LVR evaporation system control method and system based on intelligent algorithm

Abstract

The invention provides an LVR evaporation system control method and system based on an intelligent algorithm, and relates to the field of low-vacuum reboiler control. According to the method, temperature sensors, vacuum pressure sensors, conductivity sensors, turbidity sensors and liquid level sensors are arranged on an evaporator, a circulating pipeline and a vacuum system, and a scaling risk state sensing model is constructed; calculating a supersaturation index SI based on the acquired signal and soft measurement, and judging and identifying a metastable or critical deposition working condition through a threshold value; performing risk level classification on the multi-dimensional state vector by using a reinforcement learning algorithm (PPO), and outputting risk confidence and a trigger signal; after the risk is triggered, bypass induced nucleation is realized through PID shunt control, and the shunt proportion, the bypass temperature difference and the stirring intensity are accurately adjusted; meanwhile, a strategy optimization module is introduced, and self-adaptive updating of a control strategy is achieved through a return function and an online training mechanism; and finally, long-period stable operation and low scaling rate of the evaporation system are realized through hysteresis control and an energy consumption self-optimization mechanism.

Description

Technical Field

[0001] This invention relates to the field of low vacuum reboiler control, and more particularly to a control method and system for an LVR evaporation system based on intelligent algorithms. Background Technology

[0002] LVR (Low Vacuum Reboiler) evaporation systems are a type of energy-saving evaporation device commonly used for the treatment and resource recovery of high-concentration, high-salinity wastewater. These devices significantly reduce boiling point temperature by maintaining a low vacuum environment, thereby reducing energy consumption, and are widely used in chemical, pharmaceutical, metallurgical, environmental protection, and zero-discharge industrial wastewater industries. However, during long-term continuous operation, scaling and crystal nucleation deposits easily occur on the heat exchange surface of the evaporator, leading to decreased heat exchange efficiency, increased energy consumption, and even equipment shutdown and safety hazards.

[0003] Current methods for controlling evaporative scaling in LVRs mainly rely on the following: Firstly, the scale layer can be removed by periodically shutting down the machine for cleaning, but this method will cause system interruption and high maintenance costs; Secondly, methods such as constant flow induced crystallization or online mechanical flushing are used to reduce scaling, but due to the lack of real-time operating condition identification and dynamic control capabilities, problems of "over-control" or "control lag" often occur. Third, some systems introduce PID feedback regulation, but PID is difficult to cope with the nonlinear and multivariate coupling characteristics of the evaporation process, and is prone to control deviation and efficiency loss under high salt conditions. Summary of the Invention

[0004] Purpose of the invention: Scaling formation has distinct induction and growth phases. Therefore, this invention proposes a control method for LVR evaporation systems based on intelligent algorithms, and a control system to implement this method. This system enables precise early warning and proactive intervention during the induction phase, thereby improving the system's operating efficiency and continuity.

[0005] In a first aspect, this invention proposes a control method for an LVR evaporation system based on an intelligent algorithm, comprising the following steps: S1. Collect data on the temperature, vacuum, conductivity, turbidity, and liquid level of the evaporator feed liquid, and calculate scaling risk indicators, including supersaturation index, scaling induction risk index, and distribution of local oversaturated areas on the wall. S2. Construct a reinforcement learning agent, input the scaling risk index as a state vector into the reinforcement learning agent, and identify three risk levels: normal operating conditions, metastability-induced operating conditions, and critical deposition operating conditions. S3. When the metastable induction condition or critical deposition condition is met, the bypass nucleation diversion regulation is triggered, and the opening of the feed liquid diversion valve and the return valve are automatically adjusted to introduce part of the feed liquid with a saturation index higher than the predetermined value into the bypass nucleation unit for controllable induced crystallization. S4. Apply a negative pressure wave field during the controllable induced crystallization process. Through the frequency conversion adjustment of the vacuum pump and the linkage with the pressure wave function generation module, generate a negative pressure wave signal with controlled amplitude and frequency to achieve dynamic stripping of the initial crystal nucleus deposition. S5. Based on the reinforcement learning strategy, the bypass diversion and pressure wave coupling parameters are optimized. The bypass diversion ratio, pressure wave amplitude and frequency are iteratively optimized online to achieve the optimal balance between the descaling effect and the system evaporation efficiency.

[0006] As a preferred embodiment, step S1 specifically includes: Temperature, pressure, conductivity, turbidity, and liquid level sensors are installed in the evaporator body, circulation pipeline, and vacuum system of the LVR evaporation system to achieve real-time acquisition of the liquid state, vacuum degree, solution conductivity, liquid level, and suspended particle concentration. Multi-source sensor signals are synchronized and outliers are removed via a data acquisition module to construct a time-consistent and noise-suppressed running feature sequence. The collected data were inverted online using thermodynamic and solubility models to obtain the corresponding supersaturation index, scaling-induced risk index, and local oversaturation distribution matrix of the wall surface.

[0007] As a preferred embodiment, step S2 specifically includes: Based on the PPO architecture, a reinforcement learning agent is constructed. The maximum value of the multi-source sensor signal, the oversaturation index, the scaling-induced risk index, and the local oversaturation distribution matrix of the wall is used to construct a state vector. This state vector is input into the reinforcement learning agent, which calls the trained risk classification policy network to classify and infer the scaling risk of the system in real time. The output probability of the policy network is used to distinguish three risk levels: normal operating condition, metastable induced operating condition, and critical deposition operating condition. When the risk level is determined to be metastable induced or critical deposition, a risk trigger signal is sent to the execution control module. The risk status change trend is tracked by a rolling time window algorithm to determine whether the diversion and induction trigger conditions are met, and the risk status information is synchronously stored in the database.

[0008] As a preferred embodiment, step S3 specifically includes: After receiving a risk trigger signal, the reinforcement learning agent outputs the corresponding diversion control action according to the current risk level. The opening of the bypass diversion valve and the return valve is automatically adjusted to introduce part of the highly supersaturated liquid into the bypass nucleation unit; the temperature and flow rate are controlled in the bypass nucleation unit so that the supersaturated material preferentially forms crystal nuclei in the bypass, thereby reducing the actual scaling driving force near the main heat exchange surface. A portion of the crystal slurry or clear liquid is refluxed back to the main evaporator at a set ratio to form a controlled solute concentration gradient; the split ratio, bypass temperature, and reflux flow rate parameters are recorded for subsequent strategy optimization.

[0009] As a preferred embodiment, step S4 specifically includes: Start the pressure wave function generation module and set the pressure wave amplitude and frequency according to the control signal of the reinforcement learning agent; By coupling the frequency conversion control of the vacuum pump with the pressure wave function generation module, periodic negative pressure fluctuations are formed and act on the evaporation kettle body; Tangential disturbances and periodic shear forces are generated in the boundary layer region of the wall, causing the initial crystal nuclei to be dynamically stripped away; The system continuously monitors turbidity changes and scaling trends under pressure wave intervention and provides real-time feedback to the reinforcement learning agent. After the stripping effect reaches the preset stable range, the waveform control parameters are kept in steady-state output mode.

[0010] As a preferred embodiment, step S5 specifically includes: The reinforcement learning agent iterates and reinforces its learning based on historical operational data and real-time feedback data, according to state, action, and reward mechanisms. The bypass diversion ratio, pressure wave amplitude and frequency control parameters are dynamically adjusted to form a multi-dimensional control strategy. The comprehensive return value is calculated in each strategy update cycle, including the descaling effect return and the evaporation efficiency return. The policy network parameters are updated using gradients based on the reward value, causing the system to tend toward the optimal control combination; when the reward reaches the preset convergence threshold, a stable policy is output for real-time operation.

[0011] As a preferred embodiment, step S6 specifically includes: The reinforcement learning agent invokes the convergence strategy during actual operation to perform real-time closed-loop adjustment of diversion and pressure wave intervention. During the continuous operation of the evaporation system, risk status monitoring and strategy optimization are carried out simultaneously to form a dynamic adaptive control closed loop. When the risk level is detected to have fallen back to the normal range, the diversion ratio and pressure wave amplitude are automatically reduced to minimize energy consumption. When a risk condition is detected to be close to the threshold, the intervention intensity is automatically increased to suppress the risk of scaling in advance; the evaporation system is kept in a low deposition state over a long period of time, the cleaning interval is significantly extended, and the low temperature and high concentration evaporation conditions are stably maintained.

[0012] Furthermore, this invention also proposes a low-vacuum reboiler control system, which includes: The data acquisition module is used to install temperature sensors, vacuum pressure sensors, conductivity sensors, turbidity meters and level gauges on the evaporator body, circulation pipeline and vacuum system of the LVR evaporation system, to acquire multi-source operating condition signals at a preset sampling frequency, and to perform timestamp synchronization, anomaly removal, smoothing filtering and standardization processing on the signals. The soft measurement and risk assessment module is used to calculate the supersaturation index of the solution based on the processed signal, and combine it with the threshold to determine whether the evaporation system is in a normal, metastable or critical deposition state, thereby identifying the risk of scaling. The reinforcement learning strategy decision-making module receives the state vector, uses the reinforcement learning strategy to classify the risk state, and outputs the risk level, confidence level and trigger signal. The bypass nucleation and diversion control module is used to control the diversion valve, bypass temperature, reflux flow rate and stirring speed under the trigger signal. It achieves solution concentration regulation and local supersaturation reduction in the main heat exchange zone through bypass-induced nucleation. The pressure wave disturbance and descaling module is used to drive the vacuum pump, proportional valve and waveform controller to apply negative pressure wave to disturb and remove the crystal nucleus layer deposited on the inner wall of the evaporator, and to achieve adaptive adjustment of the disturbance intensity through risk judgment signal. The strategy optimization and adaptive control module is used to collect state, action and feedback information generated during the control process, and to periodically update and optimize the reinforcement learning strategy according to the preset reward mechanism. The closed-loop maintenance and energy consumption optimization module is used to dynamically adjust the diversion ratio, stirring intensity and pressure wave amplitude according to the solution supersaturation and its changing trend, so as to suppress scaling risk and achieve long-term stable operation, and maintain evaporation efficiency, heat exchange efficiency and energy consumption within the set range.

[0013] Compared with the prior art, the present invention has at least the following beneficial effects: (1) This invention collects signals such as temperature, vacuum pressure, conductivity, turbidity and liquid level through a multi-source sensor array, and combines it with a soft measurement model for scaling risk to identify the supersaturated state in real time. It can accurately identify metastable or critical deposition state before scaling forms on a large scale, triggering intervention in advance. This avoids the lag problem that traditional evaporation systems can only respond passively, and realizes the transformation from "post-event handling" to "pre-event prevention" control mode.

[0014] (2) This invention utilizes intelligent diversion control and bypass-induced nucleation technology to actively introduce highly supersaturated solution into the bypass induction zone, thereby weakening the local concentration difference and deposition driving force within the main evaporator. Through this active induction, reflux, and regulation mechanism, the crystal nucleus attachment rate in the main heat exchange zone can be significantly reduced, effectively extending the cleaning cycle and reducing maintenance costs.

[0015] (3) This invention uses a variable frequency vacuum pump, a proportional valve, and a waveform controller to form a negative pressure wave field, and adaptively adjusts the disturbance intensity according to the risk level to achieve efficient stripping of the initial crystal nuclei on the wall surface. This method can maintain a low adhesion state on the wall surface without reducing the evaporation efficiency, ensuring long-term stable operation of the system, while avoiding frequent manual intervention.

[0016] (4) This invention utilizes reinforcement learning algorithms to train and update strategies online for multiple variables such as risk state, diversion ratio, disturbance amplitude, and duration, possessing autonomous learning, self-correction, and adaptive adjustment capabilities. Compared with traditional fixed logic control, this method exhibits stronger robustness and control accuracy under nonlinear, strongly coupled, and high-salt-concentration operating conditions.

[0017] (5) The present invention dynamically coordinates the diversion induction and disturbance intensity throughout the process, so that the system can reduce energy consumption fluctuations while ensuring high evaporation rate and high heat exchange efficiency, thus achieving the dual goals of energy saving and stability. The system can still maintain high evaporation efficiency and high heat exchange efficiency under high salt and high concentration conditions, significantly reduce the scaling rate, and greatly extend the operating cycle, which has good industrial application value and promotion potential. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the control method for the LVR evaporation system based on intelligent algorithms in this embodiment.

[0019] Figure 2 This is a schematic diagram of the low vacuum reboiler control system in the embodiment. Detailed Implementation

[0020] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention can be practiced without one or more of these details. In other instances, certain technical features well-known in the art have not been described in order to avoid obscuring the invention.

[0021] Example 1 like Figure 1 As shown, this embodiment provides a control method for an LVR evaporation system based on intelligent algorithms, including the following steps: S1. By setting up a heat transfer surface temperature sensor, a vacuum pressure sensor, an online conductivity sensor, a turbidity meter, and a liquid level sensor, real-time data such as the evaporator feed liquid temperature, vacuum degree, conductivity, turbidity, and feed liquid height are collected, and a soft measurement model for scaling risk is established based on the state data; the soft measurement model for scaling risk is used to calculate the supersaturation index, the scaling-induced risk index, and the distribution of local oversaturated areas on the wall. S2. Based on reinforcement learning algorithm, the scaling risk status is classified and early warning is given. The scaling risk index output by the scaling risk soft measurement model is used as the state vector input to the reinforcement learning agent. Through training on historical operating data, three risk levels are identified: normal operating condition, metastable induced operating condition and critical deposition operating condition, so as to realize the early judgment of wall deposition risk. S3. When the scaling risk reaches the preset threshold, the bypass nucleation diversion regulation is triggered. Based on the output action of the reinforcement learning agent, the opening of the feed liquid diversion valve and the return valve are automatically adjusted to introduce part of the supersaturated feed liquid into the bypass nucleation unit for controllable induced crystallization, thereby reducing the actual scaling driving force at the main heat exchange surface and inhibiting the formation of initial deposits. S4. Apply a negative pressure wave field during the induced crystallization process. Through the frequency conversion adjustment of the vacuum pump and the linkage with the pressure wave function generation module, generate a negative pressure wave signal with controlled amplitude and frequency. Act on the evaporator body to enhance the tangential shear force and disturbance degree of the wall boundary layer, and realize the dynamic stripping of the initial crystal nucleus deposition. S5. Based on reinforcement learning strategy, optimize the bypass diversion and pressure wave coupling parameters. Utilize the state, action, and reward mechanism of reinforcement learning to iteratively optimize the bypass diversion ratio, pressure wave amplitude, and frequency online, so as to achieve the optimal balance between descaling effect and system evaporation efficiency, and ensure that the system maintains a low deposition state without affecting the evaporation rate. S6. To achieve adaptive scaling inhibition and long-cycle operation of the system, the closed-loop control of the reinforcement learning agent enables the evaporation system to adaptively predict scaling risk, induce diversion and adjust pressure wave stripping in linkage during long-term continuous operation, thereby significantly reducing the wall deposition rate, extending the cleaning cycle and stably maintaining the low temperature and high concentration evaporation conditions. S1. By setting up a heat transfer surface temperature sensor, a vacuum pressure sensor, an online conductivity sensor, a turbidity meter, and a liquid level sensor, real-time data such as the evaporator feed liquid temperature, vacuum degree, conductivity, turbidity, and feed liquid height are collected. Based on this data, a soft measurement model for scaling risk is established. The soft measurement model for scaling risk is used to calculate the supersaturation index, the scaling-induced risk index, and the distribution of local oversaturated regions on the wall surface. Specifically, it includes the following sub-steps: S110. Temperature sensors (accuracy ±0.1℃, response time <1s), vacuum pressure sensors (accuracy ±0.5kPa), online conductivity sensors (range 0~100mS / cm, accuracy ±0.5%), turbidity meters (range 0~500NTU), and level gauges (range 0~2m) are installed on the evaporator body, circulation pipeline, and vacuum system of the LVR evaporation system. The sampling frequency of all sensors is set to 1–10Hz, and the sampling signals are uniformly connected to the data acquisition module via an industrial bus (such as Modbus or OPCUA).

[0022] S120: After the acquired multi-source signals are time-stamped and calibrated, outliers are removed using a ±3σ threshold. Then, denoising and stabilization are achieved through a 3-5 point moving average and a second-order low-pass filter. The filter cutoff frequency is set to 1Hz to ensure that dynamic fluctuations are captured while suppressing high-frequency interference. The preprocessed signals are then uniformly converted into a time-series structured data stream for use by the soft measurement module.

[0023] S130, the soft measurement module performs online calculations every 5 seconds, based on thermodynamics and solubility models, to solve the supersaturation index (SI) of the liquid in real time:

[0024] The IAP (Ion Activity Product) is calculated based on conductivity and the chemical composition of the feed solution, using an empirical conversion formula:

[0025] SI is the supersaturation index, which indicates the degree of deviation between the solution and the thermodynamic equilibrium state of the scaling substance. It is a key parameter for judging whether there is a risk of scaling in the system. IAP is the ion activity product, which represents the product of the actual activities of the ions in the solution that are related to the formation of the target scaling crystals. It reflects the actual state of the solution. It represents the product of ion activities when a solution reaches thermodynamic saturation equilibrium at a specific temperature T. Conductivity is a parameter characterizing the electrical conductivity of a solution, and it is directly related to the ion concentration and ion mobility in the solution. During evaporation and concentration, conductivity increases as the feed solution is concentrated, thus it can be used as an online monitoring indicator of the overall ionic strength of the solution. This refers to the concentration of specific ions in the solution associated with the formation of scale crystals. For example, in a calcium carbonate system, it is the concentration of calcium ions and carbonate ions, and in a calcium sulfate system, it is the concentration of calcium ions and sulfate ions. It is usually obtained through conductivity conversion, ion-selective electrode measurement, or chemical analysis. When the target substance is an inorganic salt solution, the activity coefficient can be further introduced. Correction of ion strength error.

[0026] The IAP (ion activity product) in the aforementioned formula can also be calculated based on a soft-sensor model. The system uses real-time collected conductivity data. With respect to temperature T, a polynomial fitting formula is used:

[0027] The ion product at the current moment is calculated, and the coefficients are experimentally calibrated; then, through: Obtain the basic supersaturation.

[0028] Constructing a model for the distribution of local oversaturated regions on the wall: To reflect the spatial differences in scaling risk, the system does not rely on a high-cost internal dense sensor array, but instead uses boundary sensor data from the inlet, outlet and circulation pipelines, combined with a pre-set fluid thermodynamic model, to calculate the temperature field T(x,y,z) at different spatial nodes (x,y,z) inside the evaporator through interpolation and inversion.

[0029] The system divides the evaporator wall into several virtual monitoring grids, calculates the local supersaturation SI(x,y,z) at each grid point, and forms a distribution matrix. Based on this, the system further calculates the gradient rate of change in the wall's normal direction. (n is the wall normal). When Furthermore, when the value exceeds the set threshold, it indicates that the local area has a significant kinetic tendency to deposit crystal nuclei onto the wall, and the system marks it as a "locally oversaturated high-risk area".

[0030] Calculating the scaling-induced risk index: The scaling-induced risk index is used to quantify how close the system is to the critical time for crystal nucleation explosion. Based on the classical nucleation induction period theory, the scaling-induced risk index is defined. as follows:

[0031] in is a dimensionless risk index (0~1); B is a thermodynamic constant related to the surface energy of the solute; Correction factor related to solution turbidity ( This formula shows that as the supersaturation (SI) increases or the turbidity (NTU) rises, the nucleation induction period shortens exponentially, and the risk index rises rapidly, thus providing a quantitative basis for early warning of the control system.

[0032] S140. Based on historical operating conditions and experimental calibration, the 95th percentile value is selected to determine the metastability threshold. With critical deposition threshold It allows for dynamic adjustment of ±0.05 to accommodate different batches of liquid; the threshold is stored in the threshold management module for the reinforcement learning agent to call during real-time inference.

[0033] S150, The standardized state vector The input is fed into the reinforcement learning agent, and to ensure that the agent can perceive a comprehensive scale risk, the state vector is constructed as follows:

[0034] in: These are process parameters acquired in real time by the sensor; SI is the supersaturation index of the bulk solution. The aforementioned scale-induced risk index is used to feed back the outbreak risk in the time dimension to the agent; The maximum value in the local oversaturation distribution matrix of the wall (or the extracted gradient feature value of the high-risk area) is used to feed back the local deposition risk of the spatial dimension to the agent.

[0035] The state input vector has a dimension of 8 and is fed into the policy network at a 5-second update frequency, serving as the basis for risk identification and action decision-making.

[0036] S2. Based on reinforcement learning algorithms, scale risk status classification and early warning are performed. The scale risk index output by the soft measurement model is used as a state vector input to the reinforcement learning agent. Through training on historical operating data, three risk levels are identified: normal operating conditions, metastable induced operating conditions, and critical deposition operating conditions, to achieve early judgment of wall deposition risk. Specifically, this includes the following sub-steps: S210, The standardized state vector The input is fed into a reinforcement learning agent built on a PPO (Proximal Policy Optimization) architecture. The policy network has an input dimension of 6, corresponding to 6 operational condition monitoring parameters, and an output dimension of 3, corresponding to: Normal operating conditions; Metastable induced condition; Critical deposition conditions.

[0037] The policy network updates every 5 seconds during inference, and uses the softmax function to map the output to a probability distribution, satisfying... .

[0038] S220: The reinforcement learning agent operates according to the maximum probability principle. Determine the current risk level and perform double verification by combining the SI index value range: If SI < 0.5 and If the value is at its maximum, it is considered a normal operating condition. If 0.5 ≤ SI < 1.0 and If the value is at its maximum, it is determined to be a metastable induced condition; If SI ≥ 1.0 and If the value is at its maximum, it is determined to be a critical deposition condition.

[0039] The control signal is triggered only when the judgment result is consistent with the SI interval, so as to avoid false triggering due to short-term abnormalities.

[0040] S230. When the judgment result is metastable or critical, the reinforcement learning agent automatically sends a risk trigger signal to the execution control module. The signals include: Risk level Used to indicate the risk status of the current operating condition; corresponding to the safe zone, metastable zone, and critical deposition zone, respectively; Risk confidence level ; indicates the reliability of the model's risk assessment, calculated from the maximum value among the probability vectors of each level output by the reinforcement learning policy network, and is used to assist in judging the trigger strength and response priority; Trigger timestamp It records the precise time of the trigger signal generation for subsequent strategy tracing, action alignment, and hysteresis control correction.

[0041] The execution module selects the initial control strategy for diversion induction and pressure wave intervention based on the type of L.

[0042] S240. Simultaneously, the reinforcement learning agent uses a 10-minute sliding time window to track the changing trend of the SI index. ;when Furthermore, if the trend lasts for more than 30 seconds, an early warning event is triggered. Even if the region has not yet entered the metastable zone, the threshold or diversion preset value can be adjusted in advance to avoid risk delay.

[0043] S250. The state input vector, output probability distribution, risk judgment result, trigger signal, and trend warning record for each cycle are synchronously written into the running database. The database adopts a time-series storage structure with a recording granularity of 5 seconds. The data is used for: The policy network is updated regularly (e.g., every 2 hours) offline or online; it supports subsequent policy playback and retraining; it provides process data auditing and traceability; this module can be deployed in DCS systems or edge computing units, with computing latency controlled within 1 second to ensure real-time performance.

[0044] S3. When the scaling risk reaches a preset threshold, the bypass nucleation diversion regulation is triggered. Based on the output action of the reinforcement learning agent, the opening of the feed liquid diversion valve and the return valve are automatically adjusted to introduce part of the highly supersaturated feed liquid into the bypass nucleation unit for controllable induced crystallization, thereby reducing the actual scaling driving force at the main heat exchange surface and inhibiting the formation of initial deposits. Specifically, this includes the following sub-steps: S310, the reinforcement learning agent receives a risk trigger signal. Then, output the shunt control action vector:

[0045] in It is an action vector, representing the set of control actions output by the reinforcement learning policy network. It is the core variable used by the system to perform intervention operations after risk identification, and directly drives the diversion valve and pressure wave disturbance unit. It is the diversion ratio, which represents the percentage of the main fluid flow that is diverted to the bypass induced nucleation zone. It is a key adjustment parameter for controlling the local supersaturation of the solution, and its value range is dynamically adjusted according to the risk level L and the confidence level C. The temperature difference between the bypass induction zone and the main evaporation zone is used to control the nucleation rate and particle size distribution. By controlling the temperature difference, the crystal nucleation environment can be precisely controlled. This indicates the flow rate of the liquid refluxed from the bypass induction zone to the main evaporation zone after induced crystallization. A reasonable reflux flow rate can reduce local supersaturation while maintaining overall thermal balance. The stirring speed is indicated. This parameter directly determines the strength of the pressure wave descaling effect and is a key variable for achieving active removal of scale from the wall surface. The control signal is sent to the valve control unit and the actuator of the bypass nucleation unit through the PLC / DCS system. The total signal transmission and response delay is controlled within 1 second.

[0046] S320: Automatically adjusts the opening of the bypass diversion valve and the return valve to ensure the diversion ratio is correct. The flow rate is controlled within the range of 5% to 25% of the main circulation flow rate; the preferred range is 10% to 15%. The flow rate is controlled by a closed-loop feedback system based on the electromagnetic flowmeter, with allowable fluctuations not exceeding ±2%. When the system risk level is metastable induced, a low diversion ratio is used; when in a critical deposition state, the diversion ratio is automatically increased to the upper limit.

[0047] S330: In the bypass nucleation unit, the temperature difference is controlled by a temperature control system. The temperature is controlled at 2–8°C in the main evaporator, and the stirring speed is adjusted using a variable frequency stirrer. The flow rate is controlled at 100–300 rpm (equivalent to a liquid tangential flow rate of 0.3–0.8 m / s) to create a stable supersaturated induction environment, promoting preferential growth of crystal nuclei in the bypass. The rate of decrease in solution conductivity and the upward trend in turbidity are monitored as auxiliary criteria for the effectiveness of induction.

[0048] S340: When the crystal nuclei form within the set particle size range (e.g., average particle size 50–100 μm), the reflux valve automatically opens, diverting 50%–80% of the bypass slurry or clarified liquid at the set flow rate. The reflux is returned to the main evaporator inlet to maintain controlled solute concentration in the main heat exchange zone and reduce local supersaturation on the wall surface. The control accuracy of the reflux flow rate is set to ±1%.

[0049] S350: Records ξ, , and The execution parameters and induction results (particle size and turbidity change curves) are synchronously stored in the runtime database. This record is used for calculating the reinforcement learning reward function and subsequent policy iteration training; at the same time, the system retains at least 48 hours of historical operation trajectory to support model playback and diagnosis.

[0050] S4. During the induced crystallization process, a negative pressure wave field is applied. Through the frequency conversion adjustment of the vacuum pump and the linkage with the pressure wave function generation module, a negative pressure wave signal with controlled amplitude and frequency is generated. This signal acts on the evaporator vessel, enhancing the tangential shear force and disturbance degree of the boundary layer on the wall, and realizing the dynamic stripping of the initial crystal nucleus deposition. Specifically, this includes the following sub-steps: S410: When the risk assessment result is metastable induced or critical deposition state, the reinforcement learning agent calls the pressure wave function generation module and sets the pressure wave amplitude according to the policy output. ,frequency Duration and waveform type The preferred waveform type is a sine wave, and its expression is:

[0051] in This represents the actual pressure value inside the vacuum system or evaporator at time t. The system's baseline pressure value represents the stable pressure under undisturbed conditions, typically corresponding to a constant vacuum during evaporation; t represents time; and the pressure wave amplitude is... Controlled between 0.5 and 5 kPa, frequency Controlled within 0.1–5 Hz, duration Set to 60–300 seconds. The strategy network automatically selects the amplitude range and duration based on the risk level: low-amplitude, short-duration perturbations are selected for metastable induced states, while high-amplitude, long-duration perturbations are selected for critical deposition states.

[0052] S420: Achieves vacuum pump speed control through closed-loop coupling between the vacuum pump frequency converter and the pressure wave function generator module. correspond Precise adjustment is required. The response time of the vacuum system is set to within 0.5 seconds to ensure that the synchronization error between the actual output pressure wave and the set waveform does not exceed ±5%. To prevent excessive fluctuations from causing liquid splashing or demister impact, a pressure safety boundary is set. .

[0053] S430: Negative pressure wave action is applied to the evaporator body to enhance the disturbance intensity of the boundary layer between the fluid inside the vessel and the wall. This generates periodic tangential shear forces locally on the wall. It can be approximated as:

[0054] in It is the tangential shear force of the wall, which represents the periodic shear force caused by pressure wave disturbance at the interface between the wall boundary layer and the main fluid. It is a key parameter for determining whether the scale layer has been peeled off. It is the dynamic viscosity of the fluid, measured in Pa·s, used to describe the fluid's resistance to shear deformation. The higher the viscosity, the more "viscous" the fluid flow, and the easier it is to form a thick boundary layer; This is the velocity gradient, representing the rate of change of flow velocity perpendicular to the wall, i.e., the rate of change of velocity from the wall to the main fluid. Because the boundary layer is relatively thin, it can be used... Approximation; It is a velocity change, representing the amplitude of fluid velocity fluctuations caused by pressure wave disturbances. It consists of the pressure wave amplitude ΔP of the negative pressure wave and its frequency. Joint decision; It is the boundary layer thickness, which represents the thickness of the region near the wall where the velocity changes significantly, and is the effective range of the pressure wave disturbance on the fluid.

[0055] Estimated by CFD or empirical formulas, when Exceeding the critical stripping threshold At that time, the initial crystal nucleus deposition layer can be effectively stripped away.

[0056] S440: During the descaling process, the system monitors turbidity changes in real time. Changes in the SI index .when and If the scaling is deemed effective, the feedback result is returned to the reinforcement learning agent for reward function calculation and policy correction. If the scaling effect is not satisfactory, ΔP is automatically adjusted. Increase the intensity of the disturbance.

[0057] S450: When the descaling effect enters a stable range ( and Keep ≥ The pressure wave function generation module automatically switches to steady-state maintenance mode, with the frequency and amplitude gradually decreasing to the low-power range, achieving three-stage control: intervention, stabilization, and maintenance. Waveform parameters, response curves, and descaling feedback records during the disturbance process are simultaneously written to the database, facilitating strategy iteration and post-event diagnosis.

[0058] S5. Optimize the bypass diversion and pressure wave coupling parameters based on reinforcement learning strategy. Utilize the state, action, and reward mechanism of reinforcement learning to iteratively optimize the bypass diversion ratio, pressure wave amplitude, and frequency online, achieving the optimal balance between descaling effect and system evaporation efficiency. This ensures low-deposition operation without affecting evaporation. Specifically, this includes the following sub-steps: S510: The reinforcement learning agent continuously receives state input vectors during the operation of the evaporation system.

[0059] And output control actions according to the policy network.

[0060] The system has 9 state dimensions and 7 action dimensions. The policy network uses a PPO architecture, consisting of two fully connected neural networks with 64 neurons per layer and ReLU activation function. The inference cycle is set to 5 seconds, and the policy update cycle is set to 2 hours or after a preset number of sampling steps. It is triggered once per hour.

[0061] S520: The reward function for reinforcement learning is defined as:

[0062] in: The reward function value represents the overall reward value after the intelligent control policy executes its actions at time step t, and is used for policy updates and optimization in reinforcement learning algorithms. The larger the value, the better the current control strategy. (Fouling rate) represents the fouling rate at time t, reflecting the rate of crystal nucleus deposition and growth on the heat exchanger wall. A higher value indicates a more severe fouling problem. This is reflected in the reward function through... This encourages a reduction in the scaling rate; (Evaporation efficiency) represents the evaporation efficiency of the evaporator at time t, reflecting the effectiveness of heat energy utilization and evaporation output. The higher the value, the more efficient the system operation. (Energy consumption cost) represents the cost corresponding to the energy consumption of equipment such as vacuum pumps and circulation pumps within this time step. The higher the energy consumption, the larger this item will be, which will have a negative impact on the reward. (Action Penalty Item) indicates the additional burden or cost incurred by performing actions (such as diversion induction, pressure wave disturbance), including mechanical wear, power consumption, or response time delay caused by frequent valve operation. Penalty items can prevent over-adjustment of the control strategy.

[0063] , , , This represents the weight of each indicator in the overall reward function, used to balance the importance of different objectives. Each weight can be dynamically adjusted through offline training or online strategy iteration. This function is used to simultaneously balance three objectives: scaling inhibition, evaporation efficiency, and energy cost, avoiding single-objective optimization that could cause the system to deviate from its actual operational feasibility.

[0064] S530: Within each policy update cycle, the agent will use the most recent... A triplet of state, action, and reward Stored in an experience buffer, the PPO algorithm performs batch training on the sampled data, with a batch size of 256, iterating for 10 epochs, and a shearing ratio. By optimizing the objective function:

[0065] in The shearing loss function, representing the PPO algorithm, is the core objective function used to train the reinforcement learning policy network. This loss function improves the system's intelligent control and decision-making capabilities while ensuring stable policy updates. The parameter vector representing the current policy network is a learning variable that is continuously updated through gradient descent. This represents the expected value of the sampled trajectory at time step t (i.e., the average of the sampled data). The ratio of the probabilities of the new and old strategies, also known as the importance sampling ratio, is defined as:

[0066] in This is the current strategy. It's an old strategy.

[0067] (Dominance Function): Measures the current action The degree of superiority relative to the average action is a direct measure of the quality of intelligent control actions; This indicates that the action is superior to the average action, and the strategy tends to reinforce it. This indicates that the action is poor and the strategy tends to suppress it.

[0068] The policy probability ratio is pruned to prevent excessively large update steps from causing training instability. It is a hyperparameter (usually taken as 0.1 to 0.3).

[0069] S540: After training, monitor the average reward. Change To determine convergence, after 10 consecutive iterations... When convergence is considered, output the convergence strategy. If training fails to converge, continue iterating until the convergence criterion is met or the maximum number of iterations is reached. .

[0070] S550: Convergence Strategy After being solidified, it enters the online control phase, performing action reasoning and control signal issuance every 5 seconds, and triggering an incremental policy update every 2 hours. After each update, the average reward before and after the update, parameter weights, policy version number, and control execution records (including...) are recorded. (etc.) are synchronously written to the database. This data can be used for subsequent strategy replay, comparative analysis, or model retraining, ensuring the traceability of the algorithm evolution process and engineering stability.

[0071] S6. Achieve adaptive scaling inhibition and long-term operation of the system. Through closed-loop control of the reinforcement learning agent, the evaporation system can adaptively predict scaling risk, induce flow diversion, and adjust pressure wave stripping in conjunction with other mechanisms during long-term continuous operation. This significantly reduces the wall deposition rate, extends the cleaning cycle, and stably maintains the low-temperature, high-concentration evaporation conditions. The specific steps include the following: S610: During the continuous operation of the evaporation system, the reinforcement learning agent invokes the latest convergence strategy. Receive status input at 5-second intervals:

[0072] And output control actions:

[0073] It achieves closed-loop automatic adjustment of bypass diversion ratio, temperature difference, reflux flow rate, stirring speed, pressure wave amplitude and frequency; the system is executed in real time through PLC or DCS, and the signal delay is controlled within 1 second.

[0074] S620: Based on risk assessment results and trend signals The system dynamically adjusts the intervention intensity. When the SI index drops to SI≤0.5 and When the value is ≤0, the diversion ratio will be automatically reduced. Reduce to the lower limit of 5% and decrease the pressure fluctuation amplitude. When the pressure drops to the 0.5 kPa range, it enters a "low-energy maintenance" mode; when the SI rises and approaches 1.0, the reinforcement learning agent automatically improves its performance. Up to 15–20%, Up to 2–3 kPa, and shorten It enters "active inhibition" mode.

[0075] S630: To avoid system oscillations caused by frequent switching, the control strategy introduces a hysteresis interval:

[0076] Intervention mode switching is triggered only when SI exceeds the hysteresis range. This hysteresis strategy can significantly reduce unnecessary frequent valve and pump operations during long-term operation, thereby reducing actuator fatigue and energy consumption.

[0077] S640: During long-term continuous operation, the system continuously monitors the following indicators: heat exchange efficiency Requirement: ≥90%; Scale Rate: ≤0.05 g / m²·h; Evaporation efficiency (EvapEff): Required to be ≥80% of baseline operating conditions; Energy Cost: ≥15% reduction compared to initial operating conditions.

[0078] If the above indicators meet the interval stability condition for ≥30 minutes, the system will maintain the current strategy without adjustment and only record the data; if any indicator deviates from the threshold interval, the strategy network will be automatically invoked to readjust the action vector.

[0079] S650: In engineering applications, the system can achieve continuous and stable operation for ≥500 hours without requiring shutdown for cleaning; heat exchange efficiency degradation is ≤10%, and the scaling thickness growth rate is controlled below 0.05 g / m²·h. The system automatically records SI throughout the entire process. , , Key parameters, as well as operating indicators such as heat exchange efficiency, energy consumption, and evaporation efficiency, are monitored, supporting strategy playback, auditing, and optimization. If a sustained deviation trend occurs during long-term operation, a "safety soft limit mode" can be triggered to limit the peak pressure wave and the diversion ratio, preventing abnormal oscillations caused by extreme operations.

[0080] Example 2 like Figure 2 As shown, this embodiment provides a low-vacuum reboiler control system based on intelligent algorithms, including: The data acquisition module is used to install temperature sensors, vacuum pressure sensors, conductivity sensors, turbidity meters and level gauges on the evaporator body, circulation pipeline and vacuum system of the LVR evaporation system, to acquire multi-source operating condition signals at a preset sampling frequency, and to perform timestamp synchronization, anomaly removal, smoothing filtering and standardization processing on the signals. The soft measurement and risk assessment module is used to calculate the supersaturation index of the solution based on the processed signal, and combine it with the threshold to determine whether the evaporation system is in a normal, metastable or critical deposition state, thereby identifying the risk of scaling. The reinforcement learning strategy decision-making module receives the state vector, uses the reinforcement learning strategy to classify the risk state, and outputs the risk level, confidence level and trigger signal. The bypass nucleation and diversion control module is used to control the diversion valve, bypass temperature, reflux flow rate and stirring speed under the trigger signal. It achieves solution concentration regulation and local supersaturation reduction in the main heat exchange zone through bypass-induced nucleation. The pressure wave disturbance and descaling module is used to drive the vacuum pump, proportional valve and waveform controller to apply negative pressure wave to disturb and remove the crystal nucleus layer deposited on the inner wall of the evaporator, and to achieve adaptive adjustment of the disturbance intensity through risk judgment signal. The strategy optimization and adaptive control module is used to collect state, action and feedback information generated during the control process, and to periodically update and optimize the reinforcement learning strategy according to the preset reward mechanism. The closed-loop maintenance and energy consumption optimization module is used to dynamically adjust the diversion ratio, stirring intensity and pressure wave amplitude according to the solution supersaturation and its changing trend, so as to suppress scaling risk and achieve long-term stable operation, and maintain evaporation efficiency, heat exchange efficiency and energy consumption within the set range.

[0081] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0082] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0083] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0084] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0085] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0086] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0087] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0088] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0089] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0090] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A control method for an LVR evaporation system based on an intelligent algorithm, characterized in that, The method comprises the following steps: S1, collecting evaporator feed liquid temperature, vacuum degree, conductivity, turbidity and feed liquid height state data, calculating scaling risk index, the scaling risk index includes supersaturation index, scaling induction risk index and wall local supersaturation region distribution; S2, constructing an enhanced learning agent, inputting the scaling risk index as a state vector into the enhanced learning agent, and identifying three risk levels of normal working condition, metastable induction working condition and critical deposition working condition; S3, triggering bypass nucleation shunt regulation when the metastable induction working condition or the critical deposition working condition is met, automatically adjusting the opening of the feed liquid shunt valve and the reflux valve, and introducing part of the feed liquid with a saturation index higher than a predetermined value into a bypass nucleation unit for controllable induced crystallization; S4, applying a negative pressure wave field in the process of controllable induced crystallization, adjusting the frequency of the vacuum pump and the pressure wave function generation module linkage to generate a negative pressure fluctuation signal with controlled amplitude and frequency, realizing dynamic stripping of initial crystal nucleus deposition; S5, optimizing the bypass shunt and pressure wave coupling parameters based on the enhanced learning strategy, online iterative optimizing the bypass shunt ratio, pressure wave amplitude and frequency, so as to achieve the optimal balance between the descaling effect and the system evaporation efficiency.

2. The control method of the LVR evaporation system based on intelligent algorithm according to claim 1, characterized in that, Step S1 specifically comprises: Temperature sensors, pressure sensors, conductivity sensors, turbidity sensors and liquid level sensors are arranged on the evaporator body, circulating pipeline and vacuum system of the LVR evaporation system to collect multiple source sensing signals including liquid state, vacuum degree, solution conductivity, liquid level and suspended particle concentration in real time; The multiple source sensing signals are synchronized and the abnormal values are removed to construct a time-consistent and noise-suppressed operation characteristic sequence; The operation characteristic sequence is inversed online by using thermodynamics and solubility model to obtain corresponding supersaturation index, scaling induction risk index and wall local supersaturation distribution matrix.

3. The control method of the LVR evaporation system based on intelligent algorithm according to claim 1 or 2, characterized in that, Step S2 specifically comprises: Based on the PPO architecture, an enhanced learning agent is constructed, the maximum values of the multiple source sensing signals, the supersaturation index, the scaling induction risk index and the wall local supersaturation distribution matrix are collectively constructed as a state vector, and the state vector is input into the enhanced learning agent to classify and infer the scaling risk of the LVR evaporation system in real time; The output probability of the enhanced learning agent is used to distinguish three risk levels of normal working condition, metastable induction working condition and critical deposition working condition; when the risk level is metastable induction working condition or critical deposition working condition, a risk trigger signal is sent.

4. The control method of the LVR evaporation system based on intelligent algorithm according to claim 3, characterized in that, Step S3 specifically comprises: After receiving the risk trigger signal, the enhanced learning agent outputs corresponding shunt control actions according to the current risk level; Adjust the opening of the bypass shunt valve and the reflux valve to introduce part of the feed liquid with a saturation index higher than a predetermined value into the bypass nucleation unit; control the temperature and flow rate in the bypass nucleation unit to make the supersaturated feed liquid form crystal nucleus in the bypass first; Part of the crystal slurry or clear liquid is returned to the main evaporator according to the set proportion to form a controlled solute concentration gradient; record the shunt ratio, bypass temperature and reflux flow parameters.

5. The control method of the LVR evaporation system based on intelligent algorithm according to claim 1, characterized in that, Step S4 specifically comprises: Start the pressure wave function generation module, and set the pressure wave amplitude and frequency according to the control signal of the enhanced learning agent; The frequency control of the vacuum pump is coupled with the pressure wave function generating module to form periodic negative pressure fluctuations, which act on the evaporation kettle body; Tangential disturbance and periodic shear force are generated in the boundary layer region of the wall surface to dynamically strip the initial crystal nucleus deposition; The turbidity change and scaling trend under the intervention of the pressure wave are continuously monitored and fed back to the enhanced learning agent in real time; after the stripping effect reaches the preset stable interval, the wave form control parameters are kept in the steady state output mode.

6. The control method of a LVR evaporation system based on intelligent algorithm according to claim 1, characterized in that, Step S5 specifically includes: The enhanced learning agent learns and iterates on the historical operation data and real-time feedback data according to the state, action, and reward mechanism; The bypass shunt ratio, pressure wave amplitude, and frequency control parameters are dynamically adjusted to form a multi-dimensional control strategy, and the comprehensive reward value is calculated in each strategy update period, including the descaling effect reward and the evaporation efficiency reward; According to the reward value, the gradient of the strategy network parameters is updated to tend to the optimal control combination; when the reward reaches the preset convergence threshold, the stable strategy is output for real-time operation.

7. The control method of the LVR evaporation system based on intelligent algorithm according to claim 1 or 2, characterized in that, Step S6 specifically includes: The enhanced learning agent calls the converged strategy in the actual operation process to perform real-time closed-loop adjustment on the shunt and pressure wave intervention; In the continuous operation of the LVR evaporation system, risk state monitoring and strategy optimization are executed synchronously to form a dynamic self-adaptive control closed loop; When the detected risk state falls back to the normal interval, the shunt ratio and the pressure wave amplitude are automatically reduced; When the detected risk state approaches the threshold, the intervention intensity is automatically increased.

8. A low vacuum reboiler control system, characterized by, It includes: The data acquisition module includes temperature sensors, pressure sensors, conductivity sensors, turbidity sensors, and liquid level sensors arranged at the evaporator body, circulating pipeline, and vacuum system of the LVR evaporation system, which is used to collect multi-source sensing signals including liquid state, vacuum degree, solution conductivity, liquid level, and suspended particle concentration in real time, and perform timestamp synchronization, abnormality rejection, smoothing filtering, and standardization processing on the multi-source sensing signals; The soft measurement and risk judgment module is used to calculate the scaling risk index and identify three risk levels of normal working condition, sub-stable induction working condition, and critical deposition working condition based on the scaling risk index; The enhanced learning strategy decision module is used to output a risk trigger signal according to the current risk level; The bypass nucleation shunt control module is used to control the shunt valve, bypass temperature, backflow flow rate, and stirring speed under the risk trigger signal to realize solution concentration adjustment and local supersaturation weakening through bypass-induced nucleation; The pressure wave disturbance descaling module is used to drive the vacuum pump, proportional valve, and wave controller to apply negative pressure waves to disturb and strip the crystal nucleus layer deposited on the inner wall of the evaporation kettle, and to realize adaptive adjustment of the disturbance intensity through the risk trigger signal; The strategy optimization and adaptive control module is used to collect the state, action, and feedback information generated in the control process, and periodically update and optimize the enhanced learning strategy according to the preset reward mechanism; The closed-loop maintenance and energy consumption optimization module is used to dynamically adjust the shunt ratio, stirring intensity, and pressure wave amplitude according to the solution supersaturation and its change trend to realize scaling risk suppression and long-period stable operation, and maintain the evaporation efficiency, heat exchange efficiency, and energy consumption within the set interval.

9. An electronic device, comprising: It includes: A processor and a memory having computer program instructions stored therein; The processor implements the intelligent algorithm-based LVR evaporation system control method according to any one of claims 1 to 7 when executing the computer program instructions.

10. A computer-readable storage medium, characterized in that, The storage medium has at least one executable instruction stored therein, and the executable instruction makes the electronic device execute the intelligent algorithm-based LVR evaporation system control method according to any one of claims 1 to 7 when running on the electronic device.

Images