A method for predicting the increased oil production of measures based on the idea of clustering and matching

Abstract

The present invention discloses a method for predicting the increased oil production of measures based on the idea of clustering and matching, including respectively constructing data sample libraries for pump replacement measures, perforation supplement measures and other measures; respectively determining the characteristic factors of pump replacement measures, perforation supplement measures and other measures based on the idea of clustering and matching; setting static data, dynamic data and the change amount of the characteristic factors of measures as inputs and the annual cumulative oil production as output, and respectively constructing a prediction model for the increased oil production of pump replacement measures, a prediction model for the increased oil production of perforation supplement measures, and a prediction model for the increased oil production of other measures; training the prediction model for the increased oil production of pump replacement measures, the prediction model for the increased oil production of perforation supplement measures, and the prediction model for the increased oil production of other measures to obtain the trained prediction model for the increased oil production of pump replacement measures, the trained prediction model for the increased oil production of perforation supplement measures, and the trained prediction model for the increased oil production of other measures. Compared with the common forward thinking mode based on the statistical feature selection method, this method can essentially reflect the changes brought by different measures.

Description

Technical Field

[0001] The present invention specifically relates to a method for predicting the increased oil production of measures based on the idea of clustering and matching, and belongs to the technical field of oilfield data processing. Background Art

[0002] Deep coalbed methane, also known as deep-seated coalbed methane, refers to hydrocarbon gases stored in coal seams at depths greater than 1500 meters, while coalbed methane at depths shallower than 1500 meters is called shallow coalbed methane (or medium-shallow coalbed methane). According to the fourth resource assessment, the coalbed methane resource volume in China at depths of 1500 - 2000 meters is 1.193 trillion cubic meters, and the coalbed methane resource volume at depths greater than 2000 meters has not been systematically evaluated; according to the evaluation by PetroChina, the deep coalbed methane resource volume in the Ordos Basin is 2.3 trillion cubic meters, which is equivalent to the national shallow coalbed methane resource volume alone. However, due to various reasons, especially technical level limitations, the development of deep coalbed methane has always been a no-go zone for coalbed methane development both at home and abroad.

[0003] Inspired by the massive fracturing of shale gas, the development of deep coalbed methane needs to form a fracture network like shale gas development to avoid following the old path of shallow coalbed methane. For a long time, in traditional concepts, coal seams cannot form a fracture network like shale because shale is a typical brittle rock, while coal seams are typical plastic strata. According to the discrimination criteria for forming a fracture network by shale fracturing, deep coal seams do not have the objective conditions to form a fracture network. However, this is not the case. Some scholars have found that due to the essential differences between deep coal seams and shale, under conditions such as large thickness (average above 6m) and high gas content (average 25.2 m³ / t) in deep coal seams, it is possible for deep coal seams to be fractured to form a fracture network. Just as natural fractures are the most important feature of shale, the randomly distributed and approximately orthogonally developed microstructures inside the coal body are the most important feature of deep coal seams. Microstructures are the characteristics that distinguish coal from other reservoirs and are also the main seepage channels for gas and water, having a decisive impact on the exploitation of coalbed methane; most importantly, microstructures are the objective basis for the formation and extension of the fracture network. Therefore, it is necessary to finely describe the morphology of microstructures. Fracture research is a hot topic at home and abroad and is very detailed. In fracture research, fractures are characterized by 5 parameters: fracture aperture, aperture roughness, spatial correlation length of the aperture field, fracture trend surface fluctuation coefficient, and spatial correlation length of the trend surface. Obviously, microstructures can also be characterized by similar parameters, but the microstructures of deep coal seams are unique, and having too many parameters will be difficult to guide field practice. Therefore, carrying out research on the quantitative characterization and prediction of microstructures can provide effective guidance for understanding and developing deep coalbed methane. Summary of the Invention

[0004] The present invention mainly overcomes the disadvantages existing in the prior art, and provides a method for predicting the increased oil production of measures based on the idea of clustering and matching.

[0005] To solve the above technical problems, the technical solution provided by the present invention is: a method for predicting the increased oil production of measures based on the idea of clustering and matching, including the following steps:

[0006] Step 1: Construct data sample libraries for pump replacement measures, perforation supplement measures, and other measures respectively;

[0007] Step 2: Determine the characteristic factors of pump replacement measures, perforation supplement measures, and other measures respectively based on the idea of clustering and matching;

[0008] Step 3: Determine the data sets of pump replacement measures, perforation supplement measures, and other measures respectively according to the characteristic factors of pump replacement measures, perforation supplement measures, and other measures, and divide the data sets into training sets and test sets;

[0009] Step 4: Set static data, dynamic data, and the change amount of the characteristic factors of measures as inputs, and the annual cumulative oil production as the output, and construct a predicted increased oil production model for pump replacement measures, a predicted increased oil production model for perforation supplement measures, and a predicted increased oil production model for other measures respectively;

[0010] Step 5: Train the predicted increased oil production model for pump replacement measures, the predicted increased oil production model for perforation supplement measures, and the predicted increased oil production model for other measures respectively through the training sets of pump replacement measures, perforation supplement measures, and other measures, and obtain the trained predicted increased oil production model for pump replacement measures, the trained predicted increased oil production model for perforation supplement measures, and the trained predicted increased oil production model for other measures;

[0011] Step 6: Predict the increased oil production of measures respectively through the trained predicted increased oil production model for pump replacement measures, the trained predicted increased oil production model for perforation supplement measures, and the trained predicted increased oil production model for other measures.

[0012] A further technical solution is that the data sample library includes static data, dynamic data, the change amount of measure factors, and the increased oil production.

[0013] A further technical solution is that the static data includes effective thickness, converted middle depth, geological reserves, original formation temperature, and original formation pressure.

[0014] A further technical solution is that the dynamic data includes monthly oil production, monthly water production, cumulative oil production, cumulative water production, and production days.

[0015] A further technical solution is that the change amount of measure factors includes the change amount of the top depth of the production interval, the change amount of the bottom depth of the production interval, the change amount of the effective thickness, the change amount of the pump depth, the change amount of the pump efficiency, the change amount of the displacement, the change amount of the stroke, the change amount of the pumping frequency, the change amount of the casing pressure, and the change amount of the flowing fluid level.

[0016] A further technical solution is that the data in the data sample library are all subjected to outlier screening and missing value filling processing.

[0017] A further technical solution is that in step 2, the variable of the characteristic factor of the measure is clustered, and the variable of the characteristic factor of the measure is randomly selected as the clustering index. The Gaussian mixture model GMM clustering and K-means clustering methods are used to obtain an unsupervised classification result. If there is a matching relationship between the result label after clustering and the measure type label, it means that the selected factor pair can characterize the influence of the measure on the prediction index, and thus it is used as the characteristic factor of the measure.

[0018] A further technical solution is that the models in step 4 all adopt neural network models; the neural network models include BP neural networks and LSTM neural networks.

[0019] Advantages of the present invention: Based on the idea of reverse pattern matching, the present invention quantifies the influence of different leading measures into the "characteristic patterns" of the factor changes caused to the oil wells by them, and combines the feature selection method to establish a prediction model for the oil production increase of the measures. Compared with the conventional forward thinking mode based on the statistical feature selection method, this method can essentially reflect the changes brought by different measures. Description of the Drawings

[0020] Figure 1 It is a schematic diagram corresponding to clustering. Detailed Embodiments

[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0022] A method for predicting the oil production increase of measures based on the idea of clustering and matching of the present invention includes the following steps:

[0023] Step 1, respectively construct data sample libraries for pump replacement measures, perforation supplement measures, and other measures;

[0024] First, the data is screened for outliers and filled for missing values, and then the "measure category" in the data table is uniformly labeled: measures related to pump replacement are uniformly labeled as "pump replacement", measures related to layer perforation supplement are uniformly labeled as "perforation supplement", and other measures are uniformly labeled as "other". The specific measure details are as follows;

[0025]

[0026] The data sample library includes static data, dynamic data, measure factor change amounts, and increased oil production. The static data includes effective thickness, converted mid-depth, geological reserves, original formation temperature, and original formation pressure. The dynamic data includes monthly oil production, monthly water production, cumulative oil production, cumulative water production, and production days. The measure factor change amounts include the change amount of the top depth of the production interval, the change amount of the bottom depth of the production interval, the change amount of the effective thickness, the change amount of the pump depth, the change amount of the pump efficiency, the change amount of the displacement, the change amount of the stroke, the change amount of the pumping frequency, the change amount of the casing pressure, and the change amount of the flowing fluid level;

[0027] Step 2: Based on the idea of clustering and matching, determine the characteristic factors of the pump replacement measure, the perforation supplement measure, and other measures respectively;

[0028] Cluster the changes in indicators after the measures, and analyze whether the change amounts brought by the measures are meaningful: Randomly select the factor change amounts brought by the measures as the clustering indicators, and use the Gaussian mixture model GMM clustering and K-means clustering methods to obtain an unsupervised classification result. If there is a matching relationship between the result labels after clustering and the measure type labels, it means that the selected factor pairs can characterize the influence of the measures on the prediction indicators and can be used as model inputs, and then construct a sample library (including features in the form of dynamic time series).

[0029] The calculation method of the matching degree between the result labels after clustering and the measure type labels is described as follows. Let the result labels after clustering be 0, 1, and 2. Corresponding the three labels to the measure labels respectively. For example, "pump replacement": 0, "perforation supplement": 1, "other": 2. Then the calculation formula for the matching degree is:

[0030]

[0031] Step 3: Determine the data sets of the pump replacement measure, the perforation supplement measure, and other measures respectively according to their characteristic factors, and divide the data sets into training sets and test sets;

[0032] Step 4: Construct an increased oil production prediction model for the pump replacement measure, an increased oil production prediction model for the perforation supplement measure, and an increased oil production prediction model for other measures respectively;

[0033] Further use a neural network model for prediction. The input and output factors of the model are: static data, dynamic data, and the change amounts of the characteristic factors of the measures are set as inputs, and the annual cumulative oil production is set as the output.

[0034] Among them, the static data is known regardless of whether it is a production well or a non-production well and is input into the BP feedforward neural network; the dynamic data intercepts the data for a certain period before the measures are taken and is input into the LSTM neural network; the change data is also input into the BP network as a quantitative value. The change data is equivalent to implementing an impulse response on the time series before the measures input into the LSTM network, reflecting the changes brought by the measures on the factors. Here, the annual cumulative oil production output is the cumulative output corresponding to the time node when the measures are completed.

[0035] The learning process of the BP neural network algorithm consists of two processes: the forward propagation of the signal and the backpropagation of the error. Its training objective is to minimize the error E, and the error E is the square of the norm distance between the actual output and the expected output:

[0036]

[0037] Among them, d k and o k correspond to the true value and the predicted value of the k-th sample respectively, and m is the total number of samples; in the forward propagation process, the variable transfer formula from the hidden layer to the output layer is as follows:

[0038]

[0039] Among them, ω jk is the value from the j-th neuron in the hidden layer to the k-th neuron in the output layer, y j is the output value of the j-th neuron in the hidden layer, b k represents the threshold of the k-th neuron in the hidden layer, d k is the output value of the k-th neuron in the hidden layer, f is the activation function, and the Sigmoid function is selected.

[0040] In the backpropagation process, the steepest descent method is selected as the training function, and the weight parameter correction formula from the output layer to the hidden layer is as follows:

[0041]

[0042]

[0043] Among them, ω′ jk and b′ k represent the updated weights and thresholds respectively, η is the learning rate, and e k is the error between the predicted value and the true value.

[0044] The basic structure of the LSTM neural network includes four components: the input gate, the output gate, the forget gate, and the memory unit. The gates and unit states of the LSTM unit are updated according to the following formulas:

[0045] it = σ(W i x t + R i h t-1 + W ci c t-1 + b i )(9)

[0046] f t = σ(W f x t + R f h t-1 + W cf c t-1 + b f )(10)

[0047] z t = g(W c x t + R c h t-1 + b c )(11)

[0048] c t = i t · z t + f t · c t-1 (12)

[0049] o t = σ(W o x t + R o h t-1 + W co c t-1 + b o )(13)

[0050]

[0051] where x t and h t ∈ R d are the input and output of the network at time t. The matrices W and R with different subscripts are the weight matrices of the input part and the recurrent part respectively. b i , b o , b f and b c are the bias vectors with respect to the three gates and the memory unit. σ, g, and are activation functions, and · denotes dot product. i t is called the input gate, f t is called the forget gate, z t represents the update gate, c t represents the state of the true output at the current time t, ot is called the output gate, h t is the output of the neural network;

[0052] Step 5: Respectively train the oil production increase prediction models for pump replacement measures, hole filling measures, and other measures through the training sets of pump replacement measures, hole filling measures, and other measures to obtain the trained oil production increase prediction models for pump replacement measures, hole filling measures, and other measures. The training process R 2 The indicators are as follows:

[0053]

[0054] Step 6: Respectively predict the oil production increase of the measures through the trained oil production increase prediction models for pump replacement measures, hole filling measures, and other measures. The true values and predicted values of the partial annual cumulative oil production increase prediction results obtained are as follows:

[0055]

[0056] As mentioned above, it is not a restriction on the present invention in any form. Although the present invention has been disclosed through the above embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications within the scope of the technical solution of the present invention by using the disclosed technical content to obtain equivalent embodiments with equivalent changes. However, as long as it does not depart from the technical content of the technical solution of the present invention, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A method for predicting the increased oil production of measures based on the idea of clustering and matching, characterized in that, it includes the following steps: Step 1: Construct data sample libraries for pump replacement measures, perforation supplement measures, and other measures respectively; The other measures include major repairs, major repair pipe unjamming, sand control, water plugging by packer, perforation extension, and bridge plug water shutoff; Step 2: Determine the characteristic factors of pump replacement measures, perforation supplement measures, and other measures respectively based on the idea of clustering and matching; Cluster the variables of the characteristic factors of the measures. Randomly select the variables of the characteristic factors of the measures as the clustering indicators, and use the Gaussian mixture model GMM clustering and K-means clustering methods to obtain an unsupervised classification result; if there is a matching relationship between the result label after clustering and the measure type label, it means that the selected factors have an impact on characterizing the measure prediction index, and they are used as the characteristic factors of the measures; Step 3: Determine the data sets of pump replacement measures, perforation supplement measures, and other measures respectively according to the characteristic factors of pump replacement measures, perforation supplement measures, and other measures, and divide the data sets into training sets and test sets; Step 4: Set the static data, dynamic data, and the change amount of the characteristic factors of the measures as inputs, and the annual cumulative oil production as the output, and construct a pump replacement measure increased oil production prediction model, a perforation supplement measure increased oil production prediction model, and an other measure increased oil production prediction model respectively; Step 5: Train the pump replacement measure increased oil production prediction model, the perforation supplement measure increased oil production prediction model, and the other measure increased oil production prediction model respectively through the training sets of pump replacement measures, perforation supplement measures, and other measures to obtain the trained pump replacement measure increased oil production prediction model, the perforation supplement measure increased oil production prediction model, and the other measure increased oil production prediction model; Step 6: Predict the increased oil production of the measures respectively through the trained pump replacement measure increased oil production prediction model, the perforation supplement measure increased oil production prediction model, and the other measure increased oil production prediction model.

2. The method for predicting the increased oil production of measures based on the idea of clustering and matching according to claim 1, characterized in that, the data sample library includes static data, dynamic data, the change amount of measure factors, and increased oil production.

3. The method for predicting the increased oil production of measures based on the idea of clustering and matching according to claim 2, characterized in that, the static data includes effective thickness, converted middle depth, geological reserves, original formation temperature, and original formation pressure.

4. The method for predicting the increased oil production of measures based on the idea of clustering and matching according to claim 2, characterized in that, the dynamic data includes monthly oil production, monthly water production, cumulative oil production, cumulative water production, and production days.

5. The method for predicting the increased oil production of measures based on the idea of clustering and matching according to claim 2, characterized in that, the change amount of measure factors includes the change amount of the top depth of the production interval, the change amount of the bottom depth of the production interval, the change amount of the effective thickness, the change amount of the pump depth, the change amount of the pump efficiency, the change amount of the displacement, the change amount of the stroke, the change amount of the pumping frequency, the change amount of the casing pressure, and the change amount of the flowing fluid level.

6. The method for predicting the increased oil production of measures based on the idea of clustering and matching according to claim 1, characterized in that, the data in the data sample library are all subjected to outlier screening and missing value filling processing.

7. A method for predicting the increased oil production of measures based on the idea of clustering and matching according to claim 1, characterized in that, the models in step 4 all adopt neural network models; the neural network models include BP neural networks and LSTM neural networks.

Images