A method and device for predicting liquid accumulation in tight gas wells

Abstract

The present invention discloses a method and device for predicting liquid accumulation in tight gas wells. The method includes: obtaining production data of the tight gas well to be predicted, preprocessing the production data of the tight gas well to be predicted, and performing normalization processing on the preprocessed production data of the tight gas well; dividing the normalized production data of the tight gas well into multiple subsequences according to time, inputting them into the trained liquid accumulation prediction model, and obtaining a liquid accumulation prediction result, where the liquid accumulation prediction result includes a predicted value or confidence interval of the liquid accumulation. This method combines the LSTM and Informer methods for predicting liquid accumulation in tight gas wells, can quickly and effectively predict the liquid accumulation situation in tight gas wells, combines the advantages of the LSTM model and the Informer model, and realizes high-precision short-term and long-term time series prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of tight gas wells, and in particular to a method and device for predicting liquid accumulation in tight gas wells. Background Technique

[0002] Liquid accumulation in tight gas wells is a phenomenon that seriously affects the normal production of oil and gas wells, and is related to the production efficiency and overall safety of oil and gas wells. Therefore, it is necessary to predict liquid accumulation in tight gas wells.

[0003] Traditional time series prediction methods such as ARIMA, SVR, etc. often have problems such as insufficient prediction accuracy and inability to effectively capture long-term dependence relationships when dealing with complex and highly non-linear time series data. As a type of recurrent neural network (RNN), LSTM effectively solves the problem of gradient vanishing or gradient explosion in long sequence training through its unique gating mechanism (forget gate, input gate, output gate), and is suitable for short-term time series prediction. However, when the prediction time span increases, the computational complexity and memory occupancy of LSTM increase significantly, limiting its application in long-term prediction. Summary of the Invention

[0004] The present invention proposes a method and device for predicting liquid accumulation in tight gas wells, which can solve key technical problems such as insufficient prediction accuracy and inability to effectively capture long-term dependence relationships existing in traditional time series prediction methods when dealing with complex and highly non-linear time series data.

[0005] In various method embodiments of the present invention, a method for predicting liquid accumulation in tight gas wells includes:

[0006] Step S1: Obtain the production data of the tight gas well to be predicted, preprocess the production data of the tight gas well to be predicted, and perform normalization processing on the preprocessed production data of the tight gas well;

[0007] Step S2: Divide the normalized production data of the tight gas well into multiple subsequences according to time, input them into the trained liquid accumulation prediction model, and obtain the liquid accumulation prediction result;

[0008] The liquid accumulation prediction model includes a bidirectional processing module and a prediction module;

[0009] The bidirectional processing module includes a first LSTM layer and a second LSTM layer;

[0010] The first LSTM layer includes N forward LSTM sub - modules connected in sequence. Each subsequence is sorted in the forward time order, and each forward LSTM sub - module receives a subsequence in order. The first forward LSTM sub - module receives the subsequence ranked first and takes the generated feature vector as the output of the first forward LSTM sub - module. For each forward LSTM sub - module from the second forward LSTM sub - module to the (N - 1) - th forward LSTM sub - module, obtain the number of this forward LSTM sub - module, denote NUM0 as the number minus 1. This forward LSTM sub - module generates a feature vector based on the output of the NUM0 - th forward LSTM sub - module and its own input, and takes this feature vector as the input of the (NUM0 + 2) - th forward LSTM sub - module. The N - th forward LSTM sub - module generates the first output feature vector based on the feature vector generated by the (N - 1) - th forward LSTM sub - module and its own input.

[0011] The second LSTM layer includes N backward LSTM sub - modules connected in sequence. Each subsequence is sorted in the reverse time order, and each backward LSTM sub - module receives a subsequence in order. The first backward LSTM sub - module receives the subsequence ranked first and the first output feature vector, and takes the generated feature vector as the output of the first backward LSTM sub - module. For each backward LSTM sub - module from the second backward LSTM sub - module to the (N - 1) - th backward LSTM sub - module, obtain the number of this backward LSTM sub - module, denote NUM0 as the number minus 1. This backward LSTM sub - module generates a feature vector based on the output of the NUM0 - th backward LSTM sub - module and its own input, and takes this feature vector as the input of the (NUM0 + 2) - th backward LSTM sub - module. The N - th backward LSTM sub - module generates the fusion feature vector based on the feature vector generated by the (N - 1) - th backward LSTM sub - module and its own input.

[0012] The prediction module includes an Informer sub - module and a self - attention sub - module. The self - attention sub - module receives the fusion feature vector and the normalized tight gas well production data to generate attention weights. The Informer sub - module generates a liquid accumulation prediction result based on the fusion feature vector, the normalized tight gas well production data, and the attention weights. The liquid accumulation prediction result includes the predicted value or confidence interval of the liquid accumulation.

[0013] Optionally, the tight gas well production data to be predicted includes date, production horizon, production time, oil pressure, casing pressure, casing pressure drop, gas production rate, and liquid accumulation situation information. The liquid accumulation situation information is a boolean value indicating whether there is liquid accumulation.

[0014] Optionally, preprocessing the production data of the tight gas well to be predicted, and normalizing the preprocessed production data of the tight gas well, including:

[0015] Organize the production data of the tight gas well to be predicted into a time series

[0016]

[0017] Among them, is the time series of the production data of the tight gas well to be predicted, that is, the production data on the th day, including the following dimensions: date, production interval, production time, tubing pressure, casing pressure, casing pressure drop, gas production, liquid holdup information;

[0018] Clean the data of the time series;

[0019] For each of the tubing pressure, casing pressure, casing pressure drop, gas production, and liquid holdup as a normalization object, the following operations are performed on the normalization object:

[0020] Calculate the mean value:

[0021]

[0022] Among them, μ t is the mean value of the normalization object, S is the data volume of this normalization object, and t i is the feature vector corresponding to the i-th data of this normalization object;

[0023] Calculate the standard deviation:

[0024]

[0025] Among them, σ t is the standard deviation of the normalization object;

[0026] Normalization processing:

[0027]

[0028] t is the feature vector corresponding to all the data of the normalization object, is the transpose of μ t , is the normalization result.

[0029] Optionally, both the forward LSTM sub-module and the backward LSTM sub-module include a first LSTM layer, a first Dropout layer, a second LSTM layer, a second Dropout layer, a third LSTM layer, a third Dropout layer, and a fully connected layer connected in sequence.

[0030] Optionally, the self-attention sub-module receives the fused feature vector and the normalized tight gas well production data to generate attention weights, and the Informer sub-module generates a liquid holdup prediction result based on the fused feature vector, the normalized tight gas well production data, and the attention weights, including:

[0031] Incorporate the fused feature vector into the normalized tight gas well production data, that is, add a feature vector to the normalized tight gas well production data to form new normalized tight gas well production data; use the new normalized tight gas well production data as the input of the self-attention sub-module; where are the respective feature vectors in the new normalized tight gas well production data;

[0032] The self-attention sub-module determines the attention weight α p ,

[0033]

[0034] where , α p is the attention weight corresponding to the p-th feature vector of the new normalized tight gas well production data, s p represents the score or feature representation associated with the p-th feature vector, s q represents the score or feature representation associated with the q-th feature vector;

[0035] The self-attention sub-module determines the offset Δ p ,

[0036]

[0037] Δ p is the offset of the p-th feature vector, indicating the amount that the p-th feature vector needs to be adjusted in the attention mechanism; r pq is the attention weight between the p-th feature vector and the q-th feature vector, and this attention weight reflects the importance of the q-th feature vector when calculating the offset of the p-th feature vector; N is the total number of feature vectors in the new normalized tight gas well production data;

[0038] Superimpose the respective offsets Δ p corresponding to the respective feature vectors in the new normalized tight gas well production data to form the superimposed tight gas well production data;

[0039] The Informer sub-module generates a liquid accumulation prediction result based on the superimposed tight gas well production data.

[0040] In each of the above method embodiments of the present invention, a device for predicting liquid accumulation in a tight gas well includes:

[0041] A data processing module: configured to obtain the production data of the tight gas well to be predicted, preprocess the production data of the tight gas well to be predicted, and perform normalization processing on the preprocessed production data of the tight gas well;

[0042] A prediction module: configured to divide the normalized production data of the tight gas well into multiple subsequences according to time, input the subsequences into the trained liquid accumulation prediction model, and obtain the liquid accumulation prediction result;

[0043] The liquid accumulation prediction model includes a bidirectional processing module and a prediction module;

[0044] The bidirectional processing module includes a first LSTM layer and a second LSTM layer;

[0045] The first LSTM layer includes N forward LSTM sub-modules connected in sequence. The subsequences are sorted in the forward order of time. Each forward LSTM sub-module receives a subsequence in sequence. The first forward LSTM sub-module receives the subsequence ranked first and uses the generated feature vector as the output of the first forward LSTM sub-module. For each forward LSTM sub-module from the second forward LSTM sub-module to the N-1 forward LSTM sub-module, obtain the number of this forward LSTM sub-module, denote NUM0 as the number minus 1. This forward LSTM sub-module generates a feature vector based on the output of the NUM0 forward LSTM sub-module and its own input, and uses this feature vector as the input of the NUM0+2 forward LSTM sub-module. The Nth forward LSTM sub-module generates a first output feature vector based on the feature vector generated by the N-1 forward LSTM sub-module and its own input;

[0046] The second LSTM layer includes N reverse LSTM sub-modules connected in sequence. Each subsequence is sorted in reverse chronological order, and each reverse LSTM sub-module receives a subsequence in sequence. The first reverse LSTM sub-module receives the subsequence ranked first and the first output feature vector, and generates a feature vector as the output of the first reverse LSTM sub-module. For each reverse LSTM sub-module from the second reverse LSTM sub-module to the (N - 1)-th reverse LSTM sub-module, obtain the number of this reverse LSTM sub-module, denote NUM0 as the number minus 1. This reverse LSTM sub-module generates a feature vector based on the output of the NUM0-th reverse LSTM sub-module and its own input, and uses this feature vector as the input of the (NUM0 + 2)-th reverse LSTM sub-module. The N-th reverse LSTM sub-module generates a fused feature vector based on the feature vector generated by the (N - 1)-th reverse LSTM sub-module and its own input.

[0047] The prediction module includes an Informer sub-module and a self-attention sub-module. The self-attention sub-module receives the fused feature vector and the normalized tight gas well production data to generate attention weights. The Informer sub-module generates a liquid holdup prediction result based on the fused feature vector, the normalized tight gas well production data, and the attention weights. The liquid holdup prediction result includes the predicted value or confidence interval of the liquid holdup.

[0048] In the above method embodiments of the present invention, a computer-readable storage medium stores multiple instructions, and the multiple instructions are used to be loaded and executed by a processor to perform the method as described above.

[0049] In the above method embodiments of the present invention, an electronic device includes: a processor for executing multiple instructions; a memory for storing multiple instructions; wherein, the multiple instructions are stored by the memory and loaded and executed by the processor to perform the method as described above.

[0050] The present invention innovatively proposes an efficient prediction model that deeply integrates the Long Short-Term Memory network (LSTM) and the Informer model, aiming to break through the limitations of traditional time series prediction methods and achieve a double leap in short-term and long-term prediction accuracy. Through a carefully designed architecture, this algorithm ingeniously combines the excellent memory ability of LSTM in processing short-term time series data with the computational efficiency and memory-friendly nature demonstrated by the Informer algorithm when dealing with ultra-long sequences, thereby constructing a precise and efficient prediction framework. In terms of short-term prediction, the LSTM layer, as one of the core components of the model, utilizes its internal gating mechanisms (forget gate, input gate, output gate) to effectively capture and store key information in the time series, especially those patterns and trends crucial for recent predictions. This mechanism ensures that the model can maintain a high level of sensitivity and accuracy in the face of complex and changing short-term fluctuations, providing solid data support for immediate decision-making. Regarding the challenges of long-term prediction, the Informer algorithm, with its deformable attention mechanism and sparsification technology, significantly reduces the computational complexity and memory consumption problems faced by traditional sequence-to-sequence models when dealing with long sequences. By only focusing on the key parts of the sequence, Informer can efficiently capture long-term dependencies, enabling the model to still maintain a high level of accuracy when predicting trends and patterns in a relatively long future time range. This ability has inestimable value for industries that require long-term planning, such as tight gas liquid loading prediction.

[0051] The present invention has the following advantages:

[0052] The method for predicting tight gas well liquid loading of the present invention combines LSTM and Informer, and can quickly and effectively predict the liquid loading situation of tight gas wells. Effective prediction of the liquid loading situation can help production units take effective measures in a timely manner, improve the production efficiency of tight gas wells, and reduce losses caused by liquid loading. The present invention uses a deformable attention mechanism to improve the Informer model and enhance the prediction accuracy. It can adapt to different time spans of prediction according to user needs.

[0053] The present invention combines the advantages of the LSTM model and the Informer model to achieve high-precision short-term and long-term time series prediction. The Informer model of the present invention can perform efficient calculations through a deformable attention mechanism and a generative decoder, significantly reducing the computational complexity and memory occupancy. The present invention can adapt to prediction tasks with different time spans according to user needs.

[0054] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Description of the Drawings

[0055] The embodiments of the present invention will be described in more detail with reference to the accompanying drawings. The above and other objects, features, and advantages of the present invention will become more apparent. The drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings, the same reference numerals generally represent the same components or steps.

[0056] Figure 1 Schematic flow diagram of the method for predicting liquid accumulation in tight gas wells of the present invention;

[0057] Figure 2 Schematic structural diagram of the time prediction module of the present invention;

[0058] Figure 3 Schematic structural diagram of the device for predicting liquid accumulation in tight gas wells of the present invention;

[0059] Figure 4 Schematic structural diagram of the electronic device for predicting liquid accumulation in tight gas wells of the present invention. Detailed implementation manners

[0060] Next, exemplary embodiments of the present invention will be described in detail with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments of the present invention. It should be understood that the present invention is not limited by the exemplary embodiments described herein. It should be noted that: unless otherwise specifically stated, the relative arrangements, numerical expressions, and values of the components and steps described in these embodiments do not limit the scope of the present invention.

[0061] Those skilled in the art can understand that terms such as "first", "second", S1, S2, etc. in the embodiments of the present invention are only used to distinguish different steps, devices or modules, etc., and neither represent any specific technical meaning nor indicate an inevitable logical order between them. It should also be understood that in the embodiments of the present invention, "a plurality of" may refer to two or more, and "at least one" may refer to one, two or more. It should also be understood that for any component, data or structure mentioned in the embodiments of the present invention, without clear limitation or contrary indication in the context, it can generally be understood as one or more. In addition, the term "and / or" in the present invention is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in the present invention generally represents an "or" relationship between the associated objects before and after. It should also be understood that the present invention emphasizes the differences between the various embodiments, and their similarities can be referred to each other. For the sake of brevity, they will not be elaborated one by one. At the same time, it should be understood that for the sake of description, the dimensions of the various parts shown in the drawings are not drawn according to the actual proportional relationship. The following description of at least one exemplary embodiment is actually only illustrative and in no way limits the present invention and its application or use. Technologies, methods and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods and devices should be regarded as part of the specification. It should be noted that: similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0062] Embodiments of the present invention can be applied to electronic devices such as terminal devices, computer systems, servers, etc., which can operate together with many other general or special computing system environments or configurations. Examples of well-known terminal devices, computing systems, environments, and / or configurations suitable for use with electronic devices such as terminal devices, computer systems, servers, etc. include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network personal computers, minicomputer systems, mainframe computer systems, and distributed cloud computing technology environments including any of the above systems, and so on. Electronic devices such as terminal devices, computer systems, servers, etc. can be described in the general context of computer system-executable instructions (such as program modules) executed by a computer system. Generally, program modules can include routines, programs, object programs, components, logic, data structures, etc., which perform specific tasks or implement specific abstract data types. The computer system / server can be implemented in a distributed cloud computing environment where tasks are executed by remote processing devices linked through a communication network. In a distributed cloud computing environment, program modules can be located on local or remote computing system storage media including storage devices.

[0063] Exemplary method

[0064] Figure 1 is a schematic flowchart of a method for predicting liquid accumulation in a tight gas well provided by an exemplary embodiment of the present invention. As Figure 1 shown, it includes the following steps:

[0065] Step S1: Obtain the production data of the tight gas well to be predicted, preprocess the production data of the tight gas well to be predicted, and perform normalization processing on the preprocessed production data of the tight gas well;

[0066] Step S2: Divide the normalized production data of the tight gas well into multiple subsequences according to time, input them into the trained liquid accumulation prediction model, and obtain the liquid accumulation prediction result;

[0067] The liquid accumulation prediction model includes a bidirectional processing module and a prediction module;

[0068] The bidirectional processing module includes a first LSTM layer and a second LSTM layer;

[0069] The first LSTM layer includes N forward LSTM sub-modules connected in sequence. Each subsequence is sorted in the forward order of time. Each forward LSTM sub-module receives a subsequence in order. The first forward LSTM sub-module receives the subsequence ranked first and uses the generated feature vector as the output of the first forward LSTM sub-module. For each forward LSTM sub-module from the second forward LSTM sub-module to the (N - 1)-th forward LSTM sub-module, obtain the number of this forward LSTM sub-module, denote NUM0 as the number minus 1. This forward LSTM sub-module generates a feature vector based on the output of the NUM0-th forward LSTM sub-module and its own input, and uses this feature vector as the input of the (NUM0 + 2)-th forward LSTM sub-module. The N-th forward LSTM sub-module generates a first output feature vector based on the feature vector generated by the (N - 1)-th forward LSTM sub-module and its own input.

[0070] The second LSTM layer includes N backward LSTM sub-modules connected in sequence. Each subsequence is sorted in the reverse order of time. Each backward LSTM sub-module receives a subsequence in order. The first backward LSTM sub-module receives the subsequence ranked first and the first output feature vector, and uses the generated feature vector as the output of the first backward LSTM sub-module. For each backward LSTM sub-module from the second backward LSTM sub-module to the (N - 1)-th backward LSTM sub-module, obtain the number of this backward LSTM sub-module, denote NUM0 as the number minus 1. This backward LSTM sub-module generates a feature vector based on the output of the NUM0-th backward LSTM sub-module and its own input, and uses this feature vector as the input of the (NUM0 + 2)-th backward LSTM sub-module. The N-th backward LSTM sub-module generates a fusion feature vector based on the feature vector generated by the (N - 1)-th backward LSTM sub-module and its own input.

[0071] The prediction module includes an Informer sub-module and a self-attention sub-module. The self-attention sub-module receives the fusion feature vector and the normalized tight gas well production data to generate attention weights. The Informer sub-module generates a liquid holdup prediction result based on the fusion feature vector, the normalized tight gas well production data, and the attention weights. The liquid holdup prediction result includes the predicted value or confidence interval of the liquid holdup.

[0072] In the present invention, the Informer sub-module is a conventional Informer model in the art. The production data of the tight gas well to be predicted is time series data, including historical observation values, timestamps and other information. By cleaning, normalizing and other operations on the production data of the tight gas well to be predicted, a standard input is provided for the liquid holdup prediction model. The bidirectional processing module constructs a short-term time series prediction model based on the LSTM network module, which is applicable to predicting data in the next few hours to days. The prediction module constructs a long-term time series prediction model through deformable attention based on the output of the bidirectional processing module, which is applicable to predicting data in the next few days to weeks. The liquid holdup prediction results include information such as predicted values and confidence intervals.

[0073] The fusion feature vector of the present invention is obtained by fusing the feature vectors generated by each reverse LSTM sub-module and the feature vectors generated by each forward LSTM sub-module. Among them, the forward LSTM sub-module can capture the trend and periodic changes of data over time, while the reverse LSTM sub-module can reveal hidden information or reverse correlations that may be ignored due to forward processing. By fusing the feature vectors from these two different perspectives, the model can comprehensively understand the time series data, thereby improving the prediction accuracy.

[0074] Further, the production data of the tight gas well to be predicted includes date, production horizon, production time, tubing pressure, casing pressure, casing pressure drop, gas production, and liquid holdup information. The liquid holdup information is a boolean value indicating whether there is liquid holdup.

[0075] In the present invention, in the production data of the tight gas well, the data of each day includes a total of 30 features. In order to reduce the data dimension and the complexity of subsequent processing, only the date, production horizon, production time, tubing pressure, casing pressure, casing pressure drop, gas production, and liquid holdup information are selected.

[0076] Preprocessing the production data of the tight gas well to be predicted, and normalizing the preprocessed production data of the tight gas well, including:

[0077] Organize the production data of the tight gas well to be predicted into a time series

[0078]

[0079] where is the time series of the production data of the tight gas well to be predicted, that is, the production data of the th day, which includes the following dimensions: date, production horizon, production time, tubing pressure, casing pressure, casing pressure drop, gas production, and liquid holdup information;

[0080] Clean the data of the time series

[0081] Each of the oil pressure, casing pressure, casing pressure drop, gas production volume, and liquid holdup is taken as a normalization object, and the following operations are performed on the normalization object:

[0082] Calculate the mean value:

[0083] where μ t is the mean value of the normalization object, S is the data volume of the normalization object, and t i is the eigenvector corresponding to the i-th data of the normalization object;

[0084] Calculate the standard deviation:

[0085]

[0086] where σ t is the standard deviation of the normalization object;

[0087] Normalization processing:

[0088]

[0089] t is the eigenvector corresponding to all the data of the normalization object, is the transpose of μ t , is the normalization result.

[0090] In the present invention, the data is organized into a time series,

[0091]

[0092] where i.e., the production data of the i-th day, which includes the following dimensions: date, production interval, production time, oil pressure, casing pressure, casing pressure drop, gas production volume, and liquid holdup information. For the data with one day as the unit, the experts annotate the data according to the changes in the oil pressure, casing pressure, casing pressure drop, production time of the specific well, and the recovery of the casing pressure after shutting in the well, and the comprehensive liquid holdup investigation report. The data marked is the production data of 30 tight gas wells in a certain block in the northwest for about eight years. The first six years of the marked data are used as the training set, and the data of the last two years are used as the validation set.

[0093] For the liquid holdup situation, for the data with one day as the unit, the experts annotate the data according to the changes in the oil pressure, casing pressure, casing pressure drop, production time of the specific well, and the recovery of the casing pressure after shutting in the well, and the comprehensive liquid holdup investigation report. The data marked is the production data of 30 tight gas wells in a certain block in the northwest for about eight years. The first six years of the marked data are used as the training set, and the data of the last two years are used as the validation set.

[0094] Secondly, data cleaning is performed. During this process, possible problems are the presence of outliers or null values in the data. For outliers, they are processed into null values and then uniformly processed with the null values. For null values, the interpolation method is adopted, and the average of the previous value and the next value of the missing value is calculated. Interpolation operations are performed on each column of the data to fill the null values. Data anomalies and null values are only occasionally seen in the production data of a few tight gas wells.

[0095] Then, the numerical feature information including tubing pressure, casing pressure, casing pressure drop, gas production, and liquid holdup is normalized, so as to make the data meet the training format requirements and eliminate or avoid the problem of gradient explosion that may occur during the training process.

[0096] For the numerical features of the production data of tight gas wells, namely tubing pressure, casing pressure, casing pressure drop, gas production, and liquid holdup, first calculate their means:

[0097]

[0098] where t i is the feature vector after removing the non-numerical feature production layer from among them. Then, calculate its standard deviation:

[0099]

[0100] Finally, normalize the numerical features of the production data of tight gas wells, namely tubing pressure, casing pressure, casing pressure drop, gas production, and liquid holdup, that is, each feature value is converted into a standard normal distribution with a mean of 0 and a variance of 1.

[0101]

[0102] For the feature of the production layer of tight gas wells, which is a non-numerical feature, the One-hot algorithm is used to encode it.

[0103] Step S2: Divide the normalized production data of tight gas wells into multiple subsequences according to time, and input them into the trained liquid holdup prediction model to obtain the liquid holdup prediction results.

[0104] Furthermore, both the forward LSTM sub-module and the backward LSTM sub-module include a first LSTM layer, a first Dropout layer, a second LSTM layer, a second Dropout layer, a third LSTM layer, a third Dropout layer, and a fully connected layer connected in sequence.

[0105] In the present invention, during the prediction of the liquid holdup situation of tight gas wells, first, the production data of tight gas wells are segmented according to the time dimension. The production data of tight gas wells form a time series X = {x1, x2,..., x T} is divided into multiple subsequences, Xi ={x si ,x si+1 ,...,x ei}。s i ,e i are the start and end subscripts of the subsequence respectively. The production data of tight gas wells is in days. The segment length is set to one month. Each subsequence retains some of the time characteristics in the original tight gas well production data, but due to its reduced scale, subsequent processing is more efficient. The core purpose of this step is to provide a more refined time perspective for the model to better capture the short-term dynamic change patterns hidden in the production data.

[0106] Next, the segmented production data of these tight gas wells are separately fed into the forward LSTM module and the backward LSTM module. The forward LSTM module processes according to the natural order of the production data of tight gas wells (i.e., the order of increasing time), and it can capture the trends and periodic changes of the data as time progresses, especially those short-term patterns that are crucial for recent predictions.

[0107]

[0108] x t is the input at the current time step, is the forward hidden state at the previous time step, LSTM forward represents the forward LSTM. Segment the cleaned data, reshape it and then import it into the forward LSTM.

[0109] Construct an LSTM model. First use the LSTM layer, and then use the Dropout layer to reduce the possibility of overfitting. Stack the LSTM layer and the Dropout layer three times repeatedly. Finally, add a fully connected layer as the output layer.

[0110] The backward LSTM module, on the contrary, traverses the subsequences formed by the production data of tight gas wells in the order of decreasing time, which helps the model to examine the data from another angle and reveal hidden information or reverse correlations that may be overlooked due to forward processing.

[0111]

[0112] is the backward hidden state at the previous time step.

[0113] The construction of the backward LSTM module is the same as that of the forward one, that is, the production data of tight gas wells segmented by month is input in reverse time.

[0114] Through this bidirectional processing mechanism, the model can more comprehensively grasp the short-term characteristics of the time series.

[0115] After the processing of the bidirectional LSTM module, the output results of the two modules, namely the liquid holdup prediction data, are effectively integrated to form a comprehensive characterization of the liquid holdup situation in short-term tight gas wells. This characterization not only includes the positive development trend of the liquid holdup in tight gas wells but also incorporates supplementary information on the development trend of the liquid holdup in tight gas wells from a reverse perspective, thus greatly enhancing the model's ability to capture short-term fluctuations.

[0116]

[0117] In the present invention, a fused feature vector is obtained by fusing the feature vectors generated by each reverse LSTM sub-module and each forward LSTM sub-module. Among them, the forward LSTM sub-module can capture trends and periodic changes in the time series, while the reverse LSTM sub-module reveals reverse correlations and hidden information that may be ignored in the forward processing. By connecting these two sets of feature vectors in chronological order, not only can the original order information of the time series be retained, but also the feature information from different perspectives is integrated, thus providing a more comprehensive input representation for the model.

[0118] Furthermore, the self-attention sub-module receives the fused feature vector and the normalized production data of the tight gas well to generate attention weights, and the Informer sub-module generates a liquid holdup prediction result based on the fused feature vector, the normalized production data of the tight gas well, and the attention weights, including:

[0119] Incorporate the fused feature vector into the normalized production data of the tight gas well, that is, add a feature vector to the normalized production data of the tight gas well, and the new normalized production data of the tight gas well is used as the input of the self-attention sub-module; where is each feature vector in the new normalized production data of the tight gas well;

[0120] The self-attention sub-module determines the attention weight α of each feature vector p ,

[0121]

[0122] where , α p is the attention weight corresponding to the p-th feature vector of the new normalized production data of the tight gas well, s p represents the score or feature representation associated with the p-th feature vector, s q represents the score or feature representation associated with the q-th feature vector;

[0123] The self-attention sub-module determines the offset Δ of each feature vector p ,

[0124]

[0125] Δ p is the offset of the p-th feature vector, indicating the amount that the p-th feature vector needs to be adjusted in the attention mechanism; r pq is the attention weight between the p-th feature vector and the q-th feature vector, which reflects the importance of the q-th feature vector in calculating the offset of the p-th feature vector; N is the total number of feature vectors in the newly normalized tight gas well production data.

[0126] Superimpose each feature vector in the newly normalized tight gas well production data with its corresponding offset Δ p , to form the superimposed tight gas well production data;

[0127] The Informer sub-module generates a liquid accumulation prediction result based on the superimposed tight gas well production data.

[0128] In the present invention, the outputs of each LSTM model, that is, the liquid accumulation conditions of the one-dimensional tight gas wells, are connected in chronological order and incorporated into the sequence of the tight gas well production data, thus equivalent to adding a new feature, together with the original production data, to form the input data of the Informer. By importing this data into the Informer module, the long-term law of the time series can be further explored. With its efficient deformable attention mechanism, the Informer module can significantly reduce the computational complexity and memory consumption when processing long sequences, while maintaining the ability to keenly capture long-term dependencies.

[0129] Assume that the input data of the Informer model is a series of feature vectors, denoted as

[0130]

[0131] where N is the length of the tight gas well production data (the length after adding the predicted liquid accumulation conditions of the tight gas wells). For each feature vector z i , by introducing a position offset Δ i , it is transformed to obtain a new feature vector . The position offset Δ i is obtained by weighted summation according to the attention weight and the feature vectors of other input data.

[0132] For the calculation of the attention weight, the score s of each feature vector is calculated through the function f iThen, the scores are exponentiated and normalized to obtain the attention weights. , where j represents the index when traversing all feature vectors.

[0133] Calculation of the position offset: According to the attention weights and the feature vectors of other input data, a weighted sum is performed to obtain the position offset Δ. i This process is achieved by multiplying each feature vector by the attention weights and then summing the resulting weighted feature vectors.

[0134] Adjust the shape of the attention mechanism: By applying the position offset to the original position of the input data, the shape of the attention mechanism can be adjusted. Specifically, the new feature vector can be expressed as the original feature vector plus the position offset.

[0135] Deformable self-attention module: In the processing of tight gas well production data, the deformable self-attention module can be used to enhance the features of tight gas well production data. This converts the input vector into a feature map and then generates a Query vector, while considering the coordinates of the reference point. By combining the offset module and the attention module, the final output result is obtained. At this stage, the Informer module deeply mines the long-term trends, seasonal patterns, and possible periodic fluctuations of liquid holdup in tight gas well production data, providing a solid foundation for long-term prediction.

[0136]

[0137] O t is the output of the Informer module at time step t.

[0138] Finally, by combining the accurate capture of short-term liquid holdup patterns by the bidirectional LSTM module and the in-depth insight into long-term liquid holdup patterns by the Informer module, the prediction model proposed by the present invention achieves comprehensive, efficient, and accurate prediction of tight gas well production data with liquid holdup.

[0139] In this embodiment, the first six years of data of the labeled tight gas well production data are used as the training set, and the data of the next two years are used as the validation set. The liquid holdup prediction model is trained using the training set data; the trained liquid holdup prediction model is verified on the validation set to obtain the accuracy rate of the tight gas well liquid holdup prediction model.

[0140] Exemplary device

[0141] Figure 3 is a schematic structural diagram of a device for predicting tight gas well liquid holdup provided by an exemplary embodiment of the present invention. As Figure 3 shown, this embodiment includes:

[0142] Data processing module: configured to obtain the production data of the tight gas well to be predicted, preprocess the production data of the tight gas well to be predicted, and perform normalization processing on the preprocessed production data of the tight gas well;

[0143] Prediction module: configured to divide the normalized production data of the tight gas well into multiple subsequences according to time, input the trained liquid holdup prediction model, and obtain the liquid holdup prediction result;

[0144] The liquid holdup prediction model includes a bidirectional processing module and a prediction module;

[0145] The bidirectional processing module includes a first LSTM layer and a second LSTM layer;

[0146] The first LSTM layer includes N forward LSTM sub-modules connected in sequence. Sort each subsequence in the forward order of time. Each forward LSTM sub-module receives a subsequence in sequence. The first forward LSTM sub-module receives the subsequence ranked first and takes the generated feature vector as the output of the first forward LSTM sub-module; for each forward LSTM sub-module from the second forward LSTM sub-module to the N-1 forward LSTM sub-module, obtain the number of this forward LSTM sub-module, denote NUM0 as the number minus 1. This forward LSTM sub-module generates a feature vector based on the output of the NUM0 forward LSTM sub-module and its own input, and takes this feature vector as the input of the NUM0+2 forward LSTM sub-module; the Nth forward LSTM sub-module generates a first output feature vector based on the feature vector generated by the N-1 forward LSTM sub-module and its own input;

[0147] The second LSTM layer includes N reverse LSTM sub-modules connected in sequence. Sort each subsequence in the reverse order of time. Each reverse LSTM sub-module receives a subsequence in sequence. The first reverse LSTM sub-module receives the subsequence ranked first and the first output feature vector, and takes the generated feature vector as the output of the first reverse LSTM sub-module; for each reverse LSTM sub-module from the second reverse LSTM sub-module to the N-1 reverse LSTM sub-module, obtain the number of this reverse LSTM sub-module, denote NUM0 as the number minus 1. This reverse LSTM sub-module generates a feature vector based on the output of the NUM0 reverse LSTM sub-module and its own input, and takes this feature vector as the input of the NUM0+2 reverse LSTM sub-module; the Nth reverse LSTM sub-module generates a fusion feature vector based on the feature vector generated by the N-1 reverse LSTM sub-module and its own input;

[0148] The prediction module includes an Informer sub-module and a self-attention sub-module. The self-attention sub-module receives the fused feature vector and the normalized tight gas well production data to generate attention weights. The Informer sub-module generates a liquid holdup prediction result based on the fused feature vector, the normalized tight gas well production data, and the attention weights. The liquid holdup prediction result includes a predicted value or a confidence interval of the liquid holdup.

[0149] Exemplary electronic device

[0150] Figure 4 FIG. 7 is a schematic diagram of the structure of the electronic device 40 provided by an exemplary embodiment of the present invention. The electronic device may be either the first device or the second device, or both, or a stand-alone device independent of them. The stand-alone device may communicate with the first device and the second device to receive the input signals collected from them. Figure 4 FIG. 8 illustrates a block diagram of an electronic device according to an embodiment of the present disclosure. As Figure 4 shown, the electronic device includes one or more processors 41 and a memory 42.

[0151] The processor 41 may be a central processing unit (CPU) or other form of processing unit having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions.

[0152] The memory 42 may include one or more computer program products, and the computer program products may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory, etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage media, and the processor 41 may run the program instructions to implement the methods of the software programs of the various embodiments of the present disclosure described above and / or other desired functions. In one example, the electronic device may further include: an input device 43 and an output device 44, and these components are interconnected through a bus system and / or other forms of connection mechanisms (not shown). In addition, the input device 43 may further include, for example, a keyboard, a mouse, etc. The output device 44 may output various information to the outside. The output device 44 may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc.

[0153] Of course, for simplicity, Figure 4Only some of the components related to the present disclosure in the electronic device are shown, and components such as buses, input / output interfaces, etc. are omitted. In addition, according to specific application scenarios, the electronic device may further include any other appropriate components.

[0154] Exemplary computer program product and computer-readable storage medium

[0155] In addition to the above methods and devices, embodiments of the present disclosure may also be computer program products, which include computer program instructions that, when run by a processor, cause the processor to execute the steps in the methods according to various embodiments of the present disclosure described in the "Exemplary Methods" section above of this specification.

[0156] The computer program product may be written in any combination of one or more programming languages for programming code to perform the operations of the embodiments of the present disclosure. The programming languages include object-oriented programming languages such as Java, C++, etc., and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, executed as a stand-alone software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0157] Furthermore, embodiments of the present disclosure may also be computer-readable storage media, on which computer program instructions are stored, and when the computer program instructions are run by a processor, the processor is caused to execute the steps in the methods according to various embodiments of the present disclosure described in the "Exemplary Methods" section above of this specification.

[0158] The computer-readable storage media may adopt any combination of one or more readable media. The readable media may be a readable signal medium or a readable storage medium. The readable storage medium may, for example, include but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or components, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include: electrical connections with one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0159] The basic principles of the present disclosure have been described in conjunction with specific embodiments. However, it should be noted that the advantages, benefits, effects, etc. mentioned in the present disclosure are only examples and not limitations. It cannot be considered that these advantages, benefits, effects, etc. are essential for each embodiment of the present disclosure. Additionally, the specific details disclosed above are only for illustrative and easy-to-understand purposes and not limitations. The above details do not limit the present disclosure to necessarily implement using the above specific details.

[0160] Each embodiment in this specification is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. For the same or similar parts among the embodiments, reference can be made to each other. For system embodiments, since they basically correspond to method embodiments, they are described relatively simply. For the relevant parts, reference can be made to the partial description of the method embodiments.

[0161] The block diagrams of the devices, apparatuses, equipment, and systems involved in the present disclosure are only illustrative examples and do not intend to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open-ended terms, meaning "including but not limited to", and can be used interchangeably with each other. The word "or" and "and" used herein refer to the word "and / or" and can be used interchangeably with each other, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to" and can be used interchangeably with each other.

[0162] The methods and apparatuses of the present disclosure can be implemented in many ways. For example, the methods and apparatuses of the present disclosure can be implemented through software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of the steps for the methods is only for illustration. The steps of the methods of the present disclosure are not limited to the above specific description order, unless otherwise specifically stated. Additionally, in some embodiments, the present disclosure can also be implemented as a program recorded in a recording medium, and these programs include machine-readable instructions for implementing the methods according to the present disclosure. Therefore, the present disclosure also covers a recording medium storing a program for executing the methods according to the present disclosure.

[0163] It should also be noted that in the devices, equipment and methods of the present disclosure, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations shall be regarded as equivalent solutions of the present disclosure. The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of the present disclosure. Therefore, the present disclosure is not intended to be limited to the aspects shown herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

[0164] The above description has been given for purposes of illustration and description. In addition, this description is not intended to limit the embodiments of the present disclosure to the forms disclosed herein. Although several example aspects and embodiments have been discussed above, those skilled in the art will recognize some of their variations, modifications, alterations, additions and subcombinations.

Claims

1. A method for predicting liquid loading in tight gas wells, characterized in that: Methods include: Step S1: obtaining the production data of the tight gas well to be predicted, preprocessing the production data of the tight gas well to be predicted, and normalizing the preprocessed production data of the tight gas well; Step S2: dividing the normalized tight gas well production data into multiple subsequences according to time, inputting the trained liquid accumulation prediction model, and obtaining liquid accumulation prediction results; The effusion prediction model includes a bidirectional processing module and a prediction module; The bidirectional processing module includes a first LSTM layer and a second LSTM layer; The prediction module includes the Informer submodule and the self-attention submodule; The self-attention submodule receives the fused feature vector generated by the second LSTM layer and the normalized tight gas well production data to generate attention weights. The informer submodule generates liquid accumulation prediction results based on the fused feature vector, the normalized tight gas well production data and the attention weights: The fused feature vector is incorporated into the normalized tight gas well production data, that is, a feature vector is added to the normalized tight gas well production data to form new normalized tight gas well production data; the new normalized tight gas well production data As the input of the self-attention submodule; where are the characteristic vectors in the new normalized tight gas well production data; The self-attention submodule determines the attention weight α of each feature vector p , ， in, , α p is the attention weight corresponding to the pth eigenvector of the new normalized tight gas well production data, s p represents the score or feature representation associated with the pth feature vector, s q represents the score or feature representation associated with the qth feature vector; The self-attention submodule determines the offset Δ of each feature vector p , ， Δ p is the offset of the p-th feature vector, indicating the amount by which the p-th feature vector needs to be adjusted in the attention mechanism; r pq is the attention weight between the p-th eigenvector and the q-th eigenvector, which reflects the importance of the q-th eigenvector in calculating the offset of the p-th eigenvector; N is the total number of eigenvectors in the new normalized tight gas well production data; Each feature vector in the new normalized tight gas well production data is superimposed with its corresponding offset Δ p , forming the superimposed tight gas well production data; The Informer submodule generates liquid loading prediction results based on the superimposed tight gas well production data; The effusion prediction results include the predicted value or confidence interval of the effusion.

2. The method according to claim 1, characterized in that The first LSTM layer includes n forward LSTM submodules connected in sequence, and each subsequence is sorted in a forward time order. Each forward LSTM submodule receives a subsequence in sequence, and the first forward LSTM submodule receives the first subsequence, and the generated feature vector is used as the output of the first forward LSTM submodule; For each forward LSTM submodule from the second forward LSTM submodule to the n-1th forward LSTM submodule, obtain the number of the forward LSTM submodule, record NUM0 as the number minus 1, generate a feature vector based on the output of the NUM0th forward LSTM submodule and the input of the forward LSTM submodule itself, and use the feature vector as the input of the NUM0+2th forward LSTM submodule; the nth forward LSTM submodule generates a first output feature vector based on the feature vector generated by the n-1th forward LSTM submodule and its own input; The second LSTM layer includes n reverse LSTM submodules connected in sequence, and each subsequence is sorted in reverse time order. Each reverse LSTM submodule receives a subsequence in sequence. The first reverse LSTM submodule receives the subsequence ranked first and the first output feature vector, and uses the generated feature vector as the output of the first reverse LSTM submodule; for each reverse LSTM submodule from the second reverse LSTM submodule to the n-1th reverse LSTM submodule, obtain the number of the reverse LSTM submodule, record NUM0 as the number minus 1, and the reverse LSTM submodule generates a feature vector based on the output of the NUM0th reverse LSTM submodule and the input of the reverse LSTM submodule itself, and uses the feature vector as the input of the NUM0+2th reverse LSTM submodule; the nth reverse LSTM submodule generates a fused feature vector based on the feature vector generated by the n-1th reverse LSTM submodule and its own input; The production data of the tight gas well to be predicted include date, production layer, production time, oil pressure, casing pressure, casing pressure drop, gas production, and liquid accumulation information. The liquid accumulation information is a Boolean value indicating whether there is liquid accumulation.

3. The method according to claim 1, characterized in that The preprocessing of the tight gas well production data to be predicted and the normalization of the preprocessed tight gas well production data include: Organize the tight gas well production data to be predicted into time series ， in, is the time series of tight gas well production data to be predicted, That is Daily production data, including the following dimensions: date, production layer, production time, oil pressure, casing pressure, casing pressure drop, gas production, and liquid accumulation information; Performing data cleaning on the time series; Each of the oil pressure, casing pressure, casing pressure drop, gas production, and liquid accumulation is taken as a normalized object, and the following operations are performed on the normalized objects: Calculate the mean: ， Among them, μ t is the mean of the normalized object, S is the amount of data of the normalized object, t i is the feature vector corresponding to the i-th data of the normalized object; Calculate the standard deviation: ， Among them, σ t is the standard deviation of the normalized object; Normalization: ， t is the eigenvector corresponding to all the data of the normalized object, μ t The transpose of is the normalized result.

4. The method according to claim 2, characterized in that The forward LSTM submodule and the reverse LSTM submodule each include a first LSTM layer, a first Dropout layer, a second LSTM layer, a second Dropout layer, a third LSTM layer, a third Dropout layer, and a fully connected layer, which are connected in sequence.

5. A device for predicting liquid loading in tight gas wells, used to execute the method according to any one of claims 1 to 4, characterized in that: The device comprises: Data processing module: configured to obtain production data of the tight gas well to be predicted, preprocess the production data of the tight gas well to be predicted, and normalize the preprocessed production data of the tight gas well; Prediction module: configured to divide the normalized tight gas well production data into multiple subsequences according to time, input the trained liquid accumulation prediction model, and obtain the liquid accumulation prediction result; The effusion prediction model includes a bidirectional processing module and a prediction module; The bidirectional processing module includes a first LSTM layer and a second LSTM layer; The first LSTM layer includes n forward LSTM submodules connected in sequence, and each subsequence is sorted in a forward time order. Each forward LSTM submodule receives a subsequence in sequence, and the first forward LSTM submodule receives the first subsequence, and uses the generated feature vector as the output of the first forward LSTM submodule; for each forward LSTM submodule from the second forward LSTM submodule to the n-1th forward LSTM submodule, the number of the forward LSTM submodule is obtained, and NUM0 is recorded as the number minus 1. The forward LSTM submodule generates a feature vector based on the output of the NUM0th forward LSTM submodule and the input of the forward LSTM submodule itself, and uses the feature vector as the input of the NUM0+2th forward LSTM submodule; the nth forward LSTM submodule generates a first output feature vector based on the feature vector generated by the n-1th forward LSTM submodule and its own input; The second LSTM layer includes n reverse LSTM submodules connected in sequence, and each subsequence is sorted in reverse time order. Each reverse LSTM submodule receives a subsequence in sequence. The first reverse LSTM submodule receives the subsequence ranked first and the first output feature vector, and uses the generated feature vector as the output of the first reverse LSTM submodule; for each reverse LSTM submodule from the second reverse LSTM submodule to the n-1th reverse LSTM submodule, obtain the number of the reverse LSTM submodule, record NUM0 as the number minus 1, and the reverse LSTM submodule generates a feature vector based on the output of the NUM0th reverse LSTM submodule and the input of the reverse LSTM submodule itself, and uses the feature vector as the input of the NUM0+2th reverse LSTM submodule; the nth reverse LSTM submodule generates a fused feature vector based on the feature vector generated by the n-1th reverse LSTM submodule and its own input; The prediction module includes an Informer submodule and a self-attention submodule. The self-attention submodule receives the fused feature vector and the normalized tight gas well production data to generate an attention weight. The Informer submodule generates a liquid accumulation prediction result based on the fused feature vector, the normalized tight gas well production data and the attention weight. The liquid accumulation prediction result includes a predicted value or a confidence interval of the liquid accumulation.

6. A computer-readable storage medium, characterized in that: The storage medium stores a plurality of instructions; the plurality of instructions are used for a processor to load and execute the method as described in any one of claims 1 to 4.

7. An electronic device, characterized in that: The electronic device comprises: A processor, which is used to execute multiple instructions; A memory for storing a plurality of instructions; The plurality of instructions are used to be stored in the memory and loaded and executed by the processor according to any one of claims 1 to 4.

Images