KR20220119939A - Apparatus and method for estimating main feed water flow rate - Google Patents

Abstract

An estimation apparatus is provided. The estimation apparatus may include: an acquisition unit that acquires state quantity data of a turbine cycle; and an estimation unit that estimates a main feed-water flow for each time domain using the state quantity data. The present invention is to provide the estimation apparatus and estimation method capable of accurately estimating the main feed-water flow of a turbine cycle.

Description

Apparatus and method for estimating main feed water flow rate

The present invention relates to an apparatus and method for estimating the main feedwater flow rate of a turbine cycle.

Thermal performance analysis to determine the amount of performance state of a turbine cycle is required for improved economic operation of a power plant.

In the thermal performance analysis, various factors such as aging of unit devices forming a turbine cycle may be reflected. The reflection of some factors may act as a factor that lowers the accuracy of performance evaluation of a specific unit device.

Korean Patent Publication No. 1185581 discloses a main water supply control system that controls the water level of a steam generator when switching the main water control valve of a nuclear power plant.

Korean Patent Publication No. 1185581

An object of the present invention is to provide an estimating apparatus and an estimating method capable of accurately estimating the main feedwater flow rate of a turbine cycle.

An estimating apparatus of the present invention includes: an acquisition unit for acquiring state quantity data of a turbine cycle; and an estimator for estimating a main feed-water flow for each time domain using the state quantity data.

The estimation method of the present invention may estimate the main feedwater flow rate for each time domain using a kernel regression model that receives a factor satisfying a set value for correlation among state quantity data of a turbine cycle.

The estimation method of the present invention includes: acquiring state quantity data of a turbine cycle; calculating a correlation coefficient of the state quantity data to give robustness; classifying a feature of the state quantity data by comparing the toughness; selecting a factor of an estimation model for estimating the main feedwater flow rate of the turbine cycle using the classified shape and classification table; It may include; estimating the main water flow rate for each time domain using the selected factor and the estimation model.

In the present invention, for useful and accurate performance analysis, a determination algorithm for each area of the main water flow rate was developed using performance data based on the industry standard ASME PTC, and an estimation algorithm for each area was developed. The estimation algorithm suggested shape classification based on the correlation of measured state quantities, and based on this, an estimation model was constructed using a kernel regression model, and compared with a neural network model to verify the superiority of support vector machine modeling. The shape classification and estimation model of main feedwater flow will contribute to improved performance analysis by providing accurate calibrated thermal performance analysis in turbine cycle.

1 is a schematic diagram showing an estimation apparatus of the present invention.
2 is a flowchart illustrating an estimation method of the present invention.
3 is a schematic diagram showing a power plant system diagram corresponding to a turbine cycle.
4 is a schematic diagram showing a calculation procedure of heat balance performance.
5 is a schematic diagram illustrating a procedure for calculating the operating thermal equilibrium of a turbine cycle.
6 is a schematic diagram showing the trend of the main feed water flow rate.
7 is a schematic diagram illustrating a process of generating an estimation model using training data in the model unit.
8 is a table showing correlation factors extracted using big data.
9 is a schematic diagram illustrating an algorithm of a support vector machine.
10 is a graph showing the calculation results of the support vector machine.
11 is a table showing error comparison results for support vector machines.
12 is a table showing error comparison results for a neural network (ANN) and a support vector machine.
13 is a diagram illustrating a computing device according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art to which the present invention pertains can easily implement them. However, the present invention may be embodied in several different forms and is not limited to the embodiments described herein. And in order to clearly explain the present invention in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

In this specification, duplicate descriptions of the same components will be omitted.

Also, in this specification, when a component is referred to as 'connected' or 'connected' to another component, it may be directly connected or connected to the other component, but other components in the middle It should be understood that there may be On the other hand, in the present specification, when it is mentioned that a certain element is 'directly connected' or 'directly connected' to another element, it should be understood that the other element does not exist in the middle.

In addition, the terms used herein are used only to describe specific embodiments, and are not intended to limit the present invention.

Also in this specification, the singular expression may include the plural expression unless the context clearly dictates otherwise.

Also, in this specification, terms such as 'include' or 'have' are only intended to designate that the features, numbers, steps, operations, components, parts, or combinations thereof described in the specification exist, and one or more It should be understood that the existence or addition of other features, numbers, steps, acts, components, parts, or combinations thereof, is not precluded in advance.

Also in this specification, the term 'and/or' includes a combination of a plurality of listed items or any of a plurality of listed items. In this specification, 'A or B' may include 'A', 'B', or 'both A and B'.

Also, in this specification, detailed descriptions of well-known functions and configurations that may obscure the gist of the present invention will be omitted.

Nuclear power generation thermal performance analysis system can calculate cycle thermal equilibrium, and performance calculation of unit equipment must be performed quickly and accurately. In other words, the performance analysis program embeds necessary calculation modules and information on the characteristics of unit devices, builds models of individual devices, and connects and integrates models to easily and accurately analyze the impact of performance changes of individual devices on the overall system. should be able These commercialized programs for major power generation facilities are PEPSE (Performance Evaluation of Power System Efficiency), which was developed for the main purpose of calculating the thermal equilibrium of boiler and steam turbine cycle, and it is possible to calculate performance under various operating conditions of gas turbine, combined cycle and cogeneration facilities. There are GR-PRO, GT-MASTER, STEAM-PRO, STEAM-MASTER similar to PEPSE, Thermoflex with high flexibility, and GateCycle, a package similar to Thermoflex.

The PEPSE model is used as basic data for the thermal performance module of Yeonggwang Units 3 and 4, and the performance calculation procedure and analysis system for thermal equilibrium analysis were developed by using it to develop predictive modules for unmeasured items and thermal performance. The turbine cycle performance analysis is based on ASME (American Society of Mechanical Engineers) PTC (Performance Test Code) 6, PTC 6A, and PTC 12 to calculate turbine cycle thermal balance and correction. The present invention intends to upgrade the performance of the thermal performance analysis system by developing an input verification model of the measurement data so that the correction performance can be accurately determined based on this procedure.

1 is a schematic diagram showing an estimation apparatus of the present invention.

As part of upgrading the performance of the thermal performance analysis system, the estimation apparatus shown in FIG. 1 includes an acquisition unit 110 , a classification unit 130 , a selection unit 150 , an estimation unit 170 , and a model unit 190 . may include.

The acquisition unit 110 may acquire state quantity data of the turbine cycle.

A turbine cycle corresponds to a schematic diagram of a power plant or a schematic diagram of a power generation system, and includes various unit devices required for power generation, such as boilers, turbines, condensers, low-pressure feedwater heaters, high-pressure feedwater heaters, and pipes and valves connecting each element. may include

The state quantity data may include a measured value measured for each unit device by using a sensor, and a calculated value calculated using the measured value. For example, the measured value may include flow rate, temperature, enthalpy, and the like. Measurements vary from plant to plant, but can be obtained in bulk of 1000 or more. The calculated value may include performance information such as turbine efficiency, a moisture separation rate of a moisture separator, a vacuum degree of a condenser, and a temperature performance of a feedwater heater.

The acquisition unit 110 may obtain the state quantity data from the sensor or may obtain the state quantity data by a user input using an input means such as a keyboard. The acquisition unit 110 may include various communication modules or interfaces for wired/wireless communication with sensors or input means.

The classification unit 130 may classify a feature based on the correlation of the state quantity data.

The model unit 190 may generate an estimation model for estimating a main feed-water flow using a kernel regression model based on the classified shape.

The classification unit 130 or the model unit 190 may use support vector machine modeling.

Specifically, the classification unit 130 may classify the shape of the main water flow rate by using a support vector machine (SVM) modeling. The model unit 190 may generate an estimated model using support vector machine modeling. In this case, the support vector machine modeling may use a linear function.

The estimator 170 may estimate a main feed-water flow for each time domain by using the state quantity data. As an example, the estimator 170 may estimate the main water flow rate for each time domain using the estimation model generated by the model unit 190 .

Since there are many types of state quantity data related to the turbine cycle, factors used in the estimation model may be limited to some state quantity data. In order to accurately estimate the main water flow rate, it is necessary to select the factors used in the estimation model. The selection unit 150 may be used for selection or selection of factors.

For example, the selector 150 may select a set number of factors of the estimation model from among the plurality of state quantity data. The selection unit 150 may select a factor of the estimation model from among the state quantity data based on the shape classification performed by the classification unit 130 .

The classification unit 130 may calculate correlation coefficients of all state quantity data to impart robustness to each state quantity data. The classification unit 130 may classify the shape by comparing the toughness of the state quantity data.

The selection unit 150 may determine the factors of the estimation model in groups by using the classification table of FIG. 8 created using the shapes classified by the classification unit 130 and big data. The selection unit 150 may select the state quantity data including the determined specific group as a factor of the estimation model.

The estimation method of the present invention may be performed by the estimation apparatus shown in FIG. 1 .

The estimation method of the present invention may estimate the main feedwater flow rate for each time domain using a kernel regression model that receives a factor satisfying a set value for correlation among state quantity data of a turbine cycle.

2 is a flowchart illustrating an estimation method of the present invention.

The estimation method shown in FIG. 2 may be performed by an estimation apparatus.

A step ( S510 ) of obtaining the state quantity data of the turbine cycle may be performed. The corresponding step may be performed by the acquisition unit 110 .

A step ( S520 ) of providing robustness by calculating a correlation coefficient of the state quantity data may be performed. The corresponding step may be performed by the classification unit 130 .

A step ( S530 ) of classifying features of the state quantity data by comparing the toughness may be performed. The corresponding step may be performed by the classification unit 130 .

Using the classified shape and classification table to select a factor of the estimation model for estimating the main feedwater flow rate of the turbine cycle (S540) may be performed. The corresponding step may be performed by the selector 150 .

The step of estimating the main water flow rate for each time domain using the selected factor and the estimation model (S550) may be performed. This step may be performed by the estimator 170 .

Looking at the schematic diagram of the power plant (FIG. 3), a boiler (steam generator) that generates the steam required by the turbine using the combustion heat of fuel, a turbine that converts the thermal energy of high-temperature and high-pressure steam from the boiler into work, and steam A cycle that generates generator output while circulating a condenser that recovers water by cooling and condensing, a low-pressure feedwater heater that converts the condensed water supplied from the condensate pump into steam by heating it again in a boiler, and a high-pressure feedwater heater that plays a similar role to the low-pressure feedwater heater Consists of. Heat balance analysis of such a turbine cycle is to analyze power generation efficiency, heat rate, and device performance by determining state quantities such as flow rate, temperature, and enthalpy. ~300 pressure and state quantity data in the turbine cycle are obtained. Although there is a difference for each power plant, about 1,500 state quantity data are obtained. Based on this data, the heat consumption rate and turbine cycle efficiency, which are representative performance indices of the turbine cycle, are obtained. . Then, for each unit device forming the turbine cycle, performance data such as the efficiency of the turbine, the moisture separation rate of the moisture separator, the vacuum degree of the condensers, and the temperature performance of the feedwater heater are obtained ( FIG. 4 ). The performance data may be calculated through state quantity data corresponding to the measured value.

In this specification, performance data may be included in state quantity data.

Thermal equilibrium calculation is based on the Law of Conservation of Mass and Energy in the turbine cycle to predict the performance variables (heat consumption rate and generator output) of the cycle, and at the same time, the flow fluid for all points in the cycle. determine the state quantities of It is composed of an optimization problem that minimizes the energy balance error by setting it as an objective function composed of 10 to 15 independent variables, and the turbine cycle is constructed using a network model of the maximum structure including all power plants. The algorithm for setting the next position in the space of the independent variable during the process of repeating the calculation is constructed by applying the heuristic provided by ASME PTC and linear and non-linear regression models for the turbine extractor pressure.

The performance variable of the turbine cycle calculated through the calculation of the operating thermal equilibrium represents the current performance of the turbine cycle, and it is difficult to understand the current performance of the turbine itself because the effects of the turbine itself as well as the accessories in the turbine cycle are compounded. have. Accordingly, a separate calibration calculation is required to determine the current performance of the turbine. Calibration calculation is an item that directly affects the turbine from the outside using the group I calibration, which calculates cycle thermal equilibrium while converting the performance of major devices in the cycle to the design condition standard, except for the turbine, and the calibration curve provided by the turbine manufacturer. It consists of a group II correction that compensates for each effect individually. In performance analysis, these corrections are handled using Curve Fitting instead of using the Design Performance module.

The thermal equilibrium convergence problem is an optimization problem using the end point of the turbine expansion line as an independent variable and the thermal equilibrium error or moisture separation rate error as an objective function. did. Iterative calculation of the low pressure turbine Expansion Line End Point (ELEP) is made, and if the difference between the calculated ELEP and the estimated ELEP at the new expansion line exceeds the ASME PTC allowable deviation of 0.1 Btu/lb, the calculated ELEP is used instead of the estimated ELEP to reconstruct the turbine expansion line and enters the iterative thermal equilibrium calculation process. The calculation procedure of such a thermal equilibrium system has already been verified a lot.

Figure 6 (a) is the actual change over a long period of time (about 160 cycles) of the main items of the generator output and main feed water flow rate among the data variables measured in the field for the performance analysis of the turbine cycle. This generator output is an important indicator to analyze turbine cycle performance along with turbine cycle efficiency. It is an important indicator to analyze the various losses that occur from the work supplied by the turbine to the turbine, such as bearing friction loss, and various losses that occur in the generator, such as the generator insulation loss. However, it is not easy to derive the result by the actual calculation formula because each term is composed of a complex formula. An abnormal measurement condition may occur due to environmental factors such as deterioration of field equipment. For example, a value larger than the actual value may be measured due to the accumulation of impurities generated in the orifice as the operation period elapses. As a result, heat output is over-calculated, and operation control variables must be adjusted according to safety standards, resulting in inefficient operation with heat output reduced from the maximum possible value. If it is determined that such a situation continues, the flow rate measurement is forcibly corrected by calibrating the instrument. After that, it is common to measure and analyze the performance according to the calibrated instrument. Figure 6 (a) shows the generator output according to the trend of the main feed water flow rate. Although it does not show any special flow, as mentioned above, there is a variation section due to forced correction due to measurement error. However, it can be seen that there is no clear change in the generator output trend, but it sharply increases in the area after the forced correction.

As shown in (b) of FIG. 6 , three factors having a high correlation with the main water flow rate were extracted, set as input factors of the neural network model, and after training, the estimated values were examined. The average error in the steady state is 1.52%, and the estimated value at which the maximum error occurs is 3.87%, which is within the data validity criteria of ASME PTC 6, but the overall average error including the forced correction section and the post-correction section is 3.16%, It was found that the data in the forced correction section except for some data did not satisfy the validity criteria, and this error data occurred even after the correction section. The reason for this is that the high error in the forced correction section is judged to occur because the generalization ability of the neural network according to the learning data in the steady state is weak. It is judged that this occurs because the correlation between and the estimated value is weakened.

Things to consider when selecting a correlation factor are: (1) if the number of factors increases, the calculation becomes complicated, slowing down learning and recognition, and (2) unnecessary factors make learning difficult. (3) Also, the precision of the model may deteriorate due to the addition of inappropriate factors.

In general, it is a step of refining complex data into simple data. It is a pre-processing process before the training step of the estimation model. Appropriate pruning of raw data should be performed.

Data reduction refers to the process of merging a large number of variable groups into a small number of variable groups. Principal component analysis, which is a commonly used factor reduction method, derives eigenvectors through correlation analysis of factors and maps large eigenvectors. This is a method of extracting reduced factors. However, when all data is used in the raw data, the precision of the model may deteriorate because inappropriate factors are also applied.

The selection of correlation factors can be largely divided into (1) a model-based method and (2) a model-free method. The model-based method is a method of finding an optimal set variable through learning using a selected factor, and is a process of selecting an optimal factor through an iterative process while changing the factor. And the method that is not model-based is a method that is performed based on the statistical relationship between the factors of the training data and the output.

In the present invention, the influence between variables was identified using the Pearson correlation coefficient in order to examine the strength of the correlation factor and the output.

The learning stage in the data-based model is the process of defining the traits or features of an object from the learning data composed of correlation factors. Since the vector is made by iterative calculation of the training data, the influence of the training data is large. The predicted value is estimated using the completed feature vector, and it is said that the generalization ability is excellent when the estimated value is similar to the actual value. The generalization ability for data-driven models is handled by feature vectors and depends on the training data. That is, it is judged that the error with the actual value can be reduced when the data of the correlation factor for the estimated value have a tendency similar to the characteristics of the training data.

In the present invention, the trend of the estimated value is investigated, and training data combined with a strong correlation factor therefor is to be constructed. The trend for the estimated value is determined by classifying the feature region for the estimated item using previously collected data, and comparing the excellent correlation factor of the estimated value with the strong correlation factor in the feature region to determine the feature region for the estimated value, that is, the shape. to understand (FIG. 7). The strong correlation factors are correlation factors of a feature region previously extracted using existing big data, as shown in FIG. 8 . These correlation factors have different factors for each shape and the number of factors having toughness. When the correlation coefficient is 0.7 or more, there are cases where the number of factors with toughness exceeds 40, and when the correlation coefficient is small, there are cases where there are fewer than 10. In the case of a large number of correlation factors, the operation time is long, so there have been many studies in which the number of input factors is determined in the neural network model. In the present invention, correlation coefficients of all factors are calculated using learning data, shapes are distinguished by comparing the toughness (0, 1) of factors, and groups (Low, Middle, High) are determined by rank of factors, and the corresponding group Sets the argument including up to as the input argument. For example, when the correlation coefficient for each factor of the learning data is calculated, toughness is given, and when compared with FIG. 8, it is determined as #5 shape, and when the arbitrary factor number is 3, the group included in the ranking is the Low group to be. In such a case, all factors including the Low group may be set as input factors.

The support vector machine aims to maximize the generalization ability by maximizing the margin of the criterion that divides the two classes. This means that the probability of misclassification is reduced when classifying any additionally generated patterns. Basically, the SVM classifies based on a hyper-plane for binary classification, and it can be expressed by Equation 1 below.

X is the feature vector, where X=(x ₁ ,...,x _d ) ^T , W is the normal vector, and b is the distance from the center to the hyperplane.

로 나타낼 수 있다.Here, it is possible to divide the feature vector as it is greater than or less than 0 based on the hyperplane d(X). The distance from any point x to the hyperplane is

can be expressed as

A vector that determines the size of the blank space from the hyperplane is called a support vector. Since the margin has a value twice the distance from the hyperplane to the nearest vector, it can be expressed as in Equation 2 below.

As a conditional optimization problem, a crystal hyperplane with maximum margin can be obtained.

의 조건일 때,

를 최대화하여야 한다.

When the condition of

should be maximized.

If expressed using the cost function J(.),

의 조건일 때,

을 최소화하는 것과 같다.

When the condition of

is equivalent to minimizing

Therefore, it is necessary to find W that minimizes J(W), which is obtained using a Lagrange multiplier. Put the Lagrange multiplier a _i in each conditional expression to obtain the vector.

라 표기하고, 라그랑제 함수를 L(.)로 두면, 다음의 수학식 3과 같이 정의될 수 있다.

, and the Lagrange function is set to L(.), it can be defined as in Equation 3 below.

The conditional optimization problem expressed by the Lagrange function can be solved using the Karush-Kuhn-Tucker (KKT) condition. According to the KKT condition, it can be expressed as Equation (4).

일 때,

을 만족하는 벡터가 서포터 벡터이다. KKT조건의 첫 번째와 두 번째 식을 라그랑제 함수에 대입하여 풀면 W와 b가 사라져 라그랑제 승수만 가지게 되는데 이를

로 두고 다시 정리하면,here,

when,

A vector that satisfies is a supporter vector. By substituting the first and second expressions of the KKT condition into the Lagrange function, W and b disappear and only the Lagrange multiplier is obtained.

If you put it back and rearrange it,

를 최대화시키는 것으로 정리될 수 있다.

It can be arranged by maximizing .

The order of finding the hyperplane with the maximum margin is to find the value that obtains the maximum value, and then obtain d(X) in the order of W and b. A dual optimization model and an optimal support vector machine regression model were applied, and a radial basis function kernel, a representative kernel function, was applied.

The support vector machine model sets an appropriate value within the range of , and the simplest method is to find a value that minimizes the prediction error of the training data used for training. However, it is known that such a setting causes an over-fitting problem, resulting in a decrease in predictive performance on untrained data. In order to prevent such a phenomenon and improve generalization performance, various techniques have been proposed. Representative methods include the VC dimension technique, Bayesian learning technique, and cross-validation technique. Among them, cross-validation is a relatively simple and excellent method. As for cross-validation, there are k-folds method, Leave-One-Out method, Bootstrap method, and the like. In the present invention, the optimal λ is selected by using the Leave-One-Out method among the cross-validation methods, and is performed by the following procedure shown in FIG. 9 .

(1) Initialize arbitrary λ as much as all training data.

(2) One data among the training data is attributed to sub-validation data, and the data is randomly rearranged using the remaining data as sub-learning data. In addition, λ inherited from the sub-learning data is constant, and λ inherited from the sub-verification data is calculated as a variable.

(3) Calculate y(x) and w ₀ according to the KKT condition and compare it with the setting error.

(4) If it is not satisfied due to the setting error, the λ inherited from the lower verification data is reset using the Simplex Method.

Here, if the inherited λ of the sub-verification data is satisfied, the sub-learning data and sub-verification data of item (2) are reset. Repeat (2) to (4) until λ correlated with all data is updated, and learning evaluation is performed through validation data.

The estimated value by the support vector machine model is shown in FIG. 10, and the average error is shown as shown in FIG. 11. All the kernel functions showed an error of less than 1% in general, and there were many values that accurately estimated the measured value (less than 0.1%). However, in some cases, a rather large error occurred. First, when implemented as a linear kernel function, the estimated value with an error rate of 1% or more in the #1 shape is 2 cycles, and 1 cycle appears in both shapes #2 and #5, and shapes #3 and #4 In , all estimated values were less than %1. In particular, it can be seen that the error rate in shape #2 has a relatively large value, so the average error rate is increased. , the average error rate for each shape was the smallest.

When implemented as a polynomial function, the estimated values with an error rate of 1% or more for shape #1 and #5 were 3 cycles, 2 cycles for shape #2, and 1 cycle for shape #3. As such, there were more estimated values with an error rate of 1% or more than other functions, so the overall average error rate was the highest, but the lowest error rate was shown in the starting section of shape #2.

When implemented as an RBF function, the estimated value with an error rate of 1% or more for shape #1 is 1 period, for shape #2 and shape #5, 2 cycles appeared, and for shape #3 and #4, all estimates Values were less than %1. In particular, it was found that the error rate in shape #2 had a relatively large value, so that the average error rate was increased.

In the support vector machine model, the linear function showed high reliability, so the generalization ability was the best.

12 is an example of an estimation model result using a support vector machine model, and some data with large errors are also included. Most of these values appeared in #2 shape and #5 shape in all functions. The measured value was 5.9786 Mkg/hr, but the estimated value was 5.91251 Mkg/hr (linear function), 5.9244 Mkg/hr ( polynomial function), 5.91382 Mkg/hr (RBF function), which is the region where the forced correction starts. Due to this large error, it was found that the average error was increased in the #2 shape, and the overall error was large in the #5 shape, indicating that the average error rate was increased.

In the present invention, validation was performed to obtain the accuracy of such measurement data, and reliable measurement data could be obtained.

1) Among the driving performance data for thermal equilibrium analysis, the performance factors judged to be correlated with the measured data were extracted and data pairs were constructed using the measured data and engineering information in the performance guide, and excellent correlation factors through correlation and data pairs were extracted.

2) In the existing big data, shapes could be separated by utilizing the characteristics of the data group, and a strong correlation factor for each shape was derived, and a window was set using the strong correlation factor suitable for the shape. In addition, the appropriate number of windows could be identified through the correlation coefficient for each data pair for each factor, and the extracted windows could be used to reconstruct the training data suitable for the estimation model.

3) In the support vector machine model, the linear function showed high reliability, indicating that the generalization ability was the best.

4) As shown in FIG. 12, the estimated value of the support vector machine function was more reliable than that of the neural network. In the collection section for each shape, the learning data for the previous shape is reflected in the estimated value, but it was found that the estimated value was derived by rapidly reducing the error as the window passed.

13 is a diagram illustrating a computing device according to an embodiment of the present invention. The computing device TN100 of FIG. 13 may be a device (eg, an estimation device, etc.) described herein.

In the embodiment of FIG. 13 , the computing device TN100 may include at least one processor TN110 , a transceiver device TN120 , and a memory TN130 . Also, the computing device TN100 may further include a storage device TN140 , an input interface device TN150 , an output interface device TN160 , and the like. Components included in the computing device TN100 may be connected by a bus TN170 to communicate with each other.

The processor TN110 may execute a program command stored in at least one of the memory TN130 and the storage device TN140. The processor TN110 may mean a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor on which methods according to an embodiment of the present invention are performed. The processor TN110 may be configured to implement procedures, functions, and methods described in connection with an embodiment of the present invention. The processor TN110 may control each component of the computing device TN100 .

Each of the memory TN130 and the storage device TN140 may store various information related to the operation of the processor TN110 . Each of the memory TN130 and the storage device TN140 may be configured as at least one of a volatile storage medium and a nonvolatile storage medium. For example, the memory TN130 may include at least one of a read only memory (ROM) and a random access memory (RAM).

The transceiver TN120 may transmit or receive a wired signal or a wireless signal. The transceiver TN120 may be connected to a network to perform communication.

On the other hand, the embodiment of the present invention is not implemented only through the apparatus and/or method described so far, and a program for realizing a function corresponding to the configuration of the embodiment of the present invention or a recording medium in which the program is recorded may be implemented. And, such an implementation can be easily implemented by those skilled in the art from the description of the above-described embodiments.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements by those skilled in the art using the basic concept of the present invention as defined in the following claims are also present. It belongs to the scope of the invention.

110...Acquisition Department 130...Classification Department
150...selection unit 170...estimation unit
190...Model Department

Claims (10)

an acquisition unit for acquiring state quantity data of a turbine cycle;
an estimator for estimating a main feed-water flow for each time domain using the state quantity data;
Estimation device comprising a.
According to claim 1,
A classification unit for classifying a feature based on the correlation of the state quantity data, a model unit for generating an estimation model for estimating the main feedwater flow rate using a kernel regression model based on the classified shape is provided,
The estimation unit is an estimation device for estimating the main water flow rate for each time domain using the estimation model.
3. The method of claim 2,
The classification unit is an estimation device for classifying the shape of the main feed water flow rate using support vector machine modeling.
3. The method of claim 2,
and the model unit generates the estimated model by using support vector machine modeling.
3. The method of claim 2,
The classification unit or the model unit uses support vector machine modeling,
The support vector machine modeling is an estimation device using a linear function.
According to claim 1,
The estimator estimates the flow rate of the main water supply using an estimation model,
An estimator having a selector configured to select a set number of factors of the estimation model from among the plurality of state quantity data.
According to claim 1,
The estimator estimates the flow rate of the main water supply using an estimation model,
An estimation apparatus provided with a classification unit for classifying a feature based on the correlation of the state quantity data, and a selection unit for selecting a factor of the estimation model from among the state quantity data based on the shape classification.
8. The method of claim 7,
The classification unit gives robustness by calculating the correlation coefficient of all state quantity data,
The classification unit classifies the shape by comparing the toughness of the state quantity data,
The selector determines the factors of the estimation model in groups using the classified shape and classification table,
The selection unit is an estimation device for selecting the state quantity data including the determined specific group as a factor of the estimation model.
An estimation method performed by an estimation device, comprising:
An estimation method for estimating the main feedwater flow rate for each time domain using a kernel regression model that receives factors that satisfy a set value for correlation among the state quantity data of the turbine cycle.
An estimation method performed by an estimation device, comprising:
acquiring state quantity data of the turbine cycle;
calculating a correlation coefficient of the state quantity data to give robustness;
classifying a feature of the state quantity data by comparing the toughness;
selecting a factor of an estimation model for estimating the main feedwater flow rate of the turbine cycle using the classified shape and classification table;
estimating the main water flow rate for each time domain using the selected factor and the estimation model;
Estimation method comprising

Images