CN103337000B

CN103337000B - A kind of safe monitoring and pre-alarming method of oil-gas gathering and transportation system

Info

Publication number: CN103337000B
Application number: CN201310301350.7A
Authority: CN
Inventors: 马猛; 陈健飞; 江文军; 王安泉; 彭星来; 宫俊峰; 冯国栋; 吕德东; 盛华
Original assignee: China Petroleum and Chemical Corp; Technology Inspection Center of Sinopec Shengli Oilfield Co
Current assignee: China Petroleum and Chemical Corp; Technology Inspection Center of Sinopec Shengli Oilfield Co
Priority date: 2013-07-18
Filing date: 2013-07-18
Publication date: 2016-06-15
Anticipated expiration: 2033-07-18
Also published as: CN103337000A

Abstract

A kind of safe monitoring and pre-alarming method of oil-gas gathering and transportation system, comprises the modeling process of GMDH algorithm and the algorithm application flow in oil-gas gathering and transportation system safe early warning system; Namely GMDH algorithm model is first set up, then according to GMDH algorithm model, have chosen the input of the working parameter of a certain period oil-gas gathering and transportation system as model data, the situation in future is predicted, obtain final early warning result, and carry out early warning measure in advance and prepare. Compared with prior art, invention has the following advantages: modeler does not exist bias in modeling process, computer is the optimum variable of Confirming model and structure automatically, ensures selecting without artificial subjective factor of model like this, such that it is able to improve the precision of data prediction.

Description

A kind of safe monitoring and pre-alarming method of oil-gas gathering and transportation system

Technical field

Present method relates to a kind of method of data prediction, a kind of Forecasting Methodology being especially applied in oil-gas gathering and transportation system safe early warning system.

Background technology

The process characteristic of oil-gas gathering and transportation system self determines that it is a hazard level height, the place that is prone to accidents. Along with the increase of oilfield transportation system scale, the contact between its structure and each subsystem inner is also more and more complicated, will utilize now some safety technique means that it carries out effective safe early warning also more and more difficult.

On the other hand; along with the development of science and technology; oil-gas gathering and transportation system engineering is constantly comprehensively developed to systems engineering and concurrent engineering direction by Engineering Speciality; with above-mentioned traditional security monitoring with evaluate theoretical and method analyze these huge and Iarge-scale system of complexity time, the complicated problem such as usually can run into that changing environment, higher-dimension high-order, fault data sample be incomplete and information is few.

Therefore, " information island " and " data rich and lack of knowledge and information statement are not directly perceived " problem is the bottleneck that oilfield transportation system is carried out effective and safe management by restriction. In the process of construction of digitizing oil field, raw security testing and monitoring data volume is big and mixes, the security how field produces is managed by decision-making level by the data of magnanimity carries out interpretation and application, thus carry out rational security and emergency response, effectively control examination and repair cost, reduce damage sequence loss, and then improve the intrinsic safety in oil field, be problem demanding prompt solution in the information-based process of current safe monitoring index system in the world.

Traditional safe early warning method has multiple, such as gray theory method, Time Series AR modelling, SVMs method etc.The fitting precision of these Forecasting Methodologies is higher, but for reality engineering predict time, be easy to that bigger error occurs. Because traditional Forecasting Methodology is that the dynamic parameter of forecasting object is regarded as static parameter, modeler needs the independent variable(s) considered in function and dependent variable and parameter and coefficient in modeling process, and these often modeler according to consulting relevant data and data obtain. So traditional modeling method all with modeler subjective thought inside, this just causes the practicality of institute's established model to reduce, and the precision predicted the outcome also reduces accordingly.

Summary of the invention

For solving the deficiencies in the prior art, the present invention provides a kind of safe monitoring and pre-alarming method of the oil-gas gathering and transportation system based on GMDH algorithm, predicts the outcome the low defect of precision to solve prior art.

The present invention comprises the modeling process of GMDH algorithm and the algorithm application flow in oil-gas gathering and transportation system safe early warning system;

First setting up GMDH algorithm model, modeling process comprises the following steps:

If input variable is x₁,x₂,…,x_n, output variable is y, sets up a high-order regression model:

Y=f(x₁,x₂,…,x_n) (1)

Wherein Y is dependent variable, and x is independent variable(s);

If x is the input of system, y is the output of system, sample count as n, the number of sample is m, the data of sample is divided into learning sample and test samples, if the number of learning sample is nt, the input and output matrix of variables W of system is as follows:

W = [\begin{matrix} y_{1} & x_{11} & x_{12} & \cdot \cdot \cdot & x_{1 m} \\ y_{2} & x_{21} & x_{22} & \cdot \cdot \cdot & x_{2 m} \\ \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot \\ y_{nt} & x_{nt, 1} & x_{nt, 2} & \cdot \cdot \cdot & x_{nt, m} \\ \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot \\ y_{n} & x_{n, 1} & x_{n, 2} & \cdot \cdot \cdot & x_{n, m} \end{matrix}] - - - (2)

Next step regression Calculation;

In formula (2) matrix, get any Two Variables x of input variable in learning sample_i,x_jPolynomial regression is carried out by following equation with output variable y:

y_{ij} = A + {Bx}_{i} + {Cx}_{j} + {Dx}_{i}^{2} + {Ex}_{j}^{2} + {Fx}_{i} x_{j}, i &NotEqual; j, 1 \leq (i, j) \leq m - - - (3)

Wherein the learning sample of A, B, C, D, E and F matrix (5) calculates according to method of least squares; After recurrence, it is possible to produceThe recurrence polynomial expression of individual higher-order, if:

z = a_{0} + b_{0} x_{i} + c_{0} x_{j} + d_{0} x_{i}^{2} + e_{0} x_{j}^{2} + f_{0} x_{i} x_{j} - - - (4)

Then can calculate from the input variable part of matrix (4) and obtain new matrix as shown in the table:

(5)

Next step optimized choice;

The process of previous step is the process that input variable x is replaced by new input variable z in fact, and the object changed like this finds out between it and inspection matrix y the relation existed;

The method used is as follows:

By formula (3), the input and output variable in the test samples matrix in formula (2) matrix is carried out polynomial regression, and the result obtained corresponding element value inner with table (1) is by the root-mean-square value of column count between them:

r_{j} = {[\frac{Σ_{i = nt + 1}^{n} {(y_{i} - z_{ij})}^{2}}{Σ_{i = nt + 1}^{n} {y_{i}}^{2}}]}^{\frac{1}{2}}, j = 1,2, \cdot \cdot \cdot, k - - - (6)

Here sequence number obtains by the sequence number in test samples, is nt+1 to n;

According to experience value, or sets itself threshold value r_g, from the matrix obtained, leave out those r_j≥r_gRow; K' row are made to meet r_j<r_g, then it is reassembled into matrix z ' with K'_k, then with z '_kM), replacing the input variable of matrix (5), this just obtains one group of new matrix, and (wherein k ' is < as shown in table

Next step is optimized inspection and calculates;

According to previous step, it is possible to obtain minimum r_j, it is designated as R_min; Then with z as variable, repeat the process of the first step and the 2nd step, obtain R_minIf, the R that this step produces_minThan the R of previous step_minLittle, then continue to repeat the first step and the 2nd step, until the R obtained_minBig than a step above, so iteration just stops; This iterative process just can reach requirement after 4 iteration; The first of input matrix z arranges the y produced_iIt is worth as follows:

\begin{matrix} {\overset{&OverBar;}{y}}_{1} = a + Σ_{i = 1}^{m} b_{i} x_{1 i} + Σ_{i = 1}^{m} Σ_{j = 1}^{m} c_{ij} x_{1 i} x_{1 j} + \cdot \cdot \cdot \\ {\overset{&OverBar;}{y}}_{2} = a + Σ_{i = 1}^{m} b_{i} x_{2 i} + Σ_{i = 1}^{m} Σ_{j = 1}^{m} c_{ij} x_{2 i} x_{2 j} + \cdot \cdot \cdot \\ \cdot \\ \cdot \\ \cdot \\ {\overset{&OverBar;}{y}}_{n} = a + Σ_{i = 1}^{m} b_{i} x_{ni} + Σ_{i = 1}^{m} Σ_{j = 1}^{m} c_{ij} x_{ni} x_{nj} + \cdot \cdot \cdot \end{matrix} - - - (8)

Y described above₁To y_n, it is obtain by n sampled data; Predictive model required by us is exactly the polynomial expression represented by the first row of matrix;

Next provides the application flow of above-mentioned algorithm in oil-gas gathering and transportation system safe early warning system, and idiographic flow is as follows:

1. set up raw data set, regard the parameter value of oil-gas gathering and transportation system on-line data acquisition as a time series, predict x by front k the value in n moment_n, can represent for x_n=f(x_n-1,x_n-2,…x_n-k), corresponding matrix is

W = \begin{matrix} [\begin{matrix} x_{k + 1} & x_{k} & \cdot \cdot \cdot & x_{1} \\ x_{k + 2} & x_{k + 1} & \cdot \cdot \cdot & x_{2} \\ \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot \\ x_{k + m} & x_{k + m - 1} & \cdot \cdot \cdot & x_{m} \\ \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot \\ x_{n} & x_{n - 1} & \cdot \cdot \cdot & x_{n - k} \end{matrix}] \end{matrix}

Wherein, k is prediction order;

2., according to GMDH algorithm process, above-mentioned raw data is divided into training set and inspection collection;

3. get the reference function of (3) formula as model, each independent variable(s) in this function is used as new independent variable(s), then there are 5 independent variable(s) according to (3) formula, then the first layer input of GMDH model can choose this 5 new independent variable(s), again these 5 variablees are combined respectively between two, it is possible to producing new model number isOn the set W of training set and system, simulate these new models respectively according to method of least squares, obtain the mid-module of GMDH network like this; Again according to the model in the middle of these, calculate the estimated value on training set corresponding to mid-module and system set W;

4. with the minimum deviation criterion in the outer criterion of packet Processing Algorithm as the formula (6), from all competitive models that the 3rd step is formed, search out optimum model;

5. by Optimality equations, the situation in future is predicted, obtains final early warning result, and carry out early warning measure in advance and prepare.

Compared with prior art, invention has the following advantages: modeler does not exist bias in modeling process, computer is the optimum variable of Confirming model and structure automatically, ensures selecting without artificial subjective factor of model like this, such that it is able to improve the precision of data prediction.

Accompanying drawing explanation

Fig. 1 is GMDH algorithm iteration step number judgment criterion schematic diagram.

Fig. 2 is the application flow block diagram of GMDH algorithm in oil-gas gathering and transportation system safe early warning system.

Fig. 3 is water trap oil liquid level prognostic chart.

Fig. 4 is water trap oil liquid level prediction partial enlargement figure.

Fig. 5 is the fuel-displaced pressure prediction figure of water trap.

Fig. 6 is the fuel-displaced pressure prediction partial enlargement figure of water trap.

Fig. 7 is the fuel-displaced temperature prediction figure of water trap.

Fig. 8 is the fuel-displaced temperature prediction partial enlargement figure of water trap

Embodiment

Below in conjunction with accompanying drawing, the invention will be further described:

A kind of safe monitoring and pre-alarming method of oil-gas gathering and transportation system, comprises the modeling process of GMDH algorithm and the algorithm application flow in oil-gas gathering and transportation system safe early warning system;

Y=f(x₁,x₂,…,x_n) (1)

Wherein Y is dependent variable, and x is independent variable(s);

W = [\begin{matrix} y_{1} & x_{11} & x_{12} & \cdot \cdot \cdot & x_{1 m} \\ y_{2} & x_{21} & x_{22} & \cdot \cdot \cdot & x_{2 m} \\ \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot \\ y_{nt} & x_{nt, 1} & x_{nt, 2} & \cdot \cdot \cdot & x_{nt, m} \\ \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot \\ y_{n} & x_{n, 1} & x_{n, 2} & \cdot \cdot \cdot & x_{n, m} \end{matrix}] - - - (2)

Next step regression Calculation;

y_{ij} = A + {Bx}_{i} + {Cx}_{j} + {Dx}_{i}^{2} + {Ex}_{j}^{2} + {Fx}_{i} x_{j}, i &NotEqual; j, 1 \leq (i, j) \leq m - - - (3)

Wherein the learning sample of A, B, C, D, E and F matrix (5) calculates according to method of least squares;

After recurrence, it is possible to produceThe recurrence polynomial expression of individual higher-order, if:

z = a_{0} + b_{0} x_{i} + c_{0} x_{j} + d_{0} x_{i}^{2} + e_{0} x_{j}^{2} + f_{0} x_{i} x_{j} - - - (4)

(5)

Next step optimized choice;

The method used is as follows:

r_{j} = {[\frac{Σ_{i = nt + 1}^{n} {(y_{i} - z_{ij})}^{2}}{Σ_{i = nt + 1}^{n} {y_{i}}^{2}}]}^{\frac{1}{2}}, j = 1,2, \cdot \cdot \cdot, k - - - (6)

Next step is optimized inspection and calculates;

According to previous step, it is possible to obtain minimum r_j, it is designated as R_min; Then with z as variable, repeat the process of the first step and the 2nd step, obtain R_minIf, the R that this step produces_minThan the R of previous step_minLittle, then continue to repeat the first step and the 2nd step, until the R obtained_minBig than a step above, so iteration just stops; See Fig. 1, it is seen that, this process just can reach requirement after 4 iteration; The first of input matrix z arranges the y produced_iIt is worth as follows:

\begin{matrix} {\overset{&OverBar;}{y}}_{1} = a + Σ_{i = 1}^{m} b_{i} x_{1 i} + Σ_{i = 1}^{m} Σ_{j = 1}^{m} c_{ij} x_{1 i} x_{1 j} + \cdot \cdot \cdot \\ {\overset{&OverBar;}{y}}_{2} = a + Σ_{i = 1}^{m} b_{i} x_{2 i} + Σ_{i = 1}^{m} Σ_{j = 1}^{m} c_{ij} x_{2 i} x_{2 j} + \cdot \cdot \cdot \\ \cdot \\ \cdot \\ \cdot \\ {\overset{&OverBar;}{y}}_{n} = a + Σ_{i = 1}^{m} b_{i} x_{ni} + Σ_{i = 1}^{m} Σ_{j = 1}^{m} c_{ij} x_{ni} x_{nj} + \cdot \cdot \cdot \end{matrix} - - - (8)

Next provides the application flow of algorithm in oil-gas gathering and transportation system safe early warning system, as follows see Fig. 2 idiographic flow:

W = \begin{matrix} [\begin{matrix} x_{k + 1} & x_{k} & \cdot \cdot \cdot & x_{1} \\ x_{k + 2} & x_{k + 1} & \cdot \cdot \cdot & x_{2} \\ \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot \\ x_{k + m} & x_{k + m - 1} & \cdot \cdot \cdot & x_{m} \\ \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot & \cdot \cdot \cdot \\ x_{n} & x_{n - 1} & \cdot \cdot \cdot & x_{n - k} \end{matrix}] \end{matrix}

Wherein, k is prediction order;

The collection in worksite data of the major equipment water trap in oil-gas gathering and transportation system have been carried out prediction checking by calculation process according to GMDH algorithm and calculation procedure, respectively to the oil liquid level of water trap, go out oil pressure and fuel-displaced temperature has been applied.

Present method have employed MATLAB and GMDH method programmed, and have chosen the input of the working parameter of a certain period water trap as model data, and the result of prediction is as shown in figures 3-8. As can be seen from the figure, after adopting the method, the predictor of the data of collection in worksite and the fitting degree of actual value are very high, and relative error is maximum is only 0.03%. Absolutely prove the feasibility that GMDH model is predicted for the safety of oil-gas gathering and transportation system online data.

Claims

1. the safe monitoring and pre-alarming method of oil-gas gathering and transportation system, it is characterised in that: comprise the modeling process of GMDH algorithm and the algorithm application flow in oil-gas gathering and transportation system safe early warning system;

Y=f (x₁,x₂,…,x_n)(1)

Wherein y is dependent variable, and x is independent variable(s);

W = [\begin{matrix} y_{1} & x_{11} & x_{12} & ... & x_{1 m} \\ y_{2} & x_{21} & x_{22} & ... & x_{2 m} \\ ... & ... & ... & ... & ... \\ y_{n t} & x_{n t, 1} & x_{n t, 2} & ... & x_{n t, m} \\ ... & ... & ... & ... & ... \\ y_{n} & x_{n, 1} & x_{n, 2} & ... & x_{n, m} \end{matrix}] - - - (2)

Next step regression Calculation;

y_{i j} = A + {Bx}_{i} + {Cx}_{j} + {Dx}_{i}^{2} + {Ex}_{j}^{2} + {Fx}_{i} x_{j}, i &NotEqual; j, 1 \leq (i, j) \leq m - - - (3)

Wherein A, B, C, D, E and F are calculated according to method of least squares by the learning sample in matrix (5);

z = a_{0} + b_{0} x_{i} + c_{0} x_{j} + d_{0} x_{i}^{2} + e_{0} x_{j}^{2} + f_{0} x_{i} x_{j} - - - (4)

Then can calculate from the input variable part of matrix (4) and obtain new matrix as follows:

Next step optimized choice;

The method used is as follows:

By formula (3), the input and output variable in the test samples matrix in formula (2) matrix is carried out polynomial regression, and the result obtained corresponding element value inner with matrix (5) is by the root-mean-square value of column count between them:

r_{j} = {[\frac{Σ_{i = n t + 1}^{n} {(y_{i} - z_{i j})}^{2}}{Σ_{i = n t + 1}^{n} {y_{i}}^{2}}]}^{\frac{1}{2}}, j = 1, 2, ..., k - - - (6)

According to experience value, or sets itself threshold value r_g, from the matrix obtained, leave out those r_j≥r_gRow; K' row are made to meet r_j< r_g, then it is reassembled into matrix z with K'_K', then use z_K'Replacing the input variable of matrix (5), this just obtains one group of new matrix (wherein K'< m), as shown in table

Next step is optimized inspection and calculates;

According to previous step, it is possible to obtain minimum r_j, it is designated as R_min; Then with z as variable, repeat the process of the first step and the 2nd step, obtain R_minIf, the R that this step produces_minThan the R of previous step_minLittle, then continue to repeat the first step and the 2nd step, until the R obtained_minBig than a step above, so iteration just stops; The first of input matrix z arranges the y produced_iIt is worth as follows:

\begin{matrix} y_{1} = a + Σ_{i = 1}^{m} b_{i} x_{1 i} + Σ_{i = 1}^{m} Σ_{j = 1}^{m} c_{i j} x_{1 i} x_{1 j} + ... \\ y_{2} = a + Σ_{i = 1}^{m} b_{i} x_{2 i} + Σ_{i = 1}^{m} Σ_{j = 1}^{m} c_{i j} x_{2 i} x_{2 j} + ... \\ \begin{matrix} . \\ . \\ . \end{matrix} \\ y_{n} = a + Σ_{i = 1}^{m} b_{i} x_{n i} + Σ_{i = 1}^{m} Σ_{j = 1}^{m} c_{i j} x_{n i} x_{n j} + ... \end{matrix} - - - (8)

Next provides the application flow of algorithm in oil-gas gathering and transportation system safe early warning system, and idiographic flow is as follows:

I. set up raw data set, regard the parameter value of oil-gas gathering and transportation system on-line data acquisition as a time series, predict x by the individual value of front k ' in n moment_n, can represent for x_n=f (x_n-1, x_n-2... x_n-k′) corresponding matrix is

W = [\begin{matrix} x_{k^{,} + 1} & x_{k^{,}} & ... & x_{1} \\ x_{k^{,} + 2} & x_{k^{,} + 1} & ... & x_{2} \\ ... & ... & ... & ... \\ x_{k^{,} + m} & x_{k^{,} + m - 1} & ... & x_{m} \\ ... & ... & ... & ... \\ x_{n} & x_{n - 1} & ... & x_{n - k^{,}} \end{matrix}]

Wherein, k ' is prediction order;

II. according to GMDH algorithm process, above-mentioned raw data is divided into training set and inspection collection;

III. get the reference function of (3) formula as model, each independent variable(s) in this function is used as new independent variable(s), then there are 5 independent variable(s) according to (3) formula, then the first layer input of GMDH model can choose this 5 new independent variable(s), again these 5 variablees are combined respectively between two, it is possible to producing new model number isOn the set W of training set and system, simulate these new models respectively according to method of least squares, obtain the mid-module of GMDH network like this; Again according to the model in the middle of these, calculate the estimated value on training set corresponding to mid-module and system set W;

IV. by the minimum deviation criterion in the outer criterion of packet Processing Algorithm such as formula, shown in (6), searching out optimum model from all competitive models that the 3rd step is formed;

V. by Optimality equations, the situation in future is predicted, obtain final early warning result, and carry out early warning measure in advance and prepare.