CN114358215A - A wellbore fluid detection method based on deep anomaly detection - Google Patents

Abstract

The invention discloses a shaft effusion detection method based on depth anomaly detection, which comprises the steps of firstly obtaining SCADA production high-frequency data and reducing the dimension of the SCADA production high-frequency data; then carrying out feature fusion on the SCADA production high-frequency data subjected to dimension reduction, the A2 data and the geological parameter data; secondly, performing data modeling and training by using the fused features, and calculating a prediction error; and finally, calculating a dynamic threshold according to the obtained prediction error, and judging whether the shaft accumulates liquid according to the dynamic threshold. According to the scheme, the SCADA second-level data is used as the characteristics, so that the model only focuses on data fluctuation between days, and the data fluctuation condition within the day is also considered, and more subtle data change can be captured. And the experimental result also proves that the data fluctuation situation in the day can better reflect the concrete actual production situation, and the method of reducing the dimension firstly is adopted in the process of processing the SCADA high-frequency data, so that the data modeling of the service model is better realized, and the occurrence of dimension disaster is also avoided.

Description

A wellbore fluid detection method based on deep anomaly detection

technical field

The invention relates to the field of gas well development, in particular to a wellbore fluid detection method based on depth abnormality detection.

Background technique

Liquid accumulation in gas wells refers to the phenomenon that liquid accumulates in the wellbore because the gas cannot effectively carry the liquid out. During the production process of a gas well, the gas-liquid two phases flow out from the formation and are produced from the surface through the wellbore. In the early stage of production, the gas production of gas wells is high, and the gas-liquid two phases flow upward in an annular flow, and the liquid is carried in two ways: liquid droplets entrained in the gas core and liquid film attached to the pipe wall. As the formation pressure decreases, the gas production of the gas well decreases, resulting in the inversion of the liquid (droplet/liquid film) flow in the wellbore, which cannot be brought out of the surface, resulting in liquid accumulation. Field well testing operations show that the fluid accumulation in the wellbore leads to a significant increase in the wellbore pressure gradient, which increases the rate of decline in production and affects the ultimate recovery of gas wells. Therefore, it is of great significance to accurately predict the liquid accumulation time of gas wells and take timely drainage and gas production measures to maintain stable production of low-yield gas wells. At present, there are still the following problems in the prediction and detection of gas production process and liquid accumulation:

1. The timing of implementation of the gas extraction process mainly depends on experience, and more precise measures and timing guidance are urgently needed. If the running time is too early, the wellbore pressure is too high, and the risk of pressure-bearing operation is high; if the running time is too late, the gas well has poor liquid-carrying capacity, and it is easy to accumulate liquid, which affects the later production capacity. Uniform standards are difficult to fully apply.

2. The mechanism of two-phase flow is complex, and the results of each model vary greatly, which makes it difficult to guide the implementation of measures. The wellbore trajectory is complex, the two-phase flow simulation is difficult, and it is difficult to accurately calculate the wellbore pressure and temperature distribution; there are many critical liquid-carrying models, the model results vary greatly, and the adaptability is poor, so it is difficult to accurately guide the implementation timing of the liquid drainage process.

3. At present, there are many studies on fluid accumulation in gas wells, but there is no consensus on its mechanism. There is a large deviation between the calculated values of different prediction models, which leads to the lack of effective guidance when designing the drainage and production process in the field. The root cause is that the modeling of each mechanism model considers a single influencing factor and lacks the comparison with the actual gas well production performance.

4. SCADA collects a lot of high-frequency production data, but fails to use it effectively. At present, although some data mining projects have been carried out in the industry, the processed low-frequency static data is still mainly used, and the high-frequency production data has not been fully utilized.

SUMMARY OF THE INVENTION

In view of the above deficiencies in the prior art, the present invention provides a wellbore fluid detection method based on depth abnormality detection.

In order to achieve the above-mentioned purpose of the invention, the technical scheme adopted in the present invention is:

A method for detecting wellbore fluid accumulation based on deep anomaly detection, comprising the following steps:

S1. Obtain high-frequency data of SCADA production and reduce its dimension;

S2. Feature fusion of SCADA production high-frequency data after dimensionality reduction with A2 data and geological parameter data;

S3, using the features fused in step S2 to perform data modeling and training, and calculate the prediction error;

S4. Calculate the dynamic threshold value according to the prediction error obtained in step S3, and judge whether the wellbore is fluid according to the dynamic threshold value.

The beneficial effect of the above solution is that the present invention uses SCADA second-level data as a feature in the prediction of the normal data trend in the next time period of wellbore fluid accumulation, so that the model not only pays attention to the data fluctuation between days, Taking into account the fluctuation of data within the day, it can capture more subtle data changes. And the experimental results also prove that the fluctuation of data within the day can better reflect the actual production situation. The invention adopts the method of reducing the dimension first in the processing of SCADA high-frequency data, so as to better serve the data modeling of the model, and also avoid the occurrence of dimension disaster.

Further, the specific method of dimensionality reduction for SCADA production high-frequency data in S1 is:

Using an autoencoder to reduce the time-series production high-frequency data features obtained from SCADA production high-frequency data to a specified dimension, wherein the autoencoder includes a multi-layer LSTM network, each layer of LSTM network has different hidden units, and each layer of LSTM The network performs feature extraction of different dimensions on the high-frequency data of SCADA production in time series.

The beneficial effect of the above scheme is that the dimensionality reduction operation on the high-frequency data produced by SCADA is a data processing operation based on the characteristics of the SCADA data itself, which can ensure that the SCADA data is fully utilized while avoiding the occurrence of the model due to the high dimensionality of the data. dimensional disaster.

Further, the Concat operation is used to perform feature fusion in the S2, and the result of dimensionality reduction of the SCADA dynamic data within one day is spliced with the A2 data and the static parameters of the geological parameters to obtain the input feature vector.

The beneficial effect of the above scheme is that in the existing wellbore effusion research algorithm, it is limited by the well type and block. In order to make the model proposed by the present invention not limited by the above factors, the feature fusion operation is carried out, and A2 is introduced. The static data of data and address parameters makes the prediction of effusion intelligent.

Further, the specific methods of data modeling and training in the S3 are:

S31, using the fused features obtained in step S2 to construct an LSTM model including a forget gate, an input gate and an output gate;

S32, using the self-circulation weight to update the internal forget gate, input gate and output gate of the LSTM model to obtain the updated LSTM model.

The beneficial effect of the above scheme is that in the actual production process, the data of effusion is less than the data of no effusion, so the method based on prediction error is used to model the data and detect the effusion, which can ensure the model It is not affected by the imbalance of positive and negative samples, that is, the data of effusion and no effusion.

Further, the update method of the forgetting gate using the self-circulation weight in the S32 is:

；

;

为第

个遗忘门在

时刻的输出，

为遗忘门的第

个偏置权重，

为

时刻第

个遗忘门对应的第

个输入变量

的输入权重，

为

时刻第

个遗忘门对应的第

个隐藏层变量

的循环权重，

为

时刻的隐藏层变量，

表示sigmoid函数。in,

for the first

a forgotten gate

time output,

for the oblivion gate

a bias weight,

for

the moment

corresponding to the forget gate

input variables

The input weights of ,

for

the moment

corresponding to the forget gate

hidden layer variables

The cycle weight of ,

for

the hidden layer variable at time,

Represents the sigmoid function.

Further, the update mode of the input gate in the S32 is:

；

;

为输入门的第

个计算偏置，

是

时刻的输入向量，

为

时刻的隐藏层向量，

为

时刻第

个输入门对应的第

个输入变量

的输入权重，

为

时刻第

个输入门对应的第

个隐藏层变量

的循环权重。in,

is the first of the input gate

a computational bias,

Yes

the input vector of moments,

for

the hidden layer vector at the moment,

for

the moment

The first input gate corresponding to the

input variables

The input weights of ,

for

the moment

The first input gate corresponding to the

hidden layer variables

the loop weight.

Further, the update method of the internal state of the LSTM cell in the S32 is:

。

.

为在

时刻的LSTM模型内部状态，

为输入门单元，

为偏置权重,

为隐藏层变量的循环权重。in,

for in

The internal state of the LSTM model at the moment,

is the input gate unit,

is the bias weight,

is the loop weight of the hidden layer variable.

Further, the output of the LSTM model is expressed as:

；

;

；

;

为第

个输出门

时刻的输出，

为第

个记忆单元

时刻的值，

为输入门的第

个计算偏置，

是

时刻的输入向量，

为

时刻的隐藏层向量，

为

时刻第

个输入门对应的第

个输入变量

的输入权重，

为

时刻第

个输入门对应的第

个隐藏层变量

的循环权重。in,

for the first

output gate

time output,

for the first

memory unit

the value of the moment,

is the first of the input gate

a computational bias,

Yes

the input vector of moments,

for

the hidden layer vector at the moment,

for

the moment

The first input gate corresponding to the

input variables

The input weights of ,

for

the moment

The first input gate corresponding to the

hidden layer variables

the loop weight.

The beneficial effect of the above-mentioned further scheme is that the long-short-term memory gate is used to model the data, and the time-series dependencies can be effectively extracted, so as to achieve the effect of time-series prediction.

Further, the method of calculating the prediction error in the S3 is:

；

;

表示时刻的平均预测误差，n表示输入输出的向量长度，

为t时刻的输入数据，

为t时刻的预测数据。in,

Represents the average prediction error at the moment, n represents the vector length of the input and output,

is the input data at time t,

is the forecast data at time t.

The beneficial effect of the above solution is that, since the characteristic data of each day is a vector, for the convenience of subsequent threshold selection and effusion prediction, the expectation of the daily prediction error vector is used to represent the prediction error value of the day.

Further, the calculation method of the dynamic threshold in the S4 is expressed as:

；

;

为动态阈值向量，

函数的意义为从动态阈值向量中选择使公式

最大化的动态阈值

，

为预测误差向量的均值，

为预测误差向量的标准差，z为权重系数，

为去掉异常值后的数据均值差，

为预测误差值向量，

为

时刻的误差值

，

为预测误差窗口大小，

为去掉异常值后的数据标准差之差，

为满足

的误差值

的集合，

为

集合中连续误差值的集合。in,

is the dynamic threshold vector,

The meaning of the function is to select the formula from the dynamic threshold vector

maximized dynamic threshold

,

is the mean of the prediction error vector,

is the standard deviation of the prediction error vector, z is the weight coefficient,

In order to remove the outliers, the mean difference of the data,

is the prediction error value vector,

for

time error

,

is the prediction error window size,

In order to remove the outliers, the standard deviation of the data,

to satisfy

error value of

collection of

for

The set of consecutive error values in the set.

The beneficial effect of the above solution is that using a dynamic method to select the threshold value avoids the one-size-fits-all model of the fixed threshold value method, so that the selected threshold value is determined by the recent production situation and conforms to the actual industrial production.

Description of drawings

FIG. 1 is a schematic flowchart of a wellbore fluid detection method based on depth anomaly detection according to the present invention.

FIG. 2 is a schematic structural diagram of a dimension reduction autoencoder according to an embodiment of the present invention.

FIG. 3 is a schematic diagram of feature fusion according to an embodiment of the present invention.

FIG. 4 is a schematic structural diagram of an LSTM model according to an embodiment of the present invention.

Detailed ways

The specific embodiments of the present invention are described below to facilitate those skilled in the art to understand the present invention, but it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes Such changes are obvious within the spirit and scope of the present invention as defined and determined by the appended claims, and all inventions and creations utilizing the inventive concept are within the scope of protection.

A wellbore fluid detection method based on depth anomaly detection, as shown in Figure 1, includes the following steps:

S1. Obtain high-frequency data produced by SCADA and reduce its dimensionality;

SCADA data is the production high-frequency data read from the SCADA production data acquisition system, and the time level is in seconds, which can reflect the data fluctuations in the production process in detail. After the data cleaning operation, there are 17,279 pieces of SCADA data per day, which has the characteristics of high dimensionality.

Due to the high-dimensional characteristics of SCADA-produced high-frequency data, it is considered that feature fusion with A2 data and geological parameter data will be carried out in the future. The influence of parameters on the model is insufficient, and SCADA data is used as second-level production data, A2 is the production data aggregated by SCADA data by day, and the time level is day level, which is mainly used to observe data fluctuations between days. A2 data is used as day-level production data, and the dimensionality reduction operation raises the time dimension of SCADA data to day-level. In this embodiment, the self-encoder is used to reduce the time-series production high-frequency data features obtained from the SCADA production high-frequency data to a specified dimension, wherein the self-encoder includes a multi-layer LSTM network, and each layer of the LSTM network has different hidden units , each layer of LSTM network performs feature extraction of different dimensions on the high-frequency data produced by SCADA in time series, as shown in Figure 2.

The autoencoder has a total of five layers, each layer is an LSTM (Long Short-Term Memory) network, and each layer of LSTM network has different hidden units to extract features of different dimensions for time series data. The extracted intermediate variables of the encoder are essentially a set of the most representative features extracted from the input high-dimensional data, which reduces the dimension of SCADA high-frequency production dynamic data features from tens of thousands of records a day to a custom number of records. , raise the time scale from seconds to days, and then perform feature fusion with A2 data and static parameters.

S2. Feature fusion of SCADA production high-frequency data after dimensionality reduction with A2 data and geological parameter data;

In this embodiment, feature fusion is performed on SCADA data after dimensionality reduction, A2 data, and geological parameter data. The geological parameters are the geological parameters of each production well and are the fixed attributes of the well. The fusion operation uses the Concat operation to concatenate it into a feature vector to diversify the features of the subsequent modeling model. Concat operation refers to splicing all features directly into a vector to achieve feature diversity and improve model modeling results. The result of one day's SCADA dynamic data dimensionality reduction is spliced with A2 data and static parameters to form the final model input eigenvector, so that the effusion prediction model takes full advantage of A2 while considering the SCADA dynamic data. data and static parameters to make more accurate predictions. The data splicing method of this model is shown in Figure 3.

S3, using the features fused in step S2 to perform data modeling and training, and calculate the prediction error;

In this embodiment, the model is first trained using normal data (data in a time period without fluid accumulation), so that the model learns how to predict the A2 data of the next day according to the normal data. The data modeling model uses the LSTM model. The input and output of this model are time series data with a specific sliding window size, which aims to capture the data fluctuation relationship between days and days, and the fluctuation relationship between days and days can reflect the characteristics of effusion. , such as the sudden drop in the daily gas production, the increase in the pressure difference of the oil jacket, etc. Since the model only uses normal data for training, when the input of the model is normal data, an output with a small prediction error will be obtained, and when the input of the model is abnormal data (data in the effusion period), an output will be obtained. Larger prediction error output. The model structure is shown in Figure 4.

In this embodiment, the LSTM model is a special type of RNN, which can learn long-term dependent information. The forget gate, the input gate and the output gate are mainly used to realize the selective passage of information, so as to realize the protection and control of the information. The operating mechanism of the forget gate is shown in Equation 1:

（公式1）

(Formula 1)

是当前输入向量，

是当前隐藏层向量，

包含所有LSTM细胞的输出。

、

、

分别是偏置、输入权重和遗忘门的循环权重，

表示sigmoid函数，将权重设置为0到1之间的值。因此LSTM单元内部状态以公式2的方式更新，其中有一个条件的自环权重：in

is the current input vector,

is the current hidden layer vector,

Contains the output of all LSTM cells.

,

,

are the bias, the input weight, and the loop weight of the forget gate, respectively,

Represents the sigmoid function, setting the weights to a value between 0 and 1. Therefore, the internal state of the LSTM unit is updated in the manner of Equation 2, which has a conditional self-loop weight:

（公式2）

(Formula 2)

、

、

分别是LSTM单元中的偏置、输入权重和循环权重。in

,

,

are the bias, input weight, and loop weight in the LSTM unit, respectively.

以类似遗忘门的方式更新，但有自身的参数，其工作机制如公式3所示：input gate unit

It is updated in a manner similar to the forget gate, but has its own parameters, and its working mechanism is shown in Equation 3:

（公式3）

(Formula 3)

The meaning of the parameters is the same as that of formula 1.

也可以由输出门

关闭，其工作机制如公式4和公式5:The output of the LSTM cell

can also be controlled by the output gate

off, it works like Equation 4 and Equation 5:

（公式4）

(Formula 4)

（公式5）

(Formula 5)

、

、

分别是偏置、输入权重和循环权重。in

,

,

are the bias, input weight, and loop weight, respectively.

After the above model is built, use normal data features to start training the model. In the model in this embodiment, Adam Optimizer is used as the optimizer, the learning rate of the optimizer is set to 0.002, and the total number of iterations is 500 times.

和预测数据

，预测误差的计算公式如下：Error calculation: The model essentially predicts the average oil pressure, average casing pressure and average daily gas production of A2 data in the future based on the input characteristics, and according to the actual production data at time t

and forecast data

, the calculation formula of prediction error is as follows:

（公式6）

(Formula 6)

表示t时刻的平均预测误差，n表示输入输出的向量长度，

为t时刻的输入数据，

为t时刻的预测数据。in,

represents the average prediction error at time t, n represents the vector length of the input and output,

is the input data at time t,

is the forecast data at time t.

According to the prediction error vector, the dynamic threshold candidate vector A is obtained, and its definition is shown in formula 7:

（公式7）

(Formula 7)

是均值，

是标准差，z是权重系数。每一个时间步的误差阈值

是动态的，计算公式如公式8-公式12所示：in,

is the mean,

is the standard deviation, and z is the weight coefficient. Error threshold at each time step

is dynamic, and the calculation formula is shown in Equation 8-Equation 12:

（公式8）

(Formula 8)

（公式9）

(Equation 9)

（公式10）

(Formula 10)

（公式11）

(Equation 11)

（公式12）

(Equation 12)

，即是

The dynamic threshold method is to first calculate the threshold candidate vector A according to the prediction error vector, and then select the best threshold from the candidate vectors.

, that is

这个公式表示根据筛选出的阈值

对历史误差值

进行划分，得出历史误差值的均值

与去掉异常值之后的数据均值

之差。When calculating the specific

This formula expresses the filtered threshold according to

historical error value

Divide to get the mean of historical error values

and the mean of the data after removing outliers

Difference.

则代表的是标准差之差。Similarly,

represents the standard deviation.

表示根据阈值

选出异常值集合。

indicates that according to the threshold

Select the set of outliers.

表示

中为连续异常值的集合。

express

is the set of continuous outliers.

The meaning of the whole dynamic threshold method can be understood as: using the least abnormal sequence and the least number of abnormal points, so that the difference between the mean and standard deviation of the sequence after removing the abnormal and the original sequence is as large as possible

It can be seen that the dynamic threshold method integrates the mean and standard deviation, and it continuously updates the threshold according to the accumulation of errors. In the whole process, only one parameter needs to be adjusted, that is, the weight coefficient z.

After finding the threshold, judge the fluid accumulation at time t+1. When the prediction error at time t+1 is greater than or equal to the dynamic threshold, it is judged that the wellbore is in the state of fluid accumulation at this time, otherwise it is judged that there is no fluid accumulation at this time.

In actual industrial production, various industrial measures will also cause the data to fluctuate greatly compared to the normal state. In order to prevent the model from misjudging the industrial measures, the dynamic threshold method is used to predict the effusion. .

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce Means for implementing the functions specified in a flow or flow of a flowchart and/or a block or blocks of a block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions The apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process such that The instructions provide steps for implementing the functions specified in the flow or blocks of the flowcharts and/or the block or blocks of the block diagrams.

In the present invention, the principles and implementations of the present invention are described by using specific embodiments, and the descriptions of the above embodiments are only used to help understand the method and the core idea of the present invention; The idea of the invention will have changes in the specific implementation and application scope. To sum up, the content of this specification should not be construed as a limitation to the present invention.

Those of ordinary skill in the art will appreciate that the embodiments described herein are intended to assist readers in understanding the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations without departing from the essence of the present invention according to the technical teaching disclosed in the present invention, and these modifications and combinations still fall within the protection scope of the present invention.

Claims (10)

1. A shaft effusion detection method based on depth anomaly detection is characterized by comprising the following steps:

s1, obtaining SCADA production high-frequency data and reducing the dimension of the SCADA production high-frequency data;

s2, carrying out feature fusion on the SCADA production high-frequency data subjected to dimensionality reduction, the A2 data and the geological parameter data;

s3, carrying out data modeling and training by using the features fused in the step S2, and calculating a prediction error;

and S4, calculating a dynamic threshold according to the prediction error obtained in the step S3, judging whether the shaft is subjected to liquid accumulation or not according to the dynamic threshold, judging that the shaft is in a liquid accumulation state at the moment when the prediction error at the moment t +1 is greater than or equal to the dynamic threshold, and otherwise judging that the shaft is not subjected to liquid accumulation at the moment.

2. The method for detecting the effusion in the wellbore based on the detection of the depth anomaly according to claim 1, wherein the step of reducing the dimension of the high-frequency SCADA production data in the step S1 comprises the following steps:

reducing the obtained time sequence production high-frequency data characteristics of the SCADA production high-frequency data to a specified dimension by using a self-encoder, wherein the self-encoder comprises a plurality of layers of LSTM networks, each layer of LSTM network is provided with different hidden units, and each layer of LSTM network carries out feature extraction of different dimensions on the time sequence SCADA production high-frequency data.

3. The method as claimed in claim 2, wherein in S2, feature fusion is performed by using Concat operation, and the result of dimensional reduction of SCADA dynamic data within one day is subjected to feature vector concatenation with a2 data and static parameters of geological parameters to obtain an input feature vector.

4. The method for detecting effusion in a wellbore based on detection of depth anomaly according to claim 3, wherein the data modeling and training in S3 are carried out by:

s31, constructing an LSTM model comprising a forgetting gate, an input gate and an output gate by using the fused features obtained in the step S2;

and S32, updating the internal forgetting gate, the input gate and the output gate of the LSTM model by using the self-circulation weight to obtain the updated LSTM model.

5. The method for detecting effusion in a wellbore based on depth anomaly detection according to claim 4, wherein the updating manner of the forgetting gate by using the self-circulation weight in S32 is as follows:

；

wherein,

is as follows

A forgetting door is

The output of the time of day is,

to forget the door

The weight of each of the offsets is determined,

is composed of

At the first moment

The first that the forgetting gate corresponds to

An input variable

The input weight of (a) is determined,

is composed of

At the first moment

The first that the forgetting gate corresponds to

Hidden layer variable

The cyclic weight of (a) is determined,

is composed of

The hidden layer variable at a time of day,

representing the sigmoid function.

6. The method according to claim 5, wherein the input gates in S32 are updated in a manner that:

；

wherein,

is as follows

An input gate is arranged at

The output of the time of day is,

is the first of the input gate

The offset is calculated by the calculation unit,

is that

The input vector of the time of day,

is composed of

The hidden layer vector at a time instant,

is composed of

At the first moment

Corresponding to each input gate

An input variable

The input weight of (a) is determined,

is composed of

At the first moment

Corresponding to each input gate

Hidden layer variable

The cyclic weight of (2).

7. The method for detecting effusion in a wellbore based on detection of depth anomaly according to claim 6, wherein the updating manner of the internal state of the LSTM cell in S32 is as follows:

；

wherein,

is at the same time

The internal state of the LSTM model at a time,

in order to input the gate unit to the gate unit,

in order to bias the weight of the weight,

is the loop weight of the hidden layer variable.

8. The method of claim 7, wherein the output of the LSTM model is represented as:

；

wherein,

is as follows

Output gate

Output of time of day，

Is as follows

A memory unit

The value of the time of day is,

is the first of the input gate

The offset is calculated by the calculation unit,

is that

The input vector of the time of day,

is composed of

The hidden layer vector at a time instant,

is composed of

At the first moment

Corresponding to each input gate

An input variable

The input weight of (a) is determined,

is composed of

At the first moment

Corresponding to each input gate

Hidden layer variable

The cyclic weight of (2).

9. The method for detecting effusion in a wellbore based on detection of depth anomaly according to claim 8, wherein the prediction error is calculated in S3 by:

；

wherein,

represents the average prediction error at time t,nwhich represents the length of the vector of the output,

for the input data at the time t,

is the predicted data at time t.

10. The method of claim 9, wherein the dynamic threshold of S4 is calculated by:

；

wherein,

in order to be a dynamic threshold value vector,

the meaning of the function is to select the formula from a dynamic threshold vector

Maximized dynamic threshold

，

Is the mean value of the prediction error vector,

is the standard deviation of the prediction error vector, z is the weight coefficient,

to remove the data mean difference after outliers,

in order to predict the vector of error values,

is composed of

Error value of time of day

，

In order to predict the size of the error window,

to remove the difference between the standard deviations of the data after outliers,

to satisfy

Error value of

The set of (a) and (b),

is composed of

A set of consecutive error values in the set.

Images