CN112016766A - Oil and gas well drilling overflow and leakage early warning method based on long-term and short-term memory network - Google Patents

Abstract

The present invention provides an oil and gas well drilling leakage early warning method based on a long short-term memory network. The early warning method may include the following steps: constructing a long short-term memory network-based leakage risk early warning model, and the leakage risk early warning model includes an input layer. , hidden layer and output layer, wherein the hidden layer includes at least two layers of long-term and short-term memory network structure; use the existing drilling risk data to train the spill risk early warning model; use the trained spill risk early warning model for oil and gas Well drilling spill warning. The beneficial effects of the invention may include: timely and accurate identification results can be given, the dependence of risk identification on prior knowledge and expert experience can be reduced, the early warning of drilling risks can be made more intelligent and efficient, and it has a good field application prospect.

Description

An early warning method for oil and gas well drilling spillage based on long short-term memory network

technical field

The invention relates to the field of well control safety in oil and gas drilling engineering, in particular, to a method for early warning of oil and gas well drilling leakage based on a long short-term memory network.

Background technique

Overflow and lost circulation are two easy risks in the drilling process, especially for oil and gas exploration blocks characterized by deep wells, ultra-deep wells, fault development, carbonate rocks, etc. and some newly developed areas where the underground pressure system is not very clear. Blocks, spill hazards are more likely to occur. Overflow and lost circulation will not only cause serious damage to the reservoir, increase development costs and reduce development efficiency, but also induce drilling accidents such as stuck pipe, well collapse, blowout, etc., resulting in heavy casualties and economic losses if the control is not effective. . Therefore, real-time monitoring and early warning of early overflow and lost circulation during drilling is of great significance to achieve safe and efficient drilling and save drilling costs.

The drilling process is a complex nonlinear dynamic process with many uncertain factors. Various drilling parameters are subject to large random interference, and the parameters are interrelated and coupled with each other. It is difficult to establish an accurate spill risk identification model and spill early warning. accuracy is limited.

SUMMARY OF THE INVENTION

Aiming at the deficiencies existing in the prior art, the purpose of the present invention is to solve one or more problems existing in the prior art mentioned above. For example, one of the objectives of the present invention is to improve the accuracy of leakage monitoring and early warning.

In order to achieve the above object, the present invention provides a method for early warning of oil and gas well drilling leakage based on a long short-term memory network. The early-warning method may include the following steps: constructing a long-short-term memory network-based spillover risk early-warning model, the spillage risk early-warning model includes an input layer, a hidden layer and an output layer, wherein the hidden layer includes at least two layers of long short-term memory Network structure; use the existing drilling risk data to train the spill risk early warning model; use the trained spill risk early warning model for oil and gas well drilling spill warning.

According to an exemplary embodiment of the present invention, the early-warning method may include the following steps: constructing a long-short-term memory network-based spillover risk early-warning model, where the spillage risk early-warning model includes an input layer, a hidden layer and an output layer, wherein, The hidden layer includes at least two layers of long-term and short-term memory network structure; using the existing drilling risk data to train the spill risk early warning model; using the trained spill risk early warning model for oil and gas well drilling spill warning.

According to an exemplary embodiment of the present invention, the long short-term memory network structure of the output layer and the hidden layer may be connected by a softmax activation function.

According to an exemplary embodiment of the present invention, the output layer is capable of outputting at least one of overflow risk, lost circulation risk, and probability of occurrence of normal operating conditions.

According to an exemplary embodiment of the present invention, the step of training the leakage risk early warning model includes: forward propagation of data and back propagation of errors.

According to an exemplary embodiment of the present invention, the step of training the spill risk early warning model includes: step A: setting training parameters of the constructed spill risk early warning model; step B: initializing the constructed spill risk early warning model Weights and biases of the spill risk early warning model; Step C: Input the drilling risk data, under the current weights and biases, the spill risk early warning model outputs actual operating conditions; Step D: Compare the actual operating conditions If the actual working conditions do not meet the expected output, backpropagation layer by layer, and the error is allocated to each layer of the leakage risk early warning model; Step E: compare the weights and Adjust the offset, and repeat steps B to D until the actual working conditions of the output meet the expected output.

According to an exemplary embodiment of the present invention, the step of training the leakage risk early warning model includes: screening out drilling risk data from drilling data, where the drilling risk data includes: mud pool volume on the surface, riser pressure, wellhead The drilling fluid outlet flow rate, the bottom hole annular pressure and the bottom hole annular temperature are determined; normalize the drilling risk data, and remove outliers in the data obtained after the normalization; use overlapping The drilling risk data is expanded by means of sampling; and then the above steps A to E are performed.

According to an exemplary embodiment of the present invention, the above steps of screening drilling risk data, normalizing, removing outliers, and data expansion can be implemented through the input layer of the leakage risk early warning model.

According to an exemplary embodiment of the present invention, the drilling risk data may include: mud pool volume at the surface, riser pressure, drilling fluid outlet flow at the wellhead, annulus pressure at the bottom hole, and annulus temperature at the bottom hole.

According to an exemplary embodiment of the present invention, before the training of the leakage risk early warning model, the method may further include the step of: normalizing the drilling risk data.

According to an exemplary embodiment of the present invention, the drilling risk data may be normalized by using an optimum method.

According to an exemplary embodiment of the present invention, the method may further include the step of removing outliers in the data obtained after the normalization process.

According to an exemplary embodiment of the present invention, before the training of the leakage risk early warning model, the drilling risk data may be augmented by means of overlapping sampling.

Compared with the prior art, the beneficial effects of the present invention may include at least one of the following:

(1) The present invention can timely and accurately give the identification result when monitoring the leakage risk.

(2) The present invention can avoid the feature extraction process of the current mainstream leakage early warning methods, reduce the dependence of risk identification on prior knowledge and expert experience, and make the early warning of drilling risks more intelligent and efficient.

(3) The input and output data of the present invention is based on the establishment of an artificial intelligence model that does not rely heavily on the theoretical mechanism model, which can reduce the requirements for expert experience.

(4) With the advancement of digital oilfield construction and the establishment of a drilling risk database, the drilling risk early warning method of the present invention has a good field application prospect.

Description of drawings

The above and other objects and features of the present invention will become more apparent from the following description in conjunction with the accompanying drawings, wherein:

Fig. 1 shows a schematic diagram of the leakage risk early warning model based on long short-term memory network of the present invention;

Fig. 2 shows a schematic diagram of the present invention for overlapping acquisition;

FIG. 3 shows a training flow chart of the spill risk early warning model of the present invention.

Detailed ways

Hereinafter, the method for early warning of oil and gas well drilling leakage based on the long short-term memory network of the present invention will be described in detail with reference to the accompanying drawings and exemplary embodiments.

At present, the leakage monitoring method adopted at the drilling site is mainly based on manual monitoring of the liquid level in the mud pool. Due to the influence of the sensitivity of the liquid level sensor and the fluctuation of the liquid level in the mud pool, the monitoring accuracy is not high. Aiming at the problems of limited real-time and accuracy and low intelligence of drilling risk monitoring such as overflow and lost circulation, the present invention combines three types of monitoring parameters of downhole, wellhead and surface, as well as a long short-term memory network (Long short-term memory network). , LSTM) is suitable for processing data with strong coupling and strong temporal correlation. An intelligent early warning method of spill risk based on LSTM network is proposed to improve the accuracy of spill monitoring and early warning.

In an exemplary embodiment of the present invention, the method for early warning of oil and gas well drilling leakage based on a long short-term memory network may include the following steps:

S10: Construct a spillage risk early warning model based on a long short-term memory network as shown in Figure 1. The spillage risk early warning model includes three parts: an input layer, a hidden layer and an output layer.

S20: Use the existing drilling risk data to train the leakage risk early warning model.

S30: Use the trained spillage risk early warning model for oil and gas well drilling spillage early warning.

In this embodiment, the input layer may have the functions of monitoring parameter selection and data preprocessing.

(1) Selection of monitoring parameters

In order to accurately monitor the risk of overflow and lost circulation, appropriate monitoring parameters need to be selected. According to the change characteristics of the relevant parameters when the leakage risk occurs, the invention can optimize the volume of the mud pool above the surface, the pressure of the riser, the outlet flow rate of the drilling fluid at the wellhead, the annular pressure and the annular temperature at the bottom of the well as overflow parameters. Parameters for leakage risk monitoring.

(2) Data preprocessing

In the actual drilling process, there are many types of sensors, and the dimensions and orders of magnitude of different sensors are different. In order to speed up the training speed of the network, improve the classification accuracy, and enhance the applicability of the model, before training the network model, first use the maximum value The method normalizes the data sequence collected by the sensor, as shown in formula (1).

为归一化之后的值，x_i为当前值，x_min和x_max分别表示采集数据序列的最小值和最大值。in,

is the normalized value, x _i is the current value, and x _min and x _max represent the minimum and maximum values of the collected data sequence, respectively.

为n个采样数据序列的均值，σ²为其方差，当第n个采样点满足

时，则认为该点为野值点，采用邻域均值将其替换。At the same time, the data acquired on site is affected by noise and random interference during the process of collection and transmission, and outliers will inevitably appear, which will affect the accuracy of the recognition model. The outlier elimination method of the criterion removes outliers. Assume

is the mean of n sampled data series, σ ² is its variance, when the nth sampling point satisfies

, the point is considered to be an outlier, and the neighborhood mean is used to replace it.

In addition, for the missing data that may exist in the monitoring data, the nearest neighbor interpolation method is used to fill up.

(3) Data expansion

Further, the input layer can have the function of data augmentation. Considering that there are relatively few risk data such as overflow and lost circulation, in order to avoid over-fitting of the network, the present invention proposes to expand the data set by means of overlapping sampling. A schematic diagram of overlapping sampling is shown in Figure 2. Among them, when the model is trained, the data expansion function (that is, this link) can be used to expand the sample data, and when the model is used for overflow monitoring, the data expansion function is not required.

In this embodiment, the hidden layer may be composed of a two-layer LSTM structure, and the number, output form, and parameters of neuron nodes contained in each layer may be as shown in Table 1.

Table 1 Hidden layer information

Layer unit Output Shape Param LSTM1 32 (None,5,32) 6784 LSTM2 16 (None,16) 3136

Among them, None means that the dimension cannot be determined, which is related to the number of batch data samples during neural network training.

In this embodiment, the output layer can provide early warning for two risks of overflow and lost circulation. Among them, each drilling risk can be corresponding to a type of label, for example, it can be represented by "one-hot vector", and the specific form is shown in Table 2.

Table 2 Risk types and representation methods

The LSTM units of the output layer and the hidden layer can be connected by the softmax activation function, and finally output the overflow risk, lost circulation risk and probability of normal working conditions respectively.

In this embodiment, in step S20, that is, during the training process, 32 sets of data can be randomly selected from the training samples to form a batch, the data of one batch is trained each iteration, and the training loss value and Test accuracy. Considering the relatively small number of samples in this experiment, in order to prevent the neural network from overfitting, an L2 regularization strategy can be introduced into the _two LSTM layers for constraints.

The specific training process includes two parts: forward propagation of data and back propagation of errors, as shown in Figure 3. First, set the training parameters of the network, initialize the weights and biases of the network, input monitoring data, and then go through the LSTM layer and the fully connected layer to calculate the actual output under the current weights and biases, and calculate the error between it and the expected output. And backpropagation layer by layer, the error is allocated to each layer, and the weights and biases of the network are adjusted by the Adam algorithm until the training conditions are met, so as to realize the supervised training of the network.

The technical scheme of the present invention introduces an advanced deep learning method into the field of oil and gas well drilling risk monitoring, and proposes a drilling leakage risk early warning method based on the LSTM network, that is, the leakage risk early warning model is obtained by learning the existing drilling risk data through the LSTM network, and the The collected raw drilling data is preprocessed and directly input to the trained LSTM network model for leakage risk monitoring, which can give timely and accurate identification results.

Although the present invention has been described above in connection with the exemplary embodiments, it will be apparent to those skilled in the art that various modifications and variations can be made in the exemplary embodiments of the present invention without departing from the spirit and scope defined by the appended claims. Change.

Claims (10)

1. a kind of oil and gas well drilling overflow early warning method based on long short-term memory network, is characterized in that, described early warning method may further comprise the steps: Build a spillover risk early warning model based on a long short-term memory network, the spillage risk early warning model includes an input layer, a hidden layer and an output layer, wherein the hidden layer includes at least two long and short-term memory network structures; Use the existing drilling risk data to train the spill risk early warning model; Using the trained spill risk early warning model for oil and gas well drilling spill early warning. 2 . The oil and gas well drilling leakage warning method based on long short term memory network according to claim 1 , wherein the long short term memory network structure of the output layer and the hidden layer is connected by a softmax activation function. 3 . 3. The oil and gas well drilling leakage warning method based on long short-term memory network according to claim 1, wherein the output layer can output at least one occurrence of overflow risk, lost circulation risk and normal operating conditions. probability. 4. The oil and gas well drilling leakage warning method based on long short-term memory network according to claim 1, wherein the step of training the leakage risk warning model comprises: forward propagation of data and back propagation of errors. 5. The oil and gas well drilling leakage warning method based on long short-term memory network according to claim 1, wherein the step of training the leakage risk warning model comprises: Step A: set the training parameters of the constructed spill risk early warning model; Step B: Initialize the weights and biases of the constructed spill risk early warning model; Step C: input the drilling risk data, under the current weight and offset, the leakage risk early warning model outputs the actual working condition; Step D: Compare the actual working condition with the expected output, and if the actual working condition does not meet the expected output, backpropagate layer by layer, and allocate the error to each layer of the leakage risk early warning model; Step E: Adjust the weights and offsets, and repeat steps B to D until the actual output conditions meet the expected output. 6. The oil and gas well drilling leakage warning method based on long short-term memory network according to claim 1, wherein the drilling risk data comprises: Mud pool volume at the surface, riser pressure, drilling fluid outlet flow at the wellhead, annulus pressure at the bottom hole, and annulus temperature at the bottom hole. 7. The oil and gas well drilling leakage warning method based on long short-term memory network according to claim 1, characterized in that, before the leakage risk warning model is trained, the method further comprises the step of: The drilling risk data is normalized. 8 . The long-short-term memory network-based oil and gas well drilling leak warning method according to claim 7 , wherein the drilling risk data is normalized by using the maximum value method. 9 . 9 . The long short-term memory network oil and gas well drilling leakage warning method according to claim 7 , wherein the method further comprises the step of removing outliers in the data obtained after the normalization process. 10 . 10. The oil and gas well drilling leakage warning method based on long short-term memory network according to claim 1, characterized in that, before the leakage risk warning model is trained, the method of overlapping sampling is used to detect the drilling risk. data is expanded.

Images