A waterworks double closed loop coagulant intelligent dosing system and a control method thereof

Abstract

The present application relates to the technical field of smart water and process control, and particularly relates to a double closed loop coagulant intelligent dosing system for waterworks and a control method thereof. The double closed loop intelligent dosing system comprises a multi-modal perception network, an edge intelligent controller, a cloud edge system optimization platform, and a safety execution and alarm unit. The control method thereof is also disclosed. When the model credibility index is higher than a set threshold, model predictive control is adopted. When the index is lower than the threshold, the control is switched to a conventional control scheme. In order to improve the automatic disposal capability for abnormal working conditions, an event-driven closed loop control is set. For abnormal working conditions beyond the conventional working conditions, the abnormal working condition disposal rule control is switched, so as to form a double closed loop system architecture. The control efficiency of the coagulant intelligent dosing system under normal working conditions can be effectively improved, and the abnormal working conditions can be effectively disposed. The problem of model failure caused by abnormal water inflow and other reasons, which leads to a sharp decline in system performance, can be avoided.

Description

Technical Field

[0001] This invention relates to the field of smart water management and process control, and in particular to a dual closed-loop intelligent coagulant dosing system and its control method for waterworks. Background Technology

[0002] Over the years, the dosing of coagulants in water treatment plants has evolved from initially relying on human experience to gradually developing into proportional control based on influent flow rate, feedback control based on effluent turbidity, simple feedforward-feedback composite control, and in recent years, more complex intelligent control based on machine learning algorithms. This has continuously driven the dosing of coagulants towards greater precision and reduced consumption, meeting the growing demand for improved quality and efficiency in water treatment plants.

[0003] However, these existing control methods still have some technical limitations that urgently need to be addressed in practice, mainly in the following aspects: The method based on the proportion of influent flow rate ignores many factors that change in influent water quality, such as influent turbidity, pH, and temperature. At the same time, it cannot respond to fluctuations in the water treatment process itself, resulting in large fluctuations in effluent turbidity. In the PID feedback control method, due to the long hydraulic retention time in the sedimentation tank (120–180 minutes), relying on the turbidity feedback of the effluent results in a serious lag, which can easily lead to an oscillating cycle of "dosing-excessive standard-dosing". The feedforward-feedback composite control method has solved the problems of the above two methods to a certain extent. However, feedforward control relies on simple empirical formulas (such as dosage = k × influent turbidity) and cannot adapt to the coupled influence of multiple factors such as pH, temperature, and organic matter. Therefore, there is still considerable room for improvement in control performance. In intelligent control methods, to address the problem that simple formulas in feedforward control cannot reflect the coupled effects of complex influent factors, a model is established to reflect the complex relationships among these factors, thereby improving the accuracy of feedforward control. Commonly used modeling methods include: data-driven models (such as LSTM, XGBoost, etc.): these require massive amounts of labeled data, resulting in long engineering cycles, and their generalization ability drops sharply when water sources change abruptly, leading to a drastic reduction in control effectiveness; Mechanistic models (such as colloid destabilization dynamics) are complex to identify, computationally intensive, and difficult to deploy in real time. As with all model-based controls, the performance of the model significantly impacts the overall control effect. In intelligent control of concrete dosing, due to the nonlinearity, large time lag, time-varying nature of the coagulation process itself, and the complex and unpredictable sudden characteristics of external disturbances, model-based intelligent control faces the following significant problems in concrete dosing intelligent control engineering: On the one hand, this leads to a complex modeling process, and the model's efficient and reliable range is limited. The established model only achieves good accuracy in local ranges, thus obtaining the desired control effect, while its accuracy is low in other ranges, failing to achieve the required control effect. On the other hand, the model cannot solve the prediction problem of ranges not covered by training data. Sudden events in the natural environment, such as sudden changes in influent water quality caused by sudden pollution of water sources, are situations that the model has never encountered during training or identification, and lacks relevant training data. When faced with such situations, the model will fail, leading to control failure. Therefore, how to solve the above difficulties in applying intelligent control in coagulant dosing is an important development direction for intelligent coagulant dosing. Summary of the Invention

[0004] The purpose of this invention is to provide a dual closed-loop intelligent coagulant dosing system and its control method for waterworks, so as to improve the control efficiency of the intelligent coagulant dosing system under normal operating conditions and effectively handle abnormal operating conditions, thereby avoiding the problem of model failure caused by abnormal water intake and other reasons, which leads to a sharp decline in system performance.

[0005] To achieve the above objectives, the present invention provides a dual closed-loop intelligent coagulant dosing system for waterworks, comprising: A multimodal sensing network is used to detect basic data of water inflow and outflow from sensors and store it in a time-series database; The edge intelligent controller includes an independently operating event-driven controller and a composite model predictive controller. The event-driven controller, based on acquired influent and effluent data, determines whether the influent and effluent are abnormal. If abnormal, it generates an influent / effluent abnormality event message, locks the composite model predictive controller, and controls the addition of coagulant according to its own abnormal control rule set. If the abnormality disappears, it generates an influent / effluent abnormality disappearance message, releases the lock on the composite predictive controller, and stops its own operation. The composite model predictive controller is used to construct predictive models for multiple sub-intervals and switches the control mode from model predictive control to a high-reliability, low-control-economic control mode based on set thresholds and confidence levels. The cloud-edge system optimization platform is used for model retraining, energy efficiency analysis, expert knowledge base updates, and the distribution of new models to edge intelligent controllers. The safety execution and alarm unit uses an independent PLC to achieve hard-wired interlocking and forms a three-level alarm mechanism.

[0006] The event-driven controller includes: The water inlet anomaly detection module analyzes and determines whether the water inlet is abnormal based on the acquired water inlet data. If it is abnormal, it sends a water inlet anomaly event message; if the anomaly disappears, it sends a water inlet anomaly disappearance message. The water outflow anomaly detection module analyzes and determines whether the water outflow is abnormal based on the acquired water outflow data. If it is abnormal, it sends out a water outflow anomaly event message; if the anomaly disappears, it sends out a water inflow anomaly disappearance message. The production process abnormal event communication module is used to communicate with the production system to obtain abnormal event messages or disappearance information during the production process; A closed-loop interlocking module for the composite predictive model controller is used to implement interlocking of the composite predictive model controller; The abnormal operating condition controller is used to call the closed-loop interlocking module of the composite predictive model controller to interlock the composite predictive model controller when it receives a message of an abnormal event, and to control the addition of coagulant according to its own corresponding abnormal event control rule set. When it receives a message of the disappearance of the abnormal event, it calls the closed-loop interlocking module of the composite predictive model controller to release the interlocking of the composite predictive model controller, and at the same time stops its own control operation.

[0007] The influent anomaly detection module analyzes and determines whether the influent is abnormal based on the acquired influent data (turbidity, pH, temperature, flow rate, etc.). If it is abnormal, it sends an influent anomaly event message; if the anomaly disappears, it sends an influent anomaly disappearance message. The effluent anomaly detection module analyzes and determines whether the effluent is abnormal based on the acquired effluent data. If it is abnormal, it sends an effluent anomaly event message; if the anomaly disappears, it sends an influent anomaly disappearance message. The production process anomaly event communication module communicates with the production SCADA system or other production systems to obtain production process anomaly event occurrence or disappearance messages. When the abnormal condition controller receives an anomaly event occurrence message, it calls the composite predictive model controller closed-loop interlocking module to interlock the composite predictive model controller and performs coagulant dosing control according to its corresponding anomaly event control rule set. When it receives an anomaly event disappearance message, it calls the composite predictive model controller closed-loop interlocking module to release the interlock on the composite predictive model controller and stops its own control operation. The composite predictive model controller closed-loop interlocking module is used to implement closed-loop interlocking of the composite predictive model controller.

[0008] Furthermore, the composite model predictive controller includes: The segmented adaptive coagulation dynamic model is configured to be divided into multiple sub-intervals according to the influent turbidity range, and corresponding prediction models are established to reduce the difficulty of modeling a unified interval and improve the overall prediction accuracy of the model. The prediction model credibility index calculation module is configured to calculate the credibility of the prediction model according to the prediction model feasibility index. When the credibility is lower than the set threshold, the control mode will be switched from model prediction control to a control mode with high credibility and low control economy. The model predictive control module is configured to perform model predictive control of coagulant dosing based on a segmented adaptive coagulation dynamic model. The high control confidence and low control economy control module is configured as a non-model conventional control method, which executes control when the confidence of the prediction model is below a threshold.

[0009] The segmented adaptive coagulation dynamic model is divided into multiple sub-intervals according to the influent turbidity range, and corresponding prediction models are established to reduce the difficulty of modeling a unified interval and improve the overall prediction accuracy of the model. The prediction model credibility index calculation module calculates the feasibility of the prediction model according to the defined prediction model feasibility index. When the credibility is lower than a set threshold, the control mode will be switched from model prediction control to a high credibility, low control economy control mode. The model prediction control module includes a rolling optimization unit and an online correction unit, which performs model prediction control on the addition of coagulant based on the segmented adaptive coagulation dynamic model. The high control credibility, low control economy control module performs control based on non-model conventional control methods when the credibility of the prediction model is lower than the threshold.

[0010] This invention also discloses a control method for a dual closed-loop coagulant intelligent dosing system in a waterworks, comprising the following steps: S1: Construct a segmented adaptive coagulation dynamic model; S2: Design an MPC optimizer with multiple constraints; S3: Design feedback correction for the MPC controller; S4: Construct a reliability index for the prediction model; S5: Construct a composite model predictive controller; S6: The control program predicts the threshold setting within the controller according to the composite model and selects the execution control mode based on the confidence level.

[0011] In step S2, the objective function for designing the MPC optimizer with multiple constraints is: in, The control objective is water quality tracking. The control objective is to inhibit drug consumption. The control objective is to achieve smooth execution.

[0012] The objective function must satisfy the hard constraint of effluent turbidity: Dosage limits: Execution speed: The optimization logic is based on the influent water quality stability index. adjust .

[0013] in, The comprehensive fluctuation rate of recent influent water quality parameters is given by T, which is a time constant used to adjust the sensitivity of the index to fluctuations (e.g., T=0.2). The detailed calculation steps are as follows: Select key influent parameters: these typically include turbidity (d1), pH (d2), and temperature (d3). Calculate the volatility of each parameter: Within a sliding window of length L (e.g., L = 30 minutes, consistent with the model confidence assessment window in the document), calculate the ratio of the standard deviation to the mean for each parameter: Calculate the overall volatility: Take a weighted average of the volatility of each parameter and normalize it so that its range is around [0,1]. in is the weight of the (m)th parameter, which can be set according to its influence on the coagulation process (e.g., turbidity has the highest weight); C is a scaling constant used to map the results to a reasonable range, which can be calibrated based on historical data.

[0014] Substituting into the exponential function: Finally, Substitution According to calculations, when the water quality is stable, i.e., S > 0.9, increasing μ prioritizes energy saving; when the water quality fluctuates, i.e., S < 0.9, increasing λ prioritizes water quality preservation.

[0015] In step S1, the construction of the segmented adaptive coagulation dynamic model specifically refers to: Divide the influent turbidity into N operating ranges; The following identification model is established independently for each interval: Among them, a i The weight of the impact of the corresponding disturbance input, b i The influence weight of the corresponding control input, c i The weights of the influence of the corresponding system inertia / historical state.

[0016] The construction of the piecewise adaptive cohesive dynamic model also includes real-time model updating using online recursive least squares method, with the specific steps as follows: Define the model parameter vector: Constructing the regression vector: Perform online identification and update: The forgetting factor λ = 0.98 and the initial covariance matrix P(0) = 1000I.

[0017] The default value is N=3, and the range of values ​​is [0, 10], [10, 50], [50, ∞).

[0018] In step S4, the construction of the prediction model credibility index specifically includes the following steps: Calculate the standard deviation normalization index of the residuals; Set a rate of change index for model parameters to detect whether there are drastic fluctuations in model parameters; Calculate the overall credibility index; Set thresholds and action logic.

[0019] In step S3, the feedback correction of the MPC controller refers to correcting future predictions every 10 minutes using the latest measured turbidity of the effluent. in, The current prediction error is α = 0.7. Furthermore, the standard deviation formula is: Among them, the sliding window length Nwin =30, y meas The meaning is the measured turbidity of the water. y pred This refers to the model's predicted value.

[0020] The normalized index is: in, The average measured turbidity over the past 30 minutes.

[0021] The aforementioned model parameter change rate index: The comprehensive credibility index is as follows: Wherein, the weighting coefficient α=0.6, RSD(k)∈[0,1], if RSD(k) is greater than 1, then set to 1, PCR(k)∈[0,1]: if PCR(k) is greater than 1, then set to 1.

[0022] This invention discloses a dual-closed-loop intelligent coagulant dosing system and its control method for waterworks. The system continuously monitors model performance. When the model confidence index exceeds a set threshold, model predictive control is employed; when it falls below the threshold, control switches to a conventional control scheme with high deterministic control effect but relatively low economic efficiency. Simultaneously, to enhance the automated handling capability of abnormal operating conditions, event-driven closed-loop control is implemented. For abnormal operating conditions exceeding normal operating conditions, the system switches to rule-based control for handling abnormal operating conditions, thus forming a dual-closed-loop system architecture. This effectively improves the control efficiency of the intelligent coagulant dosing system under normal operating conditions and effectively handles abnormal operating conditions. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the architecture of a dual closed-loop intelligent coagulant dosing system for waterworks according to the present invention.

[0025] Figure 2 This is a schematic diagram of the composite model predictive controller of a dual closed-loop coagulant intelligent dosing system for waterworks according to the present invention.

[0026] Figure 3 This is a schematic diagram of the event-driven controller of a dual closed-loop coagulant intelligent dosing system for waterworks according to the present invention.

[0027] Figure 4 This is a structural diagram of the dual closed-loop system of the intelligent dosing system for coagulants in a waterworks according to the present invention.

[0028] Figure 5 This is a schematic diagram illustrating the steps of a control method for a dual closed-loop intelligent coagulant dosing system for waterworks according to the present invention. Detailed Implementation

[0029] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0030] Please see Figures 1 to 4 A dual closed-loop intelligent coagulant dosing system for waterworks, characterized in that it includes: A multimodal sensing network is used to detect basic data of water inflow and outflow from sensors and store it in a time-series database; The edge intelligent controller includes an independently operating event-driven controller and a composite model predictive controller. The event-driven controller, based on acquired influent and effluent data, determines whether the influent and effluent are abnormal. If abnormal, it generates an influent / effluent abnormality event message, locks the composite model predictive controller, and controls the addition of coagulant according to its own abnormal control rule set. If the abnormality disappears, it generates an influent / effluent abnormality disappearance message, releases the lock on the composite predictive controller, and stops its own operation. The composite model predictive controller is used to construct predictive models for multiple sub-intervals and switches the control mode from model predictive control to a high-reliability, low-control-economic control mode based on set thresholds and confidence levels. The cloud-edge system optimization platform is used for model retraining, energy efficiency analysis, expert knowledge base updates, and the distribution of new models to edge intelligent controllers. The safety execution and alarm unit uses an independent PLC to achieve hard-wired interlocking and forms a three-level alarm mechanism.

[0031] The multimodal sensing network collects basic data such as influent and effluent turbidity, influent pH, influent temperature, influent flow rate, and COD. If necessary, additional sensors such as OPR and DO can be added for more comprehensive water quality monitoring. The collected sensor data is stored in a time-series database as the basis for modeling and real-time control. The edge intelligent controller hardware is an industrial-grade ARM / Intel processor, supporting OPC UA and Modbus TCP protocols. The software uses embedded Linux + Docker containers; the event-driven controller and the composite model prediction controller are implemented as containers, running independently and isolated from communication and storage services. The edge of the cloud-edge collaborative optimization platform performs second-level MPC rolling optimization control, while the cloud performs model retraining, energy efficiency analysis, expert knowledge base updates, and distributes new models to the edge. The safety execution and alarm unit uses independent PLCs for hard-wired interlocking, and the light alarm adopts a three-level alarm mechanism: alert (yellow), warning (orange), and interlock (red).

[0032] The event-driven controller includes: The water inlet anomaly detection module analyzes and determines whether the water inlet is abnormal based on the acquired water inlet data. If it is abnormal, it sends a water inlet anomaly event message; if the anomaly disappears, it sends a water inlet anomaly disappearance message. The water outflow anomaly detection module analyzes and determines whether the water outflow is abnormal based on the acquired water outflow data. If it is abnormal, it sends out a water outflow anomaly event message; if the anomaly disappears, it sends out a water inflow anomaly disappearance message. The production process abnormal event communication module is used to communicate with the production system to obtain abnormal event messages or disappearance information during the production process; A closed-loop interlocking module for the composite predictive model controller is used to implement interlocking of the composite predictive model controller; The abnormal operating condition controller is used to call the closed-loop interlocking module of the composite predictive model controller to interlock the composite predictive model controller when it receives a message of an abnormal event, and to control the addition of coagulant according to its own corresponding abnormal event control rule set. When it receives a message of the disappearance of the abnormal event, it calls the closed-loop interlocking module of the composite predictive model controller to release the interlocking of the composite predictive model controller, and at the same time stops its own control operation.

[0033] The composite model predictive controller includes: The segmented adaptive coagulation dynamic model is configured to be divided into multiple sub-intervals according to the influent turbidity range, and corresponding prediction models are established to reduce the difficulty of modeling a unified interval and improve the overall prediction accuracy of the model. The prediction model credibility index calculation module is configured to calculate the credibility of the prediction model according to the prediction model feasibility index. When the credibility is lower than the set threshold, the control mode will be switched from model prediction control to a control mode with high credibility and low control economy. The model predictive control module is configured to perform model predictive control of coagulant dosing based on a segmented adaptive coagulation dynamic model. The high control confidence and low control economy control module is configured as a non-model conventional control method, which executes control when the confidence of the prediction model is below a threshold.

[0034] Please see Figure 5 The present invention also discloses a control method for a dual closed-loop coagulant intelligent dosing system in a waterworks, comprising the following steps: S1: Construct a segmented adaptive coagulation dynamic model; S2: Design an MPC optimizer with multiple constraints; S3: Design feedback correction for the MPC controller; S4: Construct a reliability index for the prediction model; S5: Construct a composite model predictive controller; S6: The control program predicts the threshold setting within the controller according to the composite model and selects the execution control mode based on the confidence level.

[0035] In step S1, the construction of the segmented adaptive coagulation dynamic model specifically refers to: Divide the influent turbidity into N operating ranges; The following identification model is established independently for each interval: Where ai corresponds to the influence weight of the disturbance input, bi corresponds to the influence weight of the control input, and ci corresponds to the influence weight of the system inertia / historical state.

[0036] The construction of the piecewise adaptive cohesive dynamic model also includes real-time model updating using online recursive least squares method, with the specific steps as follows: Define the model parameter vector: Constructing the regression vector: Perform online identification and update: The forgetting factor λ = 0.98 and the initial covariance matrix P(0) = 1000I.

[0037] In step S4, the construction of the prediction model credibility index specifically includes the following steps: Calculate the standard deviation normalization index of the residuals; Set a rate of change index for model parameters to detect whether there are drastic fluctuations in model parameters; Calculate the overall credibility index; Set thresholds and action logic.

[0038] The standard deviation formula is: Among them, the sliding window length Nwin =30, y meas The meaning is the measured turbidity of the water. y pred This refers to the model's predicted value.

[0039] The normalized index is: in, The average measured turbidity over the past 30 minutes.

[0040] The aforementioned model parameter change rate index: .

[0041] The comprehensive credibility index is as follows: Among them, the weighting coefficient , If it is greater than 1, then set it to 1. If it is greater than 1, then set it to 1.

[0042] Threshold setting and action logic η(k) range Status Description Control strategy ≥ 0.8 High reliability Normal MPC optimization control [0.7, 0.8) Medium confidence Enable mild smoothing (increase R weights) <0.7 Low reliability Automatically switch to a control method that provides a certain level of control effectiveness but is not cost-effective (e.g., the feedforward-feedback controller used before the modification). Table 1 In the construction of the composite model predictive controller, The segmented prediction model is designed and implemented according to step S1 above; the rolling optimization module is designed and implemented according to step S2 above; the online correction module is designed and implemented according to step S3 above; and the prediction model credibility index module is designed and implemented according to step S4 above.

[0043] Activate emergency rule base control The sources of the emergency rule base are shown in the table below. source Content Example Historical operating data of water plant "When the influent turbidity is 50–100 NTU, the PAC dosage is 8–12 mg / L, and the effluent compliance rate is 95%." Process manual / specification Water Supply and Drainage Design Manual, Volume 3: Reference Table for Coagulant Dosage Expert experience interviews An experienced technician recounted: "When the pH is less than 6.5, an additional 10% of the chemical needs to be added; otherwise, the alum flocs will become loose." Small-scale / pilot-scale experiments Laboratory beaker tests were conducted to determine the optimal dosage for different water qualities. Table 2 Specific implementation examples: Overall hardware layout and selection like Figure 1 As shown, the system deployment in this embodiment strictly follows a hierarchical architecture of perception-edge-cloud-execution. Key hardware selection is based on the principles of industrial-grade reliability, protocol openness, and matching computing performance, with the specific configuration as follows: Hardware deployment of multimodal sensing networks: The following monitoring instruments are installed at key nodes of the water inlet pipeline to form a sensing network: Turbidity meter: Two HACH solitax turbidity meters are used (one for operation and one for standby), directly installed on the main inlet water pipe. The output signal (4-20mA and Modbus TCP) is connected to the I / O module of the field control station.

[0044] Selection criteria: 0-1000 NTU range, ±1% accuracy, can accurately capture turbidity fluctuations under working conditions such as "heavy rain and high turbidity" and "low temperature in winter", as shown in Table 3 for the assessment scenario of effluent turbidity exceeding the standard.

[0045] Integrated pH / temperature sensor: Employs the WTW pH / SensoLyt® 700 IQ sensor, installed before the mixing tank. Supports the Modbus protocol and can simultaneously provide both pH and temperature disturbance variables.

[0046] Selection criteria: IP68 protection, built-in temperature compensation, suitable for low-temperature winter scenarios, triggering conditions in Table 4: temperature <3℃.

[0047] UV 254 Organic matter analyzer: HACH UVAS plus, installed on the bypass pipe of the main inlet water pipe. Used to monitor changes in organic matter content in the inlet water, serving as a disturbance input for model feedforward compensation.

[0048] Selection criteria: Response time < 5 minutes, meeting the rapid detection requirements in algal bloom scenarios. Table 4 Triggering conditions: UV 254>0.3.

[0049] Hardware layout of the edge intelligent controller: The Advantech ARK-3530 industrial fanless PC, equipped with a processor with a main frequency of up to 3.6GHz, 8GB of memory, and 256GB of SSD, was selected and deployed in the central control room of the water plant.

[0050] Selection criteria: This configuration is designed to run multiple containerized Docker services, including data acquisition, providing sufficient computational margin for event-driven controllers, MPC controllers, and time-series databases, ensuring that the rolling optimization cycle of Model Predictive Control (MPC) can be stably kept within 5 seconds.

[0051] Network and Deployment: Equipped with dual gigabit network cards. One port connects to the water plant's real-time Ethernet control network for communication with PLCs and I / O stations; the other port connects to the water plant's management information network for communication with the cloud platform. The equipment is installed in a 19-inch standard rack in the central control room and is equipped with a 2-hour online UPS.

[0052] Hardware implementation of the safety execution and alarm unit Independent safety PLC: Rockwell AB1756 series PLC modules are used. Selection criteria: consistent with the mainstream control system brands already in the water plant, facilitating reliable interlocking through hardwiring, reducing integration complexity and maintenance costs.

[0053] Actuator: Grundfos DME 940-4 series variable frequency metering pump is used for PAC (particulate coagulant) dosing. The safety PLC controls the pump frequency through the analog output module (4-20mA) and monitors the pump's operation and fault status in real time through the digital input module.

[0054] Hard-wired safety logic: The core interlocking program is programmed in the safety PLC, and this logic is completely independent of the edge controller. In one embodiment, when a communication interruption with the edge controller is detected for more than 5 seconds, or the effluent turbidity sensor value continuously exceeds 0.5 NTU (safety limit), the system immediately and automatically switches to the preset backup PID control loop inside the PLC without disturbance, and triggers an audible and visual alarm (red light).

[0055] Edge intelligent controller Ubuntu Server 24.04 LTS was deployed and installed on the edge controller, with Docker Engine 24.0+ deployed as the basic runtime environment. A dual-closed-loop coagulant dosing control system was then deployed on top of this, with the software architecture as follows: Figure 2 As shown, sensor network data is collected via Modbus TCP protocol, with a sampling frequency of 1 time / 10 seconds.

[0056] Composite Model Predictive Controller: Based on historical data and turbidity partitioning, an LSTM prediction model is trained to output the turbidity change trend for the next 10 minutes.

[0057] Event-driven controller: The triggering conditions correspond one-to-one with the dedicated rules and are controlled by the corresponding rules, as shown in Table 3. Scene Triggering conditions Special rules Heavy rain and high turbidity Turbidity increases by more than 50% within 10 minutes. Activate high turbidity emergency mode: Dosage = 1.3 × rule base calculation value, will resume after 30 minutes. Low temperatures in winter Temperature <3℃ and turbidity <5 NTU Activate microflocculation enhancement: Dosage ≥ 3.0 mg / L (even with low turbidity) Algal bloom <![CDATA[Chlorophyll a > 30 μg / L or UV 254 > 0.3]]> In conjunction with the recommendation for potassium permanganate pre-oxidation, the PAC dosage should be increased by 15%. Sensor failure Turbidity meter signal loss Using a flow rate + historical average model: Amount applied = Average value of the same time period over the past 7 days.

[0058] Table 3 Storage service unit: TDengine is used as the basic database to store operating condition data.

[0059] Cloud System Optimization Platform Deploy a smart operation and maintenance cloud platform for waterworks (such as...) Figure 3 (As shown), the functions include: Data storage and analysis: Store historical data such as system availability, drug consumption, and exceedance rate, and generate visual reports.

[0060] Remote monitoring and parameter optimization: Engineers can adjust control parameters via the web interface. In one example, the coefficient for the high turbidity emergency mode is 1.3 for 30 minutes, and the parameters are remotely sent to the edge controller.

[0061] System debugging steps Data acquisition and debugging: Verify the accuracy of data transmission from each sensor to the edge controller. In practice, the error between the turbidity meter measurement value and the laboratory manual test value should be ≤5%.

[0062] Model prediction and debugging: Input historical data of traditional PID control. In one example, the average chemical consumption is 8.5 mg / L and the turbidity exceedance rate of 3.2% in the effluent. Train the LSTM model to ensure that the error between the predicted turbidity and the actual value is ≤10%.

[0063] Rule base debugging: In one embodiment, the scenario in Table 3 above is simulated, with heavy rain and high turbidity, to verify the accuracy of the rule triggering logic and the adjustment of the dosage. Under the heavy rain and high turbidity scenario, is the dosage adjusted according to 1.3 × the rule base calculation value?

[0064] Control output debugging: Observe the response speed of the PAC metering pump to be ≤5 seconds to ensure that the dosing command can be executed in real time.

[0065] Example Effect Verification Overall, the above implementation has achieved good results, as shown in Table 4. The average drug consumption decreased from 23.02 mg / L under traditional PID control to 19.51 mg / L, a reduction of 15.2%. The stability of effluent turbidity improved by 11.1%; The recovery time for heavy rain operations was reduced from 45 minutes to 12 minutes, a decrease of 73.3%. Annual pharmaceutical cost savings ≥ 440,000 yuan (500,000 tons / day plant). Average drug consumption (mg / L) 8.5 6.2 ↓27.1% Effluent turbidity stability 0.81 0.72 ↑11.1% Recovery time under heavy rain conditions 45 minutes 12 minutes ↓73.3% Annual pharmaceutical cost savings — ≥440,000 yuan (500,000 tons / day plant) —

[0066] Table 4 The following improvements have been made to the handling of abnormal and special scenarios: In a scenario of heavy rain and high turbidity: when the turbidity increases by more than 50% within 10 minutes, the system automatically activates the high turbidity emergency mode, with a dosage of 1.3 × the rule base calculation value, which lasts for 30 minutes before resuming operation. Actual measurements under heavy rain conditions show that the stability of effluent turbidity is improved by 11.1%.

[0067] In winter low-temperature scenarios: when the temperature is <3℃ and the turbidity is <5 NTU, the system activates micro-flocculation enhancement, and even with low turbidity, the dosage is ≥3.0 mg / L. Actual measurements show that during the winter low-temperature period, the effluent turbidity remains stable below 0.5 NTU.

[0068] Algal bloom scenario: when chlorophyll a > 30 μg / L or UV 254 When the concentration is >0.3, the system is activated in conjunction with the potassium permanganate pre-oxidation recommendation, and the PAC dosage is increased by 15%. Actual measurements show that during algae bloom periods, the effluent turbidity consistently meets standards.

[0069] Sensor failure scenario: When the turbidity meter signal is lost, the system adopts the flow rate + historical average mode, and the dosage is equal to the average value of the same period over the past 7 days. During the period of signal loss, the turbidity fluctuation of the effluent is ≤0.2 NTU.

[0070] The above-disclosed embodiments are merely some preferred embodiments of the present invention and should not be construed as limiting the scope of the invention. Those skilled in the art will understand that implementing all or part of the above-described embodiments and making equivalent changes in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A dual closed-loop intelligent coagulant dosing system for waterworks, characterized in that, include: A multimodal sensing network is used to collect sensor data during the production process. After cleaning and fusion processing, the data is transmitted to the edge controller and stored in a time-series database. The edge intelligent controller includes an independently operating event-driven controller and a composite model predictive controller. The event-driven controller, based on acquired influent and effluent data, determines whether the influent and effluent are abnormal using fault judgment rules. If abnormal, it generates an influent / effluent abnormal event message, locks the composite model predictive controller, and controls coagulant dosing according to its own abnormal control rule set. If the abnormality disappears, it generates an influent / effluent abnormality disappearance message, releases the lock on the composite predictive controller, and stops its own operation. The composite model predictive controller, based on a prediction model segmented by sub-intervals, controls coagulant dosing. When the model reliability drops below a set threshold, it switches to a high-reliability, low-control-economic control mode. The cloud-edge system optimization platform is used for model retraining, energy efficiency analysis, expert knowledge base updates, and the distribution of new models to edge intelligent controllers. The safety execution and alarm unit receives execution and alarm commands from the edge intelligent controller, controls the dosing pump through an independent PLC, and activates the alarm mechanism when an alarm occurs.

2. The intelligent dosing system for dual closed-loop coagulants in water treatment plants as described in claim 1, characterized in that, The event-driven controller includes: The water inlet anomaly detection module analyzes and determines whether the water inlet is abnormal based on the acquired water inlet data. If it is abnormal, it sends a water inlet anomaly event message; if the anomaly disappears, it sends a water inlet anomaly disappearance message. The water outflow anomaly detection module analyzes and determines whether the water outflow is abnormal based on the acquired water outflow data. If it is abnormal, it sends out a water outflow anomaly event message; if the anomaly disappears, it sends out a water inflow anomaly disappearance message. The production process abnormal event communication module is used to communicate with the production system to obtain abnormal event messages or disappearance information during the production process; A closed-loop interlocking module for the composite predictive model controller is used to implement interlocking of the composite predictive model controller; The abnormal operating condition controller is used to call the closed-loop interlocking module of the composite predictive model controller to interlock the composite predictive model controller when it receives a message of an abnormal event, and to control the addition of coagulant according to its own corresponding abnormal event control rule set. When it receives a message of the disappearance of the abnormal event, it calls the closed-loop interlocking module of the composite predictive model controller to release the interlocking of the composite predictive model controller, and at the same time stops its own control operation.

3. The intelligent dosing system for dual closed-loop coagulants in water treatment plants as described in claim 1, characterized in that, The composite model predictive controller includes: The segmented adaptive coagulation dynamic model is configured to be divided into multiple sub-intervals according to the influent turbidity range, and corresponding prediction models are established to reduce the difficulty of modeling a unified interval and improve the overall prediction accuracy of the model. The prediction model credibility index calculation module is configured to calculate the credibility of the prediction model according to the prediction model feasibility index. When the credibility is lower than the set threshold, the control mode will be switched from model prediction control to a control mode with high credibility and low control economy. The model predictive control module is configured to perform model predictive control of coagulant dosing based on a segmented adaptive coagulation dynamic model. The high control confidence and low control economy control module is configured as a non-model conventional control method, which executes control when the confidence of the prediction model is below a threshold.

4. A control method for a dual closed-loop intelligent coagulant dosing system in a waterworks, characterized in that, Includes the following steps: S1: Construct a segmented adaptive coagulation dynamic model; S2: Design an MPC optimizer with multiple constraints; S3: Design feedback correction for the MPC controller; S4: Construct a reliability index for the prediction model; S5: Construct a composite model predictive controller; S6: The control program predicts the threshold setting within the controller according to the composite model and selects the execution control mode based on the confidence level.

5. The control method of the intelligent dosing system for dual closed-loop coagulant in a waterworks as described in claim 4, characterized in that, In step S1, the construction of the segmented adaptive coagulation dynamic model specifically refers to: Divide the influent turbidity into N operating ranges; The following identification model is established independently for each interval: in, Weights corresponding to the impact of disturbance inputs. The corresponding influence weights of the control inputs, The weights of the influence of the corresponding system inertia / historical state.

6. The control method of the intelligent dosing system for dual closed-loop coagulants in a waterworks as described in claim 5, characterized in that, The construction of the piecewise adaptive cohesive dynamic model also includes real-time model updating using online recursive least squares method, with the specific steps as follows: Define the model parameter vector: Constructing the regression vector: Perform online identification and update: Among them, the forgetting factor , initial covariance matrix.

7. The control method of the intelligent dosing system for dual closed-loop coagulant in a waterworks as described in claim 4, characterized in that, In step S2, the MPC optimizer with multiple constraints is designed, and its objective function is: in, The control objective is water quality tracking. The control objective is to inhibit drug consumption. The control objective is to achieve smooth execution.

8. The control method of the intelligent dosing system for dual closed-loop coagulants in a waterworks as described in claim 7, characterized in that, Parameters in the objective function According to the influent water quality stability index adjust, in, This represents the comprehensive fluctuation rate of recent influent water quality parameters, where T is a time constant used to adjust the index's sensitivity to fluctuations. The calculation steps are as follows: Select key influent parameters: including turbidity (d1), pH (d2), and temperature (d3); Calculate the volatility of each parameter: Within a sliding window of length L, calculate the ratio of the standard deviation to the mean for each parameter. Calculate the overall volatility: Take a weighted average of the volatility of each parameter and normalize it so that its range is around [0,1]. in It is the first m The weights of each parameter can be set according to their influence on the coagulation process. C It is a scaling constant used to map the results to a reasonable range and can be calibrated based on historical data.

9. The control method of the intelligent dosing system for dual closed-loop coagulant in a waterworks as described in claim 4, characterized in that, In step S4, the construction of the prediction model credibility index specifically includes the following steps: S41: Residual-based model accuracy index RSD calculation, residual standard deviation normalization index: Among them, the sliding window length , The meaning is the measured turbidity of the water. Meaning: Model predicted value; The residual ratio is obtained after normalization. in, This represents the average measured turbidity over the past 30 minutes. S42: Set the rate of change of model parameters (PCR) to detect whether there are drastic fluctuations in model parameters; If PCR > 0.5, the parameters are unstable and the model is unreliable; S43: Credibility index of prediction model based on RSD and PCR: Among them, the weighting coefficient , If it is greater than 1, then set it to 1. If it is greater than 1, then set it to 1.

10. The control method of the intelligent dosing system for dual closed-loop coagulant in a waterworks as described in claim 4, characterized in that, The control program described in step S6 sets the threshold value within the controller based on the composite model prediction, and selects the execution control mode based on confidence level as follows: When the value is ≥0.8, it is a high-confidence state, and the control strategy is normal MPC optimization control; When the value is [0.7, 0.8), it is a medium confidence state, and the control strategy is to enable mild smoothing; When the value is less than 0.7, it is considered a low-confidence state.

Images