AI-based water purification system recycling energy-saving method and system

Abstract

The application discloses an AI-based water purification system recycling energy-saving method and system, relates to the technical field of reverse osmosis water purification equipment control, and can be used for seawater desalination treatment. Multidimensional water quality, time period and water taking data of inlet water are collected, water quality parameters are normalized and dimensionally reduced, and then input into a deep neural network to obtain a water quality grade, and pressurizing parameters, concentrated water ratio and water use hysteresis threshold values are matched. The water use probability is predicted by relying on an LSTM model, and efficient water production and energy-saving standby modes are switched through hysteresis determination. An edge computing controller links to output pump valve control instructions, and user water taking amount and membrane water production flow are respectively modeled and fed back. The recovery condition is judged according to the turbidity and conductivity of concentrated water, qualified concentrated water is returned to a pretreatment unit, and the water production deviation, concentrated water quality, cycle pressure and return ratio are combined. The application can avoid energy consumption and water loss caused by fixed pressure, frequent start-stop and disordered return, and is suitable for household water purifiers, commercial direct drinking water stations and seawater desalination reverse osmosis equipment.

Description

Technical Field

[0001] This invention relates to the field of reverse osmosis water purification equipment control technology, specifically to an AI-based water purification system recycling and energy-saving method and system. Background Technology

[0002] Reverse osmosis (RO) water purification equipment typically uses a high-pressure pump to provide a transmembrane pressure differential to the RO membrane treatment unit, separating the incoming water into permeate and concentrate. The permeate is supplied for drinking or storage, while the concentrate is discharged through a concentrate discharge line or returned to the pretreatment stage via a recycling line for further filtration. The energy consumption of the RO membrane treatment unit primarily comes from the high-pressure pump pressurization process. Permeate efficiency and membrane fouling rate are influenced by factors such as influent turbidity, hardness, pH value, total dissolved solids, residual chlorine content, and the proportion of concentrate discharged.

[0003] During periods of low influent pollution load, maintaining a fixed operating pressure causes the high-pressure pump to bear a load exceeding the actual separation requirements. Conversely, during periods of high influent pollution load, maintaining a fixed operating pressure and a fixed concentrate discharge ratio may lead to a decrease in permeate flow, increased concentration polarization, or deterioration of concentrate quality. Simply relying on operating with a fixed wastewater ratio or a fixed pressure makes it difficult to simultaneously achieve energy conservation, permeate stability, and concentrate recovery safety.

[0004] Some water purification systems activate the high-pressure pump based on the water tank level or user water intake. This type of control typically only starts water production after a user has expressed a need for water, which can easily lead to a delayed water supply response. Another type of water purification system sets pre-start or standby times based on historical water usage periods, but the water usage prediction results are usually not integrated with the reverse osmosis membrane operating pressure, concentrate discharge ratio, and concentrate recovery status into a unified control chain. When the predicted probability fluctuates around the start / stop threshold, the high-pressure pump is prone to frequent start-stop cycles, resulting in start-stop losses and water pressure surges.

[0005] Concentrate recovery can reduce direct wastewater discharge, but uncontrolled recirculation increases the fouling load on the primary filtration unit. Recirculation based solely on concentrate conductivity or turbidity fails to address the coupling issues between increased feedwater load after recirculation, continued adjustment of reverse osmosis membrane operating parameters, and sustained pressure application in permeate feedback. When permeate flow is below the target value, conventional feedback control tends to increase the high-pressure pump pressure; however, if the concentrate quality is already close to the recovery limit, further increasing the pressure or recovery ratio will increase the burden on the membrane treatment unit and may cause the water quality in the recovery branch to exceed limits.

[0006] The existing control system suffers from the problem of mixing user water intake data with the actual permeate flow rate of the reverse osmosis membrane treatment unit. User water intake data reflects the user's water demand at the intake point and is suitable for inputting into the water consumption prediction model; the actual permeate flow rate of the reverse osmosis membrane treatment unit reflects the water production capacity and is suitable for pressure feedback correction. If the permeate flow rate is directly used as the user water intake data, in a water purification system equipped with a permeate storage tank, the model may learn the water replenishment behavior rather than the user's water intake behavior, resulting in a discrepancy between the prediction results and the actual water consumption period. Summary of the Invention

[0007] The purpose of this invention is to provide an AI-based water purification system recycling energy-saving method and system. The technical problem to be solved is that when the fluctuation of influent water quality, changes in user water demand and concentrate recovery status are handled by independent logic, the reverse osmosis membrane working pressure, high-pressure pump start-up and shutdown, concentrate valve opening and concentrate recovery path lack unified constraints, which can easily lead to fixed parameter energy consumption, predictive switching jitter and backflow load runaway.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0009] AI-based water purification system recycling and energy-saving methods include:

[0010] Collect influent water quality characteristic parameters, water usage time data, and user water intake data; perform scale unification and feature dimensionality reduction processing on the influent water quality characteristic parameters to obtain dimensionality-reduced feature vectors.

[0011] The reduced feature vector is input into a deep neural network water quality assessment model to obtain the water quality level, and the basic working pressure, concentrate discharge ratio, pressure regulation boundary and water use probability hysteresis threshold of the reverse osmosis membrane treatment unit are generated based on the water quality level.

[0012] The water usage time period data and the user water consumption data are input into the long short-term memory network water usage prediction model to obtain the water usage probability of the next control period. The water usage probability is then hysteresis-determined based on the water usage probability hysteresis threshold to obtain the water purification system operation mode.

[0013] The water purification system operation mode is coupled with the basic working pressure, the concentrate discharge ratio and the pressure regulation boundary to form high-pressure pump pressure control command, concentrate ratio regulation electric valve command and circulating water pump command.

[0014] The reverse osmosis membrane treatment unit is driven to generate permeate and concentrate according to the pressure control command of the high-pressure pump and the command of the concentrate ratio adjustment electric valve. The actual permeate flow rate and concentrate quality judgment result of the reverse osmosis membrane treatment unit are collected. The concentrate that meets the recovery conditions is transported to the primary filtration unit. The actual permeate flow rate and concentrate quality judgment result of the reverse osmosis membrane treatment unit are used to constrain the pressure control command of the high-pressure pump and the command of the concentrate ratio adjustment electric valve in the next control cycle.

[0015] As a preferred embodiment of the present invention, the influent water quality characteristic parameters include influent turbidity, influent hardness, influent pH value, total dissolved solids in the influent, and influent residual chlorine content; the scale unification and feature dimensionality reduction processing includes normalizing the minimum and maximum values ​​of the influent water quality characteristic parameters, determining the principal component projection matrix using the covariance matrix of the training samples, and projecting the normalized influent water quality characteristic parameters onto the principal component projection matrix to form a dimensionality-reduced feature vector composed of multiple principal components.

[0016] As a preferred embodiment of the present invention, the deep neural network water quality assessment model includes an input layer, multiple fully connected hidden layers connected in sequence, and a water quality level output layer; the water quality level output layer outputs the classification probability of each water quality level, and takes the water quality level with the highest classification probability as the water quality level of the current control cycle; the training label of the water quality level is determined based on the target permeate water quality, the allowable concentrate discharge ratio, the target permeate flow rate of the reverse osmosis membrane treatment unit, and the reverse osmosis membrane fouling risk level.

[0017] As a preferred embodiment of the present invention, the basic working pressure is compensated based on the cumulative treated water volume and cumulative usage time of the reverse osmosis membrane treatment unit to obtain a compensated working pressure, and the compensated working pressure is limited within the pressure regulation boundary corresponding to the water quality level; when the compensated working pressure reaches the upper limit of the pressure regulation boundary, it is prohibited to continue to increase the pressure control command of the high-pressure pump.

[0018] As a preferred embodiment of the present invention, the water usage probability hysteresis threshold includes a high threshold and a low threshold; when the water usage probability reaches the high threshold, the water purification system operation mode is switched to high-efficiency water purification mode; when the water usage probability is not higher than the low threshold, the water purification system operation mode is switched to energy-saving standby mode; when the water usage probability is between the high threshold and the low threshold, the current water purification system operation mode is maintained.

[0019] As a preferred technical solution of the present invention, in the high-efficiency water purification mode, the compensation working pressure is used as the pressure control command for the high-pressure pump, and the concentrate discharge ratio is used as the command for the concentrate ratio regulating electric valve, thereby enabling the high-pressure pump and stopping the circulating water pump or putting it into a low-power maintenance state; in the energy-saving standby mode, the pressure control command for the high-pressure pump is set to the shutdown pressure, and the command for the concentrate ratio regulating electric valve is set to the closed state, thereby stopping the high-pressure pump and making the circulating water pump run according to the intermittent cycle.

[0020] As a preferred embodiment of the present invention, the concentrate discharge ratio is the ratio of the concentrate flow rate discharged through the concentrate discharge pipeline to the influent flow rate entering the reverse osmosis membrane treatment unit within a unit control cycle; the concentrate water quality determination result is generated based on the turbidity detection unit and conductivity detection unit installed in the concentrate discharge pipeline;

[0021] When the turbidity of the concentrate is not higher than the turbidity recovery threshold and the conductivity of the concentrate is not higher than the conductivity recovery threshold, the concentrate recovery electric valve and the pressure recovery pump are opened to deliver the concentrate to the inlet of the primary filtration unit; when the turbidity of the concentrate is higher than the turbidity recovery threshold or the conductivity of the concentrate is higher than the conductivity recovery threshold, the concentrate recovery electric valve is closed and the concentrate discharge electric valve is opened.

[0022] As a preferred technical solution of the present invention, in the high-efficiency water purification mode, the target permeate flow rate, the actual permeate flow rate of the reverse osmosis membrane treatment unit, the concentrate turbidity and the concentrate conductivity are collected. The correction direction of the compensation working pressure is determined based on the deviation between the target permeate flow rate and the actual permeate flow rate of the reverse osmosis membrane treatment unit. The question of whether to allow an increase in the compensation working pressure or the concentrate recovery ratio is determined based on the excess state of the concentrate turbidity and the concentrate conductivity relative to the recovery threshold. The corrected compensation working pressure is then limited within the pressure regulation boundary corresponding to the water quality level.

[0023] As a preferred technical solution of the present invention, when the actual permeate flow rate of the reverse osmosis membrane treatment unit is lower than the target permeate flow rate and the concentrate turbidity and concentrate conductivity are not exceeded, the compensation working pressure of the next control cycle is increased.

[0024] When the actual permeate flow rate of the reverse osmosis membrane treatment unit is lower than the target permeate flow rate and the concentrate turbidity or concentrate conductivity exceeds the limit, it is prohibited to increase the compensation working pressure of the next control cycle and reduce the concentrate recovery ratio; when the actual permeate flow rate of the reverse osmosis membrane treatment unit is higher than the target permeate flow rate, the compensation working pressure of the next control cycle is reduced.

[0025] As a preferred technical solution of the present invention, a combined stable quantity is generated based on the deviation of permeate flow rate, the excess of concentrate conductivity and the excess of concentrate turbidity. When the control cycle of the combined stable quantity for a consecutive preset number is not higher than the combined termination threshold, the iterative correction of the compensation working pressure is stopped and the current compensation working pressure is maintained. The concentrate recovery ratio is the ratio of the concentrate flow rate returned to the inlet of the primary filtration unit within a unit control cycle to the total concentrate flow rate generated by the reverse osmosis membrane treatment unit. The edge computing controller changes the concentrate recovery ratio by adjusting the opening of the concentrate recovery electric proportional valve or the speed of the pressure recovery pump.

[0026] As a preferred technical solution of the present invention, the deep neural network water quality assessment model and the long short-term memory network water use prediction model run on an edge computing controller connected to the water purification system; the edge computing controller performs range checks, continuity checks, and rate of change checks on the influent water quality characteristic parameters; when any influent water quality characteristic parameter fails the check, the water quality grade is not updated using the current influent water quality characteristic parameter, the pressure regulation boundary corresponding to the most recent effective water quality grade is used, and the high-pressure pump pressure control command is limited to the most recent effective basic working pressure; when no effective influent water quality characteristic parameter is obtained for multiple consecutive sampling cycles, the concentrate recovery electric valve is closed.

[0027] As a preferred embodiment of the present invention, the water purification system is a household reverse osmosis water purifier or a commercial direct drinking water station. When the water purification system is a household reverse osmosis water purifier, the user water intake data is collected by a user water intake flow meter installed at the user water intake end, or determined by the change in the liquid level of the product water storage tank and the amount of water replenished. When the water purification system is a commercial direct drinking water station, the user water intake data of multiple water terminals is aggregated according to a unified timestamp, and the conductivity of the mixed water is collected in the mixing water section after the concentrated water recovery pipeline and the raw water pipeline merge. When the conductivity of the mixed water exceeds the upper limit of the conductivity of the mixed water, the concentrated water recovery ratio is reduced.

[0028] This invention also discloses an AI-based water purification system recycling and energy-saving system, including an inlet water quality sensor array, a user water intake flow meter, a product water flow meter, a concentrate water quality detection module, a mixed water conductivity sensor, a water usage data acquisition module, a signal conditioning and analog-to-digital conversion module, an edge computing controller, a primary filtration unit, and a reverse osmosis membrane treatment unit.

[0029] The inlet water quality sensor array is located downstream of the mixing pipe section and upstream of the primary filtration unit, and outputs the inlet water quality detection signal to the signal conditioning and analog-to-digital conversion module.

[0030] The user water intake flow meter is installed at the user's water intake end and outputs the user water intake flow signal to the water usage data acquisition module;

[0031] The permeate flow meter is installed in the permeate pipeline of the reverse osmosis membrane treatment unit and outputs the permeate flow signal to the signal conditioning and analog-to-digital conversion module;

[0032] The concentrate water quality detection module is installed in the concentrate discharge pipeline of the reverse osmosis membrane treatment unit and outputs concentrate turbidity signal and concentrate conductivity signal to the signal conditioning and analog-to-digital conversion module.

[0033] The mixed water conductivity sensor is installed in the mixed water section after the concentrated water recovery pipeline and the raw water pipeline merge, and outputs the mixed water conductivity signal to the signal conditioning and analog-to-digital conversion module.

[0034] The edge computing controller includes a water quality feature processing module, a deep neural network water quality assessment module, a long short-term memory network water usage prediction module, an operation mode determination module, a control vector generation module, a concentrate recovery determination module, and a feedback correction module, which are used to realize the AI-based water purification system recycling and energy-saving method described above.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] This invention simultaneously determines the basic operating pressure, concentrate discharge ratio, pressure regulation boundary, and water usage probability hysteresis threshold for water quality classification, avoiding the use of water quality assessment results solely for display or single parameter selection. User water intake data is separated from the actual permeate flow rate of the reverse osmosis membrane treatment unit, ensuring that the input to the long short-term memory network water usage prediction model corresponds to actual water demand, and the input to the feedback correction module corresponds to actual water production capacity. Maintaining the current water purification system operation mode when the water usage probability is between the high and low thresholds reduces repeated start-stop cycles of the high-pressure pump caused by slight fluctuations in the predicted probability. The water purification system operation mode and membrane process parameters are integrated into unified pump and valve control commands, preventing conflicting commands from water quality control logic and sleep control logic to the high-pressure pump and concentrate ratio regulating electric valve. Permeate flow rate deviation is used to correct the operating pressure; excess concentrate conductivity and turbidity limits are used to restrict the correction direction, ensuring consistency between permeate flow control and concentrate recovery conditions. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0038] Figure 1 This is a flowchart of the method described in this invention.

[0039] Figure 2This is a flowchart of the concentrated water recovery determination branch of the present invention.

[0040] Figure 3 This is a flowchart of the direct discharge of concentrated wastewater according to the present invention. Detailed Implementation

[0041] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the embodiments of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0042] The following is in conjunction with the appendix Figures 1-3 The embodiments of the present invention will be described in detail below.

[0043] Example 1: This example uses the physical water purification process of a reverse osmosis water purification system as the control object. Control inputs come from the influent water quality sensor array, user water intake flow meter, product water flow meter, concentrate water quality detection module, and water usage data acquisition module. Control outputs directly act on the frequency converter driver of the high-pressure pump, the concentrate proportioning electric valve, the circulating water pump, the concentrate recovery electric valve, the concentrate discharge electric valve, the pressure recovery pump, and the bypass discharge electric valve.

[0044] Furthermore, the deep neural network water quality assessment model outputs a water quality grade, which is used to determine the basic operating pressure, concentrate discharge ratio, and pressure regulation boundary of the reverse osmosis membrane treatment unit. It is also used to select the high and low thresholds corresponding to the output probability of the long short-term memory network water use prediction model. The water use probability output by the long short-term memory network water use prediction model, after hysteresis judgment, forms the water purification system operation mode. This operation mode determines whether the operating pressure and concentrate discharge ratio corresponding to the water quality grade are sent to the physical actuators.

[0045] After the high-pressure pump and concentrate proportioning electric valve execute the pump and valve control commands, the reverse osmosis membrane treatment unit outputs the actual permeate flow rate, concentrate turbidity, and concentrate conductivity. The actual permeate flow rate is used to calculate the pressure correction, while the concentrate turbidity and conductivity are used to determine the recovery path and limit the pressure and concentrate recovery ratio for the next control cycle. This forms a closed loop from sensor input, feature processing, dual-model interaction, pump and valve execution, permeate and concentrate detection, to correction for the next control cycle.

[0046] The specific method is as follows:

[0047] Step 1: Collect influent water quality characteristics. The raw water pipeline and the concentrate recovery pipeline converge at the mixing pipe section. The influent water quality sensor array is located downstream of the mixing pipe section and upstream of the primary filtration unit. When concentrate recovery is not enabled, the influent water quality sensor array collects the quality of the external raw water. When concentrate recovery is enabled, the influent water quality sensor array collects the quality of the mixed influent water after the external raw water and the recovered concentrate are mixed. Based on the sensor locations, the water quality data entering the model is consistent with the water entering the primary filtration unit and subsequently entering the reverse osmosis membrane treatment unit.

[0048] The primary filtration unit includes a pre-filter, a polypropylene melt-blown filter cartridge, an activated carbon filter cartridge, or a composite pretreatment filter cartridge. Inlet turbidity is collected by a light-scattering turbidity sensor. Inlet hardness is collected by an ion-selective electrode, a titration-based online detection unit, or a calibrated conductivity response combination unit. Inlet pH is collected by a glass electrode pH sensor. Total dissolved solids in the inlet are calculated using a temperature-compensated conductivity sensor. Residual chlorine content in the inlet is collected by an electrochemical residual chlorine sensor.

[0049] The sensor's analog signal is converted into a digital quantity through signal conditioning and analog-to-digital conversion modules. Signal conditioning includes low-pass filtering, zero-point offset correction, and range mapping. Each set of data is written with a unified timestamp. Feed water quality characteristic parameters are collected at 1-5 minute intervals. Membrane inlet pressure, actual permeate flow rate of the reverse osmosis membrane treatment unit, and concentrate water quality parameters are collected at 1-10 second intervals. Data from different sampling frequencies are aligned according to the end time of the control cycle.

[0050] Step 2: Collect water usage time period data and user water consumption data. User water consumption data is collected by a user water flow meter installed at the user's water intake end. The user water flow meter integrates the user water consumption for each sampling period to obtain the user water consumption for the corresponding time period. Water usage time period data includes hour sequence number, weekday type, holiday identifier, and water consumption event time. The edge computing controller arranges the user water consumption of multiple consecutive time periods into a time-series input matrix according to fixed time steps.

[0051] When a household reverse osmosis water purifier includes a product water storage tank but does not have a user water flow meter, the user's water consumption data is determined by the change in the product water storage tank level and the amount of water replenished. The edge computing controller calculates the user's water consumption based on the effective cross-sectional area of ​​the product water storage tank, the drop in water level, and the amount of water replenished within the same time period. The product water flow meter in the product water pipeline is used to collect the actual product water flow of the reverse osmosis membrane treatment unit. This actual product water flow of the reverse osmosis membrane treatment unit is used for feedback correction and is not the sole source of user water consumption data.

[0052] Step 3: Perform a validity check on the influent water quality characteristic parameters. If any of the following conditions are met: the influent water quality characteristic parameter value exceeds the sensor's range; the value remains completely unchanged for multiple consecutive sampling periods while there is flow in the corresponding pipeline; the change in value between adjacent sampling periods exceeds the safe rate of change threshold; or the sensor communication update time exceeds the valid duration, the edge computing controller will mark the corresponding data as invalid data. Invalid data will not be input into the deep neural network water quality assessment model.

[0053] Step 4: Perform scale unification processing on valid influent water quality characteristic parameters. Since the units and numerical ranges of influent turbidity, hardness, pH, total dissolved solids, and residual chlorine differ, the edge computing controller performs minimum-maximum value normalization using the lower and upper limits of the parameters corresponding to the training data. When the online value is below the normalization lower limit, the value corresponding to the normalization lower limit is used; when the online value is above the normalization upper limit, the value corresponding to the normalization upper limit is used, and an out-of-bounds flag is generated simultaneously. The out-of-bounds flag enters the anomaly handling chain without directly changing the model's inference structure.

[0054] Step 5: Perform feature dimensionality reduction processing on the normalized influent water quality characteristic parameters. During the training phase, the mean vector and covariance matrix are calculated using training samples. Eigenvalue decomposition is performed on the covariance matrix, and a predetermined number of eigenvectors are selected according to their eigenvalues ​​from largest to smallest to form the principal component projection matrix. During the online operation phase, the edge computing controller projects the normalized influent water quality characteristic parameters onto the principal component projection matrix to obtain the dimensionality-reduced feature vectors.

[0055] The principal component projection matrix is ​​determined during the model training phase and written into the edge computing controller along with the parameters of the deep neural network water quality assessment model. The covariance matrix is ​​not recalculated during the online operation phase to avoid altering the projection direction due to a small number of anomalous water samples.

[0056] Step 6: Output water quality grades using a deep neural network water quality assessment model. The deep neural network water quality assessment model includes an input layer, multiple fully connected hidden layers, and a water quality grade output layer. The input layer receives dimensionality-reduced feature vectors. The fully connected hidden layers extract the nonlinear combination relationships between influent turbidity, influent hardness, influent pH, total dissolved solids, and residual chlorine content. The water quality grade output layer outputs the classification probability of each water quality grade and selects the water quality grade with the highest classification probability as the water quality grade for the current control period.

[0057] In a feasible configuration, the dimensionality-reduced feature vector comprises three principal components, and the deep neural network water quality assessment model comprises three input nodes, a first hidden layer with 64 neurons, a second hidden layer with 32 neurons, a third hidden layer with 16 neurons, and three output nodes. The three output nodes correspond to Class I, Class II, and Class III water quality, respectively.

[0058] The water quality rating training label is determined based on the target permeate quality, the permissible concentrate discharge ratio, the target permeate flow rate of the reverse osmosis membrane treatment unit, and the reverse osmosis membrane fouling risk level. If the feed water quality of the training sample can still meet the target permeate quality and flow rate under relatively low operating pressure and a relatively low concentrate discharge ratio, the training label can be designated as Level 1 water quality. If the feed water quality of the training sample requires medium operating pressure and a medium concentrate discharge ratio to meet the target permeate quality and flow rate, the training label can be designated as Level 2 water quality. If the feed water quality of the training sample requires higher operating pressure and a higher concentrate discharge ratio, and the membrane fouling risk is higher, the training label can be designated as Level 3 water quality.

[0059] Step 7: Read the membrane process parameters and water usage probability hysteresis threshold according to the water quality level. The edge computing controller reads the basic operating pressure, concentrate discharge ratio, lower pressure limit, upper pressure limit, low threshold, and high threshold corresponding to the current water quality level from the water quality level parameter table. When the water quality level is higher, the water quality level parameter table sets a higher basic operating pressure, a higher concentrate discharge ratio, and an earlier high-efficiency water purification mode wake-up condition.

[0060] The concentrate discharge ratio is the ratio of the concentrate flow rate discharged through the concentrate discharge pipeline to the feed water flow rate entering the reverse osmosis membrane treatment unit within a unit control cycle. The concentrate ratio regulating electric valve adjusts the flow rate of the concentrate discharge pipeline according to the concentrate ratio regulating electric valve command output by the edge computing controller, so that the concentrate discharge ratio within a unit control cycle is close to the set value in the water quality grade parameter table.

[0061] In a feasible configuration, the baseline operating pressure for Class I water quality is 0.60 MPa, the concentrate discharge ratio is 15%, the lower pressure limit is 0.55 MPa, the upper pressure limit is 0.75 MPa, the low threshold is 0.20, and the high threshold is 0.75. The baseline operating pressure for Class II water quality is 0.80 MPa, the concentrate discharge ratio is 25%, the lower pressure limit is 0.75 MPa, the upper pressure limit is 1.05 MPa, the low threshold is 0.20, and the high threshold is 0.70. The baseline operating pressure for Class III water quality is 1.00 MPa, the concentrate discharge ratio is 35%, the lower pressure limit is 0.90 MPa, the upper pressure limit is 1.20 MPa, the low threshold is 0.15, and the high threshold is 0.60.

[0062] The high threshold for Class II water quality is lower than that for Class I water quality, allowing for a longer preparation time under higher treatment load conditions. The low threshold for Class III water quality is relatively lower, preventing the high-pressure pump from stopping immediately after a short period of reduced water usage.

[0063] Step 8: Compensate the base operating pressure based on the cumulative treated water volume and cumulative usage time of the reverse osmosis membrane treatment unit. The cumulative treated water volume is determined by the integral value of the permeate flow meter and the estimated concentrate flow rate. The cumulative usage time is the cumulative duration for which the high-pressure pump is in enabled mode. The edge computing controller obtains the treated water volume compensation pressure from a table based on the cumulative treated water volume and the usage time compensation pressure from a table based on the cumulative usage time. The base operating pressure, treated water volume compensation pressure, and usage time compensation pressure are then combined to form the compensation operating pressure.

[0064] When the cumulative treated water volume does not reach the preset treated water volume threshold, the treated water volume compensation pressure is zero; after the cumulative treated water volume reaches the preset treated water volume threshold, the treated water volume compensation pressure increases in stages. When the cumulative usage time does not reach the preset usage time threshold, the usage time compensation pressure is zero; after the cumulative usage time reaches the preset usage time threshold, the usage time compensation pressure increases in stages.

[0065] The compensation working pressure is not unlimited. The edge computing controller compares the compensation working pressure with the lower and upper pressure limits corresponding to the current water quality level. When the compensation working pressure is lower than the lower pressure limit, the lower pressure limit is used; when the compensation working pressure is higher than the upper pressure limit, the upper pressure limit is used. When the compensation working pressure reaches the upper pressure limit, the feedback correction module no longer adds high-pressure pump pressure control commands, but instead increases the concentrate discharge ratio, extends the water production time, or provides a maintenance prompt for the output filter cartridge indicating insufficient water production.

[0066] Step 9: Utilize the Long Short-Term Memory (LSTM) network water usage prediction model to output the water usage probability for the next control period. Water usage time period data and user water consumption data are arranged into a time-series input matrix at fixed time steps. The time step can be set to 1 hour. A single training sample can include hourly user water consumption, hour number, weekday identifier, and non-weekday identifier for the previous 7 or 14 days. User water consumption is normalized according to the rated water supply capacity of the water purification system.

[0067] The Long Short-Term Memory (LSTM) network water usage prediction model comprises an input layer, a first LSTM hidden layer, a second LSTM hidden layer, a fully connected layer, and a probability output layer. The first LSTM hidden layer extracts intraday water usage changes, and the second LSTM hidden layer extracts cross-day water usage changes. The probability output layer uses a logistic activation function to form the water usage probability for the next control period.

[0068] Training labels are generated based on whether there are water withdrawal events exceeding the minimum water withdrawal amount in the next control period. Positive labels are assigned when a water withdrawal event exists, and negative labels are assigned when no water withdrawal event exists. When there are significant differences in water usage distribution between weekdays and non-weekdays, separate parameter groups for weekdays and non-weekdays can be saved. The edge computing controller selects the corresponding parameter group based on calendar data.

[0069] Step 10: Determine the water purification system operation mode based on the hysteresis threshold corresponding to the water usage probability and water quality level. The preparation time required for the reverse osmosis membrane treatment unit to reach the target permeate state varies depending on the water quality level. When the influent water quality is poor, membrane pressure build-up and pre-rinsing require more time; therefore, the high-level threshold is set relatively low to allow the high-efficiency water purification mode to start earlier. When the influent water quality is good, the high-level threshold is set relatively high to reduce unnecessary early startup.

[0070] When the probability of water usage reaches the high threshold corresponding to the current water quality level, the system switches to high-efficiency water purification mode in the next control cycle. When the probability of water usage is not higher than the low threshold corresponding to the current water quality level, the system switches to energy-saving standby mode in the next control cycle. When the probability of water usage is between the high and low thresholds, the current water purification system operation mode is maintained. The high threshold is always higher than the low threshold. The difference between the two thresholds is determined based on the prediction probability error, the shortest continuous water production time, and the allowed start frequency of the high-pressure pump.

[0071] Step 11: Combine the water purification system's operating mode with the compensation working pressure, concentrate discharge ratio, and pressure regulation boundary to form pump and valve control commands. In high-efficiency water purification mode, the edge computing controller outputs the compensation working pressure as a high-pressure pump pressure control command and the concentrate discharge ratio as a concentrate ratio regulating electric valve command, enabling the high-pressure pump and stopping or putting the circulating water pump into a low-power maintenance state. In energy-saving standby mode, the edge computing controller sets the high-pressure pump pressure control command to the shutdown pressure and sets the concentrate ratio regulating electric valve command to the closed state, stopping the high-pressure pump and causing the circulating water pump to operate intermittently.

[0072] The pump and valve control commands prevent the simultaneous occurrence of high-pressure pump stop commands and non-zero membrane working pressure commands in the same control cycle, and also prevent the concentrate ratio regulating electric valve from maintaining an unnecessary opening in energy-saving standby mode.

[0073] Step 12: The reverse osmosis membrane treatment unit is driven to produce permeate and concentrate. The effluent from the primary filtration unit enters the reverse osmosis membrane treatment unit via a high-pressure pump. A pre-membrane pressure sensor collects the actual pre-membrane pressure. The high-pressure pump frequency converter adjusts the high-pressure pump speed according to the high-pressure pump pressure control command, making the actual pre-membrane pressure close to the compensation operating pressure.

[0074] The permeate outlet of the reverse osmosis membrane treatment unit is connected to a permeate storage tank or a water terminal. A permeate flow meter is installed in the permeate pipeline. The concentrate outlet is connected to a concentrate proportioning electric valve. The concentrate proportioning electric valve adjusts the concentrate pipeline flow rate according to the concentrate discharge ratio.

[0075] Step 13: Determine the recovery path based on the concentrate quality assessment results. The concentrate quality detection module is set downstream of the concentrate proportioning electric valve and upstream of the diversion point of the concentrate recovery branch and the concentrate discharge branch, so that the same concentrate sample can be used for both recovery and discharge assessment.

[0076] The turbidity recovery threshold and conductivity recovery threshold of the concentrate are calibrated based on the allowable influent turbidity and conductivity of the primary filtration unit, as well as the mixing ratio of concentrate to raw water. When the concentrate turbidity and conductivity are not higher than the concentrate turbidity recovery threshold and the concentrate conductivity recovery threshold, the edge computing controller opens the concentrate recovery electric valve and pressure recovery pump, and closes the concentrate discharge electric valve. The concentrate enters the inlet of the primary filtration unit through the check valve and mixes with the raw water.

[0077] When the turbidity of the concentrate is higher than the turbidity recovery threshold or the conductivity of the concentrate is higher than the conductivity recovery threshold, the edge computing controller closes the electric valve for concentrate recovery and the pressure recovery pump, and opens the electric valve for concentrate discharge, allowing the concentrate to enter the wastewater discharge end.

[0078] The concentrate recovery ratio is the ratio of the concentrate flow rate returning to the inlet of the primary filtration unit within a unit control cycle to the total concentrate flow rate generated by the reverse osmosis membrane treatment unit. The concentrate recovery electric valve can be a concentrate recovery electric proportional valve, and the pressure recovery pump can be a variable frequency pump. The edge computing controller changes the concentrate flow rate returning to the inlet of the primary filtration unit within a unit control cycle by adjusting the opening degree of the concentrate recovery electric proportional valve or the speed of the pressure recovery pump.

[0079] Step 14: Adjust the control command for the next control cycle based on the actual permeate flow rate of the reverse osmosis membrane treatment unit and the concentrate quality assessment results. In high-efficiency water purification mode, the edge computing controller reads the target permeate flow rate from the water quality grade parameter table and collects the actual permeate flow rate of the reverse osmosis membrane treatment unit. The difference between the target permeate flow rate and the actual permeate flow rate of the reverse osmosis membrane treatment unit is used to determine the correction direction of the compensation working pressure. When the actual permeate flow rate of the reverse osmosis membrane treatment unit is lower than the target permeate flow rate, the pressure correction direction is to increase. When the actual permeate flow rate of the reverse osmosis membrane treatment unit is higher than the target permeate flow rate, the pressure correction direction is to decrease.

[0080] When the actual permeate flow rate of the reverse osmosis membrane treatment unit is lower than the target permeate flow rate, and the concentrate turbidity and conductivity are within limits, the feedback correction module generates a positive pressure correction based on the relative deviation of the permeate flow rate and the pressure correction step size, allowing an increase in the compensation working pressure for the next control cycle. When the concentrate turbidity or conductivity exceeds the limit, the feedback correction module sets the positive pressure correction to zero, allowing only the compensation working pressure to be maintained or reduced, and simultaneously reducing the concentrate recovery ratio. When the actual permeate flow rate of the reverse osmosis membrane treatment unit is higher than the target permeate flow rate, the feedback correction module generates a negative pressure correction based on the relative deviation of the permeate flow rate and the pressure correction step size to reduce the energy consumption of the high-pressure pump.

[0081] The feedback correction module generates a combined stabilizing factor based on permeate flow rate deviation, concentrate conductivity exceeding the limit, and concentrate turbidity exceeding the limit. If the combined stabilizing factor does not exceed the combined termination threshold for two or three consecutive control cycles, the feedback correction module stops the current pressure iteration and maintains the current compensation working pressure. If the combined stabilizing factor does not meet the termination condition, it enters the next control cycle to continue correction.

[0082] In one calculation example, the current water quality level is Class II, the basic operating pressure is 0.80 MPa, the cumulative treated water volume compensation pressure is 0.10 MPa, the cumulative usage time compensation pressure is 0.05 MPa, and the compensation operating pressure is 0.95 MPa. The pressure range corresponding to Class II water quality is 0.75 MPa to 1.05 MPa, therefore 0.95 MPa is within the allowable range. If the Long Short-Term Memory network water usage prediction model outputs a water usage probability of 0.72 for the next hour, and the high threshold corresponding to Class II water quality is 0.70, then the water purification system enters the high-efficiency water purification mode. The edge computing controller outputs a high-pressure pump pressure control command of 0.95 MPa and a concentrated water ratio adjustment electric valve command corresponding to a concentrated water discharge ratio of 25%.

[0083] Continuing with the aforementioned control cycle as an example, if the target permeate flow rate is 1.0 L / min, the actual permeate flow rate of the reverse osmosis membrane treatment unit is 0.9 L / min, the relative deviation of the permeate flow rate is 0.1, the pressure correction step size is 0.2 MPa, and the concentrate turbidity and concentrate conductivity are within limits, then the compensation working pressure in the next control cycle will increase by 0.02 MPa and be updated to 0.97 MPa. If the concentrate turbidity or concentrate conductivity exceeds the limit, the feedback correction module will not execute the positive pressure increase of 0.02 MPa, and the compensation working pressure in the next control cycle will remain at 0.95 MPa or decrease according to the negative deviation, while simultaneously reducing the concentrate recovery ratio.

[0084] Step 15: Implement anomaly protection. Set safe change rate thresholds for influent turbidity, total dissolved solids, and residual chlorine content. When the change per unit time in adjacent sampling periods exceeds the safe change rate threshold, the edge computing controller, without waiting for the deep neural network water quality assessment model to complete routine grading, directly shuts down the high-pressure pump, closes the concentrate recovery electric valve, and opens the concentrate discharge electric valve or bypass discharge electric valve.

[0085] After the influent water quality characteristics return to normal, the edge computing controller does not immediately resume concentrate recirculation. The edge computing controller requires that multiple consecutive sampling cycles pass range checks, continuity checks, and rate of change checks before recalculating the dimensionality-reduced feature vector and outputting the water quality grade.

[0086] When the permeate flow meter fails, the feedback correction module stops pressure iteration, and the control vector generation module uses the most recent effective compensation working pressure. When the concentrate turbidity sensor or concentrate conductivity sensor fails, the concentrate recovery determination module forcibly outputs a discharge control command. After communication is restored, newly acquired data must undergo a validity check before entering the feedback chain.

[0087] In this embodiment, the water quality grade controls both the pressure and concentrate ratio, as well as the high and low thresholds used in the water usage prediction results of the Long Short-Term Memory (LSTM) network. The water purification system operation mode formed by the LTM network's water usage prediction results then determines whether the pressure and concentrate ratio corresponding to the water quality grade are issued.

[0088] Example 2: This example further defines the sensors, pipelines, edge computing controller and feedback correction structure of the water purification system based on Example 1.

[0089] The raw water pipeline and the concentrate recovery pipeline converge at the mixing pipe section, which sequentially connects to the influent water quality sensor array, the primary filtration unit, the high-pressure pump, and the reverse osmosis membrane treatment unit. A sampling buffer chamber is installed between the influent water quality sensor array and the primary filtration unit. The sampling buffer chamber is used to reduce the impact of water hammer on the turbidity and pH sensors.

[0090] The high-pressure pump is a DC brushless inverter pump with a pressure closed-loop interface. A pre-membrane pressure sensor is located between the high-pressure pump outlet and the inlet of the reverse osmosis membrane treatment unit. The high-pressure pump inverter driver receives the target pressure value and enable signal output from the edge computing controller and utilizes the pre-membrane pressure sensor to complete the bottom-layer speed closed-loop.

[0091] The permeate outlet of the reverse osmosis membrane treatment unit is connected in sequence to a permeate flow meter, a check valve, and a permeate storage tank. The user's water intake flow meter is located downstream of the permeate storage tank at the user's water intake end. The permeate storage tank is equipped with a level sensor. The level sensor is used for full-water protection, water-shortage protection, and for recalculating user water intake data when no user's water intake flow meter is configured; it does not replace the long short-term memory network water usage prediction model.

[0092] The concentrate outlet of the reverse osmosis membrane treatment unit is sequentially connected to a concentrate ratio regulating electric valve, a concentrate quality detection module, and a three-way diverter. The first outlet of the three-way diverter is connected to the mixing water section via a concentrate recovery electric valve, a pressure recovery pump, and a check valve. The second outlet of the three-way diverter is connected to the wastewater discharge end via a concentrate discharge electric valve. A mixed water conductivity sensor is installed in the mixing water section to detect the conductivity of the recovered concentrate after mixing with the raw water.

[0093] The pressure recovery pump only starts when the concentrate quality meets the recovery conditions, the inlet pressure of the primary filtration unit allows backflow, and the high-pressure pump is in high-efficiency purification mode. A check valve is installed at the outlet of the pressure recovery pump to prevent raw water from flowing back into the concentrate recovery branch. When a variable frequency pump is used, the edge computing controller adjusts the concentrate recovery ratio by regulating the pump's speed. When an electric proportional valve is used for the concentrate recovery, the edge computing controller adjusts the valve's opening to change the concentrate recovery ratio.

[0094] The edge computing controller uses an embedded control board with a processor, non-volatile memory, analog input interface, digital input interface, pulse counting interface, pulse width modulation output interface, and communication interface. The deep neural network water quality assessment model, the long short-term memory network water use prediction model, the water quality grade parameter table, and the most recently effective control state are stored in the non-volatile memory.

[0095] Model inference is performed locally on the edge computing controller. Network outages do not affect water quality classification and water usage forecasting. The edge computing controller can periodically upload operational data, but this upload function is not involved in the core control chain.

[0096] The feed water quality parameters are assessed every 5 minutes. User water intake probability is updated every hour. Membrane inlet pressure, actual permeate flow rate of the reverse osmosis membrane treatment unit, concentrate turbidity, and concentrate conductivity are updated every 5 seconds. The feedback correction control cycle is set to 30 seconds. Data at different frequencies are accompanied by the acquisition time and valid status.

[0097] The concentrated water recovery electric valve and the concentrated water discharge electric valve are interlocked. Before the concentrated water recovery electric valve is opened, the concentrated water discharge electric valve remains open. After the pressure recovery pump inlet pressure stabilizes, the concentrated water discharge electric valve is closed. If the concentrated water recovery conditions fail, the concentrated water discharge electric valve is opened first, and then the concentrated water recovery electric valve and the pressure recovery pump are closed to prevent the concentrated water pipeline from being instantly shut down.

[0098] Example 3: This example is based on Example 1 and applied to a commercial direct drinking water station, as detailed below:

[0099] The commercial drinking water station is equipped with one primary filtration unit, two parallel reverse osmosis membrane treatment units, one product water storage tank, and multiple water terminals. Each water terminal is equipped with a user flow meter. The flow rates output from these multiple flow meters are aggregated using a unified timestamp to generate user water consumption data for each 15-minute period. A long short-term memory network (LSTM) water consumption prediction model uses the user water consumption data from the previous 28 days to predict the probability of water consumption in multiple future 15-minute time periods.

[0100] The weekday parameter group and the non-weekday parameter group are trained separately. The weekday parameter group is loaded when the calendar data identifies a weekday, and the non-weekday parameter group is loaded when the calendar data identifies a non-weekday. When schools, office buildings, or industrial parks have fixed work schedules, shift identifiers can also be used as input features.

[0101] Each of the two reverse osmosis membrane treatment units is equipped with a pre-membrane pressure sensor, a permeate flow meter, and a concentrate flow meter. The control vector generation module determines whether to activate one or both reverse osmosis membrane treatment units simultaneously based on the predicted water usage probability. The two reverse osmosis membrane treatment units share the same water quality grade but generate separate permeate flow deviations and pressure corrections.

[0102] The concentrate recovery branch merges into the mixing water section. A mixing water conductivity sensor is installed in the mixing water section. If the concentrate itself meets the recovery conditions but the mixing water conductivity exceeds the upper limit, the feedback correction module reduces the concentrate recovery ratio, without directly shutting down all recovery. If the mixing water conductivity remains below the mixing water conductivity recovery threshold for several consecutive control cycles, the feedback correction module allows for a gradual recovery of the concentrate recovery ratio.

[0103] During periods of low water usage, the high-pressure pump stops, the circulating water pump operates intermittently, and the sensor array maintains low-frequency sampling. When the predicted water usage probability approaches a high threshold, the influent water quality sensor array resumes its normal sampling frequency. Once the water usage probability reaches the high threshold, the high-pressure pump and the concentrate water quality detection module resume full operation. This creates a tiered wake-up mechanism for the edge computing controller, sensors, circulating water pump, and high-pressure pump.

[0104] This embodiment combines demand forecasting from multiple water terminals with start / stop selection of multiple reverse osmosis membrane treatment units, and uses the conductivity of mixed water to impose a secondary constraint on the concentrate recovery ratio, making it suitable for commercial direct drinking water scenarios with large fluctuations in water load.

[0105] This invention can be implemented using existing influent water quality sensor arrays, user water intake flow meters, product water flow meters, concentrate water quality detection modules, mixed water conductivity sensors, edge computing controllers, high-pressure pumps, concentrate proportioning electric valves, concentrate recovery electric valves, concentrate discharge electric valves, pressure recovery pumps, and bypass discharge electric valves. Deep neural network water quality assessment models and long short-term memory network water usage prediction models can be pre-trained and deployed on the edge computing controller. During system operation, the edge computing controller outputs pump and valve control commands based on sensor data, local model inference results, and feedback correction results, without relying on real-time cloud computing.

[0106] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0107] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An AI-based water purification system recycling and energy-saving method, characterized in that, include: Collect influent water quality characteristic parameters, water usage time data, and user water intake data; perform scale unification and feature dimensionality reduction processing on the influent water quality characteristic parameters to obtain dimensionality-reduced feature vectors. The reduced feature vector is input into a deep neural network water quality assessment model to obtain the water quality level, and the basic working pressure, concentrate discharge ratio, pressure regulation boundary and water use probability hysteresis threshold of the reverse osmosis membrane treatment unit are generated based on the water quality level. The water usage time period data and the user water consumption data are input into the long short-term memory network water usage prediction model to obtain the water usage probability of the next control period. The water usage probability is then hysteresis-determined based on the water usage probability hysteresis threshold to obtain the water purification system operation mode. The water purification system operation mode is coupled with the basic working pressure, the concentrate discharge ratio and the pressure regulation boundary to form high-pressure pump pressure control command, concentrate ratio regulation electric valve command and circulating water pump command. The reverse osmosis membrane treatment unit is driven to generate permeate and concentrate according to the pressure control command of the high-pressure pump and the command of the concentrate ratio adjustment electric valve. The actual permeate flow rate and concentrate quality judgment result of the reverse osmosis membrane treatment unit are collected. The concentrate that meets the recovery conditions is transported to the primary filtration unit. The actual permeate flow rate and concentrate quality judgment result of the reverse osmosis membrane treatment unit are used to constrain the pressure control command of the high-pressure pump and the command of the concentrate ratio adjustment electric valve in the next control cycle.

2. The AI-based water purification system recycling and energy-saving method according to claim 1, characterized in that, The influent water quality characteristic parameters include influent turbidity, influent hardness, influent pH value, total dissolved solids, and residual chlorine content. The scale unification and feature dimensionality reduction processing includes normalizing the minimum and maximum values ​​of the influent water quality characteristic parameters, determining the principal component projection matrix using the covariance matrix of the training samples, and projecting the normalized influent water quality characteristic parameters onto the principal component projection matrix to form a dimensionality-reduced feature vector composed of multiple principal components.

3. The AI-based water purification system recycling and energy-saving method according to claim 2, characterized in that, The deep neural network water quality assessment model includes an input layer, multiple fully connected hidden layers connected in sequence, and a water quality level output layer. The water quality level output layer outputs the classification probability of each water quality level and takes the water quality level with the highest classification probability as the water quality level of the current control cycle. The training labels of the water quality levels are determined based on the target permeate water quality, the allowable concentrate discharge ratio, the target permeate flow rate of the reverse osmosis membrane treatment unit, and the reverse osmosis membrane fouling risk level.

4. The energy-saving method for recycling a water purification system based on AI according to claim 1, characterized in that, The base working pressure is compensated based on the cumulative water volume and cumulative usage time of the reverse osmosis membrane treatment unit to obtain a compensated working pressure, and the compensated working pressure is limited within the pressure regulation boundary corresponding to the water quality level; when the compensated working pressure reaches the upper limit of the pressure regulation boundary, it is prohibited to continue to increase the high-pressure pump pressure control command.

5. The AI-based water purification system recycling and energy-saving method according to claim 1, characterized in that, The water usage probability hysteresis threshold includes a high threshold and a low threshold; when the water usage probability reaches the high threshold, the water purification system operation mode switches to high-efficiency water purification mode; when the water usage probability is not higher than the low threshold, the water purification system operation mode switches to energy-saving standby mode; when the water usage probability is between the high threshold and the low threshold, the current water purification system operation mode is maintained.

6. The AI-based water purification system recycling and energy-saving method according to claim 5, characterized in that, In the high-efficiency water purification mode, the compensation working pressure is used as the pressure control command for the high-pressure pump, and the concentrate discharge ratio is used as the command for the concentrate ratio regulating electric valve, enabling the high-pressure pump and stopping the circulating water pump or putting it into a low-power maintenance state; in the energy-saving standby mode, the pressure control command for the high-pressure pump is set to the shutdown pressure, and the command for the concentrate ratio regulating electric valve is set to the closed state, stopping the high-pressure pump and making the circulating water pump run according to the intermittent cycle.

7. The AI-based water purification system recycling and energy-saving method according to claim 1, characterized in that, The concentrate discharge ratio is the ratio of the concentrate flow rate discharged through the concentrate discharge pipeline to the influent flow rate entering the reverse osmosis membrane treatment unit within a unit control cycle; the concentrate water quality determination result is generated based on the turbidity detection unit and conductivity detection unit installed in the concentrate discharge pipeline. When the turbidity of the concentrate is not higher than the turbidity recovery threshold and the conductivity of the concentrate is not higher than the conductivity recovery threshold, the concentrate recovery electric valve and the pressure recovery pump are opened to deliver the concentrate to the inlet of the primary filtration unit; when the turbidity of the concentrate is higher than the turbidity recovery threshold or the conductivity of the concentrate is higher than the conductivity recovery threshold, the concentrate recovery electric valve is closed and the concentrate discharge electric valve is opened.

8. The AI-based water purification system recycling and energy-saving method according to claim 4, characterized in that, In the high-efficiency water purification mode, the target permeate flow rate, the actual permeate flow rate of the reverse osmosis membrane treatment unit, the concentrate turbidity, and the concentrate conductivity are collected. The correction direction of the compensation working pressure is determined based on the deviation between the target permeate flow rate and the actual permeate flow rate of the reverse osmosis membrane treatment unit. The extent to which the compensation working pressure or the concentrate recovery ratio is allowed is determined based on the excess state of the concentrate turbidity and the concentrate conductivity relative to the recovery threshold. The corrected compensation working pressure is then limited to the pressure regulation boundary corresponding to the water quality level.

9. The AI-based water purification system recycling and energy-saving method according to claim 8, characterized in that, When the actual permeate flow rate of the reverse osmosis membrane treatment unit is lower than the target permeate flow rate and the concentrate turbidity and concentrate conductivity are within limits, the compensation working pressure of the next control cycle is increased. When the actual permeate flow rate of the reverse osmosis membrane treatment unit is lower than the target permeate flow rate and the concentrate turbidity or concentrate conductivity exceeds the limit, it is prohibited to increase the compensation working pressure of the next control cycle and reduce the concentrate recovery ratio; when the actual permeate flow rate of the reverse osmosis membrane treatment unit is higher than the target permeate flow rate, the compensation working pressure of the next control cycle is reduced.

10. An AI-based water purification system with recycling and energy-saving features, characterized in that: It includes an influent water quality sensor array, a user water intake flow meter, a product water flow meter, a concentrate water quality detection module, a mixed water conductivity sensor, a water usage data acquisition module, a signal conditioning and analog-to-digital conversion module, an edge computing controller, a primary filtration unit, and a reverse osmosis membrane treatment unit. The inlet water quality sensor array is located downstream of the mixing pipe section and upstream of the primary filtration unit, and outputs the inlet water quality detection signal to the signal conditioning and analog-to-digital conversion module. The user water intake flow meter is installed at the user's water intake end and outputs the user water intake flow signal to the water usage data acquisition module; The permeate flow meter is installed in the permeate pipeline of the reverse osmosis membrane treatment unit and outputs the permeate flow signal to the signal conditioning and analog-to-digital conversion module; The concentrate water quality detection module is installed in the concentrate discharge pipeline of the reverse osmosis membrane treatment unit and outputs concentrate turbidity signal and concentrate conductivity signal to the signal conditioning and analog-to-digital conversion module. The mixed water conductivity sensor is installed in the mixed water pipe section after the concentrated water recovery pipeline and the raw water pipeline merge, and outputs the mixed water conductivity signal to the signal conditioning and analog-to-digital conversion module. The edge computing controller includes a water quality feature processing module, a deep neural network water quality assessment module, a long short-term memory network water usage prediction module, an operation mode determination module, a control vector generation module, a concentrate recovery determination module, and a feedback correction module, used to implement the AI-based water purification system recycling energy-saving method as described in any one of claims 1 to 9.

Images