A shale gas drilling cost prediction method and system

Abstract

The application discloses a shale gas drilling cost prediction method and system, including the following steps: obtaining shale gas well drilling cost related data, preprocessing the shale gas well drilling cost related data, and obtaining drilling cost influence factors; constructing an LSTM model, performing time series analysis on the drilling cost influence factors according to the LSTM model, and obtaining influence factor prediction samples; and establishing a drilling cost prediction model by using an adaptive kernel density estimation method, inputting the influence factor prediction samples into the drilling cost prediction model, and obtaining drilling cost prediction data. By combining specific drilling cost data, the main factors influencing the drilling cost are analyzed, the drilling cost is predicted and refined based on the LSTM model and the drilling cost prediction model, the prediction accuracy is improved, the cost law in the production process is obtained, a decision basis is provided for actual production, the production efficiency is improved, and the production cost is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of drilling exploration, and particularly relates to a shale gas drilling cost prediction method and system. BACKGROUND

[0002] Shale gas refers to natural gas extracted from shale layers, mainly in the form of absorption or free state in dark mud shale or high-carbon mud shale, and is a kind of unconventional natural gas. Due to large-scale drilling production, high investment and high risk, a large amount of manpower, material resources and financial resources need to be spent, and there are many unknown factors and uncertain factors in the drilling process, and the drilling production risk is also very large. Drilling operation will lead to increased drilling investment due to the influence of factors such as complex underground conditions, and the current shale gas development investment risk in China is high. Cost is a key factor restricting the development of shale gas industry. Shale gas single well cost benefit estimation is the basis for economic evaluation of shale gas block development, and is also a necessary reference for shale gas investment decision-making. It is of great significance to reduce the risk of shale gas development. Therefore, controlling drilling cost is crucial to reduce drilling engineering investment and improve economic efficiency. SUMMARY

[0003] The technical problem to be solved by the present application is that the drilling cost is large due to the complexity of the drilling process factors, and the drilling cost cannot be evaluated. The purpose is to provide a shale gas drilling cost prediction method and system. By analyzing data characteristics, fitting learning rate model, combining industry information to solve research results, preliminary shale gas well cost analysis research is completed, drilling cost is studied in detail, cost law in the production process is obtained, decision-making basis for actual production is provided, production efficiency is improved, and production cost is reduced.

[0004] The present application is realized by the following technical scheme:

[0005] The present application provides a shale gas drilling cost prediction method in the first aspect, comprising the following steps:

[0006] S1, shale gas well drilling cost related data is obtained, the shale gas well drilling cost related data is preprocessed, and drilling cost influence factors are obtained;

[0007] S2, an LSTM model is constructed, time series analysis of the drilling cost influence factors is performed according to the LSTM model, and influence factor prediction samples are obtained;

[0008] S3, a drilling cost prediction model is established by using an adaptive kernel density estimation method, the influence factor prediction samples are input into the drilling cost prediction model, and drilling cost prediction data is obtained.

[0009] The application analyzes main factors influencing drilling cost by combining specific drilling cost data, constructs an LSTM model to perform time series analysis on drilling cost influencing factors, establishes a drilling cost prediction model by using an adaptive kernel density estimation method to predict drilling cost, improves prediction accuracy, and obtains future drilling cost data from historical drilling cost data to conduct detailed research on drilling cost, obtains cost rules in the production process, provides decision basis for actual production, improves production efficiency, and reduces production cost.

[0010] Further, the preprocessing of the shale gas well drilling cost related data includes: determining the weight of each factor in the shale gas well related data according to the probability distribution, combining to obtain a first factor group set, and the first factor group set includes a plurality of combinations of factors influencing drilling cost;

[0011] The causal relationship between each factor group in the first factor group set and drilling cost is verified, and the factor groups having a causal relationship with drilling cost are saved in a second factor group set;

[0012] For each factor group in the second factor group set, the corresponding drilling cost of the each factor group in a first time period in the future is predicted according to the eigenvalues of the factors in the factor group, and the drilling cost influencing factors are determined.

[0013] Further, the time series analysis of the drilling cost influencing factors according to the LSTM model specifically includes:

[0014] The drilling cost influencing factors are normalized;

[0015] The time series data is passed through a convolution layer, and the obtained result is the extracted features;

[0016] Each dimension of the obtained features is put into the neurons of the LSTM to obtain an LSTM hidden layer output vector;

[0017] The LSTM hidden layer output vector is automatically focused on using an attention mechanism, and the parameters significantly related to the target value are automatically given greater weights;

[0018] The feature vector processed by the attention mechanism is unfolded into a one-dimensional vector to meet the input requirements of the full connection layer;

[0019] The one-dimensional vector obtained by the unfolding operation is input into the full connection layer, and the full connection layer outputs the influencing factor prediction sample.

[0020] Further, the S3 specifically includes:

[0021] The historical drilling cost is obtained as the dependent variable, and the historical drilling cost influencing factors are obtained as the independent variables to construct a vector;

[0022] determining training samples and test samples according to vectors;

[0023] establishing a drilling cost prediction model by using an adaptive kernel density estimation method, and training the drilling cost prediction model;

[0024] model verification is performed on the trained drilling cost prediction model;

[0025] drilling cost prediction is performed by using the verified drilling cost prediction model in combination with an influence factor sequence.

[0026] Further, the determination of the training sample set according to vectors comprises:

[0027] input variables and output variables are determined according to drilling cost historical data, and training samples and test samples are constructed, wherein,

[0028] cumulative well number, drilling cycle and industrial producer purchase price index are selected as input variables, and drilling cost is selected as an output variable.

[0029] Further, the establishment of the drilling cost prediction model by using the adaptive kernel density estimation method and the training of the drilling cost prediction model comprise:

[0030] normalization processing is adopted to normalize sample data of the training samples and the test samples, eliminate the influence of different dimensions and units of each factor, and normalize the input and output data of the samples so that the normalized values are in [-1, 1];

[0031] determining a kernel function, and establishing the drilling cost prediction model by using the adaptive kernel density estimation method.

[0032] Further, the determination of the kernel function specifically comprises: determining a non-sensitive loss function coefficient, a penalty parameter and a Gaussian radial basis kernel function parameter, and the determination step comprises:

[0033] determining the range of the three parameters;

[0034] determining the step length of the change of each parameter;

[0035] forming all combinations of the three parameters;

[0036] selecting the optimal parameters by traversing all combinations.

[0037] Further, the training of the drilling cost prediction model specifically comprises:

[0038] setting initial values of network weights and thresholds, calculating actual network output and hidden unit state, calculating network output error, and ending the training when the output error is less than a threshold value or a specified number of iterations is reached, otherwise performing error back propagation.

[0039] The training error is calculated, the weight value and the threshold value are corrected according to the training error, and the above steps are repeated until the training error is less than a set value.

[0040] The second aspect of the present application provides a shale gas drilling cost prediction system, comprising an interface layer, a business layer and a data layer;

[0041] The data layer is used to obtain shale gas well drilling cost related data, and the business layer is used to preprocess the shale gas well drilling cost related data to obtain drilling cost influencing factors.

[0042] The business layer is used to build an LSTM model, and the drilling cost influencing factors are analyzed in time series according to the LSTM model to obtain influencing factor prediction samples.

[0043] The business layer is used to establish a drilling cost prediction model by using an adaptive kernel density estimation method, and the interface layer is used to input the influencing factor prediction samples into the drilling cost prediction model to obtain drilling cost prediction data.

[0044] Further, the interface layer comprises a model control unit and a data maintenance unit.

[0045] The business layer comprises a data processing unit, a model building unit and a parameter optimization unit.

[0046] The data layer comprises a drilling cost related data unit and a model data unit.

[0047] The drilling cost related data unit is used to obtain shale gas well drilling cost related data, and the data is input into the business layer, and the data processing unit is used to preprocess the shale gas well drilling cost related data to obtain drilling cost influencing factors.

[0048] The model building unit is used to build an LSTM model, and the drilling cost influencing factors are analyzed in time series according to the LSTM model to obtain influencing factor prediction samples.

[0049] The model building unit is also used to establish a drilling cost prediction model by using an adaptive kernel density estimation method, the model control unit and the data maintenance unit are used to input the influencing factor prediction samples into the drilling cost prediction model, and the data processing unit is used to process the influencing factor prediction samples input into the drilling cost prediction model to obtain drilling cost prediction data.

[0050] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0051] 1. The main factors affecting the drilling cost are analyzed in combination with specific drilling cost data, an LSTM model is constructed to analyze the time series of the drilling cost influencing factors, an adaptive kernel density estimation method is used to establish a drilling cost prediction model, the drilling cost is predicted, the prediction accuracy is improved, the future drilling cost data is obtained according to the historical drilling cost data, the drilling cost is studied in detail, the cost law in the production process is obtained, the decision basis for actual production is provided, the production efficiency is improved, and the production cost is reduced.

[0052] 2. The adaptive kernel density estimation method is used to establish a drilling cost prediction model, a kernel function is introduced, a support vector machine model is established by analyzing the drilling cost influencing factors, the drilling cost is predicted, more drilling cost variables are introduced in the probability model, and the prediction accuracy is improved. On the other hand, the conditional probability can be obtained after some specific events have occurred, and the predictability can be better quantified. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the technical scheme of the exemplary embodiments of the present application, the drawings needed in the examples will be briefly introduced as follows, and it should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. Other related drawings can also be obtained by those skilled in the art without creative labor. In the drawings:

[0054] Figure 1 The method flowchart in the example of the present application;

[0055] Figure 2 The total cost and average cost of shale gas wells in the example of the present application;

[0056] Figure 3 The cumulative average cost and drilling cost factor of shale gas wells in the example of the present application;

[0057] Figure 4 The cumulative average cost and drilling cost factor of shale gas wells in the example of the present application;

[0058] Figure 5 The drilling cost factor learning curve from 2015 to 2017 in the example of the present application. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical scheme and advantages of the present application more clear and obvious, the present application will be further described in detail below in combination with examples and drawings, the exemplary embodiments of the present application and their descriptions are only used to explain the present application, and are not regarded as a limitation on the present application.

[0060] Example 1

[0061] As Figure 1 The embodiment provides a shale gas well drilling cost prediction method, and the method comprises the following steps:

[0062] S1, obtaining shale gas well drilling cost related data, preprocessing the shale gas well drilling cost related data, and obtaining drilling cost influence factors;

[0063] S2, constructing an LSTM model, performing time series analysis on the drilling cost influence factors according to the LSTM model, and obtaining influence factor prediction samples;

[0064] S3, establishing a drilling cost prediction model by using an adaptive kernel density estimation method, inputting the influence factor prediction samples into the drilling cost prediction model, and obtaining drilling cost prediction data.

[0065] By combining specific drilling cost data, the main factors affecting the drilling cost are analyzed, the LSTM model is constructed to perform time series analysis on the drilling cost influence factors, the adaptive kernel density estimation method is used to establish the drilling cost prediction model, the drilling cost is predicted, the prediction accuracy is improved, the future drilling cost data is obtained according to the drilling cost historical data, the drilling cost is researched in detail, the cost law in the production process is obtained, the decision basis for actual production is provided, the production efficiency is improved, and the production cost is reduced.

[0066] In some possible embodiments, the preprocessing of the shale gas well drilling cost related data comprises: determining the weight of each factor in the shale gas well related data according to a probability distribution, combining to obtain a first factor group set, and the first factor group set comprises a plurality of combinations of factors affecting the drilling cost;

[0067] The causal relationship between each factor group in the first factor group set and the drilling cost is verified, and the factor groups having the causal relationship with the drilling cost are saved into a second factor group set;

[0068] For each factor group in the second factor group set, the corresponding drilling cost of each factor group in a first time period in the future is predicted according to the characteristic value of each factor in the factor group, and the drilling cost influence factor is determined.

[0069] In some possible embodiments, the shale gas well drilling cost related data specifically comprises: cumulative well number, cumulative total cost, average cost, drilling cycle, industrial producer purchase price index (last year = 100), and industrial producer purchase price index (1990 = 100). The industrial producer purchase price index is used to eliminate the price influence.

[0070] It also includes the 2015-2020 shale gas well settlement table: 2015-2020 228 shale gas wells related to 228 data, including 63 indicators, provided by the Natural Gas Institute;

[0071] Producer Purchasing Price Index: Producer Purchasing Price Index (last year = 100) and Producer Purchasing Price Index (1990 = 100) downloaded from the National Bureau of Statistics network.

[0072] Factors affecting drilling costs include macroeconomic factors, technical factors, policy factors, and term structure factors, including: cumulative well number, cumulative total cost, average cost, drilling cycle, producer purchasing price index (last year = 100), producer purchasing price index (1990 = 100), and producer purchasing price index to eliminate price impact. The impact of cumulative total cost includes: financial market data, bank credit data, financial terminal data, government website information and news media information. As shown in Table 1:

[0073] Table 1 Shale gas well annual data

[0074]

[0075] Among them, the visualization shows the number of wells, total cost (billion) and average cost of shale gas wells per year, as shown in Figure 2 The average cost of shale gas wells per year does not decrease year by year.

[0076] Based on the basic idea of learning curve method (with the increase of production, the operator's proficiency increases, and the product cost will decrease, and the cost decrease is gradually accumulated, not the annual average. Observation data found that the drilling cycle changed greatly, and the average cost per day was introduced.

[0077] Based on the cumulative idea and unit cost, the cumulative well number, cumulative total cost, cumulative average cost, drilling cycle, and drilling cost per day are calculated, as shown in Table 2:

[0078] Table 2 Shale gas well annual data

[0079]

[0080] Among them, the visualization shows the cumulative average cost and drilling cost per day, as shown in Figure 3 The drilling cost per day of shale gas wells partly conforms to the trend of decreasing year by year, but an inflection point appears in 2018.

[0081] The introduction of the producer purchasing price index includes the introduction of the price factor to eliminate the price impact. The shale gas well belongs to the industrial field, so the "producer purchasing price index (1990 = 100)" is introduced. The price index uses the annual average price index, and the drilling cost factor (drilling cost per day / price index * 100). The calculation results are shown in Table 3:

[0082] Table 3 Cumulative average cost and drilling cost factor of shale gas well

[0083]

[0084] The cumulative average cost and drilling cost factor are visualized as shown in Figure 4 .

[0085] As shown in Figure 5 , the drilling cost factor decreases year by year from 2015 to 2017, which is consistent with the learning curve. The fitting results are as follows: Y = a * exp (-x / t), a = 4.927, t = 2.946, the model accuracy is very high, and it can be concluded that:

[0086] The drilling cost factor decreases year by year from 2015 to 2017, and is relatively consistent with the learning rate curve;

[0087] The financial crisis in 2018 led to a fluctuation in the cost reduction trend;

[0088] From 2019 to 2020, it gradually decreased in the new balance;

[0089] The amount of annual data is too small, and the data set needs to be expanded for verification.

[0090] First, build an LSTM model, the construction steps include:

[0091] 1. Record past information: determined by the current input and the output of the previous time, how much information the current memory unit needs to save in the past: f t = σ(W f · [h t-1 , x t ] + b f ) The activation function is often a sigmoid function, which maps numbers between 0 and 1, 1 means all past information is recorded, 0 means all past information is discarded.

[0092] 2. Record current information: determined by the current input and the output of the previous time, which new information is stored in the state unit, including two parts, one is how much current information the current memory unit needs to save, the other is the representation of current information i t = σ(W i · [ht-1 , t ]+b i , The activation function is a sigmoid function, which maps a number between 0 and 1, 1 indicating that all the current information is recorded, and 0 indicating that all the current information is discarded. For the representation of the current information, as a candidate for the current state, it is obtained by mapping through the activation function tanh.

[0093] 3. Current state: the current state is obtained from the current information and the past information This step determines how much of the previous information and the current information is retained by two weights and, if the weight is larger, the model will use more information at a farther time in the future, and if the weight is larger, the model will use more information at a closer time in the future, thereby avoiding the forgetting of information at a farther time point in the ordinary RNN network structure.

[0094] 4. Output result: the current output o is obtained from the output of the last time, the input of the current time and the current state: t =σ(W o ·[h t-1 ,x t ]+b o )h t =o t *tanh(C t )

[0095] According to the LSTM model, the time series analysis of the drilling cost influence factor is carried out, which specifically includes:

[0096] The drilling cost influence factor is normalized. Normalization is beneficial to avoid the magnitude difference between various factors and eliminate the influence of various factors due to different dimensions. The input and output data of the sample are normalized.

[0097] The time series data is passed through the convolution layer, and the obtained result is the extracted feature;

[0098] Each dimension of the obtained feature is put into the neuron of the LSTM to obtain an LSTM hidden layer output vector;

[0099] The attention mechanism is used to automatically focus on the LSTM hidden layer output vector, and the parameters significantly related to the target value are automatically given greater weights;

[0100] The feature vector processed by the attention mechanism is expanded into a one-dimensional vector to meet the input requirements of the fully connected layer;

[0101] The one-dimensional vector obtained by the expansion operation is input into the fully connected layer, and the fully connected layer outputs the influence factor prediction sample.

[0102] In the forward calculation process of the LSTM, the operation of each memory cell is controlled by an input gate, a forget gate and an output gate. Each gate performs an operation on the current input data, and determines whether to pass the operation result down according to the judgment logic of the activation function. In the current state, the memory cell receives an input vector xt and the hidden state ht-1 of the cell at the previous time, the input gate it is multiplied by newly generated information, and the cell state ft·Ct-1 processed by the forget gate ft is superimposed to generate the cell state Ct. After the operation and control of the output gate, the reserved information ot of the current cell state and the output ht of the hidden state are generated. The backpropagation error propagation of the LSTM is similar to that of the BPNN. By calculating the error of each cell, the error is propagated in the time and network level directions, the weight gradient is updated, the connection weight is optimized, and the network accuracy is improved.

[0103] A drilling cost prediction model is established by using an adaptive kernel density estimation method. A sequence of influence factors is input into the drilling cost prediction model to obtain drilling cost prediction data, which specifically includes:

[0104] Historical drilling costs are obtained as dependent variables, and historical drilling cost influence factors are obtained as independent variables to construct a vector.

[0105] Training samples and test samples are determined according to the vector. The training sample set is determined according to the vector, including: determining input variables and output variables according to historical drilling cost data, and constructing training samples and test samples, wherein,

[0106] The cumulative number of wells, drilling cycle and industrial producer purchase price index are selected as input variables, and the drilling cost is selected as the output variable.

[0107] A drilling cost prediction model is established by using an adaptive kernel density estimation method, and the drilling cost prediction model is trained.

[0108] The initial values of the weights and thresholds of the network are set, the actual output of the network and the state of the hidden unit are calculated, the output error of the network is calculated, and the training is ended until the output error is less than the threshold value or the specified iteration step number is reached, otherwise the error is backpropagated.

[0109] The training error is calculated, the weights and thresholds are corrected according to the training error, and the above steps are repeated.

[0110] The trained drilling cost prediction model is verified.

[0111] The verified drilling cost prediction model is used to predict the drilling cost in combination with the sequence of influence factors.

[0112] The adaptive kernel density estimation method is used to establish a drilling cost prediction model, and the drilling cost prediction model is trained, including: using normalization processing to normalize the sample data of the training samples and the test samples, eliminating the influence of different dimensions and units of each factor, and normalizing the input and output data of the samples to make the normalized values in [-1, 1].

[0113] The kernel function is determined, the adaptive kernel density estimation method is used to establish a drilling cost prediction model, the insensitive loss function coefficient, the penalty parameter and the Gaussian radial basis kernel function parameter are determined, and the determining step includes: determining the range of the three parameters; determining the step length of each parameter; forming all combinations of the three parameters; traversing all combinations to select the optimal parameters.

[0114] The adaptive kernel density estimation method is used to establish a drilling cost prediction model, and the drilling cost prediction model is trained, including: using normalization processing to normalize the sample data of the training samples and the test samples, eliminating the influence of different dimensions and units of each factor, and normalizing the input and output data of the samples to make the normalized values in [-1, 1].

[0115]

[0116] Wherein, {x (j ):j=1,2,3,...,M} represents M sample data for constructing a density function, n represents the dimension of the variable space, w represents the window number, k(x) represents the kernel probability density function, and specifically, Where S is the covariance of the sample set {x (j)}, In the formula, The mean value of the M sample data is expressed.

[0117] The adaptive kernel density estimation is fixed, and the window width parameter is corrected to wλ j , and the form is as follows:

[0118]

[0119] Wherein, λ j is a local bandwidth factor, and w is a window width parameter

[0120]

[0121] Wherein, f(x) represents the joint probability density of the sample x, 0≤α<1 is a sensitivity factor, and usually α=0.5, M d is the number of different samples in the M sample data (M d ≤M).

[0122] Embodiment 2

[0123] The embodiment provides a shale gas drilling cost prediction system, which comprises an interface layer, a business layer and a data layer.

[0124] The interface layer comprises a model control unit and a data maintenance unit;

[0125] The business layer comprises a data processing unit, a model building unit and a parameter optimization unit;

[0126] The data layer comprises a drilling cost related data unit and a model data unit;

[0127] The drilling cost related data unit is used to obtain shale gas well drilling cost related data, and input the data to the business layer, and the data processing unit is used to preprocess the shale gas well drilling cost related data, and obtain drilling cost impact factors;

[0128] The model building unit is used to build an LSTM model, and perform time series analysis on the drilling cost impact factors according to the LSTM model, to obtain impact factor prediction samples;

[0129] The model building unit is also used to establish a drilling cost prediction model by using an adaptive kernel density estimation method, the model control unit and the data maintenance unit are used to input the impact factor prediction samples into the drilling cost prediction model, and the data processing unit is used to process the impact factor prediction samples input into the drilling cost prediction model, to obtain drilling cost prediction data.

[0130] The data layer is used to add, modify, delete and query data;

[0131] Data addition realizes the addition function of sample data by the user, the user can manually input the basic information of drilling cost data, or import the data saved in the system into the system.

[0132] Data modification realizes the modification function of sample data by the user. The sample data does not need to be modified, but sometimes the data reported by the drilling team will have individual errors, which need to be modified by the administrator.

[0133] Data query realizes data query, which can find data meeting the conditions according to blocks, well types, technologies, well numbers, etc. The user can select to output to the table.

[0134] Data statistics realize various statistical analysis of drilling cost data, such as total cost, average cost well number, total well number in a certain period.

[0135] The business layer is used for new model creation, model modification, model deletion, model automatic training and drilling cost prediction.

[0136] New model creation creates a new prediction model. After the addition of a new block is completed, a model needs to be established, the basic information such as the required sample data range and sample change number needs to be set, so as to facilitate the user to use the model to predict data.

[0137] Modifying the model, realizing the modification of the model file by the user, such as modifying the change range of sample data, the change number of sample data and the like.

[0138] Model viewing, facilitating the user to view the model and understand the model information.

[0139] Model automatic training, realizing the training of a newly-built model, finding the optimal parameter combination, establishing a support vector machine regression model and automatically saving the optimal parameters and the training result, and simultaneously realizing the automatic modification of the model according to the data change condition set by the user to automatically find the optimal parameters.

[0140] Drilling cost prediction, realizing the drilling cost prediction function of the user, the user manually inputs or batch inputs the drilling cost influence factors, the system can automatically output the predicted value, and if there is an actual cost, the predicted result can be statistically analyzed.

[0141] User permission management, realizing the control of the system use by different users, the administrator can modify the drilling cost data and the model data, and other users can only use the system to predict the cost.

[0142] The above-described specific embodiments further specifically describe the purposes, technical solutions and beneficial effects of the present application, and it should be understood that the above-described embodiments are only specific embodiments of the present application and are not used to limit the protection scope of the present application, and any modification, equivalent replacement, improvement and the like within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A shale gas drilling cost prediction method, characterized in that: The following steps are involved: S1. Obtaining data related to shale gas well drilling costs, preprocessing the data related to shale gas well drilling costs, and obtaining drilling cost influencing factors; S2. Build an LSTM model and perform time series analysis on factors influencing drilling costs based on the LSTM model to obtain prediction samples of the factors. S3. Using the adaptive kernel density estimation method to establish a drilling cost prediction model, input the influencing factor prediction samples into the drilling cost prediction model to obtain drilling cost prediction data; The step of establishing a drilling cost prediction model using the adaptive kernel density estimation method includes: in, Represents M sample data for constructing the density function, represents the dimension of the variable space, Indicates the number of windows, represents the kernel probability density function, specifically, , where S is the sample set The covariance of , where Description The mean of the sample data; Fixed the wide kernel density function of the adaptive kernel density estimation and modified the window width parameter to , which is in the following form: in, is the local bandwidth factor, is the window width parameter, in, represents the joint probability density of sample x, is the sensitivity factor, usually, for The number of different samples in the sample data .

2. The shale gas drilling cost prediction method according to claim 1, characterized in that: Preprocessing the data related to shale gas well drilling costs includes: determining weights of various factors in the shale gas well related data according to a probability distribution, and combining the weights to obtain a first factor group set, wherein the first factor group set includes a combination of multiple groups of factors that affect the drilling costs; Testing the causal relationship between each factor group in the first factor group set and the drilling cost, and saving the factor groups that have a causal relationship with the drilling cost into the second factor group set; For each factor group in the second factor group set, the drilling cost corresponding to each factor group in the future first time period is predicted according to the characteristic value of each factor in the factor group, and the drilling cost influencing factor is determined.

3. The shale gas drilling cost prediction method according to claim 1, characterized in that: The time series analysis of factors affecting drilling costs based on the LSTM model specifically includes: Normalize the factors affecting drilling costs; Pass the time series data through the convolution layer, and the result is the extracted features; Put each dimension of features obtained into the LSTM neuron to obtain the LSTM hidden layer output vector; Use the attention mechanism to automatically focus on the LSTM hidden layer output vector, and automatically give larger weights to parameters that are significantly related to the target value; Expand the feature vector processed by the attention mechanism into a one-dimensional vector to meet the input requirements of the fully connected layer; The one-dimensional vector obtained by the expansion operation is input into the fully connected layer, and the fully connected layer outputs the impact factor prediction sample.

4. The shale gas drilling cost prediction method according to claim 1, characterized in that: The S3 specifically includes: The historical drilling cost is obtained as the dependent variable, and the influencing factors of the historical drilling cost are obtained as the independent variables to construct a vector; Determine the training samples and test samples based on the vector; The drilling cost prediction model is established using the adaptive kernel density estimation method and the drilling cost prediction model is trained; Performing model verification on the trained drilling cost prediction model; The verified drilling cost prediction model is used in combination with the influencing factor sequence to predict drilling costs.

5. The shale gas drilling cost prediction method according to claim 4, characterized in that: Determining the training sample set according to the vector includes: According to the historical data of drilling costs, the input variables and output variables are determined, and the training samples and test samples are constructed. The cumulative number of wells, drilling cycle, and industrial producer purchase price index are selected as input variables and drilling cost as output variable.

6. The shale gas drilling cost prediction method according to claim 4, characterized in that: The use of the adaptive kernel density estimation method to establish a drilling cost prediction model specifically includes: Normalization is used to normalize the sample data of training samples and test samples, eliminating the influence of different dimensions and units of each factor, and normalizing the input and output data of the sample so that the normalized values ​​are between [-1, 1]; The kernel function is determined and the adaptive kernel density estimation method is used to establish a drilling cost prediction model.

7. The shale gas drilling cost prediction method according to claim 6, characterized in that: Determining the kernel function specifically includes: determining an insensitive loss function coefficient, a penalty parameter, and a Gaussian radial basis kernel function parameter, wherein the determining step includes: Determine the range of the three parameters; Determine the step size for each parameter change; Form all combinations of three parameters; Traverse all combinations and select the optimal parameters.

8. The shale gas drilling cost prediction method according to claim 4, characterized in that: The training of the drilling cost prediction model specifically includes: Set the initial values ​​of the network weights and thresholds, calculate the actual output of the network and the state of the hidden units, and calculate the network output error until the output error is less than the threshold or reaches the specified number of iterations. End the training; otherwise, perform error backpropagation. Calculate the training error, modify the weights and thresholds according to the training error, and repeat the above steps until the training error is less than the set value.

9. A shale gas drilling cost prediction system, characterized in that: Including interface layer, business layer and data layer; The data layer is used to obtain data related to shale gas well drilling costs, and the business layer is used to pre-process the data related to shale gas well drilling costs to obtain drilling cost influencing factors; The business layer is used to build an LSTM model, perform time series analysis on factors affecting drilling costs based on the LSTM model, and obtain prediction samples of the factors affecting the drilling costs; The business layer is used to establish a drilling cost prediction model using an adaptive kernel density estimation method, and the interface layer is used to input the influencing factor prediction samples into the drilling cost prediction model to obtain drilling cost prediction data; The step of establishing a drilling cost prediction model using the adaptive kernel density estimation method includes: in, Represents M sample data for constructing the density function, represents the dimension of the variable space, Indicates the number of windows, represents the kernel probability density function, specifically, , where S is the sample set The covariance of , where Description The mean of the sample data; Fixed the wide kernel density function of the adaptive kernel density estimation and modified the window width parameter to , which is in the following form: in, is the local bandwidth factor, is the window width parameter, in, represents the joint probability density of sample x, is the sensitivity factor, usually, for The number of different samples in the sample data .

10. The shale gas drilling cost prediction system according to claim 9, characterized in that: The interface layer includes a model control unit and a data maintenance unit; The business layer includes a data processing unit, a model building unit and a parameter optimization unit; The data layer includes drilling cost related data units and model data units; The drilling cost related data unit is used to obtain shale gas well drilling cost related data and input the data into the business layer. The data processing unit is used to pre-process the shale gas well drilling cost related data and obtain drilling cost influencing factors. The model building unit is used to build an LSTM model, and perform time series analysis on drilling cost influencing factors according to the LSTM model to obtain influencing factor prediction samples; The model building unit is also used to establish a drilling cost prediction model using an adaptive kernel density estimation method. The model control unit and the data maintenance unit are used to input the influencing factor prediction samples into the drilling cost prediction model. The data processing unit is used to perform data processing on the influencing factor prediction samples input into the drilling cost prediction model to obtain drilling cost prediction data.

Images