Offshore oilfield development main control factor screening method

Abstract

The invention relates to an offshore oilfield development main control factor screening method, which comprises the following steps: S10, acquiring historical data of various associated factor parameter values of different types of offshore oilfield development, the associated factors including static factors, dynamic factors and management index factors; s20, adopting a correlation analysis algorithm to calculate an association relationship between the association factors corresponding to the historical data; s30, constructing a training set according to the historical data and the association relationship between the association factors, and training a neural network model of a preset network structure by using the training set to obtain a trained neural network model; and S40, the trained neural network model is used to predict the offshore oilfield with main control factors to be predicted, and the main control factors are obtained according to the association relationship of the association factors corresponding to the offshore oilfield to be predicted.

Description

Technical Field

[0001] The invention relates to the technical field of offshore oil and gas development, and in particular to a method for screening main control factors of different types of offshore oil field development laws. Background Art

[0002] There are many types and numbers of offshore oil fields, as well as many production wells. The dynamic data is complex and the data volume is at the big data level.

[0003] There are many factors related to the development of different types of offshore oil fields, including, for example, structure, sediment type, lithology, physical properties, liquid volume changes, pressure maintenance level, oil production rate, well-controlled reserves, etc.

[0004] In the traditional extensive mining stage, there are fewer defined correlation factors and the dimensions of each factor are low. The grasp of the development laws of offshore oil fields is based on selecting several factors as the main controlling factors through human experience, and identifying the development laws based on the main controlling factors.

[0005] However, in the era of big data for offshore development, the above-mentioned related factors are characterized by multiple indicators and dimensions, as well as time domain characteristics. It is no longer realistic to select the main controlling factors based on human experience. How to identify the main controlling factors for the development of different types of offshore oil fields has become an urgent problem to be solved. Summary of the invention

[0006] In response to the above problems, the purpose of this application is to provide a method for screening the main controlling factors of offshore oil field development, which uses a preset algorithm based on historical data to obtain a training set, then trains a neural network model based on the training set, and then uses the neural network model to identify the main controlling factors of the offshore oil field to be predicted, which can be used as a reference indicator for identifying its development rules.

[0007] To achieve the above objectives, this application adopts the following technical solutions:

[0008] In a first aspect, the present application provides a method for screening main controlling factors for offshore oilfield development, the method comprising:

[0009] S10, obtaining historical data of parameter values ​​of various correlation factors for different types of offshore oilfield development, wherein the correlation factors include static factors, dynamic factors, and management index factors;

[0010] S20, using a correlation analysis algorithm to calculate the correlation relationship between the correlation factors corresponding to the historical data;

[0011] S30, constructing a training set according to the historical data and the correlation relationship between the correlation factors, and using the training set to train a neural network model with a preset network structure to obtain a trained neural network model;

[0012] S40, using the trained neural network model to predict the offshore oil field with the main controlling factor to be predicted, wherein the main controlling factor is obtained according to the correlation relationship between the associated factors corresponding to the offshore oil field to be predicted.

[0013] In one implementation, in S10, the correlation analysis algorithm includes a correlation coefficient method.

[0014] In one implementation, the correlation analysis algorithm also includes a grey correlation analysis method.

[0015] In one implementation, the correlation analysis algorithm also includes a decision tree analysis method.

[0016] In one implementation, the neural network model of the preset network structure includes a controllable residual network GRN and a variable selection network VSN.

[0017] In one implementation, an attention mechanism module is added to the network structure of the neural network model.

[0018] In one implementation, the main controlling factors are based on the top 30% of the correlation factors in each type of correlation factors.

[0019] In a second aspect, a computer-readable storage medium is provided, storing a computer program, which, when executed by a processor, implements the method for screening main controlling factors for offshore oilfield development of the first aspect.

[0020] The present invention adopts the above technical scheme and proposes a Gated Residual Networks-Variable Selection Networks (GRN-VSN) neural network algorithm for correlation analysis, which optimizes indicators with higher correlation and discards indicators with lower correlation, so as to provide a reference for accurately understanding the development laws of oil fields of different types and at different development stages. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Correlation analysis of various indicators;

[0022] Figure 2 Decision tree and random forest algorithms;

[0023] Figure 3 GRN-VSN algorithm structure diagram;

[0024] Figure 4 The computational efficiency of GRN-VSN and traditional algorithms;

[0025] Figure 5 GRN-VSN algorithm flow chart;

[0026] Figure 6 Grey correlation algorithm crude oil recovery degree correlation calculation results;

[0027] Figure 7 The calculation results of the correlation of crude oil recovery degree by GRN-VSN neural network;

[0028] Figure 8 Comparison of verification effects between GRN-VSN and grey relational analysis;

[0029] Fig. 9 Correlation between post-average indicators and crude oil recovery degree. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.

[0031] In view of the problems existing in the prior art, the present invention provides a method for screening the main controlling factors of offshore oilfield development, the method comprising: S10, obtaining historical data of parameter values ​​of various correlation factors of different types of offshore oilfield development, the correlation factors comprising static factors, dynamic factors and management index factors; S20, using a correlation analysis algorithm to calculate the correlation relationship between the various correlation factors corresponding to the historical data; S30, constructing a training set according to the historical data and the correlation relationship between the various correlation factors, and using the training set to train a neural network model with a preset network structure to obtain a trained neural network model; S40, using the trained neural network model to predict the offshore oilfields with the main controlling factors to be predicted, wherein the main controlling factors are obtained according to the correlation relationship between the various correlation factors corresponding to the offshore oilfields to be predicted.

[0032] The above method and process are described below in more detailed embodiments in conjunction with more drawings of the present invention.

[0033] The method flow in this embodiment includes:

[0034] S10, obtaining historical data of parameter values ​​of various correlation factors for different types of offshore oilfield development, wherein the correlation factors include static factors, dynamic factors, and management index factors;

[0035] Specifically, this embodiment divides the associated factors of the oil field into static factors, dynamic factors and management index factors.

[0036] Static factors include structure, sediment type, lithology, physical properties (permeability, porosity), thickness, fluid, connectivity, etc.).

[0037] Dynamic factors include liquid volume changes, pressure maintenance level, oil production rate, well-controlled reserves, water storage rate (water consumption rate), etc.

[0038] Management indicator factors include the completion rate of dynamic detection plan and the qualified rate of injection in each injection well layer.

[0039] S20, using a correlation analysis algorithm to calculate the correlation relationship between the correlation factors corresponding to the historical data;

[0040] Specifically, the correlation analysis algorithm includes the correlation coefficient method, the grey correlation analysis method and the decision tree analysis method.

[0041] The correlation coefficient method usually evaluates the correlation between features by constructing a linear function relationship between two features. However, in offshore oil and gas development projects, it is found that the correlation between features is not always a simple linear relationship. There are also multiple correlation relationships such as exponential correlation and polynomial correlation. Therefore, there are many ways to calculate the correlation coefficient. Commonly used correlation coefficients include Pearson correlation coefficient, Spearman rank correlation coefficient and Kendall rank correlation coefficient.

[0042] Grey Relational Analysis (GRA) is a schematic diagram as shown below: Figure 1 As shown, it can be used to determine the correlation between various indicators. Different from the correlation coefficient method that analyzes data pairwise, the grey correlation analysis method regards all indicators as a whole, and the analysis of correlation is based on the overall system. Grey correlation analysis requires first clarifying the parent sequence and the subsequence. In this article, the parent sequence is the key development indicator, and the subsequence is the selected development indicator system. After clarifying the parent sequence and the subsequence, each sequence needs to be dimensionless. Commonly used methods include normalization and averaging. After completing the preprocessing operation, the correlation coefficient can be calculated. First, it is necessary to calculate the absolute difference between the subsequence and the parent sequence after preprocessing, that is, |x i -x0|,i∈[1,n], x0 is the value of the parent sequence. After interpolation of all data, the global minimum and global maximum values ​​need to be screened out:

[0043]

[0044] The calculation formula of grey correlation coefficient is:

[0045]

[0046] Among them, ρ is the resolution coefficient, which takes a value between [0,1] and is usually 0.5. The mean of the calculated ζ(k) is calculated respectively, and the calculated value is the grey correlation between the current sequence and the parent sequence. The closer the grey correlation is to 1, the stronger the correlation between the two variables.

[0047] A decision tree is a tree structure consisting of nodes and edges. Figure 2 As shown. The construction of a decision tree generally includes three parts: feature selection, tree generation, and pruning. It should be pointed out that tree generation is a recursive process. Generally speaking, as the division process continues, it is hoped that the samples contained in the branch nodes of the decision tree belong to the same category as much as possible, that is, the "purity" of the nodes is getting higher and higher. Pruning is to increase the generalization ability of the model and prevent overfitting. Multiple decision trees constitute a random forest. Each tree will produce a result, and the category with more votes is the final classification result.

[0048] S30, constructing a training set according to the historical data and the correlation relationship between the correlation factors, and using the training set to train a neural network model with a preset network structure to obtain a trained neural network model;

[0049] Based on Gated Residual Networks (GRN) and Variable Selection Networks (VSN) to perform feature correlation analysis, such as Figure 3 As shown. GRN controls the residual module by introducing the concept of gates, which can effectively prevent gradient descent and gradient explosion problems, ensure the accuracy of constructed functions, and achieve accurate results when constructing complex functions. The gate is usually composed of Sigmoid functions. The attention mechanism enables the network to pay more attention to features related to the analysis of the main controlling factors, so reliable results can be obtained when applied to other reservoir types. Therefore, the present invention adds an attention mechanism to the GRN-VSN network to enhance the robustness of network predictions and improve the generalization ability of network predictions. Figure 4 As shown in the figure, the computational efficiency of the GRN-VSN algorithm is 30 times higher than that of the traditional algorithm, where the X-axis coordinate is time and the Y-axis coordinate is the CR consistency ratio, that is, the ratio of the actual consistency index to the random consistency index.

[0050] During the training process, the data set is first preprocessed. In order to enable the computer to understand and process the categorical static indicators, it needs to be encoded using one-hot encoding. This method can retain the independence between discrete categories to a great extent, reduce the error caused by the quantization of static indicators, and perfectly combine static indicators with well data. Figure 5As shown in the figure, the processed data first enters the variable selection network (VSN), where variables are selected to create multiple sets of training subsets. Different subsets are then put into the gated residual network (GRN) for training. Finally, the importance of each feature is selected based on the accuracy of all subsets in verification.

[0051] S40, using the trained neural network model to predict the offshore oil field with the main controlling factor to be predicted, wherein the main controlling factor is obtained according to the correlation relationship between the associated factors corresponding to the offshore oil field to be predicted.

[0052] The above process can realize the identification of the main controlling factors.

[0053] In one example, the neural network model and the correlation algorithm are used to compare the oil fields to be predicted. The selected indicators and crude oil recovery data are analyzed for correlation using two correlation detection algorithms: gray correlation analysis and GRN-VSN neural network. The correlations calculated by the two methods are combined to optimize the indicators used for recovery rate prediction, and the indicators with low correlation are discarded.

[0054] In this example, if Figure 6 As shown in the figure, the correlation between (a) dynamic indicators, (b) static indicators and (c) management indicators and crude oil recovery degree was obtained through grey correlation analysis. Among the dynamic indicators, A7 and A5 are the two indicators with the highest correlation, namely oil production rate and number of adjusted wells, with correlations of 0.8197 and 0.8212 respectively. However, except for water content, the correlations of other indicators are also strong. Among the static indicators, B9, B12, B3 and B4 are the indicators with the top 30% correlation, namely reserve abundance, formation type, viscosity and effective thickness, among which viscosity has the highest correlation of 0.791. For management indicators, the two indicators with the highest correlation are C6 and C3, namely dynamic detection plan completion rate and injection qualification rate of injection wells, with correlations of 0.812 and 0.804 respectively.

[0055] like Figure 7 As shown in the figure, the correlation results obtained by GRN-VSN neural network analysis are quite different from those obtained by grey correlation algorithm. Through GRN-VSN neural network analysis, it is found that the dynamic indicators with the greatest correlation are water content and water consumption rate; the static indicators with the greatest correlation are sedimentary phase type, viscosity, reservoir type and medium-depth reservoir; the management indicators with the greatest correlation are water injection well split injection rate and oil-water well comprehensive time rate.

[0056] Since the results of the two algorithms for the correlation analysis of indicators are quite different, the real data of the indicators with the greatest correlation in the two methods are taken for verification. The results are as follows: Figure 8As shown. GRN-VSN judges that the water content has the strongest correlation with the degree of crude oil recovery, while the grey correlation analysis judges that the oil production rate has the strongest correlation with the degree of crude oil recovery. After actual verification, it is found that the relationship judged by GRN-VSN is more accurate, and there is a more obvious functional relationship between the two indicators. After research, it was found that the reason for the error in the grey correlation analysis judgment is that some data are 0 due to oil field shutdown and other reasons, which affects the data accuracy. After eliminating this part of the data, the grey correlation analysis can also analyze the correlation between indicators. In order to reduce the "black box" effect of artificial intelligence methods, the feature correlation analysis is combined with the theoretical basis, and the weighted average method is used to combine the GRN-VSN method with the grey correlation analysis method. The weight of each indicator is determined by the ranking of the indicator correlation analysis. The results are shown as follows Fig. 9 shown.

[0057] According to the average results, the final indicators for predicting the degree of crude oil recovery are selected as follows: dynamic indicators include water cut and annual oil production; static indicators include viscosity, reserve abundance, medium depth of reservoir and effective thickness; management indicators include dynamic detection plan completion rate and water injection well distribution rate.

[0058] In the several embodiments provided by the present invention, it should be understood that the disclosed method can be implemented in other ways. For example, the embodiments described above are merely illustrative.

[0059] The above-mentioned integrated unit implemented in the form of a software functional unit can be stored in a computer-readable storage medium. The above-mentioned software functional unit is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor (Processor) to perform some steps of the above-mentioned method of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store program codes.

[0060] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for screening main controlling factors for offshore oilfield development, characterized in that: The method comprises: S10, obtaining historical data of parameter values ​​of various correlation factors for different types of offshore oilfield development, wherein the correlation factors include static factors, dynamic factors, and management index factors; S20, using a correlation analysis algorithm to calculate the correlation relationship between the correlation factors corresponding to the historical data; S30, constructing a training set according to the historical data and the correlation relationship between the correlation factors, and using the training set to train a neural network model with a preset network structure to obtain a trained neural network model; S40, using the trained neural network model to predict the offshore oil field with the main controlling factor to be predicted, wherein the main controlling factor is obtained according to the correlation relationship between the associated factors corresponding to the offshore oil field to be predicted.

2. The method for screening the main controlling factors of offshore oilfield development according to claim 1, characterized in that: In S10, the correlation analysis algorithm includes a correlation coefficient method.

3. The method for screening the main controlling factors of offshore oilfield development according to claim 2, characterized in that: The correlation analysis algorithm also includes a grey correlation analysis method.

4. The method for screening the main controlling factors of offshore oilfield development according to claim 3, characterized in that: The correlation analysis algorithm also includes a decision tree analysis method.

5. The method for screening the main controlling factors of offshore oilfield development according to claim 1, characterized in that: The neural network model with the preset network structure includes a controllable residual network GRN and a variable selection network VSN.

6. The method for screening the main controlling factors of offshore oilfield development according to claim 5, characterized in that: in, In the network structure of the neural network model, an attention mechanism module is added.

7. The method for screening the main controlling factors of offshore oilfield development according to claim 1, characterized in that: The main controlling factors are based on the top 30% of the correlation factors in each type of correlation factors.

8. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the method for screening main controlling factors for offshore oil field development according to any one of claims 1 to 7 is implemented.

Images