CN108596780A - A kind of Reservoir type division methods and system based on multiple information - Google Patents

Abstract

The invention discloses a method and system for classifying reservoir types based on multivariate information. The method includes: acquiring reservoir type sensitive data from logging data at multiple sample points; reservoir type sensitive data including neutron data, resistivity, porosity, sonic time difference, natural gamma ray and compensated density ; Obtain the actual reservoir type of each sample point obtained by artificial reservoir division of each sample point; input the data sensitive to the reservoir type of each sample point into the deep neural network, adjust the structural parameters of the neural network so that the output of the neural network corresponds to The reservoir type is used to train the deep neural network; the trained deep neural network is used to classify the reservoir type according to the measured data that is sensitive to the reservoir type. The method and system disclosed in the invention can improve the accuracy of reservoir division.

Description

Method and system for classifying reservoir types based on multivariate information

technical field

The invention relates to the field of petroleum geophysical exploration, in particular to a method and system for classifying reservoir types based on multivariate information.

Background technique

Reservoir type classification is one of the important contents of reservoir description, and is the key work for analyzing oilfield development potential. In actual production, parameters closely related to reservoir characteristics such as shale content and permeability are usually used to classify reservoir types. Among them, shale content can reflect the lithology and physical properties of the reservoir, and permeability can characterize the flow of fluid in the rock. Difficulty of internal mobility. However, parameters such as shale content and permeability used for reservoir type classification cannot be directly obtained from well logging measurement data, but are obtained by crossplot method or empirical formula calculation based on actual well logging curves. The accuracy depends on the selection of model characteristic parameters, regional geological characteristics and stratification characteristics, and is greatly affected by human factors such as expert judgment and experience. Since each logging parameter has its specific applicable formation, and no logging parameter can be fully applied to all formations, only limited logging information is often used in the classification of reservoir types, resulting in the inconsistency of reservoir classification. Not very accurate.

Contents of the invention

The object of the present invention is to provide a method and system for classifying reservoir types based on multivariate information, so as to improve the precision of reservoir classifying.

To achieve the above object, the present invention provides the following scheme:

A method for classifying reservoir types based on multivariate information, including:

Acquiring data sensitive to reservoir type from logging data of multiple sample points; the data sensitive to reservoir type includes neutron data, resistivity, porosity, acoustic time difference, natural gamma ray and compensation density;

Obtaining the actual reservoir type of each sample point obtained by manually dividing each sample point into reservoirs;

Input the data sensitive to the reservoir type of each sample point into the deep neural network, adjust the structural parameters of the neural network to make the neural network output the corresponding reservoir type, thereby training the deep neural network; the trained deep neural network The network is used to classify the reservoir type according to the measured data sensitive to the reservoir type.

Optionally, input the data sensitive to the reservoir type of each sample point into the deep neural network, adjust the structural parameters of the neural network to make the neural network output the corresponding reservoir type, so as to train the deep neural network, specifically include:

Using the crossplot method and the kernel density estimation method to perform characteristic statistics on each of the data sensitive to the reservoir type, determine the mean and variance of each of the data sensitive to the reservoir type, and obtain the sample mean of each data and sample variance;

standardizing each of the reservoir type-sensitive data by using the sample mean and sample variance of each data to obtain standardized data;

The standardized data of each type of each sample point is used as the input of the deep neural network, and the linear rectification function is used as the activation function of the deep neural network. By adjusting the structural parameters of the deep neural network, the depth The difference between the reservoir type data output by the neural network and the corresponding actual reservoir type is within a preset range, thereby completing the training of the deep neural network.

Optionally, the process of classifying the reservoir type according to the measured data sensitive to the reservoir type includes:

Obtain the measured data that is sensitive to the reservoir type and obtain the measured data;

Calculating the mean and variance of each of the measured data to obtain the measured mean and the variance of each of the measured data;

Using the sample mean and sample variance of each data to calculate the mean threshold range and variance threshold range of each data;

judging whether the measured mean of each type of measured data is within the corresponding mean threshold range and whether the measured variance of each type of measured data is within the corresponding variance threshold range, to obtain a first judgment result;

If the first judgment result indicates that the measured mean of any type of measured data is outside the corresponding mean threshold range and/or the measured variance is outside the corresponding variance threshold range, re-collect the corresponding type of data;

If the first judgment result indicates that the measured mean of each measured data is within the corresponding mean threshold range and the measured variance is within the corresponding variance threshold range, then each measured data is standardized to obtain standard measured data;

The standard measured data of each type is used as the input of the deep neural network to obtain the measured reservoir type output by the neural network.

The present invention also discloses a reservoir type classification system based on multivariate information, including:

The sample data acquisition module is used to acquire data sensitive to the reservoir type from the logging data of multiple sample points; the data sensitive to the reservoir type includes neutron data, resistivity, porosity, acoustic time difference, natural gamma and compensation density;

The actual reservoir type acquisition module is used to obtain the actual reservoir type of each sample point obtained by manually dividing each sample point into reservoirs;

The neural network training module is used to input the data sensitive to the reservoir type of each sample point into the deep neural network, adjust the structural parameters of the neural network to make the neural network output the corresponding reservoir type, thereby training the deep neural network; The trained deep neural network is used to classify reservoir types according to measured data sensitive to reservoir types.

Optionally, the neural network training module specifically includes:

The sample data mean value and variance calculation unit is used to perform characteristic statistics on each of the data sensitive to the reservoir type by using the crossplot method and the kernel density estimation method, and determine the mean value of each of the data sensitive to the reservoir type and variance to get the sample mean and sample variance of each data;

A sample data standardization unit, configured to standardize each of the data sensitive to reservoir types by using the sample mean and sample variance of each data to obtain standardized data;

The training unit is used to use the standardized data of each type of each sample point as the input of the deep neural network, and use the linear rectification function as the activation function of the deep neural network, by adjusting the structural parameters of the deep neural network , so that the difference between the reservoir type data output by the deep neural network and the corresponding actual reservoir type is within a preset range, thereby completing the training of the deep neural network.

Optionally, the system for classifying reservoir types further includes a reservoir classifying module, configured to classify reservoir types according to measured data sensitive to reservoir types; the reservoir classifying module specifically includes:

The measured data acquisition unit is used to obtain the measured data sensitive to the reservoir type and obtain the measured data;

The measured data mean and variance calculation unit is used to calculate the mean and variance of each of the measured data to obtain the measured mean and the measured variance of each of the measured data;

A threshold range calculation unit, configured to use the sample mean and sample variance of each data to calculate the mean threshold range and variance threshold range of each data;

A judging unit, configured to judge whether the measured mean of each type of measured data is within a corresponding mean threshold range and whether the measured variance of each type of measured data is within a corresponding variance threshold range, to obtain a judgment result;

The data re-acquisition unit is used to re-acquire the corresponding type of data if the judgment result indicates that the measured mean of any type of measured data is outside the corresponding mean threshold range and/or the measured variance is outside the corresponding variance threshold range;

The measured data standardization unit is used to standardize each measured data to obtain standard measured data if the judgment result indicates that the measured mean of each measured data is within the corresponding mean threshold range and the measured variance is within the corresponding variance threshold range ;

The reservoir division unit is configured to use the standard measured data of each type as the input of the deep neural network to obtain the measured reservoir type output by the neural network.

According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects: The method and system for classifying reservoir types based on multivariate information disclosed in the present invention, through the combination of neutron data, resistivity, porosity, acoustic time difference, natural gamma For various data such as horse and compensation density, the learning of deep neural network is used to establish the complex nonlinear relationship between various data and reservoir types, and realize the automatic and intelligent division of reservoir types. The invention avoids the interference of human factors through automatic and intelligent division, improves the precision of reservoir type division, utilizes multiple data at the same time, realizes the fusion of multivariate information, and further improves the precision of reservoir division.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required in the embodiments. Obviously, the accompanying drawings in the following description are only some of the present invention. Embodiments, for those of ordinary skill in the art, other drawings can also be obtained according to these drawings without paying creative labor.

Fig. 1 is the method flowchart of the embodiment of the reservoir type classification method based on multivariate information of the present invention;

Fig. 2 is a map of effective distribution ranges of different logging parameters in different types of reservoirs;

Fig. 3 is the network structural diagram of the depth neural network that the present invention adopts;

Fig. 4 is a schematic diagram of a function curve in which the objective function gradually becomes gentle during the deep neural network training process;

Fig. 5 is the reservoir division result figure that adopts the reservoir type division method based on multivariate information of the present invention to obtain;

Fig. 6 is a system structure diagram of an embodiment of the system for classifying reservoir types based on multivariate information in the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

The purpose of the present invention is to make full use of multivariate logging information such as acoustic discharge to carry out direct division of reservoir types, use logging multivariate information in actual production data to establish reservoir type sample library, and use deep learning network to construct multivariate well logging information and The complex nonlinear relationship between reservoir types, comprehensive utilization of multiple logging information such as acoustic discharge, so that the reservoir types in the actual logging data can be directly divided, and the indirect calculation of parameters such as shale content and permeability can be avoided. The uncertainty existing in the classification of reservoir types can avoid the influence of human factors on the classification of reservoir types, and improve the automation and intelligence level of reservoir type prediction.

In order to make the above objects, features and advantages of the present invention more comprehensible, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

Fig. 1 is a method flowchart of an embodiment of the method for classifying reservoir types based on multivariate information in the present invention.

Referring to Fig. 1, the reservoir type classification method based on multivariate information includes:

Step 101: Obtain data sensitive to reservoir type from the logging data of multiple sample points; the data sensitive to reservoir type includes neutron data, resistivity, porosity, sonic time difference, natural gamma ray and compensation density.

By screening various types of logging information such as acoustic waves, radioactivity, and resistivity in the existing logging data in the research area, sort out natural gamma ray, resistivity, spontaneous potential, neutron, etc. that are sensitive to lithology. Well curves, neutron, density, sound wave and other related well logging curves sensitive to permeability. The present invention considers that when classifying reservoir types in actual production, it is mainly based on shale content and permeability parameters, and selects six logging parameters such as neutron data, resistivity, porosity, acoustic time difference, natural gamma ray, and compensation density as Multiple sources of information for reservoir type classification. It is also possible to further add other parameters on this basis, but the premise is that the added parameters have valid values for each reservoir type, and then eliminate the data of serious borehole wall collapse and the data caused by logging instruments or measurement errors. Abnormal values provide the basic data basis for building numerical sample libraries of reservoir types.

Step 102: Obtain the actual reservoir type of each sample point obtained by manually dividing each sample point into reservoirs.

Build a library of six logging parameters selected from neutron data, resistivity, porosity, acoustic time difference, natural gamma ray, and compensation density, carry out environmental correction according to the characteristics of each well data, and unify the units to international standards Units to ensure that the data measured by different instruments and operators in different periods can be effectively used. On this basis, according to the existing shale content, permeability parameters and oil test results, the data samples at different depths are manually divided into reservoir types, and the data of each sample point are divided into non-reservoir, type I There are three types of reservoirs (relatively ideal oil and gas reservoirs with great development potential) and Type II reservoirs (possible oil and gas reservoirs with certain development potentials), and they are marked with numbers 0, 1 and 2 respectively , to establish a reservoir type numerical sample library of multivariate logging information at each sample point. The non-reservoirs are reservoirs with large shale content, basically no development potential; the type I reservoirs are relatively ideal oil and gas reservoirs with great development potential; the type II reservoirs are possible oil and gas reservoirs It has a certain development potential.

Step 103: Input the data sensitive to the reservoir type of each sample point into the deep neural network, adjust the structural parameters of the neural network so that the neural network outputs the corresponding reservoir type, thereby training the deep neural network; the trained The deep neural network is used to classify the reservoir type according to the measured data sensitive to the reservoir type.

This step 103 specifically includes:

Using the crossplot method and the kernel density estimation method to perform characteristic statistics on each of the data sensitive to the reservoir type, determine the mean and variance of each of the data sensitive to the reservoir type, and obtain the sample mean of each data and sample variance;

standardizing each of the reservoir type-sensitive data by using the sample mean and sample variance of each data to obtain standardized data;

The standardized data of each type of each sample point is used as the input of the deep neural network, and the linear rectification function is used as the activation function of the deep neural network. By adjusting the structural parameters of the deep neural network, the depth The difference between the reservoir type data output by the neural network and the corresponding actual reservoir type is within a preset range, thereby completing the training of the deep neural network.

This step 103 can specifically be implemented in the following specific implementation manners:

After the reservoir type sample library is established, the multivariate parameters in the sample library must be characterized and standardized to ensure that the distribution range of each parameter in the sample library is uniform and reasonable, and at the same time provide parameters for the prediction of subsequent actual data reservoir types in accordance with. The present invention adopts the crossplot method and the kernel density estimation map method to jointly carry out the feature statistical work of the sample library (the crossplot method is that every two parameters are respectively analyzed for the distribution characteristics of statistical data, and the kernel density method is used in probability theory to estimate parameter density function, so as to obtain the statistical features required below), obtain the mean value μ _i and variance σ _i of each parameter in the sample library (i is the number of the data type, and there are neutron data, resistivity, porosity, acoustic wave Time difference, natural gamma ray, and compensation density are six logging parameters, so i=1, 2, ..., 6), to clarify the effective distribution range of different logging parameters in different types of reservoirs, Fig. 2 shows that in different types of reservoirs Effective range map for different logging parameters. The distribution range is used to judge whether the measured data can be used to judge the reservoir type during the actual measurement. The degree of discrimination each logging parameter has for the reservoir type class is then determined based on the kernel density estimate peak and bandwidth for each parameter. Especially when the parameter distribution range is uneven and the number of reservoir type samples is insufficient, it is necessary to continue to supplement the sample library information, that is, the number of samples does not meet the training conditions, and it is necessary to continue to collect more and more sufficient data for the construction of reservoir type and multivariate information. lay the foundation for a reasonable relationship between them.

Considering the influence of factors such as construction teams, instruments and measurements in the actual production data, it is necessary to standardize the multivariate parameter information and optimize the data structure of the reservoir type sample library. The present invention adopts the following formula to standardize the data in the multivariate information base to ensure that the data after the standardization conform to a normal distribution with a mean value of 0 and a variance of 1, namely:

Among them, μ _i is the mean value of the corresponding logging data in the sample library, σ _i is the variance of the corresponding sample data, i is the data type number (i=1, 2, ..., 6), and j represents different sample points (j= 1, 2, ..., n (n is the total number of sample points)), x _ij represents the parameters before standardization, Z _ij represents the parameters after standardization.

After the standardization is completed, the structure of the neural network needs to be defined. When designing a deep neural network, it is necessary to carry out precise design according to the characteristics of the reservoir type sample library, and specify the network structure parameters such as the number of layers of the deep neural network, the number of neurons in each layer, and the activation function. Fig. 3 is a network structure diagram of the deep neural network adopted in the present invention. Referring to Fig. 3, the present invention constructs a deep neural network model comprising 5 hidden layers, wherein the number of neurons in each hidden layer is 20, 40, 80, 40, 20 respectively. In the network design, the linear rectification function (Relu function) is used as the activation function to effectively alleviate the overfitting problem and accelerate the network convergence process, and a Softmax layer is added before the output layer. The network structure can be expressed in matrix form as:

g(W ⁽¹⁾ a ⁽¹⁾ +b ⁽¹⁾ )＝a ⁽²⁾

g(W ⁽²⁾ a ⁽²⁾ +b ⁽²⁾ )＝a ⁽³⁾

...

g(W ^(n-1) a ^(n-1) +b ^(n-1) )=y

Wherein, g(x) represents the operational form of the activation function, and the activation function adopted in the present invention is a ReLu function, i.e. g(x)=max(0,x), and n represents the number of layers of the deep neural network (n= among the present invention 6, that is, the number of hidden layers plus the sum of the output layer), b ⁽ⁱ⁾ represents the bias item, where a ⁽ⁱ⁾ (i=1,...,n-1) is the feature vector of the middle layer (when i=1 , a ⁽¹⁾ represents the multivariate logging information in the sample library used), W ⁽ⁱ⁾ represents the weight matrix mapped from the i-th layer to the i+1-th layer in the network structure, and y represents the marked type of reservoir.

After the definition of the network structure is completed, all the data in the sample library of the multivariate information storage layer is used to input into the deep neural network for training. During the training, the model parameters (that is, the weight value of the network W ⁽ⁱ⁾ and the bias item b ⁽ⁱ⁾ ), adjust the model parameter values through repeated iterations to ensure that the multivariate information in the sample library passes through the network and the output reservoir type y' and the reservoir type y in the sample library reach the objective function error The minimum, specific expression of the objective function is as follows:

In the training process, the present invention adopts a regularization method to prevent overfitting, that is, a part of neurons are randomly not involved in training during each iteration training, thereby reducing the interdependence between neurons and improving the performance of the neural network. Generalization ability; an optimization algorithm based on gradient descent is used to update the model parameters in the training iteration process, that is, the gradient of each parameter in the network model is calculated according to the objective function, and the network parameters are updated using the gradient value, and the model is set The maximum number of iterations of the parameter is 5000 times, and the training can be stopped when the training reaches the maximum number of iterations or the objective function gradually becomes smoother during the iteration process. Figure 4 shows the function that the objective function gradually becomes smoother during the training process of the deep neural network Schematic diagram of the curve. It can be seen from the figure that the slope of the objective function gradually becomes lower, which means that the objective function gradually becomes gentler. When this condition is met, the trained network model parameters W ⁽ⁱ⁾ and b ⁽ⁱ⁾ (i=1, . . . , n−1) can be saved.

After the training is completed, the training results can be used to carry out the division of reservoir types.

The process of classifying the reservoir type according to the measured data sensitive to the reservoir type includes:

Obtain the measured data that is sensitive to the reservoir type and obtain the measured data;

Calculating the mean and variance of each of the measured data to obtain the measured mean and the variance of each of the measured data;

Using the sample mean and sample variance of each data to calculate the mean threshold range and variance threshold range of each data;

judging whether the measured mean of each type of measured data is within the corresponding mean threshold range and whether the measured variance of each type of measured data is within the corresponding variance threshold range, to obtain a first judgment result;

If the first judgment result indicates that the measured mean of any type of measured data is outside the corresponding mean threshold range and/or the measured variance is outside the corresponding variance threshold range, re-collect the corresponding type of data;

If the first judgment result indicates that the measured mean of each measured data is within the corresponding mean threshold range and the measured variance is within the corresponding variance threshold range, then each measured data is standardized to obtain standard measured data;

The standard measured data of each type is used as the input of the deep neural network to obtain the measured reservoir type output by the neural network.

Specifically, the following implementation methods can be used to classify reservoir types:

When carrying out the classification of actual data reservoir types, the required multivariate information (namely neutron data, resistivity, porosity, acoustic time difference, natural gamma ray, and compensation density 6 types of data) is first screened out from a large number of logging data, and the data are eliminated. Outliers, calculate the average value μ' and variance σ' of each multivariate information in the actual data, and compare the average value μ' and variance σ' with the multivariate information statistical range in the sample library established by the present invention, in the comparison The following standards apply:

μ×(1-α)＜μ’＜μ×(1+α)

σ×(1-2α)<σ’<σ×(1+2α)

Wherein, α is the confidence parameter of the predicted data, and the present invention sets α=0.1. When the actual data characteristics do not satisfy the above formula, it indicates that the multivariate data to be divided into reservoir types differs too much from the distribution characteristics of the information in the sample database, which does not meet the basic conditions for reservoir type division, and it is necessary to recheck whether the data is reasonable. If it is unreasonable, the data needs to be collected again. When the statistical characteristics of the input actual data satisfy the above formula, the data can be directly classified into reservoir types using this method. Then, the multivariate information in the data to be predicted is standardized, and the multivariate logging information actually input is used to classify the reservoir types by using the network model parameters W and b obtained from training, so as to obtain the reservoir type classification results corresponding to the data and Output as a graph.

Fig. 5 is a result diagram of reservoir division obtained by adopting the multivariate information-based reservoir type division method of the present invention.

Referring to FIG. 5 , the reservoir type division based on multivariate information of the present invention can improve the accuracy of reservoir division.

Fig. 6 is a system structure diagram of an embodiment of the system for classifying reservoir types based on multivariate information in the present invention.

Referring to Fig. 6, the reservoir type classification system based on multivariate information includes:

The sample data acquisition module 601 is used to acquire data sensitive to the reservoir type from the logging data of multiple sample points; the data sensitive to the reservoir type includes neutron data, resistivity, porosity, acoustic time difference, Natural gamma and compensated density;

An actual reservoir type acquisition module 602, configured to acquire the actual reservoir type of each sample point obtained by manually dividing each sample point into a reservoir;

The neural network training module 603 is used to input the data sensitive to the reservoir type of each sample point into the deep neural network, and adjust the structural parameters of the neural network to make the neural network output the corresponding reservoir type, thereby training the deep neural network ; The trained deep neural network is used to classify the reservoir type according to the measured data sensitive to the reservoir type.

The neural network training module 603 specifically includes:

The sample data mean value and variance calculation unit is used to perform characteristic statistics on each of the data sensitive to the reservoir type by using the crossplot method and the kernel density estimation method, and determine the mean value of each of the data sensitive to the reservoir type and variance to get the sample mean and sample variance of each data;

A sample data standardization unit, configured to standardize each of the data sensitive to reservoir types by using the sample mean and sample variance of each data to obtain standardized data;

The training unit is used to use the standardized data of each type of each sample point as the input of the deep neural network, and use the linear rectification function as the activation function of the deep neural network, by adjusting the structural parameters of the deep neural network , so that the difference between the reservoir type data output by the deep neural network and the corresponding actual reservoir type is within a preset range, thereby completing the training of the deep neural network.

The reservoir division module 604 is used to divide the reservoir type according to the measured data sensitive to the reservoir type; the reservoir division module 604 specifically includes:

The measured data acquisition unit is used to obtain the measured data sensitive to the reservoir type and obtain the measured data;

The measured data mean and variance calculation unit is used to calculate the mean and variance of each of the measured data to obtain the measured mean and the measured variance of each of the measured data;

A threshold range calculation unit, configured to use the sample mean and sample variance of each data to calculate the mean threshold range and variance threshold range of each data;

A judging unit, configured to judge whether the measured mean of each type of measured data is within a corresponding mean threshold range and whether the measured variance of each type of measured data is within a corresponding variance threshold range, to obtain a judgment result;

The data re-acquisition unit is used to re-acquire the corresponding type of data if the judgment result indicates that the measured mean of any type of measured data is outside the corresponding mean threshold range and/or the measured variance is outside the corresponding variance threshold range;

The measured data standardization unit is used to standardize each measured data to obtain standard measured data if the judgment result indicates that the measured mean of each measured data is within the corresponding mean threshold range and the measured variance is within the corresponding variance threshold range ;

The reservoir division unit is configured to use the standard measured data of each type as the input of the deep neural network to obtain the measured reservoir type output by the neural network.

According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects: The method and system for classifying reservoir types based on multivariate information disclosed in the present invention, through the combination of neutron data, resistivity, porosity, acoustic time difference, natural gamma For various data such as horse and compensation density, the learning of deep neural network is used to establish the complex nonlinear relationship between various data and reservoir types, and realize the automatic and intelligent division of reservoir types. The invention avoids the interference of human factors through automatic and intelligent division, improves the precision of reservoir type division, utilizes multiple data at the same time, realizes the fusion of multivariate information, and further improves the precision of reservoir division.

As for the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and for the related information, please refer to the description of the method part.

In this paper, specific examples have been used to illustrate the principle and implementation of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; meanwhile, for those of ordinary skill in the art, according to the present invention Thoughts, there will be changes in specific implementation methods and application ranges. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims (6)

1. A reservoir type classification method based on multivariate information, characterized in that it comprises: Acquiring data sensitive to reservoir type from logging data of multiple sample points; the data sensitive to reservoir type includes neutron data, resistivity, porosity, acoustic time difference, natural gamma ray and compensation density; Obtaining the actual reservoir type of each sample point obtained by manually dividing each sample point into reservoirs; Input the data sensitive to the reservoir type of each sample point into the deep neural network, adjust the structural parameters of the neural network to make the neural network output the corresponding reservoir type, thereby training the deep neural network; the trained deep neural network The network is used to classify the reservoir type according to the measured data sensitive to the reservoir type. 2. the reservoir type classification method based on multivariate information according to claim 1, is characterized in that, described each sample point is input deep neural network to the data that is sensitive to reservoir type, adjusts the structure parameter of neural network to make neural network The network outputs the corresponding reservoir type, thereby training the deep neural network, specifically including: Using the crossplot method and the kernel density estimation method to perform characteristic statistics on each of the data sensitive to the reservoir type, determine the mean and variance of each of the data sensitive to the reservoir type, and obtain the sample mean of each data and sample variance; standardizing each of the reservoir type-sensitive data by using the sample mean and sample variance of each data to obtain standardized data; The standardized data of each type of each sample point is used as the input of the deep neural network, and the linear rectification function is used as the activation function of the deep neural network. By adjusting the structural parameters of the deep neural network, the depth The difference between the reservoir type data output by the neural network and the corresponding actual reservoir type is within a preset range, thereby completing the training of the deep neural network. 3. the reservoir type division method based on multivariate information according to claim 2, is characterized in that, according to the data that is sensitive to reservoir type according to actual measurement, the process that reservoir type is divided comprises: Obtain the measured data that is sensitive to the reservoir type and obtain the measured data; Calculating the mean and variance of each of the measured data to obtain the measured mean and the variance of each of the measured data; Using the sample mean and sample variance of each data to calculate the mean threshold range and variance threshold range of each data; judging whether the measured mean of each type of measured data is within the corresponding mean threshold range and whether the measured variance of each type of measured data is within the corresponding variance threshold range, to obtain a first judgment result; If the first judgment result indicates that the measured mean of any type of measured data is outside the corresponding mean threshold range and/or the measured variance is outside the corresponding variance threshold range, re-collect the corresponding type of data; If the first judgment result indicates that the measured mean of each measured data is within the corresponding mean threshold range and the measured variance is within the corresponding variance threshold range, then each measured data is standardized to obtain standard measured data; The standard measured data of each type is used as the input of the deep neural network to obtain the measured reservoir type output by the neural network. 4. A reservoir type classification system based on multivariate information, characterized in that it comprises: The sample data acquisition module is used to acquire data sensitive to the reservoir type from the logging data of multiple sample points; the data sensitive to the reservoir type includes neutron data, resistivity, porosity, acoustic time difference, natural gamma and compensation density; The actual reservoir type acquisition module is used to obtain the actual reservoir type of each sample point obtained by manually dividing each sample point into reservoirs; The neural network training module is used to input the data sensitive to the reservoir type of each sample point into the deep neural network, adjust the structural parameters of the neural network to make the neural network output the corresponding reservoir type, thereby training the deep neural network; The trained deep neural network is used to classify reservoir types according to measured data sensitive to reservoir types. 5. the reservoir type classification system based on multivariate information according to claim 4, is characterized in that, described neural network training module, specifically comprises: The sample data mean value and variance calculation unit is used to perform characteristic statistics on each of the data sensitive to the reservoir type by using the crossplot method and the kernel density estimation method, and determine the mean value of each of the data sensitive to the reservoir type and variance to get the sample mean and sample variance of each data; A sample data standardization unit, configured to standardize each of the data sensitive to reservoir types by using the sample mean and sample variance of each data to obtain standardized data; The training unit is used to use the standardized data of each type of each sample point as the input of the deep neural network, and use the linear rectification function as the activation function of the deep neural network, by adjusting the structural parameters of the deep neural network , so that the difference between the reservoir type data output by the deep neural network and the corresponding actual reservoir type is within a preset range, thereby completing the training of the deep neural network. 6. the reservoir type classification system based on multivariate information according to claim 5, is characterized in that, also comprises reservoir classification module, is used for classifying reservoir type according to the data sensitive to reservoir type according to actual measurement; The reservoir division module described above specifically includes: The measured data acquisition unit is used to obtain the measured data sensitive to the reservoir type and obtain the measured data; The measured data mean and variance calculation unit is used to calculate the mean and variance of each of the measured data to obtain the measured mean and the measured variance of each of the measured data; A threshold range calculation unit, configured to use the sample mean and sample variance of each data to calculate the mean threshold range and variance threshold range of each data; A judging unit, configured to judge whether the measured mean of each type of measured data is within the corresponding mean threshold range and whether the measured variance of each type of measured data is within the corresponding variance threshold range, to obtain a judgment result; The data re-acquisition unit is used to re-acquire the corresponding type of data if the judgment result indicates that the measured mean of any type of measured data is outside the corresponding mean threshold range and/or the measured variance is outside the corresponding variance threshold range; The measured data standardization unit is used to standardize each measured data to obtain standard measured data if the judgment result indicates that the measured mean of each measured data is within the corresponding mean threshold range and the measured variance is within the corresponding variance threshold range ; The reservoir division unit is configured to use the standard measured data of each type as the input of the deep neural network to obtain the measured reservoir type output by the neural network.