KR101896115B1 - Method and apparatus for estimating corrosion rate of oil plant facility - Google Patents

Abstract

A method of predicting erosion rate of an oil plant, comprising: obtaining corrosion data of the oil plant; Removing an abnormal point from the obtained corrosion data; And applying the corrosion data from which the anomalous points have been removed to a kriging model to predict the corrosion rate. The step of removing the anomaly point includes applying piecewise linearization and time weighted averaging to the corrosion data in consideration of the corrosion rate and the measurement time of the corrosion parameters; And performing a Q-test on the applied results to remove erosion data that deviates from a predetermined threshold. According to the present invention, since the corrosion rate can be predicted in the refinery, it is possible to utilize the corrosion evaluation method, the material selection of the new chemical processing facility, the corrosion management, and the schedule of the facility diagnosis evaluation based on the corrosion prediction result. can do.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention [0001] The present invention relates to a method and apparatus for predicting corrosion rate of oil facilities,

The present invention is intended to predict the corrosion rate of oil installations used in refinery plants. In particular, the kriging model can be used to easily and accurately predict the corrosion rate of oil installations and to compare the predicted results with various techniques And to a technique for verifying the same.

Various types of oil facilities used in oil refineries are used. When these facilities are corroded, the operation of oil refineries is very limited. Particularly, since the upper part of the distillation tower is made of carbon steel and is vulnerable to corrosion and corrosion of the upper part of the distillation tower is caused by the operation of the apparatus without shut-down for several years due to the characteristics of the equipment industry, . Therefore, it is necessary to anticipate the extent of corrosion of these facilities and take action before the accident occurs.

However, corrosion is a very slow reaction, and the mechanism is complicated, which makes it difficult to apply the theoretical model to a real refinery.

Existing models for corrosion reactions include electrochemical models, mass transfer models, and empirical models. Among these, the electrochemical model focuses on the reaction rate to the corrosion reaction on the iron surface. In addition, the mass transfer model focuses on preventing the thin layer of iron oxide produced on the iron surface from spreading corrosion-causing substances to the iron surface. That is, the mass transfer model assumes that mass transfer determines the rate of corrosion. The empirical model is a model for predicting corrosion rate based on experimental results. Especially, in the empirical model, the corrosion rate or the degree of corrosion is indirectly predicted through the specimen or current measurement, and the corrosion is predicted by regression analysis of the linearity, polynomial, and theoretical expression by statistically processing the experimental data.

However, refining plants have difficulties in knowing the exact composition of the substance and contain various impurities, so it is difficult to experiment and there is a limitation in measuring all the necessary values in the formula. In addition, simple linear or polynomial regression analysis has limitations in representing complex corrosion mechanisms. In other words, it is difficult to predict the results because existing models contain many parameters that must be experimentally determined. In addition, the corrosion reaction is very slow, so it is not easy to apply to a refinery operating over a few years, since experiments often only take a few days to several weeks. Furthermore, the crude oil used in the refining process is a complex mixture in which the exact composition is unknown, and it is difficult to set the experimental conditions under the same conditions. It is more difficult to theoretically predict the mechanism of the corrosion reaction to such an extent that the corrosion rate is reduced by 90% compared with the case where only H ₂ S is added even when only CO ₂ is present.

According to Korean Patent No. 10-1199462 entitled " Apparatus for Predicting Carbon Steel Pipe Corrosion Rate, its Prediction Method and its Recording Medium "(hereinafter referred to as Prior Art 1), corrosion data of a specific metal structure collected for various corrosion times are statistically To provide a generalized empirical formula that can predict the amount of corrosion of a particular metallic structure over time. Prior Art 1 is a description of a method of predicting corrosion rate over time in a corrosive environment, which enables to plan and execute various measures such as repair, replacement, and disposal based on predicted corrosion amount. For this purpose, the prior art is to input the corrosion data of a specific metal structure collected for various corrosion exposure times at various corrosion times, to determine the maximum value for the input data and the minimum value of the COV (Α) and a dimensional parameter (β) are calculated for the corrosion data for each corrosion exposure time for the determined corrosion amount interval, and the calculated shape parameter (α) and the dimension parameter β) to calculate the linear exponential or polynomial expression for the exposure time to approximate each parameter (α, β), and use this to complete the time-dependent subcriteria prediction equation.

However, according to Prior Art 1, since corrosion data for each metal structure is required and different data are required for various corrosion time and corrosion exposure time, there is a disadvantage that the process of measuring the corrosion amount is complicated and difficult to be practically applied.

Therefore, there is a desperate need for a technique for relatively easily and accurately predicting the erosion rate of oil installations.

Prior Art 1: Korean Patent No. 10-1199462 entitled " Apparatus for Predicting Carbon Steel Pipe Corrosion Rate, Prediction Method Therefor and Recording Medium thereof "

[Thesis 1] A.G. Journel and C.J. Huijbregts, Mining Geostatistics, 5th Ed., Academic Press, London (1991). [Thesis 2] Y. Lang, S.E. Zitney and L.T. Biegler, Comput. Chem. Eng., 35, 1705 (2011). [Thesis 3] M.A. De Oliveira, O. Possamai, L.V.O.D. Valentina and C.A. Flesch, Expert Syst. Appl., 40, 272 (2013). [Paper 4] K. Movagharnejad, B. Mehdizadeh, M. Banihashemi and M.S. Kordkheili, Energy, 36, 3979 (2011).

It is an object of the present invention to provide a method for easily and accurately predicting the corrosion rate by predicting the corrosion rate using a kriging model used in geostatistics based on actual plant data.

It is also an object of the present invention to provide an apparatus for predicting a corrosion rate of a petroleum plant capable of verifying a predicted result by comparing and analyzing a predicted result with the results of multiple linear regression analysis and artificial neural network.

를 사용하여 부식 속도를 예측하는 단계를 더 포함하며, 여기에서, z^*는 예측하려는 부식 속도, z _i 는 예측하려는 환경과 유사한 환경에서 측정된 부식 데이터, n은 상기 유사한 환경에서 측정된 부식 데이터의 수, λ_i는 z _i 에 적용되는 가중치를 나타낸다. 또한, 본 발명에 의한 유류 설비의 부식 속도 예측 방법은 예측된 결과를 다중 선형 회귀 분석 결과 및 인공 신경망 분석 결과 중 적어도 하나와 함께 제공하는 단계를 더 포함한다. 상기 유류 설비는 정류 공장의 증류탑 상부일 수 있다.According to an aspect of the present invention, there is provided a method for predicting a corrosion rate of an oil facility, the method comprising: acquiring corrosion data of the oil facility; Removing an abnormal point from the obtained corrosion data; And applying the corrosion data from which the anomalous points have been removed to a kriging model to predict the corrosion rate. The step of removing the anomaly point includes applying piecewise linearization and time weighted averaging to the corrosion data in consideration of the corrosion rate and the measurement time of the corrosion parameters; And performing a Q-test on the applied result to remove corrosion data that deviates from a predetermined threshold. The step of predicting the erosion rate comprises: obtaining corrosion parameters to be applied to the kriging model; Analyzing the correlation between the obtained corrosion parameter and the corrosion rate with a significant probability; And lowering the dimension of the hyperplane of the kriging model by removing the low-significance corrosion parameter according to the analysis result. The step of predicting the corrosion rate comprises:

Using further comprising the step of predicting the erosion rate, where, z ^* is corrosive to predict speed, z _i is the corrosion data obtained in similar to predict environmental environment, n is the corrosion data obtained in the similar environment And? _I represents a weight applied to z _i . The method of predicting erosion rate of the oil plant according to the present invention further includes the step of providing predicted results together with at least one of the results of the polynomial regression analysis and the ANN analysis. The oil facility may be above the distillation tower of the rectification plant.

를 사용하여 부식 속도를 예측하고, 여기에서, z^*는 예측하려는 부식 속도, z _i 는 예측하려는 환경과 유사한 환경에서 측정된 부식 데이터, n은 상기 유사한 환경에서 측정된 부식 데이터의 수, λ_i는 z _i 에 적용되는 가중치를 나타낸다. 본 발명의 유류 설비의 부식 속도 예측 장치는 예측된 결과를 다중 선형 회귀 분석 결과 및 인공 신경망 분석 결과 중 적어도 하나와 함께 제공하는 결과 제공부를 더 포함하며, 상기 유류 설비는 정류 공장의 증류탑 상부일 수 있다.
According to another aspect of the present invention, there is provided an apparatus for predicting a corrosion rate of a oil facility, comprising: a corrosion data obtaining unit obtaining corrosion data of the oil facility; An abnormal point removing unit for removing an abnormal point from the obtained corrosion data; And a corrosion rate predicting unit for predicting a corrosion rate by applying corrosion data from which the anomalous points are removed to a kriging model. The anomaly removing unit applies piecewise linearization and time-weighted averaging to the corrosion data in consideration of the corrosion rate and the measurement time of the corrosion parameters, performs a Q-test on the result of the application, And is configured to remove corrosion data. The corrosion rate predicting unit obtains the corrosion parameters to be applied to the kriging model, analyzes the correlation between the obtained corrosion parameter and the corrosion rate with a significant probability, removes the corrosion parameter having a low degree of significance according to the analysis result, It is configured to lower the dimension of the hyperplane. The corrosion rate predicting unit may include:

Where z ^* is the corrosion rate to be predicted, z _i is the corrosion data measured in an environment similar to the environment to be predicted, n is the number of corrosion data measured in the similar environment, and λ _i Represents a weight applied to z _i . The apparatus for predicting erosion rate of a plant according to the present invention further comprises a result providing unit for providing a predicted result together with at least one of a result of a polynomial regression analysis and an artificial neural network analysis, have.

According to the present invention, since the corrosion rate can be predicted in the refinery, it is possible to utilize the corrosion evaluation method, the material selection of the new chemical processing facility, the corrosion management, and the schedule of the facility diagnosis evaluation based on the corrosion prediction result. can do.

In addition, according to the present invention, since the kriging model, which is a statistical technique, is used to predict the corrosion rate in an oil refinery, corrosion prediction suitable for an individual refinery can be made.

Further, according to the present invention, it is possible to obtain more accurate results by comparing general statistical methods such as multiple linear regression analysis with the prediction results of the present invention, and ultimately, to operate safely and continuously for a longer time Thereby maximizing profits.

FIG. 1 is a flowchart schematically illustrating a corrosion rate predicting method of an oil plant according to an embodiment of the present invention.
FIG. 2 is a graph showing the segmental linearization used in the present invention. FIG.
3 is a graph illustrating a kriging model used in the present invention.
4 is a graph showing an experimental variogram and a theoretical variogram in the case of six corrosion variables.
FIG. 5 is a graph showing the prediction results in the case where the corrosion parameter is six.
6 is a graph showing an experimental variogram and a theoretical variogram in the case of four corrosion variables.
FIG. 7 is a graph showing prediction results in the case where four corrosion variables are used.
8 is a block diagram schematically showing an apparatus for predicting corrosion rate of a oil plant according to another embodiment of the present invention.

In order to fully understand the present invention, operational advantages of the present invention, and objects achieved by the practice of the present invention, reference should be made to the accompanying drawings and the accompanying drawings which illustrate preferred embodiments of the present invention.

Hereinafter, the present invention will be described in detail with reference to the preferred embodiments of the present invention with reference to the accompanying drawings. However, the present invention can be implemented in various different forms, and is not limited to the embodiments described. In order to clearly describe the present invention, parts that are not related to the description are omitted, and the same reference numerals in the drawings denote the same members.

FIG. 1 is a flowchart schematically illustrating a corrosion rate predicting method of an oil plant according to an embodiment of the present invention.

First, in order to predict the corrosion at the refinery, the past operation data on corrosion factors measurable in the refinery and the corrosion rate are inputted (S110).

Examples of corrosive factors measurable in refineries are pH, Cl ^- , Fe ²⁺ , H ₂ S concentration, NH ₃ concentration, and flow rate. Corrosion by CO2 and H ₂ S is a major corrosion factor in the refinery due to major corrosion reactions. There are other factors affecting various corrosion, but the precise erosion mechanism is not clarified, so the application of the theoretical model is limited. In addition, there are restrictions on the number and types of variables that can be measured in a real refinery. Different refinery plants have different operating environments and require a corrosion prediction model for each refinery. Therefore, in the present invention, corrosion rate is predicted using actual measurement data.

Once the data is obtained, the abnormal point is removed from the obtained corrosion data (S130).

In order to remove the anomalous points, a Q-test must be performed. In the present invention, piecewise linearization and time-weighted averaging techniques can be applied before performing the Q-test.

When the abnormal point is removed as described above, a corrosion parameter to which the remaining data is to be applied is obtained (S140). In the present invention, six parameters of H ₂ S, Cl ^- , Fe ²⁺ , NH ₃ , pH, and flow rate were used as the corrosion parameters. However, the present invention is not limited thereto and various corrosions It is of course possible to introduce variables.

Once the corrosion parameters are determined, the correlation between the obtained corrosion parameters and the corrosion rate is analyzed with a significant probability, and the corrosion parameters with low significance are removed (S160). Removal of the corrosion parameters in this way can reduce the dimension of the hyperplane of the kriging model.

If the dimension of the hyperplane is determined, the corrosion rate is estimated by applying the corrosion data from which the abnormal point is removed to the kriging model (S170). As described above, in the present invention, the operation data of the refinery is input using the Kriging model, which is an advanced statistical method, and the corrosion rate suitable for the operation environment of the refinery is predicted. The kriging model is generally used in a spatial coordinate system, but in the present invention, this model is applied to a hyperplane of a non-spatial coordinate system made up of corrosion factors to perform corrosion prediction.

When the corrosion rate is predicted, a polynomial linear regression analysis and an artificial neural network analysis are performed, and the result is provided to the user along with the predicted result using the kriging model (S190).

That is, in the present invention, a hyperplane composed of the non-spatial coordinates of the corrosion parameters (H ₂ S, Cl ^- , Fe ²⁺ , NH ₃ , pH and flow rate) is constituted and the number of corrosion parameters is reduced While reducing the burden, predict corrosion rate more easily than in the prior art.

Hereinafter, the present invention will be described in more detail with reference to the related drawings.

First, the operation data is measured for several years. A set of variables as described above, is pH, Cl ^- includes, Fe ^2+, H ₂ S, NH _3, and flow rate, and may include also other parameters. The rate of corrosion is measured irregularly weekly, but all other variables can be measured daily irregularly.

To control the six corrosion parameters and corrosion rate, piecewise linearization is used, and a time-weighted average technique is used to obtain the mean value of each interval.

FIG. 2 is a graph showing the segmental linearization used in the present invention. FIG.

As shown in FIG. 2, even when the measurement time is different, the variables can be linearly processed through the segmented linearization.

In order to increase the accuracy of the prediction and to facilitate processing, certain outliners may be excluded. For example, values outside the 95% confidence may be excluded from the driving data. In the present specification, 35 data out of 42 measured data were used as learning samples, and the remaining 7 data were used as verification samples. However, this is for the purpose of explanation of the present invention, and it goes without saying that fewer or more data can be used.

3 is a graph illustrating a kriging model used in the present invention.

As shown in FIG. 3, the kriging model is a technique for predicting the value of a point of interest using a linear combination of values around the adjacent points. The kriging model can be considered as an alternative to regression analysis. Unlike other statistical models, the kriging model predicts values from measured values using a spatial correlation function, variogram, which is a function that expresses the spatial association of observations commonly used in geostatistics , A covariance value is calculated using a variogram.

In general, the Kriging model is a technique used in geostatistics. It is a technique for predicting linear combinations of values at points close to the point to be predicted on the spatial coordinates. In addition to geology, geography, and meteorology, It is also applied to fields related to spatial information. For example, further discussion of the kriging model can be found in [Thesis 1] through [Thesis 4] cited herein. These papers are incorporated herein by reference in their entirety.

Can be expressed as Equation (1) of the kriging model.

In Equation (1), z ^* is a corrosion rate to be estimated, z _i is a value at an _i- th point, λ _i is a weighting parameter, and n is the number of adjacent points. Several equations must be solved to determine the weighting function. The methods for determining the weighting function may be implemented using various kriging techniques such as simple kriging, ordinary kriging, block kriging, co-kriging, or universal kriging. It depends on which model you use. It should be understood that the kriging model that may be used does not limit the invention and that all techniques known before the invention can be used.

Thus, the corrosion rate prediction using the Kriging model can be obtained by linear combination of the corrosion rate values in the similar environment among the inputted historical data.

For convenience of explanation, a simple kriging model among various methods of simple kriging is expressed by Equation (2).

In Equation (2), C (i, j) is the covariance value at the distance between the adjacent points i and j, and c (i) is the covariance value at the distance between the target point and the adjacent point i. These covariance values can be obtained using a covariance function derived from the variogram as in Equation (3).

In Equation (3), h is the separation distance between two points, and? Is the variogram. Sill is the covariance value at a distance of zero or the maximum variation of a pair of points. That is, this is the upper limit of the variogram that grows to an infinite distance due to the covariance value C (h), which decreases to zero as the distance increases. Since the weighted parameter is is obtained from the variogram, the accuracy of the variogram greatly affects the predicted performance.

The variogram describes the correlation of data as distance. The experimental variogram is half of the mean square difference between pairs of points with distance h, as shown in equation (4).

Is the number of pairs, z is the dependent variable value at some point, and (i, j) is a pair of points spaced by distance h. Theoretical variogram can be obtained from an experimental variogram.

As an example, although the present invention is described herein using an orininal kriging as in Equation 5, it is clear that the present invention is not limited to this technique, and that various kriging techniques can be employed.

Also, in this specification, as described above, a kriging model is applied using a hyperplane made of a corrosion parameter.

The accuracy of the predicted corrosion rate with the Kriging technique is compared with techniques such as MLR and ANN.

Using multiple linear regression analysis, it is easy to analyze the effect of each independent variable on the dependent variable and to recognize the relationship between the independent variable and the dependent variable. However, multiple linear regression analysis is not sufficient to predict natural phenomena with nonlinear and complex behavior.

Artificial neural network technique is an artificial intelligence method that simulates the neural network of the human brain. Because these methods are robust in solving complex nonlinear systems, they are widely used in energy consumption, oil prices, thermal conductivity, and process optimization. However, artificial neural networks have some disadvantages, the first being the problem of over-fitting. The problem is that the data fit well in the sample interval but do not show good prediction performance in other regions. Another problem is that the degree of freedom is too high and there are too many optional variables.

Therefore, it is necessary to reduce the number of variables. The first reason is that using all six corrosion variables to predict corrosion rate does not yield satisfactory predictions. Some variables may have less impact on corrosion rate than others, but it is not necessary to model them involuntarily.

In the case of the kriging model, the 42 data sets were applied to the 6-dimensional hyperplane, but the higher the dimension, the more dimensional data may be not suitable because more sample data is needed to find the proper variogram. Therefore, in the present invention, the hyperplane is constructed by reducing the number of data sets by two. The number of variables used increases as the reliability increases. In general, it can be seen that satisfactory results can be obtained by measuring the confidence interval with the results obtained by constructing the 4-dimensional hyperplane.

Six corrosion parameters were developed and validated using 35 data sets. All variables are normalized and applied to the model.

Figure 4 shows an experimental and theoretical variogram.

The spacing in the graph represents the similarity of the six corrosion variables between the training sample points. Since only 35 data sets are used, there is no variogram experiment in the range of 0 to 0.2. Therefore, the absence of a variogram experiment in this interval indicates that the theoretical variogram is not correct. Figure 4 shows that there is a spatial relationship (this is called a range) in the separation distance on the hyperplane until a separation distance of 0.6 is reached. The variogram values outside this range have no relation to the separation distance. Estimating parameters using an optimization tool yields a theoretical variogram. An exponential model is used as the theoretical variogram, which has a range of 0.574 and sill of 0.087, respectively.

The mean square error (MSE) of the Kriging model using the closest 3, 5, and 7 points and all points within the validation interval are 0.541, 0.625, 0.572, and 0.351, respectively. The results from the nearest 3, 5, and 7 points are worse because there is no near point within a distance of 0.2. As shown in (a) of FIG. 5, unlike the other models, the MSE in the learning section is exactly zero because the measured and predicted values in the corresponding section exactly match in the kriging model.

Figure 5 (b) shows the MLR model results. The MSE of the verification samples is 0.420 and the MSE of the training samples is 1.126. Although the set of parameters is determined by the training samples, the results of the verification interval show better results than the MSE. The reason is due to the range of corrosion rates: 0.456 MPY to 4.867 MPY in the learning period, but 1.217 MPY to 2.920 MPY in the verification period. Furthermore, trends in measured values are not well visible in the prediction results.

Forty different types of artificial neural network models consist of concealed neurons, presence / absence of bias neurons, and a combination of epochs. Table 1 shows the results of the artificial neural network model for MSE.

As the number of hidden neurons increases, the MSE in the learning zone has a tendency to decrease in all cases. On the other hand, up to four to six concealed neurons, the MSE in the probe area tends to increase, and the more concealed neurons become inconsistent in the tendency. However, as the number of concealed neurons increases, more parameters must be determined. Models with bias neurons typically exhibit poor predictive performance than models without bias neurons. It also affects the number and alignment of epochs.

The most desirable artificial neural network model is a model with 3 hidden neurons and 10,000 epochs without bias neurons, which is shown in Figure 5 (c). Unlike the multiple linear regression model, it provides a good predictor of the measured corrosion rate, and compared to polynomial regression analysis, the MSE value is only half that of the given learning and verification periods.

Compared with the multiple linear regression analysis model and the artificial neural network model of FIG. 5, the kriging model is similar to the multiple linear regression model. Also, looking at the Kriging model, there is a large difference between the measured and predicted values at a small corrosion rate of about 1.5 MPY in the validation period. Therefore, an artificial neural network model shows optimal results when there are six corrosion variables. However, the kriging model can be improved if more accurate variograms are obtained. To do this, you need to use more data sets or reduce the number of variables.

Thus, if the six corrosion parameters are reduced to four, a significance probability of less than 5% is obtained. Of the six variables, NH ₃ and Fe ²⁺ can be removed. The remaining variables, H, Cl ^- , H ₂ S, and flow rates are used.

The experimental variogram and the theoretical variogram for the Kriging model when four variables are used are shown in FIG. In Fig. 6, adjacent points exist within a distance of 0.2, and the number of points in the range is increased as compared with the case where the number of points is 6. As a result, the experimental variogram exhibits an apparent increasing tendency within the range interval as the separation distance increases. The Gaussian model, the theoretical variogram, fits well with the experimental variogram. The parameter set of the model is obtained through parameter estimation. The range increased from 0.574 to 0.585, compared to the use of six corrosion variables. The sill also increased from 0.087 to 0.116.

Figure 7 (a) shows the optimal kriging model among the kriging models with four variables. The MSEs associated with the nearest 3, 5, and 7 points are 0.377, 0.183, and 0.375, respectively. The Kriging model using the closest 5 points shows the lowest MSE value. As can be seen in Figures 5 (a) and 7 (a), the range of predicted values in the verification interval is larger than when the number of variables is six. Therefore, this kriging model improves predictive performance and is highly reliable even at low rates of corrosion.

The multiple linear regression model with four variables shows MSEs of 0.448 and 1.126 in the verification and learning zones, respectively. The reason that the predictive performance of a model with four variables is smaller than the model with six variables is that the set of removed variables was determined by applying the significance probability in order to obtain a data set with little correlation with the corrosion rate .

The results of the artificial neural network model with four corrosion variables are shown in Table 2. Compared to the case of six variables, four variables show very improved results. However, looking at the MSE, the optimal result is similar to the case of six variables, with a difference of 0.209: 0.232.

Table 3 shows the minimum MSE value for each model in each case. The Kriging model, which has four corrosion variables and uses the closest five points, has the best results.

8 is a block diagram schematically showing an apparatus for predicting corrosion rate of a oil plant according to another embodiment of the present invention.

The apparatus for predicting erosion rate of a plant according to another aspect of the present invention includes a corrosion data obtaining unit 810, an abnormal point removing unit 820, a Q-test unit 840, a central control unit 850, A speed predicting unit 860, a correlation analyzing unit 880, and a result providing unit 890.

The corrosion data acquisition section 810 acquires the measured corrosion data for the corrosion of the oil plant. As described above, the corrosion data may be data measured at various time intervals.

Then, the anomaly removing unit 820 removes the anomaly point from the acquired corrosion data. To this end, the anomaly removing unit 820 may apply the segmented linearization and time weighted averaging techniques to the corrosion data taking into account the corrosion rate and the measurement time of the corrosion parameters. The processed corrosion data is provided to the Q-test section 840, which performs Q-test on the data to remove corrosion data that deviates from a predetermined threshold.

The central control unit 850 controls the operation of the components of the corrosion rate prediction apparatus 800 of the oil facility.

If the abnormal point is removed, the corrosion rate predicting unit 860 predicts the corrosion rate by applying the corrosion data from which the abnormal point is removed to the kriging model. In order to facilitate the prediction of the corrosion rate, the correlation analyzing unit 880 analyzes the correlation between the corrosion rate to be applied to the kriging model and the corrosion rate with the significance probability, The dimension of the hyperplane of the kriging model can be lowered. When the dimension of the hyperplane is lowered in this manner, there is an advantage that the corrosion rate can be predicted more quickly and easily.

When the erosion rate is predicted, the result provider 890 provides the predicted result to the user along with the polynomial linear regression analysis result and the artificial neural network analysis result.

As described above, in the present invention, 35 data out of 42 data whose anomalous points have been removed are used for learning and 7 data are used for verification. As a result, Kriging model model is more correct than simple polynomial regression analysis but mean square error is worse than artificial neural network. In order to improve accuracy, linear correlation of corrosion rate and corrosion rate is analyzed with significance, The Kriging model was applied to the 4 - dimensional hyperplane where Fe ²⁺ and NH ₃ were removed.

As a result of the reduction of dimensions, 35 data can better represent the hyperplane, so the predictions of the Kriging model have been improved significantly and more accurate than multilinear regression analysis and artificial neural network. Therefore, it can be confirmed that when the kriging model is applied to the non-spatial coordinate system, the number of variables is smaller than the number of data, which is effective when the accurate prediction is difficult.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is evident that many alternatives, modifications and variations will be apparent to those skilled in the art.

In addition, the method according to the present invention can be embodied as computer-readable code on a computer-readable recording medium. A computer-readable recording medium may include any type of recording device that stores data that can be read by a computer system. Examples of the computer-readable recording medium include a ROM, a RAM, a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like, and may be implemented in the form of a carrier wave (for example, transmission via the Internet) . The computer readable recording medium may also store computer readable code that may be executed in a distributed manner by a distributed computer system connected to the network.

As used herein, the singular " include " should be understood to include a plurality of representations unless the context clearly dictates otherwise, and the terms "comprises & , Parts or combinations thereof, and does not preclude the presence or addition of one or more other features, integers, steps, components, components, or combinations thereof. The terms "part", "unit", "module", "block", and the like described in the specification mean units for processing at least one function or operation, And a combination of software.

Therefore, it should be understood that the present invention and the drawings attached hereto are only a part of the technical idea included in the present invention, and that those skilled in the art will readily understand the technical ideas included in the specification and drawings of the present invention It is obvious that all the variations and concrete examples that can be deduced are included in the scope of the present invention.

The present invention can be used to predict the erosion rate of oil installations in refineries.

800 Corrosion rate prediction device of oil facility
810 Corrosion data acquisition unit 820 Abnormal point removal
840 Q-test unit 850 Central control unit
860 Corrosion Rate Prediction Unit 880 Correlation Analysis Unit
890 Results provided

Claims (12)

An apparatus for predicting erosion rate of an oil plant,
A corrosion data obtaining unit for obtaining corrosion data of the oil facility;
An abnormal point removing unit for removing an abnormal point from the obtained corrosion data; And
And a corrosion rate predicting unit for predicting a corrosion rate by applying the corrosion data from which the abnormal point is removed to a kriging model,
Wherein the abnormal point removing unit comprises:
Applying piecewise linearization and time weighted averaging techniques to the corrosion data taking into account the corrosion rate and the measurement time of the corrosion parameters,
Wherein the corrosion rate estimator is configured to perform a Q-test on the applied result to remove corrosion data that deviates from a predetermined threshold,
Obtaining the corrosion parameters to be applied to the kriging model,
The correlation between the obtained corrosion parameters and the corrosion rate is analyzed with significance probability,
According to the results of the analysis, it is constructed to lower the dimension of the hyperplane of the kriging model by removing the low -
The corrosion rate predicting unit may include:

The corrosion rate is predicted by using < RTI ID = 0.0 >
Where z ^* is the corrosion rate to be predicted, z _i is the corrosion data measured in an environment similar to the environment to be predicted, n is the number of corrosion data measured in the similar environment, and λ _i is the weight applied to z _i ,
Further comprising a result providing unit for providing the predicted result together with at least one of the result of the multi-linear regression analysis and the result of the artificial neural network analysis, and the oil facility corresponds to the upper part of the distillation tower of the rectification plant. Device.
A method for predicting corrosion rate of an oil plant using the corrosion rate predicting device of the oil plant according to claim 1,
Obtaining corrosion data of the oil facility;
Removing an abnormal point from the obtained corrosion data; And
And a step of applying the corrosion data from which the abnormal point is removed to the kriging model to predict the corrosion rate,
The step of removing the abnormal point may include:
Applying piecewise linearization and time weighted averaging techniques to the corrosion data in consideration of the corrosion rate and the measurement time of the corrosion parameters; And
Performing a Q-test on the applied result to remove corrosion data that deviates from a predetermined threshold,
The step of predicting the corrosion rate comprises:
Obtaining corrosion parameters to be applied to the kriging model;
Analyzing the correlation between the obtained corrosion parameter and the corrosion rate with a significant probability; And
And a step of lowering the dimension of the hyperplane of the kriging model by removing the corrosion parameter having a low degree of significance according to the analysis result,
The step of predicting the corrosion rate comprises:

Further comprising the step of predicting the rate of corrosion,
Where z ^* is the corrosion rate to be predicted, z _i is the corrosion data measured in an environment similar to the environment to be predicted, n is the number of corrosion data measured in the similar environment, and λ _i is the weight applied to z _i , And providing the predicted result together with at least one of a polynomial regression analysis result and an artificial neural network analysis result,
Wherein the oil facility is above the distillation tower of the rectification plant. delete delete delete delete delete delete delete delete delete delete

Images