An intelligent control system and method for flow controller based on edge computing

Abstract

This invention belongs to the field of oil well water injection flow monitoring technology, specifically disclosing an intelligent control system and method for a flow controller based on edge computing. The system includes a cloud server, edge computing modules deployed in each water injection well, and an industrial Internet of Things (IoT) connecting the two. The edge computing modules collect and process multi-source water injection data in real time, extracting feature datasets including equipment performance degradation coefficients. Based on the matching degree between the features and pre-stored operating condition templates, the system adaptively selects or fuses and calls corresponding control sub-models to generate target control commands and drive the flow controller. The cloud server analyzes well group data, identifies inter-well interference, and generates and distributes collaborative control strategies. This invention achieves a unified approach to precise water injection well flow, adaptive control, and well group collaborative optimization, significantly improving control accuracy, system stability, and safety.

Description

Technical Field

[0001] This invention belongs to the field of oil well water injection flow monitoring technology, and specifically relates to an intelligent control system and method for a flow controller based on edge computing. Background Technology

[0002] In oilfield production, precise and stable control of the injection flow rate in injection wells is crucial for maintaining formation pressure and enhancing oil recovery. Currently, most injection well flow controllers use traditional PID controllers or PLCs with preset fixed programs. However, these control methods suffer from low accuracy, slow response, and poor adaptability. Specifically, PID controller parameters are typically fixed values ​​or only support limited manual tuning, failing to adapt to dynamic changes in formation pressure, fluctuations in injection water quality, and equipment wear. This results in poor control performance under conditions of fluctuating injection pressure and changes in formation absorbency. Furthermore, traditional PID control logic is simplistic, focusing solely on the injection flow rate setpoint as the control target. It fails to comprehensively consider the coupling relationships between multiple parameters such as injection pressure, wellhead pressure, formation pressure, and injection temperature, making it difficult to achieve precise and stable control of the injection process. PLC control logic updates with preset fixed programs rely on manual on-site debugging, which cannot adapt to dynamic changes in injection conditions and inter-well interference, impacting the effectiveness of injection development. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the existing technology and provide an intelligent control system for flow controllers based on edge computing.

[0004] In a first aspect, the present invention provides an intelligent control system for a flow controller based on edge computing, comprising a cloud server, edge computing modules installed in each injection well, and an industrial Internet of Things for connecting the cloud server and the edge computing modules; the edge computing modules have a built-in preset local control model, and the output of the local control model is connected to the flow controller. The edge computing module is used to collect multi-source water injection data of the corresponding water injection well in real time, and to process and extract features from the multi-source water injection data in real time to generate a feature dataset that represents the current water injection status and equipment status. The local control model includes multiple pre-trained control sub-models for different water injection conditions. Each control sub-model stores a corresponding condition feature template. The local control model selects a control sub-model based on the fit between the feature dataset and each condition feature template. The selected control sub-model is used as the master control model to generate target control commands. The target control commands are used to adjust the target opening of the flow controller. The cloud server receives multi-source water injection data from multiple edge computing modules, identifies the mutual influence between water injection well groups based on the multi-source water injection data, and generates collaborative control commands.

[0005] A further embodiment is that the edge computing module is connected to the sensor group of the corresponding injection well and the drive component of the flow controller. The sensor group is used to collect data on injection pressure, wellhead pressure, real-time injection flow rate, injection temperature, and formation pressure. The drive component is used to receive the target control command or the cooperative control command, and drive the flow controller to perform the opening adjustment action.

[0006] A further embodiment is that the edge computing module includes: The data acquisition unit is used to read the data of the sensor group and the current opening degree of the flow controller in real time at a first preset cycle to form the original multi-source water injection dataset; The data processing unit is used to filter, convert units and align time on the original multi-source water injection dataset, and extract and generate a feature dataset. The feature dataset includes at least: instantaneous water injection flow deviation, water injection flow deviation change rate, instantaneous value and change rate of water injection pressure, instantaneous value and change rate of wellhead pressure, formation pressure and water injection pressure difference, and equipment performance degradation coefficient determined based on the historical opening degree-water injection flow relationship. The instruction execution unit is used to send the target control instruction to the drive component to perform the opening change action and monitor the instruction execution status.

[0007] A further embodiment is that the edge computing module also includes an optimization unit, which is used to evaluate the control effect based on the working condition data collected in the second preset cycle after the opening change action is completed. If the control effect does not meet expectations, a weighting coefficient is assigned to the opening adjustment amount in the next control cycle based on the mapping relationship between the opening adjustment amount and the control effect.

[0008] A further solution is that the method for the data processing unit to extract the determination of the device performance degradation coefficient includes: Under stable water injection conditions, record the flow controller opening value and the corresponding stable water injection flow value within the historical time window; By performing curve fitting between the opening value and the stable water injection flow rate value, the current actual water injection flow rate-opening characteristic curve is obtained; The current actual water injection flow rate-opening characteristic curve is compared with the preset standard characteristic curve or the characteristic curve obtained from the previous fitting. The deviation of the curve shape or the rate of change of gain at key points is calculated to obtain the performance degradation coefficient of the equipment.

[0009] A further proposed solution is that the process of selecting a control sub-model by the local control model is as follows: The similarity of each feature in the feature dataset with several pre-stored working condition feature templates is calculated. The working condition feature templates include: normal steady-state water injection working condition, water injection pressure fluctuating rapidly working condition, formation water absorption performance gradually changing working condition, equipment performance degradation working condition, and adjacent well injection and production interference working condition. If the overall similarity of a certain working condition feature template exceeds the first similarity threshold, then the dedicated control sub-model bound to that template is selected as the main control model. If the overall similarity of all working condition feature templates is lower than the first similarity threshold, but the single feature similarity of multiple templates exceeds the second similarity threshold, then multiple control sub-models associated with these high-similarity single features are activated, and the fusion weight is calculated based on the similarity of each single feature, and the outputs of multiple control sub-models are weighted and fused.

[0010] A further embodiment is that the control sub-model includes: The adjustment control unit is used to determine the difference between the instantaneous water injection flow rate deviation and the preset deviation. ,like If the value is greater than or equal to the first difference threshold, output a first injection flow compensation amount that is linearly related to the instantaneous injection flow deviation; if... If the value is less than the first difference threshold, output a second water injection flow compensation amount that is linearly related to the instantaneous water injection flow deviation; The compensation optimization unit receives the instantaneous injection flow rate deviation, injection pressure, and formation pressure to generate a comprehensive deviation signal. Within the first preset period, it performs an accumulation calculation on the current comprehensive deviation to obtain the cumulative compensation amount. Based on the cumulative compensation amount, it sets an incremental compensation coefficient and uses the product of the cumulative compensation amount and the incremental compensation coefficient as the third injection flow rate compensation amount.

[0011] The compensation buffer unit determines the deviation trend based on the rate of change of the water injection flow rate and the rate of change of the water injection pressure. If the deviation is going to increase, it outputs a negative buffer corresponding to the deviation trend; if the deviation is going to decrease, it outputs a positive buffer corresponding to the deviation trend. The neural network unit outputs the opening adjustment amount of the flow controller based on the first water injection flow compensation amount, the second water injection flow compensation amount, and the third water injection flow compensation amount, and integrates the positive buffer and the negative buffer with the opening adjustment amount as the target control command.

[0012] A further solution is that the process of integrating the positive buffer and negative buffer with the opening adjustment amount is as follows: After assigning buffer weight coefficients to the positive or negative buffer amount output by the compensation buffer unit, they are added to the opening adjustment amount output by the neural network unit to obtain the initial control command. The initial control command is subjected to rate limiting and amplitude limiting processing to obtain the final target control command; The buffer weight coefficient is set by the compensation buffer unit according to the deviation trend.

[0013] A further proposed solution is that the process by which the cloud server generates the collaborative control strategy is as follows: The cloud server obtains the time-series pressure fluctuations of adjacent injection wells and determines whether there is interference between injection and production or formation pressure transmission. If there is interference, a collaborative strategy including peak-shifting injection, flow redistribution, or pressure balancing objectives is generated and distributed to the edge computing modules of the relevant wells.

[0014] A second aspect of the present invention provides an intelligent control method for a flow controller based on edge computing, which, when applied to the above-mentioned intelligent control system, includes the following steps: S1: Real-time acquisition of raw data from multi-source water injection; S2: Preprocess the original data and extract multidimensional features to form a feature dataset; S3: Perform fit analysis between the feature dataset and multiple locally stored control sub-models; S4: Based on the fit analysis results, generate target control instructions through selection or weighted fusion methods; S5: Send the target control command to the drive unit to perform opening adjustment; S6: Collect adjusted operating condition data and evaluate the control effect; based on the mapping relationship between the opening adjustment amount and the control effect, assign weight coefficients to the opening adjustment amount in the next control cycle.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: The edge computing module of this invention collects and processes multi-source water injection data in real time, generating a feature dataset. By matching the real-time feature dataset with the pre-stored operating condition feature templates of each sub-model, it can automatically identify the most relevant water injection condition type and dynamically switch or merge the most suitable control strategy. Based on real-time diagnosis of the state of a single well, it can automatically maintain high-precision water injection flow control in complex water injection environments. At the same time, by avoiding inappropriate control actions, it significantly reduces overshoot and oscillation, improves the stability of the water injection process and equipment lifespan, and fundamentally overcomes the inherent defects of traditional PID controllers in oilfield water injection applications.

[0016] This invention achieves multi-parameter coordination through the design of the data processing unit and control sub-model of the edge computing module. At the feature extraction level, it not only calculates the injection flow rate deviation but also simultaneously calculates the instantaneous values ​​and rates of change of injection pressure and formation pressure, as well as their deviations from safety thresholds. The formation-injection pressure difference is used as a key feature reflecting the formation's water absorption capacity and injection resistance. The injection flow rate compensation, pressure compensation, and buffering capacity calculated by the control sub-model are fused by the neural network unit to output a coordinated opening command. This precisely controls the injection flow rate while improving the dynamic stability of the entire injection system and the formation pressure maintenance effect.

[0017] This invention utilizes the data processing unit of an edge computing module to continuously collect the opening value of the water injection flow controller and the corresponding stable water injection flow value. It then updates the current actual water injection flow-opening characteristic curve online through curve fitting and compares it with a preset standard characteristic curve. By calculating the deviation of the curve shape or the gain change rate at key points, a quantitative and real-time equipment performance degradation coefficient can be calculated. This achieves self-maintenance of control performance, automatically compensating during the slow degradation of the equipment, maintaining high control accuracy and dynamic performance over a long period, and reducing water injection fluctuations caused by gradual equipment failure.

[0018] The cloud server of this invention continuously receives multi-source water injection data uploaded from the edge computing module, identifies the mutual influence between water injection well groups, generates collaborative control commands, and realizes a fundamental leap from isolated single-well control to well group collaboration in oilfield water injection flow control. This improves water injection efficiency and formation pressure balance, effectively suppresses systemic fluctuations caused by inter-well interference, and enhances the production stability and recovery rate of the entire water injection development block. Attached Figure Description

[0019] The following figures are for illustrative purposes only and are not intended to limit the scope of the invention, wherein: Figure 1 Schematic diagram of the connection of the intelligent control system for well clusters; Figure 2 Hardware architecture diagram of the intelligent control system; Figure 3 Internal structure diagram of the edge computing module; Figure 4 : Block diagram of the internal structure of the local control model and control sub-model; In the diagram: 1. Injection well; 2. Wellhead device; 3. Flow monitoring unit; 4. Flow controller; 5. Drive component; 6. Edge computing module; 7. Cloud server; 8. Industrial Internet of Things; 9. Sensor group; 10. Local control model; 11. Data acquisition unit; 12. Data processing unit; 13. Command execution unit; 14. Optimization unit; 15. Control sub-model; 16. Adjustment and control unit; 17. Compensation and optimization unit; 18. Compensation and buffer unit; 19. Neural network unit. Detailed Implementation

[0020] To make the objectives, technical solutions, design methods, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0021] Example 1 like Figure 1 and Figure 2 As shown, this invention provides an intelligent control system for a flow controller 4 based on edge computing. This system, consisting of edge computing modules 6 installed in each injection well 1 within a certain range of a well group, and an industrial Internet of Things (IoT) 8 connecting these edge computing modules 6 to a cloud server 7, achieves well group collaboration for oilfield water injection flow control. Specifically, a flow controller 4 is installed on the wellhead device 2 of each injection well 1. A flow monitoring unit 3 measures the instantaneous flow rate of water injected into the pipeline in real time. A sensor group 9 is used to collect data such as injection pressure, wellhead pressure, formation pressure, and injection temperature. In this embodiment, the flow controller 4 is an electric regulating valve. A drive component 5 is mechanically connected to the flow controller 4 to drive its valve core to change the opening degree. The sensor group 9 includes a pressure transmitter, a temperature sensor, and a formation pressure sensor. The intelligent control system includes a cloud server 7 and edge computing modules 6 installed in each injection well 1. Each edge computing module 6 has a built-in preset local control model 10, the output of which is connected to the flow controller 4.

[0022] The edge computing module 6 is used to collect multi-source water injection data of the corresponding water injection well 1 in real time, and to process and extract features from the multi-source water injection data in real time to generate a feature dataset representing the current water injection status and equipment status. The local control model 10 includes multiple control sub-models 15 pre-trained for different water injection conditions. Each control sub-model 15 stores a corresponding condition feature template. The local control model 10 selects a control sub-model 15 based on the fit between the feature dataset and each condition feature template. The selected control sub-model 15 is used as the master control model to generate target control commands. The instructions are used to adjust the target opening degree of the flow controller 4; the cloud server 7 receives multi-source water injection data from multiple edge computing modules 6, and identifies the mutual influence between water injection well groups based on the multi-source water injection data and generates collaborative control instructions. The process of the cloud server 7 generating collaborative control strategies is as follows: the cloud server 7 obtains the time-series pressure fluctuations of adjacent water injection wells 1, determines whether there is injection-production interference or formation pressure transmission influence in adjacent water injection wells 1; if there is interference, it generates a collaborative strategy that includes peak water injection, flow redistribution or pressure balancing objectives, and sends it to the edge computing modules 6 of the relevant wells.

[0023] In this embodiment, the edge computing module 6 connects to the sensor group 9 corresponding to the injection well 1 and the drive component 5 of the flow controller 4. The sensor group 9 is used to collect data on injection pressure, wellhead pressure, real-time injection flow rate, injection temperature, and formation pressure. The drive component 5 is used to receive the target control command or collaborative control command and drive the flow controller 4 to perform opening adjustment actions. The edge computing module 6 uses an industrial gateway to connect to the sensor group 9, the flow monitoring unit 3, and the drive component 5 via cable or fieldbus to realize data acquisition and command issuance. The edge computing module 6 maintains communication with the cloud server 7 through the industrial Internet of Things 8. The cloud server 7 receives and stores data from each well and identifies the mutual influence between the injection well groups. For example, by analyzing the spatiotemporal correlation of pressure fluctuations in multiple wells, it determines whether there is injection-production interference or formation pressure transmission, and generates collaborative control commands accordingly, which are then issued to the edge computing module 6 of the relevant wells through the industrial Internet of Things.

[0024] like Figure 3As shown, the edge computing module 6 includes a data acquisition unit 11, a data processing unit 12, an instruction execution unit 13, and an optimization unit 14. The data acquisition unit 11 reads data from the sensor group 9 and the current opening degree of the flow controller 4 in real time at a first preset cycle, forming an original multi-source water injection dataset. The data processing unit 12 performs filtering, unit conversion, and time alignment processing on the original multi-source water injection dataset, and extracts and generates a feature dataset. The feature dataset includes at least: instantaneous water injection flow deviation, water injection flow deviation change rate, instantaneous water injection pressure value, and... The parameters include the rate of change, the instantaneous value and rate of change of wellhead pressure, the pressure difference between formation pressure and injection pressure, and the equipment performance degradation coefficient determined based on the historical opening degree-injection flow rate relationship; the command execution unit 13 is used to send the target control command to the drive component 5 to execute the opening degree change action and monitor the command execution status; the optimization unit 14 is used to evaluate the control effect based on the operating condition data collected in the second preset cycle after the opening degree change action is completed. If the control effect does not meet expectations, the weighting coefficient is assigned to the opening degree adjustment amount in the next control cycle based on the mapping relationship between the opening degree adjustment amount and the control effect. In this embodiment, the acquisition process for instantaneous injection flow rate deviation, injection flow rate deviation change rate, instantaneous injection pressure value and change rate, wellhead pressure instantaneous value and change rate, and formation pressure and injection pressure differential is as follows: The original injection flow rate signal is low-pass filtered and then the difference is calculated with a set value to obtain the instantaneous injection flow rate deviation; using the filtered injection flow rate deviation sequence, the difference between the deviations of two consecutive sampling periods is calculated to obtain the injection flow rate deviation change rate; the original pressure signal is filtered to obtain a smooth instantaneous pressure value. The trend can be obtained by calculating the linear regression slope or simple difference mean of the filtered pressure data within a certain time window to obtain the instantaneous value of injection pressure / wellhead pressure and its trend; after filtering the injection pressure and formation pressure respectively, instantaneous subtraction is performed to obtain a stable pressure differential signal, avoiding calculation disturbances caused by asynchronous noise between the two signals. The data processing unit 12 extracts the determination method for the equipment performance degradation coefficient, which includes: under stable water injection conditions, recording the opening value of the water injection flow controller 4 and the corresponding stable water injection flow value within a historical time window; performing curve fitting on the opening value and the stable water injection flow value to obtain the current actual water injection flow-opening characteristic curve; comparing the current actual water injection flow-opening characteristic curve with a preset standard characteristic curve or a previously fitted characteristic curve, calculating the curve shape deviation or the gain change rate at key points, and obtaining the equipment performance degradation coefficient. Under stable water injection conditions, the system continuously collects the opening value of the water injection flow controller 4 and the corresponding stable water injection flow value, updates the current actual water injection flow-opening characteristic curve online through curve fitting, and compares it with a preset standard characteristic curve. By calculating the deviation of the curve shape or the gain change rate at key points, the equipment performance degradation coefficient is calculated quantitatively and in real time.It achieves self-maintenance of control performance. The system can automatically compensate during the slow degradation of equipment, so that the control accuracy and dynamic performance are kept at a high level for a long time, reducing the fluctuation of water injection process caused by gradual equipment failure.

[0025] In the above, the local control model 10 includes multiple control sub-models 15, each of which is associated with a working condition feature template. When generating the target control command, the matching and selection of the control sub-models 15 must be performed first. The selection process is as follows: The real-time extracted feature dataset is compared with pre-stored feature templates for each operating condition to calculate similarity. In this embodiment, cosine similarity calculation is used. The operating condition feature templates include: normal steady-state water injection condition, water injection pressure fluctuation condition, formation water absorption performance gradual change condition, equipment performance degradation condition, and adjacent well injection-production interference condition. If the overall similarity of a certain operating condition feature template exceeds the first similarity threshold, then the control sub-model 15 bound to that operating condition feature template is directly selected as the main control model. If the overall similarity of all operating condition feature templates is below the first threshold, but the individual feature similarity of multiple operating condition feature templates exceeds the second similarity threshold (e.g., operating condition feature template I is highly similar to the water injection pressure fluctuation template, and operating condition feature template II is highly similar to the equipment degradation template), then the system simultaneously activates these two corresponding control sub-models 15. For the multiple activated sub-models, the fusion weight is calculated based on the similarity of each individual feature. For example, sub-model A is activated due to the pressure feature, and its weight is... Sub-model B is activated due to its decaying characteristic, and its weights... After normalizing the weights, the final control command... for: ,in and These are the control quantities output by sub-models A and B, respectively.

[0026] like Figure 4 As shown, each control sub-model 15 may contain an adjustment control unit 16, a compensation optimization unit 17, a compensation buffer unit 18, and a neural network unit 19; wherein, the adjustment control unit 16 is provided with a first difference threshold, used to determine the difference between the instantaneous water injection flow deviation and a preset deviation. ,like If the instantaneous water flow rate deviation is greater than or equal to the first difference threshold, the system is determined to be in a large deviation condition. In this case, the control unit 16 directly multiplies the instantaneous water flow rate deviation by a preset first proportional coefficient to obtain the first water flow rate compensation amount. This output amount has a linear functional relationship with the instantaneous water flow rate deviation. If the instantaneous injection flow rate deviation is less than the first difference threshold, the unit determines that the system is in a small deviation condition. At this time, the adjustment control unit 16 multiplies the instantaneous injection flow rate deviation by a preset second proportional coefficient to obtain the second injection flow rate compensation amount. This output amount also has a linear functional relationship with the instantaneous injection flow rate deviation. In this embodiment, the value of the second proportional coefficient is less than the first proportional coefficient, which is intended to provide a smoother and more precise adjustment when the deviation is small. The compensation optimization unit 17 is responsible for handling the cumulative compensation of the deviation. The compensation optimization unit 17 simultaneously receives the instantaneous injection flow rate deviation and the pressure deviation calculated from the injection pressure and formation pressure, and combines the two into a comprehensive deviation signal according to a preset weight. In each control cycle, when the valve opening of the injection flow controller 4 has not reached the physical limit, and the current accumulation direction is consistent with the dominant change direction of the comprehensive deviation, the comprehensive deviation value of the current cycle is multiplied by an accumulation coefficient and added to the historical accumulation value. The compensation optimization unit 17 dynamically adjusts an incremental compensation coefficient according to the latest accumulation value, multiplies the two, and outputs the result as the third injection flow rate compensation amount. The compensation buffer unit 18 serves as a trend predictor and buffer. It receives the rate of change of injection flow deviation and the rate of change of injection pressure, and analyzes the combined trend of these two rates to predict the future direction of the injection flow deviation. If the prediction indicates that the injection flow deviation will increase in the future, the compensation buffer unit 18 outputs a negative buffer amount, the magnitude of which is directly proportional to the strength of the predicted deviation increase trend. This aims to enhance the control effect and suppress the deviation increase. If the prediction indicates that the injection flow deviation will decrease in the future, the unit outputs a positive buffer amount, the magnitude of which is also directly proportional to the strength of the predicted deviation decrease trend. This aims to provide damping to suppress possible overshoot in the system and make the adjustment process smoother. The input to the neural network unit 19 is the first or second injection flow compensation amount output by the control unit 16 and the third injection flow compensation amount output by the compensation optimization unit 17. This neural network unit 19 is pre-trained using a large amount of historical data. The input is a combination of the three compensation amounts under various operating conditions, and the output is the valve opening adjustment amount verified as optimal in practice. In real-time control, the neural network unit 19 intelligently fuses multiple input compensation quantities through learned complex nonlinear mapping relationships, directly outputting the opening adjustment quantity, and fusing the positive and negative buffers with the opening adjustment quantity as the target control command. The process of fusing the positive and negative buffers with the opening adjustment quantity is as follows: the compensation buffer unit 18 dynamically sets the buffer weight coefficient based on the currently calculated deviation trend strength. The stronger the trend, the larger the buffer weight coefficient, and the stronger the influence of the buffering effect on the final control command; when the trend is flat, the buffer weight coefficient decreases.The opening adjustment amount output by the neural network unit 19 is multiplied by the positive buffer amount (or negative buffer amount) output by the compensation buffer unit 18 by the buffer weight coefficient, and the result is arithmetically added to obtain a preliminary control command value. To ensure the physical executability of the control command and protect the equipment, the preliminary control command must be subject to safety constraints, including rate limiting and amplitude limiting. In this embodiment, rate limiting refers to calculating the difference between the preliminary control command value and the actual executed command value of the previous control cycle. This difference is compared with the maximum allowable opening change per unit time of the water injection flow controller 4. If the difference exceeds the maximum allowable change, the difference is limited to the maximum allowable change, thereby generating an intermediate command value that meets the rate safety requirements. Amplitude limiting refers to converting the command value after rate limiting into the corresponding theoretical valve opening. It is checked whether the theoretical opening is within the safe operating range formed by the minimum and maximum openings defined by the mechanical structure of the water injection flow controller 4 itself. If the opening exceeds the upper limit, it is forcibly set to the maximum opening; if it falls below the lower limit, it is forcibly set to the minimum opening. If it is within the range, the command value is retained. After rate limiting and amplitude limiting, the change rate is first ensured to not exceed the equipment capacity, and then the final position is ensured to not exceed the equipment stroke. The resulting command value is the final determined target control command. This command is then sent to the drive unit 5, which drives the valve of the water injection flow controller 4 to perform a precise and smooth opening adjustment action. In the above, the buffer weight coefficient is set by the compensation buffer unit 18 according to the deviation trend.

[0027] Example 2 Based on Example 1, this example provides an intelligent control method for a flow controller 4 based on edge computing, including the following steps: S1: Edge computing module 6 collects raw data such as injection pressure, wellhead pressure, injection flow rate, injection temperature, formation pressure, and valve opening in real time.

[0028] S2: Perform preprocessing such as filtering and alignment on the raw data, calculate instantaneous water injection flow deviation, rate of change, water injection pressure and formation pressure difference, equipment performance attenuation coefficient, etc., to form a multidimensional feature dataset.

[0029] S3: Perform similarity matching analysis between the feature dataset and multiple locally stored control sub-models 15.

[0030] S4: Based on the adaptation results, select a single best sub-model or perform weighted fusion of the outputs of multiple sub-models to generate target control instructions.

[0031] S5: After the target control command is processed by safety constraints, it is sent to the drive unit 5 to perform opening adjustment.

[0032] S6: Collect adjusted operating condition data and evaluate the control effect. Based on the effect feedback, dynamically adjust the weight coefficients of each control action component in the next cycle to achieve online self-optimization.

[0033] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. An intelligent control system for a flow controller based on edge computing, characterized in that, include: Cloud servers, edge computing modules installed in each injection well, and industrial IoT for connecting cloud servers and edge computing modules; The edge computing module has a built-in preset local control model, and the output of the local control model is connected to the flow controller. The edge computing module is used to collect multi-source water injection data of the corresponding water injection well in real time, and to process and extract features from the multi-source water injection data in real time to generate a feature dataset that represents the current water injection status and equipment status. The local control model includes multiple pre-trained control sub-models for different water injection conditions. Each control sub-model stores a corresponding condition feature template. The local control model selects a control sub-model based on the fit between the feature dataset and each condition feature template. The selected control sub-model is used as the master control model to generate target control commands. The target control commands are used to adjust the target opening of the flow controller. The cloud server receives multi-source water injection data from multiple edge computing modules, and identifies the mutual influence between water injection well groups based on the multi-source water injection data and generates collaborative control commands. The process of selecting a control sub-model by the local control model is as follows: The similarity of each feature in the feature dataset with several pre-stored working condition feature templates is calculated. The working condition feature templates include: normal steady-state water injection working condition, water injection pressure fluctuating rapidly working condition, formation water absorption performance gradually changing working condition, equipment performance degradation working condition, and adjacent well injection and production interference working condition. If the overall similarity of a certain working condition feature template exceeds the first similarity threshold, then the dedicated control sub-model bound to that template is selected as the main control model. If the overall similarity of all working condition feature templates is lower than the first similarity threshold, but the single feature similarity of multiple templates exceeds the second similarity threshold, then multiple control sub-models associated with these high-similarity single features are activated, and the fusion weight is calculated based on the similarity of each single feature, and the outputs of multiple control sub-models are weighted and fused.

2. The intelligent control system for a flow controller based on edge computing according to claim 1, characterized in that, The edge computing module is connected to the sensor group of the corresponding water injection well and the drive component of the flow controller. The sensor group is used to collect data on water injection pressure, wellhead pressure, real-time water injection flow rate, water injection temperature, and formation pressure. The drive component is used to receive the target control command or the cooperative control command, and drive the flow controller to perform the opening adjustment action.

3. The intelligent control system for a flow controller based on edge computing according to claim 2, characterized in that, The edge computing module includes: The data acquisition unit is used to read the data of the sensor group and the current opening degree of the flow controller in real time at a first preset cycle to form the original multi-source water injection dataset; The data processing unit is used to filter, convert units and align time on the original multi-source water injection dataset, and extract and generate a feature dataset. The feature dataset includes at least: instantaneous water injection flow deviation, water injection flow deviation change rate, instantaneous value and change rate of water injection pressure, instantaneous value and change rate of wellhead pressure, formation pressure and water injection pressure difference, and equipment performance degradation coefficient determined based on the historical opening degree-water injection flow relationship. The instruction execution unit is used to send the target control instruction to the drive component to perform the opening change action and monitor the instruction execution status.

4. The intelligent control system for a flow controller based on edge computing according to claim 3, characterized in that, The edge computing module also includes an optimization unit, which is used to evaluate the control effect based on the working condition data collected in the second preset cycle after the opening change action is completed. If the control effect does not meet expectations, the unit assigns weight coefficients to the opening adjustment amount in the next control cycle based on the mapping relationship between the opening adjustment amount and the control effect.

5. The intelligent control system for a flow controller based on edge computing according to claim 4, characterized in that, The method for determining the device performance degradation coefficient extracted by the data processing unit includes: Under stable water injection conditions, record the flow controller opening value and the corresponding stable water injection flow value within the historical time window; By performing curve fitting between the opening value and the stable water injection flow rate value, the current actual water injection flow rate-opening characteristic curve is obtained; The current actual water injection flow rate-opening characteristic curve is compared with the preset standard characteristic curve or the characteristic curve obtained from the previous fitting. The deviation of the curve shape or the rate of change of gain at key points is calculated to obtain the performance degradation coefficient of the equipment.

6. The intelligent control system for a flow controller based on edge computing according to claim 5, characterized in that, The control sub-model includes: The adjustment control unit is used to determine the difference between the instantaneous water injection flow rate deviation and the preset deviation. ,like If the value is greater than or equal to the first difference threshold, output a first injection flow compensation amount that is linearly related to the instantaneous injection flow deviation; if... If the value is less than the first difference threshold, output a second water injection flow compensation amount that is linearly related to the instantaneous water injection flow deviation; The compensation optimization unit receives the instantaneous injection flow rate deviation, injection pressure, and formation pressure to generate a comprehensive deviation signal. Within the first preset period, it performs an accumulation calculation on the current comprehensive deviation to obtain the cumulative compensation amount. Based on the cumulative compensation amount, it sets an incremental compensation coefficient and uses the product of the cumulative compensation amount and the incremental compensation coefficient as the third injection flow rate compensation amount. The compensation buffer unit determines the deviation trend based on the rate of change of the water injection flow rate and the rate of change of the water injection pressure. If the deviation is going to increase, it outputs a negative buffer corresponding to the deviation trend; if the deviation is going to decrease, it outputs a positive buffer corresponding to the deviation trend. The neural network unit outputs the opening adjustment amount of the flow controller based on the first water injection flow compensation amount, the second water injection flow compensation amount, and the third water injection flow compensation amount, and integrates the positive buffer and the negative buffer with the opening adjustment amount as the target control command.

7. The intelligent control system for a flow controller based on edge computing according to claim 6, characterized in that, The process of integrating the positive buffer and negative buffer with the opening adjustment amount is as follows: After assigning buffer weight coefficients to the positive or negative buffer amount output by the compensation buffer unit, they are added to the opening adjustment amount output by the neural network unit to obtain the initial control command. The initial control command is subjected to rate limiting and amplitude limiting processing to obtain the final target control command; The buffer weight coefficient is set by the compensation buffer unit according to the deviation trend.

8. The intelligent control system for a flow controller based on edge computing according to claim 7, characterized in that, The process by which the cloud server generates the collaborative control strategy is as follows: The cloud server obtains the time-series pressure fluctuations of adjacent injection wells to determine whether there is interference between injection and production or formation pressure transmission. If there is interference, a collaborative strategy including peak-shifting injection, flow redistribution, or pressure balancing objectives is generated and distributed to the edge computing modules of the relevant wells.

9. A smart control method for a flow controller based on edge computing, characterized in that, The intelligent control system according to any one of claims 1-8 includes the following steps: S1: Real-time acquisition of raw data from multi-source water injection; S2: Preprocess the original data and extract multidimensional features to form a feature dataset; S3: Perform fit analysis between the feature dataset and multiple locally stored control sub-models; S4: Based on the fit analysis results, generate target control instructions through selection or weighted fusion methods; S5: Send the target control command to the drive unit to perform opening adjustment; S6: Collect adjusted operating condition data and evaluate the control effect; based on the mapping relationship between the opening adjustment amount and the control effect, assign weight coefficients to the opening adjustment amount in the next control cycle.

Images