of the unit are expressed as follows:
z＝(tc，Δpe，θ，Qf，p0，μ0，tr，Qt，H，η，π，Qc，P)T，
wherein, the parameters are historical steady-state data, which can be detailed in table 2.
And performing optimization iterative computation on the compressor state prediction model by adopting an SGD algorithm, and finishing iteration and training the compressor state prediction model when the optimization objective function reaches the minimum.
(1) Gaussian noise which obeys normal distribution is added to an original input vector to form data containing noise
Figure BDA0003347951370000113
Figure BDA0003347951370000114
(2) The data containing noise is processed according to
Figure BDA0003347951370000115
Is mapped to a low-dimensional hidden variable space,
wherein σ is an activation function corresponding to the encoder in the self-encoder neural network, W is a weight matrix corresponding to the encoder, and b is a bias vector corresponding to the encoder;
(3) the low-dimensional hidden variable space is decoded in terms of z ' ═ σ ' (W ' h + b ') to form a reconstructed data set z ', where each is a parameter corresponding to a decoder.
(4) Calculating a residual error:
Figure BDA0003347951370000111
and performing optimization iterative computation on the model by adopting an SGD (random gradient descent method), and finishing the model training when the iteration is terminated when the optimization objective function (namely residual error) is minimum.
The model is used for real-time monitoring of the gas compressor of the gas turbine, and a real-time monitoring data vector at a certain moment is set as x ═ t (t)c，Δpe，θ，Qf，p0，μ0，tr，Qt，H，η，π，Qc，P)TThe reconstructed prediction monitoring data vector passing through the denoising coder model at the moment is
Figure BDA0003347951370000116
The reference value under the working condition corresponding to the real-time monitoring vector at the moment is
Figure BDA0003347951370000117
The health index is calculated:
Figure BDA0003347951370000112
the health index is calculated in real time, whether the current operation condition of the gas compressor of the gas turbine is normal or not can be measured, when the deviation between the real-time operation value of the gas compressor and the health standard is overlarge, a scaling fault is considered to occur, and off-line water washing is planned.
The contradiction between the steady-state screening of the data in the current unit judgment is that if the screening condition is too strict, only a very small amount of data is obtained, which is not beneficial to model training. If the screening condition is too loose, the fluctuation range of the data after steady-state screening is still too large, and the accuracy of the model is influenced.
According to the thermal operation principle of the unit, the change amplitude of data required for defining the steady state in the steady state screening is determined, so that the SGD algorithm is adopted for solving the linear programming problem.
The fifth step is specifically as follows:
inputting real-time unit operation data of a gas turbine compressor into a compressor state prediction model, calculating a compressor health index, and when the compressor health index is smaller than a preset value, scaling faults occur in the compressor;
and (4) combining the offline washing income of the unit and the actual maintenance plan of the power plant, making an offline washing strategy and performing offline washing on the compressor.
A general prediction model such as algorithms of RNN-LSTM, PSO-SVR and the like can only predict a single result, and the compressor state prediction model needs to be combined with fault analysis, so that the reduction of different parameters is probably caused by different results, and the prediction of the single result cannot well cover different fault conditions. The prediction model can predict all indexes of the compressor (including important state indexes of the compressor) through the input of real-time unit operation data, so that the abnormal state of the compressor scaling can be found earlier, and the deviation between the prediction and the actual operation value is measured by adopting a KDE nuclear density estimation method to obtain the evaluation of the scaling degree of the current compressor so as to provide reference for an offline water washing decision.
The off-line washing profit calculation process specifically comprises the following steps:
calculating unit additional cost and shutdown washing cost of the gas-steam combined cycle unit by combining historical data of a power plant;
the unit additional cost is calculated by the formula:
Figure BDA0003347951370000121
wherein M isrunRepresents the unit operating parasitic loss, Δ W1Indicating a low power generation during the operating time after a combustion engine fouling failure, WrunIndicates the generating capacity of the unit after the scaling fault occurs, delta F1Indicating that the combustion engine consumes more natural gas fuel after scaling failure, CeIndicating the price of electricity on the Internet, CfRepresents the price of the fuel;
the calculation formula of the shutdown washing cost is as follows:
Mstop＝ΔW2Ce-ΔF2Cf+Mw，
wherein, Δ W2Indicating the amount of power, Δ F, of the combined cycle unit during shutdown2Representing the amount of fuel saved during shutdown, MwRepresents the cost of water washing;
calculating the offline washing profit by combining unit operation additional loss and shutdown washing cost;
the off-line washing yield calculation formula is as follows:
E＝Mrun×We-Mstop，
wherein E represents the net gain of the washing, and WeAnd (4) the expected power generation amount of the combustion engine at the next stop time after the washing.
The foregoing shows and describes the general principles and broad features of the present invention and advantages thereof. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.

Claims (10)

1. A gas turbine compressor detection and water washing optimization method is characterized by comprising the following specific steps:
the method comprises the following steps: constructing a gas turbine compressor fault model by utilizing an FMEA (failure mode analysis) and FTA (fiber to the array) analysis tool to obtain parameters influencing compressor scaling and measuring points for monitoring the parameters;
step two: acquiring historical operating data of the gas turbine according to parameters influencing scaling of the gas compressor and measuring points for monitoring the parameters;
step three: performing data steady-state screening on the historical operating data by adopting a sliding window algorithm, and acquiring health benchmarks of the historical operating data under different working conditions by adopting a DBSCAN clustering algorithm;
step four: training and optimizing to obtain a compressor state prediction model according to the acquired steady-state data set and the health benchmark;
step five: inputting the real-time unit operation data of the gas turbine compressor into a compressor state prediction model to obtain the evaluation of the current compressor scaling degree, and providing a reasonable offline water washing decision according to the evaluation of the compressor scaling degree.
2. The gas turbine compressor detection and water wash optimization method according to claim 1, wherein the first step is specifically: firstly, analyzing gas compressor equipment of a gas turbine to form an equipment tree; and secondly, forming equipment fault analysis about the gas compressor according to knowledge accumulation and theoretical analysis of operation and maintenance of the gas power plant, and obtaining a fault mode, a generation part, fault parameter symptoms and fault reasons.
3. The gas turbine compressor detection and water wash optimization method according to claim 1, wherein the second step specifically comprises: and (4) combining the step one and actual experience analysis to obtain a plurality of parameter types and measuring points which affect the scaling of the compressor, and storing the acquired data to form a historical database.
4. The gas turbine compressor detection and water washing optimization method according to claim 1, wherein the data steady state screening in the third step comprises the specific steps of: determining the length of a sliding window and a stable data stability index in a sliding window algorithm through linear programming of data;
the objective function of the linear program is:
Figure FDA0003347951360000011
the constraint conditions are as follows:
Figure FDA0003347951360000012
0≤w1≤10，
0≤w1≤10，
wherein epsiloniThe steady state stability index of the ith parameter is obtained; n represents the percentage of the currently screened steady-state data volume to the total data volume; Δ l represents the sliding window length in min; n ═ f (Δ l) represents the sliding window length Δ l as a function of n; w is a1、w2Is a weight coefficient;
given an initial value of iteration epsiloniΔ l and set weight coefficient w1、w2(ii) a Performing steady-state screening on each parameter by adopting a steady-state screening algorithm based on CST, wherein the screening result is a steady state or an unsteady state; the steady state time periods of all the parameters are subjected to union gathering to obtain a total steady state time period, and steady state data are obtained corresponding to the time periods; calculating to obtain a linear programming objective function; changing the variable ε using the PSO algorithmiAnd delta l, and finally solving the relevant parameters through iteration to obtain a steady-state data set.
5. The method of claim 4, wherein the steady state stability indicator ε of the ith parameteriThe calculation formula is as follows:
Figure FDA0003347951360000021
wherein, XiRepresenting a steady-state data column which is screened by the ith parameter according to the currently given sliding window length and the steady-state data stability index; max refers to the maximum value of the sequence; min refers to the minimum value of the number sequence; average refers to the average of the number series.
6. The gas turbine compressor detection and water wash optimization method according to claim 4, wherein the percentage n of the currently screened steady-state data volume to the total data volume is calculated as follows:
Figure FDA0003347951360000022
wherein N is the number of the steady-state data screened currently, and N is0The total number of data pieces in the historical database.
7. The method of claim 1, wherein the step three of calculating the health benchmark comprises the following specific steps:
with the unit load as an object, performing DBSCAN clustering on the steady-state data set based on the density to obtain an optimal clustering number S, a clustering interval and a unit load clustering cluster;
for the unit load cluster, with the ambient temperature as an object, performing DBSCAN clustering on the steady-state data set of each unit load cluster based on the density to obtain the optimal cluster number K and a cluster interval;
the clustering interval of the unit load clustering cluster is a working condition dividing unit, and the clustering center of each unit load clustering cluster is a unit health standard of the working condition dividing unit.
8. The gas turbine compressor detection and water wash optimization method according to claim 1, wherein the fourth step is specifically:
inputting the steady state data set of the unit into a compressor state prediction model, and performing model training by using a desiccant self-encoder principle;
and performing optimization iterative computation on the compressor state prediction model by adopting an SGD algorithm, and finishing iteration and training the compressor state prediction model when the optimization objective function reaches the minimum.
9. The gas turbine compressor detection and water wash optimization method according to claim 1, wherein the step five specifically comprises:
inputting real-time unit operation data of a gas turbine compressor into a compressor state prediction model, calculating a compressor health index, and when the compressor health index is smaller than a preset value, scaling faults occur in the compressor;
and (4) combining the offline washing income of the unit and the actual maintenance plan of the power plant, making an offline washing strategy and performing offline washing on the compressor.
10. The method of claim 9, wherein the off-line wash gain calculation process comprises:
calculating unit additional cost and shutdown washing cost of the gas-steam combined cycle unit by combining historical data of a power plant; the unit additional cost is calculated by the formula:
Figure FDA0003347951360000031
wherein M isrunRepresents the unit operating parasitic loss, Δ W1Indicating a low power generation during the operating time after a combustion engine fouling failure, WrunIndicates the generating capacity of the unit after the scaling fault occurs, delta F1Indicating that the combustion engine consumes more natural gas fuel after scaling failure, CeIndicating the price of electricity on the Internet, CfRepresents the price of the fuel;
the calculation formula of the shutdown washing cost is as follows:
Mstop＝ΔW2Ce-ΔF2Cf+Mw，
wherein, Δ W2Indicating the amount of power, Δ F, of the combined cycle unit during shutdown2Representing the amount of fuel saved during shutdown, MwRepresents the cost of water washing;
calculating the offline washing profit by combining unit operation additional loss and shutdown washing cost;
the off-line washing yield calculation formula is as follows:
E＝Mrun×We-Mstop，
wherein E represents the water washing tapeNet income from WeAnd (4) the expected power generation amount of the combustion engine at the next stop time after the washing.
CN202111328290.9A 2021-11-10 2021-11-10 Gas turbine compressor detection and water washing optimization method Pending CN114021283A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111328290.9A CN114021283A (en) 2021-11-10 2021-11-10 Gas turbine compressor detection and water washing optimization method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111328290.9A CN114021283A (en) 2021-11-10 2021-11-10 Gas turbine compressor detection and water washing optimization method

Publications (1)

Publication Number Publication Date
CN114021283A true CN114021283A (en) 2022-02-08

Family

ID=80063461

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111328290.9A Pending CN114021283A (en) 2021-11-10 2021-11-10 Gas turbine compressor detection and water washing optimization method

Country Status (1)

Country Link
CN (1) CN114021283A (en)

Similar Documents

Publication Publication Date Title
CN106404403B (en) Method and system for analysis of a turbomachine
CN111539553B (en) Wind turbine generator fault early warning method based on SVR algorithm and off-peak degree
CN106779200A (en) Based on the Wind turbines trend prediction method for carrying out similarity in the historical data
CN104850904A (en) Analysis method for optimizing gas turbine overhaul and maintenance scheme
CN111648992B (en) Gas turbine compressor fault identification early warning method
CN113107626B (en) Load prediction method of combined cycle generator set based on multivariable LSTM
CN115294671A (en) Air compressor outlet pressure prediction method and prediction system
CN113515049A (en) Operation regulation and control system and method for gas-steam combined cycle generator set
CN115238573A (en) Hydroelectric generating set performance degradation trend prediction method and system considering working condition parameters
CN115471014A (en) Power grid enterprise electricity carbon consumption prediction method based on deep neural network
Zhu et al. Operational state assessment of wind turbine gearbox based on long short-term memory networks and fuzzy synthesis
CN115034485A (en) Wind power interval prediction method and device based on data space
CN111178602A (en) Circulating water loss prediction method based on support vector machine and neural network
CN113505909A (en) Error compensation method for short-term wind power trend prediction
CN111192163B (en) Generator reliability medium-short term prediction method based on wind turbine generator operating data
CN114021283A (en) Gas turbine compressor detection and water washing optimization method
CN116128136A (en) LSO-Catboost-based coal-fired power plant boiler NO X Emission prediction method
CN102231081B (en) Energy utilization state diagnosis method for process industrial equipment
CN115906437A (en) Fan state determination method, device, equipment and storage medium
CN112462708B (en) Remote diagnosis and optimized scheduling method and system for pump station
CN112348696B (en) BP neural network-based heating unit peak regulation upper limit evaluation method and system
CN113567164A (en) Systematic evaluation and prediction method for technical improvement demand of wind power plant
CN111461565A (en) Power supply side power generation performance evaluation method under power regulation
CN117893030B (en) Power system risk early warning method based on big data
Zhou et al. Prediction model of coal consumption based on random forest variable selection and random-grid hyperparametric optimization algorithm

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination