Method and device for determining remaining service life of electric submersible pump and electronic equipment

Abstract

The invention provides a method and device for determining the remaining service life of an electric submersible pump and electronic equipment. Based on the method, before specific implementation, a preset life prediction model at least comprising a main prediction model based on a Wavelet-LSTM structure can be obtained through deep learning training. In specific implementation, corresponding target time sequence data can be constructed according to acquired operation parameters of a target electric submersible pump and environmental parameters of a working condition environment during operation; a preset service life prediction model is used for processing the target time sequence data, and meanwhile a prediction value of the remaining service life of the target electric submersible pump and uncertainty parameters related to the prediction value of the remaining service life are determined to serve as a target prediction result; and according to the target prediction result, the remaining service life of the target electric submersible pump is determined. Therefore, the method can be well adapted to a complex operation condition scene of the electric submersible pump, and the remaining service life of the electric submersible pump can be accurately predicted.

Description

Technical Field

[0001] This manual belongs to the field of oil and gas field lift construction technology, and in particular relates to the methods, devices and electronic equipment for determining the remaining service life of electric submersible pumps. Background Technology

[0002] Electric submersible pumps (ESPs) are among the most widely used mechanical lifting devices, extensively applied in oil and gas production wells and water injection systems. Typically, these ESPs operate in complex and harsh environments such as high temperature, high pressure, and high sand content. Furthermore, the relatively complex structure of ESPs means that failure of any critical component can trigger a system-wide malfunction, leading to unplanned downtime, impacting crude oil production and water injection volume, and causing significant economic losses to oil and gas field operations. Therefore, identifying potential failure risks in advance and accurately assessing the remaining useful life (RUL) of ESPs is of paramount importance.

[0003] However, existing methods for predicting the remaining service life of electric submersible pumps often fail to adequately adapt to the complex operating conditions of these pumps, resulting in relatively large prediction errors and low reference value.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This manual provides a method, apparatus, and electronic equipment for determining the remaining service life of an electric submersible pump, which can be well adapted to the complex operating conditions of electric submersible pumps and accurately predict the remaining service life of the pumps.

[0006] This manual provides a method for determining the remaining service life of an electric submersible pump, including:

[0007] Obtain the operating parameters of the target electric submersible pump, as well as the environmental parameters of the operating environment during operation;

[0008] Based on the operating parameters and environmental parameters, construct target time series data for the target electric submersible pump;

[0009] The target time series data is processed using a preset lifetime prediction model to obtain the corresponding target prediction result; wherein, the preset lifetime prediction model includes at least a master prediction model; the master prediction model is a hybrid model based on a Wavelet-LSTM structure; the target prediction result includes at least the predicted value of the remaining lifetime of the target electric submersible pump, and the uncertainty parameter of the predicted value of the remaining lifetime.

[0010] Based on the target prediction results, the remaining service life of the target electric submersible pump is determined.

[0011] In one embodiment, the main prediction model comprises: a multi-scale decomposition layer, a first feature processing layer, a second feature processing layer, a third feature processing layer, a feature fusion layer, and a fully connected output layer;

[0012] The multi-scale decomposition layer is a model layer based on a Wavelet Transform structure, and the first feature processing layer, the second feature processing layer, and the third feature processing layer are model layers based on an LSTM structure.

[0013] The multi-scale decomposition layer is connected to the first feature processing layer through a first branch and connected to the second feature processing layer through a second branch; the first feature processing layer, the second feature processing layer, and the third feature processing layer are also connected to the feature fusion layer respectively; and the feature fusion layer is connected to the fully connected output layer.

[0014] In one embodiment, the processing of the target time series data by using the preset life prediction model to obtain a corresponding prediction result comprises:

[0015] The multi-scale decomposition layer is used to process the target time series data to decompose corresponding low-frequency component data and high-frequency component data;

[0016] The first feature processing layer is used to process the low-frequency component data to obtain long-term dependence features; the second feature processing layer is used to process the high-frequency component data to obtain short-term fluctuation features; and the third feature processing layer is used to process the target time series data to obtain comprehensive working condition features;

[0017] The feature fusion layer is used to dynamically weight and fuse the long-term dependence features, the short-term fluctuation features, and the comprehensive working condition features to obtain corresponding target fusion features;

[0018] The fully connected output layer is used to process the target fusion features to determine and output corresponding target prediction results.

[0019] In one embodiment, the fully connected output layer comprises: a mapping module, a dimension reduction module, and a prediction module; the mapping module is connected to the feature fusion layer and the dimension reduction module respectively, and the prediction module is connected to the dimension reduction module.

[0020] Correspondingly, the processing of the target fusion features by using the fully connected output layer to determine and output corresponding target prediction results comprises:

[0021] The mapping module is used to map the target fusion features into corresponding target intermediate features;

[0022] The dimension reduction module is used to perform dimension reduction operation on the target intermediate features to obtain dimension-reduced intermediate features.

[0023] The intermediate features after dimensionality reduction are processed by the prediction module to determine and output the first prediction sub-result; wherein the first prediction sub-result is used to indicate the predicted value of the remaining service life of the target electric submersible pump.

[0024] In one embodiment, the fully connected output layer further includes a reliability analysis module; wherein the reliability analysis module is connected to the prediction module and the second feature processing layer respectively;

[0025] Accordingly, the method further includes using a fully connected output layer to process the target fusion features, determine and output the corresponding target prediction result, and then using this method to determine and output the target prediction result.

[0026] The reliability analysis module determines and outputs a second prediction sub-result based on the first prediction sub-result and the short-term fluctuation characteristics; wherein the second prediction sub-result is used to indicate the uncertainty parameter of the predicted value of the remaining service life of the target electric submersible pump.

[0027] In one embodiment, the preset lifetime prediction model further includes an auxiliary prediction model and a joint prediction model; the joint prediction model is connected to the output of the auxiliary prediction model and the output of the main prediction model, respectively; wherein the auxiliary prediction model and the main prediction model are models with different structures, and the joint prediction model is used to receive and determine and output the corresponding target prediction result based on the output results of the main prediction model and the output results of the auxiliary prediction model.

[0028] In one embodiment, the auxiliary prediction model includes at least one of the following: a first auxiliary sub-model based on a CNN-LSTM structure, a second auxiliary sub-model based on a GRU structure, a third auxiliary sub-model based on a Bi-GRU-LSTM structure, and a fourth auxiliary sub-model based on a Transformer structure.

[0029] In one embodiment, determining the remaining service life of the target electric submersible pump based on the target prediction result includes:

[0030] Based on the uncertainty parameters, the predicted value of the remaining service life of the target electric submersible pump is corrected to determine the remaining service life of the target electric submersible pump.

[0031] In one embodiment, the operating parameters include at least one of the following: current, voltage, vibration amplitude, and vibration frequency.

[0032] In one embodiment, the environmental parameters include at least one of the following: temperature, pressure, sand content, and corrosion parameters.

[0033] This manual also provides a method for determining the remaining service life of an electric submersible pump, including:

[0034] Obtain the operating parameters of the target electric submersible pump, as well as the environmental parameters of the operating environment during operation;

[0035] Based on the operating parameters and environmental parameters, construct target time series data for the target electric submersible pump;

[0036] The target time series data is processed using a preset lifetime prediction model to obtain the corresponding target prediction result; wherein, the preset lifetime prediction model includes a main prediction model and an auxiliary prediction model; the main prediction model is a hybrid model based on a Wavelet-LSTM structure; the model structure of the auxiliary prediction model is different from that of the main prediction model; the target prediction result includes at least the predicted value of the remaining lifetime of the target electric submersible pump, and the uncertainty parameter of the predicted value of the remaining lifetime.

[0037] Based on the target prediction results, the remaining service life of the target electric submersible pump is determined.

[0038] This manual also provides a device for determining the remaining service life of an electric submersible pump, comprising:

[0039] The acquisition module is used to acquire the operating parameters of the target electric submersible pump, as well as the environmental parameters of the operating environment during operation.

[0040] The construction module is used to construct target time series data about the target electric submersible pump based on the operating parameters and environmental parameters.

[0041] The prediction module is used to process the target time series data using a preset lifetime prediction model to obtain the corresponding target prediction result; wherein, the preset lifetime prediction model includes at least a master prediction model; the master prediction model is a hybrid model based on a Wavelet-LSTM structure; the target prediction result includes at least the predicted value of the remaining lifetime of the target electric submersible pump, and the uncertainty parameter of the predicted value of the remaining lifetime.

[0042] The determination module is used to determine the remaining service life of the target electric submersible pump based on the target prediction results.

[0043] This specification also provides an electronic device, including a processor and a memory for storing processor-executable instructions, wherein the processor executes the instructions to implement the steps of a method for determining the remaining service life of the electric submersible pump.

[0044] This specification also provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of the method for determining the remaining service life of the electric submersible pump.

[0045] This specification also provides a computer program product comprising a computer program that, when executed by a processor, implements the steps of a method for determining the remaining service life of the electric submersible pump.

[0046] Based on the method, apparatus, and electronic equipment for determining the remaining service life of an electric submersible pump (EDP) provided in this specification, before implementation, a pre-set service life prediction model can be trained in advance using deep learning. This model includes at least a master prediction model based on a Wavelet-LSTM structure. The master prediction model includes at least a multi-scale decomposition layer, a first feature processing layer, a second feature processing layer, a third feature processing layer, and a feature fusion layer. In practice, firstly, based on the obtained operating parameters of the target EDP and the environmental parameters of the operating environment in which the target EDP operates, corresponding target time series data can be constructed. Then, the pre-set service life prediction model can be used to process the target time series data, simultaneously determining the predicted value of the remaining service life of the target EDP and the uncertainty parameters related to this predicted value, which are used as the target prediction result. Finally, based on this target prediction result, the remaining service life of the target EDP is determined. By fully utilizing the structural characteristics of the master prediction model based on Wavelet-LSTM, long-term dependency features and short-term fluctuation features can be jointly extracted and utilized to accurately determine the predicted value of the remaining service life of the target EV pump. Simultaneously, by effectively utilizing the data characteristics of short-term fluctuation features, the uncertainty parameters regarding this predicted remaining service life can also be accurately determined. This approach is well-suited to the complex operating conditions of EV pumps, accurately predicting their remaining service life, and enabling precise maintenance and upkeep of the EV pumps, extending their service life and ensuring the stability and safety of oil and gas field lifting operations. Attached Figure Description

[0047] To more clearly illustrate the embodiments of this specification, the accompanying drawings used in the embodiments will be briefly introduced below. The drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart illustrating a method for determining the remaining service life of an electric submersible pump, provided in one embodiment of this specification.

[0049] Figure 2 This is a schematic diagram of one embodiment of the method for determining the remaining service life of an electric submersible pump provided in the embodiments of this specification, applied in a scenario example.

[0050] Figure 3This is a schematic diagram of one embodiment of the method for determining the remaining service life of an electric submersible pump provided in the embodiments of this specification, applied in a scenario example.

[0051] Figure 4 This is a schematic diagram of one embodiment of the method for determining the remaining service life of an electric submersible pump provided in the embodiments of this specification, applied in a scenario example.

[0052] Figure 5 This is a schematic diagram of one embodiment of the method for determining the remaining service life of an electric submersible pump provided in the embodiments of this specification, applied in a scenario example.

[0053] Figure 6 This is a schematic diagram of one embodiment of the method for determining the remaining service life of an electric submersible pump provided in the embodiments of this specification, applied in a scenario example.

[0054] Figure 7 This is a schematic diagram of the structural composition of an electronic device provided in one embodiment of this specification;

[0055] Figure 8 This is a schematic diagram of the structural composition of a device for determining the remaining service life of an electric submersible pump, provided in one embodiment of this specification.

[0056] Figure 9 This is a schematic diagram of one embodiment of the method for determining the remaining service life of an electric submersible pump provided in the embodiments of this specification, applied in a scenario example.

[0057] Figure 10 This is a schematic diagram of one embodiment of the method for determining the remaining service life of an electric submersible pump provided in the embodiments of this specification, applied in a scenario example.

[0058] Figure 11 This is a schematic diagram illustrating one embodiment of the method for determining the remaining service life of an electric submersible pump provided in the embodiments of this specification, applied in a scenario example. Detailed Implementation

[0059] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0060] It should be noted that the information and data related to users involved in the embodiments of this specification are all information and data authorized by the user or fully authorized by the relevant parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of the relevant data all comply with relevant laws, regulations, and standards, and necessary confidentiality measures have been taken. They do not violate public order and good morals, and corresponding operation entry points are provided for users or relevant parties to choose to authorize or refuse.

[0061] It should also be noted that in the embodiments of this specification, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0062] See Figure 1 As shown in the embodiments of this specification, a method for determining the remaining service life of an electric submersible pump is provided. Specifically, this method may include the following:

[0063] S101: Obtain the operating parameters of the target electric submersible pump, as well as the environmental parameters of the operating environment during operation;

[0064] S102: Based on the operating parameters and environmental parameters, construct target time series data for the target electric submersible pump;

[0065] S103: Process the target time series data using a preset lifetime prediction model to obtain the corresponding target prediction result; wherein, the preset lifetime prediction model includes at least a master prediction model; the master prediction model is a hybrid model based on a Wavelet-LSTM structure; the target prediction result includes at least the predicted value of the remaining lifetime of the target electric submersible pump, and the uncertainty parameter of the predicted value of the remaining lifetime.

[0066] S104: Based on the target prediction results, determine the remaining service life of the target electric submersible pump.

[0067] Specifically, the aforementioned target electric submersible pumps can be understood as electric submersible pumps currently in operation for lift operations in oil and gas fields. Specifically, these target electric submersible pumps can include those used in oil and gas production wells, or those used in water injection systems.

[0068] The aforementioned operating parameters can be understood as parameter data used to reflect the operating status of the target electric submersible pump. Specifically, the aforementioned operating parameters may include at least one of the following: current, voltage, vibration amplitude, vibration frequency, etc.

[0069] The aforementioned environmental parameters can be understood as parameter data reflecting the operating environment of the target electric submersible pump. Specifically, the aforementioned environmental parameters may include at least one of the following: temperature, pressure, sand content, corrosion parameters, etc.

[0070] It should be noted that the operating parameters and environmental parameters listed above are only illustrative. In actual implementation, other types of operating parameters and environmental parameters may be included depending on the specific circumstances and processing requirements. This manual does not limit this.

[0071] In specific implementation, the acquisition of the target electric submersible pump's operating parameters and the environmental parameters of the operating environment during operation may include: acquiring parameter data sets of multiple consecutive time points within the observation period of the target electric submersible pump; wherein, each parameter data set corresponds to a time point, and each parameter data set contains the operating parameters and environmental parameters collected at the corresponding time point.

[0072] The observation period mentioned above can be a period of time in history, or it can be the current period of time that includes the current time point, such as the most recent hour.

[0073] The above-mentioned acquisition of the target electric submersible pump's operating parameters and the environmental parameters of the operating environment can be specifically implemented by: collecting operating parameters and environmental parameters for the observation period through multi-dimensional sensors deployed in the target electric submersible pump's working area.

[0074] The above-mentioned construction of target time series data about the target electric submersible pump based on the operating parameters and environmental parameters may include: arranging and splicing parameter data groups of multiple time points within the observation period in chronological order; and setting corresponding timestamps on each parameter data group to obtain target time series data about the target electric submersible pump.

[0075] The aforementioned preset lifespan prediction model can be understood as a hybrid model that is pre-trained through deep learning and can predict the remaining lifespan of the electric submersible pump based on the input time series data about the pump, while determining and outputting the uncertainty parameters corresponding to the preset remaining lifespan.

[0076] Specifically, the aforementioned master prediction models include at least a master prediction model based on the Wavelet-LSTM structure.

[0077] Specifically, the Wavelet (or Wavelet Transform) mentioned above can be understood as a model structure that performs multi-scale decomposition of data using wavelet functions. Based on Wavelet, multi-scale refinement operations can be performed on the data through scaling and translation operations, including: time subdivision at high frequencies, frequency subdivision at low frequencies, etc.

[0078] The aforementioned LSTM can be understood as a type of Long Short-Term Memory network, specifically a type of time-recurrent neural network. Based on LSTM, the temporal dependency features of data can be extracted and utilized for prediction.

[0079] The Wavelet-LSTM structure described above can be understood as a hybrid structure that simultaneously couples two different model structures: Wavelet and LSTM, i.e., a hybrid model of wavelet transform and long short-term memory network.

[0080] In this embodiment, the Wavelet-LSTM structure described above is introduced and used to construct the master prediction model to process time series data about electric submersible pumps, which can be well adapted to the complex and ever-changing operating conditions of electric submersible pumps.

[0081] Based on the above Wavelet-LSTM structure, by making full use of the structural characteristics of the two different model structures, Wavelet and LSTM, we can first use the Wavelet structure to finely decompose the time series data to extract low-frequency component data that can specifically reflect the continuous degradation trend of electric submersible pump performance due to long-term use (e.g., long-term temperature drift of electric submersible pump), and high-frequency component data that can specifically reflect the instantaneous changes caused by short-term influencing factors of the electric submersible pump itself or external factors under complex operating conditions (e.g., sudden vibration, sudden fluctuation of sand content).

[0082] It should be noted that the aforementioned low-frequency component data (e.g., Approximation) often exhibits long-term dependence on time, characterized by high determinism and low uncertainty. In contrast, the aforementioned high-frequency component data (e.g., Detail) often exhibits short-term volatility on time, characterized by strong randomness and high uncertainty.

[0083] Based on the aforementioned characteristics of high-frequency and low-frequency component data, different LSTM structures can be used to process the low-frequency and high-frequency component data independently. This allows for the extraction of long-term dependency features and short-term fluctuation features based on different dimensions, respectively. Subsequently, by combining these two different features (long-term dependency and short-term fluctuation) across multiple dimensions, the remaining service life of the electric submersible pump can be predicted more accurately and comprehensively. Furthermore, by utilizing the data characteristics of short-term fluctuation features and specifically considering their contribution to the predicted remaining service life, the uncertainty parameters regarding the predicted remaining service life can be precisely determined quantitatively. Accordingly, the prediction result output by the aforementioned pre-defined service life prediction model can simultaneously include: the predicted value of the remaining service life, and the uncertainty parameters regarding that predicted value.

[0084] Specifically, the aforementioned master prediction model may include at least the following structures: multi-scale decomposition layer, first feature processing layer, second feature processing layer, and feature fusion layer.

[0085] The aforementioned multi-scale decomposition layer is a module based on the Wavelet structure, which can be used to decompose the input time series data into at least high-frequency component data and low-frequency component data.

[0086] The first and second feature processing layers described above are two independent modules based on an LSTM structure. Specifically, the first feature processing layer can be used to process low-frequency component data to extract long-term dependency features. The second feature processing layer can be used to process high-frequency component data to extract short-term fluctuation features. Then, by using the master prediction model in combination with at least the long-term dependency features and short-term fluctuation features, the predicted value of the remaining useful life can be determined; simultaneously, the uncertainty parameters corresponding to the predicted value of the remaining useful life can be determined.

[0087] The aforementioned target prediction results include at least a predicted value for the remaining service life of the target electric submersible pump, and an uncertainty parameter regarding that predicted service life. This uncertainty parameter is used to characterize the reliability of the associated service life prediction based on a dynamically changing dimension.

[0088] In practice, the remaining service life of the target electric submersible pump can be determined by simultaneously using two data points: the predicted value of the remaining service life in the target prediction results and the uncertainty parameter regarding the predicted value of the remaining service life.

[0089] Based on the above embodiments, by fully utilizing the structural characteristics of the master prediction model based on the Wavelet-LSTM structure, and jointly extracting and utilizing long-term dependency features and short-term fluctuation features, the predicted value of the remaining service life of the target electric submersible pump is accurately determined. Simultaneously, by utilizing the short-term fluctuation features, the uncertainty parameters regarding the predicted value of this remaining service life are accurately determined. This allows for better adaptation to the complex operating conditions of electric submersible pumps and accurate prediction of their remaining service life.

[0090] In some embodiments, see Figure 2 As shown, the main prediction model may specifically include: a multi-scale decomposition layer, a first feature processing layer, a second feature processing layer, a third feature processing layer, a feature fusion layer, and a fully connected output layer;

[0091] The multi-scale decomposition layer is a model layer based on the Wavelet Transform structure, and the first feature processing layer, the second feature processing layer, and the third feature processing layer are model layers based on the LSTM structure.

[0092] The multi-scale decomposition layer is connected to the first feature processing layer via a first branch and to the second feature processing layer via a second branch; the first feature processing layer, the second feature processing layer, and the third feature processing layer are also connected to the feature fusion layer; the feature fusion layer is connected to the fully connected output layer.

[0093] The first feature processing layer, the second feature processing layer, and the third feature processing layer are independent of each other. The first feature processing layer is specifically used to process the low-frequency component data decomposed from the target time series data based on the multi-scale decomposition layer; the second feature processing layer is specifically used to process the high-frequency component data decomposed from the target time series data based on the multi-scale decomposition layer; and the third feature processing layer is specifically used to process the original target time series data that contains both high-frequency and low-frequency component data.

[0094] Specifically, the aforementioned multi-scale decomposition layer can process the received target time series data to decompose it into low-frequency component data and high-frequency component data; the low-frequency component data is sent to the first feature processing layer through the first branch; at the same time, the high-frequency component data is sent to the second feature processing layer through the second branch.

[0095] Furthermore, the aforementioned third feature processing layer can be connected to the multi-scale decomposition layer via the third branch. The multi-scale decomposition layer can send the received target time series data as the raw signal directly to the third feature processing layer via the third branch without processing.

[0096] In some embodiments, see Figure 3As shown, the above-described method uses a preset lifetime prediction model to process the target time series data and obtain the corresponding prediction results. In specific implementation, this may include the following:

[0097] S1: The target time series data is processed using a multi-scale decomposition layer to decompose the corresponding low-frequency component data and high-frequency component data;

[0098] S2: The low-frequency component data is processed using the first feature processing layer to obtain long-term dependency features; the high-frequency component data is processed using the second feature processing layer to obtain short-term fluctuation features; and the target time series data is processed using the third feature processing layer to obtain comprehensive operating condition features.

[0099] S3: The feature fusion layer is used to dynamically weight and fuse the long-term dependency features, short-term fluctuation features, and comprehensive operating condition features to obtain the corresponding target fused features;

[0100] S4: Utilize the fully connected output layer to process the target fusion features, determine and output the corresponding target prediction results.

[0101] Specifically, the aforementioned low-frequency component data can be data elements in the target time series data that change relatively slowly and have relatively long periods. For example, signal data reflecting the gradual degradation of equipment performance, such as the slow rise in temperature and stable decrease in pressure caused by component aging during long-term operation of an electric submersible pump.

[0102] The aforementioned high-frequency component data can specifically refer to data components in the target time series data that exhibit relatively drastic changes and relatively short periods. For example, signal data reflecting sudden interference or random noise, such as instantaneous vibrations caused by fluctuations in sand content or voltage spikes caused by power grid fluctuations.

[0103] In practice, the time-frequency localization characteristics of the multi-scale decomposition layer based on the wavelet transform function can be utilized to analyze and decompose the target time series data at different time and frequency resolutions to obtain low-frequency component data and high-frequency component data. Then, the low-frequency component data is sent to the first feature processing layer through the first branch, and the high-frequency component data is sent to the second feature processing layer through the second branch. At the same time, the original target time series data is directly sent to the third feature processing layer through the third branch.

[0104] The aforementioned first feature processing layer (or LSTM layer 1) adjusts network parameters such as the number of memory units and the weight of the forget gate, and combines deep learning specifically for low-frequency component data based on long-term dimensions. It focuses more on capturing long-term time-series dependencies and is good at processing slowly changing signal data such as "monthly" or "quarterly" data, so as to accurately extract long-term degradation trend features with better performance.

[0105] The aforementioned second feature processing layer (or LSTM layer 2) adjusts network parameters such as shortening the time step and adjusting the input gate weights. Combined with deep learning specifically designed for high-frequency component data based on short time dimensions, it focuses more on responding to short-term, burst signals and is good at processing rapidly changing signal data such as "minute-level" or "hour-level" signals to accurately extract short-term abnormal fluctuation features with better performance.

[0106] The aforementioned third feature processing layer (or LSTM layer 3) combines deep learning specifically designed for original signal data that simultaneously contains both short time periods and short time intervals by setting and adjusting relevant network parameters. This achieves a relatively balanced ability to capture both long and short time series signal data, preserving the memory of long-term trends while also responding to short-term anomalies. It avoids one-sided feature extraction due to an emphasis on a single dimension, thus accurately extracting comprehensive working condition features with better results.

[0107] In practice, the long-term dependency features extracted by the first feature processing layer, the short-term fluctuation features extracted by the second feature processing layer, and the comprehensive operating condition features extracted by the third feature processing layer can be fed into the feature fusion layer. Then, the feature fusion layer can be used to linearly combine the above three features based on different dimensions according to dynamic weights to obtain a high-dimensional target fusion feature, such as a 128-dimensional feature vector that contains information on different dimensions such as equipment degradation and abnormal fluctuations.

[0108] Furthermore, by utilizing the aforementioned fully connected layer to process the target fusion features, and by determining and combining the predicted value of the remaining service life of the target electric submersible pump, as well as the uncertainty parameters regarding the predicted value of the remaining service life, a target prediction result for the target electric submersible pump can be obtained.

[0109] Based on the above embodiments, the predicted value of the remaining service life of the electric submersible pump and the uncertainty parameters of the predicted value of the remaining service life can be accurately determined simultaneously by utilizing the multiple model layer structure in the master prediction model, thus obtaining a better prediction result.

[0110] In some embodiments, the fully connected output layer may specifically include a mapping module, a dimensionality reduction module, and a prediction module, etc.; wherein the mapping module is connected to the feature fusion layer and the dimensionality reduction module respectively, and the prediction module is connected to the dimensionality reduction module.

[0111] Accordingly, see Figure 4 As shown, the above-described fully connected output layer processes the target fusion features to determine and output the corresponding target prediction result. In specific implementations, this may include the following:

[0112] S1: The target fusion features are mapped into corresponding target intermediate features using the mapping module;

[0113] S2: Use the dimensionality reduction module to perform dimensionality reduction on the target intermediate features to obtain the dimensionality-reduced intermediate features;

[0114] S3: Use the prediction module to process the intermediate features after dimensionality reduction, determine and output the first prediction sub-result; wherein, the first prediction sub-result is used to indicate the predicted value of the remaining service life of the target electric submersible pump.

[0115] In practice, the mapping module can be used to perform linear calculations on the target fusion features using the pre-trained preset weight matrix and bias terms, transforming the originally relatively abstract target fusion features into target intermediate features that are closely related to the remaining service life of the electric submersible pump.

[0116] The aforementioned dimensionality reduction module may include: a first sub-fully connected layer and a second sub-fully connected layer.

[0117] Accordingly, in practical implementation, the first fully connected layer in the dimensionality reduction module can be used to perform a first compression operation on the target intermediate features to obtain the corresponding compressed intermediate features. For example, the first fully connected layer can be used to compress the 128-dimensional target intermediate features to 32 dimensions, while retaining feature information closely related to the remaining service life of the electric submersible pump, such as temperature drift and vibration peaks. Then, the second fully connected layer can be used to perform a second compression operation based on the compressed features to obtain the dimensionality-reduced intermediate features that meet the requirements. For example, the second fully connected layer can be used to further compress the 32-dimensional compressed intermediate features to 1 dimension.

[0118] In this way, the structural characteristics of the fully connected layer in the dimensionality reduction module can be fully utilized. By performing two compression operations in sequence, the feature dimension can be effectively reduced, redundant information in the intermediate features of the target can be eliminated, and the amount of data processing during subsequent prediction can be reduced.

[0119] In specific implementation, the prediction module can be used to determine and output the predicted value of the remaining service life of the target electric submersible pump as the first prediction sub-result by processing the intermediate features after dimensionality reduction; on the other hand, it can also determine and output the uncertainty parameter of the predicted value of the remaining service life as the second prediction sub-result, thereby obtaining and outputting the target prediction result that includes both the first and second prediction sub-results.

[0120] In some embodiments, the fully connected output layer may further include: a reliability analysis module; wherein the reliability analysis module is connected to the prediction module and the second feature processing layer respectively;

[0121] Accordingly, the above-mentioned use of a fully connected output layer to process the target fusion features, determine and output the corresponding target prediction results, may further include: using a reliability analysis module to determine and output a second prediction sub-result based on the first prediction sub-result and the short-term fluctuation features; wherein the second prediction sub-result is used to indicate the uncertainty parameter of the predicted value of the remaining service life of the target electric submersible pump.

[0122] Based on the above embodiments, the reliability analysis module in the fully connected output layer can be used to analyze and determine the contribution value of short-term fluctuation characteristics to the first prediction sub-result, and in a quantitative manner, accurately determine the uncertainty parameter of the predicted value of the remaining useful life as the second prediction sub-result.

[0123] In some embodiments, see Figure 5 As shown, the preset lifetime prediction model may further include an auxiliary prediction model and a joint prediction model; the joint prediction model is connected to the output of the auxiliary prediction model and the output of the main prediction model, respectively; wherein, the auxiliary prediction model and the main prediction model are models with different structures, and the joint prediction model is used to receive and determine and output the corresponding target prediction result based on the output results of the main prediction model and the output results of the auxiliary prediction model.

[0124] Accordingly, in specific implementation, the prediction model can be combined with the prediction results of the main prediction model and the auxiliary prediction model at the same time, and different prediction mechanisms can be integrated to obtain a more comprehensive and accurate prediction result that meets the requirements.

[0125] In some embodiments, the auxiliary prediction model includes at least one of the following: a first auxiliary sub-model based on a CNN-LSTM structure, a second auxiliary sub-model based on a GRU structure, a third auxiliary sub-model based on a Bi-GRU-LSTM structure, a fourth auxiliary sub-model based on a Transformer structure, etc.

[0126] It should be noted that the above-mentioned auxiliary prediction model may include one of the above-mentioned auxiliary sub-models, or it may include a parallel combination of the above-mentioned auxiliary sub-models.

[0127] Specifically, the aforementioned CNN (Convolutional Neural Network) can be understood as a type of feedforward neural network, which has relatively excellent local feature extraction capabilities.

[0128] The GRU (Gated Recurrent Unit) mentioned above can be understood as a type of neural network unit with a gating mechanism (or gated recurrent unit), specifically designed for processing sequential data and suitable for modeling tasks with long-term dependencies. The GRU can also be understood as a simplified version of the LSTM model; because it only contains two control units—a reset gate and an update gate—its model structure is relatively simpler and its training efficiency is relatively higher.

[0129] The Transformer described above can be understood as a network structure that uses a Self-Attention structure to replace the RNN network commonly used in NLP tasks.

[0130] The aforementioned CNN-LSTM structure, namely the fusion model of convolutional neural network and long short-term memory network, can effectively extract and utilize local temporal features, such as short-term stress fluctuation features.

[0131] The GRU structure described above can focus on capturing and utilizing long-term temporal dependency features, such as monthly temperature change trends.

[0132] The aforementioned Bi-GRU-LSTM structure, namely the fusion model of bidirectional gated recurrent units and long short-term memory networks, is adept at capturing and utilizing relatively complex long-term temporal dependency features.

[0133] The Transformer structure described above can learn the changes between time points by tracking the relationships in sequence data, and efficiently capture and utilize temporal features in a parallel manner.

[0134] In this way, the aforementioned auxiliary prediction model, in conjunction with the main prediction model, can be used to predict the remaining service life of the target electric submersible pump based on different prediction mechanisms and dimensions, obtaining prediction results characterized by different dimensions. Furthermore, a joint prediction model can be used to simultaneously integrate the prediction results from multiple dimensions, fully considering multiple dimensions and various influencing factors, to accurately determine the predicted value of the target electric submersible pump's remaining service life. Simultaneously, by calculating and based on the differences between multiple different prediction results and the differences between multiple different prediction results and the predicted value of the remaining service life, uncertainty parameters corresponding to the predicted value of the remaining service life can be determined, thereby obtaining a more accurate target prediction result.

[0135] In some embodiments, the reliability analysis module determines and outputs a second prediction sub-result based on the first prediction sub-result and the short-term fluctuation characteristics. In specific implementations, this may include the following:

[0136] S1: Using the reliability analysis module, determine the contribution ratio of the short-term fluctuation characteristic to the first prediction result based on the first prediction result and the short-term fluctuation characteristic;

[0137] S2: Determine and output the corresponding second prediction sub-result based on the contribution ratio of the short-term fluctuation characteristics to the first prediction sub-result.

[0138] In some embodiments, determining the remaining service life of the target electric submersible pump based on the target prediction result may specifically include the following: correcting the predicted value of the remaining service life of the target electric submersible pump based on uncertainty parameters to determine the remaining service life of the target electric submersible pump.

[0139] In practice, it can be determined whether the uncertainty parameter is greater than or equal to a preset first probability threshold. When it is determined to be greater than or equal to the preset first probability threshold, the predicted value of the remaining service life output by the model can be directly determined as the remaining service life of the target electric submersible pump.

[0140] When the probability is determined to be less than a preset second probability threshold, it is checked whether the probability is greater than or equal to the preset second probability threshold. When the probability is determined to be greater than or equal to the preset second probability threshold, a feature deviation vector can be obtained and calculated based on short-term fluctuation characteristics and long-term dependence characteristics. Then, based on the feature deviation vector, the predicted value of the remaining service life output by the model is modified, and the modified predicted value is determined as the remaining service life of the target electric submersible pump.

[0141] When the probability is determined to be less than the preset second probability threshold, an error message can be generated; and the target prediction result and the error message can be sent to the user to switch to manual judgment.

[0142] In some embodiments, the operating parameters may specifically include at least one of the following: current, voltage, vibration amplitude, vibration frequency, etc. It should be noted that the operating parameters listed above are merely illustrative. In specific implementations, other types of operating parameters may be included depending on the specific circumstances and processing requirements. This specification does not limit this.

[0143] In some embodiments, the environmental parameters may specifically include at least one of the following: temperature, pressure, sand content, corrosion parameters, etc. The corrosion parameters may include the pH of the well fluid, etc. It should be noted that the environmental parameters listed above are merely illustrative. In specific implementations, other types of environmental parameters may be included depending on the specific circumstances and treatment requirements. This specification does not limit this.

[0144] In some embodiments, after obtaining the operating parameters of the target electric submersible pump and the environmental parameters of the operating environment, the method may further include: performing preprocessing such as data cleaning, normalization, and sliding window division on the operating parameters and environmental parameters to obtain high-quality operating parameters and environmental parameters with small errors that meet the requirements.

[0145] In some embodiments, after determining the remaining service life of the target electric submersible pump based on the target prediction results, the method may further include the following:

[0146] Based on the uncertainty parameters and the remaining service life of the target electric submersible pump, a target operation and maintenance strategy matching the target electric submersible pump is determined; wherein, the target operation and maintenance strategy includes the maintenance cycle, maintenance method and maintenance frequency for the target electric submersible pump.

[0147] Specifically, based on the uncertainty parameters and the remaining service life of the target electric submersible pump, a preset strategy set can be queried to determine a matching preset operation and maintenance strategy as the target operation and maintenance strategy. The preset strategy set stores multiple preset operation and maintenance strategies, each corresponding to at least one combination of the numerical range of the uncertainty parameter and the numerical range of the remaining service life.

[0148] Before implementation, a large number of historical operation and maintenance records can be collected; then, by performing clustering learning on the above historical operation records, multiple preset operation and maintenance strategies can be determined; and by combining the above multiple preset operation and maintenance strategies, a preset strategy set can be constructed.

[0149] In some embodiments, see Figure 6 As shown, in specific implementations, the method may also include the following:

[0150] S1: Obtain and construct sample data based on the operation and maintenance records of the sample electric submersible pumps in the sample wells;

[0151] S2: Construct an initial lifetime prediction model; wherein the initial lifetime prediction model includes at least an initial master prediction model; the initial master prediction model is a hybrid model based on a Wavelet-LSTM structure;

[0152] S3: Use the sample data and the initial life prediction model to perform deep learning to obtain a preset life prediction model that meets the requirements.

[0153] Based on the above embodiments, a preset life prediction model that meets the requirements can be trained in advance using sample data and deep learning.

[0154] As can be seen from the above, based on the method for determining the remaining service life of an electric submersible pump provided in the embodiments of this specification, before specific implementation, a preset service life prediction model can be obtained in advance through deep learning training, including at least a master prediction model based on a Wavelet-LSTM structure. The master prediction model includes at least: a multi-scale decomposition layer, a first feature processing layer, a second feature processing layer, a third feature processing layer, and a feature fusion layer. In specific implementation, the target time series data can be constructed first based on the obtained operating parameters of the target electric submersible pump and the environmental parameters of the operating environment. Then, the preset service life prediction model is used to process the target time series data, simultaneously determining the predicted value of the remaining service life of the target electric submersible pump and the uncertainty parameters regarding the predicted value of the remaining service life, as the target prediction result. Based on the target prediction result, the remaining service life of the target electric submersible pump is determined. By fully utilizing the structural characteristics of the master prediction model based on a Wavelet-LSTM structure, long-term dependency features and short-term fluctuation features can be jointly extracted and utilized to accurately determine the predicted value of the remaining service life of the target electric submersible pump; simultaneously, the uncertainty parameters regarding the predicted value of the remaining service life can be accurately determined using short-term fluctuation features. This allows it to better adapt to the complex operating conditions of electric submersible pumps, accurately predict the remaining service life of the pumps, and thus enable precise operation and maintenance, extending the service life of the pumps and ensuring the stability and safety of oil and gas field lifting operations.

[0155] This specification also provides another method for determining the remaining service life of an electric submersible pump, which may include the following in specific implementation:

[0156] S1: Obtain the operating parameters of the target electric submersible pump, as well as the environmental parameters of the operating environment during operation;

[0157] S2: Based on the operating parameters and environmental parameters, construct target time series data for the target electric submersible pump;

[0158] S3: Process the target time series data using a preset lifetime prediction model to obtain the corresponding target prediction result; wherein, the preset lifetime prediction model includes a main prediction model and an auxiliary prediction model; the main prediction model is a hybrid model based on a Wavelet-LSTM structure; the model structure of the auxiliary prediction model is different from that of the main prediction model; the target prediction result includes at least the predicted value of the remaining lifetime of the target electric submersible pump, and the uncertainty parameters of the predicted value of the remaining lifetime.

[0159] S4: Based on the target prediction results, determine the remaining service life of the target electric submersible pump.

[0160] Specifically, the aforementioned auxiliary prediction model may include a combination of one or more of the following auxiliary sub-models: a first auxiliary sub-model based on the CNN-LSTM structure, a second auxiliary sub-model based on the GRU structure, a third auxiliary sub-model based on the Bi-GRU-LSTM structure, and a fourth auxiliary sub-model based on the Transformer structure.

[0161] For details, please refer to Figure 5 As shown, the aforementioned preset lifetime prediction model also includes a joint model. This joint model is connected to both the main prediction model and the auxiliary prediction model.

[0162] In practice, a joint model can be used to simultaneously receive the prediction results from the main prediction model and the auxiliary prediction model; then, based on the model characteristics of different models, multiple prediction results can be weighted and fused to generate and output the target prediction result.

[0163] The prediction results output by the main prediction model include both the predicted value of the remaining useful life and the uncertainty parameter of the predicted value of the remaining useful life; the prediction results output by the auxiliary prediction model may only include the predicted value of the remaining useful life.

[0164] When using a joint model to fuse multiple prediction results to generate a target prediction result, on the one hand, the joint model can be used to dynamically weight the predicted values ​​of remaining useful life in the prediction results output by the main prediction model and the auxiliary prediction model, combined with dynamic weighting coefficients, to determine the final predicted value of remaining useful life.

[0165] The aforementioned dynamic weighting coefficients are determined based on preset base coefficients and the scene change characteristics of the target electric submersible pump's current operating condition. The preset base coefficients were determined through prior experimental testing. Specifically, when determining the dynamic weighting coefficients, based on the scene change characteristics, when the scene change of the target electric submersible pump's current operating condition is rapid and drastic, weighting coefficients corresponding to the main prediction model are added selectively on top of the preset base coefficients.

[0166] On the other hand, a joint model can be used to first calculate the difference between the predicted values ​​of the remaining useful life output by the primary prediction model and the auxiliary prediction model; then, based on the uncertainty parameters and the difference in predicted values ​​in the prediction results output by the primary prediction model, and combined with the structural difference values ​​between the primary and auxiliary prediction models, the final uncertainty parameters can be determined. Finally, by combining the aforementioned final predicted values ​​of the remaining useful life and the final uncertainty parameters, the final target prediction result can be generated and output.

[0167] Based on the above embodiments, by combining main prediction models and auxiliary prediction models with different structures, the complex operating conditions of electric submersible pumps can be better adapted, the remaining service life of electric submersible pumps can be accurately predicted, and the corresponding operation and maintenance of electric submersible pumps can be carried out in a precise manner, thereby extending the service life of electric submersible pumps and ensuring the stability and safety of oil and gas field lifting operations.

[0168] This specification provides an electronic device through its embodiments. (See attached document.) Figure 7 As shown. The electronic device includes a network communication port 701, a processor 702, and a memory 703. These structures are connected by internal cables so that they can perform specific data interaction.

[0169] Specifically, the network communication port 701 can be used to acquire the operating parameters of the target electric submersible pump, as well as the environmental parameters of the operating environment during operation.

[0170] The processor 702 can be specifically used to construct target time series data about the target electric submersible pump based on the operating parameters and environmental parameters; process the target time series data using a preset lifetime prediction model to obtain the corresponding target prediction result; wherein, the preset lifetime prediction model includes at least a master prediction model; the master prediction model is a hybrid model based on a Wavelet-LSTM structure; the target prediction result includes at least a predicted value of the remaining lifetime of the target electric submersible pump, and uncertainty parameters about the predicted value of the remaining lifetime; and determine the remaining lifetime of the target electric submersible pump based on the target prediction result.

[0171] The memory 703 can be used to store the corresponding instruction program and related intermediate data.

[0172] Based on the above method, the relevant structural performance of electronic equipment can be effectively utilized to improve the data processing speed of electronic equipment and efficiently realize the data processing for determining the remaining service life of electric submersible pumps.

[0173] In this embodiment, the network communication port 701 can be a virtual port bound to different communication protocols, thereby enabling the sending or receiving of different data. For example, the network communication port can be a port responsible for web data communication, a port responsible for FTP data communication, or a port responsible for email data communication. Furthermore, the network communication port can also be a physical communication interface or communication chip. For example, it can be a wireless mobile network communication chip, such as GSM or CDMA; it can also be a Wi-Fi chip; or it can be a Bluetooth chip.

[0174] In this embodiment, the processor 702 can be implemented in any suitable manner. For example, the processor can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers, etc. This specification is not limiting.

[0175] In this embodiment, the memory 703 may include multiple layers. In a digital system, anything that can store binary data can be a memory. In an integrated circuit, a circuit with storage function but no physical form is also called a memory, such as RAM, FIFO, etc. In a system, a storage device with a physical form is also called a memory, such as a memory stick, TF card, etc.

[0176] This specification also provides a computer-readable storage medium based on the above-described method for determining the remaining service life of an electric submersible pump. The computer-readable storage medium stores computer program instructions that, when executed, perform the following: acquiring the operating parameters of the target electric submersible pump and the environmental parameters of the operating environment; constructing target time series data about the target electric submersible pump based on the operating parameters and environmental parameters; processing the target time series data using a preset life prediction model to obtain a corresponding target prediction result; wherein the preset life prediction model includes at least a master prediction model; the master prediction model is a hybrid model based on a Wavelet-LSTM structure; the target prediction result includes at least a predicted value of the remaining service life of the target electric submersible pump and uncertainty parameters regarding the predicted value of the remaining service life; and determining the remaining service life of the target electric submersible pump based on the target prediction result.

[0177] In this embodiment, the storage medium includes, but is not limited to, Random Access Memory (RAM), Read-Only Memory (ROM), Cache, Hard Disk Drive (HDD), or Memory Card. The memory can be used to store computer program instructions. The network communication unit can be an interface configured according to standards specified in the communication protocol for network connection communication.

[0178] In this embodiment, the specific functions and effects implemented by the program instructions stored in the computer-readable storage medium can be explained in comparison with other embodiments, and will not be repeated here.

[0179] This specification also provides a computer program product, comprising at least a computer program, which, when executed by a processor, implements the following method steps: acquiring the operating parameters of a target electric submersible pump and the environmental parameters of the operating environment; constructing target time series data about the target electric submersible pump based on the operating parameters and environmental parameters; processing the target time series data using a preset lifetime prediction model to obtain a corresponding target prediction result; wherein the preset lifetime prediction model includes at least a master prediction model; the master prediction model is a hybrid model based on a Wavelet-LSTM structure; the target prediction result includes at least a predicted value of the remaining lifetime of the target electric submersible pump and an uncertainty parameter regarding the predicted value of the remaining lifetime; and determining the remaining lifetime of the target electric submersible pump based on the target prediction result.

[0180] This specification also provides another computer program product, which includes at least a computer program that, when executed by a processor, performs the following method steps: acquiring the operating parameters of a target electric submersible pump and the environmental parameters of the operating environment; constructing target time series data about the target electric submersible pump based on the operating parameters and environmental parameters; processing the target time series data using a preset lifetime prediction model to obtain a corresponding target prediction result; wherein the preset lifetime prediction model includes a main prediction model and an auxiliary prediction model; the main prediction model is a hybrid model based on a Wavelet-LSTM structure; the model structure of the auxiliary prediction model is different from that of the main prediction model; the target prediction result includes at least a predicted value of the remaining lifetime of the target electric submersible pump and an uncertainty parameter regarding the predicted value of the remaining lifetime; and determining the remaining lifetime of the target electric submersible pump based on the target prediction result.

[0181] See Figure 8 As shown in the embodiments of this specification, a device for determining the remaining service life of an electric submersible pump is also provided. This device may specifically include the following structural modules:

[0182] The acquisition module 801 can be used to acquire the operating parameters of the target electric submersible pump, as well as the environmental parameters of the operating environment during operation.

[0183] The construction module 802 can be specifically used to construct target time series data about the target electric submersible pump based on the operating parameters and environmental parameters.

[0184] The prediction module 803 is specifically used to process the target time series data using a preset lifetime prediction model to obtain the corresponding target prediction result; wherein, the preset lifetime prediction model includes at least a master prediction model; the master prediction model is a hybrid model based on a Wavelet-LSTM structure; the target prediction result includes at least the predicted value of the remaining lifetime of the target electric submersible pump, and the uncertainty parameter of the predicted value of the remaining lifetime.

[0185] The determination module 804 can be used to determine the remaining service life of the target electric submersible pump based on the target prediction results.

[0186] In some embodiments, the main prediction model may specifically include: a multi-scale decomposition layer, a first feature processing layer, a second feature processing layer, a third feature processing layer, a feature fusion layer, and a fully connected output layer, etc.

[0187] Specifically, the multi-scale decomposition layer can be a model layer based on the Wavelet Transform structure, and the first feature processing layer, the second feature processing layer, and the third feature processing layer can be model layers based on the LSTM structure.

[0188] Specifically, the multi-scale decomposition layer can be connected to the first feature processing layer via a first branch and to the second feature processing layer via a second branch; the first feature processing layer, the second feature processing layer, and the third feature processing layer are also connected to the feature fusion layer; the feature fusion layer is connected to the fully connected output layer.

[0189] In some embodiments, when the prediction module 803 is specifically implemented, it can process the target time series data using a preset lifetime prediction model in the following manner to obtain the corresponding prediction results: The target time series data is processed using a multi-scale decomposition layer to decompose corresponding low-frequency component data and high-frequency component data; the low-frequency component data is processed using a first feature processing layer to obtain long-term dependency features; the high-frequency component data is processed using a second feature processing layer to obtain short-term fluctuation features; the target time series data is processed using a third feature processing layer to obtain comprehensive operating condition features; the long-term dependency features, short-term fluctuation features, and comprehensive operating condition features are dynamically weighted and fused using a feature fusion layer to obtain corresponding target fusion features; and the target prediction results are determined and output using a fully connected output layer by processing the target fusion features.

[0190] In some embodiments, the fully connected output layer may specifically include: a mapping module, a dimensionality reduction module, and a prediction module; wherein, the mapping module may be connected to the feature fusion layer and the dimensionality reduction module respectively, and the prediction module may be connected to the dimensionality reduction module.

[0191] Accordingly, when the prediction module 803 is specifically implemented, it can use the fully connected output layer to process the target fusion features to determine and output the corresponding target prediction results in the following manner: using the mapping module to map the target fusion features into corresponding target intermediate features; using the dimensionality reduction module to perform dimensionality reduction operation on the target intermediate features to obtain dimensionality-reduced intermediate features; using the prediction module to process the dimensionality-reduced intermediate features to determine and output the first prediction sub-result; wherein, the first prediction sub-result is used to indicate the predicted value of the remaining service life of the target electric submersible pump.

[0192] In some embodiments, the fully connected output layer may further include: a reliability analysis module, etc.; wherein the reliability analysis module is connected to the prediction module and the second feature processing layer respectively;

[0193] Accordingly, when the prediction module 803 is specifically implemented, it can also use the fully connected output layer to process the target fusion features to determine and output the corresponding target prediction results in the following way: the reliability analysis module determines and outputs the second prediction sub-result based on the first prediction sub-result and the short-term fluctuation features; wherein the second prediction sub-result is used to indicate the uncertainty parameter of the predicted value of the remaining service life of the target electric submersible pump.

[0194] In some embodiments, the preset lifetime prediction model may further include an auxiliary prediction model and a joint prediction model; the joint prediction model is connected to the output of the auxiliary prediction model and the output of the main prediction model, respectively; wherein the auxiliary prediction model and the main prediction model are models with different structures, and the joint prediction model is used to receive and determine and output the corresponding target prediction result based on the output results of the main prediction model and the output results of the auxiliary prediction model.

[0195] In some embodiments, the auxiliary prediction model may specifically include at least one of the following: a first auxiliary sub-model based on a CNN-LSTM structure, a second auxiliary sub-model based on a GRU structure, a third auxiliary sub-model based on a Bi-GRU-LSTM structure, a fourth auxiliary sub-model based on a Transformer structure, etc.

[0196] In some embodiments, when the determination module 804 is specifically implemented, the remaining service life of the target electric submersible pump can be determined according to the target prediction result in the following manner: the predicted value of the remaining service life of the target electric submersible pump is corrected according to the uncertainty parameter to determine the remaining service life of the target electric submersible pump.

[0197] In some embodiments, the operating parameters may specifically include at least one of the following: current, voltage, vibration amplitude, vibration frequency, etc.

[0198] In some embodiments, the environmental parameters may specifically include at least one of the following: temperature, pressure, sand content, corrosion parameters, etc.

[0199] This specification also provides another device for determining the remaining service life of an electric submersible pump, which may specifically include the following structural modules:

[0200] The acquisition module can be used to acquire the operating parameters of the target electric submersible pump, as well as the environmental parameters of the operating environment during operation.

[0201] The construction module can be specifically used to construct target time series data about the target electric submersible pump based on the operating parameters and environmental parameters.

[0202] The prediction module can be used to process the target time series data using a preset lifetime prediction model to obtain the corresponding target prediction result; wherein, the preset lifetime prediction model includes a main prediction model and an auxiliary prediction model; the main prediction model is a hybrid model based on a Wavelet-LSTM structure; the model structure of the auxiliary prediction model is different from that of the main prediction model; the target prediction result includes at least the predicted value of the remaining lifetime of the target electric submersible pump, and the uncertainty parameters of the predicted value of the remaining lifetime.

[0203] The determination module can be used to determine the remaining service life of the target electric submersible pump based on the target prediction results.

[0204] It should be noted that the units, devices, or modules described in the above embodiments can be implemented by computer chips or physical entities, or by products with certain functions. For ease of description, the above devices are described by dividing them into various modules according to their functions. Of course, in implementing this specification, the functions of each module can be implemented in one or more software and / or hardware, or the module that implements the same function can be implemented by a combination of multiple sub-modules or sub-units, etc. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection between the devices or units shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0205] As can be seen from the above, the device for determining the remaining service life of an electric submersible pump provided in the embodiments of this specification can be well adapted to the complex operating conditions of electric submersible pumps, accurately predict the remaining service life of the electric submersible pump, and thus accurately perform corresponding operation and maintenance on the electric submersible pump, extend the service life of the electric submersible pump, and ensure the stability and safety of oil and gas field lifting operations.

[0206] In a specific scenario example, the method for determining the remaining service life of an electric submersible pump (ESP) provided in this manual can be applied to predict and utilize the remaining service life of an ESP by integrating deep learning algorithms with uncertainty quantification. The specific implementation process can be found below.

[0207] In this scenario example, existing methods often employ Multilayer Perceptrons (MLPs) to predict the Remaining Life (RUL) of electric submersible pumps (ESPs). However, MLPs typically only consider spatial features and ignore time-series information, resulting in limited prediction accuracy and an inability to accurately predict the RUL of ESPs in complex environments. Long Short-Term Memory (LSTM) networks, an improved structure of traditional Recurrent Neural Networks (RNNs), effectively alleviate the vanishing gradient problem by introducing memory units and three gating mechanisms (input gate, forget gate, and output gate), and can also capture long-term dependencies, performing excellently in time-series modeling. However, prediction based on conventional LSTMs also has limitations; it cannot effectively handle the complex and variable operating conditions of ESPs, and it also neglects the uncertainty of the prediction results.

[0208] To address the aforementioned issues and their root causes, this scenario example considers constructing five differentiated deep learning models (CNN-LSTM, GRU, Bi-GRU-LSTM, Transformer, and Wavelet-LSTM) to enhance the ability to characterize uncertainty from the perspectives of model structure and feature processing. This approach better adapts to the complex, multi-target operating conditions of electric submersible pumps (ESPs) and effectively covers various changing scenarios. Furthermore, parallel prediction using multiple models fully leverages the advantages of different model structures to obtain and determine the remaining service life of the ESP based on the prediction results from multiple different models. Simultaneously, model variability is acquired and utilized to reflect uncertainty (e.g., to determine uncertainty parameters), as the structural characteristics of different models determine their varying abilities to fit and generalize data.

[0209] Among them, the CNN-LSTM mentioned above is good at extracting local temporal features (such as short-term pressure fluctuations); the GRU / Bi-GRU-LSTM mentioned above focuses on capturing long-term temporal dependencies (such as monthly temperature change trends); and the Wavelet-LSTM mentioned above decomposes high and low frequency components through wavelet transform (such as distinguishing between instantaneous vibration noise and long-term performance degradation signals), which can more accurately process multi-scale data under complex working conditions.

[0210] In this scenario example, multiple models can be run simultaneously to obtain multiple sets of prediction results. If the predicted values ​​from multiple models are highly consistent, it indicates low uncertainty. If the predicted values ​​have large dispersion (e.g., 90-95 hours based on GRU and 105-110 hours based on Transformer), it directly reflects the uncertainty caused by data noise and operational complexity. The Wavelet-LSTM also supports multi-branch feature fusion to quantify the uncertainty at the feature level within the model (e.g., uncertainty parameters).

[0211] As the best-performing model (primary prediction model) in this scenario example, the Wavelet-LSTM described above can achieve a fine characterization of uncertainty through parallel processing of three branches. For details, please refer to... Figure 9 As shown.

[0212] Specifically, branch 1 (corresponding to the first branch, used for transmitting and processing low-frequency component data): processes "slowly changing signals" (such as long-term temperature drift) that have degraded the performance of the equipment. These signals have high determinism and low uncertainty.

[0213] Branch 2 (corresponding to the second branch, used for transmitting and processing high-frequency component data): handles transient interference "rapidly changing signals" (such as sudden vibrations, sand content fluctuations), which are highly random and uncertain;

[0214] Branch 3 (corresponding to the third branch, used for transmitting and processing raw signals): retains the raw monitoring data without processing, reflecting the "mixed uncertainty" under real working conditions.

[0215] The model then weights and integrates the results from the three branches through a feature fusion layer. If the proportion of high-frequency components is high (e.g., frequent sudden vibrations), the prediction range of the output will be widened accordingly to reflect the uncertainty caused by instantaneous interference; if the proportion of low-frequency components is high (e.g., equipment stability degradation), the prediction range will be narrowed, and the uncertainty will be reduced. By employing a combination of various deep learning models, including CNN-LSTM, GRU, Bi-GRU-LSTM, Transformer, and Wavelet-LSTM, and performing remaining service life estimation predictions for electric submersible pumps, more practical and accurate prediction results can be obtained, providing a more reliable basis for equipment condition assessment and maintenance decisions. Specific implementation may include the following steps.

[0216] Step 1: Collect multi-dimensional sensor monitoring data of the electric submersible pump during operation (e.g., the operating parameters of the target electric submersible pump and the environmental parameters of the operating environment), including but not limited to parameters such as current, voltage, temperature, pressure, and vibration. Then, perform data cleaning, normalization, and sliding window partitioning to form time-series input samples that can be used for model training.

[0217] Step two involves constructing various deep learning model combinations, including a fusion model of convolutional neural networks and long short-term memory networks (CNN-LSTM), a gated recurrent unit (GRU) model, a fusion model of bidirectional gated recurrent units and long short-term memory networks (Bi-GRU-LSTM), a Transformer model, and a wavelet transform and long short-term memory network combination model (Wavelet-LSTM). The core innovation of Wavelet-LSTM lies in introducing wavelet transform as a feature preprocessing module; see [reference needed]. Figure 9 As shown, the model first decomposes multi-dimensional sensor time-series data (such as current, voltage, temperature, pressure, vibration, etc., sampled at a frequency of 20 minutes / time, spanning approximately one year) during the operation of the electric submersible pump into multi-scale components using PyWavelets. This decomposition results in low-frequency components (Approximation) reflecting slow performance degradation and strong determinism, and high-frequency components (Detail) reflecting transient interference and strong randomness, while preserving the original signal. Subsequently, targeted feature extraction is performed through three independent LSTM layers (LSTM layer 1 processes the low-frequency components, LSTM layer 2 processes the high-frequency components, and LSTM layer 3 processes the original signal). Finally, a feature fusion layer weightedly integrates the features from the three branches, and a fully connected output layer completes the prediction of remaining service life. This design not only overcomes the limitations of existing single-dimensional time-series modeling techniques but also accurately handles multi-scale data under complex operating conditions in multiple wells.

[0218] Step 3: Train and predict the samples using the various deep learning models mentioned above, and obtain the remaining lifespan prediction results for each model.

[0219] For specific implementation, please refer to Figure 9As shown, the first step is to collect the dataset, adjusting the multi-dimensional production record data (time series data) to a structure suitable for the model input, such as (batch_size, seq_len, input_size). Then, use wavelet transform to perform multi-scale decomposition on the data. The "multi-scale" approach here is essentially based on the time-frequency localization properties of wavelet functions, analyzing the original time series at different time and frequency resolutions: For slowly changing, long-period signal components in the time series (such as the slow temperature rise and stable pressure drop due to component aging during long-term operation of an electric submersible pump, reflecting the gradual degradation of equipment performance), the low-frequency filtering properties of wavelet transform are used to separate them into low-frequency components; while for drastically changing, short-period signal components in the time series (such as instantaneous vibrations caused by sand content fluctuations and voltage spikes caused by power grid fluctuations, reflecting sudden interference or random noise), the high-frequency filtering properties of wavelet transform are used to separate them into high-frequency components: low-frequency components are analyzed through LSTM layer 1, mainly performing low-frequency processing; high-frequency components are analyzed through LSTM layer 2, mainly performing high-frequency processing; and the original signal, which is not processed, is analyzed through LSTM layer 3, mainly performing original feature processing. The input to LSTM layer 1 is the low-frequency component after wavelet transform decomposition, and its function is to "extract long-term degradation trend features". Therefore, its network parameter settings focus more on capturing long-term time-series dependencies. For example, by adjusting the number of memory units and the weight of the forget gate, the memory of "monthly / quarterly" slowly changing signals is strengthened. The input to LSTM layer 2 is the high-frequency component, and its function is to "extract short-term abnormal fluctuation features". The parameter settings focus more on the response to short-term and sudden signals. For example, shortening the time step and adjusting the input gate weights ensure that "minute / hourly" rapidly changing signals can be captured quickly. The input to LSTM layer 3 is the original signal, and its function is to "extract comprehensive operating condition features". The parameter settings need to balance the ability to capture long and short-term time-series signals. It should retain the memory of long-term trends and be able to respond to short-term anomalies, so as to avoid one-sided feature extraction due to focusing on one dimension. Finally, feature fusion is performed on the features of the three branches. The standardized feature vectors of the three branches are linearly combined according to dynamic weights, and then linearly mapped and reduced in dimensionality through a fully connected layer to output the prediction result. The fully connected layer receives the high-dimensional feature vector (e.g., 128-dimensional, including information such as device degradation and abnormal fluctuations) after fusion of the three branches (low frequency, high frequency, and original signal) of Wavelet-LSTM. It performs linear calculations using a preset weight matrix and bias term to transform the abstract feature vector into a value directly related to lifetime. For example, the formula "feature vector × weight matrix + bias term" is used to transform the 128-dimensional feature into intermediate values. These values ​​will gradually approach the goal of "remaining lifetime", completing the transformation from "feature description" to "lifetime correlation".The fused high-dimensional features (e.g., 128-dimensional) contain redundant information. Fully connected layers eliminate redundancy and reduce computation by reducing feature dimensionality. For example, the first fully connected layer compresses the 128-dimensional features to 32-dimensional features (retaining key prediction information such as temperature drift and vibration peaks), and then the second fully connected layer further compresses them to 1-dimensional features, finally outputting a specific RUL prediction value (e.g., "95 hours remaining"), achieving "dimensionality reduction without losing key information," making the results simpler and more suitable for the actual needs of lifetime prediction.

[0220] Specifically, the aforementioned low-frequency components correspond to the "slow-changing signals" of electric submersible pump performance degradation, such as the slow rise in temperature and stable decrease in pressure caused by component aging during long-term operation. These signals are unaffected by short-term random interference and exhibit strong stability and regularity. In service life prediction, their core advantage lies in accurately capturing the long-term degradation trend of equipment from normal operation to gradual failure, which is the core basis for remaining service life (RUL) prediction. Because low-frequency components reflect the essential change patterns of equipment performance, prediction models built upon them can effectively avoid interference from short-term fluctuations, ensuring a more accurate judgment of the long-term life trajectory of the equipment. This provides a stable and reliable long-term trend reference for "predictive maintenance," avoiding misjudgments of equipment life stages due to short-term noise.

[0221] The aforementioned high-frequency components correspond to "rapidly changing signals" caused by fluctuations in operating conditions, such as instantaneous vibrations caused by fluctuations in sand content and voltage spikes caused by power grid fluctuations. Although these signals have strong randomness, they still have key advantages in service life prediction. On the one hand, high-frequency components can reflect the short-term abnormal operating condition impacts faced by equipment in real time. For example, sudden vibrations may accelerate component wear and prematurely consume the remaining service life of the equipment. By analyzing high-frequency components, these "accelerating factors of service life loss" can be captured in time, correcting long-term prediction results and avoiding prediction deviations caused by ignoring short-term damage. On the other hand, the degree of fluctuation of high-frequency components is directly related to the uncertainty of prediction results. By analyzing the intensity and frequency of high-frequency components, the impact of short-term operating condition fluctuations on prediction results can be quantified, making the final service life prediction range more consistent with actual operating conditions and improving the robustness of prediction results.

[0222] The low-frequency components extracted above represent the "long-term, gradual" degradation characteristics of equipment performance, specifically including: the slow monthly / quarterly drift trend of the ESP operating temperature, the stable decrease in outlet pressure over operating time, and the long-term small increase in motor current. These characteristics are characterized by "large time span, gradual change, and strong regularity," and can directly reflect the aging degree of core equipment components (such as motors and pump shafts). For example, the long-term temperature drift can reflect the aging rate of the motor insulation layer, and the stable decrease in pressure can reflect the wear degree of the pump impeller, serving as a core feature for judging the "benchmark value" of the equipment's remaining lifespan. The high-frequency components extract the "short-term, transient" abnormal characteristics during equipment operation, specifically including: the peak value and frequency of transient vibrations (such as a sudden increase in vibration frequency from 50Hz to 120Hz when sand content increases rapidly), voltage / current spikes (such as a 20% instantaneous voltage increase caused by power grid fluctuations), and short-term sudden changes in pressure (such as pressure fluctuations caused by wellhead throttle valve adjustments), etc. These characteristics are characterized by "short time span, drastic changes, and strong randomness," reflecting the short-term operating conditions and shocks faced by equipment. For example, instantaneous vibration peaks can reflect the impact load on components, and voltage spikes can reflect the instantaneous stress of the electrical system. These characteristics are key bases for correcting "baseline life values" and quantifying uncertainties. The original signal has not undergone wavelet transform decomposition, preserving the complete information of the mixture of high and low frequency signals. Therefore, what is extracted is a "comprehensive and all-encompassing" operating condition and performance characteristic, which includes both the long-term degradation trend in the low-frequency component and the short-term abnormal fluctuations in the high-frequency component, such as the mixed feature of "slow temperature rise (low frequency) + 3 instantaneous vibrations per day (high frequency)" within a certain period of time. These features can completely restore the operating state of the electric submersible pump under real working conditions, avoid information loss due to signal decomposition, and provide a "global reference" for the features extracted from high and low frequency components. When there are contradictions between the features of high and low frequency components (such as low frequency showing long-term stability but high frequency showing frequent abnormalities), the original signal features can help determine whether it is "short-term abnormality has not affected the long-term trend" or "short-term abnormality has begun to accelerate long-term degradation", ensuring the completeness and accuracy of feature analysis.

[0223] The following example, using an oilfield in a specific region, illustrates the application of this dataset. The dataset for this region contains multi-dimensional sensor data from 58 wells, including but not limited to parameters such as current, voltage, temperature, pressure, and vibration. Production data for each well is recorded every 20 minutes, with a time span of approximately one year.

[0224] In practice, a combination of deep learning models was used to train the dataset, with 7 wells serving as training wells, 2 as test wells, and the last well as a validation well. Several deep learning models were constructed, including a fusion model of convolutional neural network and long short-term memory network (CNN-LSTM), a gated recurrent unit (GRU) model, a bidirectional gated recurrent unit and long short-term memory network (Bi-GRU-LSTM) model, a Transformer model, and a wavelet transform and long short-term memory network (Wavelet-LSTM) model. Considering the impact of device performance and computational efficiency, the batch size was set to 32, the training epochs to 25, and the learning rate to 1e-4.

[0225] By plotting the actual lifespan curve and the model's predicted curve, the model's prediction effectiveness for pump lifespan is demonstrated. Individually plotted error curves can be compared horizontally to calculate the prediction errors of different models, reflecting differences in model performance. Specific prediction results for different models for a particular well in this region can be found in [reference needed]. Figure 10 As shown, the true value represents the actual remaining lifetime. Further details regarding the errors of different models' predictions relative to the actual remaining lifetime can be found in [reference needed]. Figure 11 As shown, the model was trained on production data from 40 wells, tested on production data from 10 wells, and validated on production data from 8 wells. The results show that the model effectively predicts the life curve of electric submersible pumps. The model combining wavelet transform and long short-term memory network achieved the best results, with a prediction error of less than 10 hours.

[0226] Through the above scenario examples, the method for determining the remaining service life of the ESP (Electric Submersible Pump) provided in this manual was verified. A complete preprocessing workflow for ESP well production data was established. For multi-dimensional sensor monitoring data during pump operation, including but not limited to parameters such as current, voltage, temperature, pressure, and vibration, data cleaning, normalization, and sliding window partitioning were performed to form time-series input samples that can be used for model training. The life prediction model constructs a combination of various deep learning models, including a fusion model of convolutional neural networks and long short-term memory networks (CNN-LSTM), a Transformer model, and a wavelet transform and long short-term memory network combination model (Wavelet-LSTM). This model effectively adapts to the complex operating conditions of ESPs, accurately predicts the remaining service life of the ESP, and thus enables precise maintenance and operation of the ESP, extending its service life and ensuring the stability and safety of oil and gas field lifting operations.

[0227] While this specification provides the steps of operation for the methods described in the embodiments or flowcharts, more or fewer steps may be included based on conventional or non-inventive means. The order of steps listed in the embodiments is merely one possible order of execution among many steps and does not represent the only possible order. In actual device or client product execution, the methods shown in the embodiments or drawings may be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment, or even a distributed data processing environment). The terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, product, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, product, or apparatus. Without further limitations, the presence of other identical or equivalent elements in a process, method, product, or apparatus that includes said elements is not excluded. The terms "first," "second," etc., are used to denote names and do not indicate any particular order.

[0228] Those skilled in the art will also know that, besides implementing the controller using purely computer-readable program code, the same functions can be achieved by logically programming the method steps, making the controller function as logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers (PLCs), and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the devices within it used to implement various functions can also be considered structures within that hardware component. Alternatively, the devices used to implement various functions can be considered as both software modules implementing the method and structures within a hardware component.

[0229] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, classes, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer-readable storage media, including storage devices.

[0230] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this specification can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solutions of this specification can essentially be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, mobile terminal, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments of this specification.

[0231] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. This specification can be used in numerous general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable electronic devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices, etc.

[0232] Although this specification has been described by way of examples, those skilled in the art will recognize that many variations and modifications are possible without departing from the spirit of this specification, and it is intended that the appended claims cover such variations and modifications without departing from the spirit of this specification.

Claims

1. A method for determining the remaining service life of an electric submersible pump, characterized in that, include: Obtain the operating parameters of the target electric submersible pump, as well as the environmental parameters of the operating environment during operation; Based on the operating parameters and environmental parameters, construct target time series data for the target electric submersible pump; The target time series data is processed using a preset lifetime prediction model to obtain the corresponding target prediction result; wherein, the preset lifetime prediction model includes at least a master prediction model; the master prediction model is a hybrid model based on a Wavelet-LSTM structure; the target prediction result includes at least the predicted value of the remaining lifetime of the target electric submersible pump, and the uncertainty parameter of the predicted value of the remaining lifetime. Based on the target prediction results, the remaining service life of the target electric submersible pump is determined.

2. The method according to claim 1, characterized in that, The main prediction model includes: a multi-scale decomposition layer, a first feature processing layer, a second feature processing layer, a third feature processing layer, a feature fusion layer, and a fully connected output layer; The multi-scale decomposition layer is a model layer based on the Wavelet Transform structure, and the first feature processing layer, the second feature processing layer, and the third feature processing layer are model layers based on the LSTM structure. The multi-scale decomposition layer is connected to the first feature processing layer via a first branch and to the second feature processing layer via a second branch; the first feature processing layer, the second feature processing layer, and the third feature processing layer are also connected to the feature fusion layer; the feature fusion layer is connected to the fully connected output layer.

3. The method according to claim 2, characterized in that, The process of processing the target time series data using a preset lifetime prediction model to obtain the corresponding prediction results includes: The target time series data is processed using a multi-scale decomposition layer to decompose the corresponding low-frequency component data and high-frequency component data. The low-frequency component data is processed using the first feature processing layer to obtain long-term dependency features; the high-frequency component data is processed using the second feature processing layer to obtain short-term fluctuation features; and the target time series data is processed using the third feature processing layer to obtain comprehensive operating condition features. The long-term dependency features, short-term fluctuation features, and comprehensive operating condition features are dynamically weighted and fused using a feature fusion layer to obtain the corresponding target fused features. The target fusion features are processed by the fully connected output layer to determine and output the corresponding target prediction results.

4. The method according to claim 3, characterized in that, The fully connected output layer includes a mapping module, a dimensionality reduction module, and a prediction module; wherein the mapping module is connected to the feature fusion layer and the dimensionality reduction module, and the prediction module is connected to the dimensionality reduction module. Accordingly, by processing the target fusion features using a fully connected output layer, the corresponding target prediction results are determined and output, including: The mapping module is used to map the target fusion features into corresponding target intermediate features; The dimensionality reduction module is used to perform dimensionality reduction on the intermediate features of the target to obtain the dimensionality-reduced intermediate features. The intermediate features after dimensionality reduction are processed by the prediction module to determine and output the first prediction sub-result; wherein the first prediction sub-result is used to indicate the predicted value of the remaining service life of the target electric submersible pump.

5. The method according to claim 4, characterized in that, The fully connected output layer further includes a reliability analysis module; wherein the reliability analysis module is connected to the prediction module and the second feature processing layer respectively; Accordingly, the method further includes using a fully connected output layer to process the target fusion features, determine and output the corresponding target prediction result, and then using this method to determine and output the target prediction result. The reliability analysis module determines and outputs a second prediction sub-result based on the first prediction sub-result and the short-term fluctuation characteristics; wherein the second prediction sub-result is used to indicate the uncertainty parameter of the predicted value of the remaining service life of the target electric submersible pump.

6. The method according to claim 1, characterized in that, The preset lifetime prediction model also includes an auxiliary prediction model and a joint prediction model; the joint prediction model is connected to the output of the auxiliary prediction model and the output of the main prediction model, respectively; wherein, the auxiliary prediction model and the main prediction model are models with different structures, and the joint prediction model is used to receive and determine and output the corresponding target prediction result based on the output results of the main prediction model and the output results of the auxiliary prediction model.

7. The method according to claim 6, characterized in that, The auxiliary prediction model includes at least one of the following: a first auxiliary sub-model based on CNN-LSTM structure, a second auxiliary sub-model based on GRU structure, a third auxiliary sub-model based on Bi-GRU-LSTM structure, and a fourth auxiliary sub-model based on Transformer structure.

8. The method according to claim 1, characterized in that, The step of determining the remaining service life of the target electric submersible pump based on the target prediction result includes: Based on the uncertainty parameters, the predicted value of the remaining service life of the target electric submersible pump is corrected to determine the remaining service life of the target electric submersible pump.

9. The method according to claim 1, characterized in that, The operating parameters include at least one of the following: current, voltage, vibration amplitude, and vibration frequency.

10. The method according to claim 1, characterized in that, The environmental parameters include at least one of the following: temperature, pressure, sand content, and corrosion parameters.

11. A method for determining the remaining service life of an electric submersible pump, characterized in that, include: Obtain the operating parameters of the target electric submersible pump, as well as the environmental parameters of the operating environment during operation; Based on the operating parameters and environmental parameters, construct target time series data for the target electric submersible pump; The target time series data is processed using a preset lifetime prediction model to obtain the corresponding target prediction result; wherein, the preset lifetime prediction model includes a main prediction model and an auxiliary prediction model; the main prediction model is a hybrid model based on a Wavelet-LSTM structure; the model structure of the auxiliary prediction model is different from that of the main prediction model; the target prediction result includes at least the predicted value of the remaining lifetime of the target electric submersible pump, and the uncertainty parameter of the predicted value of the remaining lifetime. Based on the target prediction results, the remaining service life of the target electric submersible pump is determined.

12. A device for determining the remaining service life of an electric submersible pump, characterized in that, include: The acquisition module is used to acquire the operating parameters of the target electric submersible pump, as well as the environmental parameters of the operating environment during operation. The construction module is used to construct target time series data about the target electric submersible pump based on the operating parameters and environmental parameters. The prediction module is used to process the target time series data using a preset lifetime prediction model to obtain the corresponding target prediction result; wherein, the preset lifetime prediction model includes at least a master prediction model; the master prediction model is a hybrid model based on a Wavelet-LSTM structure; the target prediction result includes at least the predicted value of the remaining lifetime of the target electric submersible pump, and the uncertainty parameter of the predicted value of the remaining lifetime. The determination module is used to determine the remaining service life of the target electric submersible pump based on the target prediction results.

13. An electronic device, characterized in that, It includes a processor and a memory for storing processor-executable instructions, wherein the processor, when executing the instructions, implements the steps of the method according to any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that, It stores computer instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 11.

15. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the steps of the method according to any one of claims 1 to 11.

Images