CN108596251A - One kind carrying out fluid identification of reservoir method based on committee machine using log data - Google Patents

Abstract

The invention discloses a method for identifying reservoir fluids based on committee machines using well logging data. The method includes the following steps: 1) selecting well logging data as input data; 2) normalizing the input data; 3) 4) Form a data set from logging data and fluid type; 5) Randomly divide the data set into training data set and test data set; 6) Select a pre-classifier; 7) For each 8) Using the test data set as input, a category is given by each classification model; 9) For each set of input data, each classification model is given The categories are combined, and the committee decision-making mechanism is adopted to give the final classification category. The method simulates the decision-making mechanism of the committee, and combining multiple pre-classifiers can reduce falling into the local minimum and make the committee's decision-making more scientific and accurate.

Description

A Reservoir Fluid Identification Method Using Well Logging Data Based on Committee Machine

technical field

The invention belongs to the technical field of reservoir fluid identification in petroleum exploration, and in particular relates to a method for identifying reservoir fluid by using logging data based on a committee machine.

Background technique

Geophysical logging is a continuous and in-situ geophysical parameter measurement technology along the wellbore. The measured data mainly include natural gamma ray, natural potential, deep and shallow resistivity, sound wave, density, neutron, etc., and reservoir division can be realized by using these data , fluid identification, porosity, permeability and saturation calculations. By studying the logging response characteristics of reservoirs containing different fluids, fluid identification of reservoirs can be carried out. Practice has proved that some reservoirs have obvious logging response characteristics, and it is easy to realize fluid identification. However, for low-porosity and low-permeability reservoirs, low-resistivity oil layers, and complex lithologic formations, the fluid in the pores makes little contribution to the logging response. There is a nonlinear relationship between reservoirs with different fluid properties and various logging responses, and it is difficult to qualitatively identify fluids by using logging responses. Therefore, in view of the low porosity and permeability of tight sandstone and the low accuracy of reservoir identification, relevant scholars at home and abroad have used intelligent algorithms such as neural networks and fuzzy systems to solve the problem of logging interpretation. At present, when the neural network method is used for fluid identification, fluid-sensitive logging data is usually selected as input, and then a single machine learning algorithm (such as BP neural network) is selected for training. The training set comes from reservoirs with known test results. After training the network, input the logging data of the unknown fluid type reservoir, and use the neural network to predict and judge the fluid type of the reservoir.

The core of the fluid identification method based on neural network using logging data is the machine learning algorithm adopted. At present, single learning algorithms such as BP network, decision tree, and support vector machine are often used. These methods have their own advantages and disadvantages. Classification through a single decision-making mechanism often leads to over-training and poor robustness. Therefore, when using a single intelligent algorithm for training and prediction, it may fall into a local minimum due to overfitting, resulting in poor generalization ability.

For this reason, considering that each intelligent algorithm has different advantages and functions, the combination of multiple methods can reduce the risk of falling into a local minimum. And through the excellent decision-making mechanism, the committee's decision-making is more scientific and accurate.

BP (back propagation) neural network is a concept proposed by scientists led by Rumelhart and McClelland in 1986. It is a multi-layer feed-forward neural network trained according to the error back propagation algorithm. It is currently the most widely used neural network.

Decision Tree (Decision Tree) is a decision analysis method for evaluating project risk and judging its feasibility by forming a decision tree to obtain the probability that the expected value of the net present value is greater than or equal to zero on the basis of knowing the probability of occurrence of various situations. A graphical method for intuitive use of probability analysis. Because this decision-making branch is drawn in a graph that resembles the branches of a tree, it is called a decision tree. In machine learning, a decision tree is a predictive model that represents a mapping relationship between object attributes and object values.

Support vector machine (Support Vector Machine, SVM) was first proposed by Corinna Cortes and Vapnik in 1995. It shows many unique advantages in solving small samples, nonlinear and high-dimensional pattern recognition, and can be extended and applied to function simulation. and other machine learning problems.

In machine learning, a support vector machine is a supervised learning model related to related learning algorithms that can analyze data and recognize patterns for classification and regression analysis.

Adaptive neuro-fuzzy reasoning system is a new type of fuzzy reasoning system structure that combines fuzzy logic and neuron network organically. It adopts the hybrid algorithm of back propagation algorithm and least square method to adjust the premise parameters and conclusion parameters, and can automatically generate If -Then rules. Adaptive Network-based Fuzzy Inference System ANFIS (AdaptiveNetwork-based Fuzzy Inference System) combines neural network and fuzzy inference organically, which not only brings into play the advantages of both, but also makes up for their respective shortcomings.

Contents of the invention

In order to solve the problem of overfitting and falling into local minimum in a single intelligent classification algorithm, which leads to poor generalization ability, the present invention provides a method for identifying reservoir fluids based on committee machines using well logging data. On the basis of an intelligent classification algorithm, a committee decision-making mechanism is adopted to realize the identification and classification of reservoir fluids. The method combines the classification results of multiple intelligent classification algorithms, uses a decision-making mechanism similar to a committee, and determines the final classification result on the basis of multiple classification results, effectively combining the advantages of different intelligent classification algorithms. Improved the accuracy of the final classification.

In order to achieve the above object, the present invention adopts the following technical solutions:

A method for identifying reservoir fluids based on committee machines using well logging data, the method comprising the following steps:

1) Select fluid-sensitive logging data as input data;

2) Normalize each attribute value of the input data;

3) Obtain the reservoir fluid type according to the oil test results;

4) Combine logging data and reservoir fluid types for each layer to form a data set;

5) Randomly divide the data set into two categories, one is the training data set, and the other is the test data set;

6) Select several intelligent classification algorithms as pre-classifiers;

7) Using the training data set as input, train each pre-classification algorithm separately to obtain the corresponding classification model;

8) Take the test data set as input, and give a category through each classification model;

9) For each set of input data, combine the categories given by each classification model, and adopt a committee decision-making mechanism to give the final classification category.

Preferably, in the step 1), acoustic time difference (AC, ), neutron density (CNL), compensation density (DEN), natural gamma ray (GR), deep induction logging (ILD), medium induction logging (ILM) and other logging data as input data.

Preferably, the reservoir types involved in step 3) include dry layers, water layers, oil-bearing water layers, oil-water layers and oil layers.

Preferably, in step 5), the training data set and the testing data set account for 80% and 20% of the total data set respectively.

Preferably, the intelligent classification algorithm adopted in the step 6) includes BP neural network, support vector machine and adaptive neural network-fuzzy reasoning system.

Preferably, the committee decision-making mechanism in step 9) uses a voting method to combine the output types of each intelligent classification model to generate the final classification output type.

Preferably, the voting method includes three types: the absolute majority voting method, if a certain type gets more than half of the votes, it is predicted to be that type, otherwise the prediction is rejected; the relative majority voting method is predicted to be the type with the most votes, if there are multiple If the type has the highest number of votes, one will be randomly selected; the weighted voting method gives each intelligent system a weight value to calculate the number of votes for each type.

The advantages and beneficial effects of the present invention are: the present invention proposes the idea of a combination of multiple intelligent classification algorithms in the selection of core classification algorithms, that is, a committee machine, which is similar to forming a committee, and each committee member corresponds to a different intelligent classification algorithm. Each classification algorithm has different strengths and features. The committee machine simulates the decision-making mechanism of the committee, and combines these single intelligent classification algorithms through an excellent decision-making mechanism, which can reduce falling into the local minimum and make the committee's decision-making more scientific and accurate. Therefore, the method is superior to a single intelligent classification system, with a high degree of training and better prediction results.

Description of drawings

Accompanying drawing 1 is the working principle diagram of the logging data-based reservoir fluid identification method of the present invention.

Accompanying drawing 2 is the working flowchart of the reservoir fluid identification method based on logging data of the present invention.

Accompanying drawing 3 is the confusion matrix diagram of the classification prediction of the reservoir fluid identification method based on logging data according to the present invention.

Accompanying drawing 4 is the comparison diagram of the classification and prediction performance of each single intelligent algorithm in the embodiment of the reservoir fluid identification method based on logging data in the present invention.

Detailed ways

Referring to accompanying drawing 1, accompanying drawing 2, a kind of method based on committee machine utilizes logging data to carry out reservoir fluid identification, described method comprises the following steps:

1) Select fluid-sensitive logging data as input data;

2) Normalize each attribute value of the input data;

3) Obtain the reservoir fluid type according to the oil test results;

4) Combine logging data and reservoir fluid types for each layer to form a data set;

5) Randomly divide the data set into two categories, one is the training data set, and the other is the test data set;

6) Select several intelligent classification algorithms as pre-classifiers;

7) Using the training data set as input, train each pre-classification algorithm separately to obtain the corresponding classification model;

8) Take the test data set as input, and give a category through each classification model;

9) For each set of input data, combine the categories given by each classification model, and adopt a committee decision-making mechanism to give the final classification category.

In the step 1), logging data such as acoustic time difference, neutron density, compensation density, natural gamma ray, deep induction logging, and medium induction logging are selected as input data.

The reservoir types involved in the step 3) include dry layers, water layers, oily water layers, oil-water layers and oil layers.

In the step 5), the training data set and the testing data set account for 80% and 20% of the total data respectively.

The intelligent classification algorithm adopted in the step 6) includes BP neural network, support vector machine and adaptive neural network-fuzzy reasoning system.

The committee decision-making mechanism in step 9) uses a voting method to combine the output types of each intelligent classification model to generate the final classification output type.

The voting method adopts any one of absolute majority voting method, relative majority voting method and weighted voting method.

The present invention will be further described below in conjunction with embodiment.

Example

The logging data and oil testing results data of several wells in the Honghe area were selected as the data set to carry out the classification committee machine experiment. Follow the steps below:

1) Select fluid-sensitive logging data such as acoustic transit time, neutron density, compensation density, gamma ray (GR), deep induction logging, medium induction logging, and eight lateral logging (LL8) as input data;

2) Normalize the data of the 7 input features, and the normalization formula is:

In the formula, x _min and x _max are the average value, minimum value, and maximum value of all data in a certain attribute, respectively, and x is the data to be normalized. After normalization processing, the data of the 7 input features are all within [-1,1];

3) Classify the reservoirs according to the oil testing results. The classification targets are five types of reservoirs, namely dry layer, water layer, oily water layer, oil-water layer and oil layer, which are represented by numbers 1 to 5 in the experiment ;

4) Combine logging data and reservoir fluid types for each layer to form a data set;

5) The data set is randomly divided into two categories, 80% of the data is used as a training set, and 20% of the data is used as a test set, part of the training set data is shown in Table 1, and part of the test set data is shown in Table 2;

6) Select BP neural network, support vector machine and neuro-fuzzy system as the pre-classifier;

7) Using the training data set as input, train each pre-classification algorithm separately to obtain the corresponding classification model;

8) Using the test data set, the three models of BP neural network, support vector machine and neuro-fuzzy system are respectively obtained through training. For each sample input xi, the three models output a classification mark respectively, O _{BPNN, i} , O _{SVM, i} and _OANFIS.i , where the value set of the classification mark is {1, 2, 3, 4, 5}, which is used as the committee machine to construct the experimental data, and some data are shown in Table 3;

9) For each set of input data, combine the categories given by each classification model, and use the relative majority voting method as the combination strategy of the committee. In the relative majority voting method, the mark with the most votes counted by the committee is used as the final output of the committee; if the marks output by the three models are different, a random selection is made from the three marks. Through the relative majority vote, the committee finally obtained the predicted category for each sample, ie completed the fluid identification of the reservoir to be explained.

Table 1 Part of the training data used for the classification committee experiments

Table 2 Part of the test data used in the classification committee experiment

Table 3 Output data of some pre-classifiers used in the construction of the classification committee

comparative example

In order to illustrate the performance of the classification committee method, the accuracy rate and mean square error are used to characterize the performance of classification prediction, and the classification performance of the committee and three committee members—BP neural network, support vector machine and neuro-fuzzy system are tested using test data. And compare the prediction results of the classification committee with the performance of the three committees - BP neural network, support vector machine and neuro-fuzzy system on the classification problem. The test results are: the classification accuracy rate of the committee model is 96.1%, the mean square error between the output value and the target value is 6.8%, and the confusion matrix of its classification prediction is shown in Figure 3; The square error is lower than the mean square error of the single committee system, and the prediction accuracy rate based on the classification committee is also higher than that of any single committee system. The comparison is shown in Figure 4.

Finally, it should be noted that obviously, the above-mentioned embodiments are only examples for clearly illustrating the present invention, rather than limiting the implementation. For those of ordinary skill in the art, other changes or changes in different forms can be made on the basis of the above description. It is not necessary and impossible to exhaustively list all the implementation manners here. However, the obvious changes or changes derived therefrom still fall within the scope of protection of the present invention.

Claims (7)

1. A method for identifying reservoir fluid based on committee machine utilizing well logging data, characterized in that, the method may further comprise the steps: 1) Select fluid-sensitive logging data as input data; 2) Normalize each attribute value of the input data; 3) Obtain the reservoir fluid type according to the oil test results; 4) Combine logging data and reservoir fluid types for each layer to form a data set; 5) Randomly divide the data set into two categories, one is the training data set, and the other is the test data set; 6) Select several intelligent classification algorithms as pre-classifiers; 7) Using the training data set as input, train each pre-classification algorithm separately to obtain the corresponding classification model; 8) Take the test data set as input, and give a category through each classification model; 9) For each set of input data, combine the categories given by each classification model, and adopt a committee decision-making mechanism to give the final classification category. 2. A kind of method for identifying reservoir fluid based on committee machine utilizing logging data according to claim 1, characterized in that: in said step 1), select acoustic time difference, neutron density, compensation density, natural gamma ray , deep induction logging, medium induction logging and other logging data as input data. 3. a kind of based on committee machine according to claim 1 utilizes logging data to carry out reservoir fluid identification method, it is characterized in that: the reservoir type involved in described step 3) comprises dry layer, water layer, oil-bearing water layer , oil-water layer and oil layer. 4. a kind of based on committee machine according to claim 1 utilizes logging data to carry out reservoir fluid identification method, it is characterized in that: in described step 5), training data set and testing data set respectively account for the total amount of data set 80% and 20%. 5. a kind of based on committee machine according to claim 1 utilizes logging data to carry out reservoir fluid identification method, it is characterized in that: the intelligent classification algorithm adopted in described step 6) comprises BP neural network, support vector machine and automatic Adapting Neural Networks - Fuzzy Inference Systems. 6. a kind of based on committee machine according to claim 1 utilizes logging data to carry out reservoir fluid identification method, it is characterized in that, the committee decision-making mechanism in the described step 9) adopts the output type of combination of each intelligent classification model by voting , yielding the final classification output type. 7. A method for identifying reservoir fluids based on committee machines using well logging data according to claim 1 or 6, wherein the voting methods include three types: the absolute majority voting method, if a certain type gets more than half of the votes , then the prediction is of this type, otherwise the prediction is rejected; relative to the majority voting method, the prediction is the type with the most votes, if there are multiple types with the highest number of votes at the same time, one of them is randomly selected; the weighted voting method gives each intelligent system a weight value To calculate the number of votes for each type.