A real-time quantitative prediction method for crude oil density based on multi-component gas logging system

Abstract

The present application provides a kind of real-time quantitative prediction method of crude oil density based on multi-component gas logging system, belongs to oil and gas exploration and development technical field.The present application includes: step one: the content of different components in drilling fluid is analyzed and detected by gas detection equipment;Step two: combined with multi-component gas logging data, the curve characteristics of gas logging multi-component data set are analyzed;Step three: the parameters and derived parameters that can best reflect the properties of crude oil density in the study horizon are selected and optimized, including the processes of screening and correction of basic data, selection and interpretation of derived parameters, etc.;Step four: correlation analysis is carried out between the optimized parameters and crude oil density, the formation oil and gas information is decomposed into hydrocarbon component unit, and the internal relationship between hydrocarbon component and derived parameters and reservoir crude oil characteristics is found out.The present application combines the related parameters such as gas logging parameters and derived parameters by machine learning method based on BP neural network model, and establishes the real-time prediction model of crude oil density of multi-component gas logging parameters of reservoir.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of oil and gas exploration and development, and particularly relates to multi-component gas logging data prediction, and in particular to a real-time quantitative prediction model for crude oil density. BACKGROUND

[0002] Reservoir crude oil density is one of the important parameters that need to be determined for oil testing scheme design, oil reservoir reserve calculation, and oil reservoir development scheme formulation before oil testing. Therefore, real-time and accurate prediction of reservoir crude oil density has important engineering significance. However, in the process of oil and gas exploration, accurate reservoir crude oil density data can only be obtained by collecting samples for experimental analysis after oil testing and oil production, which is difficult to achieve real-time perception analysis on the logging site. One of the ultimate goals of the development of reservoir crude oil density prediction from qualitative to quantitative is to predict the productivity of the reservoir. The change of crude oil properties is an important factor to determine whether the reservoir oil has industrial output capacity. Therefore, real-time perception analysis and accurate prediction of crude oil properties have important significance in production practice.

[0003] The previous prediction of crude oil density before oil testing mainly adopts the reservoir pyrolysis method to predict the density by using the correlation between the content of gaseous hydrocarbon, liquid hydrocarbon, and heavy hydrocarbon in the existing drilling oil in the research area. This method has two shortcomings: first, the high-temperature combustion of the sample in the rock pyrolysis detection process will cause a large loss of some components in the sample, resulting in a large error in the calculation of light oil index, heavy oil index, and other indicators in representing the relative content of gaseous hydrocarbon, liquid hydrocarbon, and heavy hydrocarbon; second, only the relationship between the content of gaseous hydrocarbon, liquid hydrocarbon, or heavy hydrocarbon and the crude oil density is statistically analyzed in the model establishment, without describing the comprehensive influence of the three components on the crude oil density. This makes the method have a large error in the application process, and the absolute error of the prediction result can be as high as 0.1 g / ml or more. The existing research on the prediction of crude oil density is relatively less, and most of them use empirical chart, critical point analysis, and linear regression method to predict the crude oil density, which is greatly affected by human factors, resulting in low fitting effect precision, inaccurate and comprehensive prediction of crude oil density, and inability to realize real-time quantitative prediction of crude oil density.

[0004] The prediction of crude oil density cannot be replaced by gasoline ratio. Gasoline ratio contains associated natural gas, mainly C4 and below light hydrocarbon, and trace amounts of H2S and CO2. The gas logging parameters required for crude oil density prediction not only include alkanes, but also aromatic hydrocarbons and naphthenes. The prediction of crude oil density focuses on real-time perception and quantification on the logging site, as well as high prediction accuracy.

[0005] With the continuous development of light hydrocarbon analysis technology, light hydrocarbon analysis technology is based on the concentration and distribution of light hydrocarbon compounds in oil and natural gas, the stability and the solubility in water and other physical and chemical properties, and the application of light hydrocarbon components deduces the rules of the genetic type, the degree of thermal evolution and the secondary evolution intensity of oil and natural gas, and evaluates the oil-bearing property and the properties of crude oil; the traditional gas logging system can only obtain some related data containing some components, and the quality is not high, which is not conducive to quantitative analysis; the data of the multi-component gas logging system is not affected by lithology, and can reflect the oil and gas bearing property and the physical property of the research layer in real time. The multi-component gas logging data is mostly studied from the qualitative point of view in the analysis of the properties of crude oil in complex reservoirs, for example, 'Identification of reservoir fluid based on FLAIR technology', 'New method of oil and gas layer identification based on gas logging ratio factor', 'New method of interpreting reservoir fluid properties in Bohai oilfield logging based on hydrocarbon component analysis', and there is no great breakthrough in quantitative prediction. - The data of the multi-component gas logging system is not affected by lithology, and can reflect the oil and gas bearing property and the physical property of the research layer in real time. The multi-component gas logging data is mostly studied from the qualitative point of view in the analysis of the properties of crude oil in complex reservoirs, for example, 'Identification of reservoir fluid based on FLAIR technology', 'New method of oil and gas layer identification based on gas logging ratio factor', 'New method of interpreting reservoir fluid properties in Bohai oilfield logging based on hydrocarbon component analysis', and there is no great breakthrough in quantitative prediction.

[0006] In order to overcome the above problems, the present application provides a real-time quantitative prediction method for crude oil density based on a multi-component gas logging system to realize real-time perception while drilling in logging site, and establish a real-time quantitative prediction model for crude oil density with high precision. SUMMARY

[0007] The present application aims to solve the defects of the prior art, and provides a real-time quantitative prediction method for crude oil density based on a multi-component gas logging system.

[0008] The present application adopts the following technical scheme:

[0009] A real-time quantitative prediction method for crude oil density based on a multi-component gas logging system, comprising the following steps:

[0010] 1) According to the drilling fluid sample of the research area, the content of hydrocarbon gas and specified non-hydrocarbon gas is quantitatively detected by a gas detection device;

[0011] 2) Combined with gas logging data, the curve characteristics of the gas logging component data set are analyzed;

[0012] 3) The parameters and derived parameters that can best reflect the oil abundance in the oil layer are selected and optimized;

[0013] 4) The optimized parameters are correlated with the crude oil density;

[0014] 5) According to the established gas logging data set derived parameter and crude oil density graph and regression equation, the gas logging parameters and related parameters and derived parameters such as logging and rock core are combined by a machine learning method based on BP neural network, and a multi-component gas logging parameter crude oil density quantitative prediction model of the reservoir is established.

[0015] Further, in step 1): the detection of gas components is as follows:

[0016] The hydrocarbon gases detected and analyzed by the multi-component gas logging system extend from alkanes ( , , , , , , , ) to aromatic hydrocarbons ( , ) naphthenes ( ) and non-hydrocarbon gases ( , ).

[0017] Further, in step 2): analyze the characteristics of multi-component gas logging data and compare and analyze each fluid phase;

[0018] Among the gas components, the hydrocarbon components above are in liquid state under normal conditions, so when the gas logging system detects a large amount of hydrocarbon components above , it means the presence of oil-bearing or oil layers; naphthenes and aromatic hydrocarbons are soluble in water and oil at normal temperature and pressure, but their solubility in oil is much higher than in water, so the detection value of naphthenes and aromatic hydrocarbons is higher in oil layers.

[0019] Light hydrocarbons in crude oil are composed of n-alkanes, iso-alkanes, naphthenes and aromatic hydrocarbons, and are an important component of crude oil. In particular, ~ monomer n-alkanes have the highest relative percentage content in crude oil; monomer iso-alkanes and naphthenes such as 2-methylhexane and methylcyclohexane can account for more than 1% of the components of crude oil, and benzene and toluene can account for 1.3%-1.8% of the light hydrocarbon components. The content and distribution of light hydrocarbon compounds in crude oil not only depend on the genetic type of crude oil, but also depend on the degree of thermal evolution and the intensity of secondary evolution.

[0020] Further, in step 3): derived parameters (gaseous hydrocarbons), (heavy components in liquid state at normal temperature and pressure), Pg (hydrocarbon potential), OPI (oil production index), PS (light and heavy component index), Tg (total hydrocarbons);

[0021] ;

[0022] Further, in step 4): correlation analysis of derived parameters and actual crude oil density

[0023] The most representative gas logging data in the reservoir is screened, atypical data is removed, gas logging sample data reflecting the fluid properties of the reservoir is selected, an optimized derived parameter is selected, and correlation analysis is performed on the crude oil density. The derived parameter is analyzed by a plate and a regression equation of the gas logging parameter and the derived parameter and the crude oil density, and a correlation coefficient (light component gaseous hydrocarbon), (heavy component), (hydrocarbon production potential), (oil production rate index), Tg (total hydrocarbon), Ps (light and heavy component index), and the correlation strength of the crude oil density.

[0024] Further, in step 5):

[0025] The BP neural network is composed of an input layer, a hidden layer, and an output layer, can learn and store a large number of input-output mode mapping relationships, and does not need to reveal a mathematical equation describing the mapping relationship in advance. Her learning rule is to use the steepest descent method, to continuously adjust the weights and thresholds of the network through back propagation, so that the sum of errors is minimized.

[0026] A three-layer network is defined, the input layer (0th layer), the hidden layer (1st layer), and the output layer (2nd layer), q number of neurons, the specific number is determined according to the final fitting effect, and each node is provided with a bias, and the activation function is a relu function.

[0027] The weights of the input layer to the hidden layer and the hidden layer to the output layer are respectively, and for a neural network model containing only a single hidden layer: The process of the BP neural network is mainly divided into two stages, the first stage is the forward propagation of the signal, from the input layer through the hidden layer, and finally to the output layer; the second stage is the backward propagation of the error, from the output layer to the hidden layer, and finally to the input layer, and the weights and biases of the hidden layer to the output layer and the weights and biases of the input layer to the hidden layer are adjusted in turn.

[0028] Forward propagation process:

[0029] From the input layer to the hidden layer:

[0030] From the hidden layer to the output layer:

[0031] In the formula: is the input of the hidden layer neuron; is the weight of the input layer to the hidden layer; is the input feature value, , is a bias variable, is the input of the output neuron, is the weight of the hidden layer to the output layer; The numerical coefficient of the hidden layer.

[0032] Because the parameters are random, the first calculated result has a very large error from the true result, so the parameters need to be adjusted according to the error to better fit, until the error reaches a minimum value, at which time the back propagation of the model is needed.

[0033] The back propagation process is as follows:

[0034] The network parameters are adjusted by calculating the error between the output layer and the expected value, so that the error is smaller.

[0035] Error formula:

[0036]

[0037] In the formula: is the model output value; is the actual expected value.

[0038] Weight back update: ;

[0039] In the formula: is the learning rate, is the error between the output layer and the expected value, is the reverse weight difference.

[0040] A plurality of groups of logging parameters capable of representing the density of the known crude oil are selected as input data, and are randomly divided into a training set and a validation set in proportion, and after data preprocessing and normalization processing of the obtained data, the data are input into the above-mentioned BP neural network model, and the BP neural network prediction model is trained; the prediction accuracy of the training set on the network is calculated, and the accuracy of the model is evaluated.

[0041] The beneficial effects of the present application are as follows:

[0042] The real-time quantitative prediction method for crude oil density based on the multi-component gas measurement system of the present application can obtain the parameters and related derived parameters that can best reflect the original density properties in the research horizon through the multi-component gas measurement system in real time, and input the parameters into the BP neural network for training, so as to perform real-time quantitative prediction of the crude oil density. Compared with the traditional gas measurement system prediction method, the related gas measurement parameters are more comprehensive and accurate, which provides better possibility for improving the real-time quantitative prediction accuracy of the crude oil density. Therefore, it is considered that this method is an important step for realizing more accurate real-time quantitative prediction of the crude oil density. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 is the flowchart of the embodiment of the present application;

[0044] Figure 2A schematic diagram illustrating the process of establishing a real-time quantitative prediction model for crude oil density in the study area for this invention;

[0045] Figure 3 Derived parameters are selected according to the embodiments of the present invention. , The comparison results of the prediction model based on eigenvalues ​​are shown in the figure. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention are described clearly and completely below. Obviously, the described embodiments are only some embodiments of this invention, not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0047] This invention utilizes the hydrocarbon gases detected by the gas detection equipment of a multi-component gas detection system, from alkanes ( , , , , , , , ) extended to aromatic hydrocarbons ( , )cycloalkanes ( This study explores the effective relationship between data points such as hydrocarbon components, oil abundance, derived parameter components and their contents, and crude oil density.

[0048] By decomposing formation oil and gas information into hydrocarbon component units, and finding the intrinsic relationship between hydrocarbon components and their derived parameters and reservoir characteristics, we can combine historical gas logging data and measured data such as the non-hydrocarbon gas content of hydrocarbon components with the intrinsic relationship between crude oil density.

[0049] The following section provides a detailed explanation of the real-time quantitative prediction method for crude oil density based on a multi-component gas measurement system, using relevant data from crude oil samples obtained from well HZ26-6-1.

[0050] like Figures 1-2 As shown, the present invention provides a method for real-time quantitative prediction of crude oil density based on a multi-component gas measurement system, comprising:

[0051] 1) Based on drilling fluid samples from the study area, the content of hydrocarbon gases and specified non-hydrocarbon gases was quantitatively detected using a multi-component gas logging system:

[0052] Alkanes ( , , , , , , , ) to aromatic hydrocarbons ( , ), naphthenes ( ) and non-hydrocarbon gases ( , ), in particular ~ The relative percentage of monomeric n-alkanes in the range of C15-C22 is the highest in crude oil;

[0053] 2) Combined with gas logging data, analyze and mine the curve characteristics of gas logging component data set;

[0054] Analyze the characteristics of the distribution curves of each gas logging component data of the layer, and compare and analyze each fluid phase to mine the side correlation between the fluid curve and the density of crude oil, as shown in Figure 2 ;

[0055] 3) Select the oil layer as the parameter and derived parameter that can best reflect the properties of crude oil;

[0056] (light component gaseous hydrocarbon), (heavy component liquid hydrocarbon in liquid state at normal temperature and pressure), (hydrocarbon potential), OPI (oil production index), PS (light and heavy component index), Tg (total hydrocarbon);

[0057] .

[0058] 4) Correlation analysis of derived parameters and sample crude oil density data

[0059] When selecting sample data, there are many core and rock wall analysis parameter data in the research layer section, but when establishing the fitting regression equation between the derived parameters and the density of crude oil, only one-to-one, that is, one derived parameter corresponds to one crude oil density, so in a large amount of sample data, representative data reflecting the fluid properties of the research layer are selected.

[0060] 5) According to the established gas logging data set derived parameter and crude oil density chart and regression equation, through the machine learning method based on BP neural network, the gas logging parameters and related parameters such as logging and rock core are combined with the derived parameters to establish a quantitative prediction model of the multi-component gas logging parameter crude oil density of the reservoir;

[0061] Define a three-layer network, input layer (0th layer), hidden layer (1st layer), and output layer (2nd layer). Select the multi-component gas logging parameters and derived parameters that can represent the known crude oil density as input, q number of neurons, and each node has a bias. The activation function is the relu function. ​

[0062] The process is mainly divided into two stages, the first stage is the forward propagation of signal, from the input layer through the hidden layer to the output layer; the second stage is the backward propagation of error, from the output layer to the hidden layer, and finally to the input layer, adjusting the weights and biases of the hidden layer to the output layer and the weights and biases of the input layer to the hidden layer in turn.

[0063] Forward propagation process:

[0064] From the input layer to the hidden layer: ;

[0065] From the hidden layer to the output layer: ;

[0066] In the formula: is the input of the hidden layer neuron; is the weight from the input layer to the hidden layer; is the number of feature values of the input, , is the bias variable, is the input of the output neuron, is the weight from the hidden layer to the output layer; is the hidden layer value coefficient.

[0067] The first calculated result will have a very large error from the true result, so the parameters need to be adjusted according to the error to make the parameters better fit, until the error reaches the minimum value, the back propagation of the model is needed;

[0068] Back propagation process:

[0069] Adjust the network parameters by calculating the error between the output layer and the expected value, so that the error becomes smaller.

[0070] Error formula:

[0071] In the formula: is the output value of the model; is the actual expected value.

[0072] Weight backward update: ;

[0073] In the formula: is the learning rate, is the error between the output layer and the expected value, is the reverse weight difference.

[0074] The Adam algorithm is used to quickly adjust the size of the weight, so that the loss function is constantly reduced. The Adam algorithm is based on the gradient descent method, and by calculating the first moment estimate and the second moment estimate of the gradient, independent adaptive learning rates are designed for different parameters, which can accelerate the convergence rate.

[0075] A plurality of groups of logging parameters capable of representing the density of known crude oil are selected as input data, and are randomly divided into a training set and a validation set in proportion. After data preprocessing and normalization of the obtained data, the data are input into the above-mentioned BP neural network model, and the BP neural network prediction model is trained. The prediction accuracy of the training set on the network is calculated, and the accuracy of the model is evaluated according to the training set and the control results. Figure 3 For the characteristic value, the derivative , is selected as a prediction model control result.

[0076] Table 1 is a 30-group sample set with the derivative parameter , as the characteristic value;

[0077] According to the crude oil density quantitative prediction model established according to the two preferred derivative parameters characteristic values, the prediction model is trained, and the prediction model control result Figure 3 is shown, the goodness of fit is: 0.92336, and the error is within 0.02 g / mL, which can realize the real-time quantitative prediction of the reservoir crude oil density on the logging site.

[0078] Table 1 is a 30-group sample set with the derivative parameter , as the characteristic value;

[0079]

[0080] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for real-time quantitative prediction of crude oil density based on a multi-component gas measurement system, characterized in that... It includes the following steps: Step 1. Based on drilling fluid samples from the study area, quantitatively detect the content of hydrocarbon gases and specified non-hydrocarbon gases using gas detection equipment; The detected gas components are as follows: The multi-component gas logging system detects and analyzes hydrocarbon gases, starting with alkanes. , , , , , , , Extended to aromatic hydrocarbons , Cycloalkanes non-hydrocarbon gases , ; Step 2. Combine gas measurement data to mine and analyze the curve characteristics of the gas measurement component dataset; Step 3. Select the parameters and derived parameters that best reflect the oil abundance of the oil layer and perform optimization. Step 4. Perform a correlation analysis between the preferred parameters and crude oil density; The most representative gas logging data from the reservoir were selected, and atypical data were removed. Gas logging sample data reflecting the reservoir fluid properties were selected for optimal derived parameters. Correlation analysis was then performed with crude oil density. Through plots and regression equations of gas logging parameters, derived parameters, and crude oil density, the correlation coefficients of the derived parameters were analyzed. Light components, gaseous hydrocarbons Recombination components, Hydrocarbon production potential The correlation strength between oil yield index, total hydrocarbon (Tg), light and heavy component index (Ps) and crude oil density; Step 5. Based on the established chart and regression equation of derived parameters and crude oil density of the gas logging dataset, a multi-component gas logging parameter crude oil density quantitative prediction model for the reservoir is established by combining gas logging parameters, well logging and cuttings core related parameters with derived parameters through a machine learning method based on BP neural network. A backpropagation (BP) neural network consists of an input layer, hidden layers, and an output layer. A three-layer network is defined: an input layer, a hidden layer, and an output layer, with q neurons. The exact number of neurons is determined based on the final fitting result. Each node has a bias, and the activation function is the ReLU function. These represent the weights from the input layer to the hidden layer and from the hidden layer to the output layer, respectively. For a neural network model with only a single hidden layer, the backpropagation (BP) neural network process mainly consists of two stages: the first stage is the forward propagation of the signal, from the input layer through the hidden layer and finally to the output layer; the second stage is the backward propagation of the error, from the output layer to the hidden layer and finally to the input layer, adjusting the weights and biases from the hidden layer to the output layer and from the input layer to the hidden layer in sequence. Forward propagation process: From input layer to hidden layer: ; From hidden layer to output layer: ; In the formula: This is the input to the hidden layer neurons; These are the weights from the input layer to the hidden layer; The input feature value is the numerical value. , For bias variables, For the input of the output neuron, The weights from the hidden layer to the output layer; These are the numerical coefficients of the hidden layer; Because the parameters are random, the initial calculation result will have a very large error compared to the actual result. Therefore, it is necessary to adjust the parameters according to the error to make the parameters fit better until the error reaches the minimum value. This is where backpropagation of the model is needed. Backpropagation process: By calculating the error between the output layer and the expected value, the network parameters are adjusted to reduce the error. Error formula: ; In the formula: Output values ​​for the model; Actual expected value Weighted reverse update: ; ; In the formula: For learning rate, The error between the output layer and the expected value. This is the reverse weighted difference.

2. The method for real-time quantitative prediction of crude oil density based on a multi-component gas measurement system according to claim 1, characterized in that, Step 2 also includes comparative analysis of each fluid phase: Among them, the gas-measuring components due to Hydrocarbon gas components of 2000 and above are in a liquid state under normal conditions, so when the gas logging system detects a large amount of... The presence of hydrocarbon components above + indicates the presence of an oil layer or oil reservoir. Cycloalkanes and aromatics are soluble in water and oil at room temperature and pressure, but their solubility in oil is much greater than in water, so the detection values ​​of cycloalkanes and aromatics are higher in oil reservoirs. Light hydrocarbons in crude oil consist of four parts: n-alkanes, isoalkanes, cycloalkanes, and aromatics, and are an important component of crude oil, especially... ~ Monomeric n-alkanes have the highest relative percentage content in crude oil; monomeric isoalkanes and cycloalkanes account for more than 1% of crude oil components, while benzene and toluene account for 1.3%-1.8% of light hydrocarbon components. The content and distribution of light hydrocarbon compounds in crude oil depend not only on the genetic type of crude oil but also on the degree of thermal evolution and the intensity of secondary evolution it has undergone.

3. The method for real-time quantitative prediction of crude oil density based on a multi-component gas measurement system according to claim 2, characterized in that, The cycloalkanes are 2-methylhexane and methylcyclohexane.

4. The method for real-time quantitative prediction of crude oil density based on a multi-component gas measurement system according to claim 1, characterized in that, In step 3, derived parameters: It is a gaseous hydrocarbon. The heavy components are liquid at room temperature and pressure; Pg is the hydrocarbon production potential; OPI is the oil yield index; PS is the light heavy component index; and Tg is the total hydrocarbon content. 。 5. The method for real-time quantitative prediction of crude oil density based on a multi-component gas measurement system according to claim 1, characterized in that, In step 5, the BP neural network can learn and store a large number of input-output pattern mapping relationships without revealing the mathematical equations describing these mapping relationships in advance. Its learning rule is to use the steepest descent method, based on the relevant gas measurement parameters and derived parameters of the input, to continuously adjust the weights and thresholds of the network through backpropagation, so as to minimize the sum of squared errors, and thus fit the crude oil density prediction model with the smallest error.

6. The method for real-time quantitative prediction of crude oil density based on a multi-component gas measurement system according to claim 1, characterized in that, In step 5, multiple sets of logging parameters that can characterize the known crude oil density are selected as input data and randomly divided into training and validation sets according to proportions. After data preprocessing and normalization, the obtained data is input into the BP neural network model to train the BP neural network prediction model. The prediction accuracy of the training set on the network is calculated to evaluate the accuracy of the model.

Images