Wet dust removal circulating water purification cooperative control method and system based on dust removal load feedforward compensation

Abstract

The invention discloses a wet dust removal circulating water purification cooperative control method and system based on dust removal load feed-forward compensation. The method comprises the following steps: constructing a multi-source heterogeneous data and dynamic space-time alignment database; establishing a deep learning model, and predicting flux characteristics of future pollutants according to the working condition data of the dust removal system; constructing a hydrodynamic model, and calculating the dynamic time lag of pollutant transportation; and taking the predicted flux characteristics and the dynamic time lag as feed-forward signals, performing cooperative control on agent addition and equipment of the circulating water purification system, and feeding back closed-loop correction based on the effluent quality. Through prospective prediction of the dust removal load and accurate calculation of pollutant conveying time lag, predictive control of the purification system is realized, the problem of response lag of traditional feedback control is solved, the load impact resistance and control precision of the system are improved, the medicament ratio can be optimized, and medicament consumption can be reduced.

Description

Technical Field

[0001] This application relates to the field of industrial automation control and environmental protection equipment technology, specifically to a method and system for coordinated control of wet dust removal circulating water purification based on dust removal load feedforward compensation. Background Technology

[0002] In heavy industrial production processes such as metallurgy, power generation, and chemicals, wet scrubbing systems are key environmental protection equipment for treating dust-laden waste gas. This system captures particulate matter from the waste gas through a circulating water spray scrubbing process, and the resulting dust-laden wastewater (turbid circulating water) is transported to a circulating water purification system. After undergoing a series of physicochemical treatments, such as sedimentation, coagulation, and filtration, the water quality is restored to the process requirements before being recycled back to the dust collection system.

[0003] In existing technologies, wet scrubbing systems and circulating water purification systems are typically designed as two independent control units. The scrubbing system focuses on ensuring dust removal efficiency and safe operation, while the circulating water purification system focuses on achieving compliant end-of-pipe water quality. This separate control model exposes several technical problems when dealing with the dynamic changes in modern industrial production. First, there is a close physical coupling and significant signal transmission lag between the two systems. Industrial production (such as the blowing cycle of converter steelmaking) has strong periodic and intermittent characteristics, causing the dust removal load to fluctuate drastically in a short period of time. High-concentration dust-laden wastewater, passing through long-distance pipe networks and large-volume treatment tanks, requires tens of minutes or even hours for its water quality characteristics to be transmitted to key control nodes of the water treatment system, such as chemical dosing points.

[0004] Secondly, traditional control strategies are ill-suited to effectively handle such large-lag, high-impact load changes. Currently, automatic chemical dosing control in circulating water purification systems largely relies on feedback control loops based on influent or effluent turbidity detection, such as proportional-integral-derivative (PID) control. The inherent drawback of this control method is its lag in response. When the turbidity sensor detects water quality deterioration, it means that a high-concentration polluted water mass has already entered or passed through the treatment unit. Adjusting the chemical dosage at this point, due to the lag in response, is insufficient to effectively treat this high-concentration polluted water mass, potentially leading to effluent quality exceeding standards or excessive chemical dosing.

[0005] Furthermore, existing technologies lack the means to proactively perceive and finely control the physicochemical properties of pollutants. Changes in dust removal load are not only reflected in changes in suspended solids concentration, but also in dynamic changes in the physicochemical properties of particulate matter, such as particle size distribution, density, and surface charge. These properties directly affect the effectiveness of subsequent coagulation and sedimentation, as well as the type and ratio of reagents required. Traditional control systems cannot bridge the gas, liquid, and solid multiphase media to predict subtle characteristics of end-point water quality from source production data, thus failing to achieve optimized dosing of water treatment reagents and coordinated scheduling of equipment. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for coordinated control of wet dust removal and circulating water purification based on dust removal load feedforward compensation, so as to solve the technical problems of response lag, insufficient resistance to load impact and low control accuracy caused by the separate control of wet dust removal and circulating water purification systems in the background art.

[0007] In one aspect, the present invention provides a method for coordinated control of circulating water purification in wet dust removal based on dust removal load feedforward compensation, the method comprising the following steps:

[0008] Step 1 involves constructing a multi-source heterogeneous data acquisition and dynamic spatiotemporal alignment database. This step includes: collecting dust removal-side characteristic data representing the operating status of the dust removal system, production rhythm signals representing the generation patterns of pollution sources, transmission-side characteristic data representing the pollutant transport process, and purification-side characteristic data representing the operating effect of the purification system; and using dynamic sample alignment technology based on reverse virtual tracing, calculating the actual physical transmission lag time of the purification-side water quality data samples in the pipeline network, and combining the purification-side water quality data samples with the dust removal-side characteristic data and the production rhythm signals at the traced-back source time into a spatiotemporally precisely aligned sample pair to construct a training dataset.

[0009] Step 2: Using the training dataset, train a deep learning-based dust load feature prediction model to predict the pollutant flux characteristics that will arrive at the dosing point in the circulating water purification system in the future, based on real-time dust removal side feature data and production rhythm signals.

[0010] Step 3: Construct a pollutant transport time delay calculation model based on fluid mechanics to calculate the dynamic transport time of pollutants from the dust collector to the dosing point in the circulating water purification system in real time.

[0011] Step 4: Based on the pollutant flux characteristics predicted by the dust removal load characteristic prediction model and the dynamic transmission time calculated by the pollutant transport time delay calculation model, feedforward compensation control is implemented on the circulating water purification system.

[0012] Step 5: Based on the deviation between the effluent water quality index of the circulating water purification system and the control target value, perform feedback correction on the dust removal load characteristic prediction model.

[0013] Furthermore, the specific steps for constructing the training dataset using the dynamic sample alignment technology based on reverse virtual tracing in step one are as follows: For each water quality data sample on the purification side in the historical database, using the pipeline hydraulic model, input the historical pipeline instantaneous flow velocity data sequence before the sampling time of the water quality data sample on the purification side, and calculate the actual physical transmission lag time experienced by the water mass corresponding to the water quality data sample on the purification side in the pipeline through reverse integration calculation; and use the actual physical transmission lag time to combine the water quality data sample on the purification side with the dust removal side feature data and the production rhythm signal at the source time back to the previous step into a spatiotemporally precisely aligned sample pair.

[0014] Furthermore, the dust load characteristic prediction model is a hybrid neural network that combines a long short-term memory network with an attention mechanism.

[0015] Furthermore, the pollutant flux characteristics include the total mass concentration of suspended solids characterizing the load, and the equivalent particle size distribution index characterizing the load quality; wherein, the equivalent particle size distribution index is a comprehensive indicator used to quantify the settling performance of suspended particulate matter in wastewater, obtained based on the mapping relationship between source production operating parameters and wastewater physical settling characteristics.

[0016] Furthermore, the pollutant transport time delay calculation model based on fluid mechanics in step three is implemented using the virtual tracer particle method. The virtual tracer particle method is implemented as follows: when the rate of change of the pollutant flux characteristics predicted by the dust removal load characteristic prediction model exceeds a preset threshold, a virtual tracer particle carrying a timestamp and pollutant attributes is generated at the wastewater outlet of the dust collector; the instantaneous flow velocity of the return water network is collected in real time, and the displacement of the virtual tracer particle is integrally extrapolated in real time. Its cumulative displacement is calculated by integrating the instantaneous flow velocity v(τ) from the generation time T0 to the current time t; when the cumulative displacement S(t) of the virtual tracer particle is equal to the physical length of the pipeline from the dust collector outlet to the dosing point, it is determined that the virtual tracer particle has reached the dosing point, thus obtaining the dynamic transport time.

[0017] Furthermore, the feedforward compensation control implemented in step four includes feedforward dosing correction control. Specifically, the feedforward dosing correction control involves adjusting the output of the dosing metering pump at a preset lead time before the virtual tracer particles are expected to arrive at the dosing point; and implementing a dynamic proportioning strategy. When the predicted equivalent particle size distribution index is higher than a first preset threshold, it indicates that the wastewater is mainly composed of easily settling coarse particles, so the dosage of coagulant is increased; when the predicted equivalent particle size distribution index is lower than a second preset threshold, it indicates that the wastewater is mainly composed of fine particles, so the dosage of coagulant aid is increased while increasing the amount of coagulant.

[0018] Furthermore, the feedforward compensation control implemented in step four also includes load stabilization and sludge discharge linkage control; the load stabilization and sludge discharge linkage control includes at least one of the following control methods: when the predicted peak value of the total suspended solids concentration exceeds the maximum treatment capacity of the circulating water purification system under the current operating conditions, the switching valve in the pipeline network is controlled in advance to temporarily divert the high-turbidity wastewater to the emergency water tank; based on the predicted total amount of suspended solids in the influent, the amount of sludge that will increase in the sedimentation tank in the future preset time period is calculated integrally, and before the calculated sludge level reaches the high-level alarm threshold, the sludge discharge pump is started in advance or the sludge discharge valve is opened for preventive sludge discharge.

[0019] Furthermore, the specific method for performing feedback correction in step five is as follows: when the deviation between the turbidity of the sedimentation tank effluent monitored in real time and the control target value exceeds a preset duration, an online learning or offline update mechanism is triggered, and the weight parameters of the dust removal load characteristic prediction model are fine-tuned or retrained using the latest collected data that has undergone spatiotemporal alignment processing.

[0020] Another aspect of the present invention provides a wet dust removal circulating water purification collaborative control system based on dust removal load feedforward compensation, characterized in that it includes:

[0021] The data construction module is configured to build a multi-source heterogeneous data acquisition and dynamic spatiotemporal alignment database, including: collecting dust removal side feature data characterizing the operating status of the dust removal system, production rhythm signals characterizing the generation pattern of pollution sources, transmission side feature data characterizing the pollutant transport process, and purification side feature data characterizing the operating effect of the purification system; and using dynamic sample alignment technology based on reverse virtual tracing, calculating the actual physical transmission lag time of the purification side water quality data sample in the pipeline network, and combining the purification side water quality data sample with the dust removal side feature data and the production rhythm signal at the source time traced back to form a spatiotemporally precisely aligned sample pair to construct a training dataset;

[0022] The model training module is configured to use the training dataset to train a deep learning-based dust load feature prediction model, which is used to predict the pollutant flux characteristics that will arrive at the dosing point in the circulating water purification system in the future based on real-time dust removal side feature data and production rhythm signals.

[0023] The time delay calculation module is configured to build a pollutant transport time delay calculation model based on fluid mechanics, which is used to calculate the dynamic transmission time of pollutants from the dust collector to the dosing point in the circulating water purification system in real time.

[0024] The collaborative control module is configured to implement feedforward compensation control on the circulating water purification system based on the pollutant flux characteristics predicted by the dust removal load characteristic prediction model and the dynamic transmission time calculated by the pollutant transport time delay calculation model.

[0025] The feedback correction module is configured to perform feedback correction on the dust removal load characteristic prediction model based on the deviation between the effluent water quality index of the circulating water purification system and the control target value.

[0026] By adopting the above technical solutions, this invention can bring beneficial technical effects. First, by introducing dynamic sample alignment technology based on reverse virtual tracing, the correspondence between pollution sources and end-point water quality under long-distance, variable-flow transportation conditions is accurately restored, eliminating spatiotemporal noise in training samples and providing a solid foundation for building a high-precision prediction model. Second, by utilizing a dynamic time-delay calculation model based on real-time flow velocity integral, the arrival time of high-turbidity water masses is accurately predicted, enabling precise early intervention of control actions and fundamentally solving the response lag problem of traditional feedback control in systems with large time delays. Finally, by establishing a soft measurement model of the particle size distribution index of wastewater particulate matter, refined control based on the perception of pollutant physical sedimentation characteristics is achieved. This allows for dynamic optimization of reagent ratios based on the dust characteristics generated at different production stages, effectively reducing reagent consumption while ensuring effluent quality. Attached Figure Description

[0027] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0028] Figure 1 This is a schematic flowchart of a wet dust removal circulating water purification collaborative control method based on dust removal load feedforward compensation provided in an embodiment of the present invention.

[0029] Figure 2 This is a schematic diagram of the architecture of a wet dust removal circulating water purification collaborative control system based on dust removal load feedforward compensation provided in an embodiment of the present invention.

[0030] Figure 3This is a schematic diagram illustrating the principle of dynamic sample alignment technology based on reverse virtual tracing in an embodiment of the present invention.

[0031] Figure 4 This is a schematic diagram illustrating the principle of the pollutant transport time delay calculation model based on the virtual tracer particle method in an embodiment of the present invention.

[0032] Figure 5 This is a physical layout diagram of the wet dust collector and circulating water purification system of the present invention.

[0033] Figure 6 This is a schematic diagram of the time-series superposition of multi-source asynchronous pollutants according to the present invention.

[0034] Figure 7 This is a comparison chart of the effects of the feedforward control of this invention and traditional feedback control. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this invention clearer, the specific embodiments of this invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0036] Example 1

[0037] This embodiment provides a method for coordinated control of wet dust removal circulating water purification based on dust removal load feedforward compensation, referring to... Figure 1 As shown, this method is implemented in a typical converter steelmaking wet dust removal and water treatment system. Its core objective is to overcome the inherent physical coupling and signal transmission lag of the system, and to achieve predictive and refined control of the water treatment system. The specific steps of this method are as follows:

[0038] Step S100: Construct a multi-source heterogeneous data acquisition and dynamic spatiotemporal alignment database.

[0039] To establish a mechanism model that accurately reflects the mapping relationship across the entire process from production conditions and dust removal load to circulating water quality, it is first necessary to collect and process multi-source heterogeneous data across the system. Specifically, the types of data collected include: dust removal side characteristic data, such as flue gas flow rate at the dust collector inlet measured by a differential pressure transmitter, flue gas temperature measured by a thermocouple, pressure difference across the venturi tube of the dust collector, valve position signal reflecting the opening of the venturi tube throat, and circulating water supply flow rate measured by an electromagnetic flowmeter; these data characterize the real-time operating status of the dust removal system; and production rhythm signals, such as Boolean logic signals representing blowing and standby states obtained from the converter main control programmable logic controller (PLC), height signals from oxygen lance position sensors, and mass flow meter signals from the oxygen pipeline network. These signals provide prior knowledge about the patterns of pollution source generation; transmission-side characteristic data, such as the instantaneous flow velocity measured by ultrasonic flow meters installed on the dust removal wastewater return pipeline and the ultrasonic level gauge signals in the return water open channel, are used to track the transport process of pollutants in the pipeline; and purification-side characteristic data, such as the actual dosage of coagulant and coagulant aid converted from the frequency feedback signal of the dosing pump, the online turbidity meter readings installed at the inlet and outlet of the sedimentation tank, and the sludge concentration measured by the sludge interface meter at the bottom of the sedimentation tank, serve as labels and feedback verification signals for subsequent model training.

[0040] Furthermore, to address the time misalignment between the "dust generation time at the source" and the "water quality change time at the end point" caused by the long-distance transport of dust removal wastewater through pipelines, this embodiment employs a dynamic sample alignment technique based on reverse virtual tracing to construct a high-fidelity training dataset. (Refer to...) Figure 3 This technology targets each purified water quality data sample recorded in a historical database, such as the turbidity value Y(t1) at the sedimentation tank inlet measured at time t1. Using a hydraulic model based on pipeline geometry and fluid characteristics, and inputting the historical instantaneous flow velocity data sequence v(t) from before time t1, it accurately calculates the actual transport time Δt of the water mass corresponding to that water quality sample in the pipeline network through inverse integration. This hydraulic model can be a one-dimensional pipeline flow model, comprehensively considering factors such as pipeline length, cross-sectional area, and wall roughness. The inverse integration process is equivalent to solving the equations. Where L is the total length of the pipeline network, the precise physical transmission lag time Δt is obtained. Using this dynamically calculated lag time, the water quality data Y(t1) at time t1 is combined with the precisely backtracked source dust removal operation data X(t1-Δt) at time t1-Δt to form a spatiotemporally aligned sample pair (X(t1-Δt), Y(t1)). This method effectively eliminates temporal misalignments caused by fluctuations in production load (such as blowing intervals) and circulating water volume (such as pump start-stop), providing a high-quality data foundation for subsequent deep learning model training. It is significantly superior to traditional methods using fixed lag times or simple statistical cross-correlation analysis.

[0041] Step S200: Establish a dust load characteristic prediction model based on deep learning.

[0042] Using the dynamic spatiotemporal aligned database constructed in step S100, a deep learning model is trained. This model can predict the pollutant flux characteristics of wastewater arriving at the circulating water purification system based on real-time dust removal system operating data. Preferably, the model's network architecture employs a hybrid neural network combining a long short-term memory network and an attention mechanism. The long short-term memory network, through its unique gating structure, can effectively learn and memorize long-range temporal dependencies within industrial production cycles (e.g., a complete blowing cycle may last 20-30 minutes). The attention mechanism automatically learns and assigns higher weights to key process signals that have the greatest impact on the prediction results when the model processes the input time-series data. For example, when a signal indicating the start of converter blowing is detected, the model will focus on drastic changes in strongly correlated features such as flue gas flow rate and flue gas temperature, thereby improving the sensitivity and accuracy of the prediction. The model's input is a time-series data matrix of dust removal side feature data and production rhythm signals from the current moment and a past period (e.g., 30 minutes), and the output is the predicted pollutant flux characteristics at the future time of arrival at the water treatment dosing point.

[0043] This embodiment utilizes a deep learning model to achieve dual soft measurement of both the "quality" and "quantity" dimensions of dust removal load. Specifically, the pollutant flux characteristics include not only the total suspended solids concentration (mg / L), representing the "quantity" of the load, but also the equivalent particle size distribution index, representing the "quality" of the load. This equivalent particle size distribution index is a dimensionless comprehensive indicator used to quantify the settling performance of suspended particulate matter in wastewater. For example, the index can be defined as a scalar between 0 and 1; a value closer to 1 indicates that the wastewater is dominated by large particles that settle quickly, while a value lower indicates that it is dominated by fine particles that are difficult to settle naturally. By learning from a large amount of spatiotemporally aligned sample data, this model can establish a direct mapping relationship from source smelting process parameters (such as the ignition stage in the initial blowing stage, where the oxygen lance height is low, easily producing large iron oxide scale particles, corresponding to a high equivalent particle size distribution index) to the physical settling characteristics of wastewater, overcoming the technical bottlenecks of traditional online particle size analyzers, such as high cost, difficult maintenance, and delayed response.

[0044] Step S300: Construct a time delay calculation model for pollutant transport based on fluid mechanics.

[0045] Because the flow rate of the circulating water system changes dynamically during actual operation, the transport time of pollutants from the dust collector discharge to the dosing point is not a fixed value. Therefore, this step constructs a dynamic time-delay calculation model to calculate the estimated arrival time of high-concentration pollutant water masses in real time and with high accuracy. (Refer to...) Figure 4 In this embodiment, the dynamic time delay calculation is achieved through a virtual tracer particle method. Specifically, in the digital twin environment of the system, whenever the prediction model output in step S200 shows that the pollutant load will undergo a significant change (for example, the predicted total suspended solids concentration will increase by more than 50% within 5 minutes), a virtual water mass, or virtual tracer particle, with a timestamp and pollutant attributes (such as the predicted total suspended solids concentration and equivalent particle size distribution index) is generated at the wastewater outlet of the dust collector at the software level.

[0046] Subsequently, the system collects the instantaneous flow velocity v(t) along the return water pipeline in real time and performs real-time integral deduction of the displacement of the virtual water mass. The displacement calculation formula can be expressed as: Where S(t) is the cumulative displacement from generation time T0 to the current time t, and v(τ) is the instantaneous flow velocity at time τ, which is provided in real time by the flow meter in the pipeline network. In each control cycle (e.g., 1 second), the system updates the cumulative displacement of all active virtual water masses. When the cumulative displacement S(t) of a virtual water mass calculated by the time delay calculation module is equal to the physical length L of the pipeline network from the dust collector outlet to the dosing point (this length is a preset parameter), the system determines that the virtual water mass has arrived at the dosing point. This model can update the remaining transmission time of each virtual water mass in real time. If the pipeline flow velocity increases due to heavy rainfall or variable frequency operation of the water pump, the model will automatically correct and shorten the estimated arrival time, thereby ensuring accurate synchronization between the issuance of subsequent control commands and the actual arrival time of the polluted water mass.

[0047] Step S400: Implement coordinated control of drug dosing and equipment based on feedforward compensation.

[0048] The pollutant flux characteristics predicted in step S200 and the precise arrival time calculated in step S300 are used as core feedforward signals and input into the collaborative controller to achieve predictive control of the circulating water purification system. Specific control strategies include:

[0049] Firstly, feedforward dosing correction. At a specific lead time (e.g., 60 seconds, set according to the kinetics of chemical dissolution and mixing) before the virtual water mass is expected to reach the dosing point, the co-controller automatically adjusts the output frequency or stroke of the metering pump. Furthermore, the controller implements a dynamic proportioning strategy. An internal chemical proportioning function or lookup table based on the equivalent particle size distribution index can be preset within the controller. For example, when the predicted equivalent particle size distribution index is greater than 0.8, indicating that the wastewater mainly consists of easily settling coarse particles (such as iron oxide scale from the initial stage of converter blowing), the system will primarily increase the dosage of coagulant (such as polyaluminum chloride) to enhance charge neutralization and double-layer compression, while maintaining a low proportion of coagulant aids (such as polyacrylamide). Conversely, when the prediction index is below 0.4, indicating that fine particles (such as submicron-sized dust generated during the decarbonization period of the converter) dominate the wastewater, the system will significantly increase the dosage of coagulant aid while increasing the coagulant, in order to enhance the adsorption bridging and net sweeping effects and effectively capture fine particles that are difficult to settle.

[0050] Secondly, load stabilization and sludge discharge are linked. When the peak suspended solids concentration output by the predictive model exceeds the maximum treatment capacity of the purification system under current operating conditions (e.g., the predicted total suspended solids concentration exceeds 6000 mg / L, while the system's design treatment limit is 5000 mg / L), the co-controller can issue an instruction in advance to control the electrically operated switching valves in the pipeline network, temporarily diverting this portion of high-turbidity wastewater to the emergency water tank for buffering, thereby buffering and stabilizing the load peak and avoiding impact on the main treatment process. Simultaneously, the controller can calculate the amount of sludge that will increase in the sedimentation tank over a future period (e.g., 2 hours) based on the predicted total suspended solids in the influent, and convert this into an increase in sludge level. If the predicted sludge level exceeds the high-level alarm threshold, the controller will start the sludge discharge pump or open the sludge discharge valve in advance for preventative sludge discharge, preventing sludge overturning due to excessive sludge accumulation and ensuring stable effluent quality.

[0051] Step S500: Perform closed-loop correction based on effluent water quality feedback.

[0052] To ensure the long-term stability and adaptability of the control system, this method also includes a feedback correction closed loop. The system monitors key water quality indicators such as turbidity of the sedimentation tank effluent in real time. When the actual effluent quality deviates continuously from the system's control target value, for example, if the average effluent turbidity is 20% higher than the set value for 24 consecutive hours, it indicates that the system characteristics may have drifted (e.g., due to equipment wear, pipe scaling, or seasonal water temperature changes). At this time, the system will trigger an online learning or offline update mechanism, using the latest collected and aligned operational data to fine-tune or retrain the weight parameters of the deep learning model in step S220. Specifically, the system can mark newly collected and spatiotemporally aligned data samples from the past week as a new training set, and start incremental training of the model during periods of idle system computing power (such as at night), thereby continuously improving the model's prediction accuracy and achieving adaptive correction to system changes.

[0053] like Figure 7As shown in the figure, this invention compares and analyzes the control effects of the feedforward compensation control method and the traditional PID feedback control method under load shock conditions. The horizontal axis represents time, the vertical axis represents effluent turbidity, and the horizontal dashed line represents the control target value of 1.0 NTU. A load shock occurs at t=30 min, and the two control methods show significant differences. With traditional PID feedback control, due to a detection and response lag of approximately 5 minutes, the controller only begins to adjust after the load shock occurs, causing the effluent turbidity to rise rapidly and exceed the control target value, reaching a peak of 1.80 NTU, with the exceedance lasting approximately 45 minutes. The shaded area in the figure clearly shows the exceedance range of traditional control. In contrast, with the feedforward compensation control method of this invention, through real-time monitoring and predictive analysis of the influent water quality parameters, the controller can sense the load change trend approximately 3 minutes in advance and pre-adjust the dosage, ensuring that the effluent turbidity remains below the control target value throughout the entire load shock process, with a peak value of only 0.97 NTU. The comparative results show that, compared with the traditional PID feedback control, the feedforward compensation control method of the present invention reduces the overshoot by about 46% and shortens the response time by about 85%, effectively avoiding the problem of excessive effluent under load shock conditions, and significantly improving the control accuracy and stability of the water treatment system.

[0054] Example 2

[0055] This embodiment provides a wet dust removal circulating water purification collaborative control system based on dust removal load feedforward compensation. This system is used to implement the method described in Embodiment 1. (Refer to...) Figure 2 As shown, the system can logically include: a data perception layer 10, an edge computing gateway 20, an intelligent collaborative control platform 30, and an actuator layer 40.

[0056] Data Sensing Layer 10 is responsible for collecting field data, including but not limited to ultrasonic flow sensors, thermocouple temperature sensors, and differential pressure sensors deployed in dust removal systems and circulating water networks, as well as communication interfaces for accessing production main control systems such as converters or blast furnaces via industrial Ethernet (such as Profinet or Modbus TCP / IP protocols). This layer provides a real-time and accurate data foundation for the entire collaborative control system.

[0057] Edge computing gateway 20, deployed in a control cabinet near the data source, is responsible for the initial cleaning, aggregation, and caching of the massive amounts of raw data collected. Preferably, this gateway runs the dynamic sample alignment algorithm based on reverse virtual tracing described in step S100 of embodiment 1 to complete the dynamic spatiotemporal alignment preprocessing of the data. Deploying this computationally intensive task at the edge can effectively utilize high-frequency real-time streaming data, reduce the computational burden on the cloud platform, and decrease dependence on network bandwidth.

[0058] The intelligent collaborative control platform 30 is the decision-making core of the system and is typically deployed on an industrial control computer or cloud server in the central control room. This platform integrates multiple functional modules, including: a load prediction module 31, which runs the aforementioned long short-term memory network-attention mechanism deep learning model and outputs real-time predictions of future pollutant flux characteristics (total suspended solids concentration and equivalent particle size distribution index), corresponding to step S200 in Example 1; a time delay calculation module 32, which runs a pollutant transport model based on fluid dynamics to perform trajectory extrapolation and dynamic arrival time calculation for virtual tracer particles, corresponding to step S300 in Example 1; and a decision control module 33, which receives the outputs from the prediction module 31 and the time delay calculation module 32, and generates specific control commands for operations such as chemical dosing, sludge discharge, and diversion based on preset control logic and dynamic proportioning strategies, and includes feedback correction logic, corresponding to steps S400 and S500 in Example 1.

[0059] The actuator layer 40 is responsible for receiving and executing instructions from the intelligent collaborative control platform 30. This layer includes variable frequency metering pumps connected to the dosing system, electric sludge discharge valves or pumps at the bottom of the sedimentation tank, and electric gate valves for diversion in the pipeline network, among other field automation equipment. These devices communicate with the intelligent collaborative control platform through the I / O modules of a PLC or Distributed Control System (DCS), translating the platform's decision instructions into precise actions in the physical world.

[0060] like Figure 5 As shown, the physical layout of the wet scrubber and circulating water purification system includes the following equipment and connections: The wet scrubber 1 adopts a venturi tube structure design, with a dust-laden gas inlet at the top and a wastewater outlet 2 at the bottom. Wastewater outlet 2 is transported over a long distance via a return water network 3. The return water network 3 has switching valves at the diversion nodes, allowing wastewater to be diverted to an emergency water tank 4 or directly into a sedimentation tank 5 depending on the system's operating conditions. The emergency water tank 4 serves as a diversion buffer facility, used to temporarily store wastewater during abnormal system conditions or maintenance. The sedimentation tank 5 has an inlet and an outlet; the inlet is located on one side of the tank, and the outlet is on the opposite side. Suspended solids and sludge are separated within the sedimentation tank 5. A dosing point 6 is located at the front end of the sedimentation tank 5's inlet. A dosing system 7 is located above or near the sedimentation tank 5, delivering flocculants or pH adjusters to the dosing point 6 via a dosing pipeline. A sludge removal device 8 is connected to the bottom of the sedimentation tank 5 for periodically discharging deposited sludge. The outlet of sedimentation tank 5 is connected to circulating water pump station 9 via a switching valve. Circulating water pump station 9 is equipped with a circulating water pump, which transports the purified water back to wet scrubber 1 through water supply network 10, forming a complete circulating water loop. In the diagram, solid lines indicate the flow direction of wastewater pipes, dashed lines indicate the flow direction of water supply pipes, arrows indicate the direction of water flow, and butterfly valve symbols indicate the position of the switching valve.

[0061] Example 3

[0062] This embodiment aims to illustrate the specific application of the method of the present invention in a complex industrial scenario with multiple asynchronous pollution sources, so as to demonstrate the necessity of adopting the method of this application. The scenario is a large-scale integrated production facility, which contains a No. 1 production unit and a No. 2 production unit. Both units use independent wet scrubbers, but the dust-laden wastewater generated after treatment flows into the same main pipeline network and is treated by a centralized circulating water purification system.

[0063] In this specific application scenario, there is a technical challenge that existing technologies struggle to address. Production unit one's process is characterized by a short cycle and high intensity, resulting in wastewater with high peak suspended solids concentrations but relatively large and dense particulate matter, making it relatively easy to settle. Production unit two, on the other hand, has a longer cycle and a more gradual load, but its wastewater is primarily composed of fine, low-density particles with poor settling properties. Because the production plans of the two units are independent and their start-up and shutdown times are completely asynchronous, the total wastewater entering the centralized purification system exhibits highly complex, non-periodic, random fluctuations in both suspended solids concentration and particulate matter distribution index. Traditional single-loop control systems based on inlet water quality feedback are ineffective because they cannot decouple and predict the characteristics of pollutants from different sources. This often leads to mismatches between reagent dosage and actual water quality characteristics, resulting in unstable effluent quality or reagent waste.

[0064] like Figure 6 As shown, the industrial wastewater treatment system of this invention needs to treat asynchronously discharged wastewater from multiple production units. The upper part of the figure shows the time-series concentration curve of wastewater from production unit 1. This unit exhibits short-cycle (approximately 4 hours), high peak values, and coarse particulate discharge characteristics, with drastic concentration fluctuations, reaching peak values ​​above 250 mg / L. The middle part of the figure shows the time-series concentration curve of wastewater from production unit 2. This unit exhibits long-cycle (approximately 12 hours), gradual, and fine particulate discharge characteristics, with relatively stable concentration changes. The lower part of the figure shows the total influent concentration curve after the wastewater from the two units is mixed. Due to the asynchronous production cycles of the two units, the mixed influent concentration exhibits a complex superposition effect. The area marked by the dashed box in the figure represents the complex periods of load superposition. During these periods, the high-load periods of the two units overlap, resulting in a composite peak in the total influent concentration, placing higher demands on the operation and control of subsequent treatment units. The intelligent control system of this invention can analyze this multi-source asynchronous superposition characteristic in real time, predict the arrival of complex periods, and adjust the operating parameters of each treatment unit in advance to ensure that the effluent quality consistently meets standards.

[0065] To address this problem, the method of this invention is applied to the system. First, the data acquisition and alignment process in step one involves collecting and aligning the process parameters of production units one and two, the operating data of the dust removal system, and the pipe flow velocity data of their respective wastewater before entering the main pipeline network. Through reverse virtual tracing, any water quality sample collected by the purification system can be precisely traced back to its primary contributing source (unit one or two) and the operating conditions at the time of its generation.

[0066] Secondly, in step two, a multi-input multi-output (MIMO) deep learning prediction model can be trained. The input layer of this model receives real-time operating data from two production units, while the output layer simultaneously predicts the pollutant flux characteristics contributed by unit one and unit two, respectively, when the pollutant reaches the dosing point in the future (i.e., ...). , and , ).

[0067] The key lies in the coordinated implementation of steps three and four. The system generates and independently tracks virtual tracer particles for each of the predicted high-load water masses from two different sources. Based on time-delay calculations, the decision control module obtains a sequence of water masses expected to arrive at the dosing point at various points in the future timeline. This sequence clearly indicates the source of each water mass and its corresponding pollutant characteristics. When the system predicts that a high-concentration coarse-particle water mass from Unit 1 and a medium-concentration fine-particle water mass from Unit 2 will arrive at the dosing point simultaneously at a future time, the decision control module executes a composite water quality weighted fusion algorithm. This algorithm calculates the total suspended solids concentration after mixing based on the predicted flow rate and SS concentration of the two water masses, and calculates the equivalent particle size distribution index (EPS) of the mixed particles based on their respective EPS indices and flow rates. Ultimately, based on the calculated composite water quality characteristics (total suspended solids concentration and equivalent particle size distribution index), the system generates an optimal ratio of coagulant to flocculant aid, thereby achieving precise, feedforward control of complex wastewater quality from multiple sources, asynchronously generated wastewater with significantly different characteristics.

[0068] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for coordinated control of circulating water purification in wet dust removal based on dust load feedforward compensation, characterized in that, Includes the following steps: Step 1 involves constructing a multi-source heterogeneous data acquisition and dynamic spatiotemporal alignment database. This step includes: collecting dust removal-side characteristic data representing the operating status of the dust removal system, production rhythm signals representing the generation patterns of pollution sources, transmission-side characteristic data representing the pollutant transport process, and purification-side characteristic data representing the operating effect of the purification system; and using dynamic sample alignment technology based on reverse virtual tracing, calculating the actual physical transmission lag time of the purification-side water quality data samples in the pipeline network, and combining the purification-side water quality data samples with the dust removal-side characteristic data and the production rhythm signals at the traced-back source time into a spatiotemporally precisely aligned sample pair to construct a training dataset. Step 2: Using the training dataset, train a deep learning-based dust load feature prediction model to predict the pollutant flux characteristics that will arrive at the dosing point in the circulating water purification system in the future, based on real-time dust removal side feature data and production rhythm signals. Step 3: Construct a pollutant transport time delay calculation model based on fluid mechanics to calculate the dynamic transport time of pollutants from the dust collector to the dosing point in the circulating water purification system in real time. Step 4: Based on the pollutant flux characteristics predicted by the dust removal load characteristic prediction model and the dynamic transmission time calculated by the pollutant transport time delay calculation model, feedforward compensation control is implemented on the circulating water purification system. Step 5: Based on the deviation between the effluent water quality index of the circulating water purification system and the control target value, perform feedback correction on the dust removal load characteristic prediction model.

2. The method according to claim 1, characterized in that, The specific steps for constructing the training dataset using dynamic sample alignment technology based on reverse virtual tracing in step one are as follows: For each water quality data sample from the purification side in the historical database, the pipeline hydraulic model is used. The historical pipeline instantaneous flow velocity data sequence before the sampling time of the water quality data sample is input, and the actual physical transport lag time experienced by the water mass corresponding to the water quality data sample in the pipeline is calculated back through inverse integral calculation. By utilizing the actual physical transmission lag time, the water quality data sample from the purification side is combined with the dust removal side feature data and the production rhythm signal from the source time back to form a spatiotemporally precisely aligned sample pair.

3. The method according to claim 1, characterized in that, The dust load characteristic prediction model is a hybrid neural network that combines a long short-term memory network with an attention mechanism.

4. The method according to any one of claims 1 to 3, characterized in that, The pollutant flux characteristics include the total mass concentration of suspended solids characterizing the load, and the equivalent particle size distribution index characterizing the load quality; wherein, the equivalent particle size distribution index is a comprehensive index used to quantify the settling performance of suspended particulate matter in wastewater, which is obtained based on the mapping relationship between source production operating parameters and wastewater physical settling characteristics.

5. The method according to claim 1, characterized in that, In step three, the pollutant transport time delay calculation model based on fluid mechanics is implemented using the virtual tracer particle method.

6. The method according to claim 5, characterized in that, The virtual tracer particle method is implemented as follows: When the rate of change of pollutant flux characteristics predicted by the dust removal load characteristic prediction model exceeds a preset threshold, a virtual tracer particle carrying a timestamp and pollutant attributes is generated at the wastewater outlet of the dust collector. The instantaneous flow velocity of the return water pipe network is collected in real time, and the displacement of the virtual tracer particle is integrated and extrapolated in real time. The cumulative displacement is calculated by time integration of the instantaneous flow velocity v(τ) from the generation time T0 to the current time t. When the cumulative displacement S(t) of the virtual tracer particle is equal to the physical length of the pipeline from the dust collector outlet to the dosing point, it is determined that the virtual tracer particle has reached the dosing point, thereby obtaining the dynamic transmission time.

7. The method according to claim 1, characterized in that, The feedforward compensation control implemented in step four includes feedforward dosing correction control, which specifically involves: Adjust the output of the dosing metering pump at a preset lead time before the virtual tracer particles are expected to arrive at the dosing point; A dynamic proportioning strategy is implemented. When the predicted equivalent particle size distribution index is higher than the first preset threshold, it indicates that the wastewater is mainly composed of easily settled coarse particles, so the dosage of coagulant is increased. When the predicted equivalent particle size distribution index is lower than the second preset threshold, it indicates that the wastewater is mainly composed of fine particles, so the dosage of coagulant aid is increased while increasing the dosage of coagulant.

8. The method according to claim 1 or 7, characterized in that, The feedforward compensation control implemented in step four also includes load easing and sludge discharge linkage control; the load easing and sludge discharge linkage control includes at least one of the following control methods: When the predicted peak value of total suspended solids concentration exceeds the maximum processing capacity of the circulating water purification system under the current operating conditions, the switching valve in the pipeline network is controlled in advance to temporarily divert the high turbidity wastewater to the emergency water tank. Based on the total mass concentration of suspended solids in the pollutant flux characteristics, the amount of sludge that will increase in the sedimentation tank within a preset time period is calculated by integration. Before the calculated sludge level reaches the high-level alarm threshold, the sludge discharge pump is started or the sludge discharge valve is opened for preventive sludge discharge.

9. The method according to claim 1, characterized in that, The specific method for performing feedback correction in step five is as follows: When the deviation between the turbidity of the sedimentation tank effluent monitored in real time and the control target value exceeds the preset duration, an online learning or offline update mechanism is triggered. The weight parameters of the dust removal load characteristic prediction model are fine-tuned or retrained using the latest collected data that has undergone spatiotemporal alignment processing.

10. A wet dust removal circulating water purification collaborative control system based on dust removal load feedforward compensation, characterized in that, include: The data construction module is configured to build a multi-source heterogeneous data acquisition and dynamic spatiotemporal alignment database, including: collecting dust removal side feature data characterizing the operating status of the dust removal system, production rhythm signals characterizing the generation pattern of pollution sources, transmission side feature data characterizing the pollutant transport process, and purification side feature data characterizing the operating effect of the purification system; and using dynamic sample alignment technology based on reverse virtual tracing, calculating the actual physical transmission lag time of the purification side water quality data sample in the pipeline network, and combining the purification side water quality data sample with the dust removal side feature data and the production rhythm signal at the source time traced back to form a spatiotemporally precisely aligned sample pair to construct a training dataset; The model training module is configured to use the training dataset to train a deep learning-based dust load feature prediction model, which is used to predict the pollutant flux characteristics that will arrive at the dosing point in the circulating water purification system in the future based on real-time dust removal side feature data and production rhythm signals. The time delay calculation module is configured to build a pollutant transport time delay calculation model based on fluid mechanics, which is used to calculate the dynamic transport time of pollutants from the dust collector to the dosing point in the circulating water purification system in real time. The collaborative control module is configured to implement feedforward compensation control on the circulating water purification system based on the pollutant flux characteristics predicted by the dust removal load characteristic prediction model and the dynamic transmission time calculated by the pollutant transport time delay calculation model. The feedback correction module is configured to perform feedback correction on the dust removal load characteristic prediction model based on the deviation between the effluent water quality index of the circulating water purification system and the control target value.

Images