A deep well pump operation frequency conversion control method and system

Abstract

The present invention relates to the technical field of deep well pump control in oil and gas fields, and discloses a deep well pump operation frequency conversion control method and system, wherein a deep well pump operation frequency conversion control method comprises: constructing a comprehensive energy efficiency evaluation model, and introducing a grid state factor and an electricity price factor; analyzing the harmonic characteristics of the frequency converter, obtaining the total harmonic distortion rate and each harmonic component, and determining the harmonic management strategy according to the grid sensitivity and the harmonic pollution degree; constructing a multi-objective optimization algorithm, the multi-objective optimization algorithm comprehensively considers three objectives, and uses reinforcement learning to generate an optimal frequency conversion control strategy; realizing intelligent switching of operation modes; generating a predictive operation plan based on the operation mode switching result and the grid state prediction, formulating a daily operation plan in combination with the electricity price prediction and production demand, and dynamically adjusting the plan when there is a deviation between the actual operating conditions and the prediction; the present invention reduces energy consumption under the same flow conditions by constructing a comprehensive energy efficiency evaluation model and applying the reinforcement learning optimization algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of deep well pump control in oil and gas fields, and more particularly to a deep well pump operation frequency conversion control method and system. Background Art

[0002] Deep-well pump variable frequency control technology is a key technology in modern oil and gas field production. By adjusting the motor power supply frequency, the operating state of the deep-well pump is controlled to meet production requirements under different operating conditions. With the development of oil and gas field production technology and the increasing demand for energy conservation and emission reduction, deep-well pump variable frequency control technology faces new challenges and development needs.

[0003] At present, deep well pump frequency conversion control technology mainly has the following technical problems:

[0004] The existing deep well pump variable frequency control method mainly takes maintaining flow stability as a single control target, and lacks comprehensive optimization of system energy efficiency. Traditional control systems usually use PID control algorithms to adjust the inverter output frequency according to the set flow target, but do not fully consider the coupling relationship between multiple factors such as motor efficiency, hydraulic efficiency and inverter efficiency. This single-target control method causes the system to consume too much energy and have high operating costs while meeting production needs; existing energy efficiency optimization methods are mostly based on static models and cannot adapt to dynamic changes in grid conditions. Factors such as periodic fluctuations in grid electricity prices, real-time changes in grid loads, and fluctuations in power quality directly affect the actual operating costs and efficiency of deep well pump systems, but traditional control methods lack the ability to perceive and respond to these dynamic factors, resulting in discrepancies between theoretical optimization effects and actual There is a large gap in the operating results; as a nonlinear load, the inverter will inject a large amount of harmonic current into the power grid during operation, causing harmonic pollution to the power grid. These harmonics will not only deteriorate the power quality of the power grid, but also lead to increased equipment heat generation, decreased operating efficiency, shortened service life and other problems. The existing deep well pump control systems generally lack the ability to effectively monitor and actively manage harmonics, and are unable to dynamically adjust the operating strategy according to the grid sensitivity and harmonic limit requirements, further exacerbating energy waste and equipment loss; traditional control algorithms mostly use fixed control strategies based on empirical parameters, which have poor adaptability to complex and changeable well conditions. The operating environment of deep well pumps is complex, and well parameters (such as water content, gas-liquid ratio, well depth, etc.) change frequently. Fixed parameter control algorithms are difficult to effectively respond to various unexpected working conditions and cannot achieve the global optimization of system performance.

[0005] Therefore, there is an urgent need for a deep well pump operation frequency conversion control method that can comprehensively consider energy efficiency optimization, grid adaptability and harmonic management to solve the above-mentioned problems existing in the existing technology. Summary of the Invention

[0006] The present invention provides a deep well pump operation frequency conversion control method and system, which solves the technical problems of single target control, static model optimization, insufficient harmonic pollution management and poor adaptability of fixed parameter control algorithm in the related art frequency conversion control method.

[0007] The present invention provides a deep well pump operation frequency conversion control method, comprising:

[0008] Construct a comprehensive energy efficiency evaluation model, which includes motor efficiency sub-model, hydraulic efficiency sub-model and inverter efficiency sub-model, and introduces grid status factor and electricity price factor;

[0009] Analyze the inverter harmonic characteristics based on the output results of the comprehensive energy efficiency evaluation model, obtain the total harmonic distortion rate and each harmonic component, and determine the harmonic management strategy based on the grid sensitivity and harmonic pollution level;

[0010] A multi-objective optimization algorithm is constructed based on a comprehensive energy efficiency evaluation model and harmonic management strategy. The multi-objective optimization algorithm comprehensively considers three goals: energy consumption, production efficiency, and grid friendliness, and uses reinforcement learning to generate the optimal variable frequency control strategy.

[0011] Intelligent switching of operating modes is achieved based on the optimal frequency conversion control strategy, switching between energy efficiency priority mode, low harmonic mode and peak-valley balance mode according to the grid status and electricity price forecast results;

[0012] Generate a predictive operation plan based on the operation mode switching results and grid status forecast, formulate a daily operation plan based on electricity price forecast and production demand, and dynamically adjust the plan when the actual operating conditions deviate from the forecast.

[0013] Furthermore, the steps of constructing the comprehensive energy efficiency evaluation model include:

[0014] Collect deep well pump operating parameter data and power grid status data;

[0015] Construct a motor efficiency sub-model and calculate the motor efficiency curve under different load rates and frequencies;

[0016] Construct a hydraulic efficiency sub-model to calculate the hydraulic efficiency of the pump under different flow and head conditions;

[0017] Construct an inverter efficiency sub-model to calculate the inverter efficiency under different output frequencies and loads;

[0018] The motor efficiency sub-model, hydraulic efficiency sub-model and inverter efficiency sub-model are integrated to calculate the total efficiency of the deep well pump variable frequency control system. The grid status factor and electricity price factor are introduced to construct a comprehensive energy efficiency index.

[0019] Furthermore, the harmonic management strategy includes:

[0020] Adjust the pulse width modulation parameters of the inverter;

[0021] Configure harmonic filtering device;

[0022] Optimize the inverter operating frequency to avoid harmonic resonance points;

[0023] Switch to low harmonic mode when the grid harmonic content is high and the grid sensitivity is high.

[0024] Furthermore, the steps for constructing the multi-objective optimization algorithm include:

[0025] Define the state space and action space of the deep well pump variable frequency control system;

[0026] Define a multi-objective reward function that comprehensively considers energy consumption indicators, production efficiency indicators, and grid-friendliness indicators;

[0027] Build a deep Q network model;

[0028] Train a deep Q-network model to maximize cumulative rewards;

[0029] Generate optimal variable frequency control strategy.

[0030] Furthermore, the constructed deep Q network model includes:

[0031] Input layer, receiving state parameters of deep well pump variable frequency control system;

[0032] At least one hidden layer to extract features and transform the input data;

[0033] Output layer, outputs the Q value of each action;

[0034] Experience replay buffer, which stores state transition samples;

[0035] Both the target network and the evaluation network are deep neural networks;

[0036] The constraint processing module ensures that the generated control strategy meets the constraint of minimum flow requirement.

[0037] Furthermore, the steps for implementing the intelligent switching of the operating mode include:

[0038] Define a set of operating modes for deep well pumps, including energy efficiency priority mode, low harmonic mode, and peak-valley balance mode;

[0039] Collect and process grid status data;

[0040] Build a mode switching decision algorithm to determine the most suitable operation mode based on the current grid status and electricity price information;

[0041] Implement a smooth mode switching mechanism, set the minimum time interval for mode switching, and use gradual parameter adjustment.

[0042] Furthermore, the step of generating the predictive operation plan includes:

[0043] Collect and process forecast data, including future electricity price period forecasts, production target demand forecasts, equipment maintenance plans, and historical operating data statistics;

[0044] Build a day-ahead planning model, divide the 24-hour period into sections, and assign different operation plans to each time period;

[0045] Execute the operation plan optimization algorithm to generate the optimal daily operation plan;

[0046] Develop forward-looking control strategies, preset target frequency ranges for each time period, and pre-configure mode switching time points;

[0047] Implement a dynamic plan adjustment mechanism to adjust the operation plan when the actual operating conditions deviate from the forecast.

[0048] Furthermore, the grid state factor is calculated based on the grid quality and load conditions, taking into account the ratio of the current total harmonic distortion rate to the maximum allowed harmonic distortion rate, and the ratio of the current grid load rate to the maximum allowed load rate.

[0049] Furthermore, the electricity price factor is calculated based on the ratio of the electricity price in the current period to the benchmark electricity price. When the electricity price is at its peak, the deep well pump variable frequency control system automatically reduces the load and adjusts the operating period.

[0050] The present invention provides a deep well pump operation frequency conversion control system for executing the above-mentioned deep well pump operation frequency conversion control method, comprising:

[0051] Energy efficiency evaluation module, which is used to build a comprehensive energy efficiency evaluation model, integrating motor, hydraulic and inverter efficiency sub-models, and taking into account grid status and electricity price factors;

[0052] Harmonic management module, used to analyze the harmonic characteristics of the inverter, evaluate the sensitivity of the power grid, and implement harmonic suppression measures;

[0053] The optimization control module is used to build a multi-objective optimization algorithm based on reinforcement learning to balance energy consumption, production efficiency, and grid friendliness and generate the optimal control strategy;

[0054] Mode switching module, used to intelligently switch between three operating modes: energy efficiency priority, low harmonics, and peak-valley balance;

[0055] The plan generation module is used to formulate predictive daily operation plans and dynamically adjust them according to actual operating conditions to achieve forward-looking energy management.

[0056] The beneficial effects of the present invention are: by constructing a comprehensive energy efficiency evaluation model and applying a reinforcement learning optimization algorithm, energy consumption is reduced under the same flow conditions, the high-efficiency operating range is expanded, and the annual operating cost is reduced. The present invention can automatically adjust the optimal operating frequency according to actual operating conditions, avoiding the energy waste caused by traditional fixed parameter control;

[0057] By analyzing and managing the harmonic characteristics of the inverter, harmonic pollution is reduced, effectively avoiding equipment efficiency degradation and grid quality issues. The present invention can automatically adjust the operating mode according to grid sensitivity, proactively reducing harmonic emissions during sensitive periods, and improving grid compatibility.

[0058] Through electricity price forecasting and predictive operation plan generation, the present invention can intelligently plan operation time periods and loads, reduce energy consumption during peak electricity price periods, and lower some electricity bills. It can also dynamically adjust production plans according to electricity price periods to balance production demand and cost control.

[0059] By comprehensively considering flow demand, energy efficiency, and grid status, the present invention reduces equipment losses caused by harmonics and extends equipment life. The smooth mode switching mechanism avoids system instability caused by frequent adjustments and improves overall operational reliability.

[0060] Abandoning the traditional fixed algorithm, the invention adopts reinforcement learning to generate control strategies, which enables the system to adaptively respond to various changes in working conditions. Especially under complex well conditions, the invention can autonomously adjust the control strategy according to real-time data to achieve the optimal operating state.

[0061] By generating predictive operation plans, the present invention can respond to electricity price fluctuations and changes in production demand in advance, reducing efficiency losses caused by passive responses. The rolling time domain optimization mechanism ensures the dynamic adjustment capability of the plan and improves the system's ability to cope with uncertainty. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 This is a flow chart of a frequency conversion control method for deep well pump operation in the present invention;

[0063] Figure 2 It is a line graph showing the changing trend of the unit energy consumption of the deep well pump at different time points before and after the application of the method of the present invention;

[0064] Figure 3 It is a bar graph showing the improvement of different harmonic types before and after applying the method of the present invention;

[0065] Figure 4 It is a radar chart comparing the performance of the traditional control method and the three operating modes proposed by the method of the present invention in terms of five key performance indicators;

[0066] Figure 5 It is a scatter plot of the relationship between carrier frequency and total harmonic distortion;

[0067] Figure 6 It is an area graph showing the changes in power demand, electricity price and energy cost before and after the application of the method of the present invention within 24 hours. DETAILED DESCRIPTION

[0068] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. Furthermore, features described in some examples may be combined in other examples.

[0069] At least one embodiment of the present invention discloses a method for frequency conversion control of deep well pump operation, such as Figure 1 Shown, including:

[0070] Step 1: Construct a comprehensive energy efficiency evaluation model. The comprehensive energy efficiency evaluation model includes a motor efficiency sub-model, a hydraulic efficiency sub-model, and a frequency converter efficiency sub-model, and introduces a grid status factor and an electricity price factor.

[0071] This step analyzes the operating data and efficiency characteristics of the deep well pump system to build an energy efficiency evaluation model that comprehensively considers multiple factors, providing a basis for subsequent optimization control. Specifically, it includes:

[0072] Step 1.1, collecting deep well pump operating parameter data;

[0073] Deep well pump operating parameter data includes real-time operating data such as motor current, voltage, power factor, speed, torque, downhole liquid level, outlet flow, inlet pressure, and outlet pressure, as well as grid status data such as grid voltage, frequency, harmonic content, and electricity price period;

[0074] A distributed SCADA system is used to collect data. An edge computing unit is installed at the wellhead, connecting sensors via an industrial bus (such as Modbus / PROFIBUS). Motor parameters are collected by a multifunctional power analyzer installed in the power distribution cabinet, sampling at a frequency of 10 times per second. Pressure and flow parameters are collected by corresponding sensors installed in the pipeline, and the data is transmitted via 4-20mA analog signals. A parameter validity detection mechanism is established, using a three-level filtering process (hard threshold filtering, median filtering, and rate of change detection) to handle abnormal data.

[0075] Step 1.2, construct the motor efficiency sub-model;

[0076] The motor efficiency submodel calculates the motor efficiency curve at different load factors and frequencies. Motor efficiency is calculated as the ratio of the motor's output mechanical power to its input electrical power. Mechanical power is the product of torque and angular velocity, while electrical power is determined by voltage, current, and power factor.

[0077] Step 1.3, construct the hydraulic efficiency sub-model;

[0078] The hydraulic efficiency submodel calculates the hydraulic efficiency of the pump under different flow and head conditions. Hydraulic efficiency is expressed as the ratio of the potential energy gained by the liquid to the mechanical power input to the pump shaft, taking into account factors such as liquid density, gravitational acceleration, flow rate, and head.

[0079] Step 1.4, construct the inverter efficiency sub-model;

[0080] The inverter efficiency sub-model calculates the inverter efficiency at different output frequencies and loads. The inverter efficiency is defined as the ratio of its output power to its input power.

[0081] Step 1.5, construct a comprehensive energy efficiency evaluation model;

[0082] By integrating the motor efficiency sub-model, hydraulic efficiency sub-model and inverter efficiency sub-model, the total system efficiency is calculated as the product of inverter efficiency, motor efficiency and hydraulic efficiency. At the same time, the grid state factor ( ) and electricity price factors to construct a comprehensive energy efficiency index E. This index comprehensively considers the overall efficiency of the system and its comprehensive performance under the current grid status and electricity price conditions;

[0083] The comprehensive energy efficiency evaluation model here is specifically implemented as follows:

[0084] Data Collection and Processing: Sensors installed at key points in the deep-well pump system collect real-time operating parameters of the motor, pump, and inverter. Data is typically collected over a 1-second period and smoothed using a sliding window average (60-second window size) to eliminate the effects of short-term fluctuations.

[0085] Grid status factor calculation: Grid status factor ( ) is calculated based on the grid quality and load conditions, taking into account the current total harmonic distortion (THD) and the maximum allowable harmonic distortion ( ) ratio, and the current grid load rate ( ) and the maximum allowable load rate ( ) is obtained by combining appropriate weight coefficients.

[0086] Electricity Price Factor Calculation: The electricity price factor is calculated based on the ratio of the current electricity price to the benchmark price (usually the off-peak price). The electricity price factor takes a lower value during peak electricity prices, prompting the deep well pump system to reduce energy consumption or adjust its operating hours.

[0087] Model Adaptation: Regularly adjust model parameters by comparing the deviation between predicted efficiency and actual measured efficiency. Calibration is typically performed monthly or immediately after equipment maintenance.

[0088] In specific application scenarios, the application examples of this comprehensive energy efficiency evaluation model are as follows:

[0089] Oilfield Multi-Well Scenario: In an oilfield with multiple wells, a library of different efficiency curves is established for deep-well pumps with different well conditions (such as water cut and gas-liquid ratio). The system automatically selects the most suitable efficiency curve based on real-time monitored well parameters, improving evaluation accuracy.

[0090] Seasonal Adjustment: The system introduces seasonal correction factors to account for the impact of seasonal ambient temperatures on equipment efficiency. For example, in high summer temperatures, the motor and inverter efficiency curves are adjusted downward to reflect the efficiency loss caused by the temperature increase.

[0091] Aging Compensation: To address the aging effects of long-term equipment, the system automatically adjusts the baseline of the efficiency evaluation model based on equipment operating time and maintenance records. For example, for motors that have been in operation for more than three years, the efficiency curve is adjusted downward by 5% to 10% to reflect the effects of aging.

[0092] Multi-operating condition coverage: By storing efficiency data points at different load rates and frequencies, a multidimensional efficiency lookup table is constructed. During real-time evaluation, a multivariate interpolation algorithm is used to calculate the precise efficiency value under the current operating condition, improving the accuracy of evaluations under borderline and unconventional operating conditions.

[0093] Step 2: Analyze the harmonic characteristics of the inverter based on the output of the comprehensive energy efficiency evaluation model, obtain the total harmonic distortion rate and each harmonic component, and determine the harmonic management strategy based on the grid sensitivity and harmonic pollution level;

[0094] This step analyzes and manages the harmonics generated by the inverter to reduce the negative impact of harmonics on grid quality and equipment efficiency. Specifically, it includes:

[0095] Step 2.1, establish the inverter harmonic model;

[0096] The inverter harmonic model analyzes the harmonic characteristics of the inverter at different frequencies. The inverter output voltage contains the fundamental component and each harmonic component. The amplitude and phase characteristics of each harmonic can be obtained through Fourier analysis.

[0097] Optionally, in some implementations, the harmonic model can be further refined into a carrier frequency-dependent model. This model maps harmonic characteristics at different carrier frequencies, enabling the system to automatically select the optimal carrier frequency based on sensitive grid frequency bands. For example, a function can be established that correlates carrier frequency with the amplitude of specific subharmonics to guide the system to avoid critical harmonic frequencies.

[0098] Step 2.2, analyze the impact of harmonics on the power grid;

[0099] Calculate the total harmonic distortion rate and the current distortion rate of each harmonic to assess the degree of harmonic pollution. The total harmonic distortion rate represents the ratio of the square root of the sum of the squares of the effective values ​​of all harmonic currents to the effective value of the fundamental current.

[0100] For example, in practical applications, we can focus on the 5th, 7th, 11th and 13th harmonics, which are usually the main harmonic components generated by the inverter and have a greater impact on the power grid and equipment.

[0101] Step 2.3, based on the grid sensitivity and harmonic pollution level;

[0102] Determine harmonic management strategies. Grid sensitivity can be calculated using parameters such as grid impedance and short-circuit capacity;

[0103] Grid sensitivity is determined by measuring grid impedance at specific frequencies. Using a small signal injection method, small signals of varying frequencies are injected into the grid during non-operating hours. The corresponding voltage and current responses are measured to calculate the impedance-frequency characteristic curve. The sensitivity index is defined as the ratio of the impedance at a specific frequency to the fundamental frequency impedance. A sensitive frequency band database is established, recording the frequencies at which impedance peaks occur, for use in developing harmonic management strategies.

[0104] In some implementations, harmonic management strategies can be dynamically adjusted based on grid impedance characteristics. For example, if a grid resonance risk is detected (grid impedance peaks at a certain frequency), the system automatically adjusts the inverter parameters to avoid that resonant frequency and prevent harmonic amplification.

[0105] Step 2.4, implement harmonic suppression measures;

[0106] Harmonic suppression measures include: adjusting the inverter's pulse width modulation (PWM) parameters, such as carrier frequency and modulation ratio; configuring necessary harmonic filtering devices; optimizing the inverter's operating frequency to avoid harmonic resonance points;

[0107] Optionally, harmonic suppression measures can include the use of multi-pulse inverter technology. For example, in high-power applications, a 12-pulse or 18-pulse inverter structure can be used to offset low-order harmonics through phase shifting, reducing the 5th and 7th harmonic content. Furthermore, if the load permits, random PWM technology can be used to spread the harmonic energy over a wider frequency band, reducing the harmonic peak value at a specific frequency point.

[0108] Step 3: Build a multi-objective optimization algorithm based on the comprehensive energy efficiency evaluation model and harmonic management strategy. The multi-objective optimization algorithm comprehensively considers three goals: energy consumption, production efficiency, and grid friendliness, and uses reinforcement learning to generate the optimal variable frequency control strategy.

[0109] This step applies reinforcement learning technology to build a multi-objective optimization algorithm, balancing the three goals of energy consumption, production efficiency, and grid friendliness to generate the optimal frequency conversion control strategy. Specifically, it includes:

[0110] Step 3.1, define the system state space and action space;

[0111] The system state space includes deep well pump operating parameters (flow, pressure, power, etc.) and environmental parameters (grid status, electricity price, etc.);

[0112] The action space includes control variables such as the inverter output frequency and operating time selection;

[0113] Step 3.2, define the multi-objective reward function;

[0114] The multi-objective reward function takes into account the following three aspects:

[0115] Energy consumption index: reflects energy efficiency and is positively correlated with the comprehensive energy efficiency index in step 1;

[0116] Production efficiency indicators: reflect the degree of achievement of production targets, such as flow stability, pressure satisfaction, etc.

[0117] Grid friendliness index: reflects the degree of impact on the power grid and is negatively correlated with the degree of harmonic pollution in step 2.

[0118] The comprehensive reward function is the weighted sum of the three indicators. The weight coefficient reflects the relative importance of each goal and satisfies the constraint that the sum of the weights is 1.

[0119] Step 3.3, build a reinforcement learning model based on deep Q network;

[0120] The reinforcement learning model based on the Deep Q-Network (DQN) consists of the following components:

[0121] Experience replay buffer: stores state transition samples;

[0122] Target network and evaluation network: Both are deep neural networks, with state as input and Q value of each action as output;

[0123] Constraint processing module: ensures that the generated control strategy meets constraints such as minimum flow requirements.

[0124] The specific implementation of the deep Q network model here is as follows:

[0125] Network Structure: The evaluation network and the target network share the same structure, consisting of an input layer, three hidden layers, and an output layer. The input layer has the same number of nodes as the state dimension and receives the deep-well pump's operating and environmental parameters. The hidden layers contain 128, 64, and 32 neurons, respectively, using the ReLU activation function. The output layer has the same number of nodes as the action space dimension and represents the Q-values ​​of different actions.

[0126] Optionally, in some embodiments, the network structure can adopt a dual-stream network architecture, that is, the state is divided into an operating parameter stream and an environmental parameter stream, which are processed by independent hidden layers and then merged. This structure is more suitable for processing the correlation between different types of features.

[0127] Network parameter updates: The evaluation network updates its parameters in real time using the backpropagation algorithm, while the target network parameters are periodically copied from the evaluation network using soft updates. The soft update coefficient is typically set to a small value to ensure smooth changes in the target network parameters.

[0128] In some implementations, a periodic hard update approach may be used, where the evaluation network parameters are completely copied to the target network every fixed number of steps (e.g., 1000 steps). This approach may be more stable in some scenarios.

[0129] Experience replay mechanism: Maintains a fixed-size experience pool to store state transition samples. During training, randomly sample small batches of samples for learning, reducing correlation between samples and improving training stability.

[0130] Optionally, a priority experience replay mechanism can be used to allocate sampling probabilities based on the temporal difference error of samples, so that the model pays more attention to those unexpected or important conversion samples and accelerates the learning process.

[0131] Constraint handling: For constraints such as the minimum flow rate requirement that deep well pumps must meet, a penalty term is used. When the generated control strategy violates the constraint, a negative penalty term is added to the reward function to guide the model to learn a strategy that satisfies the constraint.

[0132] For example, for a minimum flow constraint, a penalty term can be added to the reward function, where the penalty coefficient is multiplied by the larger of the maximum value of the difference between the minimum flow requirement and the current flow and zero.

[0133] In some implementations, constraints can also be handled through action space pruning, that is, actions that would violate the constraints are directly excluded during the strategy generation phase to ensure that all optional actions are within a safe range.

[0134] Step 3.4, training the reinforcement learning model;

[0135] The goal of the reinforcement learning model is to maximize the cumulative reward, that is, to maximize the expected value of the sum of rewards with a discount factor that reflects the importance of future rewards;

[0136] Training is divided into two phases: offline pre-training and online fine-tuning. Offline pre-training uses historical run data to build a simulation environment, and experience replay is used to train the DQN network. An ε-greedy strategy is used for exploration-exploitation balance, with the ε value initially set to 0.8 and gradually reduced to 0.05 during training. The training batch size is set to 64, and the Adam optimizer is used with a learning rate of 0.001, updating the target network every 100 steps. During the online fine-tuning phase, a safety range constraint is used to limit the action space to within ±3 Hz of the current frequency.

[0137] Step 3.5, generate the optimal variable frequency control strategy;

[0138] Through the trained reinforcement learning model, for a given system state, the action that maximizes the Q value is selected as the control strategy;

[0139] In specific application scenarios, the application examples of this deep Q network model are as follows:

[0140] State representation: For a deep-well pump system in an oilfield, the state vector contains the following parameters: motor power (kW), flow rate (m³ / h), downhole liquid level (m), outlet pressure (MPa), current frequency (Hz), grid harmonic content (%), current electricity price (yuan / kWh), and other parameters, forming a continuous state space.

[0141] Action definition: The action space is a discrete frequency adjustment value, including seven optional values ​​[-5Hz, -2Hz, -1Hz, 0Hz, +1Hz, +2Hz, +5Hz], which represents the adjustment amount relative to the current frequency.

[0142] Training Process: Before actual deployment, pre-training is performed based on historical operational data and a simulation environment. Then, the model is continuously optimized using online learning in the actual system. To ensure safety, a low exploration rate is initially used, and safety boundaries are set to limit the range of actions.

[0143] Adaptive adjustment: For different oil well conditions, such as high water-cut wells and high gas-liquid ratio wells, the model can automatically adjust the control strategy by observing the reward feedback under different states to adapt to the optimal operating parameters of the specific well conditions.

[0144] Step 4: Implement intelligent switching of operating modes based on the optimal variable frequency control strategy, switching between energy efficiency priority mode, low harmonic mode, and peak-valley balance mode according to the grid status and electricity price forecast results;

[0145] This step realizes intelligent switching of deep well pump operation modes based on the grid status and electricity price forecast results to adapt to different working conditions. Specifically, it includes:

[0146] Step 4.1, define the operating mode set of the deep well pump;

[0147] The set of operating modes for deep well pumps (each mode corresponds to a set of control parameter configurations) includes:

[0148] Energy efficiency priority mode ( ): Parameter configuration that gives priority to system energy efficiency;

[0149] Low harmonic mode ( ): Give priority to parameter configuration that reduces harmonic pollution;

[0150] Peak-valley balance mode ( ): Optimize the parameter configuration of the operation plan according to the electricity price period.

[0151] Step 4.2, collecting and processing grid status data;

[0152] Grid status data includes:

[0153] Grid voltage and frequency stability indicators;

[0154] Harmonic content and grid impedance characteristics;

[0155] Electricity price period division and forecast information.

[0156] Grid voltage and frequency stability data is collected using power quality analyzers installed at substations and wellsite distribution centers. Harmonic content is calculated using Fourier transforms, employing a sliding window analysis method with a window length of 10 grid cycles. Grid impedance characteristics are regularly measured using specialized test equipment or estimated through analysis of voltage and current waveforms at the grid connection point. Electricity price period information is obtained through real-time integration with the power company's information system or configured through pre-set period tables. Data processing utilizes a layered architecture: the edge layer performs preliminary calculations, while the core layer performs comprehensive analysis. The results are stored in JSON format.

[0157] Step 4.3, constructing a mode switching decision algorithm;

[0158] The mode switching decision algorithm determines the most suitable operation mode based on the current grid status and electricity price information, combined with the control strategy generated in step 3;

[0159] It should be noted that the mode switching decision function comprehensively considers the following factors:

[0160] When the grid harmonic content is high or the grid sensitivity is high, the low harmonic mode tends to be selected;

[0161] When electricity prices are at their peak, the company tends to choose peak-valley balancing mode, reducing load or adjusting operating hours;

[0162] When the grid is in good condition and electricity prices are not at peak times, the energy efficiency priority mode tends to be selected.

[0163] Step 4.4, implement the mode smooth switching mechanism;

[0164] In order to avoid system instability caused by frequent switching, a mode switching smoothing function is introduced:

[0165] Set the minimum time interval for mode switching;

[0166] Use gradual parameter adjustments instead of sudden switches;

[0167] Establish the priority and condition constraints for mode switching.

[0168] Step 5: Generate a predictive operation plan based on the operation mode switching results and grid status forecast. Combined with the electricity price forecast and production demand, formulate a daily operation plan and dynamically adjust the plan when the actual operating conditions deviate from the forecast.

[0169] This step formulates the most economical daily operation plan based on electricity price forecasts and production needs, and plans the frequency conversion control parameters of the deep well pump in advance. Specifically, it includes:

[0170] Step 5.1, collect and process prediction data;

[0171] The forecast data includes: electricity price period forecast for the next 24 hours; production target demand forecast; equipment maintenance plan; and historical operation data statistics.

[0172] Electricity price forecasts are based on integration with the power company's information system to obtain price period breakdowns for the next 24 hours. Production demand forecasts utilize time series analysis based on historical data, combined with production planning indicators. Equipment maintenance plans synchronize scheduled maintenance times and types with the enterprise asset management system. Historical data statistics utilize a three-tiered storage architecture: a real-time database stores the last seven days of data, a relational database stores three months of data, and a data warehouse stores long-term historical data. Data preprocessing includes outlier detection, missing value imputation, and time scale normalization.

[0173] Step 5.2: Build a day-ahead planning model;

[0174] The day-ahead planning model divides the 24-hour period into multiple time periods and assigns a different operation plan to each time period, with the goal of minimizing total operating costs. The calculation process considers the sum of the product of the electricity price and the corresponding power consumption in each time period, and is subject to multiple constraints.

[0175] The objective function is designed to be the sum of the product of the electricity price and the corresponding power for each time period. A dynamic time segmentation method is used to adaptively divide time periods based on price differences, with finer divisions for larger differences. Constraints are handled using the Lagrange multiplier method, converting constraints into penalty terms and adding them to the objective function. Daily total output constraints are expressed using integral equations to ensure that the production plan meets daily targets. Safety constraints are implemented by setting upper and lower limits for operating parameters, including maximum allowable current and maximum allowable temperature.

[0176] Among them, the constraints include: the total daily output reaches the production target; the equipment operating parameters are within the safe range; and the number of equipment start-up and shutdown times is considered to be limited.

[0177] Step 5.3, execute the operation plan optimization algorithm;

[0178] Generate the optimal daily operation plan. The operation plan optimization algorithm uses mixed integer programming to solve optimization problems involving discrete variables (such as equipment start and stop status) and continuous variables (such as operating frequency).

[0179] Step 5.4, develop a proactive control strategy;

[0180] Based on the daily operation plan, control parameters for each time period are configured in advance to achieve forward-looking energy management:

[0181] Preset the target frequency range for each time period;

[0182] Pre-configured mode switching time point;

[0183] Prepare load adjustment plan.

[0184] Step 5.5, implement the dynamic adjustment mechanism of the plan;

[0185] When actual operating conditions deviate from the forecast, the operation plan is dynamically adjusted to ensure system adaptability:

[0186] Establish a mechanism to detect deviations from plan execution;

[0187] Set the threshold for triggering replanning;

[0188] Implement rolling time domain optimization update.

[0189] Deviation detection utilizes a mechanism that compares real-time monitoring with predicted values, employing multi-level threshold trigger conditions. When deviations in liquid level, flow rate, or grid conditions exceed preset thresholds, a replanning process is triggered. Replanning utilizes a rolling time domain approach, preserving the plan for the next two hours and adjusting only the subsequent time periods. A three-tiered response mechanism is established: minor deviations result in parameter fine-tuning, medium deviations adjust the current mode parameters, and major deviations result in replanning. Dynamic adjustment results are transmitted to the control system via an intermediate buffer to ensure a smooth transition.

[0190] A deep well pump operation frequency conversion control system, used to execute the above-mentioned deep well pump operation frequency conversion control method, comprising:

[0191] Energy efficiency evaluation module, which is used to build a comprehensive energy efficiency evaluation model, integrating motor, hydraulic and inverter efficiency sub-models, and taking into account grid status and electricity price factors;

[0192] Harmonic management module, used to analyze the harmonic characteristics of the inverter, evaluate the sensitivity of the power grid, and implement harmonic suppression measures;

[0193] The optimization control module is used to build a multi-objective optimization algorithm based on reinforcement learning to balance energy consumption, production efficiency, and grid friendliness and generate the optimal control strategy;

[0194] Mode switching module, used to intelligently switch between three operating modes: energy efficiency priority, low harmonics, and peak-valley balance;

[0195] The plan generation module is used to formulate predictive daily operation plans and dynamically adjust them according to actual operating conditions to achieve forward-looking energy management.

[0196] Here, the present invention provides an implementation example:

[0197] The frequency conversion control method for deep-well pump operation in this embodiment was put to practical use in a cluster of high-water-cut oil wells at a certain oilfield. The oilfield has 12 deep-well pumping wells with an average water cut of 92%. The deep-well pump motors have power ranging from 45 to 75 kW, and the inverters used are all general-purpose vector inverters. The wells are located in two production blocks, and the regional power grid is unstable. The electricity price is based on a time-of-use (TOU) pricing policy, with a peak-to-valley price difference of 0.4 yuan / kWh.

[0198] Before the application, the oil field faced the following challenges:

[0199] Energy consumption remains high, with annual electricity costs for the 12 wells reaching approximately 3.2 million yuan, accounting for 42% of operating costs;

[0200] The grid harmonics problem is serious, with the inverter's total harmonic distortion rate reaching an average of 15%, causing frequent equipment overheating;

[0201] The equipment failure rate is high due to grid quality issues, with an average of 2 to 3 motors requiring replacement every quarter;

[0202] Optimization control under different well conditions is difficult to manage uniformly, and manual experience adjustment is inefficient.

[0203] This method is deployed on an industrial computer in a regional control center and connected to the frequency conversion control cabinets at each wellhead via a fieldbus network to collect data and issue control commands. Simultaneously, the system is connected to power grid monitoring equipment and an electricity price information system to obtain real-time grid parameters and electricity price data.

[0204] The specific implementation process of this embodiment in actual application is as follows:

[0205] For a typical high-water-cut oil well (numbered YJ-05, motor power 55kW), we first collected data for 30 days, including operating parameters and grid status parameters. Based on the collected data, we constructed an efficiency model specific to this well:

[0206] Motor efficiency curve: In the 40 to 50 Hz frequency range, the optimal efficiency point is 91.2% at a load factor of 70%. As the load decreases, the efficiency drops significantly, reaching only 76.5% at a load factor of 30%.

[0207] Hydraulic efficiency curve: The highest efficiency is 63.8% in the flow rate range of 40 to 45 m³ / d; the efficiency drops rapidly when the flow rate is lower than 25 m³ / d or higher than 60 m³ / d;

[0208] Inverter efficiency characteristics: The highest efficiency is 96.5% around 45Hz, and the efficiency drops below 92% when the frequency is below 30Hz;

[0209] In the calculation of the grid state factor, based on the characteristics of the regional power grid, the harmonic weight coefficient is set to 0.6 and the load weight coefficient is set to 0.4, reflecting the actual situation that harmonics have a more obvious impact on the system.

[0210] In view of the aging of the oil well (it has been in operation for 4.5 years), the motor efficiency baseline curve in the model was lowered by 8% to more accurately reflect the actual equipment condition.

[0211] Harmonic analysis of the inverter output revealed that the oilfield inverter primarily generates harmonics of the 5th (approximately 8.2%), 7th (approximately 6.5%), and 11th (approximately 4.3%) order. Combined with grid impedance testing, a resonance risk point was identified near 250 Hz, corresponding to the 5th harmonic frequency.

[0212] Based on this, the system has taken the following harmonic management measures:

[0213] Adjust the PWM carrier frequency: increase it from the original 4kHz to 8kHz, reducing the low-order harmonic content;

[0214] Configure the output side LC filter: a filter with an inductance of 0.5mH and a capacitance of 100μF is designed for the 5th harmonic;

[0215] Implement intelligent carrier frequency adjustment: when grid impedance changes are detected, it automatically searches for the optimal carrier frequency within the range of 6-10kHz;

[0216] In addition, the system automatically switches to low harmonic mode ( ), actively reduce the output power and frequency to control THD within a safe range.

[0217] The state space for the YJ-05 well contains 12 parameters, and the action space is discretized into 9 frequency adjustment options (-6 Hz to +6 Hz, with a step size of 1.5 Hz).

[0218] The deep Q network model adopts a structure suitable for the characteristics of this well:

[0219] The input layer receives 12-dimensional state parameters;

[0220] The three hidden layers have 96, 48, and 24 neurons respectively, using the LeakyReLU activation function;

[0221] The output layer is the Q value of 9 actions.

[0222] The weights of each indicator in the reward function are adjusted according to the production needs of the oil field: energy consumption index 0.45, production efficiency index 0.35, and grid friendliness index 0.2, giving priority to energy efficiency and output requirements.

[0223] The system was first pre-trained in a simulation environment using two years of historical data from the well, reaching convergence after approximately 80,000 training samples. Then, in actual deployment, an ε-greedy strategy was employed, with an initial exploration rate set to 0.1 and gradually reduced to 0.01 every 2,000 hours of operation, ensuring continuous learning and stability.

[0224] Three operating modes were deployed in this well:

[0225] Energy efficiency priority mode ( ): Applicable to periods when the power grid is in good condition and electricity prices are low, and the operating frequency is adjusted between 42 and 48 Hz;

[0226] Low harmonic mode ( ): When THD>10% or the grid impedance is abnormal, it is enabled, the operating frequency is limited to the range of 38 to 45Hz, and the PWM carrier frequency is increased to the maximum value;

[0227] Peak-valley balance mode ( ): Activated during peak electricity price periods (10:00 AM to 3:00 PM and 6:00 PM to 9:00 PM daily), the operating frequency is reduced to 35 to 40 Hz, reducing energy consumption. The system sets a minimum interval of 30 minutes between mode switches to avoid mechanical shock caused by frequent switching. Furthermore, a gradual switching mechanism is used, with the frequency change rate controlled to no more than 2 Hz / minute, ensuring a smooth transition.

[0228] The system generates a 24-hour operation plan based on the regional electricity price policy and the production demand of the well:

[0229] During the period of lowest electricity price from 1:00 AM to 5:00 AM, the system operates at the highest efficiency (frequency 46 Hz);

[0230] During the peak hours of 10:00 AM to 3:00 PM, the frequency is reduced to 38 Hz, reducing power consumption by approximately 30%.

[0231] During the period of high output demand but moderate electricity prices (5:00-9:00), maintain a medium frequency of 42Hz;

[0232] On the day before equipment maintenance (the 15th of each month), it automatically adjusts to conservative operation mode and limits the frequency to below 40Hz.

[0233] When actual well conditions deviate from the forecast, the system automatically adjusts the plan. For example, if the abnormal drop in liquid level exceeds the warning threshold (0.5m / h), replanning will be triggered, adjusting the operating frequency for subsequent periods.

[0234] The application of this implementation method in the above-mentioned oil field was evaluated after 6 months of operation, and the following two technical effects were verified:

[0235] First, the effect of improving energy efficiency;

[0236] After applying this method, the energy consumption of 12 wells in the oil field has been significantly improved. Taking the YJ-05 well as an example, while maintaining a basically stable production (daily liquid production fluctuation does not exceed ±5%), its energy consumption index changes are as follows:

[0237] Average unit energy consumption before application: 2.85kWh / m³ (electricity required to lift each cubic meter of liquid);

[0238] After one month of application: 2.42kWh / m³, a decrease of 15.1%;

[0239] After three months of application: 2.28kWh / m³, a decrease of 20.0%;

[0240] After six months of operation, the system achieved a 24.2% reduction to 2.16 kWh / m³. Particularly during peak electricity price periods, the system's intelligent scheduling reduced power consumption by an average of 28.5%, while also shifting some production load to low-price periods. This reduced the annual electricity bill for all 12 wells from 3.2 million yuan to 2.48 million yuan, saving approximately 720,000 yuan, a 22.5% reduction.

[0241] At the same time, the system's utilization of efficient operating ranges is improved:

[0242] Before application: The high-efficiency range (efficiency > 85%) accounted for only 32% of the operating time;

[0243] After application: The proportion of operating time in the efficient range increased to 78%, an increase of 46 percentage points.

[0244] It is worth noting that the energy efficiency improvement effect shows a continuous optimization trend over time, which is due to the self-optimization ability of the reinforcement learning model. As data accumulates and experience becomes richer, the control strategy continues to improve.

[0245] Secondly, harmonic management and equipment reliability improvement effects;

[0246] Harmonic pollution is the main factor affecting equipment reliability. After applying this method, the harmonic management effect is better:

[0247] THD changes:

[0248] Before application: The average THD was 15.2%, frequently exceeding the grid standard limit;

[0249] After one month of application: the average THD dropped to 7.8%, a decrease of 48.7%;

[0250] Six months after application: Average THD further decreased to 5.2%, a reduction of 65.8%.

[0251] Changes in key harmonic components (taking the 5th harmonic as an example):

[0252] Before application: the average 5th harmonic content is 8.2%;

[0253] After application: the 5th harmonic content dropped to 2.5%, a reduction of 69.5%.

[0254] The direct benefit of harmonic reduction is reduced equipment heat and extended service life:

[0255] The average operating temperature of the inverter dropped from 65°C to 52°C, a decrease of 13°C.

[0256] The motor winding temperature dropped from 95°C to 82°C, a decrease of 13°C;

[0257] The equipment failure rate has decreased, with only one motor failure occurring during the six-month period, compared to an average of six to eight during the same period previously.

[0258] Based on accelerated aging tests and life prediction models, the average equipment service life is expected to be extended by 35%.

[0259] In addition, the quality of the power grid has also improved, with the grid voltage fluctuation rate reduced from the original ±7% to ±4%, and the stability of other electrical equipment has also been improved.

[0260] Overall, this implementation achieves the dual goals of optimizing energy efficiency and managing harmonics in practical applications, reducing operating costs while also improving equipment reliability and service life, achieving excellent results. These benefits are expected to improve further as the system operates longer and data accumulates.

[0261] like Figures 2 to 6 As shown, there are respectively a line graph showing the changing trend of unit energy consumption of the deep well pump at different time points before and after the application of the method of the present invention; a bar graph showing the improvement of different harmonic types before and after the application of the method of the present invention; a radar graph showing the comparison of the performance of the traditional control method and the three operating modes proposed by the method of the present invention in terms of five key performance indicators; a scatter plot showing the relationship between carrier frequency and total harmonic distortion rate; and an area graph showing the changes in power demand, electricity price and energy cost before and after the application of the method of the present invention within 24 hours.

[0262] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make more forms of equivalent embodiments based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. A deep well pump operation frequency conversion control method, characterized in that: include: Construct a comprehensive energy efficiency evaluation model, which includes motor efficiency sub-models, hydraulic efficiency sub-models, and inverter efficiency sub-models. It also introduces a grid state factor and an electricity price factor. The grid state factor is calculated based on the grid quality and load conditions, taking into account the ratio of the current total harmonic distortion rate to the maximum allowable harmonic distortion rate, as well as the ratio of the current grid load rate to the maximum allowable load rate. The electricity price factor is calculated based on the ratio of the current time period electricity price to the benchmark electricity price. When the electricity price is at its peak, the deep well pump variable frequency control system automatically reduces the load and adjusts the operating time period. Based on the output results of the comprehensive energy efficiency evaluation model, the harmonic characteristics of the inverter are analyzed to obtain the total harmonic distortion rate and each harmonic component. The harmonic management strategy is determined according to the grid sensitivity and harmonic pollution level. The harmonic management strategy includes: Adjust the pulse width modulation parameters of the inverter; Configure harmonic filtering device; Optimize the inverter operating frequency to avoid harmonic resonance points; Switch to low harmonic mode when the grid harmonic content is high and the grid sensitivity is high; A multi-objective optimization algorithm is constructed based on a comprehensive energy efficiency evaluation model and harmonic management strategy. The multi-objective optimization algorithm comprehensively considers three goals: energy consumption, production efficiency, and grid friendliness, and uses reinforcement learning to generate the optimal variable frequency control strategy. Intelligent switching of operating modes is achieved based on the optimal variable frequency control strategy. Switching between energy efficiency priority mode, low harmonic mode, and peak-valley balance mode is performed according to the grid status and electricity price forecast results. The steps for implementing the intelligent switching of operating modes include: Define a set of operating modes for deep well pumps, including energy efficiency priority mode, low harmonic mode, and peak-valley balance mode; Collect and process grid status data; Build a mode switching decision algorithm to determine the most suitable operation mode based on the current grid status and electricity price information; Implement a smooth mode switching mechanism, set the minimum time interval for mode switching, and use gradual parameter adjustment; A predictive operation plan is generated based on the operation mode switching results and the grid state forecast. A daily operation plan is formulated in combination with the electricity price forecast and production demand. The plan is dynamically adjusted when the actual operating conditions deviate from the forecast. The steps for generating the predictive operation plan include: Collect and process forecast data, including future electricity price period forecasts, production target demand forecasts, equipment maintenance plans, and historical operating data statistics; Build a day-ahead planning model, divide the 24-hour period into sections, and assign different operation plans to each time period; Execute the operation plan optimization algorithm to generate the optimal daily operation plan; Develop forward-looking control strategies, preset target frequency ranges for each time period, and pre-configure mode switching time points; Implement a dynamic plan adjustment mechanism to adjust the operation plan when the actual operating conditions deviate from the forecast.

2. A deep well pump operation frequency conversion control method according to claim 1, characterized in that: The steps of constructing the comprehensive energy efficiency evaluation model include: Collect deep well pump operating parameter data and power grid status data; Construct a motor efficiency sub-model and calculate the motor efficiency curve under different load rates and frequencies; Construct a hydraulic efficiency sub-model to calculate the hydraulic efficiency of the pump under different flow and head conditions; Construct an inverter efficiency sub-model to calculate the inverter efficiency under different output frequencies and loads; The motor efficiency sub-model, hydraulic efficiency sub-model and inverter efficiency sub-model are integrated to calculate the total efficiency of the deep well pump variable frequency control system. The grid status factor and electricity price factor are introduced to construct a comprehensive energy efficiency index.

3. A deep well pump operation frequency conversion control method according to claim 1, characterized in that: The steps to construct a multi-objective optimization algorithm include: Define the state space and action space of the deep well pump variable frequency control system; Define a multi-objective reward function that comprehensively considers energy consumption indicators, production efficiency indicators, and grid-friendliness indicators; Build a deep Q network model; Train a deep Q-network model to maximize cumulative rewards; Generate optimal variable frequency control strategy.

4. A deep well pump operation frequency conversion control method according to claim 3, characterized in that: The constructed deep Q network model includes: Input layer, receiving state parameters of deep well pump variable frequency control system; At least one hidden layer to extract features and transform the input data; Output layer, outputs the Q value of each action; Experience replay buffer, which stores state transition samples; Both the target network and the evaluation network are deep neural networks; The constraint processing module ensures that the generated control strategy meets the constraint of minimum flow requirement.

5. A deep well pump operation frequency conversion control system, characterized in that: A method for frequency conversion control of a deep well pump according to any one of claims 1 to 4, comprising: Energy efficiency evaluation module, which is used to build a comprehensive energy efficiency evaluation model, integrating motor, hydraulic and inverter efficiency sub-models, and taking into account grid status and electricity price factors; Harmonic management module, used to analyze the harmonic characteristics of the inverter, evaluate the sensitivity of the power grid, and implement harmonic suppression measures; The optimization control module is used to build a multi-objective optimization algorithm based on reinforcement learning to balance energy consumption, production efficiency, and grid friendliness and generate the optimal control strategy; Mode switching module, used to intelligently switch between three operating modes: energy efficiency priority, low harmonics, and peak-valley balance; The plan generation module is used to formulate predictive daily operation plans and dynamically adjust them according to actual operating conditions to achieve forward-looking energy management.

Images