CN110598326A - A Well Test Interpretation Method Based on Artificial Intelligence - Google Patents

Abstract

The invention relates to a well test interpretation method based on artificial intelligence and belongs to the field of well test analysis; it solves the problems of cumbersome interpretation process, too slow efficiency, and certain interpretation errors in current well test interpretation; the technical scheme is: based on filtering Denoising preprocesses the pressure derivative, writes a program to process the training samples, establishes a well test interpretation model based on convolutional neural network, and improves the genetic algorithm to improve the fitting speed and parameter fitting accuracy. The model is adjusted through experiments and theoretical research. The optimized network structure identifies the characteristic section of the derivative curve, and combines the flow section characteristics and data trends of the well test model to intelligently diagnose the model; the present invention performs well test analysis based on artificial intelligence, and can self-adjust, self-learn, and self-associate , optimize the work process, improve work efficiency and decision-making quality, avoid the influence of subjective judgment, and realize fully automatic well test interpretation.

Description

A Well Test Interpretation Method Based on Artificial Intelligence

technical field

The invention relates to a well test interpretation method based on artificial intelligence, and belongs to the field of well test analysis of oil and gas reservoirs.

Background technique

Well testing is an important means of reservoir engineering. It is a method to study oil and gas reservoir geology and oil and gas well engineering parameters based on oil and gas seepage mechanics and by means of pressure, temperature and production testing. That is to test the wells (oil wells, gas wells and water wells), and measure the pressure and production changes of the wells (oil wells, gas wells and water wells) due to changing the working system. Productivity and completion quality, as well as reservoir and test well performance issues, and analysis of test well stimulation effects.

Comparing well testing analysis methods at home and abroad, there are some advantages and disadvantages. The advantages of foreign well testing analysis methods are: high degree of commercialization; emphasis on versatility and wide applicability; functional integration of software; numerical well testing function; software maintenance and updating in a timely manner. Its shortcomings are: adaptability to the actual field of domestic oilfields; testing methods and testing techniques; testing data quality issues; the particularity of the actual formation; the particularity of mining conditions. The advantages of domestic well testing analysis methods are: focus on specific problems or project development; strong pertinence and strong adaptability. Its shortcomings are: not paying attention to the research of unconventional well testing technology; small development scale and poor scalability; single function and few models; software maintenance and update lag behind the development of computer software technology.

The application of traditional program design in well test interpretation and well test evaluation has been quite common. Since computer technology was introduced into well test interpretation, well test interpretation has been a combination of automated and manual processes. At present, most of the well testing interpretation methods are manual and computer-assisted, but most of them have limitations such as cumbersome interpretation process, too slow efficiency, and certain interpretation errors due to experience and other reasons in the artificial fitting process. Written with non-intelligent programs, it does not have the functions of self-adjustment, self-learning, and self-association. It is limited to solving some of the tasks that are relatively easy to do. It can only help the field engineer to speed up his calculation process, and the uniqueness of the solution is poor. The most difficult part, model identification, injury identification, and uniqueness of solutions, can only be done by a very small number of experts. With the development of artificial intelligence and the innovation of oilfield intelligence, artificial intelligence technology can optimize workflow, improve work efficiency and decision-making quality, avoid the influence of subjective judgment, and realize fully automatic well test interpretation. The intelligent well test interpretation method has great practical significance.

SUMMARY OF THE INVENTION

The purpose of the present invention is: in order to solve the problems of cumbersome interpretation process, too slow efficiency, and certain interpretation errors in the current well test interpretation, the present invention performs well test analysis based on artificial intelligence, and can self-adjust, self-learn, self-associate, and optimize the work. process, improve work efficiency and decision-making quality, avoid the influence of subjective judgment, and realize fully automatic well test interpretation.

In order to achieve the above purpose, the present invention provides a well testing interpretation method based on artificial intelligence, the method includes the following steps: importing measured pressure data and production data, and processing the data based on filtering and denoising, and performing wavelet transformation algorithm. Improve optimization, preprocess the pressure derivative to form a continuous smooth diagnostic curve; use Tensorflow to implement a complete convolutional neural network, and use this convolutional neural network to recognize the handwritten digit dataset (MNIST); write programs to process training samples, Establish a well test interpretation model based on convolutional neural network; improve the genetic algorithm to improve the fitting speed and parameter fitting accuracy, the model adjusts the optimal network structure through experimental and theoretical research, identifies the characteristic segment of the derivative curve, and combines well testing The flow segment characteristics and data trends of the model, the system intelligently diagnoses the model.

In the above-mentioned method for well testing interpretation based on artificial intelligence, Fourier transform is used for the preprocessing of the pressure derivative, and feature quantities that do not depend on the translational change of the well testing data are extracted.

In the above-mentioned artificial intelligence-based well test interpretation method, the main steps of establishing a well test interpretation model based on a convolutional neural network include: defining the weight and bias of the convolution layer; defining a pooling layer; establishing a convolution layer; Establish a fully connected layer; establish an output layer; optimize the method; conduct sample training to establish a well test interpretation model based on convolutional neural network.

In the above-mentioned artificial intelligence-based well test interpretation method, the steps of establishing the convolution layer mainly include: defining the first layer of convolution, first defining the weight of this layer; defining the bias; defining the first layer of the convolutional neural network A convolutional layer h_conv1=conv2d(x_image, W_conv1)+b_conv1, and at the same time perform nonlinear processing on h_conv1, that is, the activation function; then perform pooling processing; in the same way, define the second convolutional layer, this The input of the layer is the output of the pooling layer above; then, the second convolutional layer of the convolutional neural network is defined; finally, the pooling operation is performed.

In the above-mentioned artificial intelligence-based well test interpretation method, the optimization method is to use the cross-entropy loss function to define the cost function, and use tf.train.AdamOptimizer() as the optimizer for optimization to minimize the cross_entropy.

Compared with the prior art, the present invention has the following beneficial effects: (1) the interpretation process in the well test interpretation is simple, the efficiency is high, and the interpretation error is small; (2) the well test analysis based on artificial intelligence can be self-adjusted and self-learning , Self-association; (3) Optimize work flow, improve work efficiency and decision-making quality, avoid the influence of subjective judgment, and realize fully automatic well test interpretation.

Description of drawings

In the attached image:

Figure 1 is a schematic diagram of a convolutional neural network.

Figure 2 is a schematic diagram of the structure of the convolutional layer and the pooling layer.

Fig. 3 is a data extraction interface diagram of a software compiled by using the method provided by the present invention.

Fig. 4 is a software training interface diagram compiled by the method provided by the present invention.

FIG. 5 is a diagram of a software fitting interpretation interface prepared by using the method provided by the present invention.

Detailed ways

The present invention will be further described below with reference to the embodiments and the accompanying drawings.

The invention provides a well testing interpretation method based on artificial intelligence. The method includes the following steps: importing measured pressure data and production data, processing the data based on filtering and denoising, and improving and optimizing the wavelet transform algorithm to adjust the pressure. The derivative is preprocessed to form a continuous smooth diagnostic curve; a complete convolutional neural network is implemented with Tensorflow, and the convolutional neural network is used to identify the handwritten digit dataset (MNIST); write a program to process the training samples, and build a convolutional neural network based on Network well test interpretation model; improved genetic algorithm to improve the fitting speed and parameter fitting accuracy, the model adjusted the optimal network structure through experimental and theoretical research, identified the characteristic section of the derivative curve, combined with the flow section characteristics of the well test model And data trends, the system intelligently diagnoses the model.

In the artificial intelligence-based well test interpretation method of the present invention, the preprocessing of the pressure derivative adopts Fourier transform to extract feature quantities that do not depend on the translational change of the well test data.

In an artificial intelligence-based well testing interpretation method of the present invention, the main steps of establishing a well testing interpretation model based on a convolutional neural network include: defining the weight and bias of the convolutional layer; defining a pooling layer; establishing a convolutional layer ; Establish a fully connected layer; establish an output layer; optimize the method; conduct sample training to establish a well test interpretation model based on convolutional neural network.

The present invention is mainly based on the convolutional neural network (CNN) to establish a well test interpretation model. As shown in Figure 1, the convolutional neural network is a kind of deep learning, and its weight sharing network structure makes it more similar to the biological neural network, reducing the The complexity of the network model is reduced and the number of weights is reduced. The first few layers are usually alternating convolutional and downsampling layers, and the last few layers near the output layer are usually fully connected networks. The training process of convolutional neural network is mainly to learn network parameters such as convolution kernel parameters and inter-layer connection weights of convolutional layers, and the prediction process is mainly to calculate category labels based on input images and network parameters.

The basic structure of CNN consists of an input layer, a convolutional layer, a pooling layer (also called a sampling layer), a fully connected layer and an output layer. Figure 2 shows a schematic diagram of the convolutional layer and pooling layer structure of a one-dimensional CNN. The top layer is the pooling layer, the middle layer is the convolution layer, and the bottom layer is the input layer of the convolution layer.

In an artificial intelligence-based well test interpretation method of the present invention, the steps of establishing a convolution layer mainly include: defining the first layer of convolution, first defining the weight of this layer; defining bias; defining the first layer of the convolutional neural network A convolutional layer h_conv1=conv2d(x_image, W_conv1)+b_conv1, and at the same time perform nonlinear processing on h_conv1, that is, the activation function; , the input of this layer is the output of the pooling layer above; then, the second convolutional layer of the convolutional neural network is defined; finally, the pooling operation is performed.

Pooling is an important operation in convolutional neural networks, which can reduce features while maintaining the local invariance of features. The convolution kernel and pooling kernel in CNN are equivalent to the engineering realization of the receptive field in the Hubel-Wiesel model. The convolution layer is used to simulate the simple cells of the Hubel-Wiesel theory, and the pooling layer simulates the complex cells of the theory. The size (number of neurons) DWindow of each output feature surface of each pooling layer in CNN is:

In formula (1), the size of the pooling kernel is DWindow, and the pooling layer reduces the number of connections between the convolutional layers, that is, the number of neurons is reduced by the pooling operation, which reduces the computational load of the network model.

In an artificial intelligence-based well test interpretation method of the present invention, the steps of establishing a fully connected layer mainly include: defining a fully connected layer; multiplying the flattened h_pool2_flat by the W_fc1 of this layer; considering the problem of overfitting , plus a dropout processing.

After multiple convolution layers and pooling layers, one or more fully connected layers are connected. Similar to MLP, each neuron in the fully connected layer is fully connected to all neurons in the previous layer. Connection layers can integrate class-discriminative local information in convolutional or pooling layers.

In an artificial intelligence-based well test interpretation method of the present invention, the steps of establishing an output layer mainly include: constructing an output layer. The input is 1024, the final output is 10 (because the mnist dataset is [0-9] ten classes), and the prediction is the final predicted value; using the softmax classifier (multi-classification, the output is the probability of each class), right The output is classified.

In an artificial intelligence-based well test interpretation method of the present invention, the optimization method is to use the cross-entropy loss function to define the cost function, and use tf.train.AdamOptimizer() as the optimizer for optimization to minimize the cross_entropy.

In the artificial intelligence-based well test interpretation method of the present invention, the training steps mainly include: defining a session and initializing all variables; training 1000 times, and checking the accuracy of the model every 50 times.

Finally, the C++ and Python integration modules are used to write well test interpretation software. Figure 3, Figure 4, and Figure 5 show the interface diagram of data extraction, software training interface, and fitting interpretation interface, respectively.

Compared with the prior art, the present invention has the following beneficial effects: (1) the interpretation process in the well test interpretation is simple, the efficiency is high, and the interpretation error is small; (2) the well test analysis based on artificial intelligence can be self-adjusted and self-learning , Self-association; (3) Optimize work flow, improve work efficiency and decision-making quality, avoid the influence of subjective judgment, and realize fully automatic well test interpretation.

Claims (5)

1. a well test interpretation method based on artificial intelligence, is characterized in that, this method comprises the following steps: Import the measured pressure data and production data, and preprocess the pressure derivative based on filtering and denoising to form a continuous smooth diagnostic curve; Implement a complete convolutional neural network with Tensorflow, and use this convolutional neural network to recognize handwritten digits dataset (MNIST); Write a program to process training samples and build a well test interpretation model based on convolutional neural network; Improve the genetic algorithm to improve the fitting speed and parameter fitting accuracy. The model adjusts the optimal network structure through experimental and theoretical research, identifies the characteristic segment of the derivative curve, combines the flow segment characteristics and data trends of the well test model, and makes systematic intelligent diagnosis out the model. 2. A well testing interpretation method based on artificial intelligence according to claim 1, characterized in that: the preprocessing of the pressure derivative adopts Fourier transform to extract feature quantities that do not depend on the translational change of well testing data . 3. The artificial intelligence-based well test interpretation method according to claim 1, wherein the main step of establishing a well test interpretation model based on a convolutional neural network comprises: Define the Weight and bias of the convolutional layer; Define the pooling layer; Build a convolutional layer; Build a fully connected layer; Build the output layer; Optimization; Carry out sample training to establish a well test interpretation model based on convolutional neural network. 4. A kind of well test interpretation method based on artificial intelligence according to claim 3, is characterized in that: described step of establishing convolution layer mainly comprises: Define the first layer of convolution, first define the Weight of this layer; define bias; Define the first convolutional layer h_conv1=conv2d(x_image, W_conv1)+b_conv1 of the convolutional neural network, and perform nonlinear processing on h_conv1, that is, the activation function to process; Then perform pooling processing; In the same way, define the second convolutional layer, the input of this layer is the output of the previous pooling layer; Next, define the second convolutional layer of the convolutional neural network; Finally, the pooling operation is performed, and so on. 5. a kind of well test interpretation method based on artificial intelligence according to claim 3, is characterized in that: described optimization method is to use cross entropy loss function to define cost function, use tf.train.AdamOptimizer() as optimization optimizer to minimize cross_entropy.