Oil well intelligent control terminal based on edge cloud computing technology

Abstract

The invention provides an intelligent oil well control terminal based on a side cloud computing technology. The intelligent oil well control terminal comprises an acquisition unit, an equipment control unit, a data processing unit, a communication unit and a power supply unit. By moving production recording analysis down to a side end, platform operation pressure, transmission pressure and unreliability of transmission are reduced, integrity of sample data of a production recording algorithm is guaranteed, a data secure transmission mode is provided, data transmission levels are reduced, intermediate links are reduced, data transmission efficiency and stability are improved, and field operation workload is reduced.

Description

Technical Field

[0001] The present invention belongs to the field of oil and gas well production, and particularly relates to an intelligent control terminal for oil wells based on edge cloud computing technology. Background Art

[0002] Changqing Oilfield has built the largest oil and gas production Internet of Things system in China. The main control terminal (RTU) is installed at the well site, which is mainly responsible for collecting, processing, and uploading data such as oil production well and water injection valve group data, well site pressure, and video, and communicating with the upper computer to achieve remote management. The main RTU cabinet at the well site mainly consists of components such as RTU (Remote Terminal Unit), industrial switch, various equipment power supplies, air switches, serial port servers, surge protectors, voice power amplifiers, light control switches, video servers, terminal blocks, and waterproof distribution boxes. The following problems mainly exist:

[0003] ① The main control unit belongs to an early technology product, and there is still a large room for improvement in its storage space, processing speed, interface indicators, etc., and it cannot meet the need for further improvement of functions; the communication method has poor compatibility, and there are protocol restrictions for the extended access of third-party access devices;

[0004] ② The existing main control cabinet at the well site integrates multiple power supplies inside, with many components, integrating multiple manufacturers, and the internal wiring is complex, resulting in a large amount of maintenance work; the equipment debugging and fault diagnosis have high technical content, there is a lack of effective on-site detection means for components, the component replacement volume is large, and the maintenance cost is high; the main RTU cabinet at the well site is installed on a cement pole, and maintenance operations require climbing heights, with high safety risks;

[0005] ③ Using the RS232 serial cable for on-site debugging and program upgrading, it does not have the ability of remote debugging and program upgrading; it does not have the functions of fault diagnosis and autonomous management, and cannot meet the requirements of digital oilfield and unattended management of the station.

[0006] ④ Due to limited hardware resources of early products, multifunctional expansion cannot be carried out, such as built-in embedded web server, edge computing capabilities such as data diagnosis, large-capacity data storage capabilities, and operation and maintenance log recording capabilities.

[0007] Ultimately, it leads to less sampled data uploaded, affecting the quantity and quality of the dynamometer cards drawn by the cloud based on the sampled data, and making it impossible to improve the accuracy of fault diagnosis through the dynamometer cards. Summary of the Invention

[0008] The purpose of the present invention is to provide an intelligent control terminal for oil wells based on edge cloud computing technology to realize on-site analysis of the production conditions of oil and water wells.

[0009] The object of the present invention is achieved by the following technical means. An intelligent control terminal for oil wells based on edge-cloud computing technology includes a collection unit, a device control unit, a data processing unit, a communication unit, and a power supply unit.

[0010] The collection unit is used to collect production data of oil wells and water source wells.

[0011] The device control unit is used to control oil well and water source well equipment.

[0012] The data processing unit is used to draw a dynamometer card based on the data collected by the collection unit and diagnose faults based on a residual convolutional neural network model.

[0013] The communication unit is used to interact data with the cloud, upload the fault diagnosis results of the data processing unit and the data collected by the collection unit, and receive the oil well control instructions issued by the cloud, and control the operation of oil well and water source well equipment through the device control unit.

[0014] The power supply unit supplies power to the collection unit, the device control unit, the data processing unit, and the communication unit.

[0015] The collection unit collects production data of oil wells and water source wells through wireless Zigbee and RS485 serial ports.

[0016] The device control unit conducts data interaction with oil well and water source well equipment through wireless Zigbee to control the normal start / stop and timed start / stop of pumping units and water pumps.

[0017] The communication unit is also connected to the well site monitoring camera and the well site video server through Ethernet and transmits them to the cloud through the communication unit.

[0018] The specific steps for the data processing unit to diagnose faults are as follows: perform sliding filtering on the load data queue and displacement data queue collected by the collection unit to obtain the original filtered data, draw a dynamometer card based on the original filtered data, input the dynamometer card into the residual convolutional neural network model, and output the type of the dynamometer card, that is, the fault type.

[0019] For the residual convolutional neural network model, the transfer function of the hidden layer neurons uses the tangent sigmoid, the transfer function of the output layer neurons is the logarithmic sigmoid, the training function is trainlm, the backpropagation algorithm is Levenberg-Marquadt, and the feature point data marked in each dynamometer card in the dynamometer card knowledge base is extracted and input into the neural network for training until the neural network reaches the predetermined recognition error, and the training of the residual convolutional neural network model is completed.

[0020] The specific structure of the residual convolutional neural network model is as follows: input, convolutional layer, SE-residual module, max pooling, batch normalization, SE-residual module, SE-residual module, max pooling, batch normalization, SE-residual module, SE-residual module, fully connected layer, fully connected layer, fully connected layer, output. The convolutional layer contains 16 convolutional kernels of size 3×3; the 5 SE-residual modules contain two convolutional layers, and the sizes of the convolutional kernels are 1×1 and 3×3 respectively, and the numbers of convolutional kernels are 32, 64, 64, 128, and 128 in sequence; the convolutional layer sets the convolution stride to 1, adds L2 regularization, and uses LeakyReLu as the activation function; the filter size of max pooling is 2×2, and the pooling stride is set to 2; the numbers of neurons in the fully connected layers are 1024, 512, and 4 in sequence, and the activation functions are LeakyReLu, LeakyReLu, and softmax in sequence.

[0021] It also includes a voice unit, which is connected to the wellsite alarm device through an audio interface and is used to generate voice and give an alarm prompt through the wellsite alarm device according to the diagnosed fault type.

[0022] The beneficial effects of the present invention are as follows: By moving the production record analysis to the edge end, the computing pressure, transmission pressure, and transmission unreliability of the platform are reduced. While ensuring the integrity of the sample data of the production record algorithm, it also has a data security transmission mode, reduces the data transmission level, reduces intermediate links, improves data transmission efficiency and stability, and reduces the on-site operation workload. At the same time, by directly drawing the dynamometer card at the edge end, the number of sampling points of the dynamometer card is increased, and the accuracy of the dynamometer card is improved. Description of the Drawings

[0023] Figure 1 It is a structure diagram of an oil well intelligent control terminal based on edge-cloud computing technology;

[0024] Figure 2 It is a connection diagram of an oil well intelligent control terminal based on edge-cloud computing technology and external devices;

[0025] Figure 3 It is a feature diagram and a simplified structure diagram of a data processing unit;

[0026] Figure 4 It is a structure diagram of a data processing unit module;

[0027] The present invention will be further described in detail below with reference to the drawings and embodiments. Specific Embodiments

[0028]

Embodiment 1

[0029] As Figure 1As shown in the figure, an intelligent control terminal for oil wells based on edge-cloud computing technology includes a collection unit, a device control unit, a data processing unit, a communication unit, and a power supply unit.

[0030] The collection unit is used to collect production data of oil wells and water source wells.

[0031] The device control unit is used to control oil well and water source well equipment.

[0032] The data processing unit is used to draw a dynamometer card based on the data collected by the collection unit and diagnose faults based on a residual convolutional neural network model.

[0033] The communication unit is used to interact data with the cloud, upload the fault diagnosis results of the data processing unit and the data collected by the collection unit, and receive the oil well control instructions issued by the cloud, and control the operation of oil well and water source well equipment through the device control unit.

[0034] The power supply unit supplies power to the collection unit, the device control unit, the data processing unit, and the communication unit.

[0035] As Figure 2 shown in the figure, the collection unit is connected to oil well equipment through wireless Zigbee and RS485 to collect production data of oil wells and water source wells. The main oil well data collected includes but is not limited to wireless oil pressure, casing pressure, wireless integrated load (or dynamometer card data after sampling and filtering), radio parameters, etc. The collected data includes water source well outlet pressure, water source well flow rate, water source well deep water level, etc.; sensors corresponding to data such as injection pressure, cumulative flow rate, and instantaneous flow rate.

[0036] The device control unit conducts data interaction with oil well and water source well equipment through wireless Zigbee to control the normal start / stop and timed start / stop of pumping units and water pumps. And the setting of injection volume or the realization of source supply and injection allocation through preset phased injection targets.

[0037] The communication unit is also connected to the well site monitoring camera and the well site video server through Ethernet and transmitted to the cloud through the communication unit. Realize data interaction between monitoring data and the video server of the oil production plant. Carry out fixed-point cruising of the pumping unit based on the functions of the camera. After diagnosing faults, managers can remotely conduct remote inspections of the basic working conditions of the pumping unit, improving the visualization of working condition diagnosis.

[0038] It also includes a voice unit, which is connected to the wellsite alarm device through an audio interface and is used to generate voice according to the diagnosed fault type and give an alarm prompt through the wellsite alarm device. The tts voice synthesis technology transplanted by the nuclear board is used to play the emergency level warning information within the effective well-watching range. It is controlled through an audio interface with the external alarm horn outside the wellsite to realize different voice wellsite warning functions. For example, in the conventional production mode, high-priority abnormal working conditions (such as broken load, loose belt, etc.) can be output through the wellsite power amplifier to prompt the well-watching staff to repair in time.

[0039] The equipment control unit and the data processing unit are integrated on the RTU nuclear board, which is a wide-temperature programmable control board. The embedded Linux operating system is transplanted at the system layer, and its characteristics are as follows:

[0040] (1) It provides powerful network functions, supports the TCP / IP protocol and other protocols, provides TCP / UDP / IP / PPP protocol support and a unified MAC access layer interface, and reserves interfaces for various mobile computing devices;

[0041] (2) Strong stability and weak interactivity. Once the embedded system starts running, it does not require too much intervention from the user. This requires the EOS responsible for system management to have strong stability. The user interface of the embedded operating system generally does not provide operation commands, and it provides services to user programs through system call commands;

[0042] (3) Strong real-time performance. The real-time performance is generally strong and can be used in various device controls;

[0043] (4) Unified interface. It provides various device driver interfaces;

[0044] (5) Removability. An open and scalable architecture.

[0045] An embedded web service is built in, and it can be accessed through the browser http protocol to realize functions such as configuration of the operating parameters of terminal devices, remote upgrade of devices, operation process, wellsite-level device management, and maintenance log recording.

[0046] The communication unit realizes the data interaction function with the data terminal server of the oil production plant responsible for production management in the oilfield through the ModusTCP protocol, specifically realizing the functions of online data collection and control or data regulation of the downlink, such as starting and stopping control of pumping units, balance adjustment, frequency control, or realizing layered water injection allocation, conventional water injection allocation, etc.

[0047] Through the MQTT / AMQP protocol, as the basic bearer link for secure transmission management, it is applicable to the transmission specifications of the existing Internet of Things systems in oilfields. Using MQTT / AMQP facilitates the upload of statistical analysis data results calculated by edge computing at the edge to the cloud server used by the oilfield company for central indicator monitoring. On the one hand, it can be applied to low-power or unstable network environments. In a distributed system, components need to exchange information while maintaining loose coupling, and the AMQP transmission protocol has powerful functions and complex routing capabilities. It enables these devices to communicate efficiently and reliably, improving the reliability and efficiency of uploading the computing results at the edge in the cloud-edge collaboration structure.

[0048] The power supply unit, namely the power management board, supports multiple power supply specifications such as input 220VAC / DC24V and output DC48V / DC24V / DC12V, and integrates SPD (surge protection). It is mainly used to limit the transient overvoltage (i.e., lightning surge) caused by lightning and most of the switching overvoltages in the power supply system to ensure the normal operation of the equipment in harsh environments and support the dynamic reading and monitoring functions of the power supply status.

[0049] The acquisition unit accesses third-party data through wired / wireless transparent transmission and programming, including IEEE802.15.4 protocol and zigbee protocol; wired transmission supports the transparent transmission access and programming access of wired meters RS485 to various digital devices and meters in the well site, including but not limited to wellhead collectors, water distribution room RTUs, water source well controls, digital pumping unit control cabinets, etc.

[0050] The communication unit is integrated on the network communication board, matching the communication method in the oilfield field, providing optical fiber ST interfaces, Ethernet interfaces, etc., supporting POE power supply for the bridge, and its control strategy provides functions such as network security protection, identity authentication, asymmetric encryption, danger prevention rule library, protocol filtering, black and white list control, etc. It is mainly used for the centralized access and management of well site network devices;

[0051] In the existing digital structure, the dynamometer card is uploaded layer by layer to the data terminal server of the oil production plant after sampling and filtering. During the transmission process, due to problems such as network quality and a large number of transmission nodes, the original data samples are insufficient, thus affecting the reliability of production record analysis; by drawing and diagnosing the dynamometer card on the edge device, the production record analysis is moved down to the edge to reduce the computing pressure, transmission pressure and transmission unreliability of the platform, ensuring the integrity of the sample data of the production record algorithm while also having a data security transmission mode.

[0052] The specific steps for the data processing unit to diagnose faults are as follows: perform sliding filtering on the load data queue and displacement data queue collected by the acquisition unit to obtain the original filtered data, draw a dynamometer card based on the original filtered data, and input the dynamometer card into the residual convolutional neural network model to output the type of the dynamometer card, that is, the fault type.

[0053] As Figure 3 and Figure 4 shown, for the residual convolutional neural network model, the transfer function of the hidden layer neurons uses the tangent sigmoid, the transfer function of the output layer neurons is the logarithmic sigmoid, the training function is trainlm, the backpropagation algorithm is Levenberg-Marquadt. Feature point data extracted from each dynamogram annotated in the dynamogram knowledge base is input into the neural network for training until the neural network reaches a predetermined recognition error, and the training of the residual convolutional neural network model is completed.

[0054] The specific structure of the residual convolutional neural network model is as follows: input, convolutional layer, SE-residual module, max pooling (pooling layer), batch normalization (normalization layer), SE-residual module, SE-residual module, max pooling, batch normalization, SE-residual module, SE-residual module, fully connected layer, fully connected layer, fully connected layer, output. The convolutional layer contains 16 convolutional kernels of size 3×3; the 5 SE-residual modules contain two convolutional layers, and the sizes of the convolutional kernels are 1×1 and 3×3 respectively, and the numbers of convolutional kernels are 32, 64, 64, 128, and 128 in sequence; the convolutional layer sets the convolution stride to 1, adds L2 regularization, and LeakyReLu is used as the activation function; the filter size of max pooling is 2×2, and the pooling stride is set to 2; the numbers of neurons in the fully connected layers are 1024, 512, and 4 in sequence, and the activation functions are LeakyReLu, LeakyReLu, and softmax respectively.

[0055] First, for the load data queue and displacement data queue, a sliding filtering method is adopted. The data in the load array and displacement array are compared by sliding, and a deviation comparison method is used to compare the deviation between a certain data and its adjacent data. If the deviation > 2Kn, then this data is considered an outlier, and the outliers in the array are marked. Secondly, the outlier points are removed, and the data before the outlier data is used to replace the outlier data to ensure the integrity of the data and its corresponding relationship with the load data. After the digital filtering calculation is completed, it is put back into the array to obtain the original filtered data.

[0056] According to the fact that the horizontal coordinate data of the dynamogram is obtained through a displacement sensor and the vertical coordinate data of the dynamogram is obtained through a load sensor, it is beneficial to draw the dynamogram with the original filtered data.

[0057] The model for indicator diagram recognition is a residual convolutional neural network model. The transfer function of the neurons in the hidden layer uses the tangent sigmoid, the transfer function of the neurons in the output layer uses the logarithmic sigmoid, the training function uses trainlm, and the backpropagation algorithm uses the Levenberg-Marquadt with a relatively fast convergence speed. Feature point data is extracted from each indicator diagram marked with work in the knowledge base and input into the neural network for training until the neural network reaches the predetermined recognition error. The knowledge base is a set of indicator diagram samples formed by business experts separating and marking indicator diagrams, followed by image enhancement, and continuously enriched and improved with the help of successfully recognized indicator diagrams, and finally formed.

[0058] The intelligent control terminal of the oil well based on edge-cloud computing technology integrates a residual convolutional neural network model. The drawn indicator diagram data is normalized, feature point data is extracted, and the feature point data is input into the diagnostic model to complete the preliminary intelligent working condition diagnosis. After the indicator diagram data and the diagnostic results are uploaded to the cloud, the cloud can also complete further diagnosis by comparing with the indicator diagram knowledge base to improve the diagnostic results of the indicator diagram.

[0059] The feature recalibration residual convolutional neural network model has a total of 14 layers. Among them, the convolutional layer contains 16 convolutional kernels of size 3×3; 5 SE-residual modules (the residual module embeds the Sequeeze-and-Excitation substructure) contain two convolutional layers, and the sizes of the convolutional kernels are 1×1 and 3×3 respectively, and the numbers of convolutional kernels are 32, 64, 64, 128, and 128 in sequence; the convolutional layer sets the convolution stride to 1, adds L2 regularization, and LeakyReLu is used as the activation function; the size of the pooling filter is 2×2, and the pooling stride is set to 2; the numbers of neurons in the fully connected layer are 1024, 512, and 4 in sequence, and the activation functions are LeakyReLu, LeakyReLu, and softmax in sequence. The input of the model is the indicator diagram, and the output is the type of the indicator diagram (i.e., the fault type).

Claims

1. An intelligent control terminal for oil wells based on edge-cloud computing technology, characterized in that: it includes a collection unit, a device control unit, a data processing unit, a communication unit and a power supply unit, The collection unit is used to collect production data of oil wells and water source wells; The device control unit is used to control oil well and water source well equipment, The data processing unit is used to draw a dynamometer card based on the data collected by the collection unit and diagnose faults based on a residual convolutional neural network model; The communication unit is used to interact data with the cloud, upload the fault diagnosis results of the data processing unit and the data collected by the collection unit, and receive the oil well control instructions issued by the cloud, and control the operation of the oil well and water source well equipment through the device control unit; The power supply unit supplies power to the collection unit, the device control unit, the data processing unit and the communication unit.

2. An intelligent control terminal for oil wells based on edge-cloud computing technology according to claim 1, characterized in that: The collection unit collects production data of oil wells and water source wells through wireless Zigbee and RS485.

3. An intelligent control terminal for oil wells based on edge-cloud computing technology according to claim 1, characterized in that: The device control unit interacts data with oil well and water source well equipment through wireless Zigbee to control the normal start / stop and timed start / stop of pumping units and water pumps.

4. An intelligent control terminal for oil wells based on edge-cloud computing technology according to claim 1 or 2, characterized in that: The communication unit is also connected to the well site monitoring camera and the well site video server through Ethernet and transmitted to the cloud through the communication unit.

5. An intelligent control terminal for oil wells based on edge-cloud computing technology according to claim 1, characterized in that: The specific steps for the data processing unit to diagnose faults are as follows: perform sliding filtering on the load data queue and displacement data queue collected by the collection unit to obtain the original filtered data, draw a dynamometer card based on the original filtered data, input the dynamometer card into the residual convolutional neural network model, and output the type of the dynamometer card, that is, the fault type.

6. An intelligent control terminal for oil wells based on edge-cloud computing technology according to claim 1 or 5, characterized in that: For the residual convolutional neural network model, the transfer function of the hidden layer neurons uses the tangent S-type, the transfer function of the output layer neurons is the logarithmic S-type, the training function is trainlm, the backpropagation algorithm is Levenberg-Marquadt, and the feature point data marked in each dynamometer card in the dynamometer card knowledge base is input into the neural network for training until the neural network reaches the predetermined recognition error, and the training of the residual convolutional neural network model is completed.

7. An intelligent control terminal for oil wells based on edge-cloud computing technology according to claim 6, characterized in that: The specific structure of the residual convolutional neural network model is as follows: input, convolutional layer, SE-residual module, max pooling, batch normalization, SE-residual module, SE-residual module, max pooling, batch normalization, SE-residual module, SE-residual module, fully connected layer, fully connected layer, fully connected layer, output. The convolutional layer contains 16 convolutional kernels of size 3×3; the 5 SE-residual modules contain two convolutional layers, and the sizes of the convolutional kernels are 1×1 and 3×3 respectively, and the numbers of convolutional kernels are 32, 64, 64, 128, and 128 in sequence; the convolutional layer is set with a convolution stride of 1, L2 regularization is added, and LeakyReLu is used as the activation function; the filter size of max pooling is 2×2, and the pooling stride is set to 2; the numbers of neurons in the fully connected layers are 1024, 512, and 4 in sequence, and the activation functions are LeakyReLu, LeakyReLu, and softmax in sequence.

8. An intelligent control terminal for oil wells based on edge cloud computing technology according to claim 1, characterized in that: it further includes a voice unit, and the voice unit is connected to the well site alarm device through an audio interface, and is used to generate voice and alarm through the well site alarm device according to the diagnosed fault type.

Images