Efficiency improvement and consumption reduction intelligent regulation method and system for reverse osmosis system driven by reinforcement learning AI large model

Abstract

The present disclosure provides a method and system for efficiency improvement and consumption reduction of a reverse osmosis system driven by a reinforcement learning AI large model, comprising: obtaining operating state parameters of a multi-stage reverse osmosis system, and performing cross-stage load transfer analysis on the operating state parameters to construct a cross-stage operating state sequence. The reverse osmosis system of different stages is divided into operating stages, the load evolution path is learned by an AI large model, and a cross-stage load constraint mapping set is generated by combining a reinforcement learning algorithm. The operating trend changes under different combinations of operating state parameters are predicted, and the energy-saving behavior of the previous stage is constrained and limited to determine a cross-stage collaborative regulation strategy. The operation of each stage of the reverse osmosis system is collaboratively adjusted, and the adjusted operating deviation is analyzed and corrected to obtain a cross-stage operation feedback result. According to the cross-stage operation feedback result, the cross-stage load transfer path is learned, the cross-stage collaborative regulation strategy is optimized, and the operating state parameters are updated.

Description

Technical Field

[0001] This disclosure belongs to the field of artificial intelligence technology, and more specifically, relates to a method and system for intelligent control of reverse osmosis system based on reinforcement learning AI large model to improve efficiency and reduce consumption. Background Technology

[0002] In industrial pure water deep treatment systems, a single-stage, two-stage reverse osmosis process is typically employed. When the system actively reduces the operating pressure of the upstream stage to lower overall energy consumption, although upstream energy consumption decreases significantly, the salinity of the permeate may slowly increase. This change may not exceed permissible limits in the short term, but it will accumulate in the downstream reverse osmosis system, causing the downstream membrane modules to operate at high osmotic pressure for extended periods. Ultimately, this results in energy savings in the upstream system, while the downstream system is forced to maintain high-pressure operation for an extended period.

[0003] Currently, while existing technologies provide advanced industrial pure water treatment systems that can dynamically optimize a single system, they lack long-term load transfer constraints between membrane modules to a certain extent. Therefore, they are more prone to the reverse phenomenon of "local energy saving but overall increased consumption". Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to overcome the aforementioned deficiencies and propose a method and system for intelligent control of reverse osmosis systems based on reinforcement learning AI large model to improve efficiency and reduce energy consumption.

[0005] The present invention adopts the following technical solution.

[0006] The first aspect of this invention discloses a method for intelligent regulation and control of a reverse osmosis system based on a reinforcement learning AI large model to improve efficiency and reduce energy consumption, the method comprising:

[0007] The operating status parameters of a multi-stage reverse osmosis system are obtained, and cross-stage load transfer analysis is performed on the operating status parameters to construct a cross-stage operating status sequence.

[0008] The operation stages of the reverse osmosis system in different sections are divided according to the cross-section operation state sequence, and the load evolution path between multiple membrane modules is learned through the AI ​​large model, so as to generate a cross-section load constraint mapping set by combining reinforcement learning algorithm.

[0009] Based on the cross-segment load constraint mapping set, the changes in operating trends under different combinations of operating state parameters are predicted, and the energy-saving behavior of the front end is constrained and limited in order to determine the cross-segment coordinated control strategy.

[0010] The cross-segment coordinated control strategy is used to coordinate the operation of each segment of the reverse osmosis system, and the operation deviation between the actual operating state parameters after adjustment and the target operating state parameters is analyzed and corrected to obtain cross-segment operation feedback results.

[0011] Based on the cross-segment operation feedback results, the cross-segment load transmission path is learned, and the cross-segment collaborative control strategy is optimized to update the operating status parameters of the multi-segment reverse osmosis system.

[0012] Furthermore, the step of acquiring the operating status parameters of the multi-stage reverse osmosis system and performing cross-stage load transfer analysis on the operating status parameters to construct a cross-stage operating status sequence includes:

[0013] Raw operating data of the first and second stages of reverse osmosis systems are collected according to a preset time period, and the raw operating data is time-aligned and invalid data is removed to obtain cross-stage aligned operating data segments.

[0014] Extract the first-stage permeate conductivity, the second-stage operating pressure, and the second-stage membrane pressure difference from the cross-segment aligned operation data segment, and construct the cross-segment salinity pressure transmission coefficient based on the first-stage permeate conductivity, the second-stage operating pressure, and the second-stage membrane pressure difference;

[0015] The cross-segment energy consumption inheritance coefficient is calculated based on the cross-segment salinity-pressure transmission coefficient, and the cross-segment load transmission characteristics are constructed based on the cross-segment energy consumption inheritance coefficient.

[0016] The multi-stage reverse osmosis system includes a first-stage reverse osmosis system and a second-stage reverse osmosis system, and the cross-stage operating state sequence is obtained by arranging the cross-stage load transfer characteristics in chronological order.

[0017] Furthermore, the step of dividing the reverse osmosis system into operational stages based on the cross-segment operational state sequence, and learning the load evolution path between multiple membrane modules using an AI large model, in conjunction with a reinforcement learning algorithm to generate a cross-segment load constraint mapping set, includes:

[0018] The changes in the operating status parameters of a reverse osmosis system are segmented, and each segment is marked based on the changing trend of the operating status parameters of each segment to construct a cross-segment operating stage marking sequence.

[0019] The preceding energy-saving stage and the following load inheritance stage are determined from the cross-segment operation stage marker sequence, and a multi-segment load inheritance relationship sample is constructed based on the preceding energy-saving stage and the following load inheritance stage.

[0020] Furthermore, the step of dividing the reverse osmosis system into operational stages based on the cross-segment operational state sequence, and learning the load evolution path between multiple membrane modules using an AI large model to generate a cross-segment load constraint mapping set in conjunction with a reinforcement learning algorithm, also includes:

[0021] The timing encoder of the AI ​​big model receives the operating status parameters in the multi-segment load inheritance relationship sample, and identifies the delay segment in which a first energy-saving behavior affects a second load through an attention mechanism, and outputs the evolution trend of the second load through a prediction encoder.

[0022] Based on the delay segment of the first stage of energy-saving behavior affecting the second stage of load and the evolution trend of the second stage of load, a load evolution path representation is constructed, and the cross-segment load constraint mapping set is established according to the load evolution path representation through a reinforcement learning algorithm.

[0023] The core of the AI ​​big model is a time-series big model for industrial time-series data, and the structure of the AI ​​big model is a combination of a time-series encoder, an attention mechanism, and a predictive decoder.

[0024] Furthermore, the step of predicting the changes in operating trends under different combinations of operating state parameters based on the cross-segment load constraint mapping set, and constraining and limiting the energy-saving behavior of the upstream section to determine the cross-segment coordinated control strategy, includes:

[0025] The operating status parameters of each segment of the reverse osmosis system are read from the cross-segment load constraint mapping set, and the adjustment range of the operating status parameters of a segment of the reverse osmosis system is narrowed according to the load evolution path to obtain cross-segment candidate control parameters.

[0026] The cross-segment candidate control parameters are input into the AI ​​big model to call the AI ​​big model to predict candidate operating state parameters, and the candidate operating state parameters are used to constrain and limit the front-end energy-saving behavior to obtain a constrained candidate control strategy.

[0027] The candidate control strategies for constraints are screened according to preset screening rules to determine the target operating state parameters that meet the set requirements, and a cross-segment collaborative control strategy to be executed is generated based on the target operating state parameters.

[0028] Furthermore, the cross-section coordinated control strategy is used to coordinately regulate the operation of each section of the reverse osmosis system, and the operational deviation between the adjusted actual operating parameters and the target operating parameters is analyzed and corrected to obtain cross-section operational feedback results, including:

[0029] Based on the cross-segment coordinated control strategy, the reverse osmosis system of each segment is coordinated and adjusted, and the operating status parameters of each segment are monitored to obtain the actual operating status parameters after adjustment.

[0030] Calculate the operating deviation between the adjusted actual operating state parameters and the corresponding target operating state parameters, and determine whether the operating deviation is within a preset range;

[0031] If not, then a correction parameter is generated using a reinforcement learning algorithm, and the cross-segment operation feedback result is output based on the correction parameter.

[0032] Furthermore, the step of learning the cross-section load transfer path based on the cross-section operation feedback results and optimizing the cross-section coordinated control strategy to update the operating status parameters of the multi-section reverse osmosis system includes:

[0033] Based on the cross-section operation feedback results, the adjusted actual operating parameters and correction parameters are integrated into load transfer learning samples, and continuous window analysis is performed on the load transfer learning samples to identify high-pressure risk sections in the two-stage reverse osmosis system.

[0034] Retrieve the energy-saving strategy corresponding to the high-pressure risk section, and determine the pressure compensation of the second-stage reverse osmosis system corresponding to the energy-saving behavior in the first-stage energy-saving strategy, so as to generate cross-stage collaborative optimization parameters based on the pressure compensation;

[0035] Based on the cross-segment collaborative optimization parameters, the operating status parameters of the multi-segment reverse osmosis system are periodically iteratively updated.

[0036] The second aspect of this invention discloses a smart control system for improving efficiency and reducing energy consumption in a reverse osmosis system based on a reinforcement learning AI large model, used to implement the smart control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a reinforcement learning AI large model as described in any of the first aspects, the system comprising:

[0037] The cross-section load analysis module is used to obtain the operating status parameters of the multi-section reverse osmosis system and perform cross-section load transfer analysis on the operating status parameters to construct a cross-section operating status sequence.

[0038] The load constraint mapping module is used to divide the operation stages of the reverse osmosis system of different segments according to the cross-segment operation state sequence, and to learn the load evolution path between multiple membrane modules through the AI ​​large model, so as to generate a cross-segment load constraint mapping set by combining reinforcement learning algorithm.

[0039] The control strategy generation module is used to predict the changes in operating trends under different combinations of operating state parameters based on the cross-segment load constraint mapping set, and to constrain and limit the energy-saving behavior of the front end in order to determine the cross-segment coordinated control strategy.

[0040] The operation correction feedback module is used to coordinate the operation of each section of the reverse osmosis system through the cross-section coordinated control strategy, and to analyze and correct the operation deviation between the actual operating state parameters after adjustment and the target operating state parameters, so as to obtain the cross-section operation feedback results.

[0041] The operating parameter optimization module is used to learn the cross-segment load transmission path based on the cross-segment operation feedback results and optimize the cross-segment coordinated control strategy to update the operating status parameters of the multi-stage reverse osmosis system.

[0042] A third aspect of the present invention discloses a terminal, including a processor and a storage medium;

[0043] The storage medium is used to store instructions;

[0044] The processor is configured to operate according to the instructions to perform the steps of the method described in the first aspect.

[0045] A fourth aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0046] The beneficial effects of the present invention are as follows: Compared with the prior art, the present invention has the following advantages:

[0047] (1) This invention acquires the operating status parameters of a multi-stage reverse osmosis system and performs cross-stage load transfer analysis on these parameters to construct a cross-stage operating status sequence. Then, based on this sequence, the operating stages of different reverse osmosis stages are divided, and the load evolution path between multiple membrane modules is learned using a large AI model. This is combined with a reinforcement learning algorithm to generate a cross-stage load constraint mapping set. Subsequently, based on the generated cross-stage load constraint mapping set, the operating trend changes under different combinations of operating status parameters are predicted, and upstream energy-saving behaviors are constrained to determine a cross-stage coordinated control strategy, fundamentally realizing long-term load transfer constraints between cross-stage membrane modules.

[0048] (2) After determining the cross-segment coordinated control strategy, this invention coordinates the operation of each segment of the reverse osmosis system through the cross-segment coordinated control strategy, and analyzes and corrects the operational deviation between the adjusted actual operating state parameters and the target operating state parameters to obtain cross-segment operation feedback results. Finally, based on the cross-segment operation feedback results, the cross-segment load transfer path is learned, and the cross-segment coordinated control strategy is optimized to update the operating state parameters of the multi-segment reverse osmosis system. On the basis of realizing the long-term load transfer constraint between cross-segment membrane modules, the operating state parameters of the multi-segment reverse osmosis system are further updated and optimized, realizing the long-term coordinated operation of the multi-segment reverse osmosis system with low energy consumption and high stability. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating the intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large AI model driven by reinforcement learning, provided by the present invention.

[0050] Figure 2 This is a schematic diagram of a single-stage energy-saving and two-stage load inheritance process provided by the present invention;

[0051] Figure 3 This is a schematic diagram of the coordinated constraint between front-end energy saving and back-end high voltage provided by the present invention;

[0052] Figure 4 This is an isometric schematic diagram of the multi-stage reverse osmosis cross-section coordinated adjustment device provided by the present invention;

[0053] Figure 5 This is a schematic diagram of the intelligent control system for improving efficiency and reducing energy consumption in reverse osmosis systems based on a large AI model driven by reinforcement learning, provided by the present invention. Detailed Implementation

[0054] The present application will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and should not be construed as limiting the scope of protection of the present application.

[0055] like Figure 1 As shown in one embodiment, a method for intelligent control of a reverse osmosis system based on a reinforcement learning AI large model to improve efficiency and reduce energy consumption includes the following steps:

[0056] Step S110: Obtain the operating status parameters of the multi-stage reverse osmosis system, and perform cross-stage load transfer analysis on the operating status parameters to construct a cross-stage operating status sequence.

[0057] In some embodiments, the intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large AI model driven by reinforcement learning provided by the present invention includes the following steps in step S110:

[0058] Step S111: Collect raw operating data of the first-stage reverse osmosis system and the second-stage reverse osmosis system according to the preset time period, and perform time alignment and invalid data removal on the raw operating data to obtain cross-stage aligned operating data segments.

[0059] Step S112: Extract the first-stage product water conductivity, the second-stage operating pressure, and the second-stage membrane pressure difference from the cross-segment aligned operation data segment, and construct the cross-segment salinity pressure transmission coefficient based on the first-stage product water conductivity, the second-stage operating pressure, and the second-stage membrane pressure difference.

[0060] For example, the conductivity of the permeate water in a first stage refers to a characterizing parameter of the ion conductivity in the permeate water side of a first stage reverse osmosis system, which is used to reflect the desalination effect of the first stage reverse osmosis system and the water quality status of the first stage permeate water.

[0061] Step S113: Calculate the cross-segment energy consumption inheritance coefficient based on the cross-segment salinity pressure transmission coefficient, and construct the cross-segment load transmission characteristics based on the cross-segment energy consumption inheritance coefficient.

[0062] Generally, a cross-stage salinity-pressure transfer coefficient can be constructed by first analyzing the change trend of conductivity in the first-stage product water and the change trend of operating pressure in the second-stage. This is then coupled with an analysis of the energy-saving benefits in the first stage and the increased energy consumption in the second stage. If, after a decrease in operating pressure in the first stage, the operating pressure and energy consumption per ton of water in the second stage continuously increase, it indicates that the energy-saving behavior in the first stage has been transformed into increased energy consumption in the subsequent stage, thus forming a cross-stage energy consumption inheritance coefficient. In this multi-stage reverse osmosis system, there are first-stage and second-stage reverse osmosis systems, and the cross-stage operating state sequence is obtained by arranging the cross-stage load transfer characteristics in chronological order.

[0063] In a specific embodiment, the intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a reinforcement learning AI large model provided by the present invention includes steps 1 to 5:

[0064] Step 1: Perform cross-segment operation status acquisition and load transfer feature construction processing for multi-segment reverse osmosis systems.

[0065] First, the feed water conductivity, product water conductivity, membrane pressure difference, operating pressure, high-pressure pump frequency, product water flow rate, and energy consumption per ton of water in both the first and second stage reverse osmosis systems are collected synchronously. Then, the cross-time correlation between the salinity variation trend of the first stage reverse osmosis product water and the operating pressure variation trend of the second stage reverse osmosis system is analyzed. Next, the load transfer process between the energy-saving stage of the first stage reverse osmosis system and the high-pressure operation stage of the second stage reverse osmosis system is mapped. Finally, a cross-stage load transfer feature set reflecting the long-term load inheritance relationship between multiple membrane modules is constructed, generating the corresponding cross-stage operating state sequence, including the following sub-steps:

[0066] Sub-step 1.1: Synchronous collection of full-dimensional operational data for both the first and second stages of reverse osmosis.

[0067] Specifically, the system pressure, feed water flow rate, product water flow rate, concentrate flow rate, feed water conductivity, product water conductivity, membrane pressure differential, energy consumption per ton of water, high-pressure pump frequency, chemical dosage, concentrate reflux ratio, and operating pressure are collected for both the first and second stage reverse osmosis systems. The data collection period is 5-60 seconds, and 10-30 seconds for industrial pure water systems; the data collection window is 0.5-6 hours, and the cross-stage load analysis is no less than 2 hours. Finally, the original cross-stage operation data package is output, which includes at least the product water quality, operating pressure, high-pressure pump frequency, energy consumption per ton of water for the first stage, feed water quality, operating pressure, membrane pressure differential, and energy consumption per ton of water for the second stage.

[0068] Sub-step 1.2: Cross-segment timing alignment and effective running segment filtering.

[0069] Specifically, the original operational data packets across sections are rearranged according to a unified sampling time, and the data from the first stage of reverse osmosis permeate side and the second stage of reverse osmosis feed water side are matched according to the water flow transmission time. The water flow transmission time is determined by the intermediate tank volume, pipeline volume, and the second stage feed water flow rate, and is set to 10-60 minutes. Then, unstable water production sections such as shutdowns, flushing, chemical cleaning, and abnormal discharges are removed, retaining only the effective operating sections under continuous water production conditions. Finally, an energy-saving change section and a second-stage pressure response section are extracted from each effective operating section, enabling subsequent analysis to consider the continuous change process across sections, rather than just a single-point state.

[0070] Sub-step 1.3: Calculation of the relationship between the salinity change of the first stage product water and the pressure response of the second stage.

[0071] Specifically, the conductivity of the first-stage product water is used as an engineering characteristic of the change in salinity of the first-stage product water, and the operating pressure and membrane pressure difference of the second stage are used as engineering characteristics of the change in load of the subsequent stage. The relative increase in salinity of the first-stage product water is multiplied by the relative increase in pressure of the subsequent stage, and then multiplied by the cross-stage flow carrying ratio to determine whether the slow increase in salinity of the first-stage product water is being transformed into a high-pressure load in the second stage, thus obtaining the cross-stage salinity-pressure transmission coefficient.

[0072] Sub-step 1.4: Coupled calculation of the energy-saving benefits of the first stage and the increased energy consumption costs of the second stage.

[0073] Specifically, based on the cross-segment salinity pressure transfer coefficient, it is further calculated whether the energy saving in the first stage is offset by the increased energy consumption in the second stage. In the calculation, the energy consumption pressure burden of the delayed increase in the second stage system is divided by the energy consumption pressure benefit of the current saving in the first stage system. If the obtained value is consistently high, it indicates that the energy saving behavior of the first stage pressure reduction has been transferred to the second stage system as the increased energy consumption in the later stage.

[0074] Sub-step 1.5: Generation of cross-segment load transfer feature set.

[0075] Specifically, the cross-segment salinity-pressure transfer coefficient, cross-segment energy consumption inheritance coefficient, changes in membrane pressure difference between the two segments, the decrease in the frequency of the high-pressure pump in the first segment, changes in the dosage in the first segment, and changes in the concentrate reflux ratio in the first segment are collectively constituted as a cross-segment load transfer characteristic set. Within this cross-segment load transfer characteristic set, water quality transfer, energy consumption transfer, membrane pressure difference transfer, and changes in core control parameters are uniformly compressed into a single cross-segment load transfer intensity. This intensity is used to determine whether a control action in the first segment is inducing a long-term increase in the load in the second segment. Multiple cross-segment load transfer intensities are arranged in chronological order to form a cross-segment operating state sequence.

[0076] Step S120: Divide the operation stages of the reverse osmosis system into different segments according to the cross-segment operation status sequence, and learn the load evolution path between multiple membrane modules through an AI large model, so as to generate a cross-segment load constraint mapping set by combining reinforcement learning algorithm.

[0077] In some embodiments, the intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large AI model driven by reinforcement learning provided by the present invention includes the following steps in step S120:

[0078] Step S121: The changes in the operating status parameters of a reverse osmosis system are segmented and processed, and each segment is marked based on the changing trend of the operating status parameters of each segment to construct a cross-segment operating stage marking sequence.

[0079] Step S122: Determine the preceding energy-saving stage and the following load inheritance stage from the cross-segment operation stage marker sequence, and construct a multi-segment load inheritance relationship sample based on the preceding energy-saving stage and the following load inheritance stage.

[0080] In some embodiments, the intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large AI model driven by reinforcement learning provided by the present invention further includes the following steps in step S120:

[0081] Step S123: The time encoder of the AI ​​large model receives the operating status parameters in the multi-segment load inheritance relationship samples, and identifies the delay segment of the energy-saving behavior of one segment affecting the two segments of load through the attention mechanism, and outputs the evolution trend of the two segments of load through the prediction encoder.

[0082] Step S124: Based on the delay segment of the load influencing the energy-saving behavior of the first segment and the evolution trend of the load in the second segment, construct a load evolution path representation, and establish a cross-segment load constraint mapping set based on the load evolution path representation through a reinforcement learning algorithm.

[0083] The core of the AI ​​big model is a time-series big model for industrial time-series data. The structure of the AI ​​big model is a combination of a time-series encoder, an attention mechanism, and a predictive decoder.

[0084] In a specific embodiment, the intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a reinforcement learning AI large model provided by the present invention includes the following steps: Step 2 involves performing multi-segment load inheritance relationship modeling based on the cross-segment operating state sequence. Based on the cross-segment operating state sequence output in Step 1, the pressure change behavior of the first-stage reverse osmosis system and the osmotic pressure growth behavior of the second-stage reverse osmosis system are divided into stages; subsequently, the load transfer intensity, energy consumption amplification trend, and membrane pressure inheritance state in different operating stages are correlated and analyzed; then, the long-term load evolution path between multiple membrane modules is learned using the AI ​​large model to identify the delayed load impact of upstream energy-saving behavior on downstream membrane modules; finally, a cross-segment load inheritance state space is established using a reinforcement learning algorithm to generate the corresponding cross-segment load constraint mapping set, including the following sub-steps:

[0085] Sub-step 2.1, cross-segment operation phase division processing.

[0086] Specifically, based on the cross-segment operating state sequence, the pressure change behavior of the first-stage reverse osmosis system, the high-pressure pump frequency change behavior of the first-stage system, the permeate conductivity change behavior of the first-stage system, the operating pressure change behavior of the second-stage reverse osmosis system, the membrane pressure difference change behavior of the second-stage system, and the energy consumption per ton of water change behavior of the second-stage system are continuously segmented. During the segmentation process, the segment where the operating pressure and energy consumption per ton of water in the first stage continuously decrease are marked as the upstream energy-saving stage; the segment where the permeate conductivity slowly increases but has not yet exceeded the permeate control limit is marked as the salinity latent increase stage; and the segment where the operating pressure, membrane pressure difference, or energy consumption per ton of water in the second stage continuously increases after a delay is marked as the downstream load inheritance stage. This stage division is not simply based on a single threshold, but rather on the continuous changing trends of pressure, flow rate, water quality, energy consumption, membrane pressure difference, and core control parameters. That is, it simultaneously observes whether the first-stage reverse osmosis system is reducing pressure for energy saving, whether the permeate salinity is increasing, and whether the pressure in the second stage is increasing with a delay, and amplifies the constraints through the cross-segment load transfer intensity.

[0087] Sub-step 2.2: Modeling and processing the multi-segment load inheritance intensity and energy consumption amplification trend.

[0088] Specifically, based on the cross-stage operation phase marker sequence, for each upstream energy-saving stage, the corresponding implicit salinity increase stage and downstream load inheritance stage are searched backwards to establish a continuous inheritance chain of first-stage control behavior, first-stage permeate salinity change, second-stage pressure load increase, and overall energy consumption change. This inheritance chain is used to express that after the first stage of reverse osmosis system depressurization operation, its impact is not immediately manifested as system anomaly, but rather as a slow increase in the first-stage permeate conductivity, which is then transmitted through the intermediate tank and pipelines, transforming into an increase in osmotic pressure, operating pressure, and energy consumption per ton of water in the second-stage reverse osmosis system. During execution, the downstream membrane pressure difference increase, downstream energy consumption increase, downstream permeate capacity decrease, and upstream control action intensity are coupled to determine whether the first-stage energy-saving action is substantially inherited into the second-stage load.

[0089] For example, Figure 2 This diagram illustrates the cross-stage process of energy-saving behavior being inherited by the second-stage reverse osmosis system. The upper section represents the product water output and transmission delay path from the first-stage reverse osmosis system through the intermediate water tank to the second-stage reverse osmosis system. The four curves in the lower section, from top to bottom, represent the first-stage operating pressure, first-stage product water conductivity, second-stage operating pressure and membrane pressure difference, and energy consumption per ton of water in the second stage. The dashed line marks the upper limit of product water control. The decrease in first-stage operating pressure corresponds to the preceding energy-saving stage. The slow increase in first-stage product water conductivity below the upper limit corresponds to the implicit increase in salinity. After the delay section, the second-stage operating pressure, membrane pressure difference, and energy consumption per ton of water in the second stage rise synchronously, corresponding to the subsequent load inheritance stage. Based on this, the cross-stage salinity-pressure transmission coefficient, cross-stage energy consumption inheritance coefficient, and cross-stage load transmission characteristic set are obtained.

[0090] Sub-step 2.3: Learning and processing the long-term load evolution path based on the AI ​​large model.

[0091] Specifically, a large-scale time-series model oriented towards industrial time-series data is adopted as the main body of the AI ​​large-scale model, employing a structure of time-series encoder + attention mechanism + predictive decoder. The time-series encoder receives continuous features such as pressure, flow rate, water quality, energy consumption, membrane pressure difference, high-pressure pump frequency, chemical dosage, concentrate recirculation ratio, and operating pressure from a multi-segment load inheritance relationship sample set. The attention mechanism identifies the key delay segment where energy-saving behavior in one segment affects high-pressure load in another. The predictive decoder outputs the pressure growth trend, energy consumption growth trend, and cross-segment load inheritance trend for the second segment within the subsequent operating window.

[0092] In this embodiment, the model parameters are set as follows: the historical observation window is set to 1-24 hours, and the industrial pure water system is set to 4-8 hours; the prediction window is set to 1-3 hours; the number of attention layers is set to 2-6 layers; the number of hidden units is set to 64-512; the learning rate is set to 0.0001-0.001; and the batch sample size is set to 32-256. The AI ​​large model processing includes sample window slicing, temporal feature encoding, cross-segment delayed attention calculation, subsequent load trend prediction, and prediction bias backpropagation update. This controls the AI ​​large model to focus on the critical moments in the historical window that truly affect the subsequent load, and reduces the impact of premature historical states through exponential decay.

[0093] Sub-step 2.4: Establishment of cross-segment load inheritance state space.

[0094] Specifically, based on the long-term load evolution path representation results, a cross-stage load inheritance state space is established using reinforcement learning algorithms. This cross-stage load inheritance state space is not an instantaneous set of states for a single-stage reverse osmosis system, but rather incorporates the energy-saving state of the first stage, the implicit increase in permeate salinity in the first stage, the pressure inheritance state of the second stage, the energy amplification state of the second stage, and the overall system energy consumption benefit state. Furthermore, in reinforcement learning, the state can be understood as the current cross-stage load situation; the action can be understood as the adjustment and selection of the first stage operating pressure, high-pressure pump frequency, chemical dosage, concentrate reflux ratio, and second-stage pressure compensation requirements; the reward can be understood as reducing the overall energy consumption per ton of water while ensuring stable permeate water quality and preventing deterioration of membrane pressure differential, rather than just reducing the energy consumption of the first stage system. Moreover, the control strategy is considered effective only when the overall energy consumption truly decreases, and the long-term load evolution intensity, the risk of exceeding limits in the second-stage permeate water quality, and the risk of high pressure in the second stage are all low.

[0095] Finally, the long-term load evolution patterns identified by the AI ​​model are transformed into states, actions, and reward constraints that can be used by reinforcement learning, enabling subsequent steps to generate cross-segment collaborative control strategies instead of continuing to execute single-segment local energy-saving strategies, thereby outputting a cross-segment load constraint mapping set.

[0096] Step S130: Based on the cross-segment load constraint mapping set, predict the changes in operating trends under different combinations of operating state parameters, and constrain and limit the energy-saving behavior of the front end in order to determine the cross-segment coordinated control strategy.

[0097] In some embodiments, the intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large AI model driven by reinforcement learning provided by the present invention includes the following steps in step S130:

[0098] Step S131: Read the operating status parameters of each section of the reverse osmosis system from the cross-section load constraint mapping set, and narrow the adjustment range of the operating status parameters of a section of the reverse osmosis system according to the load evolution path to obtain cross-section candidate control parameters.

[0099] Specifically, long-term load evolution paths can be used to identify the delayed impact of energy-saving behavior on high-pressure load in the second stage, and then the allowable adjustment ranges of operating pressure, high-pressure pump frequency, concentrate reflux ratio, and chemical dosage in the first stage can be limited. For example, when the membrane pressure difference in the second stage continues to increase, the concentrate reflux ratio in the first stage can be limited to continue to increase, thereby forming cross-stage candidate control parameters.

[0100] Step S132: Input the cross-segment candidate control parameters into the AI ​​big model to call the AI ​​big model to predict the candidate operating state parameters, and use the candidate operating state parameters to constrain and limit the energy-saving behavior of the front end to obtain the constrained candidate control strategy.

[0101] In a specific embodiment, the cross-segment candidate control parameters can be input into the AI ​​large model first. The timing encoder, cross-segment attention layer and prediction decoder predict the subsequent overall energy consumption per ton of water, the operating pressure of the second stage and the change trend of the membrane pressure difference of the second stage. Subsequently, if the prediction results show that energy saving in the first stage will induce long-term high-pressure operation in the second stage, the pressure reduction energy saving behavior of the first stage is restricted and a constraint candidate control strategy is generated.

[0102] Step S133: Select the candidate control strategies according to the preset screening rules to determine the target operating state parameters that meet the set requirements, and generate the cross-segment collaborative control strategy to be executed based on the target operating state parameters.

[0103] In a specific embodiment, the intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a reinforcement learning AI large model provided by the present invention includes step 3, which involves generating a collaborative control strategy based on a cross-section load constraint mapping set. Based on the cross-section load constraint mapping set output in step 2, a joint analysis is performed on the first-stage reverse osmosis operating pressure, high-pressure pump frequency, concentrate reflux ratio, and second-stage reverse osmosis pressure compensation requirements. Subsequently, the overall system energy consumption changes, downstream membrane module pressure growth trends, and long-term operational stability under different parameter combinations are predicted. Then, constraints are imposed on upstream energy-saving behaviors that may lead to long-term high-pressure operation in the downstream section. Finally, target operating parameter combinations that meet the requirements of overall low energy consumption and cross-section pressure balance are selected to generate a corresponding cross-section collaborative control strategy set, including the following sub-steps:

[0104] Sub-step 3.1: Generation and processing of cross-segment candidate control parameter combinations.

[0105] Specifically, based on the cross-section load constraint mapping set, the operating pressure, high-pressure pump frequency, concentrate reflux ratio, and chemical dosage of the current first-stage reverse osmosis system, as well as the operating pressure, membrane pressure difference, permeate conductivity, and energy consumption per ton of water of the second-stage reverse osmosis system, are first read. Then, based on the long-term load evolution intensity, cross-section constraint rewards, and downstream high-pressure risk formed in step 2, the range of adjustable operating parameters for the first-stage reverse osmosis system is narrowed. During execution, when the long-term load evolution intensity is high, the first-stage reverse osmosis system is not allowed to continue to reduce its operating pressure independently; when the membrane pressure difference in the second stage continues to rise, the excessive increase in the concentrate reflux ratio in the first stage is limited; when the permeate conductivity in the second stage approaches the control upper limit, the weight of the permeate quality constraint in the first stage is increased. Finally, multiple candidate control parameter combinations are formed, each combination including at least candidate values ​​for the first-stage operating pressure, the first-stage high-pressure pump frequency, the first-stage concentrate reflux ratio, the first-stage chemical dosage, and the second-stage pressure compensation. For example, the changes in pressure, high-pressure pump frequency, concentrate reflux ratio, and chemical dosage are uniformly converted into dimensionless regulation intensity, and the long-term load evolution intensity is superimposed. At the same time, the cross-section constraint reward is deducted. The larger the value obtained, the stronger the cross-section stability disturbance is, although the corresponding candidate parameter combination may produce energy-saving effect.

[0106] Sub-step 3.2: Prediction and processing of candidate parameter running results based on AI large model.

[0107] Specifically, the set of candidate control parameters for each stage is input into a large-scale AI model for industrial reverse osmosis time series analysis for prediction. This AI model employs a structure consisting of a time series feature encoder, a cross-stage attention layer, a parameter perturbation embedding layer, and a multi-objective predictive decoder. The time series feature encoder reads comprehensive time-series operational data from past operating windows, including pressure, flow rate, water quality, energy consumption, and membrane pressure differential. The cross-stage attention layer identifies the delayed impact of a control action in the first stage on the pressure increase in the second stage. The parameter perturbation embedding layer represents the perturbations of different candidate operating pressures, high-pressure pump frequency, concentrate reflux ratio, and chemical dosage on subsequent operating conditions. The multi-objective predictive decoder outputs the overall predicted energy consumption per ton of water, the predicted pressure increase in the second stage, the predicted membrane pressure differential increase in the second stage, the predicted conductivity of the second stage permeate, and the predicted long-term operational stability.

[0108] During the execution of the aforementioned large-scale AI model, the deviations of candidate parameter combinations from overall energy consumption, second-stage pressure increase, second-stage membrane pressure differential increase, and second-stage permeate water quality need to be examined simultaneously. First, the historical operating window is divided into equal-length time segments. Then, perturbations are injected into each candidate parameter combination to simulate the operating trend after execution. Next, the influence of first-stage parameter changes on the pressure inheritance state of the second-stage membrane module is calculated using cross-segment attention layers. Finally, the predicted results for energy consumption, water quality, pressure, and stability corresponding to different candidate combinations are output. The historical observation window for the large-scale AI model is set to 4-8 hours, the prediction window to 1-3 hours, the number of attention layers to 2-6, the number of hidden units to 64-512, and the number of candidate parameter combinations to 20-200.

[0109] Sub-step 3.3: Handling of high-voltage risk constraints in the back end of the front-end energy-saving behavior.

[0110] Specifically, based on the candidate parameter prediction results set, constraints are imposed on upstream energy-saving behaviors that may lead to prolonged high-pressure operation in the downstream stage. During execution, when the candidate parameter combination shows a decrease in operating pressure, a decrease in high-pressure pump frequency, and a decrease in energy consumption per ton of water in the first stage, but simultaneously predicts an increase in operating pressure, membrane pressure difference, or energy consumption per ton of water in the second stage, the system no longer classifies it as an effective energy-saving strategy, but instead marks it as a cross-stage load transfer strategy. For cross-stage load transfer strategies, it is necessary to limit further reduction in the operating pressure of the first stage and simultaneously add constraints on pressure compensation in the second stage, permeate water quality, or concentrate recirculation adjustment in the first stage. For example, constraints are simultaneously included on predicted pressure exceeding limits in the second stage, membrane pressure difference exceeding limits in the second stage, energy consumption exceeding limits in the second stage, and the intensity of upstream control actions. The more aggressive the energy-saving action in the first stage and the higher the intensity of multi-stage load inheritance, the greater the high-pressure risk constraint value in the downstream stage. Understandably, the synergistic constraints between upstream energy saving and downstream high pressure can be referenced... Figure 3 As shown.

[0111] Sub-step 3.4: Overall low energy consumption and cross-segment pressure balance combined screening process.

[0112] Specifically, a joint screening process is conducted on the set of constrained candidate control strategies. The screening principle is not simply to select the combination with the lowest energy consumption in the first stage, but rather, under the premise that the water quality of the second-stage permeate does not deteriorate, the membrane pressure difference in the second stage does not continuously increase, and the operating pressure of the second stage does not remain close to the control upper limit for an extended period, the combination with the lowest overall energy consumption per ton of water and a more balanced pressure distribution across stages is selected. During the screening, the overall energy consumption prediction deviation, the high-pressure risk constraint value in the later stage, the degree of pressure balance across stages, and the candidate control intensity are comprehensively scored. The candidate combination with the lowest score is determined as the target operating parameter combination. For example, by incorporating prediction deviation, high-pressure risk in the later stage, pressure balance deviation across stages, and the magnitude of control action into the screening, the selected operating parameters are not the "most energy-efficient in the first stage," but rather the "most stable and energy-efficient combination of the first and second stages."

[0113] Sub-step 3.5: Generation and processing of cross-segment collaborative control strategy set.

[0114] Specifically, based on the target combination of operating parameters, a set of cross-segment coordinated control strategies is generated for subsequent execution. This set of strategies includes at least the first-stage operating pressure setpoint, the first-stage high-pressure pump frequency setpoint, the first-stage concentrate reflux ratio setpoint, the first-stage dosing setpoint, the second-stage pressure compensation setpoint, the second-stage membrane pressure difference monitoring constraint value, the second-stage product water conductivity control constraint value, and the overall unit energy consumption target value per ton of water.

[0115] In this embodiment, to prevent sudden parameter changes from causing pressure shocks to the membrane module, adjustment rate limits and execution hold times need to be set for each parameter. Specifically, the single adjustment range of the operating pressure is controlled within 1%-5% of the current pressure; the single adjustment range of the high-pressure pump frequency is controlled within 0.5-2.5 Hz; the single adjustment range of the concentrate reflux ratio is controlled within 1%-5%; the single adjustment range of the chemical dosage is controlled within 2%-10% of the current dosage; and the strategy hold time is set to 20-60 minutes.

[0116] Finally, the set of cross-stage coordinated control strategies is output to step 4, used to actually regulate the high-pressure pump, reflux control module, chemical dosing module, and operating pressure control module in the first and second stage reverse osmosis systems. The output set of cross-stage coordinated control strategies is not a single control value, but a set of coordinated control strategies with execution constraints, hold-up times, risk limits, and feedback monitoring conditions.

[0117] Step S140: The operation of each section of the reverse osmosis system is coordinated and adjusted through a cross-section coordinated control strategy. The operational deviation between the actual operating parameters and the target operating parameters after adjustment is analyzed and corrected to obtain cross-section operation feedback results.

[0118] In some embodiments, the intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large AI model driven by reinforcement learning provided by the present invention includes the following steps in step S140:

[0119] Step S141: Based on the cross-section coordinated control strategy, the reverse osmosis system of each section is coordinated and adjusted, and the operating status parameters of the reverse osmosis system of each section are monitored to obtain the actual operating status parameters after adjustment.

[0120] Step S142: Calculate the operating deviation between the adjusted actual operating state parameters and the corresponding target operating state parameters, and determine whether the operating deviation is within the preset range.

[0121] Step S143: If not, generate correction parameters through reinforcement learning algorithm, and output cross-segment running feedback results based on correction parameters.

[0122] In a specific embodiment, the intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a reinforcement learning AI large model provided by the present invention includes step 4, which involves executing cross-section collaborative control and dynamic feedback correction processing. Based on the cross-section collaborative control strategy set output in step 3, the high-pressure pump, reflux control module, and operating pressure in the first-stage and second-stage reverse osmosis systems are coordinated and adjusted. Subsequently, the changes in salinity of the first-stage permeate, the changes in operating pressure in the second stage, and the overall energy consumption per ton of water are continuously monitored after adjustment. Then, the deviation between the actual operating results and the target collaborative operating state is analyzed. Finally, the cross-section load inheritance weight and operating reward relationship are dynamically corrected using a reinforcement learning algorithm to generate corresponding cross-section operating feedback results, including the following sub-steps:

[0123] Sub-step 4.1: Execution of cross-segment collaborative control strategy.

[0124] Specifically, based on a set of cross-stage coordinated control strategies, coordinated regulation is performed on both the first-stage and second-stage reverse osmosis systems. For the first-stage system, adjustments are made according to the setpoints for operating pressure, high-pressure pump frequency, concentrate reflux ratio, and chemical dosing from the strategy set. For the second-stage system, adjustments are made according to the setpoints for pressure compensation, membrane differential pressure monitoring, and permeate conductivity control. During execution, a step-by-step approach is used instead of instantaneous jumps, allowing the high-pressure pump frequency, operating pressure, and concentrate reflux ratio to gradually reach their target values ​​within a set holding time. Furthermore, the distance between the actual operating parameters and the target operating parameters during strategy execution is assessed, and the allowable deviation is gradually reduced as execution progresses. In this example, the single adjustment range of the operating pressure is controlled at 1%-5% of the current pressure, the single adjustment range of the high-pressure pump frequency is controlled at 1-3 Hz, the single adjustment range of the concentrate reflux ratio is controlled at 1%-5%, and the single adjustment range of the dosage is controlled at 2%-10% of the current dosage.

[0125] For example, Figure 4 The diagram shows the physical arrangement of a first-stage reverse osmosis system 1 and a second-stage reverse osmosis system 3 connected in series via an intermediate water tank 2. In the first-stage reverse osmosis system 1, a first-stage dosing module 14 is connected to the inlet side of a first-stage high-pressure pump 11 via a first-stage pipe 16. The high-pressure pump 11 pressurizes the water and sends it to a first-stage membrane module 12 via a first-stage operating pressure control module 15. The concentrate from the first-stage membrane module 12 is returned to the inlet side of the high-pressure pump 11 via a first-stage reflux control module 13, while the permeate flows into the intermediate water tank 2 via a first-stage pipe 16. The effluent from the intermediate water tank 2 enters the second-stage high-pressure pump 31 via a second-stage pipe 34, and is then sent to the second-stage membrane module 32 via a second-stage operating pressure control module 33. All pump bodies and valve bodies are connected to the pipes via flanges sharing the same side, with the direction of action pointing from the first-stage reverse osmosis system 1 to the second-stage reverse osmosis system 3.

[0126] Sub-step 4.2: Continuous monitoring and processing of the cross-segment operation status after regulation.

[0127] Specifically, after the cross-segment control execution record is formed, continuous monitoring is conducted on the following parameters: first-stage permeate conductivity, first-stage permeate flow rate, first-stage energy consumption per ton of water, second-stage operating pressure, second-stage membrane pressure differential, second-stage permeate conductivity, second-stage energy consumption per ton of water, and overall energy consumption per ton of water. The monitoring cycle is set to 10-30 seconds; the monitoring window covers at least one cross-segment water flow transmission delay time, with a value of 0.5-3 hours. Furthermore, the focus is on determining whether, after the first-stage energy-saving action is implemented, the first-stage permeate salinity rises slowly, whether the second-stage operating pressure increases with a lag, whether the second-stage membrane pressure differential shows a continuous increasing trend, and whether the overall energy consumption per ton of water truly decreases.

[0128] Sub-step 4.3: Analysis and processing of deviations between actual operating results and target coordinated state.

[0129] Specifically, the monitored sequence of cross-segment operation after regulation is compared with the target coordinated operation state corresponding to the target operating parameter combination in step 3. The deviations in water quality (stage 1), pressure (stage 2), membrane pressure difference (stage 2), and overall energy consumption are calculated. In other words, these deviations are uniformly converted into dimensionless deviations to determine whether the coordinated regulation deviates from the target state. In this example, if the salinity of the permeate in stage 1 increases slightly but the pressure in stage 2 does not increase significantly, the strategy is considered to be within an acceptable range. If the energy consumption in stage 1 decreases but the operating pressure, membrane pressure difference, and overall energy consumption per ton of water in stage 2 continuously increase, the strategy is deemed to have a risk of cross-segment load transfer.

[0130] Sub-step 4.4: Correction of the relationship between load inheritance weights and rewards based on reinforcement learning.

[0131] Specifically, based on the evaluation results of cross-segment coordination deviation, a strategy correction method from reinforcement learning is used to dynamically update the cross-segment load inheritance weights and operational reward relationships. In this example, the reinforcement learning state includes the current cross-segment coordination deviation, the second-stage pressure increase state, the overall energy consumption change state, and the long-term load evolution intensity; the actions are to increase the first-stage pressure constraint, decrease the first-stage energy saving margin, adjust the first-stage concentrate recirculation ratio, increase the first-stage chemical dosing constraint, or increase the second-stage pressure compensation constraint; the rewards are a decrease in overall energy consumption per ton of water, no increase in second-stage operating pressure, no deterioration in second-stage membrane pressure differential, and stable product water quality. The discount factor is set to 0.8-0.99, the learning rate to 0.0001-0.001, and the exploration ratio to 1%-10%.

[0132] In this embodiment, the expression for the modified strategy value is:

[0133]

[0134] In the formula, The revised strategy value. The value of the strategy before the revision. For learning rate, As a discount factor, The maximum strategy value in the next running state.

[0135] Reinforcement learning instant rewards The expression is:

[0136]

[0137] In the formula, To enhance the immediate rewards for learning, To adjust the overall energy consumption per ton of water before adjustment, To adjust the overall energy consumption per ton of water, This represents the cross-segment coordination deviation value. For long-term load evolution intensity, This is the risk constraint value for the subsequent high-pressure section. , , These are the normalized weighting coefficients. This serves as a baseline value for long-term load evolution. This serves as the baseline value for downstream high-pressure risks. In actual implementation, it should be dynamically adjusted based on the current dominant risks: when the downstream high-pressure risk increases, it should be increased. When long-term load inheritance increases, improve When the deviation in coordinated operation increases, improve The values ​​range from 0.8 to 0.95. The long-term load evolution baseline value is taken from the effective operating section where neither the second-stage operating pressure nor the second-stage membrane pressure difference has continuously increased. It is grouped according to operating load scenario, salinity fluctuation scenario, and membrane fouling stage. The upper quartile of the long-term load evolution intensity of each group is taken, and the values ​​are recalculated after each round of feedback and archiving. The high-pressure risk baseline value for the later stage is taken from the full-scale operating condition where the second-stage predicted pressure, the second-stage membrane pressure difference, and the second-stage energy consumption per ton of water simultaneously reach their respective over-limit thresholds. This condition is substituted into the formula for the high-pressure risk constraint value for the later stage, and it is updated as the threshold settings change. Both values ​​are positive, with dimensions consistent with the long-term load evolution intensity and the high-pressure risk constraint value for the later stage, respectively. They remain constant during a single round of strategy execution, ensuring that the two ratios obtained by division fall within the range of 0 to 1.

[0138] As can be seen from the above formula, reinforcement learning will only give positive rewards when the overall energy consumption decreases and the coordination deviation, long-term load evolution intensity, and high-pressure risk in the later stage are all low; if there is only energy saving in one stage and pressure increases in the second stage, the reward will be reduced or even negative.

[0139] Sub-step 4.5: Generation and processing of cross-segment operation feedback results.

[0140] Specifically, based on the reinforcement learning-corrected parameter set, cross-stage operation feedback results are generated. These feedback results include at least the overall change in energy consumption per ton of water after the current strategy execution, the change in first-stage product water salinity, the change in second-stage operating pressure, the change in second-stage membrane pressure differential, the correction result for high-pressure risk in the downstream stage, the correction result for cross-stage load inheritance weights, and the reward correction relationship required for the next strategy generation. If the feedback results show that the second-stage operating pressure and membrane pressure differential continue to rise, the restriction weight on first-stage pressure reduction and energy-saving behavior in subsequent strategies is increased; if the feedback results show that overall energy consumption decreases and the second-stage pressure remains stable, the current collaborative control direction is retained. This ultimately forms cross-stage operation feedback results that can be reused in step 5 for long-term collaborative efficiency improvement and energy consumption reduction optimization of the multi-stage reverse osmosis system.

[0141] Step S150: Learn the cross-section load transfer path based on the cross-section operation feedback results, and optimize the cross-section coordinated control strategy to update the operating status parameters of the multi-section reverse osmosis system.

[0142] In some embodiments, the intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large AI model driven by reinforcement learning provided by the present invention includes the following steps in step S150:

[0143] Step S151: Based on the cross-section operation feedback results, the adjusted actual operating parameters and correction parameters are integrated into load transfer learning samples, and continuous window analysis is performed on the load transfer learning samples to identify high-pressure risk sections in the two-stage reverse osmosis system.

[0144] Step S152: Retrieve the energy-saving strategy corresponding to the high-pressure risk section, and determine the pressure compensation of the second-stage reverse osmosis system corresponding to the energy-saving behavior in the first-stage energy-saving strategy, so as to generate cross-stage collaborative optimization parameters based on the pressure compensation.

[0145] Step S153: Based on the cross-segment collaborative optimization parameters, the operating status parameters of the multi-segment reverse osmosis system are periodically iteratively updated.

[0146] In a specific embodiment, the intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a reinforcement learning AI large model provided by the present invention includes the following sub-steps: Step 5 involves performing long-term collaborative optimization for improving efficiency and reducing energy consumption in multi-stage reverse osmosis. Based on the cross-stage operation feedback results output in Step 4, long-term learning is performed on the cross-stage load transfer paths in different industrial pure water operation load scenarios, different feed water salinity fluctuation scenarios, and different membrane fouling stages. Subsequently, the continuous high-pressure operation section of the downstream membrane module is identified. Then, the synergistic relationship between the first-stage energy-saving strategy and the second-stage pressure compensation strategy is continuously optimized. Finally, the operating parameters of the multi-stage reverse osmosis system are dynamically iteratively updated to achieve intelligent collaborative control of the multi-stage reverse osmosis system with low overall energy consumption, low pressure accumulation, and long-term stable operation.

[0147] Sub-step 5.1: Long-term learning and archiving of cross-segment operation feedback results.

[0148] Specifically, based on the feedback results of cross-segment operation, the changes in overall energy consumption per ton of water after each round of regulation, the changes in salinity of the first-stage permeate, the changes in operating pressure of the second stage, the changes in membrane pressure difference of the second stage, the correction results for high-pressure risks in the later stage, and the correction results for cross-segment load inheritance weights are archived. Subsequently, the system is classified according to industrial pure water operating load scenarios, influent salinity fluctuation scenarios, and membrane fouling stages, enabling the system to distinguish the load transfer characteristics under different operating conditions such as low load and low salinity, high load and high salinity, early stage of membrane fouling, and stage of worsening membrane fouling, thereby forming a long-term load transfer learning sample set.

[0149] Sub-step 5.2: Identification and processing of the continuous high-pressure operation section of the downstream membrane module.

[0150] Specifically, continuous window analysis is performed on the long-term load transfer learning sample set to identify sections in the two-stage reverse osmosis system where the operating pressure is consistently high, the membrane pressure differential is continuously increasing, and the energy consumption per ton of water is continuously rising. During identification, the degree of pressure exceeding the limit, membrane pressure differential exceeding the limit, and overall energy consumption increase must be measured simultaneously within the continuous window.

[0151] Sub-step 5.3 involves the coordinated optimization of a first-stage energy-saving strategy and a second-stage pressure compensation strategy.

[0152] Specifically, for the set of continuous high-pressure risk sections in the later stage, the energy-saving strategy of the corresponding time period is traced back to determine whether the decrease in operating pressure, decrease in high-pressure pump frequency, change in concentrate return ratio, and change in chemical dosage in the first stage caused long-term pressure compensation in the second stage. The higher the continuous high-pressure risk in the later stage, the more obvious the long-term load evolution, and the more aggressive the parameter adjustment in the first stage, the stronger the collaborative optimization constraint. If the cross-section constraint reward is high, it indicates that the strategy is effective overall, and the constraint strength is reduced accordingly.

[0153] Sub-step 5.4: Dynamic iteration and intelligent control of multi-stage reverse osmosis operating parameters output processing.

[0154] Specifically, based on a cross-segment collaborative optimization parameter set, the operating pressure of the first stage, the frequency of the high-pressure pump, the concentrate reflux ratio, the dosage, and the pressure compensation setpoints of the second stage are periodically updated iteratively. The long-term strategy value is updated through a reinforcement learning algorithm, with the state representing the current load scenario, salinity fluctuation scenario, membrane fouling stage, and subsequent high-pressure risk zone; the actions are adjustments to the first stage pressure, pump frequency, reflux ratio, and dosage, and the second stage pressure compensation; the rewards are a reduction in overall energy consumption per ton of water, a cumulative reduction in second-stage pressure, and stable product water quality.

[0155] In this embodiment, the long-term control strategy can only achieve positive rewards when overall energy consumption decreases, the risk of sustained high pressure in the downstream stage decreases, the pressure of collaborative constraints decreases, and the water quality of the second-stage permeate does not deteriorate. The final output is a multi-stage reverse osmosis long-term collaborative efficiency improvement and energy consumption reduction control result, achieving intelligent collaborative control with overall low energy consumption, low pressure accumulation, and long-term stable operation.

[0156] The following describes the intelligent control system for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large AI model driven by reinforcement learning, provided by this invention. The intelligent control system for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large AI model driven by reinforcement learning described below can be referred to in correspondence with the intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large AI model driven by reinforcement learning described above.

[0157] like Figure 5 As shown in one embodiment, a smart control system for improving efficiency and reducing energy consumption in a reverse osmosis system based on a reinforcement learning AI large model includes a cross-segment load analysis module, a load constraint mapping module, a control strategy generation module, an operation correction feedback module, and an operation parameter optimization module.

[0158] The cross-section load analysis module is used to obtain the operating status parameters of a multi-section reverse osmosis system and perform cross-section load transfer analysis on the operating status parameters to construct a cross-section operating status sequence.

[0159] The load constraint mapping module is used to divide the operation stages of different reverse osmosis systems according to the cross-section operation state sequence, and learns the load evolution path between multiple membrane modules through an AI large model, so as to generate a cross-section load constraint mapping set by combining reinforcement learning algorithm.

[0160] The control strategy generation module is used to predict the changes in operating trends under different combinations of operating state parameters based on the cross-segment load constraint mapping set, and to constrain and limit the energy-saving behavior of the front end in order to determine the cross-segment coordinated control strategy.

[0161] The operation correction feedback module is used to coordinate the operation of each section of the reverse osmosis system through cross-section collaborative control strategy, and to analyze and correct the operation deviation between the actual operating state parameters after adjustment and the target operating state parameters, so as to obtain cross-section operation feedback results.

[0162] The operating parameter optimization module is used to learn the cross-section load transmission path based on the cross-section operation feedback results and optimize the cross-section coordinated control strategy to update the operating status parameters of the multi-section reverse osmosis system.

[0163] The applicant of this invention has provided a detailed description of the embodiments of the invention in conjunction with the accompanying drawings. However, those skilled in the art should understand that the above embodiments are merely preferred embodiments of the invention. The detailed description is only intended to help readers better understand the spirit of the invention and is not intended to limit the scope of protection of the invention. On the contrary, any improvements or modifications made based on the inventive spirit of the invention should fall within the scope of protection of the invention.

Claims

1. A method for intelligent control of a reverse osmosis system based on a reinforcement learning AI large model to improve efficiency and reduce energy consumption, characterized in that, The method includes: The operating status parameters of a multi-stage reverse osmosis system are obtained, and cross-stage load transfer analysis is performed on the operating status parameters to construct a cross-stage operating status sequence. The operation stages of the reverse osmosis system in different sections are divided according to the cross-section operation state sequence, and the load evolution path between multiple membrane modules is learned through the AI ​​large model, so as to generate a cross-section load constraint mapping set by combining reinforcement learning algorithm. Based on the cross-segment load constraint mapping set, the changes in operating trends under different combinations of operating state parameters are predicted, and the energy-saving behavior of the front end is constrained and limited in order to determine the cross-segment coordinated control strategy. The cross-segment coordinated control strategy is used to coordinate the operation of each segment of the reverse osmosis system, and the operation deviation between the actual operating state parameters after adjustment and the target operating state parameters is analyzed and corrected to obtain cross-segment operation feedback results. Based on the cross-segment operation feedback results, the cross-segment load transmission path is learned, and the cross-segment collaborative control strategy is optimized to update the operating status parameters of the multi-segment reverse osmosis system.

2. The intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large AI model driven by reinforcement learning, as described in claim 1, is characterized in that... The process of acquiring the operating status parameters of the multi-stage reverse osmosis system and performing cross-stage load transfer analysis on these parameters to construct a cross-stage operating status sequence includes: Raw operating data of the first and second stages of reverse osmosis systems are collected according to a preset time period, and the raw operating data is time-aligned and invalid data is removed to obtain cross-stage aligned operating data segments. Extract the first-stage permeate conductivity, the second-stage operating pressure, and the second-stage membrane pressure difference from the cross-segment aligned operation data segment, and construct the cross-segment salinity pressure transmission coefficient based on the first-stage permeate conductivity, the second-stage operating pressure, and the second-stage membrane pressure difference; The cross-segment energy consumption inheritance coefficient is calculated based on the cross-segment salinity-pressure transmission coefficient, and the cross-segment load transmission characteristics are constructed based on the cross-segment energy consumption inheritance coefficient. The multi-stage reverse osmosis system includes a first-stage reverse osmosis system and a second-stage reverse osmosis system, and the cross-stage operating state sequence is obtained by arranging the cross-stage load transfer characteristics in chronological order.

3. The intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large AI model driven by reinforcement learning, as described in claim 2, is characterized in that... The process of dividing the reverse osmosis system into operational phases based on the cross-segment operational state sequence, and learning the load evolution paths between multiple membrane modules using an AI large-scale model, combined with reinforcement learning algorithms to generate a cross-segment load constraint mapping set, includes: The changes in the operating status parameters of a reverse osmosis system are segmented, and each segment is marked based on the changing trend of the operating status parameters of each segment to construct a cross-segment operating stage marking sequence. The preceding energy-saving stage and the following load inheritance stage are determined from the cross-segment operation stage marker sequence, and a multi-segment load inheritance relationship sample is constructed based on the preceding energy-saving stage and the following load inheritance stage.

4. The intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large AI model driven by reinforcement learning, as described in claim 3, is characterized in that... The step of dividing the reverse osmosis system into operational stages based on the cross-section operational state sequence, and learning the load evolution path between multiple membrane modules using an AI large model, combined with reinforcement learning algorithms to generate a cross-section load constraint mapping set, also includes: The timing encoder of the AI ​​big model receives the operating status parameters in the multi-segment load inheritance relationship sample, and identifies the delay segment in which a first energy-saving behavior affects a second load through an attention mechanism, and outputs the evolution trend of the second load through a prediction encoder. Based on the delay segment of the first stage of energy-saving behavior affecting the second stage of load and the evolution trend of the second stage of load, a load evolution path representation is constructed, and the cross-segment load constraint mapping set is established according to the load evolution path representation through a reinforcement learning algorithm. The core of the AI ​​big model is a time-series big model for industrial time-series data, and the structure of the AI ​​big model is a combination of a time-series encoder, an attention mechanism, and a predictive decoder.

5. The intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large AI model driven by reinforcement learning, as described in claim 2, is characterized in that... The method of predicting operational trend changes under different combinations of operating state parameters based on the cross-segment load constraint mapping set, and constraining and limiting upstream energy-saving behavior to determine cross-segment coordinated control strategies, includes: The operating status parameters of each segment of the reverse osmosis system are read from the cross-segment load constraint mapping set, and the adjustment range of the operating status parameters of a segment of the reverse osmosis system is narrowed according to the load evolution path to obtain cross-segment candidate control parameters. The cross-segment candidate control parameters are input into the AI ​​big model to call the AI ​​big model to predict candidate operating state parameters, and the candidate operating state parameters are used to constrain and limit the front-end energy-saving behavior to obtain a constrained candidate control strategy. The candidate control strategies for constraints are screened according to preset screening rules to determine the target operating state parameters that meet the set requirements, and a cross-segment collaborative control strategy to be executed is generated based on the target operating state parameters.

6. The intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large AI model driven by reinforcement learning, as described in claim 2, is characterized in that... The cross-section coordinated control strategy is used to coordinate the operation of each section of the reverse osmosis system, and the operational deviation between the adjusted actual operating parameters and the target operating parameters is analyzed and corrected to obtain cross-section operation feedback results, including: Based on the cross-segment coordinated control strategy, the reverse osmosis system of each segment is coordinated and adjusted, and the operating status parameters of each segment are monitored to obtain the actual operating status parameters after adjustment. Calculate the operating deviation between the adjusted actual operating state parameters and the corresponding target operating state parameters, and determine whether the operating deviation is within a preset range; If not, then a correction parameter is generated using a reinforcement learning algorithm, and the cross-segment operation feedback result is output based on the correction parameter.

7. The intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large AI model driven by reinforcement learning, as described in claim 6, is characterized in that... The step of learning the cross-section load transfer path based on the cross-section operation feedback results and optimizing the cross-section coordinated control strategy to update the operating status parameters of the multi-section reverse osmosis system includes: Based on the cross-section operation feedback results, the adjusted actual operating parameters and correction parameters are integrated into load transfer learning samples, and continuous window analysis is performed on the load transfer learning samples to identify high-pressure risk sections in the two-stage reverse osmosis system. Retrieve the energy-saving strategy corresponding to the high-pressure risk section, and determine the pressure compensation of the second-stage reverse osmosis system corresponding to the energy-saving behavior in the first-stage energy-saving strategy, so as to generate cross-stage collaborative optimization parameters based on the pressure compensation; Based on the cross-segment collaborative optimization parameters, the operating status parameters of the multi-segment reverse osmosis system are periodically iteratively updated.

8. A smart control system for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large-scale reinforcement learning AI model, characterized in that: The system is used to implement the intelligent control method for improving efficiency and reducing energy consumption in a reverse osmosis system based on a large AI model driven by reinforcement learning, as described in any one of claims 1 to 7. The system comprises: The cross-section load analysis module is used to obtain the operating status parameters of the multi-section reverse osmosis system and perform cross-section load transfer analysis on the operating status parameters to construct a cross-section operating status sequence. The load constraint mapping module is used to divide the operation stages of the reverse osmosis system of different segments according to the cross-segment operation state sequence, and to learn the load evolution path between multiple membrane modules through the AI ​​large model, so as to generate a cross-segment load constraint mapping set by combining reinforcement learning algorithm. The control strategy generation module is used to predict the changes in operating trends under different combinations of operating state parameters based on the cross-segment load constraint mapping set, and to constrain and limit the energy-saving behavior of the front end in order to determine the cross-segment coordinated control strategy. The operation correction feedback module is used to coordinate the operation of each section of the reverse osmosis system through the cross-section coordinated control strategy, and to analyze and correct the operation deviation between the actual operating state parameters after adjustment and the target operating state parameters, so as to obtain the cross-section operation feedback results. The operating parameter optimization module is used to learn the cross-segment load transmission path based on the cross-segment operation feedback results and optimize the cross-segment coordinated control strategy to update the operating status parameters of the multi-stage reverse osmosis system.

9. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-7.

Images