Thermal power plant circulating water intelligent dosing system and method based on multi-module cooperation

Abstract

The invention discloses a thermal power plant circulating water intelligent dosing method based on multi-module cooperation. The system is composed of a detection module, a control module, a chemical adding module, a sewage discharging module, a water supplementing module and an AI intelligent learning feedback module, key parameters such as the pH value, the conductivity, the temperature, the residual chlorine, the turbidity and the flow of circulating water are monitored in real time through a multi-dimensional sensing network, a dynamic water quality prediction model is established in combination with an AI algorithm, and precise closed-loop control over the chemical adding amount is achieved. The system adopts an intelligent decision-making mechanism coupled by a fuzzy PID algorithm and deep reinforcement learning (DRL), can adaptively adjust the dosing proportion of chemicals, and synchronously links a blow-down valve and a make-up pump to maintain the water balance of the system. The system solves the problems of large hysteresis quality and high manual dependence degree of a traditional dosing system, has the characteristics of real-time response, accurate control, energy conservation and consumption reduction, and provides an innovative solution for intelligent transformation of a circulating water system of a thermal power plant.

Description

Technical Field

[0001] The present invention belongs to the technical field of water treatment in thermal power plants, and specifically relates to a circulating water intelligent dosing system based on a detection module, a control module, a dosing module, a sewage discharge module, a water replenishment module and an AI intelligent learning feedback module, which is used to solve the problems of low efficiency, poor precision and reliance on manual experience in traditional dosing methods. Background Art

[0002] The circulating water system of a thermal power plant is crucial for maintaining the safe and efficient operation of its units. Its primary function is to condense turbine exhaust steam, cool equipment, and dissipate waste heat through the circulation of cooling water. The quality of circulating water is directly related to scaling, corrosion, and microbial growth in pipelines and equipment. Improper treatment can significantly reduce heat exchange efficiency, significantly increase energy consumption, and, in severe cases, lead to equipment damage and the risk of downtime. Therefore, as a core means of water quality control, the dosing system must precisely dose scale inhibitors, corrosion inhibitors, biocides, and other chemicals to maintain stable water quality.

[0003] Traditional thermal power plant circulating water dosing systems mostly rely on manual operation or semi-automatic control, and have the following technical bottlenecks: 1. Single detection method: Traditional water quality monitoring relies on manual sampling and laboratory analysis, with strong data lag, making it difficult to reflect the dynamic changes of water quality in real time, and to accurately detect complex ion concentrations (such as Ca2 + Mg2 + , Cl-), and microbial activity and other key indicators lack real-time perception capabilities, resulting in insufficient basis for dosing decisions. 2. Rigid control logic: Existing control systems mostly use fixed thresholds or empirical formulas to adjust the dosage, which cannot adapt to dynamic conditions such as water quality fluctuations, seasonal changes, or load adjustments, and can easily cause waste of agents or substandard inhibition effects. 3. Poor module synergy: Subsystems such as dosing, sewage discharge, and water replenishment operate independently and lack a linkage mechanism. 4. Lack of intelligent optimization capabilities: Traditional systems find it difficult to achieve parameter self-optimization through historical data mining and real-time feedback. When faced with complex multi-variable coupling scenarios (such as agent interactions and ambient temperature changes), their adjustment capabilities are limited, and their long-term operating efficiency gradually decreases.

[0004] In recent years, although some automated dosing systems have achieved a certain degree of closed-loop control through PLC (programmable logic controller) or PID (proportional-integral-differential) algorithms, the following defects still exist: The dosing strategy is limited by manually preset rules and cannot achieve dynamic updates driven by machine learning; at the same time, the sewage discharge and water replenishment modules have not been integrated into the unified control system, resulting in system response delays; and due to the lack of multi-source data fusion technology, it is difficult to effectively predict scaling and corrosion risks.

[0005] With the development of artificial intelligence and Internet of Things technologies, it has become possible to introduce AI intelligent learning feedback mechanisms into the field of circulating water treatment, and to build a multi-module collaborative intelligent dosing system. Through multi-dimensional real-time monitoring of the detection module, dynamic optimization of the control module, precise execution of the dosing / sewage discharge / water replenishment module, and continuous iteration of the AI module, it is possible to break through the limitations of traditional technologies and achieve a leap from "passive response" to "active prediction-optimization" in water quality control. The intelligent dosing system proposed in this patent aims to solve the above-mentioned industry pain points through modular design and algorithm empowerment, and provide technical support for energy conservation, consumption reduction and intelligent upgrading of thermal power plants. Summary of the Invention

[0006] The purpose of this invention is to achieve accurate and stable water quality parameters, economical dosage of chemicals, and energy-saving linkage of sewage discharge and water replenishment through real-time monitoring, intelligent decision-making and multi-module collaborative control, and continuously optimize control strategies based on AI learning to extend equipment service life and reduce operation and maintenance costs.

[0007] To achieve the objectives of this invention, a closed-loop control system consisting of six modules is employed: a detection module, a control module, a dosing module, a sewage discharge module, a water replenishment module, and an AI intelligent learning feedback module. This system is designed to improve operational efficiency while reducing manual intervention to ensure the efficient achievement of the invention's stated objectives.

[0008] Its main working process is as follows: the detection module collects the data of circulating water samples in real time and uploads it to the database in the control module and the AI intelligent learning feedback module. The control module combines the distributed control system (DCS) with the fuzzy PID algorithm to collaboratively generate instructions for dosing, sewage discharge and water replenishment, and transmits these instructions to the dosing module, sewage discharge module and water replenishment module. At the same time, these instructions will also be uploaded to the AI intelligent learning feedback module. According to the instructions received, the dosing module, sewage discharge module and water replenishment module perform precise operations to realize the collaborative work of the three modules, thereby promoting the intelligence and efficiency of the circulating water treatment process. Finally, the AI intelligent learning feedback module analyzes the operating data, optimizes the control algorithm, and feeds back the optimization results to the control module and the detection module to form a closed-loop control.

[0009] Detection module: By installing a pH sensor and a temperature sensor at the circulating water inlet, the pH value and temperature of the circulating water are monitored in real time to avoid affecting the effect of the agent; a turbidity sensor and a residual chlorine sensor are installed at the cooling tower location to respectively detect the accumulation of suspended matter and the microbial disinfection effect of the circulating water in the cooling tower in real time, and a conductivity sensor and a flow sensor are installed at the circulating water outlet to monitor the salinity and temperature changes of the circulating water in real time, thereby reducing the probability of corrosion and scaling. And adopting a redundant design, this system realizes comprehensive and real-time monitoring of the key parameters of the circulating water system (including pH, temperature, turbidity, residual chlorine, conductivity and flow). The monitoring data will be uploaded to the historical database of the control module and the AI intelligent learning feedback module. Among them, the 4-20mA current signal output by each sensor will be transmitted to the control module and the AI intelligent learning feedback module, and aligned with the DCS timestamp to ensure the accuracy and reliability of the data.

[0010] Control module: including the coordinated architecture of distributed control system (DCS) and PID controller, including: Distributed control system (DCS): composed of plant-level monitoring layer, control layer and equipment layer, to achieve global data coordination and strategy optimization; PID controller: deployed in the control layer, using fuzzy PID algorithm to control the execution of dosing module, sewage discharge module and water supply module, its control parameter (K p , K i , K d ) Dynamic optimization of the deep reinforcement learning DRL model in the AI intelligent learning feedback module. When the pH deviation is > 0.5, K is adjusted first. p ; When the conductivity change rate is >10%, adjust K first d ; When the steady-state error lasts for more than 30 minutes, adjust K first i ; When the turbidity mutation is >15%, the K p +K d ; When there is no significant fluctuation, maintain the current parameters.

[0011] Among them, the fuzzy PID algorithm constructs a fuzzy rule base based on the deviation and change rate of water quality parameters. The fuzzy PID processing rules include 1. Rule condition parameterization (the condition part of each fuzzy rule, such as the central value and standard deviation of the parameters in the membership function of "high pH value", "rapid increase in conductivity", "change in turbidity and residual chlorine" are defined as adjustable continuous variables). 2. Rule conclusion parameterization: the rule conclusion, such as the action intensity of "increasing the corrosion inhibitor dosage ratio", is set as an optimizable variable. 3. Dynamic allocation of rule weights: Each rule is given a dynamic weight to characterize its applicability to the current working conditions, and is adjusted in real time by the DRL model. The DRL model real-time adjustment rule base is issued through the distributed system (DCS) of the control module and works in conjunction with the fuzzy PID algorithm, including: high-weight rules are given priority in decision-making, and low-weight rules are gradually eliminated; new rules must be verified by simulation before being integrated into the real-time system; once a rule conflict or control instability is detected, the AI intelligent learning feedback module will automatically start to simplify the rule base (merge similar rules) or restore it to the historical optimal version. Subsequently, these optimization instructions are sent to the dosing module, water replenishment module and sewage discharge module through DCS to achieve efficient operation and automatic adjustment of the system.

[0012] Dosing module: The automated dosing process is implemented using the drug delivery path of drug storage tank → metering pump → dosing point. Three independent drug channels are set up, corresponding to bactericides, scale inhibitors and corrosion inhibitors, to effectively avoid cross contamination. A liquid level sensor is installed in each drug storage tank, a pulse controller is equipped in the metering pump, and a flow meter is installed on the drug delivery path. The three dosing channels use the same device. In addition, a Y-type static mixer is used to ensure that the three drugs are fully mixed with the circulating water. By using independent drug channels and pulse dosing technology, combined with a closed-loop control mechanism of flow feedback, the error in the intelligent automatic dosing process is significantly reduced.

[0013] Sewage discharge module: This system adopts an intermittent automatic sewage discharge method, and automatically controls the opening and closing of the sewage valve according to the water quality parameters of the circulating water and the operating status of the system. If the water quality index seriously exceeds the threshold by 15% and lasts for 30 minutes, the remote manual emergency sewage discharge function can be enabled. This module uses a conductivity sensor to monitor the conductivity of the sewage, controls the conductivity within the range of 2000-4000μS / cm, and achieves precise control of the sewage discharge volume through an electric regulating valve. In addition, the system sets a corresponding relationship between the sewage valve opening and the sewage discharge volume, and adjusts the valve opening degree according to the specific sewage discharge volume. At the same time, the sewage pipe is equipped with a flow meter, which can monitor the sewage discharge volume in real time, thereby ensuring the accuracy and effectiveness of the sewage discharge process. Among them, the sewage discharge volume is calculated as follows:

[0014] Where: Q is the discharge volume (m3 / h), K is the system circulating water volume (m 3 / h), C1 is the conductivity of circulating water (μS / cm), C2 is the target conductivity (μS / cm), and C0 is the conductivity of make-up water (μS / cm).

[0015] Water replenishment module: This system utilizes a multi-source intelligent switching system, with replenishment water sources including softened water, recycled water, and municipal water. Conductivity sensors and liquid level sensors are installed in each water source pipeline to monitor conductivity and liquid level in real time. The control module selects the most economical water source based on real-time water quality data and the target concentration factor (the ratio of the conductivity of the circulating water to the replenishment water). If the conductivity parameter of softened water is low, it is prioritized for high-concentration scenarios; recycled water requires pretreatment to prevent impurities from entering the system; and municipal water serves as a backup source, activated only when other water sources are insufficient. Furthermore, the system utilizes a variable-frequency replenishment pump that adjusts its speed in real time based on the replenishment volume, achieving significant energy savings. If the source water quality exceeds the standard, the system automatically closes the relevant valves and provides pretreatment prompts through the human-machine interface, such as initiating a backwash procedure. Finally, after replenishment is completed, the AI intelligent learning feedback module recalculates the equilibrium concentration of the reagent based on the newly added water volume and dynamically adjusts the dosing rate. The formula for calculating the discharge volume is:

[0016] In the formula: Q is the amount of water replenishment (m 3 / h), Q1 is the water volume predicted by LSTM water quality prediction model (m 3 / h), K1 is the water level deviation weight, K2 is the conductivity error weight, which is adaptively adjusted by the AI module according to historical working conditions to balance the priority of water replenishment and sewage discharge. It is adjusted by the AI intelligent learning feedback module according to historical working conditions. W1 is the target water level (m), W2 is the real-time water level (m), C1 is the circulating water conductivity (uS / cm), C2 is the replenishment water conductivity (uS / cm), Q2 is the sewage discharge (m 3 / h).

[0017] AI intelligent learning feedback module: The data layer utilizes multi-source data fusion and data preprocessing technology. This multi-source data fusion encompasses real-time data (such as water quality parameters, dosage, sewage discharge, and water replenishment), historical databases (including long-term water quality trends and chemical consumption records), and external data (such as ambient temperature and humidity, unit load, and seasonal characteristics). Data preprocessing involves water quality outlier detection, time series alignment, and feature engineering (extracting derived parameters such as concentration factor and corrosion rate prediction). The analysis layer combines short-term prediction models with long-term learning models. The short-term prediction model is a water quality trend prediction (LSTM model), which inputs time series data from the past 24 hours and outputs predicted water quality parameters for the next four hours. The long-term learning model is a deep reinforcement learning (DRL) model. The system automatically adjusts control parameters (Kp, Ki, Kd) every hour and updates network weights using an incremental learning strategy. Safety boundary constraints are also set (hard constraints: pH must not exceed the range of 6.5-9.0 and conductivity threshold ≤4000μS / cm). Ultimately, the data layer provides multidimensional data, which the analysis layer optimizes and analyzes, generating a dynamic control strategy matrix at the decision-making layer. This matrix includes a fuzzy control rule base for dosing and sewage and water replenishment, a coordinated optimization scheme for sewage and water replenishment, and a self-repair strategy for abnormal operating conditions, thereby achieving closed-loop control for the intelligent water dosing system. Furthermore, an emergency intervention strategy is implemented: if the AI module fails, the system automatically switches to a conservative control mode based on fuzzy PID and triggers a manual alarm.

[0018] Among them, the network structure optimization design of the deep reinforcement learning (DRL) model is as follows: 1. The branch network design of multimodal input covers water quality parameters (such as pH value, conductivity, turbidity, residual chlorine, temperature and flow), equipment status (including dosage, sewage valve opening, water replenishment and metering pump frequency), environmental variables (such as ambient temperature, unit load and seasonal characteristics) and historical prediction branches (using LSTM to output water quality prediction values for the next 4 hours). 2. Multivariable coupling modeling integrates the cross-attention mechanism (specifically using the multi-head attention mechanism (Multi-HeadAttention) to calculate the interaction weights between different branch features) and the graph neural network (GNN) assisted modeling method (by constructing a dynamic relationship graph, in which the nodes represent water quality parameters, equipment operations and environmental variables, and the edge weights are generated through self-supervised learning). 3. Safety constraints and real-time guarantees include hard constraints embedded in the output layer: (adding a projection function to the output layer of the Actor network, forcing the PID parameters (K p , K i , K d) and operational instructions comply with safety limits) and safety verification (if the DRL output action causes water quality parameters to exceed hard constraints, the model rollback is immediately triggered). 4. Simulation pre-training (simulating multivariable coupling scenarios in the digital twin platform to pre-train the DRL model) and online incremental learning (fine-tuning network parameters every hour based on the latest data, using the elastic weight consolidation (EWC) algorithm to prevent catastrophic forgetting).

[0019] The LSTM water quality prediction model's input dimensions are: a 24-hour time step (corresponding to the past 24 hours of time series data, with a sampling interval of 0.5 hours). Input features include water quality parameters (pH, conductivity, temperature, residual chlorine concentration, turbidity, and flow rate); equipment status parameters (corrosion inhibitor dosage, scale inhibitor dosage, biocide dosage, wastewater volume, make-up water volume, metering pump frequency, blowdown valve opening, and make-up water pump frequency); external environmental parameters (ambient temperature, unit load, and seasonal characteristics); and derived parameters (concentration ratio and corrosion rate prediction). To ensure model input accuracy, the data in the LSTM water quality prediction model is normalized to eliminate dimensionality differences. Seasonal features are encoded using one-hot encoding (e.g., spring: [1,0,0,0], summer: [0,1,0,0]), ensuring the model can effectively identify cyclical changes. Derived parameters (such as concentration ratio and corrosion rate) are generated through real-time calculation or preprocessing modules, enhancing the model's ability to predict water quality dynamics. During the feature engineering phase, we removed redundant or interfering data (e.g., outliers caused by transient bubbles) to ensure the reliability and temporal consistency of the input data. The data preprocessing and training strategies for the LSTM water quality prediction model are as follows: 1. Normalization: All input features are Z-score normalized to eliminate dimensionality differences; 2. Outlier handling: Transient interference is removed through sliding window mean filtering; 3. Loss function: Mean squared error (MSE), focusing on the prediction accuracy of conductivity and pH. 4. Optimizer: Adam, initial learning rate 0.001, dynamically adjusted (ReduceLROnPlateau); 5. Training data: Time series data from a historical database, split into training and validation sets with an 8:2 ratio. Model validation and deployment of the LSTM water quality prediction model utilize the following: 1. Cross-validation: Time series cross-validation (TimeSeriesSplit) is used to ensure model generalization. 2. Real-time optimization: The model is lightweight (with approximately 500,000 parameters) and inference time is kept within 10 seconds. 3. Safety constraints: The output results pass through the hard constraint layer (such as the pH value is forced to be limited to 6.5-9.0). If the predicted value exceeds the limit, an alarm is triggered and the control switches to conservative PID control.

[0020] In summary, the present invention has the following beneficial effects:

[0021] 1. This system uses pH sensors, temperature sensors, turbidity sensors, residual chlorine sensors, conductivity sensors, and flow sensors for real-time monitoring at different locations. The system is capable of simultaneous detection of multiple water quality parameters, including pH, temperature, turbidity, flow rate, and chloride ion concentration, with a data sampling frequency of seconds. By combining a self-calibration redundancy mechanism with an artificial intelligence learning feedback system, the system's error rate is significantly lower than that of traditional systems, and the probability of single-point failures is effectively reduced. In addition, the data preprocessing function in the AI intelligent learning feedback system can automatically eliminate instantaneous interference signals caused by bubbles and suspended matter, thereby significantly enhancing the data accuracy of the detection module. In addition, the LSTM water quality prediction model can effectively predict water quality, further improving the accuracy of drug delivery and effectively reducing the risk of scaling and corrosion.

[0022] 2. Three independent chemical channels are used for dosing bactericides, scale inhibitors, and corrosion inhibitors, effectively avoiding cross-contamination between chemicals, significantly improving the accuracy of chemical dosing and effectively reducing unnecessary chemical loss. In addition, the use of pulsed dosing effectively avoids the phenomenon of excessive concentration of chemicals in local areas, ensuring that the chemicals are evenly and fully dispersed in the water.

[0023] 3. The combination of intermittent automatic sewage discharge and the AI intelligent learning feedback module successfully realizes intelligent sewage discharge of the circulating water system. This combination effectively prevents water quality deterioration, improves the stability of circulating water quality, and significantly extends the service life of the equipment.

[0024] 4. The multi-source intelligent switching system combines a dynamic water replenishment model with water source optimization. Compared to single-source replenishment, this effectively reduces ineffective replenishment and avoids scaling and corrosion caused by high-salinity or high-hardness water replenishment. This optimization not only reduces equipment maintenance costs but also improves heat exchange efficiency. Furthermore, the system utilizes an AI intelligent learning feedback module to adjust various parameters in real time, adapting to changes in seasons, water sources, and operating conditions, thereby achieving water balance and stable water quality in the circulating water system.

[0025] 5. The control system used in this patent is a collaborative architecture of a distributed control system (DCS) and a PID controller. This system combines the intelligent feedback mechanism in the AI intelligent learning feedback module, which can effectively respond to conditions with sudden changes in water quality and significantly shorten the adjustment response time to less than a few minutes. Compared with traditional PID control, this optimization significantly improves the system's response speed and ability to respond to emergencies, solves the lag problem of fixed threshold adjustment, and enhances the stability and reliability of overall water quality control.

[0026] 6. By combining short-term prediction models with long-term learning models, a short-term LSTM water quality prediction model was used to effectively predict circulating water quality parameters and scaling risks. This approach not only extended equipment cleaning cycles but also significantly reduced labor costs. Furthermore, the long-term learning (RLD) model achieved closed-loop continuous optimization, enabling the system to generate performance reports and update control parameters in real time, significantly reducing annual chemical costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The system is further described below with reference to the accompanying drawings and embodiments.

[0028] Figure 1 This is the overall architecture diagram of the system;

[0029] Figure 2 This is the structure diagram of the detection module;

[0030] Figure 3 This is the control module structure diagram;

[0031] Figure 4 This is the structural diagram of the dosing module;

[0032] Figure 5 This is the structural diagram of the sewage discharge module;

[0033] Figure 6 This is the structural diagram of the water replenishment module;

[0034] Figure 7 This is the structure diagram of the AI intelligent learning feedback module;

[0035] Figure 8 It is the interaction flow chart of fuzzy PID and DRL;

[0036] Figure 9 It is the flow chart of fuzzy PID and DRL collaborative logic decision-making;

[0037] Figure 10 This is the workflow diagram of the dosing module;

[0038] In the figure, 1-flow meter, 2-Y-type static mixer, 3-metering pump, 4-dosing tank, 5-liquid level sensor. DETAILED DESCRIPTION

[0039] The present invention relates to an intelligent dosing system for circulating water in thermal power plants, specifically comprising a detection module, a control module, a dosing module, a sewage discharge module, a water replenishment module, and an AI intelligent learning feedback module. Through the collaborative operation of these multiple modules, the system achieves precise dosing of circulating water chemicals, dynamic water quality regulation, and optimized system operation, effectively improving the operating efficiency and economic efficiency of the circulating water system. The following is a specific implementation method:

[0040] When circulating water flows through the inlet, the pH sensor and temperature sensor in the detection module are embedded, directly contacting the circulating water flow area, and monitoring the pH value and temperature of the circulating water in real time. Subsequently, the turbidity sensor and residual chlorine sensor are installed in the middle section of the vertical pipe at the cooling tower outlet to avoid air bubble interference. At the circulating water outlet, the conductivity sensor and flow sensor are also embedded to monitor the conductivity and flow rate of the circulating water. All detection data is simultaneously uploaded to the control module and AI intelligent learning feedback module for subsequent data processing and analysis.

[0041] When the control module and AI intelligent learning feedback module receive the circulating water pH value, conductivity, residual chlorine, turbidity, temperature and flow rate data uploaded by the detection module, they will determine whether the total dissolved solids concentration and suspended solids content have an impact on the quality of the circulating water, and at the same time evaluate whether the dosage is sufficient, whether the agent is effective, whether there is a tendency for corrosion and scaling, and changes in the chemical reaction rate. The system adopts a collaborative control mechanism of a distributed control system (DCS) and a PID controller to trigger the corresponding logical judgment by comparing the current data with the preset threshold. Specifically, if the conductivity exceeds the set value of 3000uS / cm, the sewage discharge module is started; the circulating water pH range is 6.5-9.0. If the pH value deviates from the target range, the dosage ratio of the acidic or alkaline agent is adjusted accordingly; if the concentration of the added agent is insufficient, the metering pump will be activated for replenishment treatment. This process ensures the efficient and safe operation of the circulating water system.

[0042] The control module's instructions reach the dosing module, which then dispenses chemicals via a pipeline route from the chemical storage tank to the metering pump and then to the dosing point. The dosage is dynamically adjusted based on water quality deviations using a fuzzy PID algorithm. Three independent chemical channels are used to add biocides, scale inhibitors, and corrosion inhibitors to the circulating water. The corrosion inhibitor dosing point is located at the circulating water system inlet, the scale inhibitor dosing point is located at the heat exchanger inlet, and the biocide dosing point is located at the cooling tower inlet. The control module also calculates the wastewater discharge volume based on the circulating water's conductivity and transmits instructions to the wastewater discharge module, which then opens the module's wastewater discharge valve to discharge wastewater. If water quality indicators significantly exceed the set threshold by 15% for 30 minutes, the system will issue an alarm and can initiate a remote manual emergency wastewater discharge strategy. When the control module's instructions are transmitted to the water replenishment module, it intelligently switches between multiple water sources to maintain a balance between system water level and dilution salinity. Compared to traditional water replenishment methods, this approach breaks reliance on a single water source and dynamically selects the optimal combination based on water quality, cost, and operating conditions. The water source switching logic is as follows: when the conductivity of softened water is ≤100uS / cm and the hardness is <50mg / L, the system automatically allocates 70% of the replenishment water to softened water; recycled water must be pre-treated by reverse osmosis (suspended solids <10mg / L) before it is used, with a maximum proportion of no more than 30%; municipal water is only replenished when the water level is below 80%, and the single replenishment volume is ≤20% of the total demand; the replenishment logic stipulates that when the water level is below 85%, the system will replenish water at full speed until it reaches 90%; when the water level is between 85% and 95%, the replenishment rate will be adjusted as needed; and when the water level exceeds 95%, the replenishment will be shut off and an alarm will be issued. In addition, even if the water level is within the normal range, if the conductivity exceeds the standard, water replenishment will still be activated to dilute the salinity. These comprehensive measures effectively ensure the stable operation and safety performance of the circulating water system. When there is a conflict between sewage discharge and water replenishment, the priority judgment rules are adopted: 1. Safety constraints take precedence, and no operation must violate the hard safety boundaries (such as pH 6.5-9.0, conductivity ≤4000μS / cm, and minimum water level threshold). Once sewage discharge causes the water level to approach the lower limit, sewage discharge will be suspended and water replenishment will be prioritized until the water level returns to a safe range. 2. Dynamic weight adjustment. The AI module uses deep reinforcement learning (DRL) technology to dynamically calculate the key weight coefficients K1 and K2 in the water replenishment formula to balance the priority of sewage discharge and water replenishment in real time: if the conductivity seriously exceeds the standard (such as >4000μS / cm), the sewage discharge weight will be increased; if the water level is below the critical value (such as <85%), the water replenishment weight will be increased.3. Time-sharing operation coordination: The control module uses the distributed control system (DCS) for time-sharing scheduling: low water level priority: when the water level is <85%, water replenishment is initiated to 90% first, and then sewage discharge is triggered; high salinity priority: when the conductivity is >4000μS / cm, sewage is discharged to the safety threshold first, and then water replenishment is initiated; parallel operation optimization: if both water replenishment and sewage discharge need to be operated simultaneously, a small flow rate water replenishment and intermittent sewage discharge are adopted in parallel, and the operation volume is accurately adjusted based on the real-time feedback data from the flow meter.

[0043] Finally, the AI intelligent learning feedback module coordinates and optimizes the detection module, control module, dosing module, sewage discharge module, and water replenishment module. Water quality test data, as well as information on dosing, sewage discharge, and water replenishment in the control module, are uploaded to the AI intelligent learning feedback module's historical database. The data in the historical database undergoes data preprocessing in the AI intelligent learning feedback module and is then uploaded to the LSTM water quality prediction model and deep reinforcement learning (DRL) model. The LSTM water quality prediction model outputs water quality trend prediction data, which is then fed back to the detection module and control module for data feasibility verification. Simultaneously, the control strategy of the control module is further optimized based on the water quality prediction data. Furthermore, the DRL model automatically adjusts the detection, control, dosing, sewage discharge, and water replenishment modules every hour to continuously optimize the circulating water treatment effect. Therefore, it is precisely because the five modules, namely the detection module, control module, dosing module, sewage discharge module and water replenishment module, transmit information to the AI intelligent learning feedback module, and the AI intelligent learning feedback module feeds back information to the five modules, that the full closed-loop processing of the intelligent dosing system for circulating water in thermal power plants is achieved, which greatly improves operating efficiency and reduces manual intervention.

Claims

1. An intelligent dosing system for circulating water in a thermal power plant based on multi-module collaboration, comprising the following modules: a detection module: intended to monitor the dynamic water quality parameters of circulating water in real time, including pH value, conductivity, turbidity, temperature, residual chlorine and flow rate; a control module: this module receives data transmitted by the detection module and generates multivariable control instructions based on a preset fuzzy PID algorithm and a distributed control system (DCS); a dosing module: comprising independent dosing units for corrosion inhibitors, scale inhibitors and bactericides, which achieves precise dosing of the agents through a high-precision metering pump according to the instructions of the control module; a sewage discharge ... Block: includes an intelligent sewage valve, a flow monitoring device and a conductivity sensor, and executes the triggered sewage operation according to the instructions of the control module; water replenishment module: through the linkage of the water replenishment pump, conductivity sensor and liquid level sensor, it replenishes the circulating water in real time according to the system water loss and replenishment water conductivity to maintain the water balance of the system; AI intelligent learning feedback module: integrates historical database, data preprocessing, LSTM water quality prediction model and deep reinforcement learning (DRL), and continuously optimizes the algorithm parameters of the control module through comprehensive analysis of water quality data, equipment operating status and external environmental variables.

2. The intelligent dosing system for circulating water in thermal power plants based on multi-module collaboration according to claim 1 is characterized in that: The real-time data of the detection module is synchronously transmitted to the control module and the AI intelligent learning feedback module; the instructions of the control module coordinate the linkage operations of the dosing, sewage discharge, and water replenishment modules through the distributed control system (DCS), and the execution results are fed back to the AI intelligent learning feedback module for strategy optimization; and each module realizes data interaction through 4-20mA signal transmission and aligns with the DCS timestamp to ensure real-time and consistency.

3. The intelligent dosing system for circulating water in thermal power plants based on multi-module collaboration according to claim 1 is characterized in that: The detection module is composed of multi-parameter sensors, including distributed pH sensors, temperature sensors, turbidity sensors, residual chlorine sensors, conductivity sensors and flow sensors; the data acquisition unit adopts redundant communication to ensure the real-time and anti-interference performance of the monitoring data.

4. The intelligent dosing system for circulating water in thermal power plants based on multi-module collaboration according to claim 1 is characterized in that: The control module adopts a collaborative architecture of fuzzy PID algorithm, distributed control system (DCS) and deep reinforcement learning (DRL); wherein the fuzzy PID algorithm is used to adjust the dosage, sewage discharge frequency and water replenishment in real time; the distributed control system (DCS) globally coordinates the operations of dosage, water replenishment and sewage discharge; the deep reinforcement learning model (DRL) is an Actor-Critic network structure based on the PPO algorithm, and the input layer contains three sets of features: water quality parameters, equipment status and environmental variables. The PID parameters (K) are generated every hour based on historical data and real-time feedback. p , K i , K d ) and sends it to each module through DCS to achieve dynamic adjustment of dosing, sewage discharge and water replenishment, thereby improving the overall economy of the system.

5. The intelligent dosing system for circulating water in thermal power plants based on multi-module collaboration according to claim 1 is characterized in that: The independent dosing unit of the dosing module includes: a dosing tank equipped with a liquid level sensor with a liquid level monitoring function; a dosing pipeline is provided with a flow feedback loop; pulse dosing is adopted, combined with the flow of circulating water flow, to reduce the dosing error rate; and three independent drug channels are used, corresponding to the bactericide, scale inhibitor and corrosion inhibitor respectively, to effectively avoid cross contamination of dosing.

6. The intelligent dosing system for circulating water in thermal power plants based on multi-module collaboration according to claim 1 is characterized in that: The sewage discharge module adopts an intermittent automatic sewage discharge method and supports the following modes: automatic sewage discharge based on the sewage discharge conductivity threshold; preventive sewage discharge combined with the scaling risk predicted by the AI intelligent learning feedback module; after exceeding the sewage discharge conductivity threshold, remote manual control and emergency sewage discharge function are activated.

7. The intelligent dosing system for circulating water in thermal power plants based on multi-module collaboration according to claim 1 is characterized in that: The water replenishment module supports the following modes: dynamic water replenishment through coordinated control of conductivity sensors, liquid level sensors, flow meters and water replenishment pumps to maintain system water level and salt balance; intelligent switching of multiple water sources, access to three water replenishment sources, and generation of water source ratio plans based on water quality data, target concentration multiples and cost factors; linkage control of water replenishment and sewage discharge to calculate the optimal water replenishment volume.

8. The intelligent dosing system for circulating water in thermal power plants based on multi-module collaboration according to claim 1 is characterized in that: The AI intelligent learning feedback module includes: data preprocessing, which processes water quality information conveyed by the detection module, removes invalid information, and reduces the possibility of ineffective regulation; a historical database for storing water quality parameters, equipment operation logs, environmental data, and fault diagnosis logs; an LSTM-based water quality prediction model for warning of water quality deterioration trends and generating multi-dimensional energy efficiency analysis reports; and dynamic optimization, which uses a reinforcement learning model (DRL) to adjust the weight parameters of the control module to bring the system closer to the optimal balance between energy consumption and water quality. A safety layer constrains output actions to ensure that control instructions comply with the preset pH range (6.5-9.0) and conductivity threshold (≤4000μS / cm).

9. A control method for a thermal power plant circulating water intelligent dosing system based on multi-module collaboration, characterized in that: The following steps are involved: The detection module collects various parameters of the circulating water quality in real time; the control module combines current data with the historical database, applies the fuzzy PID algorithm to generate initial control instructions, and passes the instructions to the dosing module, sewage discharge module and water replenishment module; the AI intelligent learning feedback module triggers the deep reinforcement learning (DRL) model every hour, and generates optimization instructions for the control parameters based on the water quality data and equipment operating status of the past 24 hours; the dosing, sewage discharge and water replenishment operations are performed simultaneously, and the entire closed-loop processing is achieved through the AI intelligent learning feedback module.

Images