CN117763941B - A method for evaluating gas storage well life based on machine learning - Google Patents

Abstract

A method for assessing the life of a gas well based on machine learning, which collects attribute indicators of the gas well and sets a critical value for each attribute indicator; calculates the importance of each attribute indicator causing the failure of the gas well, and constructs a failure value equation based on the importance and the critical value; corrects the critical value of each attribute indicator based on the detection value and the failure value equation; constructs a failure prediction model based on the change of each attribute indicator, the failure value equation and the corrected critical value, and predicts the residual time period of the failure of the gas well based on the failure prediction model. The method can perform failure prediction on gas wells of various specifications, effectively reduces the hidden dangers caused by failure accidents of gas wells, and can prompt staff to replace gas wells that are about to fail in advance. It also uses a reasonable prediction method of machine learning to avoid the defects of strong subjective arbitrariness in the existing technology, and the predicted life of the gas well is highly accurate.

Description

A method for evaluating gas storage well life based on machine learning

Technical Field

The present invention belongs to the technical field of gas storage well life assessment, and in particular relates to a gas storage well life assessment method based on machine learning.

Background technique

Gas storage wells are well-shaped tubular equipment placed vertically underground for storing compressed air. The main function of gas storage wells is to store compressed natural gas. Because the underground temperature is relatively constant, the natural gas in the well rarely expands and contracts due to changes in the external temperature; and natural gas is safer to store underground. Gas storage wells are used to store gas with working media such as natural gas, nitrogen or inert gas, and air. Gas storage wells are included in the category of fixed pressure vessels in the supervision of special equipment.

In order to ensure the safe operation of gas storage wells, it is necessary to predict the life of gas storage wells so as to prevent failure problems caused by aging of gas storage wells. Just like the gas storage well life monitoring device recorded in the prior art scheme with patent application number "CN202021512322.1" and patent name "A gas storage well life monitoring device in use", the corresponding attribute values are obtained by a probe that collects the performance values of the gas tank, and then the corresponding performance values are manually predicted for the gas storage well life. Such prediction methods are often based on people's subjective understanding, with strong subjective arbitrariness, and the predicted gas storage well life has low accuracy.

Summary of the invention

In order to solve the defects in the prior art, the present invention proposes a method for assessing the life of a gas storage well based on machine learning, which collects attribute indicators of the gas storage well and sets the critical value of each attribute indicator; calculates the importance of each attribute indicator causing the failure of the gas storage well, and constructs a failure value equation based on the importance and the critical value; corrects the critical value of each attribute indicator based on the detection value and the failure value equation; constructs a failure prediction model based on the change of each attribute indicator, the failure value equation and the corrected critical value, and predicts the residual time period of the failure of the gas storage well based on the failure prediction model. Failure prediction can be performed on gas storage wells of various specifications, effectively reducing the hidden dangers caused by failure accidents of gas storage wells, and can prompt staff to replace gas storage wells that are about to fail in advance. Moreover, a reasonable prediction method of machine learning is used to avoid the defects of the prior art that the subjectivity is strong and the predicted life of the gas storage well is highly accurate.

The present invention uses the following technical solutions.

A method for assessing the life of a gas storage well based on machine learning, comprising:

Step 1: Collect the attribute indicators of the gas storage wells and set the critical value of each attribute indicator;

Step 2: Calculate the importance of each attribute index leading to the failure of the gas storage well, and construct a failure value equation based on the importance and critical value;

Step 3: Correct the critical value of each attribute indicator based on the detection value and failure value equation;

Step 4: Construct a failure prediction model based on the variation of each attribute index, the failure value equation and the corrected critical quantity, and predict the residual time period of the gas storage well as the life of the gas storage well based on the failure prediction model.

Preferably, the method of calculating the importance of each attribute index leading to the failure of the gas storage well comprises:

The estimated probability of gas storage well failure caused by the calculated attribute indicators;

The value of the operation attribute indicator is y, which is the probability of causing the gas storage well to fail;

The probability of failure of the gas storage well caused by calculating each attribute index is calculated based on the inference probability and the state probability;

The importance of each attribute indicator in causing the failure of the gas storage well is calculated according to the probability of each attribute indicator causing the failure of the gas storage well.

Preferably, the method for estimating the probability of failure of a gas storage well caused by calculating the attribute index comprises:

Use N gas storage wells as observation objects to calculate importance;

The estimated probability of gas storage well failure caused by attribute index is calculated using the equation: Q(d)=|N _d |/|N|, where Q(d) is the estimated probability, |N _d | is the number of observed objects that fail due to attribute index d, and |N| is the total number of observed objects.

Preferably, the method of calculating the probability of a state causing a gas storage well failure when the value of the computing attribute index is y comprises:

Using the equation: The state probability of causing the gas storage well to fail when the value of the operation attribute index is y, and Q(y|d _j ) in the equation is the state probability of causing the gas storage well to fail when the value of the attribute index d _j is y, /> is the number of observed objects that cause gas well failure when the value of attribute index d _j is y, /> is the number of observed objects whose attribute index d _j causes the gas storage well to fail.

Preferably, the method of calculating the probability of failure of a gas storage well caused by each attribute indicator based on the inference probability and the state probability comprises:

According to the equation: The probability of each attribute indicator causing the failure of the gas storage well is calculated. In the equation, j is the pre-set sequence code of each attribute indicator, O is the number of attribute indicators, y is the value of d _j , Z(j) is the probability of each attribute indicator causing the failure of the gas storage well, Q(d) is the estimated probability of the attribute indicator causing the failure of the gas storage well, and Q(y|d _j ) is the state probability of causing the failure of the gas storage well when the value of the attribute indicator d _j is y.

Preferably, the method of calculating the importance of each attribute indicator causing the failure of the gas storage well according to the probability of each attribute indicator causing the failure of the gas storage well comprises:

Using the equation: The importance of each attribute indicator causing the failure of the gas storage well is calculated. In the equation, _Zj is the probability of each attribute indicator causing the failure of the gas storage well, j is the pre-set sequence code of each attribute indicator, O is the number of attribute indicators, and _Xj is the importance of each attribute indicator causing the failure of the gas storage well.

Preferably, the method of constructing a failure value equation based on importance and critical quantity comprises:

The failure value equation is constructed as: In the equation, A is the failure value, j is the pre-set sequence code of each attribute indicator, O is the number of attribute indicators, _Xj is the importance of each attribute indicator leading to the failure of the gas storage well, _εj is the critical value of each attribute indicator, and _yj is the current value of the attribute indicator.

Preferably, the method of calculating the probability of failure of a gas storage well caused by each attribute indicator based on the inference probability and the state probability comprises:

Obtaining the value of the attribute index of the failed gas storage well as a detection value;

Calculate the failure values according to the detection values and the collected time sequence;

When the failure value is lower than zero and gradually approaches zero, the critical amount of each attribute index is gradually configured until the failure value is zero;

The critical amount of each attribute indicator after configuration is regarded as the critical amount of the corrected attribute indicator.

Preferably, it also comprises:

Calculate the average change of attribute indicators after C collection intervals

use Represents the change in the attribute index after C collection intervals.

Each collection of attribute index values carries a collection interval (i.e., the time interval between two adjacent collections of attribute index values), the interval is set to U, and the change of an attribute index d in the attribute index group in the interval U is defined as Δy. The value of the attribute index is not increased in proportion to the first power. To more accurately predict the value of the attribute index after the interval U, the average of the last C changes can be obtained, so that the value of the attribute index after the interval U can be more accurately predicted. The equation is: The equation contains a possible change in the attribute index after time interval U. If the current value of the attribute index d is y ₁ and the value of d after time interval U is y ₂ , then the value of d after time interval U is:/> Through the above method, the value of the attribute index d that can be predicted minutes later is:/>

Preferably, the method for constructing a failure prediction model according to the variation of each attribute index, the failure value equation and the corrected critical value comprises:

The failure prediction model is constructed as: In the equation, A is the failure value, j is the pre-set sequence code of each attribute indicator, O is the number of attribute indicators, _Xj is the importance of each attribute indicator leading to gas storage well failure, _εj is the critical value of each attribute indicator, _yj is the current value of the attribute indicator,/> It is the change in the attribute index after C collection intervals.

Preferably, the method for predicting the size of the residual period of the gas storage well life span in which the gas storage well fails according to the failure prediction mode comprises:

Calculate the corresponding failure value at each collection time point;

constructing a regression line based on the calculated failure values and corresponding collection time points;

Calculate the mean of failure values in a pre-set period;

The residual time period for failure of the gas storage well is calculated based on the failure value mean and the regression line.

Preferably, the method of calculating the corresponding failure value at each collection time point comprises:

Collect the values of the attribute indicators of the gas storage well again and record the collection time point;

The corresponding failure value at each collection time point is calculated according to the failure prediction model.

Preferably, the method for predicting the size of the residual period of the gas storage well life span in which the gas storage well fails according to the failure prediction mode comprises:

A regression line is constructed based on the calculated failure values and the corresponding collection time points.

Preferably, the method for calculating the mean number of failure values within a preset time period comprises:

Set the variable time interval M according to the current time point;

Sum up the average of each attribute index in the variable time interval L;

The failure value mean is calculated based on the mean of each attribute index and the failure prediction mode.

Preferably, the method of calculating the mean failure value based on the mean of each attribute index and the failure prediction mode comprises:

Calculate the collection time point corresponding to the mean failure value based on the mean failure value and the regression line;

The residual time period for failure of the gas storage well is calculated based on the collection time points corresponding to the failure value mean and the regression line.

Preferably, the method of calculating the residual time period of failure of a gas storage well based on the collection time points corresponding to the mean of the failure values and the regression line comprises:

The time point at which the failure value calculated based on the regression line is zero.

Preferably, the method for setting the variable time interval M according to the current time point comprises:

Set the variable time interval M to {current time point - M/2} to {current time + M/2}.

Preferably, the method for setting the critical amount of each attribute indicator includes:

Obtain a preset number of failed gas storage wells;

Sum up the values of each attribute index when each gas storage well fails;

Total the number of identical values of each attribute indicator;

The value with the highest number of identical values for each attribute indicator is regarded as the critical value of the corresponding attribute indicator.

A gas storage well life assessment device based on machine learning, comprising:

A setting module, which is used to collect attribute indicators of gas storage wells and set critical quantities of various attribute indicators;

A calculation module, which is used to calculate the importance of each attribute index leading to the failure of the gas storage well, and construct a failure value equation based on the importance and the critical amount;

A correction module, which is used to correct the critical amount of each attribute indicator according to the detection value and the failure value equation;

The prediction module is used to construct a failure prediction model based on the change amount of each attribute indicator, the failure value equation and the corrected critical amount, and predict the residual time period of the gas storage well as the life of the gas storage well based on the failure prediction model.

The beneficial effects of the present invention are that, compared with the prior art, the present invention collects attribute indicators of gas storage wells and sets critical quantities for each attribute indicator; calculates the importance of each attribute indicator leading to failure of the gas storage well, and constructs a failure value equation based on the importance and the critical quantity; corrects the critical quantity of each attribute indicator based on the detection value and the failure value equation; constructs a failure prediction model based on the change of each attribute indicator, the failure value equation and the corrected critical quantity, and predicts the residual time period of failure of the gas storage well based on the failure prediction model. Failure prediction can be performed on gas storage wells of various specifications, effectively reducing the hidden dangers caused by failure accidents of gas storage wells, and can prompt staff to replace gas storage wells that are about to fail in advance. Moreover, a reasonable prediction method of machine learning is used to avoid the defects of strong subjective arbitrariness in the prior art, and the predicted life of the gas storage well is highly accurate.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of a method for evaluating the life of a gas storage well based on machine learning according to the present invention;

FIG2 is a structural diagram of the gas storage well life assessment device based on machine learning described in the present invention.

Detailed ways

In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely expressed in combination with the drawings in the embodiments of the present invention. The embodiments expressed in this application are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.

As shown in FIG1 , a method for assessing the life of a gas storage well based on machine learning according to the present invention is run on a prediction terminal such as a computer, and includes:

Step 1: Collect the attribute indicators of the gas storage wells and set the critical value of each attribute indicator;

Step 2: Calculate the importance of each attribute index leading to the failure of the gas storage well, and construct a failure value equation based on the importance and critical value;

Step 3: Correct the critical value of each attribute indicator based on the detection value and failure value equation;

Step 4: Construct a failure prediction model based on the variation of each attribute index, the failure value equation and the corrected critical value, and predict the residual time period of the gas storage well as the life of the gas storage well according to the failure prediction model. That is to say, initially select the attribute index, set the critical value of each attribute index, calculate the importance of each attribute index, and perform the correction of the critical value, then calculate the variation of each attribute index, construct the failure prediction model based on the variation and the failure equation, and finally predict the residual time period of the gas storage well through the time interval model.

In this way, failure prediction can be performed on gas storage wells of various specifications, effectively reducing the hidden dangers caused by failure accidents of gas storage wells, and can prompt staff to replace gas storage wells that are about to fail in advance. The reasonable prediction method of machine learning is used to avoid the subjective and arbitrary defects of existing technologies, and the predicted life of gas storage wells is highly accurate.

In a preferred but non-limiting embodiment of the present invention, the method of calculating the importance of each attribute index leading to the failure of the gas storage well comprises:

The estimated probability of gas storage well failure caused by the calculated attribute indicators;

The value of the operation attribute indicator is y, which is the probability of causing the gas storage well to fail;

The probability of failure of the gas storage well caused by calculating each attribute index is calculated based on the inference probability and the state probability;

The importance of each attribute indicator in causing the failure of the gas storage well is calculated according to the probability of each attribute indicator causing the failure of the gas storage well.

In a preferred but non-limiting embodiment of the present invention, the method for estimating the probability of failure of a gas storage well caused by the computational attribute index comprises:

The importance is calculated by using N gas storage wells as observation objects; the observation objects can include gas storage wells that have failed in the past.

The equation: Q(d)＝|N _d |/|N| is used to calculate the estimated probability of the failure of the gas storage well caused by the attribute indicator. In the equation, Q(d) is the estimated probability, |N _d | is the number of observation objects that fail due to the attribute indicator d, and |N| is the total number of observation objects. The probability of collecting N observation objects to estimate the failure of the gas storage well caused by the attribute indicator is defined as Q(d _j ). The observation object N _d is used to represent the object group formed by the observation object that fails due to the attribute indicator d. Under the condition of sufficient independent observation objects of the same specification, the value of the prior probability is calculated as: Q(d)＝|N _d |/|N|. In the equation, Q(d) is the estimated probability, |N _d | is the number of observation objects that fail due to the attribute indicator d, and |N| is the total number of observation objects. The attribute indicators include: the well wall thickness indicator at different heights of the gas storage well measured by several thickness gauges equidistantly arranged from top to bottom in the gas storage well, the air pressure indicator at different heights in the gas storage well measured by several pressure sensors equidistantly arranged from top to bottom in the gas storage well, the strength indicator at different heights in the gas storage well measured by several strength testers equidistantly arranged from top to bottom on the inner wall of the gas storage well, and the soil moisture at different heights in the soil where the gas storage well is located measured by several humidity sensors equidistantly arranged from top to bottom along the outer wall of the gas storage well in the soil where the gas storage well is located. The thickness gauge, pressure sensor, humidity sensor and strength tester are all connected to the PLC or the single-chip microcomputer, and the PLC or the single-chip microcomputer is also connected to the 4G module. The PLC or the single-chip microcomputer can transmit the values measured and transmitted by the thickness gauge, pressure sensor, humidity sensor and strength tester to the prediction terminal in the 4G network via the 4G module to execute the prediction.

In a preferred but non-limiting embodiment of the present invention, the method for calculating the probability of a state causing a gas storage well failure when the value of the attribute index is y comprises:

Using the equation: The state probability of causing the gas storage well to fail when the value of the operation attribute index is y, and Q(y|d _j ) in the equation is the state probability of causing the gas storage well to fail when the value of the attribute index d _j is y, /> is the number of observed objects that cause gas well failure when the value of attribute index d _j is y, /> is the number of observed objects whose attribute index d _j causes the gas storage well to fail. The probability of attribute index causing the gas storage well to fail is defined as Q(d _j ), then the inferred state probability of each attribute index can be represented as Q(y|d _j ), where Q(y|d _j ) is the attribute index that causes the gas storage well to fail, and the probability that the value _{of d j is} y is Q(y|d _j ), using/> Represents the observed object that causes the gas storage well to fail when the attribute index in N _d is d _j and the value of d _j is y, then the state probability can be inferred as:/>

In a preferred but non-limiting embodiment of the present invention, the method for calculating the probability of failure of a gas storage well caused by calculating each attribute indicator based on the inference probability and the state probability comprises:

According to the equation: The probability of each attribute indicator causing the failure of the gas storage well is calculated. In the equation, j is the pre-set sequence code of each attribute indicator, O is the number of attribute indicators, y is the value of d _j , Z(j) is the probability of each attribute indicator causing the failure of the gas storage well, Q(d) is the estimated probability of the attribute indicator causing the failure of the gas storage well, and Q(y|d _j ) is the state probability of causing the failure of the gas storage well when the value of the attribute indicator d _j is y. The situation that often causes the failure of a gas well is a phenomenon that is mapped out by more than one attribute indicator in the attribute indicator group. Under normal circumstances, if the failure of a gas well is caused by one attribute indicator, it is easy to determine whether the gas well will fail. If the failure of a gas well is caused by several attribute indicators, it is difficult to determine when the gas well will fail, because the values of various attribute indicators often do not reach the indicator critical value ε, but the gas well often has insufficient function, gas leakage and other conditions, thus presenting a failure situation; to overcome the phenomenon that several attribute indicators cause the failure of a gas well at the same time, the state setting of the states between the attribute indicators is independent of each other, that is, each attribute indicator in the attribute indicator group is independent, and the probability of the failure of a gas well mapped by one indicator in this attribute indicator group is: The value of Z(j) can be represented as an attribute indicator in the attribute indicator group, which maps out the probability of failure of the gas storage well. In the equation, y represents the value of the attribute indicator d. If y is higher than zero and lower than the critical value, the above Q(d) and Q(y|d _j ) can be sent into the equation to obtain the probability of failure of the gas storage well caused by a single attribute indicator.

In a preferred but non-limiting embodiment of the present invention, the method of calculating the importance of each attribute indicator causing the failure of the gas storage well according to the probability of each attribute indicator causing the failure of the gas storage well comprises:

Using the equation: Calculate the importance of each attribute index leading to the failure of the gas storage well. In the equation, _Zj is the probability of each attribute index leading to the failure of the gas storage well, j is the pre-set sequence code of each attribute index, O is the number of attribute indexes, and _Xj is the importance of each attribute index leading to the failure of the gas storage well. Through the above calculations, the probability of each attribute index leading to the failure of the gas storage well can be obtained, each defined as: _Z1 , _Z2 ..., and the importance of each attribute index that can be mapped to the failure when the gas storage well fails can be calculated:/>

In a preferred but non-limiting embodiment of the present invention, the method for constructing a failure value equation based on importance and critical quantity comprises:

The failure value equation is constructed as: In the equation, A is the failure value, j is the pre-set sequence code of each attribute index, O is the number of attribute indexes, _Xj is the importance of each attribute index leading to the failure of the gas storage well, _εj is the critical value of each attribute index, and _yj is the current value of the attribute index (the current value can be the value of the corresponding attribute index currently transmitted by the controller). Through the above calculations, the probability of each attribute index leading to the failure of the gas storage well is obtained, so the failure value equation for detecting the failure of the gas storage well is:/> The failure value equation can be used to calculate the current failure value of the gas storage well according to the current value and critical value of each attribute index. When the condition of the gas storage well is normal (that is, not failed), since the values of the attribute indicators are all lower than the critical value ε _j at this time, A<0. Therefore, when the value of A gradually approaches zero, the condition of the gas storage well is in the process of changing from normal to failure.

In a preferred but non-limiting embodiment of the present invention, the method for calculating the probability of failure of a gas storage well caused by calculating each attribute indicator based on the inference probability and the state probability comprises:

Obtaining the value of the attribute index of the failed gas storage well as a detection value;

Calculate the failure values according to the detection values and the collected time sequence;

When the failure value is lower than zero and gradually approaches zero, the critical amount of each attribute index is gradually configured until the failure value is zero;

The critical value of each attribute index after configuration is regarded as the critical value of the modified attribute index. The method of setting the critical value of the attribute index is very simple, which will make the accuracy of the detected value not high, so the critical value needs to be corrected to improve the prediction accuracy of the gas storage well; select Y gas storage wells with normal functional conditions, and collect the value of the attribute index of a single gas storage well at a time interval U; use the above equation to calculate the failure value of the attribute index value after each collection, and the value of the failure value after a period of time will become close to zero or higher than zero through A＜0. In this process, the following three types of situations often occur:

When A<0, the gas storage well fails; when A=0, the gas storage well fails; when A>0, the gas storage well fails.

In the case of A＜0 and A＞0, it is attributed to the error in the prediction of gas well failure. It is necessary to configure the critical value according to the value of the attribute index collected when the gas well fails, so that the failure value is maximized and close to zero. Only the failure prediction of such gas wells can be more accurate. The critical value correction method is to extract the value of the attribute index of the failed gas well from the value of the attribute index of the gas well collected above and define it as the benchmark F. The above equation is used to calculate the failure value in the benchmark F one by one according to the time sequence of the attribute index value collection. In the process of calculating according to the benchmark data time sequence, the failure value A gradually approaches zero after the initial A＜0, and the (critical value) of each attribute index is gradually configured to meet the situation when the gas well fails, and the value at this time is very close to zero.

In a preferred but non-limiting embodiment of the present invention, it also comprises:

Calculate the average change of attribute indicators after C collection intervals

use Represents the change in the attribute index after C collection intervals.

Each collection of attribute index values carries a collection interval (i.e., the time interval between two adjacent collections of attribute index values), the interval is set to U, and the change of an attribute index d in the attribute index group in the interval U is defined as Δy. The value of the attribute index is not increased in proportion to the first power. To more accurately predict the value of the attribute index after the interval U, the average of the last C changes can be obtained, so that the value of the attribute index after the interval U can be more accurately predicted. The equation is: The equation contains a possible change in the attribute index after time interval U. If the current value of the attribute index d is y ₁ and the value of d after time interval U is y ₂ , then the value of d after time interval U is:/> Through the above method, the value of the attribute index d that can be predicted minutes later is:/>

In a preferred but non-limiting embodiment of the present invention, a method for constructing a failure prediction model based on the variation of each attribute index, the failure value equation and the corrected critical value includes:

The failure prediction model is constructed as: In the equation, A is the failure value, j is the pre-set sequence code of each attribute indicator, O is the number of attribute indicators, _Xj is the importance of each attribute indicator leading to gas storage well failure, _εj is the critical value of each attribute indicator, _yj is the current value of the attribute indicator,/> is the change in the attribute index after C collection intervals. The change in the attribute index is obtained through the above method. The same method can be used to predict the value of any attribute index in the attribute index group, and the value of A after N×U can also be predicted. Therefore, the failure prediction mode of the gas storage well can be characterized as:/> When N=0, the above equation obtains the actual failure value at the current time point, and when N＞0, the above equation obtains the predicted failure value after N×U.

In a preferred but non-limiting embodiment of the present invention, the method for predicting the residual period of the gas storage well life as the failure of the gas storage well according to the failure prediction mode comprises:

Calculate the corresponding failure value at each collection time point; the collection time point is the time point when the PLC or single-chip microcomputer transmits the value of the attribute indicator to the prediction terminal.

constructing a regression line based on the calculated failure values and corresponding collection time points;

Calculate the mean of failure values in a pre-set period;

The residual time period for failure of the gas storage well is calculated based on the failure value mean and the regression line.

In a preferred but non-limiting embodiment of the present invention, the method of calculating the corresponding failure value of each collection time point comprises:

Collect the values of the attribute indicators of the gas storage well again and record the collection time point;

The corresponding failure value at each collection time point is calculated according to the failure prediction model.

The value of the attribute index is collected once every period of time, the values of the attribute index of a cluster of gas storage wells are collected continuously, and the time point when the attribute index values are collected is registered as u; the collected attribute index values are applied to the above failure prediction model to obtain the corresponding failure value, and the time point u can be associated with the failure value A of the gas storage well through the collection and calculation of the attribute index values of a period of time, and the failure value _AN after a period of time can be calculated through the above failure prediction model.

In a preferred but non-limiting embodiment of the present invention, a method for predicting the residual period of the gas storage well life as a failure of the gas storage well according to the failure prediction mode comprises:

A regression line is constructed based on the calculated failure values and the corresponding collection time points.

Through the relationship between the above failure value and the collection time point, a regression line can be obtained by the least square method. The value of the X-axis of the rectangular coordinate system is u, and the value of the Y-axis of the rectangular coordinate system is A.

In a preferred but non-limiting embodiment of the present invention, the method for calculating the mean number of failure values within a preset time period comprises:

Set the variable time interval M according to the current time point;

Sum up the average of each attribute index in the variable time interval L;

The failure value mean is calculated based on the mean of each attribute index and the failure prediction mode.

The value of the attribute index of the gas storage well increases with the increase of usage time, but the increase of the index value is not a one-time proportional increase. It will have a significant deviation to predict the remaining time before the gas storage well fails by only one point (u _j , A _j ).

The present application uses the value of the attribute index in a time period to calculate the size of the residual period before the failure occurs. The size of the residual period can be regarded as a time interval, that is, a variable time interval. The size of the time interval is M. The variable time interval is added to the above regression line; if the value of the attribute index is collected L times in the variable time interval M, then the mean of an attribute index is: The above prediction model can be equivalent to:

According to this equation, the mean failure value can be calculated/>

In a preferred but non-limiting embodiment of the present invention, the method of calculating the mean failure value based on the mean of each attribute index and the failure prediction mode comprises:

Calculate the collection time point corresponding to the mean failure value based on the mean failure value and the regression line;

The residual time period for failure of the gas storage well is calculated based on the collection time points corresponding to the failure value mean and the regression line.

In a preferred but non-limiting embodiment of the present invention, the method of calculating the residual time period of failure of a gas storage well based on the collection time points corresponding to the mean of the failure values and the regression line comprises:

The time point when the failure value calculated according to the regression line is zero;

The difference between the time point when the failure value is zero and the corresponding collection time point of the failure value mean is used as the residual time period of the gas storage well failure. According to the failure value mean obtained by the above calculation, the corresponding time point can be obtained through the regression line, so the time point and failure value point in a variable time interval can be obtained: At the time /> The time span required during the movement toward point ( _uk , 0) (this point is the time point when the gas storage well fails) is the time span remaining before the gas storage well fails; the time point when the failure value of the gas storage well is zero can be obtained through the regression line, and the time span from the time point of the mean failure value to the time point when the failure value is zero is the time span remaining before the gas storage well fails; the value of the gas storage well attribute index collected at any time point, at this time the failure value of the gas storage well failure is A, if the time span of the gas storage well failure is predicted, the point (u, A) should be sent to the variable time interval M, where the variable time interval is uM/2 to u+M/2, and the corresponding failure value of the attribute index collected in the variable time interval M is calculated using the equation prediction model as the basis/> The time point corresponding to the regression line can be obtained/> In/> The required time period is the residual time period before the gas storage well fails.

In a preferred but non-limiting embodiment of the present invention, the method for setting the variable time interval M according to the current time point comprises:

Set the variable time interval M to {current time point - M/2} to {current time + M/2}.

In a preferred but non-limiting embodiment of the present invention, the method for setting the critical amount of each attribute indicator includes:

Obtain a preset number of failed gas storage wells;

Sum up the values of each attribute index when each gas storage well fails;

Total the number of identical values of each attribute indicator;

The value with the highest number of identical values for each attribute indicator is regarded as the critical value of the corresponding attribute indicator.

There are a set number of failed gas storage wells in the observation object. If the number of observation objects is high enough, if only one indicator in the attribute indicator set is considered and the failure of the gas storage well is determined by the indicator d, and the value of the indicator d is y when the registered gas storage well fails; when the value of y is not high, it shows that the number of gas storage wells registered as failed is not high at this time, otherwise when the value of y is high, the number of gas storage wells registered as failed is also not high, because most of the gas storage wells have failed before reaching this value. Through the above analysis, it can be obtained that the indicator value and the number of failures of the gas storage well have some relationship, which is: when the value of y is closer to the mean, the number of failed gas storage wells is higher. When y=ν, the corresponding number of failed gas storage wells is _Ozg , and the number of failures at this time is the highest, which can also map out the critical value of the attribute indicator. This method can roughly identify the critical value of each attribute indicator.

By applying the solution of the present invention, failure prediction can be performed on gas storage wells of various specifications, effectively reducing the hidden dangers of gas storage well leakage and server downtime caused by gas storage well failure, and can prompt staff in advance to replace gas storage wells that are about to fail.

As shown in FIG2 , a gas storage well life assessment device based on machine learning according to the present invention comprises:

A setting module, which is used to collect attribute indicators of gas storage wells and set critical quantities of various attribute indicators;

A calculation module, which is used to calculate the importance of each attribute index leading to the failure of the gas storage well, and construct a failure value equation based on the importance and the critical amount;

A correction module, which is used to correct the critical amount of each attribute indicator according to the detection value and the failure value equation;

The prediction module is used to construct a failure prediction model based on the change amount of each attribute indicator, the failure value equation and the corrected critical amount, and predict the residual time period of the gas storage well as the life of the gas storage well based on the failure prediction model.

The beneficial effects of the present invention are that, compared with the prior art, the present invention collects attribute indicators of gas storage wells and sets critical quantities for each attribute indicator; calculates the importance of each attribute indicator leading to failure of the gas storage well, and constructs a failure value equation based on the importance and the critical quantity; corrects the critical quantity of each attribute indicator based on the detection value and the failure value equation; constructs a failure prediction model based on the change of each attribute indicator, the failure value equation and the corrected critical quantity, and predicts the residual time period of failure of the gas storage well based on the failure prediction model. Failure prediction can be performed on gas storage wells of various specifications, effectively reducing the hidden dangers caused by failure accidents of gas storage wells, and can prompt staff to replace gas storage wells that are about to fail in advance. Moreover, a reasonable prediction method of machine learning is used to avoid the defects of strong subjective arbitrariness in the prior art, and the predicted life of the gas storage well is highly accurate.

The present disclosure can be a system, method and/or computer program product. The computer program product can include a computer-readable backup medium carrying computer-readable program instructions for causing a processor to achieve each aspect of the present disclosure.

The computer readable backup medium can be a tangible network capable of retaining and backing up the instructions used by the instruction execution network. The computer readable backup medium can be, but is not limited to, an electrical backup network, a magnetic backup network, an optical backup network, an electromagnetic backup network, a semiconductor backup network, or any suitable combination of the above. Further examples of computer readable backup media (non-exhaustive list) include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disk read-only memory (HD-ROM), digital versatile disk (DXD), memory stick, floppy disk, mechanically encoded network, such as punch cards or protrusions in grooves with instructions backed up thereon, and any suitable combination of the above. Computer readable backup media as used herein is not to be construed as transient signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through power line cables), or electrical signals transmitted through wires.

The computer readable program instructions expressed herein can be downloaded from a computer readable backup medium to each inference/processing grid line, or downloaded to an external computer or external backup grid line through a network, such as the Internet, a local area network, a wide area network, and/or a wireless network. The network can include copper transmission cables, power line transmission, wireless transmission, routers, firewalls, switches, gateway computers, and/or edge service devices. The network adapter card or network interface in each inference/processing grid line receives the computer readable program instructions from the network and forwards the computer readable program instructions for storage in the computer readable backup medium in each inference/processing grid line.

The computer program instructions for performing the operations of the present disclosure can be assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-associated instructions, microcode, firmware instructions, conditional setting values, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Sdaletal A, H++, etc., and conventional procedural programming languages such as "H" language or similar programming languages. The computer-readable program instructions can be executed entirely on the client computer, partially on the client computer, as a separate software package, partially on the client computer and partially on the remote computer, or completely on the remote computer or server. In the case of a remote computer, the remote computer can be connected to the client computer through any other network, including a local area network (LAb) or a wide area network (WAb), or can be connected to an external computer (just like using an Internet service provider to connect through the Internet). In some embodiments, each aspect of the present disclosure is achieved by customizing an electronic circuit by using state values of computer-readable program instructions, such as a programmable logic circuit, a field programmable gate array (FPGA) or a programmable logic array (PLA), which can execute computer-readable program instructions.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not deviate from the spirit and scope of the present invention should be included in the protection scope of the claims of the present invention.

Claims (5)

1. A method for assessing the life of a gas storage well based on machine learning, comprising: Step 1: Collect the attribute indicators of the gas storage wells and set the critical value of each attribute indicator; Step 2: Calculate the importance of each attribute index leading to the failure of the gas storage well, and construct a failure value equation based on the importance and critical value; Step 3: Correct the critical value of each attribute indicator based on the detection value and failure value equation; Step 4: construct a failure prediction model based on the variation of each attribute index, the failure value equation and the corrected critical quantity, and predict the residual time period of the gas storage well as the life of the gas storage well based on the failure prediction model; The method for calculating the importance of each attribute index leading to gas storage well failure includes: The estimated probability of gas storage well failure caused by the calculated attribute indicators; The value of the operation attribute indicator is y, which is the probability of causing the gas storage well to fail; The probability of failure of the gas storage well caused by calculating each attribute index is calculated based on the inference probability and the state probability; The importance of each attribute indicator causing the failure of the gas storage well is calculated according to the probability of each attribute indicator causing the failure of the gas storage well; The method for calculating the estimated probability of gas storage well failure caused by the attribute indicator includes: Use N gas storage wells as observation objects to calculate importance; The estimated probability of failure of a gas storage well caused by the attribute index is calculated using the equation: Q(d) = |N _d |/|N|, where Q(d) is the estimated probability, |N _d | is the number of observed objects that fail due to the attribute index d, and |N| is the total number of observed objects; The method for calculating the state probability of causing the gas storage well to fail when the value of the attribute indicator is y includes: Using the equation: The state probability of causing the gas storage well to fail when the value of the operation attribute index is y, and Q(y|d _j ) in the equation is the state probability of causing the gas storage well to fail when the value of the attribute index d _j is y, /> is the number of observed objects that cause gas well failure when the value of attribute index d _j is y, /> is the number of observed objects whose attribute indicator d _j causes the gas storage well to fail, and y is the value of d _j ; The method for calculating the probability of failure of a gas storage well caused by each attribute indicator based on the inference probability and the state probability includes: According to the equation: Calculate the probability of each attribute index causing the gas storage well to fail, where j in the equation is the pre-set sequence code of each attribute index, O is the number of attribute indexes, y is the value of d _j , Z(j) is the probability of each attribute index causing the gas storage well to fail, Q(d) is the estimated probability of the attribute index causing the gas storage well to fail, and Q(y|d _j ) is the state probability of causing the gas storage well to fail when the value of the attribute index d _j is y; The method for calculating the importance of each attribute indicator causing the failure of the gas storage well according to the probability of each attribute indicator causing the failure of the gas storage well includes: Using the equation: Calculate the importance of each attribute indicator leading to the failure of the gas storage well, where Z(j) is the probability of each attribute indicator leading to the failure of the gas storage well, j is the pre-set sequence code of each attribute indicator, O is the number of attribute indicators, and _Xj is the importance of each attribute indicator leading to the failure of the gas storage well; The method of constructing the failure value equation based on importance and critical quantity includes: The failure value equation is constructed as: In the equation, A is the failure value, j is the pre-set sequence code of each attribute indicator, O is the number of attribute indicators, _Xj is the importance of each attribute indicator leading to gas storage well failure, _εj is the critical value of each attribute indicator, and _yj is the current value of the attribute indicator; According to the detection value and failure value equation, the critical value of each attribute indicator is corrected, including: Obtaining the value of the attribute index of the failed gas storage well as a detection value; Calculate the failure values according to the detection values and the collected time sequence; When the failure value is lower than zero and gradually approaches zero, the critical amount of each attribute index is gradually configured until the failure value is zero; The critical amount of each attribute indicator after configuration is regarded as the critical amount of the corrected attribute indicator. 2. The method for assessing the life of a gas storage well based on machine learning according to claim 1, further comprising: Calculate the average change of attribute indicators after C collection intervals use Represents the change in the attribute index after C collection intervals. 3. The method for evaluating the life of a gas storage well based on machine learning according to claim 1 is characterized in that the method for constructing a failure prediction model according to the variation of each attribute index, the failure value equation and the corrected critical value comprises: The failure prediction model is constructed as: In the equation, A is the failure value, j is the pre-set sequence code of each attribute indicator, O is the number of attribute indicators, _Xj is the importance of each attribute indicator leading to gas storage well failure, _εj is the critical value of each attribute indicator, _yj is the current value of the attribute indicator,/> is the change in the attribute index after C collection intervals; A method for predicting the size of the residual period of the gas storage well life when the gas storage well fails according to the failure prediction model includes: Calculate the corresponding failure value at each collection time point; Construct a regression line based on the calculated failure values and the corresponding collection time points; Calculate the mean of failure values in a pre-set period; The residual time period of gas storage well failure is calculated based on the mean of failure values and the regression line; The method for calculating the corresponding failure value at each collection time point includes: Collect the values of the attribute indicators of the gas storage well again and record the collection time point; The corresponding failure value at each collection time point is calculated according to the failure prediction model. 4. The method for assessing the life of a gas storage well based on machine learning according to claim 1, wherein the method for predicting the residual period of the gas storage well life as the failure occurrence according to the failure prediction mode comprises: Construct a regression line based on the calculated failure values and the corresponding collection time points; The method for calculating the mean of failure values within a preset period of time includes: Set the variable time interval M according to the current time point; The average of each attribute index in the total variable time interval L; Calculate the mean failure value based on the mean of each attribute index and the failure prediction mode; The method for calculating the mean failure value according to the mean of each attribute index and the failure prediction mode includes: Calculate the collection time point corresponding to the mean failure value based on the mean failure value and the regression line; The residual time period for failure of the gas storage well is calculated based on the corresponding collection time points of the mean failure value and the regression line. 5. The method for assessing the life of a gas storage well based on machine learning according to claim 4 is characterized in that the method for calculating the residual time period of failure of the gas storage well based on the collection time points corresponding to the mean of the failure values and the regression line comprises: The time point when the failure value calculated according to the regression line is zero; The method of setting the variable time interval M according to the current time point includes: Set the variable time interval M to {current time point - M/2} to {current time + M/2}; The method of setting the critical amount of each attribute indicator includes: Obtain a preset number of failed gas storage wells; Sum up the values of each attribute index when each gas storage well fails; Total the number of identical values of each attribute indicator; The value with the highest number of identical values for each attribute indicator is regarded as the critical value of the corresponding attribute indicator.