A method and system for determining the end point of an online oxidation reaction of ferric pyrophosphate by ORP

Abstract

The application discloses a pyrophosphoric acid iron oxidation reaction online ORP end point determination control method and system, and belongs to the field of online detection and automatic control of industrial chemical reactions. In order to solve the problems of ORP baseline drift, fixed threshold misjudgment and field disturbance mistriggering in the pyrophosphoric acid iron oxidation reaction, the application collects ORP, temperature, pH, dosing flow, stirring speed and event signals, carries out robust filtering, batch calibration, multi-scale curve feature recognition, credible weight gating and post-execution review, realizes adaptive end point determination and stable dosing control.

Description

Technical Field

[0001] This invention relates to the field of online detection and automatic control of industrial chemical reactions, and in particular to an online ORP endpoint determination and control method and system for the oxidation of iron pyrophosphate. Background Technology

[0002] Industrial control of the ferric pyrophosphate oxidation reaction typically relies on online ORP curves, human experience, fixed thresholds, or fixed reaction times to determine the endpoint. Because the raw material batches, initial ferrous to ferric ratio, degree of pyrophosphate complexation, salt effect, temperature, and pH conditions vary with the reaction progress, the baseline position, rising range, and plateau characteristics of the ORP curve are prone to batch-to-batch drift.

[0003] Existing fixed threshold or fixed time control methods are difficult to adapt to the above-mentioned drift, which can easily lead to premature endpoint misjudgment, delayed identification, and excessive or insufficient addition of oxidant. At the same time, uneven stirring, bubble adhesion, transient impacts during feeding and replenishment, pump and valve pulsation, and short-term electrode response hysteresis can introduce spikes, jumps, and short-cycle oscillations, causing the control system to be falsely triggered or frequently start and stop.

[0004] Therefore, there is a need for an online ORP endpoint determination and control method and system for the iron pyrophosphate oxidation reaction that can overcome the shortcomings of the existing technology. Summary of the Invention

[0005] One objective of this invention is to propose an online ORP endpoint determination and control method and system for the oxidation reaction of iron pyrophosphate. Addressing the problems of existing technologies where ORP endpoint determination relies on fixed thresholds or fixed times and is difficult to adapt to batch drift and field disturbances, this invention proposes a technical solution that combines robust filtering, event confidence weighting, intra-batch baseline and temperature and pH compensation, multi-scale rate of change and platform feature identification, interval consistency confirmation, and verification control. This allows the endpoint determination to adaptively form according to the characteristics of the current batch curve, and to enter a delayed observation or re-determination phase when confidence is low or verification is abnormal.

[0006] This invention provides an online ORP endpoint determination and control method for the oxidation reaction of iron pyrophosphate, comprising: S1. Collect raw ORP sequence, temperature sequence, pH sequence, dosing flow rate sequence, stirring speed sequence, feeding event and replenishment event signals to form reaction process data; S2. Robustly filter the original ORP sequence and generate reliable weights for sampling points based on feeding events, replenishment events, feeding flow rate sequences, and stirring speed sequences to obtain the filtered ORP sequence and reliable weight sequence. S3. Determine the start time of oxidant addition based on the addition flow rate sequence, determine the initial stabilization window after that time based on the filtered ORP sequence, and determine the ORP baseline, noise scale, temperature and pH compensation amount for this batch based on this window, the filtered ORP sequence, and the pH sequence to obtain the normalized ORP sequence. S4. Within at least two time-sliding windows, calculate the rate of change, curvature, cumulative rise, and plateau volatility based on the normalized ORP sequence to obtain a multi-scale curve characteristic sequence. S5. Based on the multi-scale curve feature sequence, identify the candidate endpoints for the rising segment to enter the plateau segment. After continuous confirmation of the rate of change, confirmation of plateau fluctuation, and confirmation of interval consistency, obtain the endpoint validity result and obtain the control reference time when the result is valid or pending confirmation. S6. Based on the endpoint validity results, and when the results are valid or pending confirmation, combine the control baseline time and the credible weight sequence to output instructions for stopping oxidant addition, slowing down oxidant addition, or delaying observation. After execution, review the subsequent ORP changes.

[0007] Optionally, S1 includes: Data is collected from the ORP electrode, temperature sensor, pH sensor, oxidant dosing pump, stirring actuator, feeding valve and replenishing valve according to the preset sampling cycle; The acquisition time, sensor values, actuator status, and event trigger flags are written into the same timeline; A missing sampling point is set with a missing flag and the time order of its adjacent valid sampling points is maintained. The missing flag is used as the data quality flag for input step S2 to obtain the reaction process data.

[0008] Optionally, S2 includes: Determine the median value and absolute deviation scale of ORP within a filtering window centered on the current sampling point; When the absolute value of the difference between the current sampling point and the median value of the ORP is greater than the peak determination threshold determined by the absolute deviation scale, the current sampling point is marked as a peak point; The peak points and sampling points with the missing flag are replaced by the weighted value of adjacent valid sampling points, and the sampling points that are not peak points and do not have the missing flag are smoothed by Huber weights to obtain the filtered ORP sequence. The reliable weight sequence is generated based on the peak point marker, missing marker, feeding event, liquid replenishment event, and pump start / stop marker, flow rate change marker, and speed step marker identified by the feeding flow rate sequence and stirring speed sequence. Furthermore, the generation of the credible weight sequence includes: generating a data quality flag from the acquisition status in the reaction process data, and using the data quality flag corresponding to the missing sampling point as a missing flag; The preset sampling window surrounding the current sampling point is used as the filtering window. The peak determination threshold is determined according to the absolute deviation scale of ORP within the filtering window. Sampling points whose absolute difference between the current sampling point and the median value of ORP within the filtering window is greater than the peak determination threshold are marked as ORP peak points. The feeding event, the liquid replenishment event, the oxidant dosing pump start-stop event and flow change event identified by the feeding flow sequence, the stirring speed step event identified by the stirring speed sequence, the ORP peak point marker and the missing flag are respectively mapped to the event weight decay amount; The event weight decay is time-decayed according to the length of the freeze window following the event occurrence. When there is no event weight decay amount after time decay at the same sampling point, the aggregated event weight decay amount is determined to be zero. When there is one event weight decay amount after time decay at the same sampling point, the event weight decay amount after time decay is determined as the aggregated event weight decay amount. When there are multiple event weight decay amounts after time decay at the same sampling point, the multiple event weight decay amounts after time decay are summed and truncated according to the preset decay upper limit to obtain the aggregated event weight decay amount. The baseline weight value is the upper limit of the preset confidence weight range. The preset attenuation upper limit does not exceed the baseline weight value. The confidence weight of the sampling point is obtained by subtracting the weight attenuation of the aggregated event from the baseline weight value, and the confidence weight of the sampling point is limited to the preset confidence weight range. When the confidence weight of a sampling point is less than the confidence weight threshold, the corresponding sampling point is set as a frozen sampling point that does not trigger the endpoint determination. The frozen sampling point is not used as a candidate endpoint moment in step S5, and is not used to determine the endpoint validity result as valid. The lower limit to the upper limit of the preset trusted weight range constitutes the trusted weight value range of the sampling point, which is used for endpoint determination and control of the trusted weight threshold value in trusted judgment.

[0009] Optionally, S3 includes: The initial stable window is a continuous sampling interval within a preset observation period after the oxidant is added, where the confidence weight is greater than the initial stable window confidence weight threshold and the absolute value of the ORP change rate determined by the filtered ORP sequence within the preset window is less than the benchmark change rate threshold. The initial stable window confidence weight threshold is a preset threshold within the range of confidence weight values ​​of the sampling points, and the range of confidence weight values ​​of the sampling points is a closed interval between the preset lower limit and the preset upper limit of the confidence weight of the sampling points in the confidence weight sequence. The ORP baseline for this batch is determined based on the mean of the filtered ORP within the initial stabilization window, and the noise scale is determined based on the quantile of the filtered ORP residuals within the initial stabilization window. The temperature compensation amount is determined based on the temperature deviation of the temperature sequence relative to the average temperature of the initial stable window and a preset temperature compensation coefficient or temperature calibration compensation curve. The pH compensation amount is determined based on the pH deviation of the pH sequence relative to the average pH value of the initial stable window and a preset pH compensation coefficient or pH calibration compensation curve. The filtered ORP sequence is corrected and scaled using the ORP baseline, noise scale, temperature and pH compensation of this batch to obtain the normalized ORP sequence. If the initial stable window is not obtained within the preset observation period, a baseline unavailable flag is generated; During the period when the baseline unavailable flag exists, steps S1 to S3 continue to be executed to update the reaction process data, filtered ORP sequence, and trusted weight sequence, and to re-search for the initial stable window. Step S4, multi-scale curve feature extraction, and step S5, candidate endpoint triggering are paused. The endpoint validity result is set to invalid, and step S6 outputs a delayed observation instruction and temporarily suspends the endpoint control verification based on the normalized ORP sequence. The temporary suspension of the endpoint control verification based on the normalized ORP sequence means pausing the verification calculation based on the normalized ORP sequence until the initial stable window is obtained again, after which steps S4 to S6 and the corresponding endpoint verification are resumed.

[0010] Optionally, S4 includes: Linear fitting was performed on the sampling points in the normalized ORP sequence whose confidence weights were greater than the confidence weight threshold in the short window, medium window, and long window respectively, to obtain the rate of change of each window. The determination of the confidence weight threshold is a preset threshold that is within the range of the confidence weight values ​​of the sampling point and is independent of the initial stable window confidence weight threshold setting. When the proportion of sampling points within the window whose confidence weight is not greater than the confidence weight threshold is greater than the preset low confidence proportion threshold, the window is set as an invalid feature window that is not used in step S5. The curvature is determined based on the difference in the rate of change between adjacent time windows; The cumulative increase is determined based on the normalized ORP difference between the current window endpoint and the window start point. The platform volatility is determined based on the quantile difference of the normalized ORP residuals within the current window; The rate of change, curvature, cumulative rise, and plateau fluctuation are combined according to the sampling time to obtain the multi-scale curve feature sequence.

[0011] Optionally, S5 includes: The sampling points that do not trigger endpoint determination, determined based on the trusted weight sequence and according to the trusted weight threshold, are designated as frozen sampling points. The sampling points participating in the calculation of candidate endpoints and confirmation windows do not include the frozen sampling points. The short, medium, and long windows used for candidate endpoint triggering and continuous confirmation do not include the invalid feature window. When the medium or long window required for candidate endpoint triggering is an invalid feature window, the initial candidate endpoint time is not generated and the endpoint validity result is determined to be invalid. When the window change rate changes from being greater than the first adaptive change rate threshold to being less than the second adaptive change rate threshold, and the cumulative increase of the long window is greater than the cumulative increase threshold, the corresponding sampling time is determined as the initial candidate endpoint time. The first adaptive rate of change threshold is used to confirm the rising segment and is determined according to the noise scale and the quantile of the rate of change of the middle window within the preset learning interval after the initial stable window. The second adaptive rate of change threshold is used to confirm the plateau segment and is determined according to the noise scale and the quantile of the rate of change of the short window within the preset learning interval. The second adaptive rate of change threshold is not greater than the first adaptive rate of change threshold. Within the continuous confirmation window following the initial candidate endpoint time, if the short window change rate is continuously less than the second adaptive change rate threshold, the platform volatility is less than the platform volatility threshold, the absolute value of curvature is less than the curvature threshold, and the candidate endpoint time set generated within multiple confirmation windows satisfies the consistency condition corresponding to the interval consistency threshold, then the endpoint validity result is determined to be valid, and the median time of the candidate endpoint time set is determined as the control reference time for step S6. If the initial candidate endpoint time is not identified, the cumulative rise amount of the long window is not greater than the cumulative rise amount threshold, the established platform fluctuation confirmation result shows that the platform fluctuation degree is not less than the platform fluctuation degree threshold, the absolute value of curvature is not less than the curvature threshold, or the established candidate endpoint time set does not meet the consistency condition, then the endpoint validity result is determined to be invalid. If the initial candidate endpoint time has been identified and the aforementioned invalid determination result has not yet been formed, but the number of continuous confirmation windows has not reached the preset number of confirmation windows, the continuous confirmation of the rate of change has not been completed, the confirmation of platform fluctuation has not been formed, the confirmation of interval consistency has not been formed, or the necessary window required for continuous confirmation is an invalid feature window and the preset number of valid confirmation windows cannot be formed, then the endpoint validity result is determined to be pending confirmation, and the initial candidate endpoint time is determined as the control reference time for step S6. Furthermore, the set of candidate endpoint times generated within the multiple confirmation windows satisfies the consistency condition corresponding to the interval consistency threshold, including: forming no less than three consecutive confirmation windows with the sampling period as the step size; generating candidate endpoint times within each confirmation window according to the rule that the change rate of the middle window changes from being greater than the first adaptive change rate threshold to being less than the second adaptive change rate threshold and the cumulative increase of the long window is greater than the cumulative increase threshold, and forming the set of candidate endpoint times. Each confirmation window used for interval consistency confirmation generates a candidate endpoint time, and the number of elements in the candidate endpoint time set is not less than a preset candidate number threshold. If any confirmation window used for interval consistency confirmation fails to generate a candidate endpoint time or the number of elements in the candidate endpoint time set is less than the preset candidate number threshold, an interval consistency failure flag is output. For each confirmation window, if the rate of change of the short window within the confirmation window is continuously less than the second adaptive rate of change threshold, then the rate of change confirmation result is generated as passed; otherwise, the rate of change confirmation result is generated as failed. If the platform volatility within the confirmation window is less than the platform volatility threshold, the platform volatility confirmation result is "passed"; otherwise, the platform volatility confirmation result is "failed". If the absolute value of curvature within the confirmation window is less than the curvature threshold, the curvature confirmation result is "passed"; otherwise, the curvature confirmation result is "failed". Calculate the absolute value of the deviation between each candidate endpoint time and its median time in the candidate endpoint time set, and take the absolute value of the deviation that satisfies that no other absolute value of deviation is greater than this value as the upper bound value of the deviation; When the upper bound of the deviation is less than the interval consistency threshold, and the confirmation results of the rate of change, platform fluctuation, and curvature of each confirmation window are all passed, an interval consistency pass flag is output, and the median time is used as a candidate time for determining the control reference time. The interval consistency is defined by a flag that serves as a necessary condition for determining the validity of the endpoint validity result.

[0012] Optionally, S6 includes: When the endpoint validity result is invalid, a delayed observation instruction is output; When the endpoint validity result is valid, the set of sampling points covered by multiple confirmation windows that meet the interval consistency condition in step S5 is determined as the control decision sampling interval. When the endpoint validity result is pending confirmation, the set of sampling points covered by the currently formed continuous confirmation window after the control reference time is determined as the control decision sampling interval; If the number of sampling points in the control decision sampling interval is less than the preset minimum number of samples, then the delayed observation instruction is output. Both the sampling point trigger confidence weight threshold and the control confidence weight threshold are preset thresholds within the range of sampling point confidence weight values. The range of sampling point confidence weight values ​​is a closed interval between a preset lower limit and a preset upper limit of the sampling point confidence weight in the confidence weight sequence. The sampling point trigger confidence weight threshold is used for single sampling point trigger exclusion, and the control confidence weight threshold is used for control decision sampling interval mean exclusion. Furthermore, the control confidence weight threshold is greater than the sampling point trigger confidence weight threshold. First, determine whether the confidence weight of each sampling point within the control decision sampling interval is greater than the confidence weight threshold for triggering the sampling point. If the confidence weight of any sampling point is not greater than the confidence weight threshold for triggering the sampling point, then the delayed observation instruction is output. If the confidence weight of each sampling point is greater than the confidence weight threshold for the sampling point trigger and the validity result of the endpoint is valid, and the average confidence weight within the control decision sampling interval is greater than the control confidence weight threshold, then an oxidant stop control command is output. If the confidence weight of each sampling point is greater than the confidence weight threshold of the sampling point and the validity result of the endpoint is pending confirmation, and the average confidence weight within the control decision sampling interval is greater than the control confidence weight threshold, then an oxidant deceleration control command to reduce the oxidant dosing acceleration rate is output. If the confidence weight of each sampling point is greater than the confidence weight threshold for the sampling point trigger and the validity result of the endpoint is valid or pending confirmation, and the average confidence weight within the control decision sampling interval is not greater than the control confidence weight threshold, then the delayed observation instruction is output. The delayed observation instruction is an observation control instruction that prohibits the triggering of oxidant stopping or oxidant slowing down based on the current determination cycle, and switches the oxidant addition flow rate to the preset observation flow rate; In the review window after the control command is executed, the subsequent ORP change rate and the subsequent ORP cumulative change amount are calculated based on the normalized ORP sequence. If the subsequent ORP change rate is greater than the review change rate threshold or the subsequent ORP cumulative change amount is greater than the review cumulative change threshold, an endpoint review abnormality flag is generated. If the endpoint verification anomaly flag is generated, then when the current control command is an oxidant stop control command or an oxidant deceleration control command, the current endpoint control result is revoked and the control state of executing the delayed observation command is switched. In the delayed observation state, steps S2 to S6 are re-executed or the next determination cycle is entered. When the current control command is the delayed observation command, maintain the delayed observation state, and re-execute steps S2 to S6 or enter the next determination cycle in the delayed observation state; If the subsequent ORP change rate is not greater than the review change rate threshold and the cumulative change of the subsequent ORP is not greater than the review cumulative change threshold, then an endpoint review pass flag is generated. If the endpoint verification pass flag is generated, then when the current control command is the oxidant deceleration control command, the deceleration state is maintained and subsequent reaction process data is collected. Steps S2 to S6 are re-executed or the next judgment cycle is entered until the endpoint validity result is valid and the oxidant stop control command is output, or the delayed observation command is output when the judgment condition is not met. When the current control command is the delayed observation command, maintain the delayed observation state and continue to collect subsequent reaction process data, and re-execute steps S2 to S6 or enter the next determination cycle. When the current control command is the oxidant stop control command, maintain the control state corresponding to the current control command.

[0013] On the other hand, the present invention also provides an online ORP endpoint determination and control system for iron pyrophosphate oxidation reaction, comprising: The system includes a data acquisition module for generating reaction process data; a robust filtering and event weighting module for obtaining the filtered ORP sequence and a reliable weight sequence; an intra-batch calibration module for obtaining the normalized ORP sequence; a curve feature extraction module for obtaining the multi-scale curve feature sequence; an endpoint confirmation module for obtaining the endpoint validity result and control reference time; and a dosing control and verification module for outputting the oxidant stop control command, the oxidant deceleration control command, or the delayed observation command based on the endpoint validity result, the control reference time, and the reliable weight sequence, and verifying the endpoint status.

[0014] The beneficial effects of this invention are: 1. By determining the ORP baseline, noise scale, and temperature and pH compensation based on the initial stabilization window, the impact of different batch raw material conditions, complexation state, and process fluctuations on the fixed ORP threshold can be reduced, enabling subsequent endpoint identification to be based on the normalized online curve characteristics.

[0015] 2. By generating reliable weights for feeding, replenishment, pump start-up and shutdown, sudden changes in flow rate, step changes in stirring speed, ORP spikes and missing sampling points, and excluding low-reliability sampling points, invalid feature windows and frozen sampling points from candidate endpoints and confirmation windows, false triggering caused by spikes, bubbles, electrode hysteresis and short-period oscillations can be suppressed.

[0016] 3. By continuously confirming the rate of change, platform fluctuations, curvature, consistency of candidate endpoint time intervals, and post-execution ORP verification, a hierarchical control chain of stop feeding, deceleration feeding, and delayed observation can be formed between valid, pending confirmation, and invalid states, thereby improving the stability of endpoint stop feeding control. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of an online ORP endpoint determination and control method for the oxidation reaction of iron pyrophosphate.

[0018] Figure 2 This is a flowchart of step S2, robust filtering and reliable weight generation, of the present invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figures 1-2 A method for determining and controlling the online ORP endpoint of the ferric pyrophosphate oxidation reaction, comprising: S1. Collect raw ORP sequence, temperature sequence, pH sequence, dosing flow rate sequence, stirring speed sequence, feeding event and replenishment event signals to form reaction process data; S2. Robustly filter the original ORP sequence, and generate reliable weights for sampling points based on feeding events, replenishment events, feeding flow rate sequences, and stirring speed sequences to obtain the filtered ORP sequence and reliable weight sequence. S3. Determine the start time of oxidant addition based on the added flow rate sequence, determine the initial stabilization window after that time based on the filtered ORP sequence, and determine the ORP baseline, noise scale, temperature and pH compensation amount for this batch based on the window and the filtered ORP sequence, temperature and pH sequence, to obtain the normalized ORP sequence. S4. Within at least two time-sliding windows, calculate the rate of change, curvature, cumulative rise, and plateau volatility based on the normalized ORP sequence to obtain a multi-scale curve characteristic sequence. S5. Based on the multi-scale curve feature sequence, identify the candidate endpoint of the rising segment entering the plateau segment, obtain the endpoint validity result through continuous confirmation of the rate of change, confirmation of plateau fluctuation and confirmation of interval consistency, and obtain the control reference time when the result is valid or pending confirmation. S6. Based on the endpoint validity result, and when the result is valid or pending confirmation, combine the control baseline time and the reliable weight sequence to output the instructions to stop adding oxidant, reduce the rate of adding oxidant, or delay observation. After execution, review the subsequent ORP changes.

[0021] In this specific embodiment, S1 includes: The controller generates reaction process data on the same time axis with a fixed sampling period. The sampling time of the controller is according to the formula. Generate, where, For the first Each sampling time, This is the start time of this batch of data collection. The sampling point number starts from zero. The sampling period is 1 second. At each sampling time The controller synchronously reads the raw ORP value output by the ORP electrode through the signal acquisition interface. It measures in mV and simultaneously reads the temperature value output by the temperature sensor. and with Measuring and reading the pH value output by the pH sensor. Read the instantaneous dosing flow rate of the oxidant dosing pump. and with Metering and reading the stirring speed of the stirrer drive. And measured in rpm; To ensure that data from all sources is written to the same timeline, the controller... A data acquisition transaction is triggered, and a 200ms acquisition timeout is set for each channel. If a channel fails to return a valid value within the timeout period or the returned value exceeds the upper or lower limit of the configured engineering range for that channel, the channel will be stopped at the sampling point. The sample points are marked as missing and their values ​​are not interpolated to fill in the missing values, thus maintaining the continuity and monotonicity of the sampling point sequence in time. The controller performs edge detection on the opening and closing commands and valve position feedback of the feeding valve and the replenishing valve, and will... The change from closed to open within the interval is recorded as the feeding event trigger flag. Or rehydration event trigger flag When not triggered, they are respectively recorded as and And the oxidant dosing pump and the agitator drive are in The operation enable status, start / stop status, and frequency setpoint status at any given time are recorded as the actuator status and are written together with the sensor values ​​mentioned above into the same sampling record. The controller generates a data quality flag for each sampled record. When any critical channel ORP raw value Missing time setting Otherwise and will With the time of collection Sensor values and event trigger flags Data is written to the reaction process data table in a fixed field order, ensuring that missing sampling points occupy positions on the time axis and that the order of their adjacent valid sampling points remains unchanged, and the data quality flag is set accordingly. The input data quality flags used in step S2 to generate missing flags and confidence weights are used for subsequent robust filtering and endpoint determination control.

[0022] In this specific embodiment, S2 includes: The controller at each sampling time Receive reaction process data and analyze the original ORP sequence Perform robust filtering and event credibility weighting, where, For the first Each sampling time, The sampling point number, Sampling time The corresponding raw ORP value, As a marker of data quality and when When missing otherwise This is a flag that triggers the feeding event. This is a flag indicating that a fluid resuscitation event has been triggered. The oxidant dosing flow rate, This refers to the stirring speed; The controller uses the current sampling point Construct a filter window centered on Filter window Includes from arrive sampling points and The controller is in the filtering window Intra-selection satisfies The sampling points are used to form a valid set, and the original ORP values ​​in this valid set are sorted to calculate the median value of the ORP. When the number of valid set elements is odd Take the middle element after sorting; when the number of valid elements in the set is even. Take the arithmetic mean of the two middle elements after sorting; The controller further calculates the absolute deviation scale. ,in For filtering window The original values ​​of each ORP in the inner valid set and The median of the absolute values ​​of the difference Used to characterize the robust noise scale within the window, and when When equal to zero Set to 0.1mV to avoid peak threshold degradation; The controller generates ORP peak markers for the current sampling point based on the peak determination threshold. And determine according to the following formula: ; in, Mark the peak of ORP and when The time indicates whether the current sampling point is a peak or a missing point. For absolute value operations, Let be the peak determination coefficient and take This is the median value of the ORP. The absolute deviation scale; when The time controller generates a filtered ORP value by weighting the adjacent valid sampling points for the current sampling point. ,in For filtering ORP sequences at sampling points The value at that point is replaced by searching forward along the timeline for the nearest satisfying condition. and Sampling point index and search backwards for the nearest satisfied and Sampling point index The search scope is limited to and and When index With index Simultaneously, when both endpoints exist, a linear weighting is applied based on the time distance, and... Set as the weighted result of the two endpoints, where the front endpoint is taken as And the endpoint is taken When only indexing When it exists Set as When only indexing When it exists Set as When index With index When neither exists Set to the filtered ORP value of the previous sampling point. This ensures continuous output of the filtered ORP sequence; when Time controller for filtering window The residuals of each sampling point in the inner valid set Calculate Huber weights and smooth them, where Sampling points The residual, Huber threshold Pick ,when When the corresponding Huber weight is 1, The corresponding Huber weight is taken at that time. The controller assigns Huber weights to the valid set. Weighted average And write it into the filtered ORP sequence; While generating the filtered ORP sequence, the controller generates a reliable weight sequence based on the reaction process data. ,in Sampling points The trusted weights are determined and limited to a preset trusted weight range. Inner and The controller maps the following events to event weight decay amounts and adds them together with time decay. These events include material feeding events. Fluid replacement incident Oxidant dosing pump start / stop events Traffic mutation events Stirring speed step event ORP spike marker and missing markers ; Oxidant dosing pump start / stop events The controller determines the flow rate threshold based on the input flow rate. Generate pump operation indicator ,when hour otherwise and in Time setting Otherwise set ; Traffic mutation event It is determined by the difference between adjacent sampling points, when Time setting Otherwise set ,in ; Stirring speed step event It is determined by the difference between adjacent sampling points, when Time setting Otherwise set ,in ; The controller handles each event type Set a specific event weight decay amount. With the length of the frozen window Among them, the feeding event corresponds to fluid replacement event response Pump start / stop events correspond to Traffic mutation events correspond to , Stirring step event corresponds to ORP spike events correspond to Missing events correspond to ; When an event occurs at time After it occurs, the controller at any subsequent sampling point Calculate the time decay contribution of the event and apply a linear decay rule for decay. The linear decay rule is as follows: when... At that time, the contribution of the event was zero. The contribution of this event at that time was The controller contributes all events within their respective freeze windows to the sampling points. The summation yields the aggregated event weight decay, which is then adjusted according to a preset decay limit. Cut off to form the final aggregate decay, then use the baseline weight value. Subtracting the aggregation attenuation amount yields the confidence weight of the sampling point. and will Limited to The reliable weight sequence can be obtained from this. The controller further determines the reliability weight threshold. Generate frozen sampling point markers ,in For freezing sampling point markers and when hour otherwise and filter the ORP sequence Confidential weight sequence ORP spike marker With frozen sampling point marker By sampling time The reaction process data is written back for subsequent steps S3 to S6 for in-batch calibration, feature extraction, endpoint determination, and control verification.

[0023] In this specific embodiment, S3 includes: The controller is based on the filtered ORP sequence. Confidential weight sequence and temperature sequence pH sequence and the added flow sequence Complete in-batch calibration and generate normalized ORP sequences. ,in, The sampling point number, Sampling points The filtered ORP value, Sampling points Credibility weight, Sampling points Temperature value, Sampling points pH value, Sampling points The oxidant dosing flow rate; The controller determines the threshold at the start of the application. Determine the sampling point number corresponding to the start time of oxidant addition. The judgment rule is that when the condition is met... and The smallest The injection begins at the lock point, if the current batch collection start point meets the requirements. Then place ; since Start at the preset observation time The initial stable window is searched within the corresponding observation interval, and the window duration of the initial stable window is taken as... And the corresponding number of sampling points The controller enumerates each candidate window in a sliding manner. A window is confirmed as an initial stable window only if both types of conditions are met simultaneously. The first type of condition is the confidence condition, which means that the confidence weight of any sampling point within the window satisfies the following conditions. ,and ; The second type of condition is the baseline rate of change condition, which means that the absolute value of the linear rate of change of the filtered ORP within the window does not exceed the baseline rate of change threshold. The linear rate of change is determined by the change within the window. The slope representation obtained by performing least-squares linear fitting. The sampling time is determined by step S1. , For the sampling period and take ; The controller determines the earliest candidate window that meets the above conditions as the initial stable window and locks its start and end numbers. and ,in ; After obtaining the initial stabilization window, the controller calculates the ORP baseline for this batch. Reference temperature Compared with the reference pH value ,in Take the initial stable window The arithmetic mean, Take the initial stable window The arithmetic mean, Take the initial stable window The arithmetic mean; The controller further calculates the noise scale. ,in From the residual within the initial stable window The quantiles are determined, and the controller calculates the residuals. quantiles and quantiles And order ,when When Set to 0.5mV to avoid scaling up noise; The controller uses a defined temperature compensation coefficient. pH compensation coefficient Calculate temperature and pH compensation amounts and correct and scale the filtered ORP sequence at each sampling point. Generate normalized ORP values The formula for its calculation is: ; in, For normalized ORP values, For filtering ORP values, This serves as the baseline for the ORP of this batch. This is the temperature compensation coefficient. Temperature at the sampling point As the reference temperature, This is the pH compensation coefficient. The pH value at the sampling point. As the reference pH value, For noise scale; If an initial stable window satisfying both the confidence condition and the baseline rate of change condition cannot be found within the observation interval, the controller generates a baseline unavailable flag. No output and The baseline is unavailable flag. During the existence period, the controller continuously executes steps S1 to S3 to perform rolling updates. and The initial stable window is searched again using the latest data. At the same time, the multi-scale curve feature extraction in step S4 and the candidate endpoint triggering in step S5 are paused, and the endpoint validity results are set to invalid. In step S6, a delayed observation instruction is output, and all ORP sequences dependent on normalized ORP sequences are paused. The endpoint control verification calculation continues until the initial stable window is regained and the baseline unavailable flag is cleared. Then, resume execution of steps S4 to S6 and continuously generate normalized ORP sequences for subsequent sampling points. .

[0024] In this specific embodiment, S4 includes: Controller not available at baseline Under the condition of normalized ORP sequences Multi-scale curve features are extracted online using a sliding window to form a multi-scale curve feature sequence, where... The sampling point number, Sampling points The normalized ORP value, The sampling time and satisfying The start time of data collection. The sampling period is 1 second. The confidence weights for the sampling points are determined by setting a confidence weight threshold. And determine the confidence weight threshold and the initial stable window confidence weight threshold in step S3. They are independent of each other; The controller samples each point. Construct short, medium, and long windows respectively, with the short window length set to... And the corresponding number of sampling points The length of the middle window is taken And the corresponding number of sampling points The length of the long window is taken as And the corresponding number of sampling points It adopts a causal window format with the current sampling point as the window endpoint, and the short window index range is... The middle window index range is The long window index range is ,when If the index is less than the minimum index required for the corresponding window, the window will not generate features and will be marked as an invalid feature window; In any window Within, the controller only selects those that meet the requirements. sampling points Forming the effective point set for fitting And calculate the low confidence ratio. ,in For the window to satisfy The ratio of the number of sampling points to the total number of sampling points in the window, with a low confidence ratio threshold of [value missing]. ,when or The window is then set as an invalid feature window and an invalid feature window flag is generated. ,in The number of sampling points for the corresponding window and The number of elements in the set of valid points. The invalid feature window flag is set when the window is available. ; When the window is available At that time, the controller selects the set of valid points within this window. Perform least-squares linear fitting to obtain the rate of change. The formula for its calculation is: ; in, For window At the sampling point rate of change at that point Sampling points Sampling time, Sampling points The normalized ORP value, Set of valid points Within each The arithmetic mean, Set of valid points Within each The arithmetic mean, For summation operations; the controller determines the curvature feature based on the difference in the rate of change of adjacent time windows and uniformly uses the rate of change of the middle window for curvature calculation. The curvature feature is denoted as... The window is at the sampling point and All are available and Time to take When any window is unavailable, Set to null and mark as unusable for curvature features in S5; The controller uses the difference between endpoints of a long window to determine the cumulative increase characteristic and records it as... When a long window is available Time to take When long windows are unavailable Set to null and mark as unusable for the cumulative increase feature of S5; The controller determines the platform volatility characteristics based on the quantile difference of the fitted residuals within a short window and denotes it as follows. When the short window is available At that time, the controller first uses a short window to fit a straight line at each effective point. Calculate the residuals at the following locations: ; in For residuals, The slope is the fitting slope for the short window. The short window fitting intercept is determined by least squares and The only one that is certain is Subsequently calculate quantiles and quantiles And order When the short window is unavailable Set to null and mark as unusable for platform volatility features in step S5; The controller will sample points rate of change of the short window at the location , medium window change rate Long window rate of change Curvature characteristics Cumulative increase characteristics Platform volatility characteristics and invalid feature window flag By sampling time The feature sequence of the multi-scale curve is combined and written, and any feature item with a corresponding invalid feature window flag or null value is treated as an unusable feature in the subsequent step S5 to avoid low confidence window interfering with the triggering and confirmation of candidate endpoints.

[0025] In this specific embodiment, S5 includes: Controller not available at baseline The endpoint identification and confirmation are performed based on the multi-scale curve feature sequence, and the endpoint validity result and control reference time are output. The sampling point number, Sampling time, For the short window rate of change, For the rate of change of the middle window, This represents the cumulative increase over the long window. For platform volatility, Curvature characteristics These are the invalid feature window flags for short, medium, and long windows, respectively, with a value of 1 if the corresponding window is invalid and a value of 0 otherwise. The sampling point is frozen and the confidence weight of the sampling point is lower than the confidence weight threshold. hour otherwise ; The controller will satisfy all The sampling points are used as frozen sampling points and removed from the candidate endpoint triggering, continuous confirmation window calculation and interval consistency confirmation, and any features that need to be used are... or or Then the sampling point is considered unavailable in that feature dimension; The controller uses the sampling point number corresponding to the start time of oxidant addition. Starting from the initial stable window and ending with the sequence number. To define the learning interval, the starting point of the learning interval is... The learning interval endpoint is taken Only select those that meet the requirements within the learning interval. and The sampling points are used to construct the first adaptive rate of change threshold. And only select those that satisfy and The sampling points are used to construct the second adaptive rate of change inter-value. ; First adaptive rate of change threshold According to the rate of change of the middle window within the learning interval of Quantiles and Noise Scale It was jointly determined that, among which The noise scale obtained from S3 is in mV, and the controller calculates the noise level within the learning interval. of quantiles And order Second adaptive rate of change inter-value Based on the rate of change of the short window within the learning interval of Quantiles and Noise Scale Together, it was determined that the controller would calculate within the learning interval. of quantiles And order Thus satisfying ; Controller settings for cumulative increase threshold Platform volatility threshold Curvature threshold Interval consistency threshold ,in Units and Consistent and normalized dimensionless quantity, Let be the normalized dimensionless quantity of platform volatility. Units and Consistent and consisting of the rate of change difference component per sampling period, The number of sampling points and the corresponding time are and ; The controller satisfies and Candidate endpoint triggering is performed at the sampling points. The triggering rule is that the rate of change in the middle window crosses from the rising segment to the plateau segment at adjacent sampling points, i.e., it satisfies... and And simultaneously satisfy the cumulative increase over the long window. When the medium or long window used for triggering is an invalid feature window, i.e. or When an initial candidate endpoint is not generated, the endpoint validity result is directly set to invalid. When the sampling point number that first satisfies the triggering rule exists, it is recorded as the initial candidate endpoint number. Then it enters the continuous confirmation process, and the length of the continuous confirmation window is taken as... A continuous confirmation window is formed by sampling points and rolling with the sampling period as the step size, and the threshold for the number of confirmation windows is set to... And the confirmation window used for interval consistency verification is from Start continuous A confirmation window; Within each confirmation window, the controller again generates candidate endpoint numbers according to the triggering rules and combines the candidate endpoint numbers from each confirmation window into a candidate endpoint time set. ,in The set is a collection of candidate endpoint indices, where each element is a sampling point indices. The threshold for the number of candidates is set to... Furthermore, each confirmation window used for interval consistency verification must generate a candidate endpoint sequence number; otherwise, an interval consistency failure flag will be output and the endpoint validity result will be set to invalid. Within each confirmation window, the controller performs a pass / fail judgment on the continuous confirmation of rate of change, platform fluctuation confirmation, and curvature confirmation. The continuous confirmation of rate of change is judged as all conditions met within that confirmation window. and All sampling points satisfy The platform fluctuation confirmation judgment is based on all conditions that are met within the confirmation window. and All sampling points satisfy The curvature confirmation determination is based on all conditions within the confirmation window that satisfy the following criteria. and All sampling points satisfy Furthermore, when the short or medium window required for continuous confirmation cannot provide enough available sampling points within the confirmation window, thus preventing the formation of continuous [confirmation / continuity], [further issues arise]. When there is a valid confirmation window, the endpoint validity result is set to pending confirmation and the sampling point number corresponding to the control reference time is set to [value]. ; When the candidate endpoint time set The number of elements is not less than Furthermore, when the rate of change, platform fluctuation, and curvature confirmations for each confirmation window are all passed, the controller calculates the interval consistency of the candidate endpoint time set and uses the median of the candidate endpoint sequence number as the consistency center. The interval consistency is calculated according to the formula... Calculate, where, This is the upper bound of the deviation, and the unit is the number of sampling points. To perform the maximum value operation, Set of candidate endpoint times Any candidate endpoint number in the list, For the set of candidate endpoint indices, For absolute value operations, For set All The median index; when The output interval consistency flag is checked and the endpoint validity result is set to valid. Simultaneously, the sampling point number corresponding to the control reference time is set to [value]. ,when The output interval consistency failure flag is displayed, and the endpoint validity result is invalidated. When the initial candidate endpoint number has been identified Furthermore, no invalid judgment result has been formed yet, but the number of confirmation windows is insufficient. or candidate endpoint time set Insufficient number of elements If any confirmation window has not yet completed the confirmation of rate of change, platform fluctuation, or curvature, set the endpoint validity result to pending confirmation and set the sampling point number corresponding to the control reference time to [unclear]. .

[0026] In this specific embodiment, S6 includes: The controller is based on the endpoint validity results. Sampling point number corresponding to the control reference time And combined with a reliable weight sequence This generates dosing control commands and verification / rollback logic, among which... The sampling point number, Sampling time, The confidence weights of the sampling points are within a certain range. valid, pending, invalid These represent valid, pending confirmation, and invalid, respectively. The sequence number corresponding to the control reference time determined in step S5. The normalized ORP value generated in step S3; when At this time, the controller does not enter the stop or speed reduction judgment but directly outputs a delayed observation command and switches the feed pump flow rate setpoint to the observation flow rate. ,in The constant dosing flow rate is set under the delayed observation state, and data on the subsequent reaction process is collected under this state before entering the next judgment cycle; when When valid, the controller determines the control decision sampling interval as the union of the sets of sampling points covered by multiple confirmation windows that meet the interval consistency condition in step S5 and have not been removed by the frozen sampling point flag. ,when At that time, the controller will... The union of the sets of sampling points covered by the established continuous confirmation window that have not been removed by the frozen sampling point flag is determined as the control decision sampling interval. ,in It is a set of sampling point indices, and all its elements are sampling point indices. The controller sets a minimum number of samples for the control decision sampling interval. ,when Output a delayed observation command and switch the dosing pump flow rate setpoint to ,in For set The number of elements; when At that time, the controller performs dual threshold gating on the trusted weights and sets the sampling point to trigger the trusted weight threshold. With control of the confidence weight threshold ,in Used for single-sampling-point trigger exclusion. Used to control the exclusion of the mean of the decision sampling interval and satisfying ; The controller checks first. Does any sampling point within satisfy...? If present, a delayed observation command will be output and the setpoint of the injection pump flow rate will be switched to [value]. ; when All sampling points within the range satisfy At that time, the controller calculates Mean of internally credible weights ,in To be All Summing and dividing by The obtained arithmetic mean; when and At that time, the controller outputs an oxidant stop control command and sets the dosing pump flow rate setpoint to [value missing]. ,in Set the injection flow rate value for the stopped injection state; when pending and At this time, the controller outputs an oxidant deceleration control command and sets the dosing pump flow rate setpoint to the current dosing pump flow rate setpoint. The deceleration injection command reduces the injection flow rate to half of the current setting while keeping the injection pump running, thus entering a cautious injection state. when valid, pending and At that time, the controller outputs a delayed observation command and switches the setpoint of the injection pump flow rate to... ; The sampling point number issued by the controller under any control command is denoted as: And perform the final review within the review window, the length of which is [value missing]. And the corresponding number of sampling points The threshold for the rate of change was verified. The cumulative change threshold for review is taken as follows. And calculate the review indicators according to the following formula: , ; in, To verify the subsequent rate of change of ORP within the review window, To verify the cumulative change in ORP within the verification window, To control the normalized ORP value at the time of instruction issuance, To verify the normalized ORP value at the end of the verification window, To verify the number of sampling points in the window, For the sampling period and take ; when or At that time, the controller generates an endpoint verification anomaly flag. ,in This serves as a check for abnormalities, and when a check is abnormal... otherwise and in Furthermore, when the current control command is an oxidant stop control command or an oxidant deceleration control command, the current endpoint control result is revoked and the control state corresponding to the delayed observation command is switched. In the delayed observation state, steps S2 to S6 are re-executed with the latest data or the next judgment cycle is entered. when The system generates an endpoint verification pass flag and maintains the status. When the current control command is the oxidant deceleration control command, it maintains the deceleration state, continues to collect subsequent reaction process data, and repeats steps S2 to S6 until the endpoint validity result becomes valid. Then, it outputs the oxidant stop control command or outputs the delayed observation command when the judgment condition is not met. When the current control command is the delayed observation command, it maintains the delayed observation state, continues to collect subsequent reaction process data, and repeats steps S2 to S6. When the current control command is the oxidant stop control command, it maintains the stop state and continues to record subsequent reaction process data for batch traceability.

[0027] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0028] This invention combines intra-batch normalization of online ORP curves, multi-scale feature extraction, and event credibility weight gating into the same control flow, so that endpoint candidates no longer depend on a single fixed threshold, but are jointly limited by the rising segment, plateau segment, and consistency of multiple confirmation windows of the current batch response curve.

[0029] The key improvement of this invention lies in transforming on-site disturbances such as feeding, liquid replenishment, pump and valve pulsation, agitation changes, and ORP jumps into reliable weights, frozen windows, or invalid feature windows. By waiting for confirmation of deceleration, delayed observation, and verification of anomalies, a control backtracking path is formed, thereby better achieving stable identification and addition control of the ferric pyrophosphate oxidation reaction endpoint.

Claims

1. A method for determining and controlling the online ORP endpoint of the ferric pyrophosphate oxidation reaction, characterized in that, include: S1. Collect raw ORP sequence, temperature sequence, pH sequence, dosing flow rate sequence, stirring speed sequence, feeding event and replenishment event signals to form reaction process data; S2. Robustly filter the original ORP sequence and generate reliable weights for sampling points based on feeding events, replenishment events, feeding flow rate sequences, and stirring speed sequences to obtain the filtered ORP sequence and reliable weight sequence. S3. Determine the start time of oxidant addition based on the addition flow rate sequence, determine the initial stabilization window after that time based on the filtered ORP sequence, and determine the ORP baseline, noise scale, temperature and pH compensation amount for this batch based on this window, the filtered ORP sequence, and the pH sequence to obtain the normalized ORP sequence. S4. Within at least two time-sliding windows, calculate the rate of change, curvature, cumulative rise, and plateau volatility based on the normalized ORP sequence to obtain a multi-scale curve characteristic sequence. S5. Based on the multi-scale curve feature sequence, identify the candidate endpoints of the rising segment entering the plateau segment. After continuous confirmation of the rate of change, confirmation of plateau fluctuation, and confirmation of interval consistency, obtain the endpoint validity result, and obtain the control reference time when the result is valid or pending confirmation. S6. Based on the endpoint validity results, and when the results are valid or pending confirmation, combine the control baseline time and the credible weight sequence to output instructions for stopping oxidant addition, slowing down oxidant addition, or delaying observation. After execution, review the subsequent ORP changes.

2. The method for determining and controlling the online ORP endpoint of the ferric pyrophosphate oxidation reaction according to claim 1, characterized in that, S1 includes: Data is collected from the ORP electrode, temperature sensor, pH sensor, oxidant dosing pump, stirring actuator, feeding valve and replenishing valve according to the preset sampling cycle; The acquisition time, sensor values, actuator status, and event trigger flag are written into the same timeline; a missing flag is set for missing sampling points while maintaining the time order of their adjacent valid sampling points, and the missing flag is used as the data quality flag for input step S2 to obtain the reaction process data.

3. The method for determining and controlling the online ORP endpoint of the ferric pyrophosphate oxidation reaction according to claim 2, characterized in that, S2 includes: Within a filtering window centered on the current sampling point, the median value of the ORP and the absolute deviation scale are determined. When the absolute value of the difference between the current sampling point and the median value of the ORP is greater than the peak determination threshold determined by the absolute deviation scale, the current sampling point is marked as a peak point. The peak point and the sampling point with the missing label are replaced by the weighted value of adjacent valid sampling points, and the sampling points that are not peak points and do not have the missing label are smoothed by Huber weights to obtain the filtered ORP sequence. The reliable weight sequence is generated based on the peak point label, the missing label, the feeding event, the liquid replenishment event, and the pump start / stop label, the flow change label, and the speed step label identified by the feeding flow rate sequence and the stirring speed sequence.

4. The method for determining and controlling the online ORP endpoint of the ferric pyrophosphate oxidation reaction according to claim 1, characterized in that, S3 includes: selecting a continuous sampling interval within a preset observation period after the oxidant is added, where the confidence weight is greater than the initial stabilization window confidence weight threshold and the absolute value of the ORP change rate determined according to the filtered ORP sequence within the preset window is less than the baseline change rate threshold, as the initial stabilization window; the initial stabilization window confidence weight threshold is a preset threshold within the range of the confidence weight values ​​of the sampling points, and the range of the confidence weight values ​​of the sampling points is a closed interval between a preset lower limit and a preset upper limit of the confidence weight of the sampling points in the confidence weight sequence; determining the ORP baseline of this batch based on the filtered ORP mean within the initial stabilization window; determining the noise scale based on the filtered ORP residual quantile within the initial stabilization window; determining the temperature compensation amount based on the temperature deviation of the temperature sequence relative to the temperature mean of the initial stabilization window and a preset temperature compensation coefficient or temperature calibration compensation curve; and determining the pH compensation amount based on the pH deviation of the pH sequence relative to the pH mean of the initial stabilization window and a preset pH... The pH compensation amount is determined by the compensation coefficient or pH calibration compensation curve; the filtered ORP sequence is corrected and scaled using the current batch ORP baseline, noise scale, temperature and pH compensation amount to obtain the normalized ORP sequence; when the initial stable window is not obtained within the preset observation period, a baseline unavailable flag is generated; during the period when the baseline unavailable flag exists, steps S1 to S3 are continued to update the reaction process data, filtered ORP sequence, and reliable weight sequence, and the initial stable window is searched again, the multi-scale curve feature extraction in step S4 and the candidate endpoint triggering in step S5 are paused, the endpoint validity result is set to invalid, and a delayed observation instruction is output in step S6, and the endpoint control verification dependent on the normalized ORP sequence is temporarily not executed. The temporary suspension of the endpoint control verification dependent on the normalized ORP sequence means that the verification calculation based on the normalized ORP sequence is paused until the initial stable window is obtained again and steps S4 to S6 and the corresponding endpoint verification are resumed.

5. The method for determining and controlling the online ORP endpoint of the ferric pyrophosphate oxidation reaction according to claim 4, characterized in that, S4 includes: Linear fitting was performed on the sampling points in the normalized ORP sequence whose confidence weights were greater than the confidence weight threshold in the short window, medium window, and long window respectively, to obtain the rate of change of each window. The determination confidence weight threshold is a preset threshold that is within the range of the confidence weight values ​​of the sampling points and is independent of the initial stable window confidence weight threshold. When the proportion of sampling points whose confidence weights are not greater than the determination confidence weight threshold is greater than the preset low confidence proportion threshold, the window is set as an invalid feature window that is not used in step S5. The curvature is determined based on the difference in the rate of change between adjacent time windows; the cumulative increase is determined based on the difference in normalized ORP between the current window endpoint and the window start point; the plateau volatility is determined based on the quantile difference of the normalized ORP residuals within the current window; and the rate of change, curvature, cumulative increase, and plateau volatility are combined according to the sampling time to obtain the multi-scale curve feature sequence.

6. The method for determining and controlling the online ORP endpoint of the ferric pyrophosphate oxidation reaction according to claim 5, characterized in that, S5 includes: Sampling points that do not trigger endpoint determination, determined based on the trusted weight sequence and according to the trusted weight threshold, are designated as frozen sampling points. Sampling points participating in the candidate endpoint and confirmation window calculations do not include these frozen sampling points. The short, medium, and long windows used for candidate endpoint triggering and continuous confirmation do not include the invalid feature window. When the medium or long window required for candidate endpoint triggering is an invalid feature window, no initial candidate endpoint time is generated, and the endpoint validity result is determined as invalid. When the medium window change rate changes from greater than the first adaptive change rate threshold to less than the second adaptive change rate threshold, and the cumulative increase of the long window is greater than the cumulative increase threshold, the corresponding sampling time is determined as the initial candidate endpoint time. The first adaptive change rate threshold is used for rising segment confirmation and is determined based on the noise scale and the quantile of the medium window change rate within the preset learning interval after the initial stable window. The second adaptive change rate threshold is used for platform segment confirmation and is determined based on the noise scale and the quantile of the short window change rate within the preset learning interval. The second adaptive change rate threshold is not greater than the first adaptive change rate threshold. Within the continuous confirmation window after the initial candidate endpoint time, if the short window change rate is continuously less than the second adaptive change rate threshold... If the rate of change threshold, platform volatility less than the platform volatility threshold, absolute curvature less than the curvature threshold, and the candidate endpoint time set generated within multiple confirmation windows satisfies the consistency condition corresponding to the interval consistency threshold, then the endpoint validity result is determined to be valid, and the median time of the candidate endpoint time set is determined as the control reference time for step S6. If the initial candidate endpoint time is not identified, the cumulative increase in the long window is not greater than the cumulative increase threshold, the formed platform volatility confirmation result indicates that the platform volatility is not less than the platform volatility threshold, the absolute curvature is not less than the curvature threshold, or the formed candidate endpoint time set does not satisfy the consistency condition, then the endpoint validity result is determined to be invalid. If the initial candidate endpoint time has been identified but the aforementioned invalid determination result has not yet been formed, but the number of continuous confirmation windows has not reached the preset number of confirmation windows, the continuous confirmation of the rate of change has not been completed, the platform volatility confirmation has not been formed, the interval consistency confirmation has not been formed, or the necessary window required for continuous confirmation is an invalid feature window and cannot form a preset number of valid confirmation windows, then the endpoint validity result is determined to be pending confirmation, and the initial candidate endpoint time is determined as the control reference time for step S6.

7. The method for determining and controlling the online ORP endpoint of the ferric pyrophosphate oxidation reaction according to claim 6, characterized in that, S6 include: When the endpoint validity result is invalid, a delayed observation instruction is output; when the endpoint validity result is valid, the set of sampling points covered by multiple confirmation windows that meet the interval consistency condition in step S5 is determined as the control decision sampling interval; when the endpoint validity result is pending confirmation, the set of sampling points covered by the currently formed continuous confirmation window after the control reference time is determined as the control decision sampling interval; if the number of sampling points in the control decision sampling interval is less than the preset minimum number of samples, the delayed observation instruction is output; both the sampling point trigger confidence weight threshold and the control confidence weight threshold are within the range of the sampling point confidence weight. The preset threshold within the value range, the range of the sample point confidence weight is a closed interval between the preset lower limit and the preset upper limit of the sample point confidence weight in the confidence weight sequence, the sample point trigger confidence weight threshold is used for single sample point trigger exclusion, the control confidence weight threshold is used for control decision sampling interval mean exclusion, and the control confidence weight threshold is greater than the sample point trigger confidence weight threshold; firstly, it is determined whether the confidence weight of each sample point in the control decision sampling interval is greater than the sample point trigger confidence weight threshold; if the confidence weight of any sample point is not greater than the sample point trigger confidence weight threshold, then the delayed observation instruction is output; If the confidence weight of each sampling point is greater than the confidence weight threshold triggered by the sampling point and the endpoint validity result is valid, and the average confidence weight within the control decision sampling interval is greater than the control confidence weight threshold, then an oxidant stop control command is output; if the confidence weight of each sampling point is greater than the confidence weight threshold triggered by the sampling point and the endpoint validity result is pending confirmation, and the average confidence weight within the control decision sampling interval is greater than the control confidence weight threshold, then an oxidant deceleration control command is output to reduce the oxidant dosing rate; if the confidence weight of each sampling point is greater than the confidence weight threshold triggered by the sampling point and the endpoint validity result is valid or pending confirmation... If it is confirmed that the mean of the confidence weights within the control decision sampling interval is not greater than the control confidence weight threshold, then the delayed observation instruction is output; the delayed observation instruction is an observation control instruction that prohibits triggering oxidant stopping or oxidant slowing down based on the current judgment period and switches the oxidant dosing flow rate to a preset observation flow rate; in the review window after the control instruction is executed, the subsequent ORP change rate and subsequent ORP cumulative change are calculated based on the normalized ORP sequence. If the subsequent ORP change rate is greater than the review change rate threshold or the subsequent ORP cumulative change is greater than the review cumulative change threshold, then an endpoint review anomaly flag is generated; if the endpoint is generated... If the verification anomaly flag is detected, and the current control command is an oxidant stop control command or an oxidant rate reduction control command, the current endpoint control result is revoked and the control state of executing the delayed observation command is switched. In this delayed observation state, steps S2 to S6 are re-executed or the next judgment cycle is entered. If the current control command is the delayed observation command, the delayed observation state is maintained, and steps S2 to S6 are re-executed or the next judgment cycle is entered. If the subsequent ORP change rate is not greater than the verification change rate threshold and the subsequent cumulative ORP change is not greater than the verification cumulative change threshold, an endpoint verification pass flag is generated. If the endpoint verification pass flag is generated, then when the current control command is the oxidant deceleration control command, the deceleration state is maintained and subsequent reaction process data is collected. Steps S2 to S6 are re-executed or the next judgment cycle is entered until the endpoint validity result is valid, at which point the oxidant stop control command is output, or the delay observation command is output when the judgment condition is not met. When the current control command is the delay observation command, the delay observation state is maintained and subsequent reaction process data is collected. Steps S2 to S6 are re-executed or the next judgment cycle is entered. When the current control command is the oxidant stop control command, the control state corresponding to the current control command is maintained.

8. The method for determining and controlling the online ORP endpoint of the ferric pyrophosphate oxidation reaction according to claim 5, characterized in that, The generation of the reliable weight sequence includes: generating data quality flags from the acquisition status in the reaction process data, and using the data quality flags corresponding to missing sampling points as missing flags; using a preset sampling window around the current sampling point as a filtering window, determining a peak judgment threshold based on the ORP absolute deviation scale within the filtering window, and marking sampling points where the absolute value of the difference between the current sampling point and the ORP median value within the filtering window is greater than the peak judgment threshold as ORP peak point markers; mapping the feeding event, liquid replenishment event, oxidant dosing pump start / stop event and flow change event identified by the feeding flow sequence, stirring speed step event identified by the stirring speed sequence, the ORP peak point markers, and the missing flags as event weight decay amounts respectively; performing time decay on the event weight decay amount according to the length of the freeze window after the event occurrence time; when there is no time decay event weight decay amount for the same sampling point, the aggregation event weight decay amount is determined to be zero; when there is a time decay for the same sampling point... When calculating the event weight decay amount after time decay, the event weight decay amount after time decay is determined as the aggregated event weight decay amount. When there are multiple event weight decay amounts after time decay for the same sampling point, the event weight decay amounts after time decay are summed and truncated according to a preset decay upper limit to obtain the aggregated event weight decay amount. The benchmark weight value is the upper limit of the preset credible weight range. The preset decay upper limit does not exceed the benchmark weight value. The credible weight of the sampling point is obtained by subtracting the aggregated event weight decay amount from the benchmark weight value, and the credible weight of the sampling point is limited to the preset credible weight range. When the credible weight of the sampling point is less than the credible weight threshold, the corresponding sampling point is set as a frozen sampling point that does not trigger the endpoint determination. The frozen sampling point is not used as the candidate endpoint time in step S5, and is not used to determine the endpoint validity result as valid. The lower limit to the upper limit of the preset credible weight range constitutes the credible weight value range of the sampling point, which is used for endpoint determination and control of the credible weight threshold value in the credible judgment.

9. The method for determining and controlling the online ORP endpoint of the ferric pyrophosphate oxidation reaction according to claim 6, characterized in that, The candidate endpoint time set generated within the multiple confirmation windows satisfies the consistency condition corresponding to the interval consistency threshold, including: forming at least three consecutive confirmation windows with the sampling period as the step size; generating candidate endpoint times within each confirmation window according to the rule that the change rate of the middle window changes from greater than the first adaptive change rate threshold to less than the second adaptive change rate threshold and the cumulative increase of the long window is greater than the cumulative increase threshold, and forming the candidate endpoint time set; each confirmation window used for interval consistency confirmation generates candidate endpoint times, and the number of elements in the candidate endpoint time set is not less than a preset candidate number threshold; if any confirmation window used for interval consistency confirmation does not generate a candidate endpoint time or the number of elements in the candidate endpoint time set is less than the preset candidate number threshold, then an interval consistency failure flag is output; for each confirmation window, if the change rate of the short window within the confirmation window is continuously less than the second adaptive change rate threshold, then a change rate is generated. If the confirmation result is passed, the rate of change confirmation result is failed; if the platform volatility within the confirmation window is less than the platform volatility threshold, the platform volatility confirmation result is passed, otherwise the platform volatility confirmation result is failed; if the absolute value of curvature within the confirmation window is less than the curvature threshold, the curvature confirmation result is passed, otherwise the curvature confirmation result is failed; calculate the absolute value of the deviation between each candidate endpoint time in the candidate endpoint time set and its median time, and take the absolute value of the deviation that satisfies that no other absolute value of deviation is greater than this value as the upper bound value of the deviation; when the upper bound value of the deviation is less than the interval consistency threshold, and the rate of change confirmation result, platform volatility confirmation result, and curvature confirmation result of each confirmation window are all passed, output the interval consistency pass flag, and take the median time as a candidate time for determining the control reference time; the interval consistency pass flag is a necessary condition for determining the endpoint validity result as valid.

10. An online ORP endpoint determination and control system for iron pyrophosphate oxidation reaction, used to execute the online ORP endpoint determination and control method for iron pyrophosphate oxidation reaction as described in any one of claims 1 to 9, characterized in that, include: A data acquisition module is used to generate the reaction process data; a robust filtering and event weighting module is used to obtain the filtered ORP sequence and the reliable weight sequence. An in-batch calibration module is used to obtain the normalized ORP sequence; A curve feature extraction module is used to obtain the multi-scale curve feature sequence; The endpoint confirmation module is used to obtain the endpoint validity result and control reference time; the dosing control and verification module is used to output the oxidant stop control command, oxidant deceleration dosing control command or delayed observation command according to the endpoint validity result, control reference time and reliable weight sequence, and verify the endpoint status.

Images