An AI-based tidal flow artificial wetland automatic adjustment control method and system

Abstract

The application discloses a kind of tidal current artificial wetland automatic regulation control method and system based on AI, and wherein method includes: real-time acquisition of the multi-source sensing data of tidal current artificial wetland system, data is preprocessed and feature extraction;Based on the denitrification efficiency prediction model of LSTM network construction determines TN removal rate prediction value and optimal DO threshold value;According to TN removal rate prediction value and optimal DO threshold value, based on DRL algorithm, establish joint reward function, determine multi-objective optimization strategy, dynamically generate the adjustment parameter of tidal current artificial wetland system;According to adjustment parameter, adjust water inlet / drainage frequency, accurately add carbon source, adjust aeration intensity according to gradient;Microbial activity feedback data is obtained, and the control strategy is optimized according to the microbial activity feedback data, so that the effluent TN reaches the preset standard.The application can simultaneously consider multiple optimization goals, so as to realize the efficient, energy-saving and environmentally friendly control of artificial wetland.

Description

Technical Field

[0001] This invention relates to the field of constructed wetland environmental control technology, and in particular to an AI-based automated regulation and control method and system for tidal flow constructed wetlands. Background Technology

[0002] Modern water treatment often requires the synergistic action of physical, chemical, and biological processes to remove pollutants, especially nitrogen removal, which depends on the balance between nitrification and denitrification. Traditional constructed wetland systems typically rely on manual monitoring and adjustment, which is cumbersome and inefficient. AI-based automated control methods can achieve real-time monitoring and intelligent control, reducing human intervention and improving system efficiency and stability.

[0003] Currently, artificial intelligence (AI) technology demonstrates significant advantages in water quality prediction, parameter optimization, and real-time control. For example, Asia Pulp & Paper's AI water system achieves millisecond-level response through high-precision sensors and dynamic algorithms. However, its application is limited to industrial aeration and chemical dosing, failing to cover the multi-dimensional regulation of wetland ecosystems. Existing research lacks integration of AI with bio-hydraulic coupling models of tidal flow wetlands, based on real-time water quality data (such as NH4). + -N, NO3 - Adjusting the flooding / idle time (using -N concentration) balances nitrification and denitrification requirements, achieving dynamic optimization of the tidal cycle. Furthermore, existing systems struggle to adapt to complex environmental changes and lack the ability to adjust to dynamic changes in real time.

[0004] Therefore, there is a need to provide an AI-based automated regulation and control method and system for tidal flow constructed wetlands. By using LSTM networks to predict denitrification efficiency and considering the balance between energy consumption and carbon source addition, the regulation strategy of constructed wetland systems can be more comprehensive, taking into account multiple optimization objectives at the same time, thereby achieving efficient, energy-saving and environmentally friendly control of constructed wetlands. Summary of the Invention

[0005] In view of this, the present invention provides an AI-based automated regulation and control method and system for tidal flow constructed wetlands, in order to solve the technical problem that current constructed wetland control methods are unable to adapt to complex environmental changes and lack the ability to adjust to dynamic changes in real time.

[0006] To achieve the above-mentioned technical objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides an AI-based automated regulation and control method for tidal flow constructed wetlands, comprising:

[0008] Real-time acquisition of multi-source sensor data from tidal flow constructed wetland systems;

[0009] The multi-source sensor data is preprocessed and features are extracted to obtain key driving factors;

[0010] A nitrogen removal efficiency prediction model was built based on LSTM network. Based on historical operating data and key driving factors collected in real time, the predicted value of TN removal rate and the optimal DO threshold were determined.

[0011] Based on the predicted TN removal rate and the optimal DO threshold, a joint reward function is established based on the DRL algorithm to maximize denitrification efficiency, minimize carbon source addition, and minimize energy consumption. This determines a multi-objective optimization strategy and dynamically generates adjustment parameters for the tidal flow constructed wetland system.

[0012] According to the adjustment parameters, the inlet / outlet frequency is adjusted by an electric gate controller, the carbon source is precisely added by a peristaltic pump and a slow-release carbon source, and the aeration intensity is adjusted according to the predicted DO concentration value in a gradient manner.

[0013] Obtain microbial activity feedback data, and optimize the control strategy based on the microbial activity feedback data to make the effluent TN reach the preset standard.

[0014] Furthermore, the multi-source sensing data includes: water quality sensing data, environmental perception data, and biofilm activity monitoring data;

[0015] The water quality parameters include NH4. + -N, NO3 - -N, DO, pH, COD, temperature, flow rate and water level data are used to characterize the health status of water bodies and changes in pollutants;

[0016] The environmental sensing data includes light intensity, rainfall, and wind speed, which are used to characterize the impact of external climate on water quality changes.

[0017] The biofilm activity monitoring data includes dynamically monitored denitrifying bacteria metabolic activity and nitrification / denitrification functional gene abundance, used to characterize the impact of microbial communities in water bodies on water quality.

[0018] Furthermore, the multi-source sensing data is preprocessed and features are extracted to obtain key driving factors, including:

[0019] The multi-source sensor data is normalized and candidate regulation and control features are extracted.

[0020] Principal component analysis and mutual information screening were used to construct a matrix of driving factors for denitrification efficiency based on the candidate regulatory control characteristics, thereby obtaining the key driving factors affecting denitrification efficiency.

[0021] Furthermore, a nitrogen removal efficiency prediction model is constructed based on an LSTM network. Based on historical operating data and key driving factors collected in real time, the predicted TN removal rate and the optimal DO threshold are determined, including:

[0022] Water quality sensor data, environmental perception data, and biofilm activity monitoring data are used as inputs to the LSTM network;

[0023] The time series input is constructed by setting a preset time window, and the bidirectional LSTM is used to capture the temporal dependencies between the preceding and following time series. The predicted value of TN removal rate and DO threshold are output for the preset time period in the future.

[0024] Furthermore, based on the predicted TN removal rate and the optimal DO threshold, a joint reward function is established using the DRL algorithm to maximize nitrogen removal efficiency, minimize carbon source addition, and minimize energy consumption. This determines a multi-objective optimization strategy and dynamically generates adjustment parameters for the tidal flow constructed wetland system, including:

[0025] The predicted TN removal rate and the optimal DO threshold output by the LSTM network are used as extended states, which together with the real-time parameters constitute the multi-dimensional state space of DRL.

[0026] The action space is defined by the tidal cycle adjustment rate, aeration intensity classification, and carbon source input acceleration rate.

[0027] The multidimensional state space, action space, and joint reward function of DRL are deeply coupled to form a closed-loop optimization mechanism.

[0028] Furthermore, the reward function is determined based on the joint optimization of denitrification efficiency, carbon source consumption, and aeration energy consumption, and the strategy is iteratively updated using the Q-learning algorithm.

[0029] On the other hand, the present invention also provides a tidal flow artificial wetland system, comprising a sensing layer, an AI decision-making layer, an execution layer and a wetland function layer connected in sequence;

[0030] The sensing layer is used to collect multi-source sensor data of the tidal flow constructed wetland system in real time.

[0031] The AI ​​decision layer is used to preprocess and extract features from the multi-source sensor data to obtain key driving factors. Based on the LSTM network, a denitrification efficiency prediction model is constructed. According to historical operating data and real-time collected key driving factors, the predicted value of TN removal rate and the optimal DO threshold are determined. Based on the predicted value of TN removal rate and the optimal DO threshold, a multi-objective optimization strategy is determined based on the DRL algorithm with the joint reward function of maximizing denitrification efficiency, minimizing carbon source addition, and minimizing energy consumption. The adjustment parameters of the tidal flow constructed wetland system are dynamically generated.

[0032] The execution layer is used to adjust the inlet / outlet frequency through an electric gate controller according to the adjustment parameters, to accurately add carbon source through a peristaltic pump and a slow-release carbon source, and to adjust the aeration intensity according to the predicted DO concentration value in a gradient manner.

[0033] The wetland functional layer adopts a layered filler optimization layer and a plant root oxygenation structure to construct a multi-pathway denitrification system.

[0034] Furthermore, the layered filler optimization layer of the wetland functional layer includes an upper aerobic zone and a lower anoxic zone; the upper aerobic zone is filled with a composite of volcanic rock and iron-carbon composite material to enhance the synergistic effect of nitrifying bacteria enrichment and chemical oxidation; the lower anoxic zone is filled with a mixture of biochar and sulfur autotrophic denitrification matrix to promote the coupling of sulfur-driven denitrification and biochar adsorption.

[0035] The plant root biomimetic oxygenation structure achieves its oxygen supply function by planting emergent plants.

[0036] Furthermore, the sensing layer includes a water quality sensing module, an environmental sensing module, and a biofilm activity monitoring module;

[0037] The water quality sensing module includes multiple sensors, which are deployed in the water distribution tank, the layered filler area and the water collection tank to collect water quality data in real time.

[0038] The environmental sensing module is used to collect data on light intensity, rainfall, and wind speed.

[0039] The biofilm activity monitoring module is used to collect microbial functional gene abundance data based on electrochemical impedance spectroscopy and qPCR technology.

[0040] Furthermore, the execution layer includes a tidal gate controller, a gradient aeration system, and a dual-mode carbon source dosing device;

[0041] The tidal gate controller is used to regulate and control the inflow rate;

[0042] The gradient aeration system is used to control the aeration intensity in different areas to optimize the dissolved oxygen level in the water.

[0043] The dual-mode carbon source dosing device is used to add organic carbon sources to support the nitrogen removal process of microorganisms.

[0044] Compared with existing technologies, the AI-based automated regulation and control method and system for tidal flow constructed wetlands proposed in this invention have the following advantages:

[0045] (1) By using LSTM networks to train historical data and real-time sensor data, the temporal characteristics of water quality changes can be effectively captured, and the TN (total nitrogen) removal rate and DO (dissolved oxygen) threshold can be accurately predicted, making the prediction of denitrification efficiency more accurate.

[0046] (2) Optimization based on the DRL algorithm not only considers the improvement of denitrification efficiency, but also takes the amount of carbon source added and energy consumption as optimization targets. It can maximize denitrification efficiency while reducing the excessive use of carbon source and energy consumption, ensuring the dual optimization of environmental and economic benefits. Through the precise addition of peristaltic pump and slow-release carbon source, the amount of carbon source added can be accurately controlled, avoiding carbon source waste and reducing operating costs.

[0047] (3) By acquiring microbial activity feedback data, the control strategy can be further adjusted. This gives the system a strong self-optimization capability, enabling it to adjust the water treatment strategy according to the actual operating conditions and ensure that the effluent TN meets the preset standards.

[0048] In summary, this invention achieves precise control over the amount of carbon source added through a closed-loop mechanism of "data-driven decision-making" and "feedback-driven learning," avoiding carbon source waste, reducing operating costs, and improving the long-term stability of equipment operation, thus providing an efficient, low-carbon, and reliable solution for the wastewater treatment field. Attached Figure Description

[0049] Figure 1 A flowchart illustrating the AI-based automated regulation and control method for constructed wetlands based on tidal flow provided by this invention;

[0050] Figure 2 A flowchart illustrating the establishment and dynamic optimization of the prediction model provided by this invention;

[0051] Figure 3 A schematic diagram of the structure of the AI-based automated regulation and control system for constructed wetlands provided by this invention;

[0052] Figure 4 This is a schematic diagram showing the connection relationships of the various parts of the system provided by the present invention;

[0053] In the diagram, 1-Water inlet, 2-Connecting wire, 3-Electric water inlet valve, 4-Support, 5-Light sensor, 6-Rain gauge, 7-Anemometer, 8-Leaf photosynthesis detector, 9-Water distribution tank, 10-Water inlet quality sensor, 11-Carbonation device, 12-Water inlet pump, 13-Water quality sensor for upper packing zone, 14-Bioactive probe for upper packing zone, 15-Water quality sensor for lower packing zone, 16-Bioactive probe for lower packing zone, 17-AI decision layer, 18-Emerging plants, 19-Upper packing, 20-Microporous aeration network, 21-Aeration hole, 22-Lower packing, 23-Solar aerator, 24-Water outlet pump, 25-Water outlet quality sensor, 26-Collection tank, 27-Electric water outlet valve, 28-Water outlet. Detailed Implementation

[0054] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0055] Example 1

[0056] Please see Figure 1 This embodiment provides an AI-based automated regulation and control method for tidal flow constructed wetlands, including:

[0057] Step S101: Real-time acquisition of multi-source sensor data from the tidal flow constructed wetland system;

[0058] Step S102: Preprocess and extract features from the multi-source sensing data to obtain key driving factors;

[0059] Step S103: Construct a nitrogen removal efficiency prediction model based on LSTM network, and determine the predicted value of TN removal rate and the optimal DO threshold based on historical operating data and key driving factors collected in real time.

[0060] Step S104: Based on the predicted TN removal rate and the optimal DO threshold, establish a multi-objective optimization strategy with the joint reward function of maximizing denitrification efficiency, minimizing carbon source addition, and minimizing energy consumption based on the DRL algorithm, and dynamically generate the adjustment parameters of the tidal flow constructed wetland system.

[0061] Step S105: Adjust the inlet / outlet frequency according to the adjustment parameters via the electric gate controller, accurately add carbon source via peristaltic pump and slow-release carbon source, and adjust the aeration intensity according to the predicted DO concentration value in a gradient manner.

[0062] Step S106: Obtain microbial activity feedback data, and optimize the control strategy based on the microbial activity feedback data so that the effluent TN reaches the preset standard.

[0063] The method in this embodiment utilizes an LSTM network to train on historical and real-time sensor data, effectively capturing the temporal characteristics of water quality changes and accurately predicting TN (total nitrogen) removal rate and DO (dissolved oxygen) threshold, resulting in more precise predictions of denitrification efficiency. Optimization based on the DRL algorithm not only considers improving denitrification efficiency but also optimizes carbon source dosage and energy consumption, maximizing denitrification efficiency while reducing excessive carbon source use and energy consumption, ensuring optimal environmental and economic benefits. Precise addition of peristaltic pumps and slow-release carbon sources allows for precise control of carbon source addition, avoiding waste and reducing operating costs. By acquiring microbial activity feedback data, the control strategy can be further adjusted. This gives the system a strong self-optimization capability, enabling it to adjust water treatment strategies based on actual operating conditions to ensure that the effluent TN meets preset standards.

[0064] In a preferred embodiment, in step S101, the multi-source sensing data includes: water quality sensing data, environmental perception data, and biofilm activity monitoring data;

[0065] The water quality parameters include NH4. + -N, NO3 - -N, DO, pH, COD, temperature, flow rate and water level data are used to characterize the health status of water bodies and changes in pollutants;

[0066] The environmental sensing data includes light intensity, rainfall, and wind speed, which are used to characterize the impact of external climate on water quality changes.

[0067] The biofilm activity monitoring data includes dynamically monitored denitrifying bacteria metabolic activity and nitrification / denitrification functional gene abundance, used to characterize the impact of microbial communities in water bodies on water quality.

[0068] In a preferred embodiment, in step S102, the multi-source sensing data is preprocessed and feature extracted to obtain key driving factors, including:

[0069] The multi-source sensor data is normalized and candidate regulation and control features are extracted.

[0070] Principal component analysis and mutual information screening were used to construct a matrix of driving factors for denitrification efficiency based on the candidate regulatory control characteristics, thereby obtaining the key driving factors affecting denitrification efficiency.

[0071] In a preferred embodiment, in step S103, a nitrogen removal efficiency prediction model is constructed based on an LSTM network. Based on historical operating data and key driving factors collected in real time, the predicted TN removal rate and the optimal DO threshold are determined, including:

[0072] Water quality sensor data, environmental perception data, and biofilm activity monitoring data are used as inputs to the LSTM network;

[0073] The time series input is constructed by setting a preset time window, and the bidirectional LSTM is used to capture the temporal dependencies between the preceding and following time series. The predicted value of TN removal rate and DO threshold are output for the preset time period in the future.

[0074] In a preferred embodiment, in step S104, based on the predicted TN removal rate and the optimal DO threshold, a joint reward function is established using the DRL algorithm to maximize denitrification efficiency, minimize carbon source addition, and minimize energy consumption. This determines a multi-objective optimization strategy and dynamically generates adjustment parameters for the tidal flow constructed wetland system, including:

[0075] The predicted TN removal rate and the optimal DO threshold output by the LSTM network are used as extended states, which together with the real-time parameters constitute the multi-dimensional state space of DRL.

[0076] The action space is defined by the tidal cycle adjustment rate, aeration intensity classification, and carbon source input acceleration rate.

[0077] The multidimensional state space, action space, and joint reward function of DRL are deeply coupled to form a closed-loop optimization mechanism.

[0078] In a preferred embodiment, the reward function is determined based on the joint optimization of denitrification efficiency, carbon source consumption, and aeration energy consumption, and the strategy is iteratively updated using the Q-learning algorithm.

[0079] As a specific example, in the LSTM water quality prediction model: the input layer contains 12-dimensional real-time parameters (NH4+). + -N, NO3 - -N, DO, pH, COD, temperature, flow rate, water level, light intensity, rainfall, wind speed, and functional gene abundance), in the optimization engine based on deep reinforcement learning (DRL), the dynamic prediction model outputs the predicted value of TN removal rate and DO threshold suggestion for the next 2 hours, which forms a closed-loop optimization mechanism through deep coupling with the state space, action space and reward function of DRL.

[0080] Specifically, the LSTM dynamic prediction model's real-time output of the next 2 hours' total nitrogen (TN) removal rate and dissolved oxygen (DO) threshold suggestions serves as an extended state variable input to the DRL framework. This input, along with real-time data on water quality parameters, meteorological data, operational parameters, and microbial gene abundance, constructs a multi-dimensional state space. The action space is designed to encompass tidal cycle adjustments (±10%), aeration intensity grading (0-5 levels), and carbon source addition acceleration rate (0-20 mL / min). The reward function is constructed to jointly optimize nitrogen removal efficiency (TN removal rate weight 0.6), carbon source consumption (weight -0.3), and aeration energy consumption (weight -0.1), with the strategy iteratively updated using a Q-learning algorithm.

[0081] In some embodiments, this method further includes enhancing transfer learning to pre-configure emergency plans. By analyzing historical data to construct a model of sudden high ammonia nitrogen scenarios, intelligent control plans are generated in advance (such as dynamically adjusting tidal frequency to enhance water reoxygenation efficiency and precisely starting and stopping emergency carbon source pools to balance the carbon-nitrogen ratio). The plan parameters are then dynamically optimized using transfer learning technology. This method triggers pre-emptive control by predicting water quality fluctuation trends, reducing redundant operation of the aeration system, and simultaneously optimizing emergency chemical dosing strategies. While efficiently reducing ammonia nitrogen shock loads, it significantly reduces energy and resource consumption, shortens recovery time under abnormal conditions, and ultimately achieves a comprehensive improvement and refined management of the wastewater treatment process's resilience to shocks.

[0082] Specifically, when NH4 enters the water... + When the -N concentration abruptly exceeds the threshold (>30 mg / L), the transfer learning module calls upon the historical contingency plan library to generate emergency control strategies (such as shortening the tidal cycle to 0.5 hours or initiating slow-release carbon source pulse dosing). Based on real-time biofilm activity data (EIS impedance change rate >15%), the contingency plan parameters are adjusted through online learning to ensure the system quickly and dynamically corrects and gradually restores steady state.

[0083] In addition, to improve the system's reliability, efficiency, and stability, and reduce operating costs and equipment failures, sensor calibration (standard liquid calibration), aeration disc backwashing (high-pressure water gun pulse), and electric valve sensitivity checks (based on water level change monitoring) are performed automatically every month. Operational data is uploaded to the cloud quarterly, the global AI model is updated through learning, and then distributed to local edge terminals.

[0084] It should be noted that the slow-release carbon source mentioned in step S105 refers to a substance that can gradually release carbon for denitrifying bacteria to utilize, thereby improving nitrogen removal efficiency. Specifically, common slow-release carbon sources include: acetate, polymer-based slow-release carbon sources, organic waste, and propionate, etc.

[0085] like Figure 2 As shown, Figure 2 The flowchart illustrating the establishment and dynamic optimization of the prediction model for this method is presented.

[0086] Example 2

[0087] This invention provides a tidal flow constructed wetland system, comprising a sensing layer, an AI decision-making layer, an execution layer, and a wetland function layer connected in sequence.

[0088] The sensing layer is used to collect multi-source sensor data of the tidal flow constructed wetland system in real time.

[0089] The AI ​​decision layer is used to preprocess and extract features from the multi-source sensor data to obtain key driving factors. Based on the LSTM network, a denitrification efficiency prediction model is constructed. According to historical operating data and real-time collected key driving factors, the predicted value of TN removal rate and the optimal DO threshold are determined. Based on the predicted value of TN removal rate and the optimal DO threshold, a multi-objective optimization strategy is determined based on the DRL algorithm with the joint reward function of maximizing denitrification efficiency, minimizing carbon source addition, and minimizing energy consumption. The adjustment parameters of the tidal flow constructed wetland system are dynamically generated.

[0090] The execution layer is used to adjust the inlet / outlet frequency through an electric gate controller according to the adjustment parameters, to accurately add carbon source through a peristaltic pump and a slow-release carbon source, and to adjust the aeration intensity according to the predicted DO concentration value in a gradient manner.

[0091] The wetland functional layer adopts a layered filler optimization layer and a plant root oxygenation structure to construct a multi-pathway denitrification system.

[0092] The system in this embodiment constructs an integrated system of "intelligent perception, dynamic decision-making, precise execution, and functional enhancement," which achieves a dual improvement in denitrification efficiency and operational energy efficiency through multi-dimensional collaborative optimization.

[0093] As a preferred embodiment, such as Figure 3 As shown, Figure 3 The actual structural diagram of the system in this embodiment is shown. The layered filler optimization layer of the wetland functional layer includes an upper aerobic zone and a lower anoxic zone; the upper aerobic zone is filled with volcanic rock (particle size 10-20mm) and iron-carbon composite material (Fe). 30 A layer of biochar (with a surface area ≥800 m² / kg) with a mass ratio of 1:2 was applied, 40 cm thick. Biochar (with a specific surface area ≥800 m²) was then filled into the lower anoxic zone. 2 The mixture consists of a sulfur-based autotrophic denitrification substrate (sulfur powder / limestone volume ratio 3:1), with a thickness of 60 cm. The plant root-inspired oxygen-enriching structure provides oxygen by planting emergent aquatic plants.

[0094] Emergent plants are planted in the root oxygenation structure to provide oxygen, and aerators are used to meet the metabolic needs of denitrifying bacteria.

[0095] By optimizing the layered packing material (volcanic rock / iron-carbon + biochar / sulfur matrix) and synergistically combining it with a biomimetic oxygenation structure, a multi-pathway nitrogen removal system of chemical oxidation-biological adsorption-sulfur autotrophic denitrification is constructed, which significantly improves the TN removal rate and effectively enhances the contribution rate of simultaneous nitrification and denitrification.

[0096] The upper packing zone uses volcanic rock-iron-carbon composite packing to enhance the enrichment capacity of nitrifying bacteria, while the lower packing zone uses a combination of biochar and sulfur matrix to improve the resistance to water quality fluctuations, ensuring the stable operation of the system under water quality fluctuations.

[0097] In a preferred embodiment, the sensing layer includes a water quality sensing module, an environmental sensing module, and a biofilm activity monitoring module.

[0098] like Figure 3 As shown, the water quality sensing module includes multiple sensors, which are deployed in the water distribution pool, the layered filler area and the water collection pool of the wetland functional layer for real-time collection of water quality data.

[0099] The environmental sensing module is used to collect data on light intensity, rainfall, and wind speed.

[0100] The biofilm activity monitoring module is used to collect microbial functional gene abundance data based on electrochemical impedance spectroscopy and qPCR technology.

[0101] In some embodiments, the water quality sensing module collects water quality and environmental data in real time by deploying multi-parameter high-precision online sensors and dynamically adjusts the sampling frequency (5-30 minutes / time). Through a high-precision sensor array, NH4 is monitored. + -N, NO3 - -N, DO, pH, COD, temperature, flow rate, water level, etc. are monitored and collected.

[0102] The environmental sensing module integrates sensors for light intensity, rainfall, and wind speed, enabling simultaneous acquisition of meteorological data and the construction of a coupled meteorological-water quality dataset. A leaf photosynthesis detector is used to monitor plant photosynthesis and related aspects.

[0103] The biofilm activity monitoring module, based on the combination of electrochemical impedance spectroscopy (EIS) and qPCR technology, dynamically monitors the metabolic activity of denitrifying bacteria and the abundance of nitrification / denitrification functional genes (such as amoA, nirS, and nosZ), providing dynamic response data at the microbial level for AI decision-making.

[0104] As a specific implementation, light intensity sensors, rain gauges, and anemometers are installed around the wetland to collect meteorological data and transmit it to an edge computing terminal. Leaf photosynthesis detectors are also installed around the wetland to monitor the photosynthetic intensity of emergent plants in real time through spectral analysis. Combined with light intensity and water level data, this data is dynamically fed back to the AI ​​decision-making layer to optimize aeration and tidal cycle strategies.

[0105] Electrochemical impedance spectroscopy (EIS) probes were embedded in the packing layer, and qPCR analysis was performed by periodic sampling to dynamically monitor the abundance of functional genes of nitrifying bacteria (amoA gene) and denitrifying bacteria (nirS, nosZ genes).

[0106] In actual operation, each sensor uploads data to the edge computing terminal every 10 minutes, and abnormal data triggers an encrypted retransmission mechanism.

[0107] As a specific embodiment, the AI ​​decision-making layer includes data fusion and feature extraction, dynamic prediction models, and intelligent optimization engines.

[0108] The data fusion and feature extraction process utilizes edge computing terminals to normalize multi-source heterogeneous data and extract key driving factors for nitrogen removal efficiency (such as DO gradient, C / N ratio, and hydraulic retention time). The dynamic prediction model employs an LSTM neural network to construct a nitrogen removal efficiency prediction model. Input variables include real-time water quality data, environmental factors, and historical operating parameters. Outputs are the predicted TN removal rate and the optimal DO threshold. The intelligent optimization engine establishes a multi-objective optimization strategy based on deep reinforcement learning (DRL) algorithms. It uses maximizing nitrogen removal efficiency, minimizing carbon source dosage, and minimizing energy consumption as the joint reward function to dynamically generate the optimal solution set for tidal cycle, aeration intensity, and carbon source dosage.

[0109] The bidirectional cyclic optimization mode of the AI ​​decision-making layer in this embodiment achieves continuous model updates and equipment maintenance through a closed-loop mechanism of "data-driven decision-making" and "feedback-driven learning," ensuring long-term operational stability. This mode upgrades the traditional static control system into a dynamic, adaptive intelligent ecosystem, providing an efficient, low-carbon, and reliable solution for the wastewater treatment field.

[0110] Furthermore, predictions and decisions are generated based on real-time data. After execution, data is received and feedback is obtained. Data is obtained through continuous execution and feedback for autonomous training, model updates and upgrades, and decision optimization is achieved to carry out a new round of execution.

[0111] In a preferred embodiment, the execution layer includes a tidal gate controller, a gradient aeration system, and a dual-mode carbon source dosing device;

[0112] The tidal gate controller is used to regulate and control the inflow rate;

[0113] The gradient aeration system is used to control the aeration intensity in different areas to optimize the dissolved oxygen level in the water.

[0114] The dual-mode carbon source dosing device is used to add organic carbon sources to support the nitrogen removal process of microorganisms.

[0115] like Figure 4 As shown, Figure 4 A schematic diagram showing the connection relationship of each part of the system in this embodiment is provided.

[0116] Specifically, the tidal gate controller adjusts the inlet / outlet frequency (adjustable from 0.5 to 24 hours) via an electric gate controller to create a tidal inlet. Several electric valves are installed in the outlet channel from top to bottom (the last valve is 5cm from the wetland bottom), spaced 10cm apart. AI controls the opening of each electric valve to regulate the outlet water level. Electric gate controllers are installed in the wetland's inlet and outlet pipes, using AI to control the gate opening and closing time and frequency, combined with a water level sensor to create a tidal inlet (adjustable cycle from 0.5 to 24 hours).

[0117] The gradient aeration system uses micro aerators linked with the aeration pipeline network. The aeration intensity is adjusted according to the predicted DO concentration in a gradient (0.5-6 mg / L). A spatiotemporal differential control mode of "enhanced aeration in the initial stage of submersion and trace oxygen supply in the final stage" is designed to prioritize enhancing the oxygen transfer efficiency of the nitrification zone.

[0118] The dual-mode carbon source dosing device is designed for denitrification and uses a peristaltic pump in conjunction with a slow-release carbon source (polycaprolactone / starch composite granules) device for dosing. Polycaprolactone (PCL) provides mechanical strength, while the starch composite granules (5–10 mm in diameter) serve as a biodegradable carbon source. Microorganisms degrade the PCL / starch composite granules, slowly releasing dissolved organic carbon (DOC) and maintaining a dynamic equilibrium of a C / N ratio of 3.5–5.0.

[0119] Based on EIS electrochemical impedance spectroscopy and qPCR molecular monitoring data, a response surface model of dynamic C / N ratio (3.5-5.0) and carbon source addition ratio (5%-20%) was established by synergistically adding carbon source through a peristaltic pump and a slow-release carbon source (polycaprolactone / starch composite particles) device. This reduces carbon source waste compared to traditional methods, and at the same time reduces the demand for exogenous carbon through sulfur autotrophic denitrification, achieving synergistic efficiency in pollution reduction and carbon reduction.

[0120] A microporous aeration network is laid at the bottom of the aerobic zone, and gas is released in conjunction with a micro aerator to meet the basic oxygenation needs. The aeration intensity is dynamically adjusted by the AI ​​decision layer according to the preset gradient (0.5-6 mg / L).

[0121] A peristaltic pump and a slow-release carbon source dosing device (polycaprolactone / starch composite granules) are configured in the water distribution tank. The peristaltic pump controls the compression frequency of the hose through a high-precision stepper motor to achieve precise dosing of carbon source liquid (such as sodium acetate solution) (rate range: 0–20 mL / min). The starch composite granules are decomposed and slowly release dissolved organic carbon (DOC) to provide the basic carbon source. The dosing amount is dynamically adjusted based on an AI model.

[0122] As a specific embodiment, the AI ​​decision-making layer has autonomous training capabilities, generating initial control instructions based on LSTM prediction results; the DRL engine optimizes instruction parameters by combining historical data and real-time feedback; and the execution layer synchronously adjusts the tidal cycle, aeration intensity, and carbon source dosage. During system operation, deviations between predicted and actual values ​​trigger online model retraining, while the experience playback mechanism continuously optimizes the strategy, forming a closed-loop iteration of "data-driven prediction → prediction-guided decision-making → decision feedback to correct prediction," enabling the system to have forward-looking control (such as predicting high loads and adjusting parameters in advance), dynamic adaptability (such as reducing aeration when there is sufficient sunlight), and self-optimization capabilities.

[0123] The system employed in this embodiment utilizes advanced deep learning and deep reinforcement learning technologies, combined with multi-source sensor data, to provide an intelligent, automated, efficient, and energy-saving method for water quality management and optimization. Through precise prediction and dynamic adjustment, it can effectively improve denitrification efficiency, reduce carbon sources and energy consumption, and ensure the efficient operation of the water treatment system and the achievement of environmental protection goals. By deeply coupling the intelligent sensing layer, decision optimization layer, and execution reinforcement layer, a smart denitrification paradigm of "prediction-decision-execution-feedback" is constructed. Compared with traditional processes, while maintaining effluent TN levels consistently superior to the Class IV surface water standard, it reduces the cost per ton of water treated and decreases the carbon footprint, providing a disruptive and innovative solution for the deep purification of effluent from urban wastewater treatment plants, demonstrating significant environmental and economic benefits.

[0124] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. An AI-based tidal flow artificial wetland automatic adjustment control method, characterized in that, The method comprises the following steps: Real-time acquisition of multi-source sensing data of a tidal flow constructed wetland system; Preprocessing and feature extraction of the multi-source sensing data to obtain key driving factors; Based on the LSTM network, a denitrification efficiency prediction model is constructed, and according to the historical operation data and the real-time acquisition of the key driving factors, the TN removal rate prediction value and the optimal DO threshold value are determined; According to the TN removal rate prediction value and the optimal DO threshold value, a joint reward function is established based on the DRL algorithm to maximize the denitrification efficiency, minimize the carbon source dosage and minimize the energy consumption, determine the multi-objective optimization strategy, and dynamically generate the adjustment parameters of the tidal flow constructed wetland system; According to the adjustment parameters, the water inflow / outflow frequency is adjusted through the electric gate controller, the carbon source is accurately dosed through the peristaltic pump and the slow-release carbon source, and the aeration intensity is adjusted according to the DO concentration prediction value by gradient; Obtain microbial activity feedback data, and optimize the regulation and control strategy according to the microbial activity feedback data, so that the effluent TN reaches the preset standard. 2.The AI-based tidal flow constructed wetland automatic adjustment control method of claim 1, wherein, The multi-source sensing data includes water quality sensing data, environmental perception data and biofilm activity monitoring data; The water quality parameters include NH4 + -N, NO3 - -N, DO, pH, COD, temperature, flow rate and water level data for characterizing the health status of the water body and the change of pollutants; The environmental perception data includes light intensity, rainfall and wind speed, which are used to represent the influence of external climate environment on water quality change; The biofilm activity monitoring data includes denitrifying bacteria metabolic activity and nitrification / denitrification functional gene abundance obtained by dynamic monitoring, which are used to represent the influence of microbial community in water body on water quality. 3.The AI-based tidal flow constructed wetland automatic adjustment control method of claim 1, wherein, The preprocessing and feature extraction of the multi-source sensing data obtain the key driving factors, including: Normalizing the multi-source sensing data and extracting candidate adjustment control features; Using principal component analysis and mutual information screening method to construct a denitrification efficiency driving factor matrix based on the candidate adjustment control features to obtain the key driving factors affecting the denitrification efficiency. 4.The AI-based tidal flow constructed wetland automatic adjustment control method of claim 2, wherein, Based on the LSTM network, a denitrification efficiency prediction model is constructed, and according to the historical operation data and the real-time acquisition of the key driving factors, the TN removal rate prediction value and the optimal DO threshold value are determined, including: Taking the water quality sensing data, the environmental perception data and the biofilm activity monitoring data as the input of the LSTM network; Through a preset time window, a time series input is constructed, a bidirectional LSTM is used to capture the time sequence dependence before and after, and the TN removal rate prediction value and the DO threshold value in the future preset period are output. 5.The AI-based automatic adjustment control method of tidal flow constructed wetland according to claim 4, characterized in that, According to the TN removal rate prediction value and the optimal DO threshold value, a joint reward function is established based on the DRL algorithm to maximize the denitrification efficiency, minimize the carbon source dosage and minimize the energy consumption, determine the multi-objective optimization strategy, and dynamically generate the adjustment parameters of the tidal flow constructed wetland system, including: Taking the TN removal rate prediction value and the optimal DO threshold value output by the LSTM network as the extended state, and combining with the real-time parameters to form a multi-dimensional state space of DRL; Taking the tidal cycle adjustment rate, the aeration intensity grading and the carbon source dosing rate as the action space; Deeply coupling the multi-dimensional state space, the action space and the joint reward function of DRL to form a closed-loop optimization mechanism. 6.The AI-based automatic adjustment control method of tidal flow constructed wetland according to claim 5, wherein, The reward function is determined according to the joint optimization of denitrification efficiency, carbon source consumption and aeration energy consumption, and the strategy is updated iteratively through Q-learning algorithm.

7. A tidal flow constructed wetland system characterised in that, The system comprises a sensing layer, an AI decision layer, an execution layer, and a wetland function layer connected in sequence. The sensing layer is configured to collect multi-source sensing data of the tidal flow constructed wetland system in real time. The AI decision layer is configured to preprocess and extract features from the multi-source sensing data to obtain key driving factors, construct a denitrification efficiency prediction model based on an LSTM network, determine a TN removal rate prediction value and an optimal DO threshold value based on historical operation data and the real-time collected key driving factors, and determine a multi-objective optimization strategy based on a DRL algorithm to maximize denitrification efficiency, minimize carbon source dosage, and minimize energy consumption as a joint reward function, and dynamically generate adjustment parameters of the tidal flow constructed wetland system. The execution layer is configured to adjust the water inflow / outflow frequency through an electric gate controller, accurately dose the carbon source through a peristaltic pump and a slow-release carbon source, and adjust the aeration intensity according to the DO concentration prediction value. The wetland function layer adopts a layered filler optimization layer and a plant root oxygenation structure to construct a multi-path denitrification system.

8. The tidal flow constructed wetland system of claim 7, wherein, The layered filler optimization layer of the wetland function layer comprises an upper aerobic zone and a lower anoxic zone. The upper aerobic zone is filled with volcanic rock and iron-carbon composite material to strengthen the enrichment of nitrifying bacteria and the synergistic effect of chemical oxidation.

9. The tidal flow constructed wetland system of claim 7, wherein, The lower anoxic zone is filled with biochar and sulfur autotrophic denitrification substrate to promote sulfur-driven denitrification and biochar adsorption coupling. The plant root oxygenation structure is achieved by planting emergent plants to provide oxygen. The sensing layer comprises a water quality sensing module, an environment perception module, and a biofilm activity monitoring module. The water quality sensing module comprises a plurality of sensors arranged in the water distribution pool, the layered filler area, and the water collection pool to collect water quality data in real time.

10. The tidal flow constructed wetland system of claim 7, wherein, The environment perception module is configured to collect light intensity, rainfall, and wind speed data. The biofilm activity monitoring module is configured to collect microbial functional gene abundance data based on electrochemical impedance spectroscopy and qPCR technology. The execution layer comprises a tidal gate controller, a gradient aeration system, and a carbon source dual-mode dosing device. The tidal gate controller is configured to adjust and control the water inflow. The gradient aeration system is configured to control the aeration intensity in different regions to optimize the dissolved oxygen level in the water body. The carbon source dual-mode dosing device is configured to dose organic carbon sources to support the microbial nitrogen removal process.

Images