CN-121554019-B - Drinking water production line water quality real-time monitoring and intelligent regulating system based on Internet of things

CN121554019BCN 121554019 BCN121554019 BCN 121554019BCN-121554019-B

Abstract

The invention relates to the technical field of production line control, and discloses a drinking water production line water quality real-time monitoring and intelligent adjusting system based on the Internet of things, wherein registration equipment and units are unified, NTP time alignment eliminates deviation, shift/CIP calibration zero range, EWMA monitoring drift weight reduction, construction of a state vector containing ID/batch and audit warehousing, median/bilateral filtering denoising, unified grid interpolation, sliding window multi-scale aggregation, adaptive sampling according to curvature and events, near-time training window normalization and interpolation, soft sensor estimation of difficult-to-measure indexes, residual chlorine characterization by a mechanism model, uncertainty weighted fusion, output estimation and interval and audit. The invention realizes the whole process treatment of 'real-time monitoring-quick response-traceable compliance', obviously reduces the risks caused by shortage of chemicals or proportioning errors and environmental protection constraint, reduces the productivity loss caused by cleaning and planning conflict, and improves the stability and the treatment transparency of the water quality of the delivery factory.

Inventors

JIANG BOWEN
GAO HONGJIE
WANG LI

Assignees

辽宁岭秀山矿泉饮品有限公司

Dates

Publication Date: 20260508
Application Date: 20260121

Claims (10)

1. Drinking water production line quality of water real time monitoring and intelligent regulation system based on thing networking, its characterized in that includes: The data access and self-calibration module is used for integrating registration equipment and units, eliminating deviation when NTP pairs are performed, correcting zero range by shift/CIP, monitoring drift weight reduction by EWMA, constructing a state vector containing ID/batch and auditing and warehousing; The edge preprocessing and self-adaptive sampling module is used for denoising through median/bilateral filtering and uniform grid interpolation; sliding window multi-scale aggregation, and self-adaptive sampling according to curvature and events; The soft sensor and mechanism fusion module comprises a near-time training window normalization and interpolation, a soft sensor estimation difficult-to-measure index, a mechanism model characterization residual chlorine, uncertainty weighted fusion, an output estimation and interval and audit; The anomaly detection and root cause positioning module is used for standardizing a base line, identifying anomaly and change points by using a Mahalanobis distance and CUSUM/PH, sequencing root causes based on a process topology score, and generating treatment suggestion pushing and trace leaving; The prediction layer module is used for extracting multi-scale characteristics and outputting short-medium term prediction and uncertainty to support MPC and early warning; The intelligent regulation module compensates dead zone/hysteresis and applies NPSH constraint, risk arbitration MPC and RL, minimum residence and CBF navigation protection, overtime degradation drive-by-wire, support gray rollback and manual/stop audit; The CIP scheduling optimization module is used for triggering differential pressure/flux criteria, restraining stock, emission and window, selecting a robust formula through membrane punishment, calculating flushing volume and time sequence, generating scheduling, auditing and writing back a knowledge graph; the HMI interaction and alarm linkage module is used for presenting trends, prediction intervals, arbitration weights and suggestions, alarm grading merging, delay confirmation, interlocking and double approval, mode management and full audit and supporting track playback.
2. The system for real-time monitoring and intelligent regulation of water quality in a potable water production line based on the internet of things according to claim 1, characterized by executing the following steps through a data access and self-calibration module: S110, unifying equipment registration and units, namely checking field equipment and sensors, collecting model numbers/measuring ranges/units and installation positions, establishing equipment channel unit mapping, performing unit standardization and conversion, generating global unique equipment ID and configuration items, writing the configuration items into a configuration center and an audit library, and outputting equipment lists and calibration metadata; S120, clock synchronization and time stamp alignment, wherein a gateway periodically clocks NTP or a local time service source, calculates clock deviation of equipment or the gateway, corrects a new sample time stamp at a data receiving end, records a correction value and a source, alarms and requires equipment to reset or re-clock if the deviation exceeds a threshold value, and ensures that cross-equipment data are comparable on the same time grid; S130, zero point/measuring range calibration, namely collecting data of a stable section at a shift/CIP boundary, calculating zero point drift or a scaling factor, applying the zero point drift or the scaling factor to a corresponding channel in batch, executing two-point calibration, checking and calibrating effects, generating a calibration report and difference comparison, updating equipment state and version number, and keeping audit records; triggering in shift or CIP boundary event, correcting original value; if the proportion correction is needed, two-point calibration is adopted; s140, drift monitoring and self-diagnosis, namely real-time calculating the EWMA drift of each channel and comparing the EWMA drift with a threshold value, marking health labels when the channel is overrun and reducing the contribution of the channel to control and estimation, automatically creating maintenance worksheets and notices, and switching to a redundant sensor when closed loop input is unstable; s150, constructing a state vector, namely carrying out consistency check and missing fill-in on each source data, splicing the latest measurement into the state vector, attaching equipment ID, batch/shift and time stamp, and writing into a time sequence library and a message bus for consumption by an abnormality detection, prediction and control module.
3. The system for monitoring and intelligently adjusting the water quality of the drinking water production line based on the internet of things according to claim 1, wherein the following steps are executed through an edge preprocessing and self-adaptive sampling module: s210, denoising, namely selecting a median or bilateral filtering parameter according to channel noise characteristics, suppressing mutation and burr, outputting a denoising sequence, and marking filtered points in a data quality label; S220, time alignment, namely establishing a uniform time grid, performing linear or spline interpolation according to a source data time stamp, cleaning and resampling repeated and jump time stamps, outputting an aligned multi-source sequence and recording interpolation proportion; s230, sliding aggregation, namely maintaining a ring buffer area to store the latest window data, and calculating statistics of a mean value, a variance, an extremum and RMS to form window characteristics; S240, self-adaptive sampling, namely calculating the change rate/gradient and event indication of the time sequence in real time, updating the sampling frequency according to a formula, and issuing the updated sampling frequency to a gateway/PLC acquisition task, monitoring the occupation of CPU and bandwidth, and automatically limiting current and reducing sampling when triggering a resource protection threshold.
4. The system for monitoring and intelligently adjusting the water quality of the drinking water production line based on the internet of things according to claim 1, wherein the following steps are executed through a soft sensor and mechanism fusion module: s310, constructing a training window, namely constructing a characteristic tensor by using a sliding window, executing dimension normalization and missing value interpolation, generating a data batch for the soft sensor and the prediction model, and caching the data batch at the edge; s320, soft sensor estimation, namely calling a selected soft sensor model to perform forward reasoning, outputting on-line estimation and uncertainty of a target index, continuously recording estimation errors, and performing incremental fine adjustment or model version rolling update according to a plan; S330, a mechanism model, namely adaptively updating residual chlorine attenuation coefficient and administration gain according to temperature, pH and process section conditions, calculating residual chlorine prediction at the next moment, checking whether the residual chlorine prediction meets a regulation threshold, and inputting a result as control constraint; s340, uncertainty weighted fusion, namely evaluating the variances of the two paths of estimation of the mechanism and the data driving, performing inverse variance weighted fusion, generating final estimation and confidence and pushing the final estimation and confidence to an anomaly detection and prediction and control module to form consistent state input.
5. The system for monitoring and intelligently adjusting the water quality of the drinking water production line based on the internet of things according to claim 1, wherein the following steps are executed by the anomaly detection and root cause positioning module: S410, normalization and baseline, namely calculating historical mean and standard deviation of each parameter, generating a normalization sequence and maintaining seasonal or batch baselines, and periodically reviewing and updating to reduce drift influence; S420, counting distance alarm, namely calculating the Marsh distance in real time and comparing the Marsh distance with a threshold value, generating an alarm event by exceeding the threshold value, entering a queue, and simultaneously recording influence parameters and context to trigger root cause positioning; s430, detecting a change point, namely applying CUSUM/PH test on a standardized sequence, outputting a change point position and a type label, and carrying out time proximity correlation with a process event; s440, root cause positioning, namely executing sub-graph search and causal scoring based on the process topology, generating a root cause ordering and suggested action list, pushing to an HMI and control layer to perform 'automatic/semi-automatic/manual' execution selection and recording audit.
6. The system for monitoring and intelligently adjusting the water quality of the drinking water production line based on the internet of things according to claim 1, wherein the following steps are executed through the prediction layer module: s510, modeling a state space, namely obtaining system matrix and noise item setting through data-driven identification or mechanism deduction, completing model verification and cross check, outputting a prediction model capable of being used for MPC and registering a version; s520, multi-scale feature extraction and attention, namely constructing a four-layer hierarchical structure, calculating attention weight and significance score, generating an explanatory heat map and a key feature list, and feeding back to operation and control strategy optimization; And S530, predicting and uncertainty, namely rolling a plurality of steps in the future of prediction, calculating variance and confidence interval, issuing to a controller and an HMI for early adjustment and alarm suppression, and recording prediction errors for continuous learning.
7. The system for monitoring and intelligently adjusting the water quality of the drinking water production line based on the internet of things according to claim 1, wherein the following steps are executed through an intelligent adjusting module: S610, dead zone/hysteresis compensation of the actuator, namely identifying the dead zone and hysteresis characteristics of the actuator through a step/slope test, and applying a compensation function and saturation protection on a real-time control signal; S620, pump cavitation safety constraint, namely calculating an available net positive suction pressure head in real time and comparing the available net positive suction pressure head with a minimum net positive suction pressure head required by equipment, if the available net positive suction pressure head is lower than a safety margin, reducing pump frequency or adjusting valve position/bypass, generating a protection alarm and recording a triggering reason; s630, constraint softening and objective function, namely constructing constraint and cost functions with softening variables, calling an optimization solver to return a control sequence, checking feasibility and constraint violation degree, and recording penalty guarantee; S640, performing MPC/RL hybrid arbitration and minimum residence, namely calculating arbitration weight according to risk scores, fusing MPC and RL output, enforcing minimum residence time, and recording switching events and reasons; S650, solving dimension reduction and overtime degradation, namely monitoring the time consumption and convergence state of solving, shortening the prediction domain and the variable dimension and starting at the same time under the condition that the preset triggering condition is met; The preset triggering conditions comprise that the time consumption of solving reaches a soft threshold, the convergence quality is not up to standard, the progress is stopped, the time is continuously closed for a plurality of times but not overtime, the time consumption of solving exceeds a hard threshold, the iteration reaches an upper limit and still not converged, and the soft triggering continuously occurs to reach an upper limit; s660, reinforcing a safety barrier, namely evaluating CBF inequality and safety set constraint on candidate control quantity, and correcting or refusing the action if the CBF inequality and the safety set constraint are not met, so as to ensure that all issued commands are positioned in a safety executable control quantity set; s670, RL security and online policy, namely pre-training the policy on offline replay data and embedding security constraints, online and rollable production environment gray scales, online small-step fine tuning and gating output by an arbiter; s680, manual coverage daemon, namely continuously monitoring HMI mode bit, entering manual/stop, namely freezing automatic control and executor output and recording audit.
8. The system for monitoring and intelligently adjusting the water quality of a potable water production line based on the Internet of things according to claim 7, wherein the calculation formula of the MPC/RL hybrid arbitration and minimum residence is as follows: ; ; ; Wherein, the For the purpose of risk scoring, Is a temperature coefficient of the silicon carbide material, In order for the minimum residence time to be a minimum, In order to arbitrate the weights, the weight of the data, Is a Sigmoid function; the time interval for switching the control sources is two adjacent times; the control amounts of the MPC and the RL are respectively, Is a control amount; The reward function for RL security and online policy is as follows: ; Wherein, the Corresponding to water quality loss, energy consumption, medicine consumption, action switching and safety constraint rewards respectively; Is the water quality deviation loss; in order to satisfy the indication function of the constraint, Is a prize value; In order to be able to consume energy, In order to achieve the desired level of consumption of the medicament, To control the number of source/action switches.
9. The system for monitoring and intelligently adjusting the water quality of the drinking water production line based on the internet of things according to claim 1, wherein the following steps are executed through a CIP scheduling optimization module: s710, triggering criteria, namely periodically evaluating transmembrane pressure difference and membrane flux, setting hysteresis, generating CIP candidate events and entering a constraint evaluation stage; S720, checking chemical inventory, emission indexes and production windows, calculating feasible sets and priorities, and outputting executable time windows and bill of materials; S730, selecting formula concentration and estimating membrane damage penalty in a formula concentration robust interval, weighing cleaning effect and membrane service life, and screening candidate schemes; S740, calculating the required flushing volume and time according to the residual model, generating a flushing program and a valve pump action time sequence, and synchronizing the flushing program and the valve pump action time sequence for execution of the HMI and the PLC; S750, comprehensively optimizing, namely constructing an optimization model containing resource/environmental protection/plan constraint, obtaining opportunity, formula and stage duration by adopting MILP or heuristic solution, issuing execution and collecting effect data closed-loop optimization parameters.
10. The system for real-time monitoring and intelligent regulation of water quality of a potable water production line based on the Internet of things according to claim 1, characterized by executing the following steps through an HMI interaction and alarm linkage module: s810, trend and prediction display, namely superposing a real-time curve and a prediction interval on the HMI, displaying arbitration weights and suggested actions, providing one-key execution or semi-automatic options and displaying security assessment; s820, alarm grading and restraining, namely grading, de-duplicating and merging the alarms, setting a delay confirmation and restraining strategy, supporting batch confirmation and remarks, and reducing the risk of alarm storm; s830, interlock and secondary confirmation, wherein key actions trigger the check of interlock conditions, which can be executed after the double confirmation, and the execution and failure record events and reasons for tracing; S840, mode management and audit, namely providing mode switching guidance and risk prompt, recording operator, time and reason, automatically synchronizing the controller state and checking the safety boundary; S850, traceable and replayable, save anomaly and control trajectories to a timing/audit library, provide search screening and export capabilities, generate compliance reports and event replay to support auditing and improvement.

Description

Drinking water production line water quality real-time monitoring and intelligent regulating system based on Internet of things Technical Field The invention relates to the field of production line control, in particular to a drinking water production line water quality real-time monitoring and intelligent regulating system based on the Internet of things. Background Along with the continuous evolution of the drinking water industry to large-scale, continuous and high-standard quality supervision, the traditional water quality control mode with offline test as a core is difficult to meet the overall process treatment requirement of 'real-time monitoring-quick response-traceable compliance'. In the traditional water quality control method, when dead zones and hysteresis (valve blocking and pump cavitation) of an actuator occur, continuous and discrete actions are mutually interfered to cause oscillation, or chemicals are in shortage or concentration ratio error, CIP liquid discharge disposal is limited and environmental protection rules are changed, and cleaning beats collide with a production plan to cause productivity loss. Disclosure of Invention The invention provides a drinking water production line water quality real-time monitoring and intelligent adjusting system based on the Internet of things, which solves the technical problems of dead zone and hysteresis of an actuator, shortage of chemicals or concentration proportioning error in the related art. The invention provides a drinking water production line water quality real-time monitoring and intelligent regulating system based on the Internet of things, which comprises the following components: The data access and self-calibration module is used for integrating registration equipment and units, eliminating deviation when NTP pairs are performed, correcting zero range by shift/CIP, monitoring drift weight reduction by EWMA, constructing a state vector containing ID/batch and auditing and warehousing; The edge preprocessing and self-adaptive sampling module is used for denoising through median/bilateral filtering and uniform grid interpolation; sliding window multi-scale aggregation, and self-adaptive sampling according to curvature and events; The soft sensor and mechanism fusion module comprises a near-time training window normalization and interpolation, a soft sensor estimation difficult-to-measure index, a mechanism model characterization residual chlorine, uncertainty weighted fusion, an output estimation and interval and audit; The anomaly detection and root cause positioning module is used for standardizing a base line, identifying anomaly and change points by using a Mahalanobis distance and CUSUM/PH, sequencing root causes based on a process topology score, and generating treatment suggestion pushing and trace leaving; The prediction layer module is used for extracting multi-scale characteristics and outputting short-medium term prediction and uncertainty to support MPC and early warning; The intelligent regulation module compensates dead zone/hysteresis and applies NPSH constraint, risk arbitration MPC and RL, minimum residence and CBF navigation protection, overtime degradation drive-by-wire, support gray rollback and manual/stop audit; The CIP scheduling optimization module is used for triggering differential pressure/flux criteria, restraining stock, emission and window, selecting a robust formula through membrane punishment, calculating flushing volume and time sequence, generating scheduling, auditing and writing back a knowledge graph; the HMI interaction and alarm linkage module is used for presenting trends, prediction intervals, arbitration weights and suggestions, alarm grading merging, delay confirmation, interlocking and double approval, mode management and full audit and supporting track playback. Further, the following steps are performed by the data access and self-calibration module: S110, unifying equipment registration and units, namely checking field equipment and sensors, collecting model numbers/measuring ranges/units and installation positions, establishing equipment channel unit mapping, performing unit standardization and conversion, generating global unique equipment ID and configuration items, writing the configuration items into a configuration center and an audit library, and outputting equipment lists and calibration metadata; S120, clock synchronization and time stamp alignment, wherein a gateway periodically clocks NTP or a local time service source, calculates clock deviation of equipment or the gateway, corrects a new sample time stamp at a data receiving end, records a correction value and a source, alarms and requires equipment to reset or re-clock if the deviation exceeds a threshold value, and ensures that cross-equipment data are comparable on the same time grid; S130, zero point/measuring range calibration, namely collecting data of a stable section at a shift/CIP boundary, calculating zero point dri