Pump inspection period prediction method and device, equipment, storage medium and program product

Abstract

The invention relates to a pump inspection period prediction method and device, equipment, a storage medium and a program product. The method comprises the following steps: acquiring original production data of a rod-pumped well to be detected; inputting the original production data into a preset prediction network for prediction to obtain a pump inspection period of the rod-pumped well to be detected; the preset prediction network is used for performing period prediction based on the manual features, the spatial-temporal features and the time sequence features corresponding to the original production data. The preset prediction network in the embodiment of the invention can perform period prediction based on the manual features, the spatial-temporal features and the time sequence features corresponding to the original production data, that is, the pump inspection period of the rod pumped well to be detected can be predicted based on the features of three dimensions, so that the pump inspection period of the rod pumped well to be detected can be predicted; and a more accurate pump detection period of the rod pumped well to be detected can be obtained.

Description

Technical Field

[0001] The present application relates to the field of detection technology, and in particular to a method, device, equipment, storage medium and program product for predicting a pump detection cycle. Background Art

[0002] With the development of crude oil production and detection technology, rod pumps can be used for oil production. Due to the particularity of the working environment of rod pumps and the complexity of the downhole environment during the oil production process, the failure rate of rod pumps during the oil production process is relatively high. The pump inspection cycle of the wells with in-depth detection of rod pumps can effectively predict the occurrence of failures. Therefore, accurately predicting the pump inspection cycle of the wells is one of the effective means to reduce the failure rate of rod pumps during the oil production process.

[0003] However, the current prediction method of pump inspection cycle has the problem of low accuracy. Summary of the invention

[0004] Based on this, it is necessary to provide a method, device, equipment, storage medium and program product for predicting the pump inspection period, which can improve the accuracy of predicting the pump inspection period, in order to solve the above technical problems.

[0005] In a first aspect, the present application provides a method for predicting a pump inspection cycle, comprising:

[0006] Obtaining the original production data of the pumping well to be tested;

[0007] The original production data is input into a preset prediction network for prediction to obtain the pump inspection cycle of the pumping well to be tested; the preset prediction network is used to perform cycle prediction based on manual features, spatiotemporal features and time series features corresponding to the original production data.

[0008] In the above-mentioned prediction method of the pump inspection cycle, the original production data of the oil pumping well to be tested is obtained; the original production data is input into the preset prediction network for prediction to obtain the pump inspection cycle of the oil pumping well to be tested; the preset prediction network is used to perform cycle prediction based on the manual features, spatiotemporal features and time series features corresponding to the original production data. Since the preset prediction network in the embodiment of the present application can perform cycle prediction based on the manual features, spatiotemporal features and time series features corresponding to the original production data, that is, it can predict the pump inspection cycle of the oil pumping well to be tested based on the features of three dimensions, therefore, a more accurate pump inspection cycle of the oil pumping well to be tested can be obtained.

[0009] In one embodiment, the preset prediction network includes a data processing module, a feature extraction module and a prediction module, and the raw production data is input into the preset prediction network for prediction to obtain the pump inspection cycle of the pumping well to be tested, including:

[0010] Inputting the original production data into the data processing module for data processing and outputting a time window sequence;

[0011] Input the time window sequence into the feature extraction module to perform multi-dimensional feature extraction, and output a feature set; the feature set includes the manual features, the spatiotemporal features, and the time series features;

[0012] The feature set is input into the prediction module for cycle prediction, and the pump inspection cycle of the pumping well to be tested is output.

[0013] In this embodiment, the original production data can be input into the data processing module for data processing, and the time window sequence can be output. The time window sequence can be input into the feature extraction module for multi-dimensional feature extraction and output as a feature set, so as to accurately obtain manual features, spatiotemporal features and time series features. Thus, the accurate manual features, spatiotemporal features and time series features can be input into the prediction module for cycle prediction, so as to more accurately output the pump inspection cycle of the pumping well to be tested.

[0014] In one embodiment, the feature extraction module includes a manual feature extraction subnetwork, a spatiotemporal feature extraction subnetwork and a temporal feature extraction subnetwork, and the time window sequence is input into the feature extraction module for multi-dimensional feature extraction, and a feature set is output, including:

[0015] Inputting the time window sequence into the manual feature extraction subnetwork to perform manual type feature extraction to obtain the manual features;

[0016] Performing noise reduction processing on the time window sequence to obtain a noise-reduced time window sequence;

[0017] Inputting the denoised time window sequence into the spatiotemporal feature extraction subnetwork to extract spatiotemporal type features to obtain the spatiotemporal features;

[0018] The denoised time window sequence is input into the time series feature extraction subnetwork to extract the time series type features to obtain the time series features.

[0019] In this embodiment, the time window sequence can be directly input into the manual feature extraction subnetwork for manual feature extraction to obtain accurate manual features. The time window sequence can also be subjected to denoising to obtain a denoised time window sequence, which can then be input into the spatiotemporal feature extraction subnetwork and the temporal feature extraction subnetwork for feature extraction, thereby obtaining more accurately the spatiotemporal features and temporal features based on the denoised and smoothed time window sequence.

[0020] In one embodiment, the spatiotemporal feature extraction subnetwork includes multiple convolutional layers, a bidirectional long short-term memory convolutional layer and a first attention layer, and the denoised time window sequence is input into the spatiotemporal feature extraction subnetwork to extract spatiotemporal type features to obtain the spatiotemporal features, including:

[0021] Inputting the denoised time window sequence into each of the convolutional layers to extract spatial features, and performing feature fusion on each of the extracted first features to obtain a first fused feature; the size of the convolution kernel in each of the convolutional layers is different;

[0022] Inputting the first fusion feature into the bidirectional long short-term memory convolutional layer to extract spatiotemporal features to obtain a second feature;

[0023] Inputting the second feature into the first attention layer for weighted processing to obtain a third feature;

[0024] The third feature and the first fusion feature are fused to obtain the spatiotemporal feature.

[0025] In this embodiment, since the size of the convolution kernel in each convolution layer is different, the denoised time window sequence is input into each convolution layer for spatial feature extraction, and the first extracted features are fused to obtain the first fused feature. Convolution kernels of different sizes can be used to extract the degradation information contained in the data from different scales, and more diverse representative features can be mined. After that, the first fused feature is input into the bidirectional long short-term memory convolution layer and the first attention layer for spatiotemporal feature extraction and weighted processing to obtain the third feature, which can be assigned a larger weight to more important spatiotemporal features by adding the first attention layer Multi-Attention after the bidirectional long short-term memory convolution layer Bi-LSTM network layer, so that the spatiotemporal feature extraction subnetwork can be more focused on information useful for pump cycle prediction. Then, the third feature is fused with the first fused feature, which can reduce the loss of spatial features in the spatiotemporal feature extraction subnetwork during the construction process, thereby effectively mining the spatiotemporal features contained in the denoised time window sequence.

[0026] In one embodiment, the temporal feature extraction subnetwork includes a second attention layer and a time domain convolution layer, and the denoised time window sequence is input into the temporal feature extraction subnetwork to extract the feature of the temporal type to obtain the temporal feature, including:

[0027] Inputting the denoised time window sequence into the second attention layer for weighted processing to obtain a processed time window sequence;

[0028] The processed time window sequence is input into the time domain convolution layer for feature extraction to obtain the time series feature.

[0029] In this embodiment, the denoised time window sequence is input into the second attention layer for weighted processing to obtain a processed time window sequence, and the second attention layer Multi-Attention can be used to assign a larger weight to the key features in the more important time window sequence. Thus, the processed time window sequence is input into the time domain convolution layer for feature extraction, and more accurate time series features can be extracted from the weighted time window sequence.

[0030] In one embodiment, the manual feature extraction subnetwork includes a feature extraction layer and an output layer, and the step of inputting the time window sequence into the manual feature extraction subnetwork to perform manual feature extraction to obtain the manual features includes:

[0031] Inputting the time window sequence into the feature extraction layer to extract features and obtain initial manual features;

[0032] The initial manual features are input into the output layer for inner product merging processing to obtain the manual features.

[0033] In this embodiment, the time window sequence can be input into the feature extraction layer for feature extraction to obtain initial manual features, and then the initial manual features can be input into the output layer for inner product merging processing to accurately obtain the manual features, thereby providing richer and more effective degradation information for pump inspection cycle prediction, which helps to improve the accuracy of pump inspection cycle prediction.

[0034] In one embodiment, the prediction module includes a fusion sub-network and a regression sub-network, and the feature set is input into the prediction module for period prediction, and the pump inspection period of the pumping well to be tested is output, including:

[0035] Inputting the feature set into a fusion subnetwork for feature fusion to obtain a second fusion feature;

[0036] The second fusion feature is input into the regression sub-network for prediction to obtain the pump inspection cycle of the oil pumping unit to be tested.

[0037] In this embodiment, the feature set can be input into the fusion subnetwork for feature fusion to obtain a second fused feature. Thus, the second fused feature is input into the regression subnetwork for prediction. Based on the fusion subnetwork and the regression subnetwork, the pump inspection cycle of the oil pump to be tested can be obtained more accurately.

[0038] In a second aspect, the present application also provides a pump inspection cycle prediction device, comprising:

[0039] An acquisition module is used to acquire the original production data of the pumping well to be tested;

[0040] The prediction module is used to input the original production data into a preset prediction network for prediction to obtain the pump inspection cycle of the pumping well to be tested; the preset prediction network is used to perform cycle prediction based on the manual features, spatiotemporal features and time series features corresponding to the original production data.

[0041] In a third aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method in any one of the embodiments of the first aspect are implemented.

[0042] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method in any one of the embodiments of the first aspect are implemented.

[0043] In a fifth aspect, the present application further provides a computer program product, wherein the computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the method in any one of the embodiments of the first aspect are implemented.

[0044] The above-mentioned prediction method, device, equipment, storage medium and program product of the pump inspection cycle obtains the original production data of the pumping well to be tested; inputs the original production data into the preset prediction network for prediction to obtain the pump inspection cycle of the pumping well to be tested; the preset prediction network is used to perform cycle prediction based on the manual features, spatiotemporal features and time series features corresponding to the original production data. Since the preset prediction network in the embodiment of the present application can perform cycle prediction based on the manual features, spatiotemporal features and time series features corresponding to the original production data, that is, it can predict the pump inspection cycle of the pumping well to be tested based on the features of three dimensions, it is possible to obtain a more accurate pump inspection cycle of the pumping well to be tested. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0046] Figure 1 An application environment diagram of a method for predicting a pump inspection cycle in an embodiment;

[0047] Figure 2 A schematic diagram of a flow chart of a method for predicting a pump inspection cycle in one embodiment;

[0048] Figure 3 A schematic diagram of a process flow of a cycle prediction step in an embodiment;

[0049] Figure 4 A schematic diagram of a process flow of a multi-feature extraction step in an embodiment;

[0050] Figure 5 A schematic diagram of the structure of a spatiotemporal feature extraction subnetwork in one embodiment;

[0051] Figure 6 A schematic diagram of the structure of a time series feature extraction subnetwork in one embodiment;

[0052] Figure 7 A schematic diagram of the structure of a manual feature extraction subnetwork in one embodiment;

[0053] Figure 8 It is a schematic flow chart of a method for predicting a pump inspection cycle in an optional embodiment;

[0054] Fig. 9 FIG. 4 is a structural block diagram of a device for predicting a pump inspection cycle in an embodiment. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" in the specification and claims of this application and the above-mentioned figure descriptions and any variations thereof are intended to cover non-exclusive inclusions.

[0057] In the description of the embodiments of the present application, the technical terms "first", "second", etc. are only used to distinguish different objects, and cannot be understood as indicating or implying relative importance or implicitly indicating the number, specific order or primary and secondary relationship of the indicated technical features. In the description of the embodiments of the present application, the meaning of "multiple" is more than two, unless otherwise clearly and specifically defined.

[0058] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0059] With the development of crude oil production and detection technology, rod pumps can be used for oil production. Due to the particularity of the working environment of rod pumps during oil production and the complexity of the downhole environment, the failure rate of rod pumps during oil production is high, which makes it impossible to carry out crude oil production normally. The pump inspection cycle of the wells with in-depth detection of rod pumps can effectively predict the occurrence of failures. Therefore, accurately predicting the pump inspection cycle of the wells is one of the effective means to reduce the failure rate of rod pumps during oil production, which is crucial to improving the production capacity and economic benefits of oil wells.

[0060] The pump inspection cycle prediction network used in traditional technology has the following problems: First, the traditional pump inspection cycle prediction network can only extract the features of the original production data from a single scale. However, the degradation information reflected by the original production data obtained by different types of sensors is different, so the traditional technology will ignore the diversity of degradation information contained in the original production data; Second, the traditional pump inspection cycle prediction network only uses the features learned in the last time step to predict the pump inspection cycle. However, the features learned at different time steps have different contributions to the prediction results of the final pump inspection cycle; Second, the features extracted by the traditional pump inspection cycle prediction network may have feature loss, and there is a problem of unbalanced feature extraction in time and space. Therefore, the current pump inspection cycle prediction method has the problem of low accuracy. After introducing the background technology of the pump inspection cycle prediction method provided in the embodiment of the present application, the implementation environment involved in the pump inspection cycle prediction method provided in the embodiment of the present application will be briefly described below.

[0061] The pump inspection cycle prediction method provided in the embodiment of the present application can be applied to Figure 1In the computer device shown. The computer device can be a terminal or a server, and the computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be realized through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for predicting a pump inspection cycle is realized. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0062] Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific terminal may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0063] In one embodiment, Figure 2 As shown, a method for predicting the pump inspection cycle is provided. Figure 1 The computer device in the example is used to illustrate, including the following steps:

[0064] S201, obtaining original production data of the pumping well to be tested.

[0065] Among them, the original production data of the pumping well to be tested may include but is not limited to at least one of the production data of the pumping well to be tested, such as minimum load, maximum load, pump efficiency, pump depth, pump diameter, strokes, stroke, dynamic liquid level, downward current, upward current, displacement, daily liquid production, daily water production, daily oil production, flow pressure, casing pressure, oil pressure, water content and wellhead temperature.

[0066] In the embodiment of the present application, the computer device can optionally obtain the original production data of the pumping well to be tested in real time; or the computer device can also obtain the original production data of the pumping well to be tested at a fixed time; or the computer device can also obtain the original production data of the pumping well to be tested when receiving the periodic prediction instruction sent by the user terminal. Of course, the embodiment of the present application does not limit the specific implementation method of obtaining the original production data of the pumping well to be tested.

[0067] S202, inputting the original production data into a preset prediction network for prediction to obtain the pump inspection cycle of the pumping well to be tested; the preset prediction network is used to perform cycle prediction based on manual features, spatiotemporal features and time series features corresponding to the original production data.

[0068] In the embodiment of the present application, optionally, the computer device can directly input the original production data of the pumping well to be tested into the preset prediction network for feature extraction, obtain the manual features, spatiotemporal features and time series features corresponding to the original production data, and then use the preset prediction network to perform cycle prediction based on the manual features, spatiotemporal features and time series features corresponding to the original production data, and obtain the pump inspection cycle of the pumping well to be tested; or, the computer device can also first perform data processing on the original production data of the pumping well to be tested, and then input the original production data of the pumping well to be tested after data processing into the preset prediction network for feature extraction, obtain the manual features, spatiotemporal features and time series features corresponding to the original production data, and then use the preset prediction network to perform cycle prediction based on the manual features, spatiotemporal features and time series features corresponding to the original production data, and obtain the pump inspection cycle of the pumping well to be tested. Of course, the embodiment of the present application does not limit the specific implementation method of cycle prediction. Among them, the manual features, spatiotemporal features and time series features corresponding to the original production data can be extracted through the preset prediction network, and the cycle prediction can be performed based on the manual features, spatiotemporal features and time series features corresponding to the original production data. In this way, a more accurate pump inspection cycle prediction can be performed based on the characteristics of the three dimensions.

[0069] In the above-mentioned prediction method of the pump inspection cycle, the original production data of the oil pumping well to be tested is obtained; the original production data is input into the preset prediction network for prediction to obtain the pump inspection cycle of the oil pumping well to be tested; the preset prediction network is used to perform cycle prediction based on the manual features, spatiotemporal features and time series features corresponding to the original production data. Since the preset prediction network in the embodiment of the present application can perform cycle prediction based on the manual features, spatiotemporal features and time series features corresponding to the original production data, that is, it can predict the pump inspection cycle of the oil pumping well to be tested based on the features of three dimensions, therefore, a more accurate pump inspection cycle of the oil pumping well to be tested can be obtained.

[0070] In one embodiment, the preset prediction network includes a data processing module, a feature extraction module and a prediction module. Based on this, a method for implementing cycle prediction is provided, namely, "inputting the original production data into the preset prediction network for prediction to obtain the pump inspection cycle of the pumping well to be tested" in the above S202. Figure 3 As shown, including:

[0071] S301, input the original production data into the data processing module for data processing, and output the time window sequence.

[0072] In the embodiment of the present application, the computer device can input the original production data of the pumping well to be tested into the data processing module in the above-mentioned preset prediction network for data processing, and output a time window sequence. Exemplarily, the computer device can first normalize the original production data to obtain the normalized original production data, and then use the sliding window method to divide the normalized original production data into the input time window sequence.

[0073] S302, input the time window sequence into the feature extraction module to perform multi-dimensional feature extraction, and output a feature set; the feature set includes manual features, spatiotemporal features, and time series features.

[0074] Among them, the feature set includes manual features, spatiotemporal features and time series features. In the embodiment of the present application, optionally, the computer device can simultaneously input the time window sequence into the feature extraction module in the above-mentioned preset prediction network to perform multi-dimensional feature extraction and output a feature set; or, the computer device can also first input the time window sequence into the feature extraction module in the above-mentioned preset prediction network to perform manual type feature extraction to obtain manual features, and then input the time window sequence into the feature extraction module in the above-mentioned preset prediction network to perform spatiotemporal type and time series type feature extraction, and output a feature set. Of course, the embodiment of the present application does not limit the specific implementation method of multi-dimensional feature extraction.

[0075] S303, input the feature set to the prediction module for cycle prediction, and output the pump inspection cycle of the pumping well to be tested.

[0076] In the embodiment of the present application, optionally, the computer device can directly input the feature set into the prediction module for cycle prediction, and output the pump inspection cycle of the pumping well to be tested; or, the computer device can first process the feature set, and then input the processed feature set into the prediction module for cycle prediction, and output the pump inspection cycle of the pumping well to be tested. Of course, the embodiment of the present application does not limit the specific implementation method of cycle prediction.

[0077] In this embodiment, the original production data can be input into the data processing module for data processing, and the time window sequence can be output. The time window sequence can be input into the feature extraction module for multi-dimensional feature extraction and output as a feature set, so as to accurately obtain manual features, spatiotemporal features and time series features. Thus, the accurate manual features, spatiotemporal features and time series features can be input into the prediction module for cycle prediction, so as to more accurately output the pump inspection cycle of the pumping well to be tested.

[0078] In one embodiment, the feature extraction module includes a manual feature extraction subnetwork, a spatiotemporal feature extraction subnetwork, and a temporal feature extraction subnetwork. Based on this, a method for implementing multi-dimensional feature extraction is provided, namely, "inputting the time window sequence into the feature extraction module for multi-dimensional feature extraction and outputting a feature set" in the above S302. Figure 4 As shown, including:

[0079] S401, input the time window sequence into the manual feature extraction sub-network to perform manual type feature extraction to obtain manual features.

[0080] In the embodiment of the present application, optionally, the computer device can directly input the time window sequence into the manual feature extraction subnetwork to perform manual feature extraction to obtain manual features; or, the computer device can first input the time window sequence into the manual feature extraction subnetwork to perform manual feature extraction to obtain initial manual features, and then input the initial manual features into the manual feature extraction subnetwork for processing to obtain manual features. Of course, the embodiment of the present application does not limit the specific implementation method for extracting manual features.

[0081] S402, performing noise reduction processing on the time window sequence to obtain a noise-reduced time window sequence.

[0082] In the embodiment of the present application, the computer device can calculate the mean value of each sensor monitoring data in each time window in the time window sequence. and standard deviation , and can be calculated based on the mean and standard deviation Set the first threshold and the second threshold , thus, the computer device can perform noise reduction and smoothing processing on the time window sequence according to the following formula (1) to obtain the noise-reduced time window sequence, which can be understood as follows: if the sensor monitoring data is greater than the first threshold or less than the second threshold , the current sensor monitoring data can be modified to half of the sum of the previous sensor monitoring data and the next sensor monitoring data corresponding to the current sensor; if the sensor monitoring data is not greater than the first threshold and not less than the second threshold , the current sensor monitoring data can be kept unchanged.

[0083] (1)

[0084] In the formula, For the Window The first of the sensors Sensor monitoring data; For the Window The mean value of the monitoring data of each sensor; For the Window The standard deviation of the sensor monitoring data; is the weight coefficient.

[0085] S403, inputting the denoised time window sequence into the spatiotemporal feature extraction subnetwork to extract spatiotemporal type features to obtain spatiotemporal features.

[0086] In the embodiment of the present application, optionally, the computer device can directly input the denoised time window sequence into the spatiotemporal feature extraction subnetwork to extract the spatiotemporal type of features to obtain spatiotemporal features; or, the computer device can first process the denoised time window sequence, and then input the processed denoised time window sequence into the spatiotemporal feature extraction subnetwork to extract the spatiotemporal type of features to obtain spatiotemporal features. Of course, the embodiment of the present application does not limit the specific implementation method for extracting spatiotemporal features.

[0087] S404: Input the denoised time window sequence into the time series feature extraction sub-network to extract the time series type features to obtain time series features.

[0088] In the embodiment of the present application, optionally, the computer device can directly input the denoised time window sequence into the time series feature extraction subnetwork to extract the time series type features to obtain the time series features; or, the computer device can first process the denoised time window sequence, and then input the processed denoised time window sequence into the time series feature extraction subnetwork to extract the time series type features to obtain the time series features. Of course, the embodiment of the present application does not limit the specific implementation method for extracting the time series features.

[0089] In this embodiment, the time window sequence can be directly input into the manual feature extraction subnetwork for manual feature extraction to obtain accurate manual features. The time window sequence can also be subjected to denoising to obtain a denoised time window sequence, which can then be input into the spatiotemporal feature extraction subnetwork and the temporal feature extraction subnetwork for feature extraction, thereby obtaining more accurately the spatiotemporal features and temporal features based on the denoised and smoothed time window sequence.

[0090] In one embodiment, the spatiotemporal feature extraction subnetwork includes multiple convolutional layers, bidirectional long short-term memory convolutional layers and a first attention layer. Based on this, a method for implementing spatiotemporal feature extraction is provided, that is, the step of "inputting the denoised time window sequence into the spatiotemporal feature extraction subnetwork to extract spatiotemporal type features to obtain spatiotemporal features" in the above S403 includes:

[0091] The denoised time window sequence is input into each convolution layer for spatial feature extraction, and each extracted first feature is fused to obtain a first fused feature; the size of the convolution kernel in each convolution layer is different.

[0092] The first fusion feature is input into the bidirectional long short-term memory convolutional layer to extract spatiotemporal features to obtain the second feature.

[0093] The second feature is input into the first attention layer for weighted processing to obtain the third feature.

[0094] The third feature is fused with the first fusion feature to obtain the spatiotemporal feature.

[0095] In the present application embodiment, Figure 5 As shown, Figure 5 The schematic diagram of the structure of the spatiotemporal feature extraction subnetwork in one embodiment is shown in FIG. 1 . For example, the spatiotemporal feature extraction subnetwork may be a Conv-Bi-LSTM (Convolutional bidirectional Long Short Term Memory) network, and the spatiotemporal feature extraction subnetwork may include but is not limited to multiple convolutional layers (1D-CNN, 1D-Convolutional Neural Network, one-dimensional convolutional neural network), bidirectional long short-term memory convolutional layer (Bi-LSTM) and first attention layer (Multi-Attention, multi-head attention mechanism). The computer device may input the denoised time window sequence into each convolutional layer 1D-CNN for spatial feature extraction, obtain the first features corresponding to each convolutional layer extracted by each convolutional layer, and perform feature fusion on the first features corresponding to each convolutional layer extracted to obtain the first fused features. It should be noted that the size of the convolution kernel in each convolutional layer is different. In this way, a multi-scale learning strategy can be adopted to use convolution kernels of different sizes to extract the degradation information contained in the data from different scales, so that more diverse representative features can be mined.

[0096] Thus, the computer device can input the first fused feature into the bidirectional long short-term memory convolution layer Bi-LSTM to extract the spatiotemporal feature and obtain the second feature. After that, the computer device can input the second feature into the first attention layer Multi-Attention for weighted processing to obtain the third feature. Then, the computer device can fuse the third feature with the first fused feature, and input the fused third feature and the first fused feature into the fully connected layer for processing to obtain the spatiotemporal feature. Among them, the first feature refers to the spatial feature corresponding to each convolution layer, the first fused feature refers to the multi-scale spatial feature, the second feature refers to the multi-scale spatiotemporal feature, the third feature refers to the spatiotemporal feature after the weight is assigned, and the spatiotemporal feature refers to the multi-scale weighted spatiotemporal feature.

[0097] In this embodiment, since the size of the convolution kernel in each convolution layer is different, the denoised time window sequence is input into each convolution layer for spatial feature extraction, and the first extracted features are fused to obtain the first fused feature. Convolution kernels of different sizes can be used to extract the degradation information contained in the data from different scales, and more diverse representative features can be mined. After that, the first fused feature is input into the bidirectional long short-term memory convolution layer and the first attention layer for spatiotemporal feature extraction and weighted processing to obtain the third feature, which can be assigned a larger weight to more important spatiotemporal features by adding the first attention layer Multi-Attention after the bidirectional long short-term memory convolution layer Bi-LSTM network layer, so that the spatiotemporal feature extraction subnetwork can be more focused on information useful for pump cycle prediction. Then, the third feature is fused with the first fused feature, which can reduce the loss of spatial features in the spatiotemporal feature extraction subnetwork during the construction process, thereby effectively mining the spatiotemporal features contained in the denoised time window sequence.

[0098] In one embodiment, the above-mentioned temporal feature extraction subnetwork includes a second attention layer and a time domain convolution layer. Based on this, a method for implementing temporal feature extraction is provided, that is, the "inputting the denoised time window sequence into the temporal feature extraction subnetwork to extract the features of the temporal type to obtain the temporal features" in the above-mentioned S404 includes:

[0099] The denoised time window sequence is input into the second attention layer for weighted processing to obtain the processed time window sequence.

[0100] The processed time window sequence is input into the time domain convolution layer for feature extraction to obtain the time series features.

[0101] In the present application embodiment, Figure 6 As shown, Figure 6FIG. 1 is a schematic diagram of the structure of a temporal feature extraction subnetwork in an embodiment. For example, the temporal feature extraction subnetwork includes a second attention layer Multi-Attention and a temporal convolution layer TCN. The specific structure of the temporal convolution layer TCN is as follows: Figure 6 As shown in the TCN Block in the figure, the time-domain convolution layer TCN includes Dilated CausalConvolutions (dilated causal convolution) sublayer, WeighNorm (weight normalization) sublayer, ReLU sublayer and Dropout sublayer. TCN is a convolutional neural network used to process time series data. TCN adapts to time series data through causal convolution and dilated convolution, which can effectively capture long-distance dependencies and significantly improve prediction accuracy. In addition, TCN also has the advantages of parallel computing, long-term memory retention, and simple and easy-to-understand models. Therefore, TCN can not only improve the performance of the time series feature extraction subnetwork, but also reduce the computational complexity during training and reasoning.

[0102] The computer device can input the denoised time window sequence into the second attention layer Multi-Attention for weighted processing to obtain a processed time window sequence, so that a larger weight can be assigned to a more important time window sequence. Thus, the computer device can input the processed time window sequence into the time domain convolution layer TCN to extract the time series type features and obtain the time series features. Among them, the processed time window sequence refers to the time window sequence after the weight is assigned, and the time series features refer to the weighted time series features.

[0103] In this embodiment, the denoised time window sequence is input into the second attention layer for weighted processing to obtain a processed time window sequence, and the second attention layer Multi-Attention can be used to assign a larger weight to the key features in the more important time window sequence. Thus, the processed time window sequence is input into the time domain convolution layer for feature extraction, and more accurate time series features can be extracted from the weighted time window sequence.

[0104] In one embodiment, the manual feature extraction subnetwork includes a feature extraction layer and an output layer. Based on this, a method for implementing manual feature extraction is provided, that is, the step of "inputting the time window sequence into the manual feature extraction subnetwork to perform manual feature extraction to obtain manual features" in S401 above, including:

[0105] The time window sequence is input into the feature extraction layer for feature extraction to obtain the initial manual features.

[0106] The initial manual features are input into the output layer for inner product merging to obtain the manual features.

[0107] In the present application embodiment, Figure 7As shown, Figure 7 The figure is a schematic diagram of the structure of a manual feature extraction subnetwork in an embodiment. For example, the manual feature extraction subnetwork may include but is not limited to a feature extraction layer and an output layer, etc. The output layer includes an activation function ReLU, a SoftMax function, and an inner product merging sublayer. Among them, the manual feature is a feature artificially designed based on domain knowledge and experience, which is highly targeted and explanatory. The manual feature can provide richer and more effective degradation information for the pump inspection cycle prediction, which helps to improve the accuracy of the pump inspection cycle prediction. For example, the manual feature may include but is not limited to the mean feature and the trend coefficient feature. The mean feature represents the overall level of the sensor parameter, that is, the magnitude of the sensor parameter, and the trend coefficient feature reflects the degradation trend of the sensor parameter.

[0108] Exemplarily, the processing formulas of the manual feature extraction sub-network are shown in the following equations (2) and (3).

[0109] (2)

[0110] (3)

[0111] In the formula, is a weighted manual feature; is the initial manual feature; is the matrix inner product operation; For the final hand-crafted features; is the weight matrix; is the bias term.

[0112] That is, it can be understood that the computer device can input the time window sequence into the feature extraction layer for feature extraction to obtain the initial manual features. After that, the initial manual features are processed by the activation function ReLU in the output layer, and the weights corresponding to the initial manual features are calculated by the SoftMax function in the output layer. Thus, the initial manual features can be input into the inner product merging sublayer in the output layer for inner product merging processing, that is, the initial manual features and the features processed by the SoftMax function can be inner-product merged, and then the features after the inner product merging processing can be processed by the fully connected layer to obtain the final manual features.

[0113] In this embodiment, the time window sequence can be input into the feature extraction layer for feature extraction to obtain initial manual features, and then the initial manual features can be input into the output layer for inner product merging processing to accurately obtain the manual features, thereby providing richer and more effective degradation information for pump inspection cycle prediction, which helps to improve the accuracy of pump inspection cycle prediction.

[0114] In one embodiment, the prediction module includes a fusion subnetwork and a regression subnetwork. Based on this, a method for implementing regression prediction is provided, namely, the step of "inputting the feature set into the prediction module for period prediction and outputting the pump inspection period of the pumping well to be tested" in S303 includes:

[0115] The feature set is input into the fusion subnetwork for feature fusion to obtain the second fusion feature.

[0116] The second fusion feature is input into the regression subnetwork for prediction to obtain the pump inspection cycle of the oil pumping unit to be tested.

[0117] In an embodiment of the present application, the computer device may input a feature set into a fusion subnetwork for feature fusion to obtain a second fused feature. Optionally, the computer device may directly input manual features, spatiotemporal features, and time series features into the fusion subnetwork at the same time for feature fusion to obtain a second fused feature; or, the computer device may first input spatiotemporal features and time series features into the fusion subnetwork at the same time for feature fusion, and then input the fused features and manual features into the fusion subnetwork at the same time for feature fusion to obtain a second fused feature. Afterwards, the computer device may input the second fused feature into the regression subnetwork for prediction of the pump inspection cycle to obtain the pump inspection cycle of the oil pump to be tested. Among them, the fusion subnetwork includes a Concatenate function, the second fused feature refers to a fused feature obtained by fusing the feature set, and the regression subnetwork may include but is not limited to a fully connected layer, etc.

[0118] In this embodiment, the feature set can be input into the fusion subnetwork for feature fusion to obtain a second fused feature. Thus, the second fused feature is input into the regression subnetwork for prediction. Based on the fusion subnetwork and the regression subnetwork, the pump inspection cycle of the oil pump to be tested can be obtained more accurately.

[0119] Based on all the above embodiments, the present application also provides a complete pump inspection cycle prediction method, such as Figure 8 As shown, the method includes:

[0120] S21, obtaining the original production data of the pumping well to be tested;

[0121] S22, inputting the original production data into the data processing module for data processing, and outputting a time window sequence;

[0122] S23, inputting the time window sequence into the manual feature extraction sub-network to perform manual type feature extraction to obtain manual features;

[0123] Among them, S23 includes:

[0124] Input the time window sequence into the feature extraction layer for feature extraction to obtain initial manual features;

[0125] The initial manual features are input into the output layer for inner product merging to obtain manual features;

[0126] S24, performing noise reduction processing on the time window sequence to obtain a noise-reduced time window sequence;

[0127] S25, inputting the denoised time window sequence into the spatiotemporal feature extraction subnetwork to extract spatiotemporal type features to obtain spatiotemporal features;

[0128] Among them, S25 includes:

[0129] The denoised time window sequence is input into each convolution layer to extract spatial features, and each extracted first feature is fused to obtain a first fused feature; the size of the convolution kernel in each convolution layer is different;

[0130] The first fusion feature is input into the bidirectional long short-term memory convolution layer to extract spatiotemporal features to obtain the second feature;

[0131] The second feature is input into the first attention layer for weighted processing to obtain the third feature;

[0132] The third feature is fused with the first fusion feature to obtain a spatiotemporal feature;

[0133] S26, inputting the denoised time window sequence into the time series feature extraction subnetwork to extract the time series type features to obtain time series features;

[0134] Among them, S26 includes:

[0135] The denoised time window sequence is input into the second attention layer for weighted processing to obtain a processed time window sequence;

[0136] The processed time window sequence is input into the time domain convolution layer for feature extraction to obtain the time series features;

[0137] S27, inputting the feature set into the fusion subnetwork for feature fusion to obtain a second fusion feature; the feature set includes manual features, spatiotemporal features, and time series features;

[0138] S28, inputting the second fused feature into the regression sub-network for prediction, and obtaining the pump inspection cycle of the oil pumping unit to be tested.

[0139] In addition, the training process of the preset prediction network is as follows: in the first step, the computer equipment can collect historical production data within the pump inspection cycle of the pumping well; in the second step, the historical production data within the pump inspection cycle of the pumping well is preprocessed to generate a training data set; in the third step, an initial prediction model of the pump inspection cycle of the pumping well based on a deep learning neural network is constructed; in the fourth step, the initial prediction model of the pump inspection cycle of the pumping well is trained using the training data set to obtain a trained preset prediction network.

[0140] In the above-mentioned pump inspection cycle prediction method, since the preset prediction network in the embodiment of the present application can perform cycle prediction based on the manual features, spatiotemporal features and time series features corresponding to the original production data, that is, it can predict the pump inspection cycle of the pumping well to be tested based on the characteristics of three dimensions, therefore, a more accurate pump inspection cycle of the pumping well to be tested can be obtained. In this way, reference suggestions can be provided for fault diagnosis and preventive maintenance of the pumping well based on the prediction results of the pump inspection cycle.

[0141] Based on the above embodiments, it can be seen that, first, the embodiments of the present application can effectively reduce the noise interference in the original production data, obtain relatively stable feature data, and thus improve the prediction accuracy of the model pump inspection cycle; second, it can fully explore the diverse representative features contained in the original production data, and can effectively alleviate the problem of low prediction accuracy of the pump inspection cycle caused by insufficient extraction of degradation information; third, a multi-scale learning strategy can be adopted to construct a spatiotemporal feature extraction subnetwork based on Conv-Bi-LSTM and the first attention layer Multi-Attention, which alleviates the problems of insufficient and unbalanced extraction of degradation information and low utilization of learning features in traditional networks; fourth, based on TCN and the second attention layer Multi-Attention, weighted time series features in the original production data are extracted, which can provide the preset prediction network with the changing trends and laws of the original production data; fifth, manual features are extracted according to the characteristics of sensor data, and additional information of different importance can be provided for the pump inspection cycle prediction, further improving the prediction accuracy of the pump inspection cycle.

[0142] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0143] Based on the same inventive concept, the embodiment of the present application also provides a pump check cycle prediction device for implementing the pump check cycle prediction method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the embodiments of one or more pump check cycle prediction devices provided below can refer to the limitations of the pump check cycle prediction method above, and will not be repeated here.

[0144] In an exemplary embodiment, Fig. 9 As shown, a pump inspection cycle prediction device is provided, including: an acquisition module 31 and a prediction module 32, wherein:

[0145] The acquisition module 31 is used to acquire the original production data of the pumping well to be tested.

[0146] The prediction module 32 is used to input the original production data into the preset prediction network for prediction to obtain the pump inspection cycle of the pumping well to be tested; the preset prediction network is used to perform cycle prediction based on the manual features, spatiotemporal features and time series features corresponding to the original production data.

[0147] In one embodiment, the preset prediction network includes a data processing module, a feature extraction module and a prediction module, and the prediction module 32 includes:

[0148] A data processing unit is used to input the original production data into the data processing module for data processing and output a time window sequence;

[0149] The feature extraction unit is used to input the time window sequence into the feature extraction module for multi-dimensional feature extraction and output a feature set; the feature set includes manual features, spatiotemporal features and time series features;

[0150] The cycle prediction unit is used to input the feature set into the prediction module for cycle prediction and output the pump inspection cycle of the pumping well to be tested.

[0151] In one embodiment, the feature extraction module includes a manual feature extraction subnetwork, a spatiotemporal feature extraction subnetwork and a temporal feature extraction subnetwork, and the feature extraction unit includes:

[0152] A manual feature extraction subunit is used to input the time window sequence into the manual feature extraction subnetwork to perform manual type feature extraction to obtain manual features;

[0153] A denoising subunit is used to perform denoising on the time window sequence to obtain a denoised time window sequence;

[0154] The spatiotemporal feature extraction subunit is used to input the denoised time window sequence into the spatiotemporal feature extraction subnetwork to extract the spatiotemporal type features and obtain the spatiotemporal features;

[0155] The time series feature extraction subunit is used to input the denoised time window sequence into the time series feature extraction subnetwork to extract the features of the time series type and obtain the time series features.

[0156] In one embodiment, the spatiotemporal feature extraction subnetwork includes a plurality of convolutional layers, a bidirectional long short-term memory convolutional layer and a first attention layer, and the spatiotemporal feature extraction subunit is specifically used to:

[0157] The denoised time window sequence is input into each convolution layer to extract spatial features, and each extracted first feature is fused to obtain a first fused feature; the size of the convolution kernel in each convolution layer is different;

[0158] The first fusion feature is input into the bidirectional long short-term memory convolution layer to extract spatiotemporal features to obtain the second feature;

[0159] The second feature is input into the first attention layer for weighted processing to obtain the third feature;

[0160] The third feature is fused with the first fusion feature to obtain the spatiotemporal feature.

[0161] In one embodiment, the temporal feature extraction subnetwork includes a second attention layer and a temporal convolution layer, and the temporal feature extraction subunit is specifically used for:

[0162] The denoised time window sequence is input into the second attention layer for weighted processing to obtain a processed time window sequence;

[0163] The processed time window sequence is input into the time domain convolution layer for feature extraction to obtain the time series features.

[0164] In one embodiment, the manual feature extraction subnetwork includes a feature extraction layer and an output layer, and the manual feature extraction subunit is specifically used for:

[0165] Input the time window sequence into the feature extraction layer for feature extraction to obtain initial manual features;

[0166] The initial manual features are input into the output layer for inner product merging to obtain the manual features.

[0167] In one embodiment, the prediction module includes a fusion sub-network and a regression sub-network, and the period prediction unit includes:

[0168] A feature fusion subunit, used for inputting the feature set into the fusion subnetwork for feature fusion to obtain a second fusion feature;

[0169] The prediction subunit is used to input the second fusion feature into the regression subnetwork for prediction, so as to obtain the pump inspection cycle of the oil pumping unit to be tested.

[0170] Each module in the above pump inspection cycle prediction device can be implemented in whole or in part by software, hardware and their combination. Each module can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the corresponding operations of each module.

[0171] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 1 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be realized through WIFI, a mobile cellular network, near field communication (NFC) or other technologies. When the computer program is executed by the processor, a method for predicting a pump inspection cycle is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0172] Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0173] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0174] Obtaining the original production data of the pumping well to be tested;

[0175] The original production data is input into the preset prediction network for prediction to obtain the pump inspection cycle of the pumping well to be tested; the preset prediction network is used to perform cycle prediction based on the manual features, spatiotemporal features and time series features corresponding to the original production data.

[0176] In one embodiment, the preset prediction network includes a data processing module, a feature extraction module and a prediction module. The original production data is input into the preset prediction network for prediction to obtain the pump inspection cycle of the pumping well to be tested. When the processor executes the computer program, the following steps are also implemented:

[0177] Input the original production data into the data processing module for data processing and output the time window sequence;

[0178] The time window sequence is input into the feature extraction module for multi-dimensional feature extraction, and a feature set is output; the feature set includes manual features, spatiotemporal features, and time series features;

[0179] The feature set is input into the prediction module for cycle prediction, and the pump inspection cycle of the pumping well to be tested is output.

[0180] In one embodiment, the feature extraction module includes a manual feature extraction subnetwork, a spatiotemporal feature extraction subnetwork, and a temporal feature extraction subnetwork. The time window sequence is input into the feature extraction module for multi-dimensional feature extraction, and a feature set is output. When the processor executes the computer program, the following steps are also implemented:

[0181] Input the time window sequence into the manual feature extraction sub-network to extract manual features and obtain manual features;

[0182] Performing denoising on the time window sequence to obtain a denoised time window sequence;

[0183] The denoised time window sequence is input into the spatiotemporal feature extraction subnetwork to extract spatiotemporal type features and obtain spatiotemporal features;

[0184] The denoised time window sequence is input into the time series feature extraction subnetwork to extract the time series type features and obtain the time series features.

[0185] In one embodiment, the spatiotemporal feature extraction subnetwork includes multiple convolutional layers, a bidirectional long short-term memory convolutional layer, and a first attention layer. The denoised time window sequence is input into the spatiotemporal feature extraction subnetwork to extract spatiotemporal features to obtain spatiotemporal features. When the processor executes the computer program, the following steps are also implemented:

[0186] The denoised time window sequence is input into each convolution layer to extract spatial features, and each extracted first feature is fused to obtain a first fused feature; the size of the convolution kernel in each convolution layer is different;

[0187] The first fusion feature is input into the bidirectional long short-term memory convolution layer to extract spatiotemporal features to obtain the second feature;

[0188] The second feature is input into the first attention layer for weighted processing to obtain the third feature;

[0189] The third feature is fused with the first fusion feature to obtain the spatiotemporal feature.

[0190] In one embodiment, the temporal feature extraction subnetwork includes a second attention layer and a time domain convolution layer, and the denoised time window sequence is input into the temporal feature extraction subnetwork to extract the time series type feature to obtain the time series feature, and the processor further implements the following steps when executing the computer program:

[0191] The denoised time window sequence is input into the second attention layer for weighted processing to obtain a processed time window sequence;

[0192] The processed time window sequence is input into the time domain convolution layer for feature extraction to obtain the time series features.

[0193] In one embodiment, the manual feature extraction subnetwork includes a feature extraction layer and an output layer, and the time window sequence is input into the manual feature extraction subnetwork to perform manual type feature extraction to obtain manual features. When the processor executes the computer program, the following steps are also implemented:

[0194] Input the time window sequence into the feature extraction layer for feature extraction to obtain initial manual features;

[0195] The initial manual features are input into the output layer for inner product merging to obtain the manual features.

[0196] In one embodiment, the prediction module includes a fusion subnetwork and a regression subnetwork, inputs the feature set into the prediction module for cycle prediction, and outputs the pump inspection cycle of the pumping well to be tested. When the processor executes the computer program, the following steps are also implemented:

[0197] Input the feature set into the fusion subnetwork for feature fusion to obtain the second fusion feature;

[0198] The second fusion feature is input into the regression subnetwork for prediction to obtain the pump inspection cycle of the oil pumping unit to be tested.

[0199] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0200] Obtaining the original production data of the pumping well to be tested;

[0201] The original production data is input into the preset prediction network for prediction to obtain the pump inspection cycle of the pumping well to be tested; the preset prediction network is used to perform cycle prediction based on the manual features, spatiotemporal features and time series features corresponding to the original production data.

[0202] In one embodiment, the preset prediction network includes a data processing module, a feature extraction module and a prediction module. The original production data is input into the preset prediction network for prediction to obtain the pump inspection cycle of the pumping well to be tested. When the computer program is executed by the processor, the following steps are also implemented:

[0203] Input the original production data into the data processing module for data processing and output the time window sequence;

[0204] The time window sequence is input into the feature extraction module for multi-dimensional feature extraction, and a feature set is output; the feature set includes manual features, spatiotemporal features, and time series features;

[0205] The feature set is input into the prediction module for cycle prediction, and the pump inspection cycle of the pumping well to be tested is output.

[0206] In one embodiment, the feature extraction module includes a manual feature extraction subnetwork, a spatiotemporal feature extraction subnetwork, and a temporal feature extraction subnetwork. The time window sequence is input into the feature extraction module for multi-dimensional feature extraction, and a feature set is output. When the computer program is executed by the processor, the following steps are also implemented:

[0207] Input the time window sequence into the manual feature extraction sub-network to extract manual features and obtain manual features;

[0208] Performing denoising on the time window sequence to obtain a denoised time window sequence;

[0209] The denoised time window sequence is input into the spatiotemporal feature extraction subnetwork to extract spatiotemporal type features and obtain spatiotemporal features;

[0210] The denoised time window sequence is input into the time series feature extraction subnetwork to extract the time series type features and obtain the time series features.

[0211] In one embodiment, the spatiotemporal feature extraction subnetwork includes multiple convolutional layers, a bidirectional long short-term memory convolutional layer, and a first attention layer. The denoised time window sequence is input into the spatiotemporal feature extraction subnetwork to extract spatiotemporal features to obtain spatiotemporal features. When the computer program is executed by the processor, the following steps are also implemented:

[0212] The denoised time window sequence is input into each convolution layer to extract spatial features, and each extracted first feature is fused to obtain a first fused feature; the size of the convolution kernel in each convolution layer is different;

[0213] The first fusion feature is input into the bidirectional long short-term memory convolution layer to extract spatiotemporal features to obtain the second feature;

[0214] The second feature is input into the first attention layer for weighted processing to obtain the third feature;

[0215] The third feature is fused with the first fusion feature to obtain the spatiotemporal feature.

[0216] In one embodiment, the temporal feature extraction subnetwork includes a second attention layer and a time domain convolution layer, and the denoised time window sequence is input into the temporal feature extraction subnetwork to extract the time series type feature to obtain the time series feature, and the computer program further implements the following steps when executed by the processor:

[0217] The denoised time window sequence is input into the second attention layer for weighted processing to obtain a processed time window sequence;

[0218] The processed time window sequence is input into the time domain convolution layer for feature extraction to obtain the time series features.

[0219] In one embodiment, the manual feature extraction subnetwork includes a feature extraction layer and an output layer, and the time window sequence is input into the manual feature extraction subnetwork to perform manual type feature extraction to obtain manual features. When the computer program is executed by the processor, the following steps are also implemented:

[0220] Input the time window sequence into the feature extraction layer for feature extraction to obtain initial manual features;

[0221] The initial manual features are input into the output layer for inner product merging to obtain the manual features.

[0222] In one embodiment, the prediction module includes a fusion subnetwork and a regression subnetwork, inputs the feature set into the prediction module for cycle prediction, and outputs the pump inspection cycle of the pumping well to be tested. When the computer program is executed by the processor, the following steps are also implemented:

[0223] Input the feature set into the fusion subnetwork for feature fusion to obtain the second fusion feature;

[0224] The second fusion feature is input into the regression subnetwork for prediction to obtain the pump inspection cycle of the oil pumping unit to be tested.

[0225] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0226] It should be noted that the user information (including but not limited to user terminal information, user device information, user personal information, etc.) and data (including but not limited to the original production data of the tested pumping wells used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0227] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0228] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0229] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for predicting a pump inspection cycle, characterized in that: The method comprises: Obtaining the original production data of the pumping well to be tested; The original production data is input into a preset prediction network for prediction to obtain the pump inspection cycle of the pumping well to be tested; the preset prediction network is used to perform cycle prediction based on manual features, spatiotemporal features and time series features corresponding to the original production data.

2. The method according to claim 1, characterized in that The preset prediction network includes a data processing module, a feature extraction module and a prediction module. The raw production data is input into the preset prediction network for prediction to obtain the pump inspection cycle of the pumping well to be tested, including: Inputting the original production data into the data processing module for data processing and outputting a time window sequence; Input the time window sequence into the feature extraction module to perform multi-dimensional feature extraction, and output a feature set; the feature set includes the manual features, the spatiotemporal features, and the time series features; The feature set is input into the prediction module for cycle prediction, and the pump inspection cycle of the pumping well to be tested is output.

3. The method according to claim 2, characterized in that The feature extraction module includes a manual feature extraction subnetwork, a spatiotemporal feature extraction subnetwork and a temporal feature extraction subnetwork. The time window sequence is input into the feature extraction module for multi-dimensional feature extraction, and a feature set is output, including: Inputting the time window sequence into the manual feature extraction subnetwork to perform manual type feature extraction to obtain the manual features; Performing noise reduction processing on the time window sequence to obtain a noise-reduced time window sequence; Inputting the denoised time window sequence into the spatiotemporal feature extraction subnetwork to extract spatiotemporal type features to obtain the spatiotemporal features; The denoised time window sequence is input into the time series feature extraction subnetwork to extract the time series type features to obtain the time series features.

4. The method according to claim 3, characterized in that The spatiotemporal feature extraction subnetwork includes a plurality of convolutional layers, a bidirectional long short-term memory convolutional layer and a first attention layer. The de-noised time window sequence is input into the spatiotemporal feature extraction subnetwork to extract the spatiotemporal type of features to obtain the spatiotemporal features, including: Inputting the denoised time window sequence into each of the convolutional layers to extract spatial features, and performing feature fusion on each of the extracted first features to obtain a first fused feature; the size of the convolution kernel in each of the convolutional layers is different; Inputting the first fusion feature into the bidirectional long short-term memory convolutional layer to extract spatiotemporal features to obtain a second feature; Inputting the second feature into the first attention layer for weighted processing to obtain a third feature; The third feature and the first fusion feature are fused to obtain the spatiotemporal feature.

5. The method according to claim 3, characterized in that: The temporal feature extraction subnetwork includes a second attention layer and a time domain convolution layer, and the denoised time window sequence is input into the temporal feature extraction subnetwork to extract the feature of the temporal type to obtain the temporal feature, including: Inputting the denoised time window sequence into the second attention layer for weighted processing to obtain a processed time window sequence; The processed time window sequence is input into the time domain convolution layer for feature extraction to obtain the time series feature.

6. The method according to claim 3, characterized in that The manual feature extraction subnetwork includes a feature extraction layer and an output layer, and the time window sequence is input into the manual feature extraction subnetwork to perform manual type feature extraction to obtain the manual features, including: Inputting the time window sequence into the feature extraction layer to extract features and obtain initial manual features; The initial manual features are input into the output layer for inner product merging processing to obtain the manual features.

7. The method according to any one of claims 2 to 6, characterized in that: The prediction module includes a fusion sub-network and a regression sub-network. The feature set is input into the prediction module for period prediction, and the pump inspection period of the pumping well to be tested is output, including: Inputting the feature set into the fusion subnetwork for feature fusion to obtain a second fusion feature; The second fusion feature is input into the regression sub-network for prediction to obtain the pump inspection cycle of the oil pumping unit to be tested.

8. A pump inspection cycle prediction device, characterized in that: The device comprises: An acquisition module is used to acquire the original production data of the pumping well to be tested; The prediction module is used to input the original production data into a preset prediction network for prediction to obtain the pump inspection cycle of the pumping well to be tested; the preset prediction network is used to perform cycle prediction based on the manual features, spatiotemporal features and time series features corresponding to the original production data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images