Horizontal well yield prediction method based on extended long short-term memory neural network

Abstract

The invention discloses a horizontal well yield prediction method based on an extended long short-term memory neural network, and relates to the technical field of yield prediction. Comprising the following steps: obtaining output data of a developed horizontal well and characteristic parameters formed by output influencing data; on the basis of empirical mode decomposition and wavelet transform, decomposing the feature parameters on two levels of time domain and frequency domain to obtain a three-dimensional feature tensor; decomposing the three-dimensional feature tensors based on TCN and STL decomposition, and fusing decomposition results through a dynamic feature selection gate to obtain fused features; a prediction model is established, the prediction model is trained based on the fusion features, and the trained model can perform yield prediction; the main body of the prediction model is a bidirectional LSTM model, a space-time attention mechanism is utilized to provide a weight for fusion features, and an optimized sparrow search algorithm is utilized to select better hyper-parameters for the bidirectional LSTM model. According to the method, the yield of the horizontal well can be effectively predicted, and the prediction result is high in accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of production prediction, and in particular to a method for predicting the production of a horizontal well based on an extended long short-term memory neural network. Background Art

[0002] In the field of oil and gas production, horizontal well production prediction has always been a challenging task because it is affected by many factors, such as reservoir characteristics, wellhead pressure, liquid flow, etc. With the continuous advancement of data acquisition technology and information technology, how to accurately predict the production of horizontal wells through historical data has become one of the keys to oilfield management and decision-making optimization. Traditional production prediction methods often rely on physical models or statistical methods, which are difficult to deal with complex nonlinear relationships and high-dimensional data problems, especially when applied to tight oil production prediction. Due to the large difference between tight oil reservoirs and conventional oil reservoirs, the application of traditional methods is more difficult and less accurate. Therefore, methods based on machine learning and deep learning have gradually been widely used.

[0003] As a deep learning model for processing time series data, the Long Short-Term Memory (LSTM) neural network has been widely used in various prediction tasks. However, its application in oilfield production prediction still faces some challenges, such as the model's training efficiency, overfitting problems, and limited ability to capture complex data features. Summary of the invention

[0004] To solve at least one of the above problems, the present invention proposes a method for predicting horizontal well production based on an extended long short-term memory neural network.

[0005] The technical solution of the present invention is: a method for predicting horizontal well production based on an extended long short-term memory neural network, comprising the following steps: S1. Obtaining the production data of developed horizontal wells and characteristic parameters that affect the production data composition; S2. Based on empirical mode decomposition and wavelet transform, the feature parameters are decomposed in both time domain and frequency domain to obtain a three-dimensional feature tensor; S3, decompose the three-dimensional feature tensor based on TCN and STL decomposition, and fuse the decomposition results through a dynamic feature selection gate to obtain fused features; S4. Establish a prediction model and train the prediction model based on the fusion features. The trained model can be used for production prediction of other horizontal wells in this block. The prediction model is mainly a bidirectional LSTM model, which uses the spatiotemporal attention mechanism to provide weights for the fusion features, and uses the optimized sparrow search algorithm to select better hyperparameters for the bidirectional LSTM model.

[0006] Beneficial effects: The method of the present invention can effectively predict the production of horizontal wells in tight oil reservoirs, and the prediction results are highly accurate, the calculation time is short, and it can provide good guidance for actual production. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 This is a comparison chart of several common models and the model in this paper in 100-day production forecast; Figure 2 for Figure 1 Enlarged view of the red framed part. DETAILED DESCRIPTION

[0008] The specific implementation modes of the present invention will be described clearly and completely below with reference to examples and drawings. Obviously, the described examples are only some embodiments of the present invention, rather than all embodiments.

[0009] Example 1, a method for predicting horizontal well production based on an extended long short-term memory neural network, comprising the following steps: S1. Obtaining the production data of developed horizontal wells and characteristic parameters that affect the production data composition; Specifically, in this embodiment, the production data can be selected as the daily production of a single well, which is mainly to facilitate a more detailed analysis of production changes; the data affecting the production mainly include dynamic liquid level, bottom hole flow pressure, casing pressure, stroke, stroke frequency and water content, which are obtained by the inventor after a large number of experiments. Those skilled in the art can also select different data according to the different blocks where the oil and gas fields are located.

[0010] S2. Based on empirical mode decomposition and wavelet transform, the feature parameters are decomposed in both time domain and frequency domain to obtain a three-dimensional feature tensor; As a powerful time-frequency analysis tool, wavelet transform can decompose signals in both time and frequency dimensions at the same time. It has strong localization characteristics and can extract features of different scales in time series signals. The empirical mode decomposition (EMD) method decomposes complex signals into a series of intrinsic mode functions (IMFs) and residual terms in an adaptive manner, allowing us to better capture the local and global characteristics of the signal. Therefore, in this embodiment, combining EMD and wavelet transform will help enhance the feature extraction capability of oilfield historical data and further improve the effect of the prediction model. The inventor calls it the MD-Wavelet hybrid decomposition method, which combines empirical mode decomposition and wavelet transform, decomposes feature parameters at both the time domain and frequency domain levels, and obtains a richer and more detailed feature expression.

[0011] The method of this step comprises the following sub-steps: S21. Perform empirical mode decomposition on characteristic parameters: , where x (t) means t Characteristic parameters of the moment; IMF k ( t )for t Moment k indivual IMF Components, representing the frequency content of the signal, k =1,2,…,n, where n represents the maximum number of intrinsic mode functions; Res ( t )for t The residual term at the moment represents the long-term trend of the signal; through this decomposition, the periodic components, noise and trend components in the data are extracted separately, which can help the model better understand the nonlinear and multi-scale characteristics in the data.

[0012] S22. Apply wavelet transform to the intrinsic mode function to extract time-frequency features , where W n Indicates n indivual IMF The time-frequency characteristics of the component at time t after wavelet transformation, yes t The improved Meyer wavelet basis function of time, a and b are the scale and location parameters of the wavelet function; S23, concatenate the residual term of S21 and the time-frequency features of S22 to form a three-dimensional feature tensor: , where Represents the concatenation operation, combining each IMF component with its corresponding wavelet transform result to form a three-dimensional feature representation. The three-dimensional feature tensor contains the time-frequency information of the signal at different time steps, different frequency scales, and different feature dimensions, and can fully describe the multi-scale characteristics of the signal; IMF n Indicates the nth IMF Quantity. Res global Represents the residual forming a three-dimensional tensor, which is the residual set of all time steps and is used as subsequent input.

[0013] The final feature tensor X It is a three-dimensional tensor that contains the characteristics of oil field data at different frequencies and time scales, helping the model better understand the complex patterns in oil field production data.

[0014] S3, decompose the three-dimensional feature tensor based on TCN and STL decomposition, and fuse the decomposition results through a dynamic feature selection gate to obtain fused features; Oil and gas field production data is usually affected by multiple factors, which can cause fluctuations on different time scales. For example, seasonal changes (such as climate, production cycle, etc.) can affect oil field production on a longer time scale, while equipment failures and production strategy adjustments in the short term may cause fluctuations in a shorter period. In order to solve these problems, in this embodiment, the three-dimensional feature tensor is processed based on TCN and STL decomposition, and the processing results are fused through a dynamic feature selection gate to finally obtain a fused feature, which expresses both the short-term changes and long-term trends of the features, and is more in line with the actual situation of oil and gas field production.

[0015] This step includes the following sub-steps: S31, through TCN to capture the characteristics of three-dimensional features at different time scales ; Decompose the production data through STL decomposition ; TCN is a convolution operation used to capture long-term and short-term dependencies. Compared with ordinary convolution, TCN can capture dependencies on long time series by expanding the receptive field while maintaining the causality of the data, that is, ensuring that the model only uses past time steps when making predictions to avoid introducing interference from future data. The main advantage of TCN is that it can effectively expand the receptive field of the convolution kernel when processing time series data, thereby obtaining dependency features with a longer time span. In this step, when processing all the data in the three-dimensional feature tensor through TCN, The convolution operation formula is as follows: , where express t Features captured by TCN at all times; Indicates l Tier k convolution kernel weights; Is input at time point tk The value of indicates that the input is from t Backward Delay k Step input; d is the dynamic expansion coefficient; b l It is l The bias term of the layer; σ is the activation function ReLU. At the same time, in this step, the size of the convolution kernel can be adjusted by the dynamic expansion coefficient to expand its receptive field to capture features at different time scales.

[0016] STL decomposition is a time series decomposition method based on local weighted regression, which can decompose time series data into trend components ( T t ), seasonal component ( St ) and the residual component ( R t The advantage of STL decomposition is that it can adaptively decompose data when nonlinear and seasonal changes are more significant, thereby extracting the main patterns of the data. In this step, after using STL decomposition to decompose the oil field production data, The expression is as follows: = T t + S t + R t , where express Production data decomposed by STL at all times; T t It is the trend component, which reflects the long-term trend of data; S t is the seasonal component, capturing cyclical fluctuations; R t represents the residual component after STL decomposition, which represents the remaining part not explained by trend and seasonality, and represents noise or random fluctuations; t Indicates a point in time.

[0017] S32, through the dynamic feature selection gate, the TCN output features and Y t Fusion: , where H f Indicates fusion features; g Represents the dynamic gating fusion function.

[0018] Since TCN and STL decomposition extract features from different perspectives. TCN focuses on local short-term changes, while STL decomposition focuses on long-term trends and seasonal fluctuations. Therefore, how to effectively fuse these two types of features is the key to this module. After a large number of experiments, the inventors found that the introduction of a dynamic feature selection gate can well fuse the two. The fused features express both the short-term changes and long-term trends of the features, which is more in line with the actual situation of oil and gas field production. The two types of features are adaptively weighted and fused through a gating mechanism.

[0019] The calculation formula of g is as follows: , where represents the weight parameter in the dynamic control model; Represents the bias term of the dynamic feature selection gate; [ ] combines these two types of features into a new feature matrix.

[0020] S4. Establish a prediction model and train the prediction model based on the fusion features. The trained model can be used for production prediction of other horizontal wells in this block. The prediction model is mainly a bidirectional LSTM model (BiLSTM). The spatiotemporal attention mechanism is used to provide weights for the fusion features, and the optimized sparrow search algorithm is used to select better hyperparameters for the bidirectional LSTM model.

[0021] BiLSTM is an extension of traditional LSTM, which can process both the forward and backward dependencies of sequence data at the same time. In this way, it can not only capture past information, but also utilize future contextual information, thereby improving the model's ability to understand time series data.

[0022] The bidirectional LSTM network performs t Input x t Perform forward and backward propagation to generate t Forward output at time and t Backward output at time The formula is as follows:

[0023] in, and Respectively represent the output of forward and backward propagation at time t; LSTM express LSTM algorithm; and They represent the output of forward propagation at time t-1 and the output of backward propagation at time t+1 respectively.

[0024] However, in this embodiment, the input features include temporal information and spatial information. Therefore, in order to reflect its role, the inventor introduced the spatiotemporal attention mechanism and set a spatiotemporal gate function to fuse the output of the bidirectional LSTM with the weighted information of the spatiotemporal attention mechanism to finally form a spatiotemporal feature representation.

[0025] Among them, the spatiotemporal attention mechanism can simultaneously focus on temporal and spatial information through the joint attention mechanism. In the weighting process of spatiotemporal features, we combine temporal information and spatial position encoding to query Q and key K The model automatically allocates attention to different time steps and spatial locations based on the importance of the data.

[0026] Specifically, the spatiotemporal attention mechanism first calculates the spatiotemporal joint attention weights a i,j, which represents the correlation between spatial position and time step. The formula for calculating the spatiotemporal joint attention weight is:

[0027] in, a i,j Indicates that at time step i and spatial location j The weight on Q and K They are the query and key matrices respectively. d is the dimension of query and key. The spatiotemporal position encoding matrix P The spatial location information will be incorporated into the calculation to ensure the influence of spatial features on spatiotemporal weighting; softmax is the expression of the softmax function.

[0028] The process of fusing the output of the bidirectional LSTM with the weighted information of the spatiotemporal attention mechanism is:

[0029] in, Γ is a fusion function that calculates the weights based on the spatiotemporal attention mechanism a i,j The output of the bidirectional LSTM is adjusted so that the spatiotemporal features of the final output can better reflect the relationship between temporal and spatial features.

[0030] At the same time, in the BiLSTM model of this embodiment, the selection of hyperparameters has a crucial impact on the training process and final prediction performance of the model, such as network architecture parameters (such as the number of layers, the number of units in each layer), training parameters (such as learning rate, batch size), and parameters in the optimization process (such as momentum, weight decay, etc.).

[0031] Manually selecting these hyperparameters is often time-consuming and laborious, and may fall into local optimal solutions. To overcome these problems, this module introduces an improvement to the sparrow search algorithm (SSA), using a dynamic search space and quantized search mechanism to optimize and extend the hyperparameters in the BiLSTM network. Through this optimization strategy, the model training process can be more efficient, and the optimization process can be more comprehensive and flexible, thereby improving the accuracy and stability of the oilfield production prediction task.

[0032] Dynamic search space: In traditional optimization algorithms, a fixed search range is usually used for optimization. In our solution, a dynamic search space is used to dynamically adjust the search range for each hyperparameter. The core of this method is to use the mean value of a hyperparameter in the historical training process. μ k and standard deviation σk To determine the search range of the hyperparameter. The formula is:

[0033] in ， For the s Hyperparameters are iterated t+ 1 when the candidate value reaches the hyperparameter, Indicates s The mean value of the hyperparameters in historical iterations, Indicates s The standard deviation of the hyperparameters in the historical iterations, α Represents the dynamic adjustment coefficient.

[0034] This dynamic adjustment of the search space enables the optimization process to be carried out in a more targeted range, enhancing the global search capability and avoiding falling into the local optimal solution.

[0035] Quantum search mechanism: In order to balance exploration and development in the search process, we introduced a quantum search mechanism. The quantum search mechanism guides the search process through quantum states, so that the optimization process can fully explore the solution space and quickly develop to the potential optimal solution area. It is a mixed state, which is composed of the exploration state (explore) and the development state (exploit) in proportion:

[0036] in, α and β are the coefficients of the exploration state and the development state respectively; Indicates the state it is in. By controlling these two coefficients, the quantized search mechanism can dynamically adjust the balance between exploration and development during the training process to avoid premature convergence.

[0037] Fitness function design: In the optimization sparrow search algorithm, the design of the fitness function determines the quality of the search process. In order to improve the optimization efficiency, we designed a composite fitness function, which not only considers the model error (such as mean square error MSE), but also introduces gradient stability and feature entropy constraints. The gradient stability constraint can prevent the gradient explosion or gradient vanishing problem, while the feature entropy constraint ensures the diversity of features and avoids overfitting.

[0038] The design of the fitness function is as follows:

[0039] in, MSE is the mean square error, is the norm of the gradient, Entropyrepresents the feature entropy, ω 1. ω 2 and ω 3 respectively represent MSE , and Entropy The weight of .

[0040] After obtaining the above model, the prediction output module converts the model's feature output into a specific oilfield production forecast value through the regression layer. At the same time, in order to improve the interpretability of the model, we added an interpretability enhancement layer to the prediction process. By quantifying the contribution of features, decision makers can clearly understand the contribution of each feature to the final prediction result. In addition, a post-processing mechanism is introduced to smooth the prediction fluctuations and improve the stability and consistency of the model output.

[0041] Regression output layer: The regression layer is responsible for the feature tensor obtained from the spatiotemporal fusion module H Mapping to predicted values ​​of oil field production y pred Through the fully connected layer, the model maps the spatiotemporal features into specific prediction values. The output prediction value represents the production prediction result of the oil field, and the regression coefficient is optimized through the training process.

[0042] The specific formula is:

[0043] in, W reg is the weight matrix of the regression layer, b reg is the bias term, H It is a space-time feature.

[0044] Interpretability enhancement layer: To improve the interpretability of the model, a feature contribution quantification method is designed. This layer helps users understand the reasons for the model prediction by quantifying the contribution of each input feature to the prediction result. Specifically, we calculate the partial derivative of each feature to evaluate its influence on the prediction result.

[0045] The feature contribution measurement formula is:

[0046] In the formula, I ( x i )express x i Contribution to the prediction results, measuring the influence of the feature on the prediction results; x i Indicates the first i Features y Represents the prediction results of the model; Represents the prediction results x i Gradient of the fused features.

[0047] This approach is able to generate importance scores for each input feature and provide decision makers with clear explanations behind the predictions.

[0048] Post-processing mechanism: In order to correct possible abnormal fluctuations and prediction errors, the post-processing mechanism smoothes the predicted values ​​through methods such as sliding average to ensure that the final prediction results are more stable and reasonable. Sliding average is a common time series smoothing technology that can effectively remove small fluctuations and make the prediction results more stable.

[0049] The sliding average formula is:

[0050] in, y smooth ( t ) is the smoothed forecast value at time t, N is the window size, y pred ( t )express t The original predicted value at time .

[0051] In order to further illustrate the advantages of the method of the embodiment of the present invention, a specific test example is given below.

[0052] Take the relevant data of a certain oil field for the whole year of 2023. Due to the large amount of data, only part of the data is given below, as shown in Table 1.

[0053] Table 1 Part of the original data table

[0054] The method of the embodiment of the present invention is compared with existing methods such as ARIMA, LSTM and TCN. The final result is as follows: Figure 1 , Figure 2 And as shown in Table 2.

[0055] Table 2 Performance comparison of different models

[0056] from Figure 1 , Figure 2 As can be seen from Table 2, compared with the existing model, the model of the present invention has higher accuracy and requires less training time.

[0057] In order to illustrate the rationality of each step in the embodiment of the present invention, an ablation experimental model comparison is carried out, that is, a certain module or step in the embodiment of the present invention is removed, or a conventional module is used to replace the module of the embodiment of the present invention, and a new method is formed, and the prediction accuracy of the new method is compared with the embodiment of the present invention. The final results are shown in Table 3.

[0058] Table 3 Comparison of ablation experimental models

[0059] It can be seen from Table 3 that any module or step of the method in the embodiment of the present invention is necessary, and the method in the embodiment of the present invention has a better effect only when all modules exist.

[0060] The above description is only a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as a preferred embodiment as above, it is not intended to limit the present invention. Any technician familiar with the profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the scope of the technical solution of the present invention.

Claims

1. A method for predicting horizontal well production based on an extended long short-term memory neural network, characterized in that: The following steps are included: S1. Obtaining the production data of developed horizontal wells and characteristic parameters that affect the production data composition; S2. Based on empirical mode decomposition and wavelet transform, the feature parameters are decomposed in both time domain and frequency domain to obtain a three-dimensional feature tensor; S3, decompose the three-dimensional feature tensor based on TCN, decompose the production data based on STL decomposition, and fuse the decomposition results through a dynamic feature selection gate to obtain fused features; S4. Establish a prediction model and train the prediction model based on the fusion features. The trained model can be used for production prediction of other horizontal wells in this block. The prediction model is mainly a bidirectional LSTM model, which uses the spatiotemporal attention mechanism to provide weights for the fusion features, and uses the optimized sparrow search algorithm to select better hyperparameters for the bidirectional LSTM model.

2. The method according to claim 1, characterized in that The characteristic parameters include dynamic liquid level, bottom hole flow pressure, casing pressure, stroke, stroke frequency and water content.

3. The method according to claim 1, characterized in that: S2 includes the following sub-steps: S21. Perform empirical mode decomposition on characteristic parameters: , where x (t) means t Characteristic parameters of the moment; IMF k ( t )for t Moment k indivual IMF Components, representing the frequency content of the signal, k =1,2,…,n, where n represents the maximum number of intrinsic mode functions; Res ( t )for t The residual term at time t represents the long-term trend in the signal; S22. Apply wavelet transform to the intrinsic mode function to extract time-frequency features: , where W n Indicates n indivual IMF The time-frequency characteristics of the component at time t after wavelet transformation, yes t The improved Meyer wavelet basis function of time, a and b are the scale and location parameters of the wavelet function; S23, concatenate the residual term of S21 and the time-frequency features of S22 to form a three-dimensional feature tensor: , where represents the concatenation operation, combining each IMF component with its corresponding wavelet transform result to form a three-dimensional feature representation; IMF n Indicates the nth IMF Quantity; Res global Represents the residual forming a three-dimensional tensor, which is the set of residuals for all time steps.

4. The method according to claim 1, characterized in that S3 includes the following sub-steps: S31, through TCN to capture the characteristics of three-dimensional features at different time scales ; Decompose the production data through STL decomposition to obtain ; S32, through the dynamic feature selection gate, and Fusion: , where H f Indicates fusion features; g represents the dynamic gating fusion function; express t Features captured by TCN at all times; express t Production data decomposed by STL at all times.

5. The method according to claim 4, characterized in that In S31, The calculation formula is as follows: , where Indicates l Tier k convolution kernel weights; Is input at time point tk The value of indicates that the input is from t Backward Delay k Step input; d is the dynamic expansion coefficient; b l It is l The bias term of the layer; σ is the activation function ReLU; The calculation formula is as follows: , where T t It is the trend component, which reflects the long-term trend of data; S t is the seasonal component, capturing cyclical fluctuations; R t represents the residual component after STL decomposition, which represents the remaining part not explained by trend and seasonality, and represents noise or random fluctuations; t Indicates time; The calculation formula is as follows: , where θ g represents the weight parameter in the dynamic control model; b g Represents the bias term of the dynamic feature selection gate; [ ] combines these two types of features into a new feature matrix.

6. The method according to claim 1, characterized in that In S4, a quantized search mechanism is introduced into the optimized sparrow search algorithm: , where represents the quantum state, explore and exploit Represent the exploration state and the development state respectively; α and β are the coefficients of the exploration state and the development state respectively; Indicates the status.

7. The method according to claim 1, characterized in that In S4, when the spatiotemporal attention mechanism is used to provide weights for fusion features, the weight calculation formula is: , where a i,j Indicates that at time step i and spatial location j The weight on Q is the query matrix, express K The transposed matrix of the key, d are the dimensions of the query and key, P is the spatiotemporal position encoding matrix; softmax is the expression of the softmax function.

8. The method according to claim 1, characterized in that: In S4, an interpretability enhancement layer for evaluating the influence of features on the prediction results and a post-processing mechanism for increasing the stability of the prediction values ​​are also added to the prediction model; In the interpretability enhancement layer, the feature contribution measurement formula is used to calculate the influence of features on the prediction results: , where I ( x i )express x i Contribution to the prediction results, measuring the influence of the feature on the prediction results; x i Indicates the first i Features y Represents the prediction results of the model; Represents the prediction results x i Gradient of fused features; In the post-processing mechanism, the smoothness of the prediction results is increased by the sliding average formula: , where is the smoothed forecast value at time t; N is the window size; Indicates time t The original predicted value at time .

Images