A method, apparatus, equipment and storage medium for pumping and sand removal

Abstract

This invention relates to the field of hydrogeological exploration and groundwater environment monitoring technology, and discloses a method, apparatus, equipment, and storage medium for groundwater pumping and sampling for sand removal. The invention first introduces raw groundwater into a gravity sedimentation unit for coarse sand removal pretreatment. Based on operational requirements, either a high-efficiency sand removal mode or a fine sampling mode is selected, and corresponding control targets are set. Cascaded dual-loop PI control is executed according to the selected mode. Simultaneously, based on real-time operating data, the PI control parameters are iteratively optimized using a self-tuning model with preset reinforcement learning parameters. This invention can simultaneously meet the needs of high-flow-rate, high-efficiency sand removal and high-precision laboratory sampling, significantly improving adaptability to complex operating conditions, and substantially increasing sand removal efficiency and sampling accuracy. It can be widely applied to various groundwater pumping, sampling, and sand removal operations.

Description

Technical Field

[0001] This invention relates to the field of hydrogeological exploration and groundwater environment monitoring technology, specifically to a method, apparatus, equipment and storage medium for pumping, sampling and sand removal. Background Technology

[0002] Existing sand removal control technologies mostly employ single-mode control with fixed parameters, which cannot simultaneously meet the dual requirements of high-flow-rate, high-efficiency water intake and high-precision laboratory sampling. They also exhibit poor adaptability to complex and fluctuating field conditions, resulting in unstable sand removal performance, insufficient control precision, and weak automated operation and maintenance capabilities, making it difficult to meet the actual needs of field operations. Therefore, there is an urgent need for a water pumping and sampling sand removal method, device, equipment, and storage medium to improve the adaptability of sand removal control under various operating conditions, while balancing water intake efficiency and sampling accuracy. Summary of the Invention

[0003] In view of this, the present invention provides a method, apparatus, equipment and storage medium for water pumping and sand removal, in order to solve the problems in the prior art where water pumping and sand removal control cannot take into account both water intake efficiency and sampling accuracy, has poor adaptability to complex and fluctuating working conditions in the field, has insufficient sand removal effect and control accuracy, and has weak automated operation and maintenance capabilities.

[0004] In a first aspect, the present invention provides a method for pumping water and sampling to remove sand, applied to a pumping water and sampling system for removing sand, the method comprising: Groundwater is fed into a gravity sedimentation unit, where large particles of sand and gravel are removed through gravity sedimentation, completing the coarse sand removal pretreatment. In response to operational needs, the working mode is selected and control targets are set. For large-volume water intake, a high-efficiency sand removal mode is selected and a target pumping flow rate is set. For laboratory sampling, a fine sampling mode is selected and a target effluent turbidity is set. Based on the pumping and sampling requirements, the dual-loop PI (Proportional Integral) system is switched and executed according to the selected mode. Proportional-Integral (PI) Control: The high-efficiency sand removal mode adopts dual-loop PI control of flow rate and valve opening, while the fine sampling mode adopts dual-loop PI control of turbidity and centrifugal sand remover speed. During system operation, based on real-time operating data, the PI control parameters are iteratively optimized through a preset reinforcement learning parameter self-tuning model to output the optimal control parameters adapted to the current operating conditions. The reinforcement learning parameter self-tuning model has a preset multi-dimensional reward and penalty function adapted to the pumping sampling sand removal scenario. The multi-dimensional reward and penalty function includes at least three reward items: turbidity accuracy, sand removal efficiency, and operational stability. During system operation, the fluctuation range of the operating conditions is monitored in real time. When the fluctuation exceeds a preset threshold, the parameter iteration optimization is paused and the current optimal control parameters are locked. The self-tuning process is resumed after the operating conditions stabilize.

[0005] In one optional implementation, the execution logic of the dual-loop PI control is as follows: For the high-efficiency sand removal mode, the target pumping flow rate is used as the outer loop setpoint and the real-time collected water pipeline flow rate is used as the outer loop feedback value to calculate and output the target valve opening value. Then, the target valve opening value is used as the inner loop setpoint and the real-time valve opening feedback is used as the inner loop feedback value to output a control signal to adjust the valve opening. For the fine sampling mode, the target effluent turbidity is used as the outer loop setpoint and the real-time collected effluent turbidity is used as the outer loop feedback value to calculate and output the target centrifugal sand remover speed. Then, the target centrifugal sand remover speed is used as the inner loop setpoint and the real-time centrifugal sand remover speed feedback is used as the inner loop feedback value to output a control signal to adjust the centrifugal sand remover speed.

[0006] In one optional implementation, the reinforcement learning parameter self-tuning model is a Q-Learning (Q-learning is a value-iteration-based reinforcement learning algorithm) model customized for the pumping sampling and desanding scenario. It has a preset state space, action space, and multi-dimensional reward and punishment function that matches the scenario. The state space includes turbidity deviation, rotation speed deviation, and desanding efficiency-related features. The action space includes the PI parameter adjustment amount of the dual-loop PI control. The multi-dimensional reward and punishment function includes three types of reward items: turbidity accuracy, desanding efficiency, and operational stability.

[0007] In one optional implementation, before the system is officially run, the model is pre-trained through a square wave step test, and the initial value of the Q-table (a two-dimensional table storing the value of state-action pairs in the Q-Learning algorithm) is calculated based on the step response accuracy. During the parameter self-tuning process, an ε-greedy strategy (an action selection strategy that balances the exploration and utilization capabilities of reinforcement learning) that decreases with running time is used to complete the action selection, and an iterative convergence termination condition is preset. If any iterative convergence termination condition is met, the iteration is terminated and the optimal PI control parameters are locked.

[0008] In one optional implementation, the rule for determining the fluctuation range of the operating conditions is as follows: the turbidity deviation change rate and the inlet water pressure change rate of multiple consecutive system sampling cycles are used as the determination indicators. When any indicator exceeds a preset threshold, it is determined that the fluctuation range of the operating conditions exceeds the limit.

[0009] In one optional implementation, during the coarse sand removal pretreatment process, the liquid level data of the gravity sedimentation unit is collected in real time. When the liquid level reaches the preset threshold, the raw water inlet equipment is automatically shut down and the overflow protection is triggered. After a single operation is completed, the system automatically starts the backwashing module and the sewage discharge module to perform backwashing and sewage discharge operations on the flow channels and sand accumulation bins of the gravity sedimentation unit, the centrifugal sand remover, and the sand accumulation bin, respectively.

[0010] Secondly, the present invention provides a water pumping and sand removal device, the device comprising: The system comprises the following modules: a pretreatment module for introducing groundwater into a gravity sedimentation unit to remove large sand particles and complete coarse sand removal pretreatment; a mode matching module for selecting the operating mode and setting control targets in response to operational needs, selecting the high-efficiency sand removal mode and setting the target pumping flow rate for large-volume water intake, and selecting the fine sampling mode and setting the target effluent turbidity for laboratory sampling; a dual-loop control module for executing corresponding dual-loop PI control according to the selected mode, outputting valve opening control signals for the high-efficiency sand removal mode and centrifugal sand remover speed control signals for the fine sampling mode; a parameter self-tuning module for iteratively optimizing PI control parameters based on real-time operating data using a preset reinforcement learning parameter self-tuning model during system operation, outputting the optimal control parameters adapted to the current operating conditions to the dual-loop control module, wherein the model is configured with a multi-dimensional reward and punishment function adapted to the pumping, sampling and sand removal scenario; and an operating condition monitoring and protection module for real-time monitoring of operating condition fluctuations, pausing parameter iteration optimization and locking the current optimal control parameters when fluctuations exceed a preset threshold, and resuming the self-tuning process after the operating conditions stabilize.

[0011] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the pumping, sampling and sand removal method of the first aspect or any corresponding embodiment described above.

[0012] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions, the computer instructions being used to cause a computer to perform the pumping, sampling and sand removal method of the first aspect or any corresponding embodiment described above.

[0013] Fifthly, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the pumping, sampling and sand removal method of the first aspect or any corresponding embodiment described above. Attached Figure Description

[0014] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of the overall process of a water pumping and sand removal control method according to an embodiment of the present invention; Figure 2This is a schematic flowchart of a water pumping sampling and sand removal control method according to an embodiment of the present invention. Figure 3 This is a schematic diagram showing the details of the dual-loop PI control execution logic of a water pumping sampling and sand removal control method according to an embodiment of the present invention. Figure 4 This is a schematic diagram of the Q-Learning reinforcement learning parameter self-tuning algorithm for a water pumping sampling and sand removal control method according to an embodiment of the present invention. Figure 5 This is a structural block diagram of a water pumping, sampling and sand removal device according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] According to an embodiment of the present invention, a method for controlling water pumping, sampling and sand removal is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here. Example

[0018] According to an embodiment of the present invention, a method for controlling water pumping, sampling and sand removal is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0019] This embodiment provides a water pumping and sand removal control method, which can be used in the aforementioned water pumping and sand removal control system, such as intelligent hydrological monitoring terminals, automated water pumping and sand removal equipment, and field hydrological sampling devices. Figure 1 This is a flowchart of a water pumping, sampling, and sand removal control method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Select the working mode in response to the operation requirements and collect real-time operating data of the pumping and sand removal system; the working mode includes a high-efficiency sand removal mode and a fine sampling mode. The high-efficiency sand removal mode uses the target pumping flow rate as the control target, and the fine sampling mode uses the target effluent turbidity as the control target; the real-time operating data includes the water supply pipeline flow rate, effluent turbidity, centrifugal sand remover speed, inlet water pressure, and gravity sedimentation unit liquid level.

[0020] This step is used to determine the control objectives based on actual operational needs and to comprehensively collect core operating condition information such as flow rate, turbidity, equipment speed, pipeline pressure, and liquid level. This provides a real and complete data foundation for subsequent mode switching, dual-loop PI control, and reinforcement learning parameter optimization, ensuring that the control strategy matches the operational needs and the actual state of the system.

[0021] Step S102: Switch to the selected mode and execute the corresponding dual-loop PI control; the high-efficiency sand removal mode adopts dual-loop PI control of flow rate and valve opening, with the target pumping flow rate as the outer loop set value and the real-time flow rate as the outer loop feedback, to adjust the opening of the flow regulating valve; the fine sampling mode adopts dual-loop PI control of turbidity and centrifugal sand remover speed, with the target effluent turbidity as the outer loop set value and the real-time turbidity as the outer loop feedback, to adjust the centrifugal sand remover operating speed.

[0022] This step utilizes a dual-loop PI control architecture bound to the operating mode to achieve efficiency-priority control during high-flow water intake and turbidity accuracy-priority control during laboratory sampling. This approach balances water intake efficiency and water sample accuracy in the control structure, solving the problem that a single control logic cannot adapt to the two operating requirements.

[0023] In step S103, during system operation, based on standardized operating condition data, the PI control parameters are iteratively optimized through a preset reinforcement learning parameter self-tuning model. The reinforcement learning parameter self-tuning model is configured with a multi-dimensional reward and punishment function adapted to the pumping sampling and sand removal scenario, which includes at least a turbidity accuracy reward item, a sand removal efficiency reward item, and an operational stability reward item. The model takes the operating condition deviation as the state and the PI parameter adjustment amount as the action, continuously iterates to find the optimal PI control parameters under the current operating condition, and outputs the optimal PI control parameters under the current operating condition.

[0024] This step utilizes a scenario-customized reinforcement learning self-tuning strategy, combined with a multi-dimensional reward and punishment mechanism to guide the model optimization direction. During operation, the PI parameters are adaptively adjusted in real time to avoid the problems of insufficient accuracy and poor adaptability of manual tuning, thereby improving the control effect and stability under complex working conditions.

[0025] Step S104: Monitor the fluctuation range of the operating condition in real time. When the fluctuation exceeds the threshold, pause the parameter iteration optimization and lock the current optimal PI parameter. Resume the self-tuning process after the operating condition stabilizes. Use the turbidity deviation change rate and the influent pressure change rate of multiple consecutive sampling cycles as the judgment indicators. If any indicator exceeds the preset threshold, it is determined that the operating condition fluctuation exceeds the limit. At this time, pause the iteration and keep the currently optimized control parameter. Restart the parameter self-tuning after the operating condition falls back to the stable range.

[0026] This step sets up a protection mechanism for complex working conditions such as sudden changes in groundwater pressure and fluctuations in sand content in the field, so as to avoid the reinforcement learning model from diverging and parameter jumps caused by drastic working condition disturbances, ensure the continuous and stable operation of the system, and eliminate the risk of water sample exceeding the standard and equipment impact.

[0027] Step S105: Drive the actuator to act according to the optimal PI parameter, and perform safety protection and automatic sewage flushing when the conditions are met; send the optimal PI parameter to the actuator such as the flow regulating valve and the centrifugal sand remover to complete the control action; automatically shut off the water inlet and trigger the overflow protection when the liquid level of the gravity sedimentation unit exceeds the threshold; automatically start the backwash and sewage discharge module after the single operation is completed to flush and discharge sewage from the sedimentation unit, the sand removal equipment channel and the sand accumulation bin.

[0028] This step completes the closed-loop execution of control commands, while integrating liquid level safety protection and automated operation and maintenance functions, further improving system operation safety and automated operation and maintenance capabilities, and reducing the need for manual on-site supervision.

[0029] The water pumping sampling and sand removal control method provided in this embodiment, through a complete technical solution of patterned dual-loop PI control, reinforcement learning self-tuning combined with multi-dimensional rewards and penalties, and adaptive protection against operating condition fluctuations, can simultaneously meet the needs of efficient water intake and high-precision laboratory sampling under complex and variable operating conditions in the field, avoid algorithm divergence and control instability, and significantly improve the operating condition adaptability, control accuracy and automation level of the water pumping sampling and sand removal system. Example

[0030] This embodiment describes in detail the specific implementation of the pumping sampling and sand removal system and control method based on real-time feedback of operating conditions. The technical solution described therein corresponds completely to the protection scope defined by the claims of this application. It can solve the technical problems of insufficient sand removal stability and low sampling accuracy caused by fixed parameters and poor adaptability to operating conditions in existing sand removal control methods. Based on the content described in this embodiment, those skilled in the art can fully implement the technical solution of this application without creative effort.

[0031] This embodiment is applicable to water pumping and sand removal operations for groundwater resource surveys, hydrogeological exploration of mining areas, environmental monitoring, and engineering geological exploration. The water pumping and sand removal system includes an inlet and gravity sedimentation unit, an electric regulating valve, a flow sensor, a centrifugal sand remover, a turbidity sensor, and a control unit MCU (Microcontroller Unit). The system is also equipped with a backwashing module, a sewage discharge module, and multiple sets of electric regulating valves to realize system flushing, sewage discharge, and abnormal liquid level protection functions. The inlet water and gravity sedimentation unit is used for coarse sand removal pretreatment of raw water. It is equipped with a liquid level detection component. When the liquid level exceeds a preset threshold, the control unit (MCU) controls the inlet water pump and inlet valve to stop water intake, preventing equipment malfunctions caused by abnormal liquid levels. The electric regulating valve receives the opening control signal from the control unit (MCU) and adjusts its own opening, while simultaneously feeding back the valve position signal to the MCU. A flow sensor is installed in the pipeline between the electric regulating valve and the centrifugal sand separator, feeding back the pumping flow signal to the control unit (MCU). The centrifugal sand separator is used for fine sand removal treatment of raw water. It receives the speed control signal from the control unit (MCU) and feeds back the actual speed signal to the control unit (MCU). A turbidity sensor is installed inside the centrifugal sand separator, feeding back the turbidity signal to the control unit (MCU).

[0032] The control method described in this embodiment has two working modes: a high-efficiency sand removal mode and a fine sampling mode, to adapt to different water pumping and sampling requirements. The high-efficiency sand removal mode is used to achieve large-volume water pumping and sand removal, while the fine sampling mode is used to achieve high-precision water quality testing and sampling. The high-efficiency sand removal mode adopts a dual-loop control method combining an outer loop for flow rate and an inner loop for the opening of the electric regulating valve at the centrifugal sand remover inlet. The outer loop calculates the difference between the target pumping flow rate and the current pumping flow rate, and the calculation result is input to the PI controller PI1. PI1 outputs a reference value for the opening of the electric regulating valve to the inner loop. The inner loop calculates the difference between the reference value for the opening of the electric regulating valve and the valve position feedback, and the calculation result is input to the PI controller PI2. PI2 outputs an opening control signal to dynamically adjust the valve opening, thereby achieving stable control of the pumping flow rate and suppressing the impact of flow fluctuations on the sand removal effect. The fine sampling mode employs a dual-loop control method combining an outer loop for turbidity and an inner loop for centrifugal desander speed. The outer loop calculates the difference between the target turbidity and the turbidity feedback, and the result is input to the PI controller PI3. PI3 outputs the target speed of the centrifugal desander to the inner loop for speed. The inner loop calculates the difference between the target speed and the speed feedback, and the result is input to the PI controller PI4. PI4 outputs a speed control signal to dynamically adjust the speed of the centrifugal desander, thereby achieving high-precision and stable control of turbidity and improving the accuracy of water sample collection.

[0033] The switching between the two working modes adopts a smooth transition scheme without disturbance: when a mode switching command is received, the control unit MCU pauses the parameter self-tuning process and locks the current actuator control signal; with a step size of 500ms, the opening of the electric regulating valve and the speed of the centrifugal sand remover are linearly adjusted to complete the smooth transition from the steady-state value of the original mode to the initial set value of the new mode within 5s; after the transition is completed and the system is stable for 3 sampling cycles, the dual-loop PI control and parameter self-tuning process corresponding to the new mode are started to avoid drastic fluctuations in flow and turbidity and equipment impact caused by mode switching.

[0034] The parameter design basis and determination method for the above-mentioned non-disruptive smooth transition scheme are as follows: First, it perfectly matches the system hardware characteristics and sampling control logic. The 500ms adjustment step size adapts to the step response characteristics of electric regulating valves and variable frequency drive centrifugal sand removers, avoiding overcurrent, wear, and water hammer in pipelines caused by frequent actuator movements; the 5s transition time matches the system's first-order inertia + pure time delay dynamic characteristics, balancing switching smoothness and operating efficiency; the steady-state determination rule of 3 sampling cycles perfectly matches the system's 100ms sampling cycle, eliminating sampling noise interference and ensuring that closed-loop control is started only after the system is stable.

[0035] Secondly, it conforms to the professional specifications of industrial fluid control and cascade PI control. The parameter values ​​are the general engineering optimal values ​​for seamless mode switching in the field of water conservancy and water treatment. Verified by a large number of engineering practices, it can completely avoid drastic fluctuations in flow and turbidity and equipment shock caused by mode switching.

[0036] Third, it is fully adaptable to field survey operations. The parameter design takes into account the power stability, equipment protection requirements, and operational efficiency requirements of field operations, without placing additional burden on on-site operations, and can stably and smoothly switch between the two working modes under complex field conditions.

[0037] Fourth, the on-site calibration method is clearly defined. Before operation, parameter calibration is completed through a square wave step test. A mathematical model of the controlled object on-site is established using the step response method. The initial values ​​of the PI parameters adapted to the on-site working conditions are calculated by substituting these models into the tuning formula. Fine-tuning is then performed during trial operation to ensure the parameters fall within the aforementioned range. Those skilled in the art can complete parameter calibration and system debugging without any creative effort. The PI parameter recalibration trigger rule for the high-efficiency sand removal mode is as follows: when the steady-state deviation of the flow rate exceeds ±5% for 10 consecutive sampling cycles, or the groundwater pressure change exceeds ±30% of the rated operating condition, the system automatically suspends water intake operation, performs a square wave step test to recalibrate the PI parameters, ensuring the parameters are adapted to the current working conditions; after calibration, the high-efficiency sand removal mode operation is automatically resumed. The above recalibration rule, designed for the actual operating environment where groundwater pressure and aquifer characteristics fluctuate drastically over time, accurately identifies the unstable state of system operation through the deviation trigger mechanism, effectively avoiding control oscillations or response lag problems caused by fixed parameters, and ensuring the continuous and stable operation of the system under dynamic conditions. This rule also serves as a safety fallback mechanism for the embedded control system, providing overload protection for core actuators such as electric regulating valves and centrifugal desanders. It prevents damage caused by instantaneous exceedances in flow and water pressure, ensuring equipment safety and operational continuity during field operations. Furthermore, through a closed-loop logic of "pause operation - parameter recalibration - system recovery," the algorithm can always perform calculations based on the latest field conditions and system model, significantly improving the environmental adaptability, operational reliability, and overall control efficiency of the high-efficiency desanding mode under complex and fluctuating conditions. For the fine sampling mode, due to the characteristics of the controlled objects in its turbidity outer loop and centrifugal desander speed inner loop, which fluctuate drastically with groundwater sand content and water pressure, fixed PI parameters cannot adapt to changing conditions and are prone to causing significant fluctuations in sampling accuracy. Therefore, this embodiment employs a learning-based control parameter determination method to adaptively optimize the PI control parameters to solve the problem of fixed parameters being unable to adapt to changing conditions.

[0038] In this embodiment, the high-efficiency sand removal mode adopts fixed PI parameter control. The proportional coefficient of the PI parameter ranges from 1.2 to 3.0, and the integral coefficient ranges from 0.2 to 1.0. The initial values ​​of the parameters are calibrated through on-site step tests. The design basis and determination method of this parameter range are as follows: (1) Matching the core control objectives and characteristics of the controlled object in the high-efficiency sand removal mode. The high-efficiency sand removal mode is a cascaded double-loop PI control of the flow outer loop and the valve opening inner loop. The core objective is to achieve stable water intake with large flow, rapid response to fluctuations in operating conditions, and ensure stable sand removal efficiency. The controlled object is a first-order inertial + pure time-delay flow system composed of an electric regulating valve, a water pipeline, and a centrifugal sand remover. This parameter range is fully compatible with the dynamic characteristics of the controlled object and the typical operating conditions of groundwater pressure of 0.2 to 0.8 MPa and target flow of 5 to 50 m³ / h in the field.

[0039] (2) Determination based on theoretical calculations using general engineering tuning methods in the field of automatic control. A mathematical model of the system is established using the step response method. The Ziegler-Nichols tuning method and the attenuation curve method, which are commonly used in the field of hydrological and hydraulic fluid control, are used to calculate that the theoretical feasible range of the proportional coefficient is 1.0 to 3.2 and the theoretical feasible range of the integral coefficient is 0.1 to 1.2. The optimal range of the above engineering coefficients is determined by combining the robustness and anti-interference requirements of the actual engineering.

[0040] (3) Verification based on orthogonal comparative tests under typical field conditions. Through comparative tests under multiple conditions, it was verified that when the proportional coefficient is below 1.2, the system response is lagging, the steady-state flow deviation is large, and the sand removal efficiency cannot be stably met; when it is above 3.0, the system is prone to flow oscillation and frequent valve operation, which can cause water hammer and liquid level fluctuations in the pipeline, resulting in a decrease in sand removal efficiency. When the integral coefficient is below 0.2, the steady-state flow error cannot be eliminated, and the water intake cannot meet the operating quota; when it is above 1.0, it is easy to cause integral saturation, leading to system oscillation and loss of control. Within this parameter range, the requirements for flow control accuracy, response speed, and operational stability can be met simultaneously, and the sand removal efficiency can be stably achieved at over 95%.

[0041] First, define the state space S and the action space A, clear the experience replay set D used to store historical interaction data, and initialize all parameters of the Q-Learning algorithm.

[0042] The state space S is a five-dimensional feature vector reflecting the operating state of the system, expressed as: ; in, Turbidity deviation of the outer ring output (turbidity outer ring output deviation of centrifugal desander), unit NTU, value range: -10 to +10 NTU; The turbidity deviation change rate of the outer ring turbidity (turbidity deviation change rate of the outer ring turbidity of the centrifugal sand remover), unit NTU / s, value range: -5~+5NTU / s; The deviation of the inner ring speed of the centrifugal sand remover is expressed as a percentage of the rated speed, with a range of -20% to +20% of the rated speed. The value represents the rate of change of the inner ring rotational speed deviation of the centrifugal sand remover, expressed as % of rated speed / s, with a range of -10% to +10% of rated speed / s. This represents the system's real-time sand removal efficiency, expressed as a percentage, with a range of 0–100%.

[0043] For water intake, This refers to the inflow rate; The state space dimension setting is based on: covering all core control objectives such as turbidity control, rotation speed control, deviation, rate of change, and sand removal efficiency, ensuring that the algorithm input is comprehensive and without omission, and adapting to the multi-objective control requirements of water pumping and sampling.

[0044] In this invention, the sand removal efficiency The system water recovery rate is used to characterize the proportion of effective water that can be used for testing after the sand removal operation, relative to the total influent. The higher the value, the less water loss during the system's sand removal process, and the higher the efficiency of water intake operations, which can effectively ensure the efficiency of field water pumping and sampling operations and the representativeness of water samples.

[0045] The state space discretization rules are as follows: For each dimension of the five-dimensional feature vector, combined with the fluctuation range of typical field conditions, the state space is divided into a finite number of discrete intervals using an equal-interval discretization method: Turbidity deviation Fluctuation range ±10 NTU, discrete into 20 intervals; Turbidity deviation change rate Fluctuation range ±5 NTU / s, discrete into 20 intervals; Speed ​​deviation Fluctuation range ±20% of rated speed, discrete into 20 intervals; Speed ​​deviation change rate Fluctuation range ±10% of rated speed / s, discrete into 20 intervals; Sand removal efficiency The fluctuation range is 80% to 100%, and it is discrete into 10 intervals. After discretization, it covers the entire working condition range of field pumping and sampling. Those skilled in the art can adjust the number and range of discrete intervals according to the on-site hardware range.

[0046] The design basis and determination method for the above discretization range and number of intervals are as follows: First, it fully matches the system hardware characteristics and industry engineering specifications. The fluctuation range strictly conforms to the safe operating range of the centrifugal desander, the industry qualified standard for turbidity of groundwater exploration samples, covers the full operating range of field pumping and sampling, and has no redundant or invalid states; the number of intervals matches the system's 100ms sampling period, the adjustment accuracy of the industrial frequency converter, and the steady-state accuracy requirements of turbidity control. Second, it meets the convergence theory requirements of the Q-Learning algorithm. The number of core effective states after discretization is 1000-2000 sets, which fully matches the 1500-5000 iterations set in the scheme, ensuring that the algorithm stably converges to the global optimal solution within the field operation time, balancing control accuracy and convergence speed. Third, it is verified by orthogonal comparison tests under typical field operating conditions. Through comparative tests under multiple working conditions, it was verified that under this discretization rule, the algorithm can converge to the optimal PI parameter with turbidity accuracy ≥98% within 2000 iterations. Compared with coarser / fineer discretization schemes, the convergence speed is improved by more than 40% and the control accuracy is improved by more than 30%, which is fully adaptable to complex and fluctuating working conditions in the field.

[0047] The action space A is a set consisting of PI parameter adjustment values, expressed as: ; in: The proportional gain adjustment value for the turbidity outer loop PI3 controller in fine sampling mode is 0.01, with a range of -0.1 to +0.1. The integral coefficient adjustment for the turbidity outer loop PI3 controller in fine sampling mode is set with an adjustment step of 0.01 and a range of -0.1 to +0.1. The proportional coefficient adjustment value of the PI4 controller in the inner loop of the centrifugal sand remover in fine sampling mode is 0.01, with a value range of -0.1 to +0.1. This is the integral coefficient adjustment value of the PI4 controller in the inner loop of the centrifugal sand remover in fine sampling mode. The adjustment step size is 0.01, and the value range is -0.1 to +0.1.

[0048] Regarding the setting of motion space parameters, this embodiment strictly limits the parameters based on engineering tuning specifications and actual hardware characteristics. The adjustment step size is set to 0.01, which is 0.3% to 0.8% of the initial PI parameters. This falls within the standard fine-tuning range of industrial PI self-tuning, ensuring the convergence speed of the reinforcement learning algorithm while effectively avoiding system overshoot and oscillation caused by excessive adjustment. Simultaneously, this step size setting perfectly matches the control resolution of the electric regulating valve and the centrifugal sand remover. Both the inverter frequency resolution and the regulating valve position resolution support fine adjustments at the 0.01 level, ensuring precise execution without overshoot or impact for every parameter adjustment, thus preventing overload damage to the actuator hardware.

[0049] Regarding the range of parameter adjustments, this embodiment limits it to -0.1 to +0.1. This range strictly controls the parameter adjustment within ±10% of the initial PI parameter. Calculations show that for the proportional coefficient, this translates to a fine-tuning range of ±3.3% to ±8.3%, and for the integral coefficient, it is controlled within a safe range of ±10% to ±50%. This limitation not only complies with engineering safety specifications, ensuring that the system always operates within the rated safe adjustment range of the electric regulating valve and the centrifugal sand remover, avoiding equipment damage due to frequent large-amplitude movements, but also, from a control theory perspective, limits the overexploration of the reinforcement learning algorithm, ensuring the system stability and control robustness of the algorithm during the optimization process.

[0050] Furthermore, the motion space dimension setting only selects four core parameters of the dual-loop PI controller: the proportional and integral coefficients of the turbidity loop and the proportional and integral coefficients of the speed loop, which are adjusted online. Throughout the adjustment process, the original dual-loop cascade control structure remains unchanged; parameter optimization is performed only within the established framework. This approach ensures that the control logic of the pumping sampling system remains consistent under complex nonlinear dynamic conditions, preventing stability issues caused by structural changes and enabling it to smoothly adapt to various dynamic conditions in field pumping operations.

[0051] Before the official commencement of pumping and sampling operations, Q-table initialization is performed using square wave step test data to establish the correlation between the system's dynamic characteristics and initial control parameters, thereby reducing ineffective exploration by the Q-Learning algorithm and accelerating convergence. The initialization process begins by calculating the step response accuracy P based on the square wave test data, expressed as:

[0052] in For target turbidity, The actual output turbidity is used; then, based on the step response accuracy P and the initial PI parameter, the initial value of the Q-table is calculated, and the expression is:

[0053] in This is the initial state. This is the initial action corresponding to the initial PI parameter adjustment amount. , The initial PI parameters obtained from the square wave step test calibration are identical in calibration method, test system, and operating condition to the initial PI parameter values ​​calibrated in the high-efficiency sand removal mode. Those skilled in the art can simultaneously complete the initial PI parameter calibration for both modes using the same square wave step test procedure. , The PI parameter weights are set to meet the proportioning requirements, where k is a proportionality coefficient set based on system characteristics.

[0054] In this embodiment, the following is taken =0.7、 =0.3, the sum of their weights is 1, which conforms to the general engineering specifications for weight allocation; the proportional coefficient k=5, and the design basis and determination method of this set of parameters are as follows: First, it conforms to the general engineering specifications for PI control in industrial processes. In cascade dual-loop PI control, the proportional coefficient K p The proportional gain is a core parameter that determines the system's response speed and dynamic deviation adjustment capability. The integral coefficient Kᵢ is an auxiliary parameter used to eliminate the system's steady-state error. Therefore, a higher weight needs to be assigned to the proportional gain during pre-training initialization. The 0.7 / 0.3 ratio is a common engineering ratio for PI parameter pre-tuning in the field of hydrological and hydraulic fluid closed-loop control, and it is fully compatible with the dynamic characteristics of the first-order inertia + pure time delay controlled objects, such as the electric regulating valve and centrifugal desander, in this application. Second, it fully matches the core design goal of the pre-training stage of this application. The core purpose of the pre-training in this application is to establish the correlation between the system's dynamic characteristics and the initial control parameters, reduce the invalid exploration of the Q-Learning algorithm, and enable the algorithm to directly start iterating from the near-optimal solution. This set of parameters can ensure that the calculated initial Q value is within a reasonable range, avoiding excessively aggressive algorithm and control oscillation caused by frequent jumps in PI parameters due to an excessively large initial Q value, or excessively slow algorithm convergence and excessive invalid exploration due to an excessively small initial Q value, thus fully achieving the design goal of pre-training. Third, the on-site calibration method is clear. Those skilled in the art can adaptively adjust the parameters based on the system response characteristics obtained from the on-site square wave step test. After adjustment, it is necessary to ensure... and The sum of their weights is 1.

[0055] The experience replay set D is used to store effective state-action pairs, corresponding PI parameter combinations, and operating condition characteristics that meet the requirements of turbidity accuracy ≥98% and sand removal efficiency ≥95% during historical iterations. After a single operation, the system stores the optimal parameter combination converged in this operation into the experience replay set D. When the next pumping and sampling operation is started, the system prioritizes matching the historical optimal parameters consistent with the current operating condition characteristics from the experience replay set D as a supplementary reference for Q-table initialization, further reducing invalid exploration by the algorithm and accelerating the convergence speed.

[0056] The system's real-time feature values ​​are calculated, and a multi-dimensional composite reward and penalty function is set to define clear optimization objectives and evaluation criteria for the Q-Learning algorithm. This balances turbidity control accuracy, sand removal efficiency, and system operational stability, guiding the algorithm to converge towards the optimal control strategy. The expression for the reward and penalty function is as follows:

[0057] Among them, the turbidity accuracy award item This is used to adjust the reward value based on the magnitude of the turbidity deviation, so that the system is optimized in the direction of improving turbidity accuracy; Sand removal efficiency bonus This is used to ensure the sand removal efficiency of the system and avoid insufficient sand removal effect due to only pursuing turbidity accuracy. Stability Bonus Used to improve system stability. The standard deviation of turbidity deviation The The turbidity deviation is calculated in real time using the turbidity deviation sequence within a sliding time window: taking the turbidity deviation of the most recent n sampling periods.

[0058] (i=1,2,...,n, where n is the sampling window size. Those skilled in the art can select an integer within the corresponding range according to the system response speed. In this embodiment, the sampling window n is 10 (range 5 to 20). This value is determined based on the response speed of the turbidity sensor and the operation cycle of the controller, which can ensure statistical reliability without causing calculation delay.) Let be the turbidity deviation at the i-th sampling time. This represents the actual effluent turbidity value at the i-th sampling time (i.e., turbidity feedback). Target turbidity; First, calculate the average value of the deviation sequence. , Then through the formula Calculated; The smaller the value, the more stable the turbidity control of the system and the higher the control stability, which can effectively ensure the accuracy and representativeness of water sample collection.

[0059] , , These are the weighting coefficients for the corresponding reward items.

[0060] In this embodiment, the following is taken =0.5、 =0.25、 =0.25, the sum of the weights of the three is 1. The design basis and determination method of this set of weight coefficients are as follows: First, it fully matches the core invention objective of the fine sampling mode of this application. The core purpose of this mode is to sample high-precision water samples for laboratory testing. The turbidity accuracy of the effluent is the core judgment indicator for qualified water samples in the "Technical Specification for Groundwater Environmental Monitoring", and it is the first priority of the algorithm optimization of this application. Therefore, it is assigned the highest weight. The sand removal efficiency is the bottom line requirement to ensure the long-term safe operation of the equipment and the basic quality of the water sample. The operational stability is the key constraint to avoid the continuous fluctuation of turbidity and equipment wear caused by frequent speed adjustment of the centrifugal sand remover. The two are equally important to the engineering of continuous field sampling operations. Therefore, the remaining weights are equally distributed. Second, it conforms to the design specifications of the reward and punishment function of the Markov decision process of reinforcement learning. This set of weight ratios can ensure that the algorithm optimization direction is completely consistent with the core technical objective of this application, avoid the problems of algorithm optimization direction deviation, control oscillation or iteration non-convergence caused by weight imbalance, and can stably guide the algorithm to iterate to the optimal PI parameter that is suitable for the current working conditions. Third, the field adaptation rules are clear. Those skilled in the art can adaptively adjust the weight coefficients within the range of 0.1 to 0.8 according to the priority requirements of turbidity accuracy, sand removal efficiency, and operational stability under the on-site working conditions. After adjustment, the sum of the weights of the three factors must be 1.

[0061] Select action A corresponding to state space S from the current Q-value output, and execute the action selection using an ε-greedy strategy that decreases over time: initial exploration probability. The number of iterations decreases exponentially, and the decreasing formula is as follows: Where k is the current iteration number and β is the decay coefficient, in this embodiment The lower bound of the exploration probability is ,when After decaying to the lower limit, it remains constant. Through the exploration and utilization capabilities of this strategy's balancing algorithm, effective operating conditions are fully explored in the early stages, and convergence to the optimal control strategy in the later stages. After selecting the action, the eigenvalues ​​corresponding to the new state S′ are obtained. The Q-Learning algorithm is used for coarse adjustment of the PI parameters. The Q-value update formula for the Q-Learning algorithm is:

[0062] Where α is the learning rate, which is 0.1 in this embodiment and ranges from 0.05 to 0.2; γ is the discount factor, which is 0.9 in this embodiment and ranges from 0.8 to 0.95; and R is the instant reward value. The update formula, which is the maximum Q value for the next state, is designed based on the Bellman optimal equation and the temporal difference (TD) learning algorithm. It can be optimized online in real time without the need to build an accurate system model, making it suitable for nonlinear systems such as pumping sampling.

[0063] The above-mentioned hyperparameter ranges and typical values ​​are all customized for the PI parameter self-tuning scenario of the present invention for water pumping, sampling and sand removal. The design basis and determination method are as follows: First, the design of the sampling period and iteration count strictly matches the system hardware characteristics and field operation scenarios. The sampling period strictly conforms to the Shannon sampling theorem for industrial control and matches the system hardware characteristics. The core controlled objects of this invention include electric regulating valves and centrifugal sand removers, with a minimum step response time of 100ms. According to the Shannon sampling theorem, the sampling period must be ≤1 / 2 of the minimum response time of the controlled object. Combining industrial control engineering margins, a standard sampling period of 100ms is determined. This period also matches the 10Hz standard output frequency of the industrial turbidity and flow sensors, occupying only about 30% of the computing power of the embedded control unit. This ensures that all tasks such as data acquisition, control calculation, actuator driving, and anomaly protection can operate in parallel and stably, balancing control accuracy and system stability. Comparative experiments have verified that under this sampling period, the steady-state deviation of flow control is ≤±1.5%, and the steady-state deviation of turbidity control is ≤±1.0NTU, fully meeting the control objectives of the invention.

[0064] The Q-Learning algorithm of this invention, after discretization of the state space, has approximately 1000 valid states. The minimum number of iterations required for the algorithm to converge needs to be at least 1.5 times the total number of state-action pairs. Therefore, 1500 iterations are determined as the lower limit to ensure that the algorithm fully explores the core working conditions, and the turbidity accuracy reaches more than 85% of the preset threshold after coarse adjustment. Considering the standard operation time of 5-10 minutes for a single pumping sampling in geological exploration, and a sampling period of 100ms, 5000 iterations are determined as the upper limit to ensure that the algorithm can be completed within the operation time, and the change in Q value after iteration is less than 10⁻. 4 The algorithm has fully converged, and further iterations offer no improvement in accuracy. Experiments have verified that 2000 iterations are sufficient to converge to the optimal PI parameters with turbidity accuracy ≥98%, balancing convergence speed and generalization capability across all operating conditions. Those skilled in the art can adjust the sampling period within the range of 50ms to 200ms and the number of iterations within the range of 1500 to 5000, depending on the on-site hardware model and operating time, achieving parameter adaptation without any creative effort.

[0065] Second, the core control objectives and scenario characteristics of this invention are matched. The Q-Learning algorithm of this invention is used for adaptive parameter tuning of cascaded dual-loop PI control in fine sampling mode. Its core objective is to quickly converge to the optimal PI parameters under complex conditions of groundwater sediment content and water pressure fluctuations in the field, while ensuring that turbidity remains consistently within acceptable limits throughout the sampling period and that the system operates stably over the long term. The learning rate α is used to balance the algorithm's convergence speed and steady-state accuracy, and the discount factor γ is used to balance the immediate control effect and the long-term stability of the system. The parameter values ​​are fully adapted to the requirements of this scenario, and the invention features a 100ms sampling period and 1500–5000 iterations.

[0066] Third, it conforms to the fundamental theories and convergence requirements of reinforcement learning. According to the stochastic approximation theory of reinforcement learning, the necessary and sufficient condition for the Q-Learning algorithm to converge to the global optimum is:

[0067] and

[0068] Based on the iteration number and sampling period of this invention, the theoretical feasible range of α is calculated to be 0.01 to 0.3, which is optimized and narrowed to the engineering optimal range of 0.05 to 0.2. Based on the Bellman optimal equation of the Markov decision process, the continuous process control scenario of this invention needs to prioritize the long-term operational stability. Therefore, γ is taken in a relatively high range of 0.8 to 0.95, which fully complies with the professional theoretical specifications of reinforcement learning.

[0069] Fourth, orthogonal comparative experiments were conducted based on typical field conditions for verification. Multiple sets of comparative experiments verified that: when α is below 0.05, the algorithm converges too slowly and lags in response to changes in conditions; when α is above 0.2, the algorithm is overly sensitive to sampling noise, and frequent jumps in the PI parameter cause turbidity oscillations. When γ is below 0.8, the algorithm focuses excessively on immediate reward values, and frequent speed adjustments of the centrifugal desander lead to excessive turbidity fluctuations; when γ is above 0.95, the algorithm's response to changes in conditions is severely lagging, and it cannot quickly adapt to field conditions. Within this parameter range, the requirements for algorithm convergence speed, turbidity control accuracy, and system operational stability can be simultaneously met. α=0.1 and γ=0.9 are the optimal values ​​under typical conditions, balancing convergence speed and long-term stability, achieving convergence within 2100 iterations and a turbidity control accuracy ≥98%.

[0070] Fifth, the rules for on-site adaptive adjustments are clearly defined. Those skilled in the art can make adaptive adjustments within the above range according to the on-site working conditions: when the working conditions are stable and the sand content fluctuates little, α can be decreased and γ can be increased to prioritize steady-state accuracy; when the working conditions are complex and the fluctuation range is large, α can be increased and γ can be decreased to prioritize adaptability to the working conditions, and parameter debugging can be completed without creative effort.

[0071] After each Q-value iteration update, the turbidity accuracy score and comprehensive score corresponding to the current parameter are calculated simultaneously for subsequent selection of the optimal PI parameter. The expression for the turbidity accuracy score is:

[0072] And calculate the comprehensive scoring function F, the expression of which is:

[0073] This formula, based on multi-objective optimization theory, weights and integrates turbidity accuracy and sand removal efficiency to quantitatively evaluate control effectiveness. For turbidity accuracy weight, The weight for sand removal efficiency is used to balance the control objectives of turbidity accuracy and sand removal efficiency, and both must satisfy... + =1. In this embodiment, we take... =0.8, =0.2, the design basis and determination method of this weighting ratio are as follows: Based on the core invention objective of the fine sampling mode, this mode takes high-precision water sample sampling for laboratory testing as its core objective. Turbidity accuracy is a veto item for water sample qualification, and sand removal efficiency is a constraint item to meet the equipment operation and basic water sample quality. Therefore, the weight is absolutely tilted towards turbidity accuracy. Based on orthogonal comparative experiments of typical field pumping and sampling conditions, it was determined that, through comparative experiments with multiple weight ratios such as 0.9 / 0.1, 0.8 / 0.2, and 0.7 / 0.3, this ratio can simultaneously meet the core requirement of turbidity control accuracy ≥98%, while ensuring the fastest convergence speed of the Q-Learning algorithm and the best control stability under operating condition fluctuations, without the problems of control oscillation or iteration non-convergence. Completely matching the design logic of the comprehensive scoring function, this weighting ratio can prioritize ensuring the accuracy of turbidity control based on the entry threshold of sand removal efficiency ≥95%, while effectively constraining the turbidity disturbance caused by excessively increasing sand removal efficiency, thus achieving the control objective of prioritizing accuracy while taking efficiency into account.

[0074] Those skilled in the art can, based on the actual analytical accuracy requirements and on-site working conditions, [make adjustments / adjustments]. =0.7~0.9、 The weighting ratio is adaptively adjusted within the range of 0.1 to 0.3.

[0075] The iterative process continues to execute eigenvalue calculation, action selection, parameter optimization, and Q-table update steps until any of the following preset termination conditions are met: Sampling turbidity accuracy And sand removal efficiency It remains stable for more than 50 sampling cycles. The change in the Q-value function over 100 consecutive iterations is less than The algorithm has fully converged. The number of iterations reached the preset maximum of 5000 iterations.

[0076] The design basis and determination method of the above termination conditions are as follows: First, it matches the core invention objectives of industry standards and fine sampling models. A turbidity accuracy threshold of 98% is stricter than the turbidity requirements for water samples in the "Technical Specification for Groundwater Environmental Monitoring" (HJ164-2020), ensuring the representativeness of the water samples. A sand removal efficiency threshold of 95% meets the general engineering qualification standards for hydrogeological exploration pumping operations, balancing equipment protection and basic water sample quality. The steady-state determination rule for 50 sampling cycles eliminates interference from sampling noise and instantaneous fluctuations in operating conditions, ensuring the system achieves true steady-state compliance. Second, it meets the professional theoretical requirements for the convergence of reinforcement learning algorithms. The change in Q-value over 100 consecutive iterations is less than... This is a universal criterion for the complete convergence of the Q-Learning algorithm, ensuring that the algorithm iterates to the globally optimal control strategy and avoiding insufficient control accuracy due to premature convergence. Third, it is fully adaptable to field operation scenarios and hardware characteristics. The maximum number of iterations of 5000 matches the typical operation time of 5-10 minutes for a single pumping and sampling in geological exploration, while also adapting to the computing power capacity of industrial MCUs. This serves as a fallback termination condition, preventing the algorithm from iterating indefinitely under extreme conditions and ensuring the safe and controllable operation of the system.

[0077] After reaching the termination state, stop the iteration, and select the parameter combination with the largest comprehensive score F as the optimal PI coefficient. The optimal PI coefficient satisfies:

[0078] The final PI parameter output is calculated based on the optimal PI parameter combination, expressed as follows:

[0079] in, This is the PI parameter adjustment value, used to dynamically adjust the four parameters of the dual-loop PI controller in fine sampling mode; The curly braces contain four independent PI parameter update formulas: This is the proportionality coefficient after the turbidity outer ring is updated. This represents the current proportionality coefficient of the outer ring turbidity. This is the adjustment amount for the turbidity outer ring proportional coefficient; These are the integral coefficients after the turbidity outer ring update. The current integral coefficient of the outer ring of turbidity. This is the adjustment amount for the turbidity outer ring integral coefficient; This is the proportional coefficient after the inner loop of rotational speed is updated. This represents the current proportional coefficient of the inner speed loop. This is the adjustment amount of the proportional coefficient within the speed inner loop; The integral coefficient after the inner loop speed change is updated. This represents the current integral coefficient of the inner speed loop. This is the adjustment amount of the integral coefficient of the inner loop of the rotational speed; This parameter combination is then applied to a dual-loop PI controller to achieve dynamic adjustment of PI parameters in fine-sampling mode; a Q-table is used to store the PI parameter combinations corresponding to each state-action pair, expressed as:

[0080] The rules for determining the amplitude of operating condition fluctuations are as follows: The core indicators are the rate of change in turbidity deviation and the rate of change in influent pressure over three consecutive sampling periods. If either indicator meets any of the following conditions, the operating condition is considered to be fluctuating drastically: The turbidity deviation change rate exceeds ±20% / s; The inlet water pressure variation exceeds ±30% of the rated inlet water pressure.

[0081] The parameters of the above-mentioned judgment rules are designed to strictly conform to the actual operating environment and system hardware characteristics of field pumping and sampling: the judgment window of three consecutive sampling cycles can effectively eliminate the interference of sampling noise and instantaneous operating condition fluctuations, ensuring the accuracy of operating condition judgment; the turbidity deviation change rate threshold of ±20% / s is four times the normal operating condition turbidity fluctuation range, which can accurately identify extreme operating conditions such as aquifer abrupt changes and large-scale sand ingress; the influent pressure change amplitude threshold of ±30% matches the safe operating pressure range of electric regulating valves and centrifugal desanders, avoiding equipment damage caused by excessive pressure. When the operating condition is judged to be drastic, the control unit pauses parameter iteration optimization and locks the current optimal PI parameter. After the operating condition indicators of five consecutive sampling cycles recover to the normal range, the parameter self-tuning process is restarted. The recovery judgment rule of five consecutive sampling cycles can completely eliminate the interference of instantaneous rebound of operating conditions, ensuring that parameter optimization is started only after the system enters a stable state, avoiding control model divergence and parameter jumps caused by drastic operating condition fluctuations, and ensuring the stability and reliability of algorithm operation.

[0082] Throughout the system operation and algorithm iteration process, real-time monitoring and anomaly protection functions are executed synchronously: When the liquid level in the gravity sedimentation unit exceeds the preset threshold, the liquid level protection is triggered. The control unit shuts down the water inlet component and starts the sewage discharge procedure. Once the liquid level returns to the safe range, the unit automatically resumes operation to protect the equipment.

[0083] After a single pumping and sampling operation is completed, the system automatically starts the backwashing and sewage discharge process. The accumulated sand is discharged through the sewage discharge module, and then the gravity sedimentation unit and centrifugal sand remover are backwashed with high pressure clean water through the backwashing module to remove residual sand in the flow channels and components. After the rinsing is completed, all actuators are shut down and the system enters standby mode.

[0084] The technical solution described in this embodiment, for operating environments with drastic fluctuations in groundwater pressure and sand content, achieves the following technical improvements compared to traditional fixed-parameter single-mode sand removal control technology: In high-efficiency sand removal mode, the system can maintain a stable high sand removal efficiency, and the flow control accuracy is greatly improved, effectively avoiding control oscillations caused by fluctuations in operating conditions; In the fine sampling mode, the accuracy and stability of turbidity control are significantly improved, which can meet the sampling requirements of high-precision water samples and greatly enhance the adaptability of working conditions. The Q-Learning parameter self-tuning algorithm achieves faster convergence speed and higher operating efficiency through pre-training and experience replay mechanisms, effectively reducing invalid exploration and improving the algorithm's adaptability and reliability under complex working conditions.

[0085] This embodiment describes the control method and hardware compatibility: 1. Input signals: turbidity sensor (4-20mA), valve position sensor (0-100%); 2. Output signals: Electric regulating valve control signal (0~10V), centrifugal sand remover speed signal (0~50Hz); 3. Communication method: Direct acquisition of analog signals, without additional communication protocols; 4. Interface compatibility: Fully compatible with gravity sedimentation units, electric regulating valves, and centrifugal sand remover hardware interfaces.

[0086] The real-time feedback-based pumping and sand removal device described in this embodiment includes a pretreatment module, a pattern matching module, a dual-loop control module, and a parameter self-tuning module. These modules work together to execute the aforementioned control method. The real-time feedback-based pumping and sand removal device also includes a memory and a processor that are communicatively connected. The memory stores computer instructions, and the processor executes these instructions to implement all steps of the control method described in this embodiment. The computer-readable storage medium described in this embodiment stores computer instructions, which, when executed, cause the computer to implement the control method described in this embodiment. The computer program product described in this embodiment includes computer instructions, which, when executed, cause the computer to implement the control method described in this embodiment.

[0087] Example 3 This embodiment also provides a pumping, sampling, and sand removal control device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. The term "module" as used below refers to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0088] This embodiment provides a water pumping, sampling, and sand removal control device, such as... Figure 5 As shown, it includes: Module 501, pretreatment module, is used to pass groundwater into the gravity sedimentation unit, and complete the coarse sand removal pretreatment by removing large particles of sand and gravel through gravity sedimentation. Module 502, the pattern matching module, is used to select the working mode and set the control target in response to the operation requirements. For large water intake, it selects the high-efficiency sand removal mode and sets the target pumping flow rate. For laboratory sampling, it selects the fine sampling mode and sets the target effluent turbidity. Module 503, the dual-loop control module, is used to execute the corresponding dual-loop PI control according to the selected mode. It outputs valve opening control signals for the high-efficiency sand removal mode and outputs centrifugal sand remover speed control signals for the fine sampling mode. Module 504, parameter self-tuning module, is used in system operation to iteratively optimize PI control parameters based on real-time operating data through a preset reinforcement learning parameter self-tuning model, and output the optimal control parameters adapted to the current operating conditions to the dual-loop control module. The model is configured with a multi-dimensional reward and punishment function adapted to the pumping, sampling and sand removal scenario. Module 505, the operating condition monitoring and protection module, is used to monitor the fluctuation range of operating conditions in real time. The turbidity deviation change rate and inlet water pressure change rate of multiple consecutive system sampling cycles are used as the judgment indicators. When the fluctuation exceeds the preset threshold, the parameter iteration optimization is paused and the current optimal control parameters are locked. The self-tuning process is resumed after the operating conditions stabilize. At the same time, it realizes the overflow protection of the gravity sedimentation unit, backwashing and sand discharge control after the operation.

[0089] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0090] Example 4 In this embodiment, a water pumping, sampling, and sand removal control device is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above-mentioned functions.

[0091] This invention also provides a computer device having the above-described features. Figure 6 The image shows a water pumping, sampling, and sand removal control device.

[0092] Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 6 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 6 Take a processor 10 as an example.

[0093] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0094] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0095] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0096] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0097] The computer device also includes an input device 30 and an output device 40. The processor 10, memory 20, input device 30, and output device 40 can be connected via a bus or other means. Figure 6 Taking the example of a connection between China and Israel via a bus.

[0098] Input device 30 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the computer device, such as a touchscreen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 40 may include display devices, auxiliary lighting devices (e.g., LEDs), and haptic feedback devices (e.g., vibration motors). The aforementioned display devices include, but are not limited to, liquid crystal displays, light-emitting diodes, displays, and plasma displays. In some alternative embodiments, the display device may be a touchscreen.

[0099] Example 5 This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0100] Example 6 A portion of this application can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to this application through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0101] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for pumping water for sand removal, characterized in that, The method, applied to a pumping and sand removal system, includes: Groundwater is fed into a gravity sedimentation unit, where large particles of sand and gravel are removed by gravity sedimentation to complete the coarse sand removal pretreatment. The working mode is selected and the control target is set in response to the operation requirements. For large water intake, the high-efficiency sand removal mode is selected and the target pumping flow rate is set. For laboratory sampling, the fine sampling mode is selected and the target effluent turbidity is set. Based on the requirements of water pumping and sampling operations, switch and execute dual-loop PI control according to the selected mode: the high-efficiency sand removal mode adopts dual-loop PI control of flow rate and valve opening, and the fine sampling mode adopts dual-loop PI control of turbidity and centrifugal sand remover speed. During system operation, based on real-time operating data, the PI control parameters are iteratively optimized through a self-tuning model with preset reinforcement learning parameters, and the optimal control parameters adapted to the current operating conditions are output. The reinforcement learning parameter self-tuning model is pre-set with a multi-dimensional reward and punishment function adapted to the pumping sampling and sand removal scenario. The multi-dimensional reward and punishment function includes at least three reward items: turbidity accuracy, sand removal efficiency, and operational stability. During system operation, the fluctuation range of operating conditions is monitored in real time. When the fluctuation exceeds the preset threshold, the parameter iteration optimization is paused and the current optimal control parameters are locked. The self-tuning process is resumed after the operating conditions stabilize.

2. The method according to claim 1, characterized in that, The execution logic of the dual-loop PI control is as follows: For the high-efficiency sand removal mode, the target pumping flow rate is used as the outer loop setpoint, and the real-time collected water pipeline flow rate is used as the outer loop feedback value to calculate and output the target valve opening value. Then, the target valve opening value is used as the inner loop setpoint, and the real-time valve opening feedback is used as the inner loop feedback value to output the control signal to adjust the valve opening. For the fine sampling mode, the target effluent turbidity is used as the outer loop setpoint, and the real-time collected effluent turbidity is used as the outer loop feedback value to calculate and output the target centrifugal sand remover speed. Then, the target centrifugal sand remover speed is used as the inner loop setpoint, and the real-time centrifugal sand remover speed feedback is used as the inner loop feedback value to output the control signal to adjust the centrifugal sand remover speed.

3. The method according to claim 1, characterized in that, The reinforcement learning parameter self-tuning model is a Q-Learning model customized for the pumping sampling and sand removal scenario. It has a pre-set state space, action space, and multi-dimensional reward and punishment function that matches the scenario. The state space includes turbidity deviation, speed deviation, and sand removal efficiency related features. The action space includes the PI parameter adjustment of the dual-loop PI control. The multi-dimensional reward and punishment function includes three types of reward items: turbidity accuracy, sand removal efficiency, and operational stability.

4. The method according to claim 3, characterized in that, Before the system is officially put into operation, the model is pre-trained through square wave step test, and the initial value of Q-table is calculated based on the step response accuracy. During the parameter self-tuning process, the ε-greedy strategy that decreases with the running time is used to complete the action selection, and the iterative convergence termination condition is preset. The iteration is terminated and the optimal PI control parameters are locked when any iterative convergence termination condition is met.

5. The method according to claim 1, characterized in that, The rule for determining the fluctuation range of the operating conditions is as follows: the turbidity deviation change rate and the inlet pressure change rate of multiple consecutive system sampling cycles are used as the determination indicators. When any indicator exceeds the preset threshold, it is determined that the fluctuation range of the operating conditions exceeds the limit.

6. The method according to any one of claims 1 to 5, characterized in that, During the coarse sand removal pretreatment process, the liquid level data of the gravity sedimentation unit is collected in real time. When the liquid level reaches the preset threshold, the raw water inlet equipment is automatically shut down and the overflow protection is triggered. After a single operation is completed, the system automatically starts the backwashing module and the sewage discharge module to complete the backwashing and sewage discharge operations of the gravity sedimentation unit, the flow channel of the centrifugal sand remover and the sand accumulation bin, respectively.

7. A water pumping, sampling, and sand removal device, characterized in that, The device includes: The pretreatment module is used to pass groundwater into the gravity sedimentation unit, where large particles of sand and gravel are removed by gravity sedimentation to complete the coarse sand removal pretreatment. The pattern matching module is used to select the working mode and set the control target in response to the operation requirements. For large water intake, the high-efficiency sand removal mode is selected and the target pumping flow rate is set. For laboratory sampling, the fine sampling mode is selected and the target effluent turbidity is set. The dual-loop control module is used to execute the corresponding dual-loop PI control according to the selected mode. It outputs valve opening control signals for the high-efficiency sand removal mode and outputs centrifugal sand remover speed control signals for the fine sampling mode. The parameter self-tuning module is used during system operation to iteratively optimize PI control parameters based on real-time operating data through a preset reinforcement learning parameter self-tuning model, and output the optimal control parameters adapted to the current operating conditions to the dual-loop control module. The model is configured with a multi-dimensional reward and punishment function adapted to the pumping, sampling and sand removal scenario. The operating condition monitoring and protection module is used to monitor the fluctuation range of the operating condition in real time. When the fluctuation exceeds the preset threshold, the parameter iteration optimization is paused and the current optimal control parameters are locked. The self-tuning process is resumed after the operating condition stabilizes.

8. A water pumping, sampling, and sand removal device, characterized in that, The device is a computer-controlled device, including a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the pumping, sampling and sand removal method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the pumping, sampling and sand removal method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, Includes computer instructions for causing a computer to perform the pumping, sampling, and sand removal method according to any one of claims 1 to 6.

Images