Electrolytic disinfection control method based on artificial intelligence

Abstract

The application discloses an electrolytic disinfection control method based on artificial intelligence and relates to the technical field of electrolytic control. The application establishes a conveying distribution core of an electrolytic tank outlet and two detection sections downstream thereof to map the effective chlorine mass flow history of a first detection section into the non-decay reference effective chlorine mass flow of a second detection section. According to the Faraday theory, the effective chlorine mass flow of the first detection section is used to determine the side electron equivalent conversion loss mass flow generated, and the non-decay reference and the effective chlorine mass flow of the second detection section are used to determine the effective chlorine loss mass flow maintained. A double-loss attribution neural network generates a candidate control amount. The sources and evaluation of the candidate control amount to the effective chlorine improvement amount of the second detection section are guided by the counterfactual prediction decomposition of the candidate control amount through the generation path freezing and the maintenance path freezing. The application can distinguish whether the effective chlorine deficiency occurs in the generation stage or the post-tank maintenance stage under different working conditions and is controlled in a targeted manner.

Description

Technical Field

[0001] This invention relates to the field of electrolytic disinfection control technology, and in particular to an electrolytic disinfection control method based on artificial intelligence. Background Technology

[0002] The on-site electrolysis of brine to produce sodium hypochlorite typically utilizes diaphragm-less or diaphragm-based electrolysis units to convert sodium chloride solution into a disinfectant containing available chlorine. Available chlorine yield is not only related to the electrolysis current but is also affected by salt concentration, liquid flow rate, temperature, pH, electrode surface condition, and residence time in the electrolyzer. When the current density is too high, chloride ion mass transfer is insufficient, or the temperature rises, a portion of the input charge may enter non-target reactions such as oxygen evolution, further oxidation of chlorine-containing species, and cathode reduction, causing the actual available chlorine output from the electrolyzer to be lower than the theoretical yield estimated according to Faraday's law.

[0003] Even after being generated in the electrolyzer, available chlorine may still decay in pipelines, buffer containers, circulation sections, or before addition. This decay is related to temperature, pH, available chlorine concentration, and liquid residence time. When insufficient available chlorine is detected downstream, there are at least two different physical reasons: first, available chlorine was not fully generated in the electrolyzer according to the theoretical electron equivalent; second, available chlorine was formed at the electrolyzer outlet but was lost during subsequent transport and storage. The appropriate control actions for these two reasons differ.

[0004] Existing feedback control typically adjusts the electrolysis current, brine flow rate, or dilution water flow rate based on the effective chlorine concentration at the generator outlet or a single downstream point. Neural networks and model predictive control can improve nonlinear prediction capabilities, but if the network directly learns the mapping between "operating data—effective chlorine—control quantity," it may still interpret the retention loss downstream as insufficient chlorine production on the generator side. In this case, increasing the electrolysis current will not eliminate the retention loss; instead, it may increase side reactions, power consumption, and temperature rise.

[0005] Using two detection locations provides a basis for spatial loss separation, but aligning the measurements from both locations using only a single delivery time with a delay can still introduce significant errors when buffer containers, backflow, branch switching, or rapid flow changes are present. The liquid actually reaching the second detection section typically contains a mixture of components of different ages, and the degree of age dispersion corresponding to the same average delivery time may also differ. If this distribution characteristic is ignored, the transient differences caused by mixing during delivery will be incorrectly included in the available chlorine loss on the holding side.

[0006] Furthermore, measurement noise, response delay, and calibration bias of the two sets of available chlorine analyzers and flow meters can also generate spurious differences. Therefore, based on the two physical loss boundaries, the following issues need to be addressed: first, how to establish a non-attenuating reference for the second detection section under variable flow and mixed transport conditions; second, how to prevent measurement uncertainty from being mistaken for the true loss; third, how to ensure that the two types of losses only update the corresponding physical latent states; and fourth, how to truly constrain the loss attribution results to candidate control actions, rather than merely using them as additional inputs to the neural network. Summary of the Invention

[0007] In order to solve the above-mentioned technical problems, or at least partially solve the above-mentioned technical problems, and in view of the above-mentioned shortcomings, the present invention provides an electrolytic disinfection control method based on artificial intelligence.

[0008] In a first aspect, the present invention provides an electrolytic disinfection control method based on artificial intelligence, comprising: The electrolysis current, electrolysis voltage, brine flow rate, dilution water flow rate, salt concentration or conductivity corresponding to the salt concentration, temperature and pH of the sodium hypochlorite electrolysis generating unit are collected. The first available chlorine detection amount at the first detection section at the outlet of the electrolytic cell and the second available chlorine detection amount at the second detection section downstream of the first detection section are obtained, and the target available chlorine mass flow rate corresponding to the disinfection object is obtained. Construct or update the transport distribution kernel between the first and second detection sections, use the transport distribution kernel to perform a weighted mapping on the historical sequence of the effective chlorine mass flow rate of the first detection section to obtain the attenuation-free reference effective chlorine mass flow rate of the second detection section, and calculate the normalized deficit rate on the generating side and the normalized deficit rate on the maintaining side. The generation side electron equivalent depletion mass flow rate is determined based on the difference between the effective chlorine mass flow rate corresponding to the current electrolysis current and the effective chlorine mass flow rate of the first detection section. The maintenance side effective chlorine depletion mass flow rate is determined based on the difference between the non-attenuated reference effective chlorine mass flow rate of the second detection section and the effective chlorine mass flow rate of the second detection section. Candidate control variables are generated using a dual-loss attribution neural network. The network generates isolated-corrected latent states for generation efficiency and available chlorine retention using the generation-side and retention-side state sequences. Based on the latent states for generation efficiency, an allocation ratio is generated to model the generation-side reaction flux, resulting in the predicted available chlorine mass flow rate for the first detection cross-section. A non-negative available chlorine decay rate is determined based on the available chlorine retention latent state, and a weighted summation of the predicted available chlorine mass flow rates for different liquid ages at the first detection cross-section is performed according to the transport distribution kernel to obtain the predicted available chlorine mass flow rate for the second detection cross-section. The dual-loss attribution neural network generates multiple sets of candidate control variables based on the current latent states for generation efficiency, current latent states for available chlorine retention, target available chlorine mass flow rate, transport distribution kernel, and dominant loss type. Each set of candidate control variables includes candidate electrolysis current, candidate brine flow rate, and candidate dilution water flow rate. For each group of candidate control quantities, generation path freezing counterfactual prediction and hold path freezing counterfactual prediction are performed respectively to decompose the generation side contribution and hold side contribution of the candidate control quantity to the improvement of the predicted effective chlorine mass flow rate of the second detection section and screen the candidate control quantity. When the hold side loss is the dominant loss type, the candidate control quantity whose hold side contribution does not meet the preset requirements and mainly relies on increasing the candidate electrolysis current to increase the generation side contribution is restricted. Based on the deviation between the predicted effective chlorine mass flow rate and the target effective chlorine mass flow rate of the second detection section, the defective mass flow rate converted from the electron equivalent on the generation side, the defective mass flow rate of effective chlorine on the maintenance side, the power consumption, the degree of side reaction, and the magnitude of the change in the control quantity, the candidate control quantity after being screened by counterfactual prediction is evaluated, the target candidate control quantity is selected, and the operation of the rectifier power supply, brine supply actuator, and dilution water supply actuator is controlled under the conditions of satisfying the constraints of electrolyzer temperature, current density, allowable amount of side reaction, and rate of change of control quantity.

[0009] Furthermore, constructing the delivery distribution kernel includes: establishing a discrete material transfer model based on the effective liquid volume and flow history between the first and second detection sections; recursively updating the liquid inventory corresponding to different liquid ages between the first and second detection sections according to the control cycle to determine the inventory ratio of each liquid age component in the current control cycle; determining the probability weight of the corresponding liquid age output between the first and second detection sections based on the inventory ratio of each liquid age component, and constructing the delivery distribution kernel for the current control cycle from the probability weights corresponding to each liquid age; when the delivery time between the two detection sections is approximately a single, fixed delivery time, the delivery distribution kernel degenerates into a single-point delay kernel.

[0010] Furthermore, constructing the transport distribution kernel includes: obtaining a baseline residence time distribution through tracer experiments; determining a time scaling factor based on real-time flow rate and real-time liquid level; scaling the time axis of the baseline residence time distribution using the time scaling factor; determining a weight correction function corresponding to each residence time segment based on real-time flow rate and real-time liquid level; and performing weighted and normalized processing on the scaled residence time distribution to obtain the transport distribution kernel. When the two detection sections are approximately equal to a single, fixed transport time, the transport distribution kernel degenerates into a single-point delay kernel.

[0011] Furthermore, the calculation process for the generation-side electron equivalent deducted mass flow rate and the generation-side electron equivalent deducted mass flow rate includes: According to the electrolysis current Calculate the effective chlorine mass flow rate according to Faraday theory : Among them, electrolysis current For the first Electrolysis current for each control cycle The molar mass used when calculating the effective chlorine mass based on chlorine gas equivalents. It is Faraday's constant; According to Faraday's theory, the effective chlorine mass flow rate With the effective chlorine mass flow rate of the first detection section The difference between them determines the generation side electron equivalent reduced loss mass flow rate. : ; Based on the second detection section without attenuation, the reference effective chlorine mass flow rate With the second detection section, the effective chlorine mass flow rate The difference between them determines the available chlorine loss mass flow rate on the maintenance side. : ; in, To generate the side uncertainty tolerance, To maintain the tolerance of lateral uncertainty, it is determined based on the measurement uncertainty of the effective chlorine detection device, the flow detection device, and the current detection device, as well as the recent calibration residual.

[0012] Furthermore, the dual-loss attribution neural network includes a generator-side state encoding branch, a hold-side state encoding branch, an isolation correction and attribution layer, a response flux calculation layer, and a control decoding layer; The generation-side state encoding branch outputs the generation efficiency hidden state based on the generation-side state sequence, and the hold-side state encoding branch outputs the effective chlorine hold-side hidden state based on the hold-side state sequence. The isolation correction and attribution layer uses the generation-side electronic equivalent depleted mass flow rate and the hold-side effective chlorine depleted mass flow rate to correct the efficiency hidden state and the effective chlorine hold-side hidden state through correction interfaces, respectively. Among them, the two correction interfaces adopt gated residual units, and the gate coefficient of the gated residual unit is determined by the state of the corresponding branch and the corresponding normalized deficit rate. The generation-side normalized deficit rate is determined by the ratio of the generation-side electronic equivalent depleted mass flow rate to the Faraday theoretical effective chlorine mass flow rate, and the hold-side normalized deficit rate is determined by the ratio of the hold-side effective chlorine depleted mass flow rate to the unattenuated reference effective chlorine mass flow rate of the second detection section. The reaction flux calculation layer determines the main chlorine production flux and at least one side reaction flux based on the hidden state generation efficiency distribution ratio and the current electrolysis current. It also uses the mass transfer allowable flux determined based on salt concentration, total liquid flow rate and electrolyzer residence time to limit the main chlorine production flux, thus obtaining the predicted effective chlorine mass flow rate of the first detection section. The non-negative effective chlorine decay rate is determined based on the effective chlorine remaining in a latent state. Then, based on the transport distribution, the predicted effective chlorine mass flow rate of the first detection section at different liquid ages is calculated by performing a weighted summation after decay evolution to obtain the predicted effective chlorine mass flow rate of the second detection section. The control decoding layer generates multiple sets of candidate control quantities based on the current generation efficiency hidden state, the current effective chlorine remaining hidden state, the target effective chlorine mass flow rate, the transport distribution kernel, and the dominant loss type. Each set of candidate control quantities includes candidate electrolysis current, candidate brine flow rate, and candidate dilution water flow rate.

[0013] Furthermore, the reaction flux calculation layer generates the main chlorine production reaction allocation ratio, the anode secondary oxidation allocation ratio, the cathode reduction allocation ratio, and other loss allocation ratios based on the generation efficiency hidden state, and ensures that each allocation ratio is not less than 0 and the sum is 1. Based on the unified decomposition method of electrolytic current, the Radic electron equivalent flux; The fluxes for main chlorine production, anode-side oxidation, cathode reduction, and other losses are calculated based on the distribution ratios of the main chlorine production reaction, anode-side oxidation, cathode reduction, and other losses, as well as the Faraday electron equivalent flux. Set the mass transfer allowable flux limit for the main chlorine production flux, and the actual main chlorine production flux is the smaller of the main chlorine production flux and the mass transfer allowable flux.

[0014] Furthermore, the process by which the dual-loss attribution neural network obtains the predicted effective chlorine mass flow rate for the second detection cross-section includes: Use the Softplus function or exponential function to keep available chlorine in a hidden state. Transformed into a non-negative effective chlorine decay rate, For the delay index in the delivery distribution kernel The corresponding liquid age component, based on the non-negative effective chlorine decay rate Calculate the available chlorine retention coefficient Represented as: ; Based on the delivery distribution, the effective chlorine mass flow rate is predicted by checking the first detection section at different liquid ages. After separate attenuation, the effective chlorine retention coefficient is used. The weighted summation yields the predicted available chlorine mass flow rate for the second detection section. : ; The predicted effective chlorine mass flow rate for the first detection section is obtained by converting the effective chlorine mass equivalent based on the actual main chlorine production flux and the side reaction flux.

[0015] Furthermore, a preliminary dominant loss is determined based on the normalized loss rate on the generation side and the normalized loss rate on the maintenance side, and the attribution confidence is calculated. Active perturbation identification is then performed based on the dominant loss determination results and the calculated attribution confidence. Based on the normalized loss rate of the generating side and maintain the normalized loss ratio Preliminary dominant loss determination: When the normalized loss rate on the generation side exceeds the generation side determination threshold and is higher than the normalized loss rate on the maintenance side, the generation side loss is determined to be the dominant loss type; when the normalized loss rate on the maintenance side exceeds the maintenance side determination threshold and is higher than the normalized loss rate on the generation side, the maintenance side loss is determined to be the dominant loss type. The attribution confidence is determined by the interval between the two types of normalized loss rates, the uncertainty tolerance on the generation side, the uncertainty tolerance on the maintenance side, the stability of recent judgments, and the uncertainty output by the dual-loss attribution neural network. When the attribution confidence is lower than the preset confidence threshold, or when the dominant loss type alternates within multiple consecutive control cycles, the controller enters active disturbance identification: within the range allowed by the safety constraint mapping, a current sensing disturbance with limited amplitude and duration is first applied, followed by a limited hydraulic sensing disturbance; a sensing response matrix is ​​established based on at least two of the responses of the two types of sensing disturbances to the effective chlorine mass flow rate of the first sensing section, the effective chlorine mass flow rate of the second sensing section, the effective chlorine deficit mass flow rate on the holding side, and the change in the equivalent resistance of the tank; if the current sensing response matrix has a significant effect on the first sensing section but a small effect on the deficit ratio on the holding side, the attribution confidence on the generating side is increased; if the hydraulic sensing significantly changes the transport distribution kernel and reduces the effective chlorine deficit mass flow rate on the holding side, the attribution confidence on the holding side is increased.

[0016] Furthermore, the generation path freeze counterfactual prediction maintains the effective chlorine in a hidden state and the delivery distribution kernel does not update with the candidate control quantity, allowing the generation side state and reaction flux to respond to the candidate control quantity to obtain generation side contribution; the hold path freeze counterfactual prediction maintains the generation efficiency in a hidden state and the first detection section prediction of effective chlorine mass flow rate does not update with the candidate control quantity, allowing the hold side state, delivery distribution kernel, or effective chlorine decay rate to respond to the candidate control quantity to obtain hold side contribution.

[0017] Furthermore, the overall training loss consists of the prediction error of the first detection section, the prediction error of the second detection section, the electron equivalent conservation residual, the transport distribution consistency residual, the age distribution maintenance evolution residual, the cross-sensitivity penalty, and the control smoothing term. The transport distribution consistency residual is used to constrain the transport distribution kernel to be consistent with the flow rate, liquid level, and tracer calibration results. The cross-sensitivity penalty term is used to reduce the direct sensitivity of the latent state of generation efficiency to the normalized deficit rate on the maintenance side, and to reduce the direct sensitivity of the latent state of available chlorine maintenance to the normalized deficit rate on the generation side.

[0018] Secondly, the present invention provides an electrolytic disinfection control device based on artificial intelligence, comprising: at least one processing unit, the processing unit being connected to a storage unit via a bus unit, the storage unit being used for a computer-executable program, and the processing unit implementing the aforementioned electrolytic disinfection control method based on artificial intelligence by running the computer-executable program stored in the storage unit.

[0019] The technical solutions provided in the embodiments of the present invention have the following advantages compared with the prior art: This invention utilizes a first and a second detection cross-section to form two spatial loss boundaries, and forms a non-attenuation reference for the second detection cross-section through a delivery distribution kernel, thus decoupling the delivery timing broadening from the residual of actual effective chlorine decay. This invention uses generation-side uncertainty tolerance and maintenance-side uncertainty tolerance to suppress spurious losses caused by sensor noise, response delay, and calibration bias, improving long-term online attribution stability. The isolation correction and attribution layer of this invention restricts generation-side and maintenance-side losses to different data paths, and can combine cross-sensitivity penalties to reduce hidden state cross-interpretation, enhancing network physical interpretability. This invention uses the delivery distribution kernel and non-negative effective chlorine decay rate to jointly execute age distribution maintenance evolution, distinguishing between buffer and mixed operating conditions with the same average delivery time but different liquid age dispersion. The dual-path counterfactual contribution evaluation of this invention decomposes the improvement of downstream effective chlorine by candidate control quantities into generation-side and maintenance-side contributions, constraining loss attribution candidate actions and reducing ineffective current surges under maintenance-side loss conditions. When the attribution confidence is insufficient, restricted active identification and detection are adopted, and the cause determination is updated by using the response of different excitation paths, thereby reducing misattribution when the two types of losses are close or the operating conditions switch. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating an artificial intelligence-based electrolytic disinfection control method provided in an embodiment of the present invention; Figure 2 A flowchart illustrating the construction process of a delivery distribution core is provided in an embodiment of the present invention. Figure 3 A flowchart illustrating another method for constructing a delivery distribution core, as provided in an embodiment of the present invention. Figure 4 This is an architecture diagram of the dual-loss attribution neural network provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of an artificial intelligence-based electrolytic disinfection control device provided in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] It should be noted that, in this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0025] Example 1 like Figure 1 As shown, this invention provides an artificial intelligence-based electrolytic disinfection control method, which specifically includes the following steps: S100 collects the electrolysis current, electrolysis voltage, brine flow rate, dilution water flow rate, salt concentration or the conductivity corresponding to the salt concentration, temperature and pH of the sodium hypochlorite electrolysis generating unit, obtains the first available chlorine detection quantity at the first detection section at the outlet of the electrolytic cell and the second available chlorine detection quantity at the second detection section downstream of the first detection section, and obtains the target available chlorine mass flow rate corresponding to the disinfection object.

[0026] The equipment for preparing low-concentration sodium hypochlorite disinfectant by diaphragm-free electrolysis of brine includes: a brine supply unit, a dilution water supply unit, a mixing section, an electrolytic cell, a DC rectifier power supply, a downstream conveying pipeline, an optional buffer container, and a dosing unit. A first detection section is located at the electrolytic cell outlet or a short-distance outlet section representing the immediate output of the electrolytic cell. A second detection section is located downstream of the first detection section, preferably after the downstream conveying or temporary storage process and before the target of dosing. The first detection section is equipped with a first available chlorine detector and a first flow rate detector to obtain the first available chlorine concentration. The effective chlorine concentration of the first detection section in each control cycle and the liquid volume flow rate at the first detection section The second detection section is equipped with a second available chlorine detection device and a second flow rate detection device to obtain the second available chlorine level. The effective chlorine concentration in the second detection section of each control cycle Second detection section liquid volume flow rate The two effective chlorine detection devices at the two detection sections were calibrated using a unified effective chlorine equivalent. The first and second effective chlorine detection quantities were converted into effective chlorine mass flow rates at the first and second detection sections, respectively; wherein, the effective chlorine mass flow rate at the first detection section... for: The effective chlorine mass flow rate at the second detection section for: .

[0027] S200, acquire the effective liquid volume between the two detection sections, flow history, and at least one of valve position, liquid level, or branch flow rate, and construct or update the... The transport distribution core between the first and second detection sections of each control cycle Transport distribution core Indicates the first The liquid reaches the second detection section in one control cycle, corresponding to... The normalized contribution weight of the liquid that passed through the first detection section before each control sampling cycle. The delay index represents the discrete liquid age from the first detection section to the second detection section, and the length of the delivery distribution kernel. The maximum effective transport time covering the two detection sections.

[0028] The historical sequence of effective chlorine mass flow rate at the first detection section is weighted and mapped using the transport distribution kernel to obtain the attenuation-free reference effective chlorine mass flow rate at the second detection section. The normalized deficit rate on the generating side and the normalized deficit rate on the maintaining side are then calculated. The attenuation-free reference effective chlorine mass flow rate at the second detection section represents the theoretically expected effective chlorine mass flow rate that the second detection section should receive under the current transport mixing state, assuming no effective chlorine chemical attenuation after the first detection section.

[0029] The first and second detection sections are connected by straight pipes, mixers, buffer containers, branches, circulation loops, or combinations thereof. Different pipe structures and operating states between the first and second detection sections result in different residence time distributions and mixing states, further causing effective chlorine of different liquid ages to undergo different reaction times, thus affecting the actual effective chlorine decay at the second detection section. This invention incorporates effective chemical decay into subsequent control to better meet the disinfection requirements of subsequent disinfection targets.

[0030] A discrete material transfer model is established based on the effective liquid volume and flow history. The inventory percentage of each liquid age component between the first and second detection sections is updated periodically. A delivery distribution kernel is then formed based on the inventory percentage of each liquid age component between the first and second detection sections. Constructing the delivery distribution kernel includes: establishing a discrete material transfer model based on the effective liquid volume and flow history between the first and second detection sections; recursively updating the liquid inventory corresponding to different liquid ages between the first and second detection sections according to the control cycle to determine the inventory percentage of each liquid age component in the current control cycle; determining the probability weight of the corresponding liquid age output between the first and second detection sections based on the inventory percentage of each liquid age component, and constructing the delivery distribution kernel for the current control cycle from the probability weights corresponding to each liquid age; when the delivery time between the two detection sections is approximately a single, fixed delivery time, the delivery distribution kernel degenerates into a single-point delay kernel. In specific implementation processes, such as... Figure 2 As shown, the process includes: A discrete material transfer model is established based on the effective liquid volume and flow history between the first and second detection sections. In the discrete material transfer model, the first... The discrete transfer ratio corresponding to each control cycle is expressed as: ; in, Let be the discrete transfer ratio, representing the th The proportion of liquid transferred downstream between the first and second detection sections within a control cycle to the effective liquid inventory. For the first The effective liquid volume between the first and second detection sections within each control cycle; For the first The average flow rate between the first and second detection sections within one control cycle. To control the cycle length.

[0031] The liquid inventory corresponding to different liquid age components between the first and second detection sections is recursively updated according to the control cycle. Specifically, the liquid between the first and second detection sections is divided into discrete liquid ages. Each liquid age component is updated using the discrete transfer ratio, based on the relationship between the liquid age increasing with the control cycle between the first and second detection sections. Liquid age corresponds to the liquid inventory level: , ;in, Indicates the first , The delay index at the start of each control cycle is: The liquid age component corresponds to the liquid inventory. The inventory corresponding to the liquid age component with k=0 is determined by the liquid newly entering between the first and second detection sections in the current control cycle, and is expressed as: ;in, Indicates the first The inlet flow rate enters between the first and second detection sections during each control cycle. The updated inventory levels for each liquid age are volumetrically normalized.

[0032] Determine the inventory percentage of each liquid age between the first and second detection sections based on its inventory level: ; in, Indicates the first The percentage of liquid age in the total effective liquid inventory between the first and second detection sections, where the control cycle delay index is k.

[0033] Based on the inventory percentage of each liquid age between the first and second detection sections, the relative probability of different liquid ages leaving the first and second detection sections with the current outflow is determined, thus forming a transport distribution core between the first and second detection sections: ; in, This represents the outflow selection weight corresponding to the liquid age with delay indices k and j. In one example, all liquid ages are assumed to have the same outflow selection probability; that is, the inventory percentage of each liquid age between the first and second detection sections is directly used as the transport distribution kernel, satisfying the normalization condition. The transport distribution kernel can reflect the changes in transport delay distribution caused by changes in the actual effective liquid volume and historical flow rate between the first and second detection sections. All weights in the transport distribution kernel are not less than 0, and the sum of the weights is 1.

[0034] In another implementation, a baseline residence time distribution is obtained through tracer experiments. A time scaling factor is determined based on real-time flow rate and real-time liquid level. The time axis of the baseline residence time distribution is scaled using this time scaling factor. A weight correction function corresponding to each residence time segment is determined based on real-time flow rate and real-time liquid level. The scaled residence time distribution is then weighted and normalized to obtain the delivery distribution kernel. Figure 3 As shown, the specific process includes: The baseline residence time distribution obtained through tracer experiments is as follows: Real-time traffic Real-time liquid level Flow rate corresponding to the benchmark experiment and liquid level Determine the time scaling factor: ; in, This is a correction factor for the impact of flow rate on the delivery timescale. This is a correction factor for the effect of liquid level on the transport timescale. To prevent positive numbers with a denominator of zero.

[0035] The baseline residence time distribution, after time scaling, yields an intermediate distribution that satisfies the probability normalization condition, denoted as: ; The delivery distribution kernel is obtained by weighting and normalizing the distribution of the intermediate distribution across different time segments based on real-time operating conditions, and is represented as follows: ; in, This is a weighting adjustment function for the different time segments corresponding to the delay indices k and j at the current time t. As an example: ; in, , The weighted correction strengths are for flow rate and liquid level, respectively. , These represent the sensitivity of different residence time periods to changes in flow rate and liquid level.

[0036] When the time between the two detection sections is approximately a single, fixed delivery time, the delivery distribution core degenerates into a single-point delay core. For example, in a straight pipe with approximate piston flow and slow flow rate change, the delivery distribution core... Set as a single-point delay core. The second detection section has no attenuation and references the effective chlorine mass flow rate. Effective chlorine mass flow rate at the first detection section Historical sequence according to transport distribution core Weighted summation: . S300, based on the Faraday theory effective chlorine mass flow rate corresponding to the current electrolysis current. With the effective chlorine mass flow rate of the first detection section The difference between them determines the generation side electron equivalent reduced loss mass flow rate. Based on the second detection section without attenuation, the effective chlorine mass flow rate is used as a reference. With the second detection section, the effective chlorine mass flow rate The difference between them determines the available chlorine loss mass flow rate on the maintenance side.

[0037] The controller is based on the electrolysis current. Calculate the effective chlorine mass flow rate according to Faraday theory : Among them, electrolysis current For the first Electrolysis current for each control cycle The molar mass used when calculating the effective chlorine mass based on chlorine gas equivalents. The effective chlorine mass flow rate is the Faraday constant; and is determined by the Faraday theory. The effective chlorine mass flow rate at the first detection section And generation-side uncertainty tolerance Calculate the defective mass flow rate by converting the electron equivalent on the generating side. : ; Calculation of defective mass flow rate using electron equivalent on the generation side The physical boundary ends at the first detection section.

[0038] The controller uses the second detection section as a reference for the effective chlorine mass flow rate without attenuation. The second detection section has an effective chlorine mass flow rate. and maintain the side uncertainty tolerance Calculate the available chlorine deficit mass flow rate on the hold side. : Maintain the available chlorine deficit mass flow rate on the side. Describe the effective chlorine loss after the first detection section and before the second detection section.

[0039] Generative side uncertainty tolerance The uncertainty is determined by error propagation from the current measurement uncertainty, the first flow rate measurement uncertainty, the first effective chlorine concentration measurement uncertainty, and their recent calibration residuals. Maintaining the side uncertainty tolerance. The uncertainty is determined by the transport distribution kernel estimation error, the second flow measurement uncertainty, the second available chlorine concentration measurement uncertainty, and the relative calibration residuals of the two available chlorine analyzers.

[0040] When the first and second available chlorine detectors exhibit a fixed proportional deviation under low-loss reference conditions for an extended period, they preferentially enter instrument calibration mode. The updated error statistics are then used to redetermine the uncertainty tolerance on the holding side. .

[0041] Generate normalized loss rate Defect mass flow rate calculated from the electron equivalent on the generating side The ratio of the effective chlorine mass flow rate determined by Faraday's theory is used to maintain the side-normalized deficit rate. The effective chlorine deficit mass flow rate on the maintenance side The ratio of the effective chlorine mass flow rate to the reference flow rate without attenuation at the second detection section is used to determine the accuracy.

[0042] S400 uses a dual-loss attribution neural network to generate candidate control variables.

[0043] The dual-loss attribution neural network generates isolated-corrected hidden states for generation efficiency and available chlorine retention using the generation-side state sequence and the retention-side state sequence. Based on the hidden state for generation efficiency, an allocation ratio is generated to model the generation-side reaction flux, resulting in the predicted available chlorine mass flow rate for the first detection section. A non-negative available chlorine decay rate is determined based on the available chlorine retention state, and a weighted summation of the predicted available chlorine mass flow rates for different liquid ages at the first detection section is performed according to the transport distribution kernel, resulting in the predicted available chlorine mass flow rate for the second detection section. Multiple candidate control quantities are generated based on the current hidden state for generation efficiency, the current hidden state for available chlorine retention, the target available chlorine mass flow rate, the transport distribution kernel, and the dominant loss type. Each candidate control quantity includes candidate electrolysis current, candidate brine flow rate, and candidate dilution water flow rate.

[0044] like Figure 4 As shown, the dual-loss attribution neural network includes a generator-side state encoding branch, a hold-side state encoding branch, an isolation correction and attribution layer, and a control decoding layer; The dual-loss attribution neural network utilizes a generation-side state encoding branch, a hold-side state encoding branch, and an isolation correction and attribution layer to obtain the hidden states of generation efficiency and available chlorine hold-side state with isolation correction. The generation-side state sequence, which includes at least electrolysis current, electrolysis voltage, salt concentration, brine flow rate, and changes in the tank's equivalent resistance, is input into the generation-side state encoding branch to output the hidden state of generation efficiency. The hold-side state sequence, which includes at least temperature, pH, total liquid flow rate, and liquid age statistics obtained from the delivery distribution kernel, is input into the hold-side state encoding branch to output the hidden state of available chlorine hold-side state. The isolation correction and attribution layer inputs the generation-side electron equivalent converted deficit mass flow rate into the generation-side correction interface and the hold-side available chlorine deficit mass flow rate into the hold-side correction interface to correct the hidden states of generation efficiency and available chlorine hold-side state, respectively.

[0045] In specific implementation, the generation-side state coding branch receives the generation-side state sequence and outputs the generation efficiency hidden state. The generation-side state sequence includes at least the electrolytic current. Electrolysis voltage Salt concentration or conductivity, salt flow rate And the change in the equivalent resistance of the tank. The equivalent resistance of the tank is estimated by the ratio of electrolysis voltage to electrolysis current, and preferably uses the change in relative baseline or the residual corrected for conductivity as input. The state encoding branch on the generation side adopts any one or more of the following: causal convolutional network, gated recurrent unit, long short-term memory network, or temporal attention network.

[0046] The hold-side state encoding branch receives the hold-side state sequence and outputs the effective chlorine to maintain the hidden state. The maintenance side state sequence should include at least temperature, pH, total liquid flow rate, transport distribution, and mean age of the liquid. and the dispersion of nuclear liquid age in transport distribution It may also include the effective chlorine concentration of the first detection section. The buffer container liquid level, ambient temperature, and pipeline circulation status are monitored. The state encoding branch on the maintenance side employs any one or more of the following: causal convolutional network, gated recurrent unit, long short-term memory network, or temporal attention network. The average liquid age of the delivery distribution kernel is... and liquid age dispersion The average liquid age is introduced, which is determined by the first moment and the second central moment of the transport distribution core, respectively. and liquid age dispersion To distinguish conveying states with the same average conveying time but different levels of mixing.

[0047] Isolation correction and attribution layer will generate side-normalized loss rate Modify the hidden state of generation efficiency via the generation-side correction interface The normalized loss rate will be maintained. The effective chlorine is maintained in a latent state by modifying the calibration interface on the holding side. The two correction interfaces use gated residual units, and the gate coefficients of the gated residual units are determined by the state of the corresponding branch and the corresponding normalized loss rate.

[0048] The reaction flux calculation layer generates an allocation ratio based on the implicit state of the generation efficiency to model the reaction flux on the generation side, obtaining the predicted effective chlorine mass flow rate for the first detection section. The reaction flux calculation layer generates an allocation ratio based on the implicit state of the generation efficiency, and determines the main chlorine production flux and at least one side reaction flux in conjunction with the current electrolysis current. It then uses the mass transfer allowable flux determined based on salt concentration, total liquid flow rate, and electrolyzer residence time to limit the main chlorine production flux, obtaining the predicted effective chlorine mass flow rate for the first detection section. Specifically: The reaction flux calculation layer allocates the hidden state generation ratio based on the generation efficiency.

[0049] Specifically, the reaction flux calculation layer calculates the hidden state based on the generation efficiency. Output forms the main chlorine production reaction distribution ratio Anode secondary oxidation distribution ratio Cathode reduction distribution ratio Other loss allocation ratio : ; Wherein, parameter matrix and bias vector For training parameters. Distribution ratio of the main chlorine-producing reaction. Anode secondary oxidation distribution ratio Cathode reduction distribution ratio Other loss allocation ratio All are not less than 0 and their sum is 1.

[0050] According to the electrolysis current Unified Folding Algorithm: Radial Electron Equivalent Flux .

[0051] Based on the distribution ratio of the main chlorine-producing reaction Anode secondary oxidation distribution ratio Cathode reduction distribution ratio Other loss allocation ratio and Faraday electron equivalent flux Calculate the main chlorine flux Anode auxiliary oxidation flux Cathode reduction flux Other lost flux : ; To avoid excessive chlorine flux in main production Unconstrained growth under low salt concentration or high current conditions, setting the allowable mass transfer flux. Limiting the main chlorine production flux Among them, the allowable mass transfer flux It is determined by at least two of the following: salt concentration, total liquid flow rate, electrolyzer temperature, and electrolyzer residence time.

[0052] The actual chlorine flux is the smaller of the chlorine flux produced and the allowable mass transfer flux. .

[0053] The non-negative effective chlorine decay rate is determined based on the available chlorine remaining in a latent state. Then, based on the transport distribution, the predicted effective chlorine mass flow rate for different liquid ages at the first detection cross-section is weighted and summed after decay evolution to obtain the predicted effective chlorine mass flow rate for the second detection cross-section. The specific process includes: determining the effective chlorine remaining in a latent state... Output non-negative effective chlorine decay rate To ensure an effective chlorine decay rate If the effective chlorine is not less than 0, use the Softplus function or the exponential function to keep the effective chlorine in a hidden state. Transformed into non-negative effective chlorine decay rate .

[0054] For the transport distribution core Medium-delay index The corresponding liquid age, based on the non-negative effective chlorine decay rate Calculate the available chlorine retention coefficient Represented as: ; Based on the delivery distribution, the effective chlorine mass flow rate is predicted by checking the first detection section at different liquid ages. After separate attenuation, the effective chlorine retention coefficient is used. The weighted summation yields the predicted available chlorine mass flow rate for the second detection section. : ; Specifically, the predicted effective chlorine mass flow rate for the first detection section is obtained by converting the effective chlorine mass equivalent based on the actual main chlorine production flux and side reaction flux. For equipment with circulating flow or significant in-tank inventory, a dynamic mass conservation term for the equivalent mixing volume of the electrolytic cell is added.

[0055] Control the decoding layer based on the current generation efficiency hidden state The current available chlorine remains in a hidden state. Target available chlorine mass flow rate , transport distribution core In addition, multiple sets of candidate control variables are generated based on the dominant loss type. Each set of candidate control variables includes candidate electrolysis current, candidate brine flow rate, and candidate dilution water flow rate.

[0056] Among them, based on the normalized loss rate of the generating side and maintain the normalized loss ratio Perform preliminary dominant loss discrimination, calculate attribution confidence, and identify active perturbations based on the dominant loss discrimination results and calculated attribution confidence, including: When generating the normalized loss rate Exceeding the threshold for generation and higher than the normalized loss rate for maintenance When the loss on the generating side is determined to be the dominant loss type, the loss on the maintaining side is normalized. Exceeding the holding-side threshold and higher than the generation-side normalized loss rate At that time, the hold-side loss was determined to be the dominant loss type.

[0057] Based on the interval between the two types of normalized loss rates and the uncertainty tolerance on the generation side Maintaining tolerance for lateral uncertainty The attribution confidence is jointly determined by the recent determination stability and the uncertainty of the output of the dual-loss attribution neural network. Attribution confidence The lower the value, the less stable the current measurement is in distinguishing between the two types of loss.

[0058] Attribution confidence level When the level falls below a preset confidence threshold, or when the dominant loss type alternates across multiple consecutive control cycles, the controller enters active disturbance identification. Within the limits allowed by the safety constraint mapping, active disturbance identification first applies a current sensing disturbance with limited amplitude and duration, followed by a restricted hydraulic sensing disturbance. The restricted hydraulic sensing disturbance modifies the dilution water flow rate for a short period. Or achieve the total liquid flow rate while maintaining the safety requirements for effective chlorine dosing. The effective chlorine mass flow rate at the first detection section is determined based on two types of detection disturbances. The second detection section has an effective chlorine mass flow rate. Maintain the available chlorine deficit mass flow rate on the side. Establish a detection response matrix from at least two of the response quantities in the equivalent resistance change of the tank. If current sensing significantly affects the response of the first detection section but has little impact on the deficit ratio on the maintenance side, then increase the attribution on the generation side; if hydraulic sensing significantly alters the transport distribution core... And reduce the available chlorine deficit mass flow rate on the maintenance side. This increases the retention-side attribution.

[0059] S500 performs generation path freeze counterfactual prediction and hold path freeze counterfactual prediction for each group of candidate control quantities to decompose the generation side contribution and hold side contribution of the candidate control quantity to the improvement of the effective chlorine mass flow rate predicted in the second detection section and screen the candidate control quantity; when the hold side loss is the dominant loss type, the candidate control quantity whose hold side contribution does not meet the preset requirements and mainly relies on increasing the candidate electrolysis current to increase the generation side contribution is restricted.

[0060] For each set of candidate control variables, perform a full rolling prediction to obtain the predicted effective chlorine mass flow rate of the second detection section under the full prediction.

[0061] Execute the generated path freeze counterfactual prediction: maintain available chlorine and maintain the hidden state. , transport distribution core and effective chlorine decay rate By fixing the generator-side state and reaction flux and allowing only the generator-side state and reaction flux to vary with candidate control variables, the generator-side frozen result is obtained. The portion of the improvement in the complete prediction relative to the current baseline that can be explained by the generator-side frozen result is taken as the generator-side contribution.

[0062] Perform path-freezing counterfactual predictions: maintain generation efficiency hidden states With the first detection section predicting a fixed effective chlorine mass flow rate, only the effective chlorine is allowed to remain in a latent state. , transport distribution core or effective chlorine decay rate The hold-through path freeze results are obtained by considering the candidate brine flow rate, candidate dilution water flow rate, and the resulting changes in hydraulic state. The portion of the improvement in the complete prediction relative to the current baseline that can be explained by the hold-through path freeze results is defined as the hold-through contribution.

[0063] The candidate control variables are screened by generating side contribution and maintaining side contribution, thereby conducting preliminary screening of candidate control variables under the guidance of contribution.

[0064] When the hold-side loss is determined to be the dominant loss type, if a candidate control variable increases the candidate electrolytic current, but the hold-side contribution does not reach the preset minimum contribution ratio and the predicted hold-side normalized loss rate does not decrease, then the corresponding candidate control variable is marked as a causal inconsistency candidate variable and restricted from entering the final control cost ranking. Therefore, the controller cannot simply mask the hold-side loss that persists before the second detection section by increasing the generation amount of the first detection section.

[0065] When the generation-side loss is determined to be the dominant loss type, the candidate control quantity is allowed to be controlled by the brine flow rate. With electrolysis current Synergistically improve generation capacity, but still need to meet mass transfer allowable flux. And constraints on the allowable amount of side reactions. If an increase in the candidate electrolysis current leads to an increase in the predicted anode side oxidation flux... Cathode reduction flux Or other loss flux If the total proportion exceeds the threshold, the corresponding candidate control quantity will be downgraded in priority.

[0066] S600 evaluates the candidate control quantities after counterfactual prediction screening based on the deviation between the predicted effective chlorine mass flow rate and the target effective chlorine mass flow rate of the second detection section, the defective mass flow rate converted from the electron equivalent on the generation side, the defective mass flow rate of effective chlorine on the maintenance side, power consumption, the degree of side reaction, and the change range of the control quantity. It selects the target candidate control quantity and controls the operation of the rectifier power supply, brine supply actuator, and dilution water supply actuator under the conditions of meeting the constraints of electrolyzer temperature, current density, allowable amount of side reaction, and change rate of control quantity.

[0067] The control costs of candidate control quantities are formed based on the deviation between the predicted and target available chlorine mass flow rates at the second detection cross-section, the deficit mass flow rate calculated from the electron equivalent on the generation side, the deficit mass flow rate of available chlorine on the maintenance side, energy consumption, the degree of side reactions, and the magnitude of changes in control quantities. The control costs of these candidate control quantities are then used to conduct rolling evaluations of the candidate control quantities selected through causal consistency screening over one or more future prediction periods. Represented as: ; in, The target tracking error represents the difference between the predicted available chlorine mass flow rate and the target available chlorine mass flow rate at the second detection cross section. Deviation; To predict the generation-side loss, the predicted generation-side electron equivalent loss mass flow rate corresponding to the candidate control quantity is represented by the predicted generation-side electron equivalent loss mass flow rate. To predict the hold-side loss, this represents the predicted hold-side available chlorine loss mass flow rate corresponding to the candidate control quantity; The cost of electricity is expressed as the unit of available chlorine energy consumption. For side effects, this indicates a predicted severity of the side effects; The control change vector represents the change of the candidate control variable relative to the current control variable; This is a causal consistency penalty, representing the degree of inconsistency between the candidate control variable and the currently dominant loss type. Weights To weight All weights are non-negative.

[0068] Once the target candidate control quantity is selected, it enters the safety constraints. The safety constraints include the electrolyzer outlet temperature, electrolysis current density, current change rate, allowable side reaction, allowable salt concentration range, allowable liquid flow rate range, and safe upper limit of hydrogen concentration.

[0069] The training data for the dual-loss attribution neural network includes electrolysis current, electrolysis voltage, brine flow rate, dilution water flow rate, salt concentration or conductivity, temperature, pH, effective chlorine concentration at the first detection cross-section, effective chlorine concentration at the second detection cross-section, liquid flow rate at the two detection cross-sections, buffer container level, valve position, and equipment status data. During training, a delivery distribution kernel is established according to timestamps. Calculate the unaffected reference available chlorine mass flow rate at the second detection cross section. Calculation of the loss mass flow rate by the electron equivalent on the generation side and maintain the available chlorine deficit mass flow rate on the side .

[0070] The overall training loss consists of prediction errors from the first and second detection sections, electron equivalent conservation residuals, transport distribution consistency residuals, age distribution preservation evolution residuals, cross-sensitivity penalties, and control smoothing terms. The transport distribution consistency residuals are used to constrain the transport distribution kernel. Consistent with flow rate, level, and tracer calibration results. Cross-sensitivity penalty generates an efficiency latent state through automatic differentiation or finite difference estimation. To maintain the normalized loss rate Direct sensitivity and available chlorine retention in a latent state Normalized loss rate on the generation side The direct sensitivity is determined, and its square norm is penalized.

[0071] Example scenario of the present invention: Scenario 1: Decreased salt concentration leads to increased losses on the generation side. Effective chlorine mass flow rate at the first detection section. Significantly lower than the effective chlorine mass flow rate according to Faraday theory The generated side electron equivalent is converted to a loss mass flow rate. Increase; effective chlorine mass flow rate at the second detection cross section With no attenuation compared to the second detection section, the reference effective chlorine mass flow rate The flow rate remains relatively close, maintaining a deficit in available chlorine mass flow rate. The losses are relatively small. The system determines that the generation-side loss is dominant, prioritizes restoring the brine ratio, and operates within the allowable mass transfer flux. Increase electrolysis current when there is still a margin .

[0072] Scenario 2: An increase in the liquid level in the buffer container significantly increases the age of some liquids. The average age of the transported liquids is shown in the figure. Increase and transport distribution of nuclear liquid age dispersion Increase. Effective chlorine mass flow rate at the first detection section. Compared with Faraday's theory of available chlorine mass flow rate Basically the same, but the effective chlorine mass flow rate of the second detection section No attenuation reference effective chlorine mass flow rate below the second detection cross section Maintain the mass flow rate of available chlorine on the side. Increase. The system determines that the maintenance-side loss is dominant and prioritizes adjusting the dilution water flow rate. The total liquid flow rate or the operation mode of the buffer container can shorten the liquid age.

[0073] Scenario 3: Two sets of available chlorine analyzers exhibit slight zero-point drift. The difference between the first and second detection sections is within the maintenance uncertainty tolerance. Internally, maintain the available chlorine deficit mass flow rate on the side. If the value remains zero or low, the system will not interpret this difference as a true hold-side loss. When the drift continues to exceed the calibration allowable range, the system enters instrument calibration mode.

[0074] Scenario 4: Generating the Normalized Loss Rate With maintaining the normalized loss rate Proximity. Attribution confidence. The system performs both confined current detection and confined hydraulic detection within a safety margin. If confined current detection significantly changes the effective chlorine mass flow rate at the first detection section, while confined hydraulic detection primarily reduces the effective chlorine deficit mass flow rate on the maintenance side, then the detection response matrix... The system exhibits a distinguishable response direction, and updates the dominant loss type accordingly.

[0075] Example 2 See Figure 5 As shown, this embodiment of the invention provides an artificial intelligence-based electrolytic disinfection control device, comprising: at least one processing unit, the processing unit being connected to a storage unit via a bus unit, the storage unit serving as a computer-readable storage medium, and capable of storing software programs, computer-executable programs, and modules, such as the software program, computer-executable program, and modules corresponding to the artificial intelligence-based electrolytic disinfection control method in this embodiment of the invention. The processing unit implements the aforementioned artificial intelligence-based electrolytic disinfection control method by running the software program, computer-executable program, and modules stored in the storage unit.

[0076] Of course, the computer program stored in the storage unit of the artificial intelligence-based electrolytic disinfection control system provided in the embodiments of the present invention is not limited to the method operation described above, but can also execute related operations in the artificial intelligence-based electrolytic disinfection control method provided in any embodiment of the present invention.

[0077] Example 3 This invention provides a computer-readable storage medium storing a computer program that, when executed, implements the artificial intelligence-based electrolytic disinfection control method.

[0078] The computer-readable storage medium provided in this embodiment of the invention stores a computer program that is not limited to the method operation described above, but can also execute related operations in an artificial intelligence-based electrolytic disinfection control method provided in any embodiment of the invention.

[0079] In the embodiments provided by this invention, it should be understood that the disclosed structures and methods can be implemented in other ways. For example, the structural embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, structures, or units, and may be electrical, mechanical, or other forms.

[0080] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0081] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0082] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. An electrolytic disinfection control method based on artificial intelligence, characterized in that, include: The electrolysis current, electrolysis voltage, brine flow rate, dilution water flow rate, salt concentration or conductivity corresponding to the salt concentration, temperature and pH of the sodium hypochlorite electrolysis generating unit are collected. The first available chlorine detection amount at the first detection section at the outlet of the electrolytic cell and the second available chlorine detection amount at the second detection section downstream of the first detection section are obtained, and the target available chlorine mass flow rate corresponding to the disinfection object is obtained. Construct or update the transport distribution kernel between the first and second detection sections, use the transport distribution kernel to perform a weighted mapping on the historical sequence of the effective chlorine mass flow rate of the first detection section to obtain the attenuation-free reference effective chlorine mass flow rate of the second detection section, and calculate the normalized deficit rate on the generating side and the normalized deficit rate on the maintaining side. The generation side electron equivalent depletion mass flow rate is determined based on the difference between the effective chlorine mass flow rate corresponding to the current electrolysis current and the effective chlorine mass flow rate of the first detection section. The maintenance side effective chlorine depletion mass flow rate is determined based on the difference between the non-attenuated reference effective chlorine mass flow rate of the second detection section and the effective chlorine mass flow rate of the second detection section. Candidate control variables are generated using a dual-loss attribution neural network. The network generates isolated-corrected latent states for generation efficiency and available chlorine retention using the generation-side and retention-side state sequences. Based on the latent states for generation efficiency, an allocation ratio is generated to model the generation-side reaction flux, resulting in the predicted available chlorine mass flow rate for the first detection cross-section. A non-negative available chlorine decay rate is determined based on the available chlorine retention latent state, and a weighted summation of the predicted available chlorine mass flow rates for different liquid ages at the first detection cross-section is performed according to the transport distribution kernel to obtain the predicted available chlorine mass flow rate for the second detection cross-section. The dual-loss attribution neural network generates multiple sets of candidate control variables based on the current latent states for generation efficiency, current latent states for available chlorine retention, target available chlorine mass flow rate, transport distribution kernel, and dominant loss type. Each set of candidate control variables includes candidate electrolysis current, candidate brine flow rate, and candidate dilution water flow rate. For each group of candidate control quantities, generation path freezing counterfactual prediction and hold path freezing counterfactual prediction are performed respectively to decompose the generation side contribution and hold side contribution of the candidate control quantity to the improvement of the predicted effective chlorine mass flow rate of the second detection section and screen the candidate control quantity. When the hold side loss is the dominant loss type, the candidate control quantity whose hold side contribution does not meet the preset requirements and mainly relies on increasing the candidate electrolysis current to increase the generation side contribution is restricted. Based on the deviation between the predicted effective chlorine mass flow rate and the target effective chlorine mass flow rate of the second detection section, the defective mass flow rate converted from the electron equivalent on the generation side, the defective mass flow rate of effective chlorine on the maintenance side, the power consumption, the degree of side reaction, and the magnitude of the change in the control quantity, the candidate control quantity after being screened by counterfactual prediction is evaluated, the target candidate control quantity is selected, and the operation of the rectifier power supply, brine supply actuator, and dilution water supply actuator is controlled under the conditions of satisfying the constraints of electrolyzer temperature, current density, allowable amount of side reaction, and rate of change of control quantity.

2. The artificial intelligence-based electrolytic disinfection control method according to claim 1, characterized in that, The construction of the transport distribution kernel includes: establishing a discrete material transfer model based on the effective liquid volume and flow history between the first and second detection sections; recursively updating the liquid inventory corresponding to different liquid ages between the first and second detection sections according to the control cycle to determine the inventory ratio of each liquid age component in the current control cycle; determining the probability weight of the corresponding liquid age output from the first and second detection sections based on the inventory ratio of each liquid age component, and constructing the transport distribution kernel for the current control cycle from the probability weights corresponding to each liquid age; when the transport time between the two detection sections is approximately a single, fixed transport time, the transport distribution kernel degenerates into a single-point delay kernel.

3. The artificial intelligence-based electrolytic disinfection control method according to claim 1, characterized in that, The construction of the transport distribution kernel includes: obtaining a baseline residence time distribution through tracer experiments; determining a time scaling factor based on real-time flow rate and real-time liquid level; scaling the time axis of the baseline residence time distribution using the time scaling factor; determining the weight correction function corresponding to each residence time segment based on real-time flow rate and real-time liquid level; and performing weighted and normalized processing on the scaled residence time distribution to obtain the transport distribution kernel. When the two detection sections are approximately equal to a single, fixed transport time, the transport distribution kernel degenerates into a single-point delay kernel.

4. The electrolytic disinfection control method based on artificial intelligence according to claim 1, characterized in that, The calculation process for the generation-side electron equivalent deducted mass flow rate and the generation-side electron equivalent deducted mass flow rate includes: According to the electrolysis current Calculate the effective chlorine mass flow rate according to Faraday theory : Among them, electrolysis current For the first Electrolysis current for each control cycle The molar mass used when calculating the effective chlorine mass based on chlorine gas equivalents. It is Faraday's constant; According to Faraday's theory, the effective chlorine mass flow rate With the effective chlorine mass flow rate of the first detection section The difference between them determines the generation side electron equivalent reduced loss mass flow rate. : ； Based on the second detection section without attenuation, the reference effective chlorine mass flow rate With the second detection section, the effective chlorine mass flow rate The difference between them determines the available chlorine loss mass flow rate on the maintenance side. : ; in, To generate the side uncertainty tolerance, To maintain the tolerance of lateral uncertainty, it is determined based on the measurement uncertainty of the effective chlorine detection device, the flow detection device, and the current detection device, as well as the recent calibration residual.

5. The artificial intelligence-based electrolytic disinfection control method according to claim 1, characterized in that, The dual-loss attribution neural network includes a generator-side state encoding branch, a hold-side state encoding branch, an isolation correction and attribution layer, a response flux calculation layer, and a control decoding layer. The generation-side state encoding branch outputs the generation efficiency hidden state based on the generation-side state sequence, and the hold-side state encoding branch outputs the effective chlorine hold-side hidden state based on the hold-side state sequence. The isolation correction and attribution layer uses the generation-side electronic equivalent depleted mass flow rate and the hold-side effective chlorine depleted mass flow rate to correct the efficiency hidden state and the effective chlorine hold-side hidden state through correction interfaces, respectively. Among them, the two correction interfaces adopt gated residual units, and the gate coefficient of the gated residual unit is determined by the state of the corresponding branch and the corresponding normalized deficit rate. The generation-side normalized deficit rate is determined by the ratio of the generation-side electronic equivalent depleted mass flow rate to the Faraday theoretical effective chlorine mass flow rate, and the hold-side normalized deficit rate is determined by the ratio of the hold-side effective chlorine depleted mass flow rate to the unattenuated reference effective chlorine mass flow rate of the second detection section. The reaction flux calculation layer determines the main chlorine production flux and at least one side reaction flux based on the hidden state generation efficiency distribution ratio and the current electrolysis current. It also uses the mass transfer allowable flux determined based on salt concentration, total liquid flow rate and electrolyzer residence time to limit the main chlorine production flux, thus obtaining the predicted effective chlorine mass flow rate of the first detection section. The non-negative effective chlorine decay rate is determined based on the effective chlorine remaining in a latent state. Then, based on the transport distribution, the predicted effective chlorine mass flow rate of the first detection section at different liquid ages is calculated by performing a weighted summation after decay evolution to obtain the predicted effective chlorine mass flow rate of the second detection section. The control decoding layer generates multiple sets of candidate control quantities based on the current generation efficiency hidden state, the current effective chlorine remaining hidden state, the target effective chlorine mass flow rate, the transport distribution kernel, and the dominant loss type. Each set of candidate control quantities includes candidate electrolysis current, candidate brine flow rate, and candidate dilution water flow rate.

6. The artificial intelligence-based electrolytic disinfection control method according to claim 5, characterized in that, The reaction flux calculation layer generates the main chlorine production reaction allocation ratio, the anode secondary oxidation allocation ratio, the cathode reduction allocation ratio, and other loss allocation ratios based on the generation efficiency hidden state, and ensures that each allocation ratio is not less than 0 and the sum is 1. Based on the unified decomposition method of electrolytic current, the Radic electron equivalent flux; The fluxes for main chlorine production, anode-side oxidation, cathode reduction, and other losses are calculated based on the distribution ratios of the main chlorine production reaction, anode-side oxidation, cathode reduction, and other losses, as well as the Faraday electron equivalent flux. Set the mass transfer allowable flux limit for the main chlorine production flux, and the actual main chlorine production flux is the smaller of the main chlorine production flux and the mass transfer allowable flux.

7. The artificial intelligence-based electrolytic disinfection control method according to claim 1, characterized in that, The process by which the dual-loss attribution neural network obtains the predicted effective chlorine mass flow rate for the second detection cross-section includes: Use the Softplus function or exponential function to keep available chlorine in a hidden state. Transformed into a non-negative effective chlorine decay rate, For the delay index in the delivery distribution kernel The corresponding liquid age component, based on the non-negative effective chlorine decay rate Calculate the available chlorine retention coefficient Represented as: ； Based on the delivery distribution, the effective chlorine mass flow rate is predicted by checking the first detection section at different liquid ages. After separate attenuation, the effective chlorine retention coefficient is used. The weighted summation yields the predicted available chlorine mass flow rate for the second detection section. : ； The predicted effective chlorine mass flow rate for the first detection section is obtained by converting the effective chlorine mass equivalent based on the actual main chlorine production flux and the side reaction flux.

8. The artificial intelligence-based electrolytic disinfection control method according to claim 1, characterized in that, Preliminary dominant loss determination is performed based on the normalized loss rate of the generation side and the normalized loss rate of the maintenance side. Attribution confidence is calculated, and active perturbation identification is performed based on the dominant loss determination results and the calculated attribution confidence. Based on the normalized loss rate of the generating side and maintain the normalized loss ratio Preliminary dominant loss determination: When the normalized loss rate on the generation side exceeds the generation side determination threshold and is higher than the normalized loss rate on the maintenance side, the generation side loss is determined to be the dominant loss type; when the normalized loss rate on the maintenance side exceeds the maintenance side determination threshold and is higher than the normalized loss rate on the generation side, the maintenance side loss is determined to be the dominant loss type. The attribution confidence is determined by the interval between the two types of normalized loss rates, the uncertainty tolerance on the generation side, the uncertainty tolerance on the maintenance side, the stability of recent judgments, and the uncertainty output by the dual-loss attribution neural network. When the attribution confidence is lower than the preset confidence threshold, or when the dominant loss type alternates within multiple consecutive control cycles, the controller enters active disturbance identification: within the range allowed by the safety constraint mapping, a current sensing disturbance with limited amplitude and duration is first applied, followed by a limited hydraulic sensing disturbance; a sensing response matrix is ​​established based on at least two of the responses of the two types of sensing disturbances to the effective chlorine mass flow rate of the first sensing section, the effective chlorine mass flow rate of the second sensing section, the effective chlorine deficit mass flow rate on the holding side, and the change in the equivalent resistance of the tank; if the current sensing response matrix has a significant effect on the first sensing section but a small effect on the deficit ratio on the holding side, the attribution confidence on the generating side is increased; if the hydraulic sensing significantly changes the transport distribution kernel and reduces the effective chlorine deficit mass flow rate on the holding side, the attribution confidence on the holding side is increased.

9. The artificial intelligence-based electrolytic disinfection control method according to claim 1, characterized in that, The generation path freeze counterfactual prediction keeps the available chlorine in a hidden state and the transport distribution kernel does not update with the candidate control, allowing the generation side state and reaction flux to respond to the candidate control to obtain the generation side contribution; The hold-path freeze counterfactual prediction maintains the hidden state of the generation efficiency and the first detection section predicts the effective chlorine mass flow rate without updating with the candidate control quantity, allowing the hold-side state, delivery distribution kernel, or effective chlorine decay rate to respond to the candidate control quantity to obtain the hold-side contribution.

10. The artificial intelligence-based electrolytic disinfection control method according to claim 1, characterized in that, The overall training loss consists of the prediction error of the first detection section, the prediction error of the second detection section, the electron equivalent conservation residual, the transport distribution consistency residual, the age distribution maintenance evolution residual, the cross-sensitivity penalty, and the control smoothing term. The transport distribution consistency residual is used to constrain the transport distribution kernel to be consistent with the flow rate, liquid level, and tracer calibration results. The cross-sensitivity penalty term is used to reduce the direct sensitivity of the latent state of generation efficiency to the normalized deficit rate on the maintenance side, and to reduce the direct sensitivity of the latent state of available chlorine maintenance to the normalized deficit rate on the generation side.

Images