Adaptive filtering-based ultrapure water quality abnormal data processing method and system

Abstract

The application discloses a kind of based on adaptive filtering's ultrapure water quality abnormal data processing method and system, it is related to data processing technical field, method includes: obtaining target water quality detection channel sampling record, eliminating missing sampling, communication abnormal sampling and keeping display sampling, generate effective sampling chain;Combining historical stable window generates sorting position record and symbol section record, and based on adjacent process section similar detection channel generates process section receipt record;Again generate processing mark and filter output data, and update historical stable window and dynamic boundary record.The application solves the problem that ultrapure water quality instantaneous impact, back and forth oscillation, sustained drift and keeping display sampling are difficult to distinguish, reduces the influence of abnormal data on dynamic boundary.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method and system for processing abnormal ultrapure water quality data based on adaptive filtering. Background Technology

[0002] Ultrapure water pipelines are typically arranged with water quality testing points in the order of process sections such as reverse osmosis outlet, electro-deionization module outlet, and refining and polishing bed outlet, and continuously collect sampling records of target water quality testing channels such as conductivity, total organic carbon, and dissolved oxygen.

[0003] In actual operation, the sampling records of the target water quality monitoring channels are not always equivalent to the actual water quality changes. For example, the conductivity detection channel at the outlet of the refining and polishing bed may be stable around 0.0550 μS / cm during normal operation, but due to electromagnetic interference or communication jitter, a single point value of 0.0815 μS / cm may suddenly appear in a certain sampling cycle, and then return to 0.0551 μS / cm in the next sampling cycle. If an alarm is triggered only based on the sudden increase in value, it will cause unnecessary alarm output. As another example, the response cycle of the total organic carbon detection channel is relatively long. The plant monitoring interface may refresh the display using the previous output value of 2.8 μg / L when no new valid sample value is obtained. If this displayed sample is mistakenly regarded as a new valid sample value and included in the stabilization window calculation, it will incorrectly increase the number of repeated stabilization samples, making the subsequent dynamic boundary records too narrow. For example, the conductivity detection channel at the outlet of the polishing bed may slowly move from 0.0550 μS / cm to 0.0570 μS / cm. If there is no similar change in the same direction at the outlet of the upstream electro-deionization module, it may be due to local probe contamination or abnormal detection link. If the upstream detection channel with the same direction changes first and is transmitted to the downstream after a reasonable propagation and waiting relationship, it may be due to actual process fluctuations.

[0004] Meanwhile, existing fixed threshold, simple mean filtering, or single-point residual judgment methods mostly use whether the value exceeds the limit as the processing basis. It is difficult to simultaneously utilize the changes in sorting position, symbol segment structure, and sequence retrieval relationship of the ultrapure water process section within the same detection channel. This makes it difficult to distinguish between sensor disturbances, channel oscillations, real process fluctuations, and local link anomalies. Furthermore, abnormal values, held display values, or temporary correction values ​​are prone to entering subsequent historical stable windows and dynamic boundary records. Summary of the Invention

[0005] This invention provides a method and system for processing abnormal ultrapure water quality data based on adaptive filtering, which solves the technical problem that it is difficult to distinguish between instantaneous impacts, reciprocating oscillations, continuous drift and sustained display sampling in the target water quality detection channel in ultrapure water closed-loop pipelines, and that abnormal outputs can easily affect subsequent dynamic boundary records.

[0006] A method for processing abnormal ultrapure water quality data based on adaptive filtering includes: acquiring sampling records of target water quality detection channels; removing missing samples, communication anomaly samples, and samples that are kept on display according to data quality markers; generating a valid sampling chain according to process segment name and detection channel name; generating sorting position records based on the valid sampling chain and the historical stable window of the same process segment and the same detection channel, and generating symbol segment records based on changes in adjacent sorting positions; retrieving symbol segment records of the same type of detection channel in adjacent process segments based on the process segment sequence records of the symbol segment records and the reverse osmosis outlet, the electro-deionization module outlet, and the refining polishing bed outlet, and generating process segment receipt records; generating processing markers based on the symbol segment records and process segment receipt records; generating filtered output data based on the processing markers; and updating the historical stable window and dynamic boundary records according to the benchmark participation markers; the target water quality detection channel includes at least one of conductivity detection channel, total organic carbon detection channel, and dissolved oxygen detection channel.

[0007] The beneficial effects of this invention are reflected in:

[0008] By generating an effective sampling chain, missing samples, communication anomaly samples, and samples held for display are prevented from entering subsequent processing, avoiding the misinterpretation of repeated values ​​from low-frequency detection channels as stable and valid samples. Through sorting position records and symbol segment records, continuous numerical changes in the target water quality detection channel are converted into structured segments of rising, falling, and stable symbols, without merging through positive and negative cancellation. Therefore, single-point impacts, reciprocating oscillations, and continuous drift can be processed separately according to the segment structure. Through process segment feedback records, the symbol segment records of the target process segment can be sequentially verified with the symbol segment records of similar detection channels in adjacent process segments, thereby determining whether changes propagate sequentially along the process segments of the reverse osmosis outlet, electro-deionization module outlet, and refining and polishing bed outlet. By processing markers, filtering output data, and benchmark participation markers, real-time display, temporary correction, final backfill, limited output, and historical stable window updates are separated, preventing abnormal original values, temporary correction values, held display values, channel oscillation limited outputs, and maintenance limited outputs from entering the dynamic boundary records. Attached Figure Description

[0009] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0010] Figure 1 This is a schematic diagram illustrating the steps of the ultrapure water quality anomaly data processing method based on adaptive filtering according to the present invention; Figure 2This is a schematic diagram of some steps in S1 of the ultrapure water quality anomaly data processing method based on adaptive filtering of the present invention; Figure 3 This is a schematic diagram of the data flow timing in S2 of the ultrapure water quality anomaly data processing method based on adaptive filtering in this invention; Figure 4 This is a schematic diagram of part of step S2 in the ultrapure water quality anomaly data processing method based on adaptive filtering of the present invention; Figure 5 This is a schematic diagram of the data flow timing in S3 of the ultrapure water quality anomaly data processing method based on adaptive filtering in this invention; Figure 6 This is a schematic diagram of part of step S3 in the ultrapure water quality anomaly data processing method based on adaptive filtering of the present invention; Figure 7 This is a schematic diagram of the data flow timing in S4 of the ultrapure water quality anomaly data processing method based on adaptive filtering in this invention; Figure 8 This is a schematic diagram of part of step S4 in the ultrapure water quality anomaly data processing method based on adaptive filtering of the present invention; Figure 9 This is a schematic diagram of the module structure and data flow of the ultrapure water quality anomaly data processing system based on adaptive filtering according to the present invention. Detailed Implementation

[0011] 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. It should be understood that the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0012] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0013] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0014] This invention provides a method for processing abnormal ultrapure water quality data based on adaptive filtering, such as... Figure 1 As shown, in one embodiment, the method includes: S1. Obtain the sampling records of the target water quality testing channel, remove missing samples, communication anomaly samples, and samples that are kept on display according to the data quality mark, and generate a valid sampling chain according to the process section name and the testing channel name; S2. Generate sorting position records based on the effective sampling chain and the historical stable window of the same process section and the same detection channel, and generate symbol segment records based on the changes of adjacent sorting positions; S3. Based on the symbol segment records and the process segment sequence records of reverse osmosis outlet, electro-deionization module outlet, and fine polishing bed outlet, retrieve the symbol segment records of the same detection channels in adjacent process segments that actually exist, generate process segment receipt records, and generate processing marks based on the symbol segment records and process segment receipt records. S4. Generate filtered output data based on the processing markers, and update the historical stable window and dynamic boundary records according to the benchmark participation markers.

[0015] The target water quality testing channels include at least one of the following: conductivity testing channel, total organic carbon testing channel, and dissolved oxygen testing channel. Process section names include reverse osmosis outlet, electro-deionization module outlet, and refining and polishing bed outlet. To ensure that testing data of different dimensions do not undergo unreasonable merging, this implementation method only generates historical stability windows and sorting position records within the same process section and the same testing channel; between different process sections, only the symbol segment direction, segment start point, segment end point, and time sequence of similar testing channels are compared, without directly adding or weighting the sampled values ​​of conductivity, total organic carbon, and dissolved oxygen.

[0016] When an ultrapure water pipeline is configured with only one target water quality detection channel, the generation of valid sampling chains, sorting position records, symbol segment records, and process segment receipt records are performed only on that target water quality detection channel. When multiple target water quality detection channels are configured, each target water quality detection channel performs the above processing independently, without directly comparing the sampled values ​​between different target water quality detection channels. If an adjacent process segment is not configured with the same target water quality detection channel as the target process segment, a state of "no similar detection channel" is recorded, and an isolated record or a record to be confirmed is generated according to the case where no similar detection channel exists in the adjacent process segment. For example, when only a conductivity detection channel is configured, the conductivity detection channel can independently generate a process segment receipt record; when both a conductivity detection channel and a total organic carbon detection channel are configured, the two types of detection channels generate their own symbol segment records and process segment receipt records respectively.

[0017] Furthermore, adaptive filtering refers to not directly replacing sampled values ​​with fixed water quality thresholds, but generating sorted position records and dynamic boundary records based on the historical stability windows of the same process section and the same detection channel. It then determines whether the current sampled record is a normal output record, a temporary correction output, a final backfill output, a channel oscillation limit output, a process fluctuation limit output, a maintenance limit output, or a record pending confirmation based on the processing flag. Among these, outputs that are allowed to participate in the benchmark update are included in the subsequent historical stability window, while outputs that are not allowed to participate in the benchmark update are not included in the historical stability window.

[0018] In this embodiment, it should be noted that in S1, the sampling records of the target water quality detection channel are first decomposed into processable data objects. Each sampling record includes at least the process section name, detection channel name, sampling time, sampling value, and data quality marker. Taking a scenario with a uniform processing cycle of 5 seconds as an example, if the conductivity detection channel at the outlet of the refining polishing bed outputs 0.0549 μS / cm, 0.0550 μS / cm, and 0.0551 μS / cm at 10:00:00, 10:00:05, and 10:00:10 respectively, and the data quality markers are all valid sampling markers, then these records can enter the same valid sampling chain. If the total organic carbon detection channel outputs 1.8 μg / L at 10:00:00, but only refreshes the display using the previous output value at 10:00:05 and 10:00:10, then the latter two refresh records are marked as maintaining the display sampling and do not enter the valid sampling chain. Through this elimination logic, S1 separates the sampling records with valid sampling marks from the samples that are kept for display, the samples with communication anomalies, and the missing samples, thereby avoiding the increase in the number of stable samples by the repeated display values ​​of the low-frequency detection channel.

[0019] Data quality marker filtering precedes the generation of sorting position records and symbol segment records. If a sampling record has a missing sampling marker, a communication anomaly sampling marker, or a sustained display sampling marker, then that sampling record will not enter the valid sampling chain, nor will it participate in the generation of impact segment records, oscillation segment records, or drift segment records. Only sampling records with valid sampling markers will enter the subsequent generation of sorting position records. For example, if the conductivity sampling value of 0.0665 μS / cm at 10:00:15 also has a communication anomaly sampling marker, then that record will be discarded and no sensor disturbance processing marker will be generated; if the record has a valid sampling marker, then it will enter the sorting position records, and an impact segment record can be generated based on the subsequent recovery status.

[0020] In S2, a sorting position record is generated based on the effective sampling chain obtained in S1 and the historical stable window of the same process section and detection channel. The historical stable window only contains normal output records that have been processed and are allowed to participate in the benchmark update, as well as the final backfill output records. For example, the historical stable window of the conductivity detection channel at the exit of the finishing polishing bed contains stable values ​​such as 0.0548 μS / cm, 0.0549 μS / cm, 0.0550 μS / cm, 0.0551 μS / cm, and 0.0552 μS / cm. When the current effective sampling value is 0.0665 μS / cm, its sorting position after being inserted into the historical stable window moves to the upper boundary of the sorting; when the next effective sampling value is 0.0551 μS / cm, its sorting position returns to the main sorting position range.

[0021] S2 generates rising, falling, and stable symbols based on changes in adjacent sorting positions, and compresses consecutive identical symbols into symbol segment records. This step transforms the single-point judgment of "whether the value is greater than a fixed threshold" into a structural judgment of "how the sorting position moves and how the symbol segments are arranged," allowing shocks, oscillations, and drifts to be handled separately according to segment structures.

[0022] In S3, the symbol segment records generated in S2 are not directly used as the final anomaly conclusion. Instead, the process segment records of the reverse osmosis outlet, the electro-deionization module outlet, and the refining polishing bed outlet are combined to retrieve the symbol segment records of the same type of detection channel in adjacent process segments. For example, if the conductivity detection channel at the refining polishing bed outlet forms a single-point rising impact segment at 10:00:15, while the conductivity detection channel at the electro-deionization module outlet does not have a unidirectional impact segment or a unidirectional drift segment within the corresponding historical propagation waiting record range, an isolated record is generated, which further generates a sensor disturbance processing mark.

[0023] For example, if the total organic carbon (TOC) detection channel at the outlet of the electro-deionization module shifts from 2.2 μg / L to 3.0 μg / L at 14:20:00, and the TOC detection channel at the outlet of the refining and polishing bed shifts from 1.6 μg / L to 2.3 μg / L at 14:20:30, and both fall within the historical propagation waiting period for recording within 30 seconds, then an upstream or downstream acknowledgment record is generated, and a process fluctuation processing flag is further generated. This step uses the process segment sequence of the ultrapure water pipeline to identify whether the change conforms to the historical propagation relationship, avoiding directly treating water quality changes propagating along the process segment as sensor disturbances.

[0024] In S4, filtered output data is generated based on the processing markers generated in S3, and the historical stability window and dynamic boundary records are updated through the reference participation marker control. If a sensor disturbance processing marker is generated, a temporary correction output is generated based on the previous stable segment of the impact segment. The final backfill output is generated after the next stable segment of the impact segment is formed. The temporary correction output is only used for real-time display and does not participate in the historical stability window update. The final backfill output can replace the temporary correction output and is allowed to enter the historical stability window. If a channel oscillation processing marker is generated, a channel oscillation limiting output is generated, prohibiting it from entering the historical stability window.

[0025] If a process fluctuation handling flag is generated, a process fluctuation limit output is first generated. This output is only allowed to participate in baseline updates after the corresponding acknowledgment record remains valid throughout the period from the start of the drift segment to the closure of the next stable segment of the target process segment, and after candidate output verification is completed. If a local link anomaly handling flag is generated, a maintenance limit output and a maintenance prompt record are generated. This process separates display control, backfill control, and dynamic boundary update control to prevent abnormal original sampled values, temporary correction values, display records corresponding to maintained display samples, and unconfirmed drift values ​​from entering subsequent dynamic boundary records.

[0026] In summary, this invention constructs a data processing link through an effective sampling chain, sorting position records, symbol segment records, process segment feedback records, processing markers, filtered output data, and benchmark participation markers. This enables the segmentation, verification, and output control of abnormal sampling in the ultrapure water target water quality detection channel. This method does not rely on training models or uninterpretable inference processes and can be implemented by those skilled in the art using conventional computer data processing methods. Furthermore, it can reduce unnecessary alarm outputs, prevent outliers from entering dynamic boundary records, and preserve genuine process fluctuations consistent with the propagation relationships of process segments in plant ultrapure water closed-loop pipeline scenarios.

[0027] like Figure 2 As shown, in one specific embodiment, S1 includes:

[0028] S11. Generate aligned sampling records based on the sampling time and uniform processing cycle in the sampling records.

[0029] S12. Based on the data quality markers, retain the aligned sampling records with valid sampling markers and remove the aligned sampling records with missing sampling markers, communication abnormal sampling markers, or retained display sampling markers.

[0030] S13. Based on the process segment name and detection channel name in the retained alignment sampling record, generate the valid sampling chain for the corresponding process segment and the corresponding detection channel respectively.

[0031] In this embodiment, it should be noted that in S11, the sampling records do not directly enter the subsequent algorithms. Instead, aligned sampling records are first generated based on the sampling time and the unified processing cycle. The unified processing cycle is used to group the sampling records of different detection channels into the same processing time axis, but this grouping process does not change the sampling values ​​themselves. The unified processing cycle is determined by the refresh cycle of the plant monitoring interface, the effective sampling interval of each target water quality detection channel, and the arrival interval of industrial bus messages. If the effective sampling intervals of multiple detection channels are the same, the unified processing cycle takes that effective sampling interval; if the effective sampling intervals of different detection channels are different, the unified processing cycle takes a processing interval that can cover the main sampling refresh cycle without changing the original sampling time sequence. When there are no new sampling values, the low-frequency detection channel only forms a display sampling record and does not form an effective sampling chain record.

[0032] For example, if the conductivity detection channel generates a valid sample every 5 seconds, the total organic carbon detection channel generates a valid sample every 30 seconds, and the plant monitoring interface refreshes every 5 seconds, then the unified processing cycle can be set to 5 seconds. Interface refresh values ​​formed between two valid total organic carbon samples are considered as displayed samples and do not enter the valid sampling chain. Taking a unified processing cycle of 5 seconds as an example, the conductivity detection channel at the outlet of the refining polishing bed outputs 0.0549 μS / cm at 10:00:00.2, 0.0550 μS / cm at 10:00:05.1, and 0.0551 μS / cm at 10:00:10.1. These sampling times fall within the unified processing cycles corresponding to 10:00:00, 10:00:05, and 10:00:10, respectively, thus forming corresponding aligned sampling records.

[0033] The reason for adopting this logic is that the arrival times of data from different instruments in the ultrapure water field may have slight deviations. If the sampling times are not aligned first, it is impossible to determine the sequential relationship between changes in a certain process segment and changes in adjacent process segments in subsequent steps. This step ensures that subsequent valid sampling chains, symbol segment records, and process segment acknowledgment records are all generated based on a unified timeline, reducing the impact of message arrival time differences on the determination of process segment acknowledgment relationships.

[0034] In S12, aligned sampling records are filtered based on data quality tags. Aligned sampling records with valid sampling tags are retained; those with missing sampling tags, communication error sampling tags, or persistent display sampling tags are discarded. Data quality tags are generated based on the sampling message reception result, sampling value refresh status, and device status word. If no sampling message for the corresponding process segment and corresponding detection channel is received within the unified processing cycle, a missing sampling tag is generated; if a sampling message is received but message verification fails, the device status word indicates a communication failure, or the sampling value field cannot be parsed, a communication error sampling tag is generated.

[0035] If the current displayed value is the same as the previous output value, and there are no new sampling times or new sampling numbers within the current unified processing cycle, a "hold-to-display" sampling flag is generated. If the sampling message verification passes, the sampling value field can be parsed, and the sampling time belongs to the current unified processing cycle but is not a "hold-to-display" sampling, a "valid" sampling flag is generated. For example, if the total organic carbon detection channel at the reverse osmosis outlet outputs 2.2 μg / L at 14:20:00, the message verification passes and the sampling number is updated, generating a valid sampling flag; if there is no new sampling number at 14:20:05 but the interface still displays 2.2 μg / L, a "hold-to-display" sampling flag is generated; if a message is received at 14:20:10 but verification fails, a communication anomaly sampling flag is generated. This step ensures that subsequent processing only uses sampling records with valid sampling flags, avoiding "hold-to-display" sampling and communication anomaly sampling from entering the anomaly judgment link.

[0036] In S13, a valid sampling chain is generated based on the process segment name and detection channel name in the retained aligned sampling records. A valid sampling chain refers to a sequence of valid sampling records arranged by sampling time within the same process segment and detection channel, after removing missing samples, communication anomaly samples, and samples kept for display. Its input data consists of sampling records and data quality markers. The generation method involves grouping and sorting by process segment name, detection channel name, and sampling time. The output is a similar sampling record chain that can be used to calculate the sorting position. The unit is determined by the sampling value unit of the corresponding detection channel. It is used in stages S1 to S2, and its technical function is to provide a continuous and similar data foundation for subsequent symbol segment generation.

[0037] Effective sampling chains are established separately for the same process segment and the same detection channel. For example, the conductivity detection channel at the outlet of the refining polishing bed forms one effective sampling chain, and the conductivity detection channel at the outlet of the electro-deionization module forms another effective sampling chain; the two chains will not be mixed. If the retained effective sampling values ​​are 0.0549 μS / cm, 0.0550 μS / cm, 0.0551 μS / cm, 0.0665 μS / cm, and 0.0550 μS / cm respectively, then these records are arranged according to sampling time to form an effective sampling chain. This chain retains the relationship between the previous and subsequent effective sampling records, providing a basis for judging the changes in adjacent sorting positions in S2.

[0038] In a supplementary implementation method, if multiple sampling records with valid sampling markers appear in the same process segment and the same detection channel within the same unified processing cycle, the sampling record with the sampling time closest to the center of the unified processing cycle can be selected as the aligned sampling record, and the remaining records can be used as reference records for the same cycle. If there are no valid sampling records within the same unified processing cycle, but there are displayed sampling records, only the display record corresponding to the displayed sampling will be generated, and no valid sampling chain record will be generated.

[0039] If multiple sampling records with valid sampling markers exist within the same process segment and detection channel in the same unified processing cycle, and these sampling records fall into different sorting position ranges after being inserted into the historical stability window, then the aligned sampling records of that unified processing cycle will not be used as normal output records that are allowed to participate in the benchmark update. Instead, a same-cycle fluctuation record will be generated for reference. In the next unified processing cycle, it will continue to determine whether to form an impact segment record, oscillation segment record, or drift segment record based on the new valid sampling records. The input data for this rule are multiple valid sampling records and the historical stability window within the same unified processing cycle. The output results are aligned sampling records, same-cycle reference records, or same-cycle fluctuation reference records. The usage phase is from S1 to S2. The technical function is to avoid the erroneous compression of multiple sampling differences within the same processing cycle into a single stable sample. For example, if two valid samples of 0.0550 μS / cm and 0.0665 μS / cm appear in the conductivity detection channel during the unified processing cycle from 10:00:15 to 10:00:20, and these two samples are located in the main sorting position range and the upper boundary of the sorting, respectively, then a record of the same period fluctuation is generated for reference during this cycle and is not written into the historical stability window.

[0040] like Figure 3 and Figure 4 As shown, in one specific embodiment, S2 includes:

[0041] S21. Based on the historical output records and benchmark participation markers of the corresponding process segment and corresponding detection channel, select the normal output records and final backfill output records that have been processed and are allowed to participate in benchmark updates, and generate a historical stable window; insert the current valid sampled values ​​in the valid sampling chain into the sampled value arrangement of the historical stable window to generate a sorting position record.

[0042] S22. Generate symbol records based on the change in the sorting position of two adjacent valid sampling records in the valid sampling chain, and compress valid sampling records corresponding to consecutive identical symbols into symbol segment records based on the symbol records.

[0043] S23. Generate change segment records within the channel based on symbol segment records.

[0044] In this embodiment, it should be noted that in S21, the historical stability window is not a simple truncation of the current effective sampling chain, but rather formed from the historical output records of the corresponding process segment and the corresponding detection channel. A normal output record refers to a record directly output when the current effective sampling record has not formed an impact segment record, oscillation segment record, drift segment record, local link anomaly processing mark, or pending confirmation record, and the corresponding data quality mark is a valid sampling mark. The input data for the normal output record consists of the current effective sampling record in the effective sampling chain, the corresponding sorting position record, and the symbol segment record. The generation method is to use the current effective sampling value as the output value and assign a reference participation mark that allows participation in reference updates when the current effective sampling record is in a closed stable segment or has not triggered an anomaly processing mark. The output results include the process segment name, detection channel name, sampling time, output value, output type, and reference participation mark; the unit is consistent with the sampling value unit of the corresponding target water quality detection channel.

[0045] For example, if the conductivity detection channel at the exit of the finishing polishing bed outputs 0.0549 μS / cm, 0.0550 μS / cm, and 0.0551 μS / cm at 10:00:00, 10:00:05, and 10:00:10, respectively, and no abnormality processing mark is generated, then these outputs are considered normal output records. Historical stability windows are maintained separately for the same process segment and the same detection channel. Their input data consists of normal output records with reference participation marks and final backfilled output records from the historical output records. The output result is a set of stable sampled values ​​arranged by sampling time and usable for ranking and comparison.

[0046] The window capacity of the historical stability window is not determined by a fixed manual alarm threshold, but rather by the number of effective samples taken by the corresponding detection channel in the most recent maintenance cycle. When the number of output records with reference participation markers in the most recent maintenance cycle is greater than the minimum accumulation number, the most recent records are selected as the historical stability window. When the number is insufficient, an initial accumulation marker is generated. The minimum accumulation number is determined by the minimum number of consecutive effective samples that can form a stable section for the corresponding detection channel in the most recent maintenance cycle, and different detection channels can be determined separately. For example, if the conductivity detection channel at the outlet of the polishing bed forms 360 uniform processing cycles in the most recent 30 minutes, and there are 330 normal output records and final backfill output records with reference participation markers, then the most recent 300 records can be selected to generate the historical stability window; if there are only 12 records, the initial accumulation state is maintained, and sorted position records and dynamic boundary records are not generated. After using sorted position records, short-term fluctuations of the same detection channel can be represented by relative position changes, without relying solely on a single fixed threshold.

[0047] In S22, a symbolic record is generated based on the change in the sorting position of two adjacent valid sampling records. The sorting position record refers to the relative arrangement position record formed after the current valid sample value is inserted into the historical stable window of the same process segment and the same detection channel. Its input data are the current valid sample value and the historical stable window. The generation method involves arranging the current valid sample value and the sample values ​​in the historical stable window in numerical order. The output includes the sorting position interval of the current valid sample value, the direction of movement relative to the previous valid sample value, and the corresponding sampling time. The unit is a position-type record, without changing the unit of the sample value. The usage stage is S2, and its technical function is to convert numerical changes into relative position changes.

[0048] The sorting position interval is formed by arranging the sampled values ​​within the historical stable window in ascending order. The sorted sampled values ​​are divided into a lower sorting boundary, a main sorting position interval, and an upper sorting boundary. The lower sorting boundary corresponds to the interval containing smaller sampled values ​​within the historical stable window, the upper sorting boundary corresponds to the interval containing larger sampled values ​​within the historical stable window, and the main sorting position interval corresponds to the interval of sampled values ​​within the historical stable window located between the lower and upper sorting boundaries. After the current valid sampled value is inserted into the sorting, its sorting position interval and the direction of its interval movement relative to the previous valid sampled value are recorded.

[0049] If the ranking position of the subsequent valid sample record moves towards the upper boundary of the historical stable window relative to the previous valid sample record, an ascending symbol is generated; if it moves towards the lower boundary, a descending symbol is generated; if they remain within the same ranking position range, a stable symbol is generated. A stable symbol is generated when the previous and subsequent valid sample values ​​are equal, or when they remain within the same ranking position range after being inserted into the historical stable window. If the two sample values ​​are not equal but still within the same ranking position range, neither an ascending nor descending symbol is generated; instead, a stable symbol is generated. For example, in the historical stable window of the conductivity detection channel at the outlet of a polishing bed, the main ranking position range covers 0.0549 μS / cm to 0.0551 μS / cm. If adjacent valid sample values ​​change from 0.0550 μS / cm to 0.0551 μS / cm, but both remain within the main ranking position range, a stable symbol is generated.

[0050] In S23, a change segment record within the channel is generated based on the symbol segment record. The symbol segment record refers to the segment record formed by compressing the valid sample records corresponding to consecutive identical symbols. Its input data consists of rising, falling, and stable symbols formed by changes in adjacent sorting positions. The generation method is to merge consecutive identical symbols according to the sampling time. The output results include the segment start point, segment end point, segment symbol, segment length, previous segment identifier, and next segment identifier, with the unit being a segment type record. It is used in stages S2 to S3, and its technical function is to preserve the direction of change, duration, and segment order.

[0051] The channel change segment record refers to the impact segment record, oscillation segment record, or drift segment record identified based on the symbol segment record. Its input data consists of closed symbol segment records. The generation method determines the change type based on the arrangement order of stable, rising, and falling segments. The output includes the change segment type, change segment start point, change segment end point, change direction, and corresponding sorting position interval. The unit is the segment type record. It is used in stages S2 to S3, and its technical function is to provide a basis for change type identification for process segment feedback verification and processing mark generation.

[0052] If the preceding and following segments of a given rising or falling segment are both stable, and the rising or falling segment does not continue to extend towards the same sorting boundary, an impact segment record is generated. If at least two rising segments and at least two falling segments alternate in chronological order, and the sorting position intervals corresponding to each segment move back and forth between the upper and lower sorting boundaries, an oscillation segment record is generated. If two or more consecutive segments move towards the same sorting boundary, or move towards the same sorting boundary and then form a stable segment near that boundary, a drift segment record is generated. For example, if the symbol segment sequence of a conductivity detection channel is stable segment, rising segment, stable segment, an impact segment record is generated. If the symbol segment sequence is rising segment, falling segment, rising segment, falling segment, and the sorting position moves back and forth between the upper and lower sorting boundaries, an oscillation segment record is generated. If the symbol segment sequence is rising segment, rising segment, rising and then stable segment, a drift segment record is generated.

[0053] In a supplementary implementation, if no subsequent segment has appeared for the current symbol segment, the current symbol segment is marked as an unclosed symbol segment. Unclosed symbol segments can be used for real-time display, but must not participate in historical stability window updates, nor can they be used as stability recovery records. If the historical stability window is empty or there are no sortable historical sample values ​​in the historical stability window, an initial accumulation marker is generated, and the generation of symbol segment records and dynamic boundary records is paused.

[0054] When the same change segment has a unidirectional section and it is not yet possible to determine whether it belongs to an impact segment record or a drift segment record, if a subsequent stable segment forms after the unidirectional section and the subsequent stable segment returns to the sorting position interval of the previous stable segment, then an impact segment record is generated; if a subsequent stable segment does not form after the unidirectional section and returns to the sorting position interval of the previous stable segment, and the subsequent sorting position continues to be on the same sorting boundary side, then a drift segment record is generated. This supplementary method can prevent unclosed change segments from participating in dynamic boundary updates too early, and also prevent unidirectional changes that have not yet recovered stability from being prematurely treated as impacts.

[0055] like Figure 5 and Figure 6 As shown, in one specific embodiment, S3 includes:

[0056] S31. Based on the channel change segment record and process segment sequence record of the target process segment, determine the actual upstream or downstream adjacent process segment of the target process segment; based on the detection channel name of the target process segment, retrieve the symbol segment record of the same type of detection channel from the upstream or downstream adjacent process segments, and generate candidate records of adjacent process segments.

[0057] S32. Generate process segment receipt records based on candidate records of adjacent process segments.

[0058] S33. Generate processing markers based on symbol segment records and process segment receipt records.

[0059] In this embodiment, it should be noted that in S31, changes in the target process segment are not judged in isolation, but are combined with the process segment sequence record for adjacent process segment retrieval. The reverse osmosis outlet, the electro-deionization module outlet, and the refining and polishing bed outlet constitute a fixed pipeline sequence. If the target process segment is the refining and polishing bed outlet, then the upstream adjacent process segment is the electro-deionization module outlet; if the target process segment is the electro-deionization module outlet, then the upstream adjacent process segment is the reverse osmosis outlet, and the downstream adjacent process segment is the refining and polishing bed outlet; if the target process segment is the reverse osmosis outlet, then the downstream adjacent process segment is the electro-deionization module outlet.

[0060] For a target process segment located at the end of the process segment sequence record, only actually existing adjacent process segments are retrieved. When the target process segment is a reverse osmosis outlet, only the downstream adjacent electro-deionization module outlet is retrieved; when the target process segment is a refining polishing bed outlet, only the upstream adjacent electro-deionization module outlet is retrieved; when the target process segment is an electro-deionization module outlet, both the upstream adjacent reverse osmosis outlet and the downstream adjacent refining polishing bed outlet are retrieved. No corresponding acknowledgment records are generated for non-existent upstream or downstream adjacent process segments, nor are they treated as communication anomalies.

[0061] For example, when a drift segment record is formed in the conductivity detection channel at the outlet of the refining polishing bed, only the symbol segment record of the conductivity detection channel at the outlet of the electro-deionization module is retrieved, and downstream process segments that do not exist are not searched. If there is no drift segment in the same direction in the conductivity detection channel at the outlet of the electro-deionization module, an isolated record is generated. Only similar detection channels are searched during the retrieval process. For example, when a drift segment record is formed in the conductivity detection channel at the outlet of the refining polishing bed, only the symbol segment record of the conductivity detection channel at the outlet of the electro-deionization module is retrieved, and total organic carbon detection channels or dissolved oxygen detection channels are not searched. This is because the dimensions and physical meanings of different target water quality detection channels are different, and direct numerical comparisons are not possible; however, the direction of change and the order of the starting points of similar detection channels between adjacent process segments can be used to determine whether a pipeline propagation relationship exists. This step transforms the ultrapure water pipeline sequence into a data retrieval range.

[0062] In S32, process segment receipt records are generated based on candidate records from adjacent process segments. A process segment receipt record is a record generated after verifying the consistency of time sequence and direction between the change segment record within the target process segment and the symbol segment record of the same type of detection channel in adjacent process segments. Its input data includes the change segment record within the target process segment, the process segment sequence record, the symbol segment record of the same type of detection channel in adjacent process segments, and historical propagation waiting records. The generation method compares whether segments in the same direction occur sequentially according to the process segment sequence and fall into the historical propagation waiting records. Output results include upstream receipt records, downstream receipt records, or isolated records, with the unit being a status type record. The usage stage is S3, and its technical function is to distinguish between process propagation changes and local abnormal changes.

[0063] A co-directional drift segment refers to a situation where the same type of detection channel in both the target process segment and the adjacent process segment forms a drift segment record, and the segment symbols of the two drift segment records are in the same direction. When the drift segment of the target process segment is in an upward direction and the drift segment of the adjacent process segment is also in an upward direction, they are considered co-directional drift segments. Similarly, when the drift segment of the target process segment is in a downward direction and the drift segment of the adjacent process segment is also in a downward direction, they are considered co-directional drift segments. Determining a co-directional drift segment does not require that the sampling value amplitudes of the two process segments be the same, nor does it require that the sampling value ranges of different process segments be the same. For example, if the total organic carbon detection channel at the outlet of the electro-deionization module shifts from 2.2 μg / L to 3.0 μg / L, and the total organic carbon detection channel at the outlet of the refining and polishing bed shifts from 1.6 μg / L to 2.3 μg / L, although the numerical ranges of the two process segments are different, the segment symbols of both are in an upward direction, therefore they belong to the co-directional drift segment.

[0064] Historical propagation wait records are generated separately for adjacent process sections and similar detection channels. The input data includes records of closed drift segments during historical stable operation, corresponding upstream or downstream feedback records, and the starting point of each drift segment. The data processing unit extracts the time difference of the starting point of similar detection channels with the same drift direction between adjacent process sections from the historical records. After removing time difference records with communication anomaly sampling markers, maintained display sampling markers, or maintenance-limited output, the remaining time difference records are used as historical propagation wait records. The unit of measurement for historical propagation wait records is time, and they are used during the process section feedback verification stage. Their technical function is to determine whether changes in the target process section conform to the propagation wait relationships that have previously occurred between adjacent process sections.

[0065] For example, if the starting point of the rising drift segment of the conductivity detection channel at the outlet of the electro-deionization module is 20s, 25s, 30s, and 35s earlier than that of the conductivity detection channel at the outlet of the refining polishing bed, and these records all come from historical segments without communication anomalies or maintenance restrictions, then a historical propagation waiting record for that adjacent process segment and that detection channel is formed; when the new time difference is 25s or close to the historical record range, an upstream receipt record is allowed to be generated.

[0066] In S33, processing tags are generated based on symbol segment records and process segment feedback records. A processing tag is a processing type identifier generated based on change segment records and process segment feedback records within the channel. Its input data includes impact segment records, oscillation segment records, drift segment records, upstream feedback records, downstream feedback records, and isolated records. The generation method is based on logical combination judgments of text. Output results include sensor disturbance processing tags, channel oscillation processing tags, process fluctuation processing tags, and local link anomaly processing tags, with units of tag type records. It is used in stages S3 to S4, and its technical function is to determine the subsequent filter output path and reference participation rules.

[0067] When impact segment records and isolated records exist simultaneously, a sensor disturbance processing mark is generated; when oscillation segment records and isolated records exist simultaneously, a channel oscillation processing mark is generated; when drift segment records of the same target process segment, the same detection channel, and the same drift direction exist simultaneously with the corresponding upstream or downstream feedback records, a process fluctuation processing mark is generated; when drift segment records and isolated records exist simultaneously, a local link anomaly processing mark is generated.

[0068] When the conditions for generating multiple processing markers are met simultaneously within the same target process segment, the same detection channel, and the same processing cycle, the system first determines whether there is an oscillating structure with alternating rising and falling segments. If so, a channel oscillation processing marker is generated first. If no oscillating structure exists, the system then determines whether there is an impact structure consisting of a previous stable segment, a unidirectional segment, and a subsequent stable segment. If so, a sensor disturbance processing marker is generated. If no impact structure exists, the system then determines whether there is a drift structure continuously moving towards the same sorting boundary side, and combines this with the process segment feedback record to generate a process fluctuation processing marker or a local link anomaly processing marker. The input data for this rule are the channel change segment record and the process segment feedback record. The output result is a unique processing marker or a record to be confirmed. The usage stage is S3, and the technical function is to avoid assigning multiple conflicting processing markers to the same change segment simultaneously. For example, if a conductivity detection channel first shows alternating rising and falling segments and then short-term stability, it is processed first according to the channel oscillation processing marker, not the sensor disturbance processing marker.

[0069] In a supplementary implementation method, if historical propagation waiting records have not yet been formed, upstream or downstream acknowledgment records are not forcibly generated. Instead, the corresponding changes are retained as pending review, and a pending confirmation record or process fluctuation limit output is generated in S4. If there are co-directional segments in the candidate records of adjacent process segments but the co-directional segments have not yet closed, formal acknowledgment records are not generated temporarily. Instead, the acknowledgment relationship is determined after the segments close.

[0070] The end-process segment receipt generation rule is used to handle situations where the target process segment has no upstream or downstream adjacent process segments. When the target process segment is a reverse osmosis outlet, no upstream receipt record is generated; the generation of a downstream receipt record is determined solely by the existence of a subsequent co-directional segment in the downstream adjacent process segment. When the target process segment is a polishing bed outlet, no downstream receipt record is generated; the generation of an upstream receipt record is determined solely by the existence of a pre-occurring co-directional segment in the upstream adjacent process segment. When the target process segment is an electro-deionization module outlet, both upstream and downstream receipt records are allowed to be generated. The input data for this rule includes the target process segment name, process segment sequence record, and symbolic segment record of the same type of detection channel in adjacent process segments. The output results are upstream receipt record, downstream receipt record, isolated record, or pending confirmation record. The usage stage is S3, and its technical purpose is to prevent the end-process segment from being misjudged as having communication anomalies or missing processing due to the absence of a process segment in a certain direction. For example, when the total organic carbon detection channel at the reverse osmosis outlet forms a drift segment, instead of searching for non-existent upstream process segments, it searches for the same-direction drift segment that occurs after the total organic carbon detection channel at the outlet of the electro-deionization module exists.

[0071] like Figure 7 and Figure 8As shown, in one specific embodiment, S4 includes:

[0072] S41. Generate a temporary correction output based on the sensor disturbance processing mark and the previous stable section before the impact segment. After the next stable section after the impact segment is formed, generate the final backfill output based on the previous stable section and the next stable section. If the previous stable section before the impact segment does not exist, generate a record to be confirmed.

[0073] S42. Generate channel oscillation limit output based on channel oscillation processing markers; generate process fluctuation limit output based on process fluctuation processing markers; generate maintenance limit output based on local link anomaly processing markers.

[0074] S43. Based on the filtered output data and the benchmark, update the historical stable window and dynamic boundary records.

[0075] In this embodiment, it should be noted that in S41, for the impact segment corresponding to the sensor disturbance processing mark, it is first determined whether the previous stable segment exists. If it exists, a temporary correction output is generated based on the previous stable segment. For example, the output of the conductivity detection channel at the outlet of the polishing bed in the previous stable segment of the impact segment is 0.0549 μS / cm, 0.0550 μS / cm, and 0.0551 μS / cm, while the original sampling value of the impact segment is 0.0665 μS / cm. Instead of using 0.0665 μS / cm as an output that can participate in the benchmark update, a temporary correction output is generated first for real-time display.

[0076] The temporary correction output is generated from the previous stable segment preceding the impact segment. If multiple normal output records exist in the previous stable segment, the normal output record whose time is closest to the start of the impact segment is selected as the temporary correction output; if this record does not exist, a record to be confirmed is generated. The final backfill output is generated jointly by the previous and subsequent stable segments. If, after the subsequent stable segment is formed, its stable output is in the same sorting position interval as the final output of the previous stable segment, the first normal output record of the subsequent stable segment is selected as the final backfill output; if they are not in the same sorting position interval, the record to be confirmed is retained, and the original sampled value of the impact segment is not written into the historical stable window.

[0077] For example, if the output at the end of the previous stable segment before the impact segment is 0.0551 μS / cm, the original sampled value of the impact segment is 0.0665 μS / cm, and the first normal output of the next stable segment after the impact segment is 0.0550 μS / cm, and 0.0551 μS / cm and 0.0550 μS / cm are within the main sorting position range, then the temporary correction output can initially use 0.0551 μS / cm, and the final backfill output uses 0.0550 μS / cm. This step generates a temporary correction output when a previous stable segment exists, and generates a record to be confirmed when a stable source is lacking.

[0078] In S42, different limiting outputs are generated for different treatment markers. The filtered output data is not a single field containing only water quality values, but a composite output record containing output value, output type, reference participation marker, and additional status information. The input data for the filtered output data includes treatment markers, valid sampling records, symbolic segment records, process segment feedback records, and historical stability windows. The generation method is to select, according to the treatment marker, normal output record, temporary correction output, final backfill output, channel oscillation limiting output, process fluctuation limiting output, maintenance limiting output, or pending confirmation record. The output result includes at least the process segment name, detection channel name, sampling time, output value, output type, segment boundary, and reference participation marker. The unit is consistent with the sampling value unit of the corresponding target water quality detection channel. If the output type is a pending confirmation record or maintenance prompt record, the output value can retain the original sampling value and include an additional status description.

[0079] The channel oscillation limitation output, process fluctuation limitation output, and maintenance limitation output all retain the original sampled values, segment start point, segment end point, segment direction, and processing markers, but their reference participation markers differ. The channel oscillation limitation output is used to display to the plant monitoring interface that the channel is in a round-trip change state and prohibits entry into the historical stability window; the process fluctuation limitation output is used to retain the real changes propagating sequentially along the process segment and determines whether to allow entry into the long-term reference after the stability recovery record is formed; the maintenance limitation output is used to retain the original sampled values ​​and maintenance prompt records corresponding to local link anomalies and prohibits entry into the historical stability window.

[0080] For example, if the dissolved oxygen detection channel records an oscillation segment, the output value can still display the original sampled value of 7.2 μg / L or 5.1 μg / L for the corresponding period, but its reference participation flag will be disabled. If the total organic carbon detection channel records a process fluctuation processing flag, it will first display a process fluctuation limit output of 3.0 μg / L or 2.3 μg / L, and then determine whether to participate in the long-term reference after stabilization. This step ensures that different anomaly types enter different output paths, avoiding the use of the same output processing for different change types.

[0081] In S43, the historical stability window and dynamic boundary records are updated based on the filtered output data and the reference participation flag. The reference participation flag is a flag used to control whether filtered output data is allowed to enter the historical stability window. Its input data are filtered output data and processing flags. The generation method is to mark normal output records and final backfill outputs as allowed to participate in the reference update, mark temporary correction outputs, channel oscillation limit outputs, maintenance limit outputs, pending confirmation records, and display records corresponding to the hold display sampling as prohibited from participating in the reference update, and mark process fluctuation limit outputs that meet the stability recovery conditions as candidate participants in the reference update. The output results are three types of flags: allowed, prohibited, or candidate. The usage stage is S4, and its technical function is to block the impact of abnormal outputs on the historical stability window and dynamic boundary records.

[0082] Normal output records and final backfill outputs are allowed to enter the historical stability window; temporary correction outputs, channel oscillation limit outputs, maintenance limit outputs, pending confirmation records, and display records corresponding to the maintained display sampling are prohibited from entering the historical stability window; process fluctuation limit outputs are only allowed to participate in long-term benchmark updates if their corresponding upstream or downstream receipt records are continuously valid from the start of the drift segment of the target process segment to the closure of the next stable segment of the target process segment.

[0083] A stable recovery record refers to a closed stable segment record formed after a drift segment in the target process section and the same detection channel. Its input data is the symbolic segment record after the drift segment, and its output includes the stable recovery start point, stable recovery end point, the sorted position interval after recovery, and the corresponding processing marker. If the valid sampling records after the drift segment are continuously within the same sorted position interval and no new rising, falling, or oscillating segments are formed, a stable recovery record is generated. If isolated oscillating segments, communication anomalies, or maintenance-limited outputs still occur after the drift segment, a stable recovery record is not generated. For example, after the total organic carbon detection channel at the outlet of the refining polishing bed moves from 1.6 μg / L to 2.3 μg / L, and subsequent valid sampling cycles all form a stable segment around 2.2 μg / L to 2.3 μg / L, and the upstream receipt record of the total organic carbon detection channel at the outlet of the electro-deionization module remains valid during this period, a stable recovery record is generated, and the corresponding process fluctuation limit output is converted into a candidate output.

[0084] In a supplementary implementation, candidate outputs are not immediately written into the historical stability window upon formation, but are instead reviewed in the next dynamic boundary update cycle. The candidate output review cycle is determined by the number of effective samples required for the corresponding detection channel to form a stable recovery record, without using manually fixed alarm thresholds. If the target water quality detection channel continuously forms closed stable sections after the candidate output, and all effective sampling records within this stable section have valid sampling markers, and no maintenance-limited outputs, channel oscillation-limited outputs, or pending confirmation records are formed, then the review passes. If communication anomalies, display records that maintain dominant display sampling, or new isolated change segments appear during the review cycle, then the review fails.

[0085] The input data for this rule includes candidate outputs, subsequent valid sampling chains, stable recovery records, and processing flags. The output result is either an output record that allows participation in benchmark updates or a flag that cancels benchmark participation. The usage stage is S4, and its technical function is to prevent unstable process fluctuations from prematurely altering the historical stability window. For example, if the total organic carbon detection channel stabilizes at 2.2 μg / L to 2.3 μg / L for several consecutive valid sampling cycles after 14:23:00, and no new abnormal processing flag is generated, then the candidate output verification passes. If the final backfill output and the temporary correction output cause a change in the plant monitoring display status, a backfill record is generated; if the final backfill output and the temporary correction output do not cause a change in the display status, only the internal record is updated. If, after a process fluctuation limit output becomes a candidate output, its subsequent stable recovery record is interrupted by a new isolated oscillation segment, then the benchmark participation flag for that candidate output is canceled.

[0086] In another supplementary implementation method, dynamic boundary records are generated separately for the same process segment and the same detection channel. The input data consists of the sampled value arrangement within the historical stable window and the normal output records and final backfill output records within the closed stable section. Dynamic boundary records include at least a lower sorting boundary record, a main sorting position interval record, and an upper sorting boundary record, used to subsequently determine which sorting boundary the current valid sampled value should move towards. When updating dynamic boundary records, output records without reference marking are first excluded. Then, the remaining sampled values ​​are arranged in numerical order. The stable sampled value interval closest to the smaller value is taken as the lower sorting boundary record, the stable sampled value interval closest to the larger value is taken as the upper sorting boundary record, and the stable sampled value interval between the two is taken as the main sorting position interval record.

[0087] For example, if the historical stable window of the conductivity detection channel at the outlet of the polishing bed is mainly distributed between 0.0548 μS / cm and 0.0552 μS / cm, then the sampling value range near 0.0548 μS / cm forms the lower boundary record of the sorting, the sampling value range near 0.0552 μS / cm forms the upper boundary record of the sorting, and the intermediate sampling values ​​form the main sorting position range record; if there is only maintenance restriction output in the current period, then the data of this period will not be used to update the dynamic boundary record.

[0088] When generating sorting position records and symbol segment records within the current processing cycle, the historical stable window and dynamic boundary records already formed before the start of the current processing cycle are used. At the end of the current processing cycle, the historical stable window and dynamic boundary records are updated based on the filtered output data and the reference participation marker. The updated historical stable window and dynamic boundary records are used for the next processing cycle. Historical stable windows that have not yet been updated are not used to reverse-engineer the sorting position records already generated in the current cycle within the same processing cycle. For example, if the conductivity sample value at 10:00:15 is 0.0665 μS / cm, its sorting position record is generated based on the historical stable window formed before 10:00:15. When this sample value is determined to be a sensor disturbance and is prohibited from participating in the reference update, the update result only affects the historical stable window and dynamic boundary records of processing cycles from 10:00:20 onwards.

[0089] In another supplementary implementation method, before the maintenance reminder record corresponding to the maintenance restriction output is closed, the sampling records generated by the corresponding process section and the corresponding detection channel can still be used for real-time display and maintenance traceability, but they must not enter the historical stability window, nor can they be used to generate historical propagation waiting records. After the maintenance reminder record is closed, the initial accumulation or stability recovery judgment process is re-entered, and the historical stability window update is resumed after a new closed stable section is formed.

[0090] The input data for this rule includes maintenance restriction output, maintenance reminder record closing status, and subsequent valid sampling records. The output results are a prohibited reference participation marker, an initial accumulation marker, or a stable recovery record. The usage phase is S4. Its technical function is to prevent local link anomalies during maintenance from affecting the historical stability window and historical propagation wait records. For example, if the conductivity detection channel at the refined polishing bed outlet generates a maintenance restriction output at 16:00:00, the conductivity sampling values ​​from 16:00:00 until the maintenance reminder record closes will not enter the historical stability window, nor will they be used as samples of the propagation wait time from the electro-deionization module outlet to the refined polishing bed outlet.

[0091] This invention also provides a system for processing abnormal ultrapure water quality data based on adaptive filtering, such as... Figure 9As shown, the system includes water quality sensor nodes, an industrial bus, a data processing unit, a communication transmission module, and a plant monitoring system. The data processing unit includes an effective sampling chain generation module, a symbol segment generation module, a process segment receipt verification module, and an output boundary control module. The effective sampling chain generation module generates an effective sampling chain based on sampling records and data quality markers; the symbol segment generation module generates sorting position records based on the effective sampling chain and historical stability window, and generates symbol segment records based on changes in adjacent sorting positions; the process segment receipt verification module generates process segment receipt records based on symbol segment records and process segment sequence records, and generates processing markers based on symbol segment records and process segment receipt records; the output boundary control module generates filtered output data based on processing markers, and updates the historical stability window and dynamic boundary records according to the baseline participation markers; the communication transmission module sends the filtered output data to the plant monitoring system.

[0092] In one specific implementation method, water quality sensor nodes can be arranged at the reverse osmosis outlet, the electro-deionization module outlet, and the polishing bed outlet, respectively. Each water quality sensor node corresponds to at least one target water quality detection channel. The industrial bus sends sampling records to the data processing unit, and the plant monitoring system displays filtered output data, maintenance prompt records, pending confirmation records, and backfill results. The hold-up display samples generated by the plant monitoring system are only saved as display records and traceability records, and are not sent back to the data processing unit as new sampling records of the water quality sensor nodes to participate in the generation of effective sampling chains, historical stability windows, sorting position records, symbol segment records, and dynamic boundary records. The hold-up display samples do not correspond to the new sampling time or new sampling sequence number of the current sampling period; their values ​​come from the previous output value, and therefore cannot represent the actual water quality changes in the current sampling period. This restriction can prevent the display refresh value from being repeatedly counted in the stable sampling quantity, prevent the historical stability window from being compressed by repeated values, and prevent the output results of the data processing unit from back-contaminating subsequent sampling and processing links.

[0093] Within a single processing cycle, the water quality sensor node generates a sampling record containing the process section name, detection channel name, sampling time, sampled value, and equipment status word. The industrial bus transmits the sampling record to the data processing unit, which generates a valid sampling chain, symbol segment record, process section receipt record, processing mark, and filtered output data. The communication transmission module sends the filtered output data, maintenance prompt record, pending confirmation record, and backfill record to the plant monitoring system. The plant monitoring system only displays the above output results and does not send the refreshed and retained sampled data back to the data processing unit as a valid sampling record.

[0094] The data sent by the communication transmission module includes at least the process section name, detection channel name, sampling time, output value, output type, reference participation flag, maintenance prompt record identifier, and backfill record identifier. The plant monitoring system displays normal output, temporary correction output, final backfill output, channel oscillation limit output, process fluctuation limit output, maintenance limit output, or pending confirmation record based on the output type. However, the refresh result displayed by the plant monitoring system is not considered a new sampling record for the water quality sensor node. This data structure can be implemented through industrial communication messages, database table records, or plant monitoring system interface fields, without requiring changes to the hardware structure of the water quality sensor node.

[0095] The following example, using a scenario with specific parameters and data, illustrates the processing method and system for handling abnormal ultrapure water quality data based on adaptive filtering.

[0096] In the ultrapure water pipeline of a semiconductor plant, target water quality monitoring channels are located at the reverse osmosis outlet, the electro-deionization module outlet, and the refining and polishing bed outlet. The normal output of the conductivity monitoring channel at the refining and polishing bed outlet is concentrated around 0.0548 μS / cm to 0.0552 μS / cm; the normal output of the conductivity monitoring channel at the electro-deionization module outlet is concentrated around 0.0580 μS / cm to 0.0584 μS / cm; the common stable output of the total organic carbon monitoring channel at the electro-deionization module outlet is 2.0 μg / L to 2.4 μg / L; the common stable output of the total organic carbon monitoring channel at the refining and polishing bed outlet is 1.4 μg / L to 1.8 μg / L; the historical stable window of the dissolved oxygen monitoring channel at the electro-deionization module outlet under low-oxygen control conditions is mainly distributed between 5.8 μg / L and 6.6 μg / L. The above data are used to illustrate the processing logic and do not involve directly adding or weighting the values ​​of different monitoring channels.

[0097] The unified processing cycle is 5 seconds, with the conductivity detection channel generating a valid sample every 5 seconds, the total organic carbon detection channel generating a valid sample every 30 seconds, and the plant monitoring interface refreshing every 5 seconds. If the total organic carbon detection channel only repeats the previous output value between two valid samples, the repeated display value is marked as a retained display sample and does not enter the valid sampling chain.

[0098] In S1, the conductivity detection channel at the outlet of the refining and polishing bed outputs values ​​of 0.0549 μS / cm, 0.0550 μS / cm, 0.0551 μS / cm, 0.0665 μS / cm, 0.0550 μS / cm, and 0.0551 μS / cm at 10:00:00, 10:00:05, 10:00:10, 10:00:15, 10:00:20, and 10:00:25 respectively, and all of these are valid sampling markers. Therefore, these records form a valid sampling chain. Meanwhile, the total organic carbon detection channel does not generate any new valid samples at 10:00:05 and 10:00:10, only maintaining a display of 1.8 μg / L. Therefore, these maintained display samples do not enter the valid sampling chain.

[0099] If the 0.0665 μS / cm at 10:00:15 also has a communication anomaly sampling marker, then the record is discarded in S1 and will no longer participate in the subsequent impact segment identification; in this scenario, the record has a valid sampling marker, so it enters the subsequent sorting position record generation.

[0100] In S2, the historical stable window of the conductivity detection channel at the outlet of the refining polishing bed includes 0.0548 μS / cm, 0.0549 μS / cm, 0.0550 μS / cm, 0.0551 μS / cm, and 0.0552 μS / cm. Among them, the sampling value interval close to 0.0548 μS / cm is the lower boundary of the sorting, the sampling value interval close to 0.0552 μS / cm is the upper boundary of the sorting, and the sampling value interval close to 0.0550 μS / cm is the main sorting position interval.

[0101] When a sample value of 0.0665 μS / cm is inserted into the historical stable window, its sorting position is located at the upper boundary of the sorting; when the next valid sample value of 0.0550 μS / cm is inserted into the historical stable window, its sorting position returns to the main sorting position range. Therefore, the processing forms a previous stable segment, a single-point rising segment, and a subsequent stable segment, further generating an impact segment record. The entire judgment does not use detection channel values ​​other than conductivity, and there is no dimensional mixing.

[0102] In S3, the processing procedure retrieves the symbol segment records of the conductivity detection channel at the outlet of the electro-deionization module within the corresponding time range. If the conductivity detection channel at the outlet of the electro-deionization module remains stable between 0.0580 μS / cm and 0.0584 μS / cm between 10:00:00 and 10:00:30, and no unidirectional impact segment or unidirectional drift segment is formed in the sorting position, then the impact segment of the conductivity detection channel at the outlet of the finishing polishing bed corresponds to an isolated record. Based on this, the processing procedure generates a sensor disturbance processing mark, instead of a process fluctuation processing mark.

[0103] In S4, since the preceding stable segment of the impact segment has values ​​of 0.0549 μS / cm, 0.0550 μS / cm, and 0.0551 μS / cm, and the subsequent stable segment has values ​​of 0.0550 μS / cm and 0.0551 μS / cm, the value of 0.0551 μS / cm, which is closest to the start of the impact segment in the preceding stable segment, is first used as a temporary correction output. Then, after the subsequent stable segment is formed, 0.0550 μS / cm is used as the final backfill output. 0.0665 μS / cm is not included in the historical stabilization window, nor is the dynamic boundary record updated. The final backfill output is included in the historical stabilization window, ensuring that subsequent dynamic boundaries still form around a stable value near 0.0550 μS / cm.

[0104] During another time period, the total organic carbon (TOC) detection channel at the electro-deionization module outlet shifted from 2.2 μg / L to 3.0 μg / L between 14:20:00 and 14:22:00, while the TOC detection channel at the refining polishing bed outlet shifted from 1.6 μg / L to 2.3 μg / L between 14:20:30 and 14:22:30. Historical propagation waiting records show that the TOC detection channel shifted from the electro-deionization module outlet to the refining polishing bed outlet at 25s, 30s, and 35s. Therefore, the 30-second sequence in this instance can generate a process segment feedback record.

[0105] The process generates a process fluctuation handling flag, but does not directly delete the drift or immediately change the dynamic boundary record. Instead, it generates a process fluctuation limit output. Subsequently, if the total organic carbon detection channel at the outlet of the electro-deionization module stabilizes at 2.9 μg / L to 3.0 μg / L from 14:23:00 to 14:25:00, and the total organic carbon detection channel at the outlet of the refining and polishing bed stabilizes at 2.2 μg / L to 2.3 μg / L from 14:23:30 to 14:25:30, and the corresponding upstream receipt record remains valid during this period, a stable recovery record is generated, and the corresponding process fluctuation limit output is converted into a candidate output. If an isolated oscillation segment or maintenance limit output subsequently occurs, it is not converted into a candidate output.

[0106] During another time period, the historical stable window of the dissolved oxygen detection channel at the outlet of the electro-deionization module was mainly distributed between 5.8 μg / L and 6.6 μg / L. The sampling value range near 5.8 μg / L was the lower boundary of the sorting, and the sampling value range near 6.6 μg / L was the upper boundary of the sorting. When the current valid sampling values ​​were 7.2 μg / L, 5.1 μg / L, 7.4 μg / L, 5.0 μg / L, and 7.3 μg / L respectively, the sorting position moved back and forth between the upper and lower boundaries of the sorting, thus generating an oscillating segment record.

[0107] Within the same timeframe, the effective sampling values ​​of the dissolved oxygen detection channel at the reverse osmosis outlet remained between 6.1 μg / L and 6.4 μg / L, while those at the polishing bed outlet remained between 6.0 μg / L and 6.5 μg / L. Neither channel exhibited alternating rising and falling oscillation segments. Therefore, the oscillation segment record of the dissolved oxygen detection channel at the electrodeionization module outlet lacked similar oscillation feedback in adjacent process segments, resulting in an isolated record and further generating a channel oscillation processing marker. This channel oscillation limitation output can be displayed but is not included in the historical stability window, preventing the oscillation data from expanding the dynamic boundary record.

[0108] In another specific scenario, the conductivity detection channel at the outlet of the refining polishing bed continuously shifted from 0.0550 μS / cm to 0.0564 μS / cm between 16:00:00 and 16:05:00, and continued to move towards the upper boundary of the sorting position record, forming a drift segment record. Within the same time range and the corresponding historical propagation waiting record range, the conductivity detection channel at the outlet of the electro-deionization module remained between 0.0580 μS / cm and 0.0584 μS / cm, without forming a unidirectional drift segment record. The processing generates isolated records, and based on the drift segment records and isolated records, generates local link anomaly handling markers, subsequently generating maintenance restriction outputs and maintenance prompt records.

[0109] The maintenance reminder record includes the exit of the fine polishing bed, the conductivity detection channel, the drift start point 16:00:00, the drift direction as upward, and the corresponding section boundary. Before the maintenance reminder record is closed, the sampling records generated by this detection channel can be used for display and traceability, but they are not included in the historical stability window, nor are they used as samples awaiting historical propagation.

[0110] The preferred embodiments of this disclosure have been described in detail above with reference to the accompanying drawings. However, this disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this disclosure, various simple modifications can be made to the technical solutions of this disclosure, and these simple modifications all fall within the protection scope of this disclosure.

[0111] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, this disclosure will not describe the various possible combinations separately.

[0112] Furthermore, various different embodiments of this disclosure can be combined in any way, as long as they do not violate the spirit of this disclosure, they should also be regarded as the content disclosed in this disclosure.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for processing abnormal ultrapure water quality data based on adaptive filtering, characterized in that, The methods include: Acquire sampling records for the target water quality testing channel, remove missing samples, communication anomaly samples, and samples that are kept on display based on data quality markers, and generate valid sampling chains according to process section name and testing channel name; Based on the effective sampling chain and the historical stable window of the same process section and the same detection channel, a sorting position record is generated, and a symbol segment record is generated based on the changes of adjacent sorting positions; Based on the symbol segment records and the process segment sequence records of the reverse osmosis outlet, electro-deionization module outlet, and polishing bed outlet, retrieve the symbol segment records of the same detection channel in the adjacent process segments that actually exist, and generate process segment receipt records. A processing flag is generated based on the symbol segment record and the process segment receipt record. Filtered output data is generated based on the processing flag, and the historical stable window and dynamic boundary record are updated according to the benchmark participation flag. The target water quality detection channel includes at least one of the following: conductivity detection channel, total organic carbon detection channel, and dissolved oxygen detection channel.

2. The method for processing ultrapure water quality anomaly data based on adaptive filtering according to claim 1, characterized in that, Based on data quality markers, missing samples, communication anomaly samples, and samples that are retained for display are removed. A valid sampling chain is generated according to the process section name and detection channel name, including: An aligned sampling record is generated based on the sampling time and uniform processing cycle in the sampling record; Based on the data quality markers, retain aligned sampling records with valid sampling markers and discard aligned sampling records with missing sampling markers, communication anomaly sampling markers, or retained display sampling markers; Based on the process segment name and detection channel name in the retained alignment sampling records, generate valid sampling chains for the corresponding process segment and detection channel respectively; The held display sample is a record formed by using the previous output value when no new valid sample value is obtained in the current refresh cycle. The held display sample is only used for display and does not enter the valid sample chain.

3. The method for processing ultrapure water quality anomaly data based on adaptive filtering according to claim 1, characterized in that, Based on the effective sampling chain and the historical stable window of the same process section and the same detection channel, a sorting position record is generated, including: Based on the historical output records of the corresponding process segment and the corresponding detection channel and the benchmark participation mark, select the normal output records and the final backfill output records that have been processed and are allowed to participate in the benchmark update, and generate the historical stability window. Insert the current valid sample value in the valid sampling chain into the sample value arrangement of the historical stable window to generate the sorting position record; When there are no sortable historical sample values ​​in the historical stable window, an initial accumulation marker is generated, and the generation of the symbol segment record and the dynamic boundary record is paused. The historical stability windows are established separately for the same process segment and the same detection channel, and are not shared between different target water quality detection channels.

4. The method for processing ultrapure water quality anomaly data based on adaptive filtering according to claim 1, characterized in that, Symbolic segment records are generated based on changes in adjacent sorting positions, including: Symbol records are generated based on the change in the sorting position of two adjacent valid sampling records in the valid sampling chain; when the sorting position from the previous valid sampling record to the next valid sampling record moves toward the upper boundary of the historical stable window, an ascending symbol is generated; when the sorting position from the previous valid sampling record to the next valid sampling record moves toward the lower boundary of the historical stable window, a descending symbol is generated; when the previous valid sampling record and the next valid sampling record are in the same sorting position interval, a stable symbol is generated. Based on the symbol record, the valid sampling records corresponding to consecutive identical symbols are compressed into the symbol segment record; The symbol segment record includes the segment start point, segment end point, segment symbol, segment length, previous segment identifier, and next segment identifier; The rising symbol, the falling symbol, and the stable symbol are recorded separately and are not merged by mutual cancellation.

5. The method for processing ultrapure water quality anomaly data based on adaptive filtering according to claim 4, characterized in that, After generating the symbol segment record, the process also includes: Generate a channel change segment record based on the symbol segment record; When the same sign direction segment is located between the previous stable segment and the next stable segment, and the next stable segment returns to the sorting position interval of the previous stable segment, an impact segment record is generated. When rising and falling segments alternate, and the sorting position moves back and forth between the upper and lower sorting boundaries of the historical stable window, an oscillation segment record is generated. When an ascending or descending segment extends continuously, or moves continuously toward the same sorting boundary and then becomes a stable segment, a drift segment record is generated. When the same change segment has a single-direction segment and it cannot yet be determined whether it belongs to the impact segment record or the drift segment record, if a subsequent stable segment is formed after the single-direction segment and the subsequent stable segment returns to the sorting position interval of the previous stable segment, then an impact segment record is generated; if a subsequent stable segment is not formed after the single-direction segment and returns to the sorting position interval of the previous stable segment, and the subsequent sorting position continues to be located on the same sorting boundary side, then a drift segment record is generated.

6. The method for processing ultrapure water quality anomaly data based on adaptive filtering according to claim 5, characterized in that, Based on the symbol segment records and the process segment sequence records of the reverse osmosis outlet, electro-deionization module outlet, and polishing bed outlet, the symbol segment records of the same type of detection channel in adjacent process segments are retrieved, and process segment receipt records are generated, including: Based on the channel change segment record and the process segment sequence record of the target process segment, determine the actual upstream or downstream adjacent process segment of the target process segment. Based on the detection channel name of the target process segment, retrieve symbol segment records of the same type of detection channel from the upstream adjacent process segment or the downstream adjacent process segment to generate candidate records of adjacent process segments; Generate the process segment receipt record based on the adjacent process segment candidate records; Specifically, when the drift segment of the target process segment has a preceding drift segment of the same type of detection channel in the upstream adjacent process segment, and the temporal sequence of the two falls within the historical propagation waiting record of the corresponding adjacent process segment and the corresponding detection channel, an upstream receipt record is generated; when the drift segment of the target process segment has a subsequent drift segment of the same type of detection channel in the downstream adjacent process segment, and the temporal sequence of the two falls within the historical propagation waiting record of the corresponding adjacent process segment and the corresponding detection channel, a downstream receipt record is generated; when there is no same-direction segment of the same type of detection channel in the adjacent process segment, or when there is a same-direction segment but the temporal sequence does not fall within the historical propagation waiting record, an isolated record is generated.

7. The method for processing ultrapure water quality anomaly data based on adaptive filtering according to claim 6, characterized in that, Processing markers are generated based on the symbol segment records and the process segment receipt records, including: When both the impact segment record and the isolated record exist simultaneously, a sensor disturbance processing flag is generated; When both the oscillation segment record and the isolated record exist simultaneously, a channel oscillation processing flag is generated; When a drift segment record with the same target process segment, the same detection channel, and the same drift direction exists simultaneously with the corresponding upstream or downstream feedback record, a process fluctuation processing mark is generated. When both the drift segment record and the isolated record exist, a local link anomaly handling flag is generated; When the conditions for generating multiple processing markers are met simultaneously within the same target process segment, the same detection channel, and the same processing cycle, the processing markers are determined in the order of channel oscillation processing marker, sensor disturbance processing marker, local link anomaly processing marker, and process fluctuation processing marker.

8. The method for processing ultrapure water quality anomaly data based on adaptive filtering according to claim 7, characterized in that, Generate filtered output data based on the processing markers, including: A temporary correction output is generated based on the sensor disturbance processing mark and the previous stable section before the impact segment. After the next stable section after the impact segment is formed, a final backfill output is generated based on the previous stable section and the next stable section. When the previous stable section before the impact segment does not exist, a record to be confirmed is generated. A channel oscillation limit output is generated based on the channel oscillation processing flag; The process fluctuation processing flag is used to generate a process fluctuation limit output; Generate maintenance constraint output based on the local link anomaly handling flag; The temporary correction output is used for real-time display and does not participate in the historical stability window update. The final backfill output is used to replace the temporary correction output and is allowed to participate in the historical stability window update. The maintenance restriction output is used to generate a maintenance prompt record containing the process section name, detection channel name, drift start point, drift direction, and section boundary.

9. The method for processing ultrapure water quality anomaly data based on adaptive filtering according to claim 8, characterized in that, The historical stable window and dynamic boundary records are updated according to the baseline participation marker, including: Based on the filtered output data and the baseline participation flag, normal output records and final backfill outputs are allowed to enter the historical stability window; Based on the filtered output data and the reference participation flag, temporary correction output, channel oscillation limit output, maintenance limit output, pending confirmation record, and display record corresponding to the display sample are prohibited from entering the historical stability window; When the upstream or downstream receipt record corresponding to the process fluctuation limit output is continuously valid from the start of the drift segment of the target process segment to the closure of the next stable segment of the target process segment, the process fluctuation limit output is converted into a candidate output that is allowed to participate in the long-term benchmark update. The candidate outputs are reviewed. If no new isolated oscillation segment, maintenance-limited output, or communication anomaly sampling occurs after the candidate output, the candidate output is converted into an output record that is allowed to participate in the benchmark update. The dynamic boundary record is updated based on the closed stable segment and the historical stable window; when the historical stable window is empty or the dynamic boundary record cannot be updated, the previous dynamic boundary record is used; when there is no previous dynamic boundary record, the corresponding detection channel is kept in the initial accumulation state.

10. A system for processing abnormal ultrapure water quality data based on adaptive filtering, characterized in that, It includes water quality sensor nodes, industrial bus, data processing unit, communication transmission module and plant monitoring system; The data processing unit includes an effective sampling chain generation module, a symbol segment generation module, a process segment feedback verification module, and an output boundary control module. The effective sampling chain generation module is used to generate an effective sampling chain based on sampling records and data quality markers; The symbol segment generation module is used to generate sorted position records based on the effective sampling chain and the historical stable window, and to generate symbol segment records based on changes in adjacent sorted positions. The process segment receipt verification module is used to generate a process segment receipt record based on the symbol segment record and the process segment sequence record, and to generate a processing mark based on the symbol segment record and the process segment receipt record; The output boundary control module is used to generate filtered output data according to the processing marker, and update the historical stable window and dynamic boundary record according to the benchmark participation marker; The communication transmission module is used to send the filtered output data to the plant monitoring system.

Images