CN113806902A - Artificial intelligent early warning method for pipeline corrosion - Google Patents

Abstract

The invention discloses an artificial intelligent early warning method for pipeline corrosion, and belongs to the technical field of pipeline safety. The method comprises the following steps: firstly, collecting basic data of the pipeline, judging the condition of the pipeline for continuous service, then predicting the corrosion rate and the ultrasonic side thickness estimation corrosion rate by using a long-time memory neural network, predicting the residual service life of the pipeline, judging the safe service condition of the pipeline and making a targeted warning. The method is based on the prediction of the residual service life of the pipeline, adopts an artificial intelligence method to predict the future operating condition of the pipeline, gives out warning to abnormal conditions and adopts a coping method in time, so that the oil and gas field can be prevented from getting ill in the bud, and the intelligent oil and gas field is provided with a technical support of a corrosion early warning layer for the construction of the intelligent oil and gas field by applying medicines according to symptoms.

Description

Artificial intelligent early warning method for pipeline corrosion

Technical Field

The invention belongs to the technical field of pipeline safety, and particularly relates to a method for establishing artificial intelligent early warning of pipeline corrosion.

Background

In the oil gas field pipeline production operation process, pipeline corrosion can make pipe wall defect position attenuation, causes the pipeline bearing capacity to reduce, leads to pipeline corrosion inefficacy, causes the pipeline to leak, and the accident that causes often has proruption nature and disguise, can cause huge loss. Therefore, before a pipeline corrosion accident occurs, the serious corrosion defect part needs to be warned, overhauled and maintained in time, and corresponding protective measures are taken to scientifically guide the safe operation management of the pipeline.

At present, the existing corrosion early warning method still has some problems to be solved: the establishment of the corrosion early warning method usually depends on a severe network infrastructure, a complex database needs to be established as a basis, the data acquisition principle is fuzzy, the difficulty of early warning work is increased, and the operability is low; the corrosion early warning method is mainly based on the pipeline which cannot be continuously used, pipeline leakage data obtained by analysis and detection are mined, probability judgment is carried out on future corrosion conditions, pipeline accidents cannot be effectively prevented, the early warning result accuracy is low, and protection measures are lack of pertinence.

Therefore, in order to reduce the risk of pipeline corrosion accidents, an algorithm needs to be established to predict the residual service life of the pipeline, judge the safety service condition of the pipeline, determine the corrosion early warning level, and actively take corresponding measures in advance.

Disclosure of Invention

The invention aims to provide an artificial intelligent early warning method for pipeline corrosion, which aims to solve the problem of risk early warning caused by reduced pressure resistance after the pipeline is corroded.

An artificial intelligent early warning method for pipeline corrosion is characterized by comprising the following steps:

step 1: collecting basic data of the pipeline:

pipe outside diameter D_wMm; (2) inner diameter D of pipeline_nMm; (3) yield strength sigma of pipe material_sMPa; (4) the wall thickness d, mm of the pipeline; (5) the design pressure P, MPa of the pipeline; (6) the corrosion allowance C and mm of the pipeline; (7) pipeline production run time T_sA; (8) pipeline design service life T_u。

Step 2: dividing a pipeline corrosion defect area:

detecting corrosion defects of the pipeline, and meshing the areas according to the axial direction and the annular direction: axially divided into m parts and respectively C₁、C₂…C_i…C_mAnd n parts are divided annularly into L₁、L₂…L_j…L_nSo as to discretely divide the corrosion defect into m × n wall thickness measurement points A_ij(i＝1、2、3…m；j＝1、2、3…n)；

Wherein: m is the number of axially delimited regions, C₁、C₂…C_mEach part of the axially delimited area corresponds to an axial measuring point of one defect; n is the number of the annularly defined areas, L₁、L₂…L_nEach part of the axially defined area corresponds to a circumferential measuring point of one defect; a. the_ijM n wall thickness measurement points discretely divided for corrosion defects.

And step 3: the method comprises the following steps of:

(a) solving for axially required minimum wall thickness by equation (1)

By passingEquation (2) solving the circumferential required minimum wall thickness

The calculated result is processed

And

substituting formula (3) to determine the minimum required wall thickness t of the pipeline_min：

In the formula:

the minimum wall thickness is required in the axial direction, mm;

the minimum wall thickness is required in the circumferential direction and is mm; t is t_minThe minimum required wall thickness of the pipeline is mm;

(b) counting the wall thickness value a of the m multiplied by n wall thickness measuring points of the pipeline corrosion defect area_ijWherein the measurement gives the wall thickness value at the minimum is a_minSolving the average value t of the wall thickness of all measured points through the formula (4)_am：

In the formula: a is_ijDiscrete division of m by n for corrosion defectsWall thickness value, mm, of each measurement point; t is t_amThe average value of the wall thickness of all measured points is mm;

(c) solving the residual wall thickness ratio R of the pipeline through the formula (5)_t：

In the formula: r_tThe residual wall thickness ratio of the pipeline is used;

(d) solving the length L of the maximum allowable corrosion defect in the axial direction of the pipeline:

if R is_tNot less than 0.793, then

If R is_tIf less than 0.793, then

Wherein: l is the length value of the axial maximum allowable corrosion defect, mm;

(e) determining the safe service condition of the pipeline:

if La is less than or equal to L, the pipeline can be continuously in service;

if La is greater than L and tam-C is more than or equal to 0.9tmin, the pipeline can continue to be in service;

if La is greater than L and tam-C is less than 0.9tmin, the pipeline cannot be in service continuously, the pipeline is judged to be in first-level early warning level, and the early warning mark is displayed in red;

wherein: l is_aThe axial length of the corrosion defect of the wall thickness section of the pipeline is mm.

And 4, step 4: predicting the residual life of the pipeline:

(a) predicting the residual life of the pipeline based on online monitoring:

1) predicting the corrosion rate of the pipeline, and specifically comprises the following steps:

arranging an on-site corrosion monitoring data set: monitoring time data set alpha (t) of a certain time period₁、t₂、t₃…t_n) (ii) a Monitoring time-correlated corrosion rate data setsβ(V₁、V₂、V₃…V_n)；

Wherein: t is t₁、t₂、t₃…t_nMonitoring time of one step length according to time sequence; v₁、V₂、V₃…V_nThe corrosion rate monitoring value is mm/a corresponding to the monitoring time;

and secondly, observing the corrosion monitoring data set to supplement missing data:

arbitrarily take 3 times t in the corrosion monitoring time data set alpha_a、t_b、t_cTaking the corresponding corrosion rate monitoring value V in the corrosion rate data set beta_a、V_b、V_cSubstituting equation (6) to obtain the missing time t_xCorresponding corrosion rate value V_x：

In the formula: t is t_a、t_b、t_cMonitoring any three times in the time data set alpha; v_a、V_b、V_cFor t in the corrosion rate data set beta_a、t_b、t_cCorresponding corrosion rate monitoring value, mm/a; t is t_xMonitoring time for absence; v_xIs t_xCorresponding corrosion rate, mm/a;

constructing a long-time memory neural network:

i complete corrosion monitoring dataset: monitoring time data set alpha' (t) containing missing values of corrosion data₁、t₂、t₃…t_x…t_n) And a corresponding corrosion rate data set β' (V)₁、V₂、V₃…V_x…V_n) Wherein α 'is an input value and β' is an output value;

II long-time memory neural network model inner structure contains forgetting gate, input gate, output gate:

i last time t_i-1Corrosion rate monitoring of_i-1Passing through t_iTimeForgetting gate f (t) of long-time and short-time neural network_i) Update forgetting is performed by equation (7):

in the formula: f (t)_i) Expressed as a forgetting gate function; sigma is a neural network activation function, and a sigmoid function is selected; w is a_fAnd u_fIs a forgetting gate weight coefficient matrix; b_fIs a network offset value; t is t_iPredicting a time for the target;

represents t_iTime data corresponding to time; t is t_i-1A certain time for monitoring the time data set alpha'; v_i-1Represents the corrosion rate in the corresponding corrosion rate dataset β', mm/a;

iit_icorresponding time data

t_i-1Time-corresponding corrosion monitoring value V_i-1Enter t_iInput gate i (t) of time-interval neural network_i) Equation (8) and alternatives

In the formula: i (t)_i) A decision coefficient representing an input gate; sigma is a neural network activation function, and a sigmoid function is selected; w is a_iAnd u_iDetermining a coefficient weight coefficient matrix for the input gate; b_iAn offset value representing the input gate decision coefficient matrix;

representing input gate alternative content; w is a_cRepresenting an input gate alternative content weight matrix; b_cA bias value representing the input gate alternate content; tanh represents a hyperbolic tangent excitation function;

iii using t_iForgetting door f (t) at time_i) Determining the coefficient i (t) with the input gate_i) Alternative content

Updating the current state of the neuron to obtain t_iTemporal neuron update function

Formula (10):

in the formula:

represents t_i-1A neural update function of time;

represents t_iA neural update function of time;

ivt_ineuron update function of time

t_iCorresponding time data

And last time t_i-1Corrosion rate monitoring value V corresponding to time_i-1Passing through t_iTime output gate

Formula (11), output CorrosionRate prediction value V_iFormula (12):

in the formula:

representing an output gate decision coefficient; sigma represents a neural network activation function, and a sigmoid function is selected; w is a_oRepresenting an output gate weight matrix; b_oA bias value representing an output gate; v_iRepresents the target time t_iThe predicted value of the corrosion rate of (1), mm/a;

2) predicting the corrosion rate V obtained by the prediction_iMinimum required wall thickness t of corroded pipeline_minAnd the average value t of the wall thickness of all measured points_amSubstituting formula (13) to solve pipeline residual life T'_L：

In the formula: t'_LIn order to monitor the residual service life of the pipeline on line, a;

k is a safety coefficient, and when La is less than or equal to L, the value of K is 1; when La is more than L, K is 0.9;

(b) predicting the residual life of the pipeline based on ultrasonic thickness measurement:

1) estimating pipeline corrosion rate V_μFormula (14):

in the formula: v_μFor corrosion rate estimation, mm/a; delta d is the difference of wall thickness before and after the same point thickness measurement, mm; delta T is the time difference before and after thickness measurement，a；

2) Solving the pipeline to calculate the wall thickness epsilon formula (15):

in the formula: epsilon is the calculated wall thickness of the steel pipe, mm; p is design pressure, MPa; sigma_sThe yield strength of the steel pipe is MPa;

3) estimating the corrosion rate V_μSubstituting the calculated wall thickness epsilon of the pipeline into a calculation formula (16) to solve the corrosion residual life T'_L：

In the formula: t'_LMeasuring the residual life of the pipeline based on ultrasonic waves, a;

(b) determining the remaining life T of a pipeline_L：

T_L＝min(T’_L,T”_L) (17)

In the formula: t is_LThe remaining life of the pipeline, a.

And 5: judging the safe service condition of the pipeline:

(a) if T_L≥T_u-T_sIf the pipeline can continue to be in service safely, the pipeline is judged to be in a five-level early warning level, and the early warning mark is displayed to be green;

(b) if T_L＜T_u-T_sFurther judging the corrosion early warning level;

wherein: t is_sRunning time for pipeline production, a; t is_uDesign age, a.

Step 6: setting conditions of corrosion early warning levels:

(a) if T_s<T_L<T_u-T_sIf so, judging the early warning level as four-level, and displaying the early warning mark as blue;

(b) if 10<T_L<T_uJudging the early warning level to be three-level, and displaying an early warning markIs yellow;

(c) if 3<T_L<When 10, judging the early warning level as a secondary early warning level, and displaying an early warning mark as orange;

(d) if T_L<And 3, judging the early warning level as the first-level early warning level, and displaying the early warning mark in red.

And 7: early warning treatment:

(a) for the first-level early warning, the corrosion degree is very serious, and a corrosion pipeline needs to be replaced;

(b) for the secondary early warning, the corrosion degree is serious, and the pipeline needs to stop running and be repaired;

(c) for the third-level early warning, the corrosion degree is relatively heavy, and the pipeline is depressurized, operated and repaired;

(d) for the four-stage early warning, observing the corrosion condition of the pipeline;

(e) for the five-level early warning, the corrosion control condition of the pipeline is better, and the pipeline can be safely produced and operated.

The invention has the following beneficial effects:

(1) the corrosion early warning method collects pipeline basic data which are easy to obtain, establishes a pipeline residual life algorithm on the premise that the pipeline can be continuously used, simplifies corrosion early warning steps and improves system operability.

(2) The corrosion early warning method is based on the prediction of the residual service life of the pipeline, and adopts the methods of artificial intelligent soft measurement and ultrasonic thickness measurement estimation to form redundancy, so that the defect that the calculation condition of the residual service life of the pipeline is incomplete can be overcome, and the corrosion early warning accuracy is improved.

(3) The corrosion early warning method can accurately predict the residual service life of the pipeline, judge the corrosion early warning level of the pipeline, is beneficial to the oil and gas field to actively control the corrosion condition of the pipeline in advance, adopts targeted protective measures, promotes the safe, economic and reliable operation of the pipeline, and provides technical support of a corrosion early warning layer for the construction of the intelligent oil and gas field.

Drawings

FIG. 1 is a flow chart of corrosion warning operations;

FIG. 2 is a flow chart of message transmission in the neural network of long and short memories;

FIG. 3 is a schematic view of pipe corrosion defect meshing.

Detailed Description

The invention is described in detail below with reference to the drawings and the specific embodiments.

Example 1:

taking the No. 1 pipeline in the X station as an example, the pipeline is subjected to corrosion early warning. The method comprises the following specific implementation steps:

step 1: pipeline base data No. 1 was collected as follows: (1) pipe outside diameter D_w141.3 mm; (2) inner diameter D of pipeline_n1132.5 mm; (3) yield strength sigma of pipe material_s1240 MPa; (4) wall thickness d of pipe₁8.8 mm; (5) pipeline corrosion allowance C₁4 mm; (6) design pressure P of pipeline₁＝20MPa。

Step 2: dividing a No. 1 pipeline corrosion defect area: detecting No. 1 pipeline corrosion defects, and meshing the area according to the axial direction and the annular direction: is divided axially into 4 parts and respectively is C₁、C₂、C₃、C₄And are divided circumferentially into 5 parts each of L₁、L₂、L₃、L₄、L₅So that the corrosion defect is discretely divided into 20 wall thickness measurement points a_1×1＝9.36mm、a_1×2＝9.52mm、a_1×3＝9.12mm、a_1×4＝9.41mm、a_1×5＝9.33mm、a_2×1＝9.15mm、a_2×2＝8.86mm、a_2×3＝9.49mm、a_2×4＝8.81mm、a_2×5＝9.51mm、a_3×1＝9.57mm、a_3×2＝9.71mm、a_3×3＝9.25mm、a_3×4＝9.39mm、a_3×5＝9.26mm、a_4×1＝9.66mm、a_4×2＝9.29mm、₆a_4×3＝9.50mm、a_4×4＝9.62mm、a_4×5＝9.52mm。

And step 3: and (3) judging the continuous service condition of the No. 1 pipeline:

(a) solving for axially required minimum wall thickness by equation (1)

Through a maleEquation (2) solving the circumferential required minimum wall thickness

Substituting the calculation result into the formula (3) to determine the minimum required wall thickness t of the corroded pipeline_min1＝7.67mm；

(b) Counting the wall thickness values of 20 wall thickness measuring points of the No. 1 pipeline corrosion defect area, and determining that a is the smallest measured wall thickness value_min1Substituting the equation (4) for 8.81mm, and solving to obtain the average value t of the wall thickness of all the measured points_am1＝9.37mm；

(c) Solving the residual wall thickness ratio R of the pipeline through the formula (5)_t1＝0.63mm；

(d) Due to R_t1< 0.793, length of No. 1 pipe axially allowed maximum corrosion defect

(e) Determining the safe service condition of the No. 1 pipeline: axial length L of corrosion defect on wall thickness section of No. 1 pipeline_a180mm due to L_a1＞L₁And t is_am1-C₁＝5.37mm＜0.9t_minAnd when the pipeline is 6.903mm, the pipeline cannot be in service continuously, the pipeline is judged to be in a first-level early warning grade, and the early warning is displayed in red.

And 4, step 4: carrying out primary early warning treatment on the No. 1 pipeline: the operation is stopped and the pipeline is replaced.

Example 2:

taking the No. 2 pipeline in the X station as an example, the pipeline is subjected to corrosion early warning. The method comprises the following specific implementation steps:

step 1: pipeline base data No. 2 was collected as follows: (1) pipe outside diameter D_w2168.8 mm; (2) inner diameter D of pipeline_n2157.3 mm; (3) yield strength sigma of pipe material_s2360 MPa; (4) wall thickness d of pipe₂11.5 mm; (5) pipeline design age T_u220 a; (6) pipeline production run time T_s210 a; (7) pipeline corrosion allowance C₂3.0 mm; (8) design pressure P of pipeline₂＝9.6MPa。

Step 2: no. 2 divided pipeline rotten productEtching the defect area: detecting No. 2 pipeline corrosion defects, and meshing the area according to the axial direction and the annular direction: are axially divided into 4 parts which are respectively C'₁、C’₂、C’₃、C’₄L 'are circumferentially divided into 4 parts'₁、L’₂、L’₃、L’₄So that the corrosion defects are discretely divided into 16 wall thickness measurement points, respectively, of'_1×1＝12.55mm、a’_1×2＝13.58mm、a’_1×3＝14.87mm、a’_1×4＝14.38mm、a’_2×1＝11.84mm、a’_2×2＝13.48mm、a’_2×3＝14.33mm、a’_2×4＝14.35mm、a’_3×1＝11.76mm、a’_3×2＝13.78mm、a’_3×3＝16.50mm、a’_3×4＝15.10mm、a’_4×1＝11.27mm、a’_4×2＝13.44mm、a’_4×3＝15.92mm、a’_4×4＝15.28mm。

And step 3: and (3) judging the continuous service condition of the No. 2 pipeline:

(a) solving for axially required minimum wall thickness by equation (1)

The minimum wall thickness is required annularly through formula (2)

Substituting the calculation result into the formula (3) to determine the minimum required wall thickness t of the corroded pipeline_min2＝2.91mm；

(b) Counting the wall thickness values of 16 wall thickness measuring points of No. 2 pipeline corrosion defect area, and determining that a is the smallest measured wall thickness value_min2Substituting the equation (4) to obtain the average value t of the wall thickness of all the measured points_am2＝13.90mm；

(c) Solving the residual wall thickness ratio R of the pipeline through the formula (5)_t2＝2.84mm；

(d) Due to R_t2Not less than 0.793, the maximum allowable axial corrosion defect length of No. 2 pipeline

(e) Determining the safe service condition of the No. 2 pipeline: axial length L of No. 2 pipeline wall thickness section corrosion defect_a280mm due to L_a1≤L₁The pipeline may continue to be in service.

And 4, step 4: predicting the residual life of the No. 2 pipeline:

(a) predicting the residual life of the No. 2 pipeline based on online monitoring:

1) predicting the corrosion rate of the pipeline, and specifically comprises the following steps:

arranging an on-site corrosion monitoring data set: data set alpha of monitoring time from 1/2018 to 1/7/2021₂(1/2018, 2/1/2018, 3/1/2018, … 2019/4/1/2019, 6/1/2019, … 2021/7/1/2011); monitoring time-corresponding corrosion rate data set beta₂(0.01367mm/a, 0.04719mm/a, 0.03254mm/a … 0.0423.0423 mm/a, 0.05478mm/a … 0.05718mm/a) are shown in Table 1:

table 12 corrosion monitoring data for pipe 2018, month 1, and year 2021, month 7, and day 1

α2Time set	β2Set of corrosion rates (mm/a)
		1 month and 1 day of 2018	0.01367
2 month and 1 day of 2018	0.04719
		3 month and 1 day of 2018	0.03254
…	…
		4 month and 1 day of 2019	0.04230
6 months and 1 day in 2019	0.05478
		…	…
Year 2020, 7 and 1	0.05718

(vii) observe table 1 corrosion monitoring dataset replenishment missing data: in Table 1, 2 corrosion rate deletions need to be supplemented, and are respectively the corrosion rate values V of No. 2 pipeline 2019, 5 months and 1 day_xCorrosion rate V of 1/3/2020_x′：

First, monitoring time data t of 2019, 4 months and 1 day is extracted_a401 and corresponding etch rate V_a0.0423 mm/a; monitoring time data t of 6 month and 1 day in 2019_b601 and corresponding corrosion rate V_b0.05478 mm/a; monitoring time data t of 7 month and 1 day in 2019_c701 and corresponding etch rate V_c0.02537mm/a, the above numerical value is substituted into formula (7) to supplement and calculate the time data t of 2019 in 5 months_xCorrosion rate value V corresponding to 501_x：

Similarly, extracting data t of monitoring time of 2 months and 1 day of 2020_a' (201) and corresponding corrosion rate value V_a' -0.0233 mm/a,; monitoring time data t of 1 day in 2020, 4 months_b401 and corresponding corrosion rate value V_b' -0.04523 mm/a; monitoring time data t of 5 month and 1 day in 2020_c' (501) and corresponding corrosion rate value V_c' 0.03754mm/a, the above numerical value is substituted into formula (7) to complement and calculate the monitoring time data t of 3 months in 2020_xCorrosion rate value V corresponding to' 301_x′＝0.0179mm/a；

Constructing a long-time memory neural network, and specifically comprising the following steps:

i complete pipeline No. 2 corrosion monitoring dataset: monitoring time data set alpha 'containing corrosion data missing value'₂(1/2018/2/1/2018/3/1/… 2019/5/1/20184/… 2020/3/1/… 2021/7/1/2018), monitoring the time-dependent corrosion rate dataset β'₂(0.01367mm/a、0.04719mm/a、

0.03254mm/a … 0.033.033 mm/a … 0.0179mm/a … 0.05718mm/a) see table 2, where α 'is the input value and β' is the output value;

complete corrosion monitoring data for pipeline # 22 from 2018, 1 month, 1 day to 2021, 7 months, 1 day

Input value of alpha'2	Output value of beta'2(mm/a)
		1 month and 1 day of 2018	0.01367
2 month and 1 day of 2018	0.04719
		3 month and 1 day of 2018	0.03254
…	…
		4 month and 1 day of 2019	0.04230
5 months and 1 day in 2019	0.03300
		6 months and 1 day in 2019	0.05478
…	…
		Year 2020, 3 and 1	0.01790
…	…
		Year 2020, 7 and 1	0.05718

II long-time memory neural network model inner structure contains forgetting gate, input gate, output gate:

iI with 1 month as step length, corrosion rate monitoring value V corresponding to the time data 101 in 2018, 1 month and 1 day₁₀₁0.01367(mm/a) in 2018, 2/1 day and time data x^t2Forgetting gate f (t) of long-time and short-time neural network as 201₂₀₁) Carry over into (8) and update forgetting:

f(201)＝sigmod(w_f*0.1367+u_f*201+b_f) (8)

ii time data corresponding to 2 month and 1 day of 2018

201, 2018, 1 monthCorrosion monitoring value V corresponding to 1 day time data 101₁₀₁Input gate i (201) of long-term neural network with 0.01367(mm/a) entry time data 201, formula (9) and alternative content

i(201)＝sigmod(w_i*0.1367+u_i*201+b_i) (9)

iii forgetting gate f (201) and input gate determination coefficient i (201) using 2018 year 2 month 1 day time data 210, and candidate content

Updating the current state of the neuron to obtain a neuron updating function C of the data 201 of 2 months and 1 day in 2018₂₀₁Formula (11):

neuron update function C of 2 month and 1 day in IV2018₂₀₁Corrosion rate monitoring values V corresponding to the data 201 of 2 month, 1 day and 8 days in 2018 and the data 101 of the last time₁₀₁Output gate O of elapsed time data 201 (mm/a) 0.01367₂₀₁Equation (12), the predicted value V of the corrosion rate of 2018, 2 months and 1 day is output₂₀₁0.0028mm/a formula (13):

O₂₀₁＝sigmod(w_o*0.01367+u_o*201+b_o) (12)

V₂₀₁＝O₂₀₁·tanh(C₂₀₁) (13)

2) predicting the corrosion rate by a value V₂₀₁0.028mm/a, minimum required wall thickness t_min22.91mm and the average value t of the wall thickness of all measured points_am2Carry-in (14) to solve No. 2 pipeline residual life 13.90mm

Wherein L is_a≤L，K₂The value is 1;

(b) predicting the residual life of the No. 2 pipeline based on ultrasonic thickness measurement:

1) wall thickness d of the pipe₂11.5mm, minimum wall thickness measurement a_min211.27mm and production run time T_s2Substitution of 10a into equation (14) estimates the etch rate V_μ2＝0.023mm/a；

2) Design pressure P of pipeline₂Outer diameter of 9.6MPa₁₀D_w2168.8mm, yield strength sigma_s2360MPa and corrosion allowance C₂Calculation of wall thickness epsilon by substituting 3.0mm into equation (15)₂＝4.17mm；

3) Calculating the result V_μ20.023mm/a and epsilon₂Solving residual lifetime T ″, substituting 4.17mm into equation (16) "_L2＝308.7a；

(c) Determination of No. 2 pipeline residual life T by formula (17)_L：T_L2＝min(T’_L2,T”_L2)＝308.7a。

And 5: and (3) judging the safe service condition of the No. 2 pipeline: due to T_L2≥T_u2-T_s2And (5) if the pipeline is 10a, the pipeline can be continuously in service safely, the pipeline is judged to be in five-level early warning level, and the early warning mark is displayed in green.

Step 6: the No. 2 pipeline carries out five-stage early warning treatment: the corrosion control condition of the pipeline is better, and the pipeline can be produced and operated safely.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (1)

1. An artificial intelligent early warning method for pipeline corrosion is characterized by comprising the following steps:

step 1: collecting basic data of the pipeline:

pipe outside diameter D_wMm; (2) inner diameter D of pipeline_nMm; (3) yield strength sigma of pipe material_sMPa; (4) the wall thickness d, mm of the pipeline; (5) the design pressure P, MPa of the pipeline; (6) the corrosion allowance C and mm of the pipeline; (7) pipeline production run time T_sA; (8) pipeline design service life T_u。

Step 2: dividing a pipeline corrosion defect area:

detecting corrosion defects of the pipeline, and meshing the areas according to the axial direction and the annular direction: axially divided into m parts and respectively C₁、C₂…C_i…C_mAnd n parts are divided annularly into L₁、L₂…L_j…L_nSo as to discretely divide the corrosion defect into m × n wall thickness measurement points A_ij(i＝1、2、3…m；j＝1、2、3…n)；

Wherein: m is the number of axially delimited regions, C₁、C₂…C_mEach part of the axially delimited area corresponds to an axial measuring point of one defect; n is the number of the annularly defined areas, L₁、L₂…L_nEach part of the axially defined area corresponds to a circumferential measuring point of one defect; a. the_ijM n wall thickness measurement points discretely divided for corrosion defects.

And step 3: the method comprises the following steps of:

(a) solving for axially required minimum wall thickness by equation (1)

Solving the circumferential required minimum wall thickness by equation (2)

The calculated result is processed

And

substituting formula (3) to determine the minimum required wall thickness t of the pipeline_min：

In the formula:

the minimum wall thickness is required in the axial direction, mm;

the minimum wall thickness is required in the circumferential direction and is mm; t is t_minThe minimum required wall thickness of the pipeline is mm;

(b) counting the wall thickness value a of the m multiplied by n wall thickness measuring points of the pipeline corrosion defect area_ijWherein the measurement gives the wall thickness value at the minimum is a_minSolving the average value t of the wall thickness of all measured points through the formula (4)_am：

In the formula: a is_ijThe wall thickness values of m multiplied by n measuring points which are discretely divided for corrosion defects are mm; t is t_amThe average value of the wall thickness of all measured points is mm;

(c) solving the residual wall thickness ratio R of the pipeline through the formula (5)_t：

In the formula: r_tBeing conduitsThe residual wall thickness ratio;

(d) solving the length L of the maximum allowable corrosion defect in the axial direction of the pipeline:

if R is_tNot less than 0.793, then

If R is_tIf less than 0.793, then

Wherein: l is the length value of the axial maximum allowable corrosion defect, mm;

(e) determining the safe service condition of the pipeline:

if La is less than or equal to L, the pipeline can be continuously in service;

if La is greater than L and tam-C is more than or equal to 0.9tmin, the pipeline can continue to be in service;

if La is greater than L and tam-C is less than 0.9tmin, the pipeline cannot be in service continuously, the pipeline is judged to be in first-level early warning level, and the early warning mark is displayed in red;

wherein: l is_aThe axial length of the corrosion defect of the wall thickness section of the pipeline is mm.

And 4, step 4: predicting the residual life of the pipeline:

(a) predicting the residual life of the pipeline based on online monitoring:

1) predicting the corrosion rate of the pipeline, and specifically comprises the following steps:

arranging an on-site corrosion monitoring data set: monitoring time data set alpha (t) of a certain time period₁、t₂、t₃…t_n) (ii) a Monitoring a time-corresponding corrosion rate data set beta (V)₁、V₂、V₃…V_n)；

Wherein: t is t₁、t₂、t₃…t_nMonitoring time of one step length according to time sequence; v₁、V₂、V₃…V_nThe corrosion rate monitoring value is mm/a corresponding to the monitoring time;

and secondly, observing the corrosion monitoring data set to supplement missing data:

arbitrarily take 3 times t in the corrosion monitoring time data set alpha_a、t_b、t_cTaking the corresponding corrosion rate monitoring value V in the corrosion rate data set beta_a、V_b、V_cSubstituting equation (6) to obtain the missing time t_xCorresponding corrosion rate value V_x：

In the formula: t is t_a、t_b、t_cMonitoring any three times in the time data set alpha; v_a、V_b、V_cFor t in the corrosion rate data set beta_a、t_b、t_cCorresponding corrosion rate monitoring value, mm/a; t is t_xMonitoring time for absence; v_xIs t_xCorresponding corrosion rate, mm/a;

constructing a long-time memory neural network:

i complete corrosion monitoring dataset: monitoring time data set alpha' (t) containing missing values of corrosion data₁、t₂、t₃…t_x…t_n) And a corresponding corrosion rate data set β' (V)₁、V₂、V₃…V_x…V_n) Wherein α 'is an input value and β' is an output value;

II long-time memory neural network model inner structure contains forgetting gate, input gate, output gate:

i last time t_i-1Corrosion rate monitoring of_i-1Passing through t_iForgetting gate f (t) of time-interval neural network_i) Update forgetting is performed by equation (7):

in the formula: f (t)_i) Expressed as a forgetting gate function; sigma is a neural network activation function, and a sigmoid function is selected; w is a_fAnd u_fIs a forgetting gate weight coefficient matrix; b_fIs a network offset value; t is t_iPredicting a time for the target;

represents t_iTime data corresponding to time; t is t_i-1A certain time for monitoring the time data set alpha'; v_i-1Represents the corrosion rate in the corresponding corrosion rate dataset β', mm/a;

iit_icorresponding time data

t_i-1Time-corresponding corrosion monitoring value V_i-1Enter t_iInput gate i (t) of time-interval neural network_i) Equation (8) and alternatives

Formula (9):

in the formula: i (t)_i) A decision coefficient representing an input gate; sigma is a neural network activation function, and a sigmoid function is selected; w is a_iAnd u_iDetermining a coefficient weight coefficient matrix for the input gate; b_iAn offset value representing the input gate decision coefficient matrix;

representing input gate alternative content; w is a_cRepresenting an input gate alternative content weight matrix; b_cA bias value representing the input gate alternate content; tanh represents a hyperbolic tangent excitation function;

iii using t_iForgetting door f (t) at time_i) Determining the coefficient i (t) with the input gate_i) Alternative content

Updating the current state of the neuron to obtain t_iTemporal neuron update function C_tiFormula (10):

in the formula:

represents t_i-1A neural update function of time;

represents t_iA neural update function of time;

ivt_ineuron update function of time

t_iCorresponding time data

And last time t_i-1Corrosion rate monitoring value V corresponding to time_i-1Passing through t_iTime output gate

Equation (11), output predicted value V of corrosion rate_iFormula (12):

in the formula:

representing an output gate decision coefficient; sigma represents a neural network activation function, and a sigmoid function is selected; w is a_oRepresenting an output gate weight matrix; b_oA bias value representing an output gate; v_iRepresents the target time t_iThe predicted value of the corrosion rate of (1), mm/a;

2) predicting the corrosion rate V obtained by the prediction_iMinimum required wall thickness t of corroded pipeline_minAnd the average value t of the wall thickness of all measured points_amSubstituting formula (13) to solve pipeline residual life T'_L：

In the formula: t'_LIn order to monitor the residual service life of the pipeline on line, a;

k is a safety coefficient, and when La is less than or equal to L, the value of K is 1; when La is more than L, K is 0.9;

(b) predicting the residual life of the pipeline based on ultrasonic thickness measurement:

1) estimating pipeline corrosion rate V_μFormula (14):

in the formula: v_μFor corrosion rate estimation, mm/a; delta d is the difference of wall thickness before and after the same point thickness measurement, mm; delta T is the time difference before and after thickness measurement, a;

2) solving the pipeline to calculate the wall thickness epsilon formula (15):

in the formula: epsilon is the calculated wall thickness of the steel pipe, mm; p is design pressure, MPa; sigma_sThe yield strength of the steel pipe is MPa;

3) estimating the corrosion rate V_μSubstituting the calculated wall thickness epsilon of the pipeline into a calculation formula (16) to solve the corrosion residual life T ″)_L：

In the formula: t ″)_LMeasuring the residual life of the pipeline based on ultrasonic waves, a;

(b) determining the remaining life T of a pipeline_L：

T_L＝min(T′_L,T″_L) (17)

In the formula: t is_LThe remaining life of the pipeline, a.

And 5: judging the safe service condition of the pipeline:

(a) if T_L≥T_u-T_sIf the pipeline can continue to be in service safely, the pipeline is judged to be in a five-level early warning level, and the early warning mark is displayed to be green;

(b) if T_L＜T_u-T_sFurther judging the corrosion early warning level;

wherein: t is_sRunning time for pipeline production, a; t is_uDesign age, a.

Step 6: setting conditions of corrosion early warning levels:

(a) if T_s<T_L<T_u-T_sIf so, judging the early warning level as four-level, and displaying the early warning mark as blue;

(b) if 10<T_L<T_uIf so, judging the early warning level to be a third-level early warning level, and displaying the early warning mark to be yellow;

(c) if 3<T_L<When 10, judging the early warning level as a secondary early warning level, and displaying an early warning mark as orange;

(d) if T_L<And 3, judging the early warning level as the first-level early warning level, and displaying the early warning mark in red.

And 7: early warning treatment:

(a) for the first-level early warning, the corrosion degree is very serious, and a corrosion pipeline needs to be replaced;

(b) for the secondary early warning, the corrosion degree is serious, and the pipeline needs to stop running and be repaired;

(c) for the third-level early warning, the corrosion degree is relatively heavy, and the pipeline is depressurized, operated and repaired;

(d) for the four-stage early warning, observing the corrosion condition of the pipeline;

(e) for the five-level early warning, the corrosion control condition of the pipeline is better, and the pipeline can be safely produced and operated.

Images