Layered water taking intelligent control system and method based on multi-modal fusion and autonomous evolution

Abstract

The invention relates to a layered water taking intelligent control system and method based on multi-modal fusion and autonomous evolution. Comprising an intelligent water intake tower and an intelligent decision-making unit, the intelligent water intake tower is of a vertical multi-layer structure, and each layer is provided with a movable deformable water intake, a multi-mode sensor array, a water intake valve arranged at the deformable water intake, a deformable water intake optimization module, a data acquisition module of the multi-mode sensor array and a self-energy-supply module driven by green energy. The intelligent decision-making unit comprises a deep reinforcement learning dynamic weight distribution module, a multi-objective optimization solving module, a model prediction control module based on digital twinning, a cross-layer coupling compensation algorithm module, an intelligent compensation mechanism module and an intelligent control execution module. According to the layered water taking intelligent control system and method based on multi-modal fusion and autonomous evolution, parameters such as water temperature, turbidity, dissolved oxygen and pollutant concentration of downstream effluent are controlled, and efficient and accurate control over layered water taking is achieved.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of smart water conservancy and intelligent control, specifically relating to a hierarchical water intake intelligent control system and method based on multimodal fusion and autonomous evolution. Background Technology

[0002] Stratified water intake technology utilizes the stratification of water temperature and quality to regulate the temperature or quality of outflowing water by selecting water sources at different depths, making it an important reservoir management method. However, existing stratified water intake systems tend to be singular in their control dimensions, typically relying on a single parameter such as water temperature or flow rate as the basis for regulation, lacking a systematic consideration of the coupling relationships between multiple parameters. For example, some reservoirs have only a single intake at a fixed elevation, which, while controlling the outflowing water temperature to some extent, lacks simultaneous control measures for other key water quality indicators such as turbidity, dissolved oxygen, and pollutant concentration. This single-indicator-oriented control method struggles to address multiple objectives in complex aquatic environments, often resulting in an inability to optimize water quality across multiple aspects.

[0003] Furthermore, many existing stratified water intake systems rely on pre-set static control models, scheduling based on historical experience or fixed rules, lacking the ability to adaptively adjust to real-time conditions. This static control mode shows significant shortcomings in dynamic scenarios: when the water source environment undergoes sudden changes (such as a sudden upstream pollution event) or some sensors in the system malfunction, existing systems often cannot adjust their water intake strategies in a timely manner, leading to delayed response or control failure, requiring manual intervention. Research indicates that to cope with dynamic fluctuations in water quality, intelligent monitoring and control measures should be adopted to achieve real-time monitoring and adjustment of multiple water quality parameters such as turbidity and dissolved oxygen. However, traditional technologies lack such dynamic adaptive mechanisms, making it difficult to respond promptly to complex and changing water conditions.

[0004] In terms of structural design and operation, existing stratified water intake facilities also have significant shortcomings. Many reservoirs use stacked beam gates for water intake, which require heavy machinery such as cranes to lift and stack the gate panels one by one for opening and closing. This traditional method is cumbersome, often taking a long time to open and close, resulting in a slow system response. Because it is mainly operated manually on-site, opening and closing a single intake can often take several hours, making it difficult to quickly adjust the water intake depth according to environmental changes. As a result, when water temperature, water quality abnormalities, or other emergencies occur, the water intake system cannot respond in time, and its control lags behind environmental changes. Although automation and intelligent technologies have begun to be applied in the field of stratified water intake in recent years, and some new devices can achieve electric control and remote operation, completing the opening and closing of gates within minutes to improve response speed, traditional systems still have a significant gap in flexibility and response speed.

[0005] Furthermore, rigid, fixed water intake structures are prone to clogging and maintenance difficulties. In waters with high sediment content or frequent aquatic activity, intakes often become clogged due to siltation or the attachment of aquatic plants and algae. Low-lying intakes, in particular, are more susceptible to being covered by suspended sediment, hindering water intake. Without effective debris interception and removal measures, floating debris and biological deposits can clog the intake grilles, affecting water intake safety. Meanwhile, in water source projects in remote areas, insufficient power supply is a major limiting factor. Due to the lack of a stable power source, some water intake systems cannot be equipped with continuously operating automated control and cleaning equipment, relying instead on intermittent manual maintenance and limited on-site energy support. This insufficient energy supply further restricts the automation and intelligence of water intake systems, making it difficult for the systems to perform long-term, continuous autonomous monitoring and adjustment.

[0006] Existing stratified water intake systems generally lack the ability to fuse and intelligently analyze multi-source heterogeneous data. In traditional solutions, control decisions often rely on data from a single type of sensor, failing to effectively combine information from different types of sensors and monitoring methods. For example, a water intake control system may only monitor water temperature or flow rate, without integrating the ability to analyze multi-source data such as water quality parameters (e.g., turbidity, pH, dissolved oxygen), flow velocity, meteorological data, and even on-site images. This limitation prevents the system from comprehensively depicting the reservoir environment, making it difficult to accurately predict and determine the optimal water intake level, and restricting the implementation of more advanced intelligent control strategies. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of the aforementioned background technology and provide a stratified water intake intelligent control system and method based on multimodal fusion and autonomous evolution. This achieves multi-objective dynamic optimization control of water quality parameters and efficient and precise control of stratified water intake.

[0008] The technical solution adopted in this invention is: a hierarchical water intake intelligent control system based on multimodal fusion and autonomous evolution, including an intelligent water intake tower and an intelligent decision-making unit;

[0009] The intelligent water intake tower is a vertical multi-layer structure. Each layer is equipped with a movable and deformable water intake, a multi-modal sensor array, a water intake valve located at the deformable water intake, a deformable water intake optimization module, a data acquisition module of the multi-modal sensor array, and a green energy-driven self-powered module.

[0010] The deformable water intake adopts a biomimetic flexible structure; the deformable water intake optimization module is driven by both hydraulic and shape memory alloys to dynamically adjust the shape of the deformable water intake opening; the data acquisition module of the multimodal sensor array controls the multimodal sensor array to collect data in real time, including at least water level, water temperature, flow rate, turbidity, dissolved oxygen, and pollutant concentration, and transmits the collected data to the intelligent decision-making unit; the green energy-driven self-powered module utilizes solar and hydropower to provide the energy required for the operation of the intelligent water intake tower.

[0011] The intelligent decision-making unit includes a deep reinforcement learning dynamic weight allocation module, a multi-objective optimization solution module, a digital twin-based model prediction and control module, a cross-layer coupling compensation algorithm module, an intelligent compensation mechanism module, and an intelligent control execution module.

[0012] The system comprises the following modules: a deep reinforcement learning dynamic weight allocation module, which integrates data from a multimodal sensor array data acquisition module and dynamically adjusts the weights of each parameter based on downstream demand, urgency, and water quality standards; a multi-objective optimization solution module, which uses a constrained quadratic programming algorithm to minimize downstream parameter deviations and calculate the intake valve opening; a digital twin-based model prediction control module, which simulates downstream parameters under different intake valve openings and outputs predicted downstream parameter values ​​to predict future water quality changes and optimize intake valve actions and control strategies for future periods; a cross-layer coupling compensation algorithm module, which adjusts the control strategy based on the coupling effect during water intake from different water layers in the intelligent water intake tower and compensates for the mutual influence between water intake operations at each water layer; an intelligent compensation mechanism module, which dynamically compensates for sudden pollution events reported downstream and adjusts the intake valve openings of each layer in real time; and an intelligent control execution module, which sends control commands to the intelligent water intake tower according to the optimized control strategy to control the intake valve openings of each layer, thereby achieving precise layered water intake and accurate control of downstream water quality parameters.

[0013] A preferred formula for matching the intake opening shape with the flow field is:

[0014] ;

[0015] Constraint: Q(s) ≥ Q min The opening stress σ≤σ SMA ;

[0016] Where s is the opening shape parameter; v is the flow velocity; and M is the turbulence suppression weight matrix; is the friction factor; sedimentation rate (s) is the sedimentation rate when the opening shape parameter is s; Q(s) is the intake flow rate when the opening shape parameter is s; Q minσ is the minimum water intake flow rate; σ is the intake opening stress; σ SMA The allowable stress for shape memory alloys.

[0017] Preferably, the power supply formula for the green energy-driven self-powered module is as follows:

[0018] P total =η solar •A • I sun +η hydro • ρ • Q • g • H;

[0019] Where: P total η represents the total electricity generated from sunlight and water flow. solar A represents the solar energy conversion efficiency; A represents the heat collection area of ​​the outer wall of the intelligent water intake tower; I represents the solar energy conversion efficiency. sun η is the light intensity; hydro ρ is the solar energy conversion efficiency; Q is the water density; H is the flow rate; g is the head height; and g is the gravitational acceleration.

[0020] An energy management algorithm is adopted to prioritize the use of real-time generated energy;

[0021] ;

[0022] in: This represents the amount of electricity required. Power supply for the water intake valve; Power supply to the sensor.

[0023] Preferably, the deep reinforcement learning dynamic weight allocation module combines the dynamic weight allocation module with deep reinforcement learning to construct a state-action-reward model, defining its state space, action space, and reward function, including...

[0024] State space: s t =[T i S i O i C i ΔQ target P consumed ], i = 1, 2, ..., n;

[0025] Among them, s t Let T be the state-space parameter set at time t; i S i O i C i These represent the water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C of the i-th layer, respectively. The value of i is determined by the water level H collected from the water intake. ΔQ target The adjustment range for the target flow rate at the water intake tower outlet; Pconsumed This represents the amount of electricity required.

[0026] Action space: α t =Δw T Δw S Δw O Δw C , satisfying ∑w=1;

[0027] Where, α t The action space parameter set at time t; w T w S w O w C Δw is a weighting coefficient for water temperature (T), turbidity (S), dissolved oxygen (O), and pollutant concentration (C), reflecting the importance of each parameter; T Δw S Δw O Δw C The adjustment range of the weighting coefficients for water temperature T, turbidity S, dissolved oxygen O and pollutant concentration C;

[0028] Reward function: a weighted sum of the rate of reduction in downstream parameter deviation and the rate of reduction in energy consumption.

[0029] r t =−(α1‖y out -y target || 2 +α2∑∣Δu i |+α3P consumed );

[0030] Where, r t Let be the reward function parameter set at time t; α1 is the action space parameter representing the difference between the downstream outlet parameter set (composed of water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C) and the target parameter set; α2 is the action space parameter representing the total change in valve opening; α3 is the action space parameter representing the required electricity consumption; y out =[T out S out O out C out ] T y out This is a parameter set consisting of downstream outlet water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C; y target =[T target S target O target C target ] T y target This is the set of target parameters to be achieved for the downstream outlet water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C; Δu iP represents the change in the opening degree of the water intake valve. consumed This represents the amount of electricity required.

[0031] Preferably, the calculation of the water intake valve opening includes

[0032] Assume the flow rate Q at each water intake point i With respect to the opening degree of the water intake valve u i The relationship is:

[0033] Q i =k i • u i ;

[0034] Where, k i For flow coefficient;

[0035] The downstream parameter is a flow-weighted average of the parameters at each layer:

[0036] ;

[0037] ;

[0038] ;

[0039] ;

[0040] in, , , These are the downstream outlet water temperature (T), turbidity (S), dissolved oxygen (O), and pollutant concentration (C), respectively; T i S i O i C i The water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C of the i-th layer are respectively; Q i Let be the water intake volume of the water intake valve at the i-th layer;

[0041] The goal is to minimize the deviation of downstream parameters from the target value while satisfying the total flow constraint.

[0042] Objective function:

[0043] ;

[0044] Constraints:

[0045] Total flow constraint: ;

[0046] Water intake valve opening limit: 0≤u i ≤1 (i=1,2,…,n);

[0047] Where: w T wS w O w C The weighting coefficients for water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C; , , These are the downstream outlet water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C, respectively. , , , These are the target values ​​for water temperature (T), turbidity (S), dissolved oxygen (O), and pollutant concentration (C), respectively; Q i Q represents the water intake volume of the water intake valve at the i-th layer; target The target water intake for downstream export;

[0048] Quadratic programming: Expand the objective function into a quadratic form and solve it using a solver;

[0049] Objective function standardization: ;

[0050] in: u i Let H be the opening degree of the water intake valve at the i-th layer; H is the Hessian matrix, determined by the weights of each parameter w. T w S w O w C and T i S i O i C i The composition is as follows: f is a linear term, which is related to the deviation between the target value and the current parameter.

[0051] A preferred approach to compensate for the mutual influence between water intake operations in different water layers includes establishing an interlayer hydraulic interference model:

[0052] ;

[0053] in, k represents the actual water intake volume of the water intake valve at level i; i U is the flow coefficient; i β represents the opening degree of the water intake valve at the i-th layer; ij The interlayer interference coefficient is the effect of the opening degree of the water intake valve on the flow velocity of other layers; u ij The coupling degree between the opening degree of the water intake valve of a certain floor and the opening degree of the water intake valve of other floors; This represents the probability of this situation occurring.

[0054] A preferred approach is to dynamically compensate for sudden pollution events reported from downstream sources using the following formula:

[0055] ;

[0056] in, The opening degree of the water intake valve at the i-th layer after feedback correction from the downstream sensor; The opening degree of the water intake valve at the i-th layer before feedback correction from the downstream sensor; C is the conversion coefficient between pollutant concentration and water intake valve opening. target The target pollutant concentration to be achieved; C measured This represents the actual measured concentration of pollutants.

[0057] This invention also provides a stratified water intake intelligent control method based on multimodal fusion and autonomous evolution, including...

[0058] The system collects real-time data including at least water level, water temperature, flow velocity, turbidity, dissolved oxygen, and pollutant concentration. It integrates this data and dynamically adjusts the weights of each parameter based on downstream demand, urgency, and water quality standards. The control strategy is then adaptively adjusted according to these weights. A constrained quadratic programming algorithm is used to calculate the intake valve opening, aiming to minimize downstream parameter deviations. Downstream parameters under different intake valve openings are simulated, and predicted downstream parameter values ​​are output to predict future water quality changes and optimize intake valve actions and control strategies for future periods. The control strategy is adjusted based on the coupling effect during water intake from different water layers in the intelligent water intake tower, compensating for the mutual influence between water intake operations at each layer. Dynamic compensation is provided for sudden pollution events reported downstream, and the intake valve openings at each layer are adjusted in real-time. Control commands are sent to the intelligent water intake tower based on the optimized control strategy to control the intake valve openings at each layer, achieving precise stratified water intake and accurate control of downstream water quality parameters.

[0059] Compared with the prior art, the beneficial effects of the present invention are:

[0060] 1. Deformable Intake Structure and Adaptive Flow Field Matching. An innovative deformable intake structure using biomimetic flexible materials and shape optimization design is proposed. The opening shape is changed (e.g., from circular to slit-shaped) via micro-hydraulic drive to adapt to different flow velocities and turbidity conditions. Based on flow field simulation results, the intake morphology is dynamically adjusted to reduce turbulence. Traditional rigid intakes cannot adapt to changes in flow regime; this invention improves water intake efficiency and equipment lifespan through "structure-control" synergistic optimization. The flexible structure reduces the risk of aquatic organism inhalation, aligning with eco-friendly design trends. Sediment deposition is reduced by 35% at the same flow rate, extending the maintenance cycle to over 2 years.

[0061] 2. Dynamic weight allocation and adaptive mechanisms based on Deep Reinforcement Learning (DRL). Weight coefficient w T w S w O w CThe system is dynamically adjusted based on the real-time water quality exceedance level. For example, if C... out >C 安全阈值 Then w C Automatically boosting to the maximum value prioritizes pollutant suppression. Weight adjustment rules can be embedded in fuzzy logic or a rule engine. This enables intelligent switching from "fixed priority" to "emergency response." Traditional methods rely on manual rules or fuzzy logic, while DRL dynamically generates optimal weight strategies through self-learning historical data and real-time feedback. This invention solves the problem of incomplete rule base coverage in sudden pollution events, such as automatically identifying the control priority of emerging pollutants. Compared to traditional static rules, deep reinforcement learning reduces control errors by more than 40% in complex scenarios (simulation data).

[0062] 3. Multimodal Sensor Data Fusion and Anomaly Detection. A multimodal neural network is employed to fuse time-series data and spatial distribution characteristics of water temperature, turbidity, and dissolved oxygen. This multimodal data is used for dynamic optimization control of water quality parameters. Water temperature, turbidity, dissolved oxygen, and pollutant concentration are incorporated into a unified optimization model. Multi-parameter coupled control is achieved through a flow-weighted formula. An anomaly detection model and edge computing are introduced to improve reliability, addressing different application scenarios and technical objectives. This approach breaks through the traditional single-parameter control mode, solving the challenge of complex water quality coordinated regulation. A lightweight anomaly detection model is designed to identify sensor faults or data conflicts in real time (such as logical contradictions between dissolved oxygen and pollutant concentrations in a certain layer). Traditional methods only perform threshold judgments on single-dimensional data; this invention improves data reliability through multimodal fusion, avoiding erroneous regulation due to sensor faults. The anomaly detection model is embedded in edge computing devices, achieving millisecond-level fault alarms and data repair. The accuracy of data anomaly identification is improved to 99.2%, and the error rate is reduced by 60%.

[0063] 4. Hierarchical Collaborative Optimization Algorithm and Intelligent Control. The optimization problem of water intake valve opening is transformed into a constrained quadratic programming problem, and the global optimal solution is calculated in real time to ensure coordinated operation of multiple water intake valves. This avoids parameter oscillations caused by local optimization and improves system stability. It focuses on multi-parameter coupled control (water temperature, turbidity, dissolved oxygen, and pollutants) and designs deformable water intakes to adapt to changes in the flow field, thus expanding the technical dimensions.

[0064] 5. A stratified water intake virtual mapping system based on digital twins. A high-fidelity digital twin model is constructed, integrating fluid dynamics simulation (CFD) and water quality diffusion models to simulate downstream parameters under different intake valve openings in real time. The digital twin system interacts bidirectionally with the physical system: physical sensor data calibrates the virtual model, and virtual prediction results guide the optimization algorithm. Breaking away from the traditional "physical trial-and-error" model, this system avoids risky operations in the actual system (such as frequent start-stop of intake valves) by rehearsing control strategies through virtual simulation. It supports long-term strategy optimization, such as predicting seasonal stratification patterns based on historical data to generate contingency plans in advance. The speed of control strategy generation is increased by 5 times, and long-term energy consumption is reduced by 18%.

[0065] 6. Intelligent Decision-Making and Compensation Mechanisms. Intelligent decision-making is achieved through deep reinforcement learning; the deformable intake structure adapts to complex flow fields; and the self-powered design overcomes deployment limitations. Downstream sensor feedback is introduced to correct model prediction errors, enhancing the system's anti-interference capabilities and adapting to sudden pollution events. A digital twin-driven simulation model provides high-precision pre-simulation, and physical system feedback calibrates the model, forming an autonomous evolution capability.

[0066] 7. Cross-layer Coupling Compensation Algorithm: This algorithm quantifies the inter-layer interference effect of stratified water intake for the first time, breaking through the traditional assumption of independent regulation. An inter-layer hydraulic interference model is established to quantify the impact of a water intake valve's operation on the flow velocity of other layers (e.g., an increase in the opening of the middle layer leads to a decrease in pressure at the bottom layer). A coupling compensation term is introduced to correct the predicted flow rate. This invention incorporates inter-layer coupling into the control model, avoiding the negative feedback loop of "opening optimization - actual deviation". Under complex operating conditions, downstream parameter stability is improved by 50%, and the number of water intake valve operations is reduced by 30%.

[0067] 8. Multi-scale, second-level optimization and rapid response: Real-time control is achieved through electric water intake valve adjustment, policy updates are achieved through deep reinforcement learning, and long-term pattern learning is achieved through digital twin historical data analysis.

[0068] 9. Digital Twin and Artificial Intelligence Applications. This invention further combines deep reinforcement learning dynamic weight allocation and cross-layer coupling compensation algorithms, and integrates a flexible water intake structure to form a "virtual-real fusion closed-loop control" system, significantly increasing the technical complexity.

[0069] 10. Green Energy-Driven Self-Powered Water Intake System. A flexible solar film and micro-hydrogen generator are integrated into the outer wall of the water intake tower, utilizing water flow energy and sunlight to power sensors and intake valves. An energy management algorithm is designed to prioritize the use of real-time generated energy, solving the power supply problem for water intake facilities in remote areas and achieving carbon-neutral operation of the entire system. The energy algorithm supports a "low-power mode," automatically reducing the sampling frequency of non-critical sensors on cloudy days / at night. External grid dependence is reduced to below 10%, making it suitable for areas without grid coverage.

[0070] This invention achieves innovative hierarchical leaps: ① From "single-objective control" to "multimodal game theory": Through DRL and digital twins, global optimization with multiple parameters, constraints, and time scales is realized. ② From "independent adjustment" to "coupled compensation": For the first time, inter-layer hydraulic interference is quantified, breaking through the theoretical assumptions of traditional stratified water intake. ③ From "rigid system" to "bionic intelligence": Flexible water intake and self-powered design promote the upgrading of water conservancy facilities towards ecological and sustainable development. ④ From "manual operation and maintenance" to "autonomous evolution": Anomaly detection, parameter identification, and energy management constitute a closed-loop autonomous system, reducing the need for manual intervention.

[0071] In summary, the present invention has the following characteristics:

[0072] 1. Precise control: This invention uses deep reinforcement learning and digital twin pre-simulation to reduce the deviation of downstream parameters of water intake from reservoirs by 30% to 50%, meeting the requirements for high-precision water quality.

[0073] 2. Rapid Response: The speed of generating control strategies is increased by 5 times, and the control time for sudden pollution events is shortened to within 5 minutes.

[0074] 3. Energy saving and consumption reduction: By optimizing the coordinated operation of water intake valves, energy consumption is reduced by more than 15%. Dependence on the external power grid is reduced to below 10%, making it suitable for areas without power grid coverage.

[0075] 4. Improve water intake quality: Through deformable water intake and dynamic sensing adjustment, the accuracy of anomaly detection is increased to 99.2% and the sediment deposition of flexible structure is reduced by 35%.

[0076] This invention utilizes a multimodal fusion and autonomous evolution-based intelligent control system and method for stratified water intake to intelligently control parameters such as water temperature, flow rate, turbidity, dissolved oxygen, and pollutant concentration in downstream effluent. Through multi-parameter fusion, dynamic optimization, and intelligent compensation mechanisms, it achieves efficient and precise control of stratified water intake. This invention can be widely applied in fields such as water environment management, industrial water supply, smart water management, and ecological protection, and has greater scalability. Attached Figure Description

[0077] Figure 1 A schematic diagram of a layered water intake intelligent control system based on multimodal fusion and autonomous evolution;

[0078] Figure 2 Flowchart of intelligent water intake tower control system and method;

[0079] Figure 3 This is a flowchart of a layered water intake intelligent control system and method based on multimodal fusion and autonomous evolution. Detailed Implementation

[0080] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that these descriptions are for the purpose of aiding understanding the present invention, but do not constitute a limitation thereof. Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0081] like Figure 1 As shown, this invention provides a hierarchical intelligent water intake control system based on multimodal fusion and autonomous evolution, which consists of two main parts: an intelligent water intake tower and an intelligent decision-making unit. Specifically, it includes a deformable water intake optimization module, a data acquisition module integrating a multimodal sensor array, a green energy-driven self-powered water intake system module, a deep reinforcement learning (DRL) dynamic weight allocation module, a multi-objective optimization solution module, a digital twin-based model prediction control module, a cross-layer coupling compensation algorithm module, an intelligent compensation mechanism module, and an intelligent control execution module.

[0082] like Figure 2 As shown, the intelligent water intake tower has a multi-layer vertical structure. Each layer is equipped with a movable and deformable water intake, a multi-modal sensor array (collecting data such as water level H, water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C), and a water intake valve installed at the deformable water intake. The intelligent water intake tower also includes a deformable water intake optimization module, a data acquisition module integrating a multi-modal sensor array, and a green energy-driven self-powered water intake system module.

[0083] The deformable water intake optimization module uses a combination of shape memory alloy (SMA) and hydraulic drive for dynamic adjustment of the opening shape. The main frame consists of the water intake shell, deformable movable blades, and hydraulic drive components.

[0084] The water intake shell integrates sensors to collect parameters such as water level H, water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C around the water intake.

[0085] Deformable moving blades: 3-4 sets of retractable / bendable blades are embedded inside the water intake, and shape memory alloy (SMA) wires are embedded in the core of the blades.

[0086] The hydraulic drive assembly consists of two parts: ① hydraulic cylinders, with each set of blades connected to an independent miniature hydraulic cylinder; ② hydraulic lines, which connect the hydraulic cylinders, control unit, and hydraulic valves. The collected data can be input into the control unit for calculation, and the results can be fed back to the hydraulic valves to control the opening shape.

[0087] Sensors around the water intake collect parameters such as water level H, water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C. These parameters are calculated by the central processing unit (CPU) and output to control the shape of the water intake opening via shape memory alloy (SMA) current and hydraulic pressure. This controls the opening shape to reduce turbulence around the intake, minimize the risk of aquatic organism inhalation, and reduce downstream sediment deposition. Simultaneously, the collected water level H data is used to determine the intake location, specifically the value of the i-th layer intake.

[0088] This structure achieves deformable intake design and adaptive flow field matching. The intake is designed with biomimetic flexible materials and its shape can be changed (e.g., from circular to slit-like) via micro-hydraulic drive to adapt to different flow velocities and turbidity conditions. Based on flow field simulation results, the intake morphology is dynamically adjusted to reduce turbulence. Through "structure-control" co-optimization, the risk of aquatic organism inhalation is reduced, sediment deposition is reduced by 35% at the same flow rate, water intake efficiency and equipment lifespan are improved, and the maintenance cycle is extended to more than 2 years.

[0089] Intake opening shape - flow field matching formula: The optimization objective is the weighted sum of turbulence intensity and sedimentation rate.

[0090] ;

[0091] Constraints:

[0092] Q(s)≥Q min The opening stress σ≤σ SMA ;

[0093] Where s is the opening shape parameter, v is the flow velocity, and M is the turbulence suppression weight matrix. This is the friction coefficient. Let Q(s) be the sedimentation rate when the opening shape parameter is s, and Q(s) be the intake flow rate when the opening shape parameter is s. min Let σ be the minimum water intake flow rate, and σ be the intake opening stress. SMA The allowable stress for shape memory alloys (SMA).

[0094] The integrated multimodal sensor array data acquisition module enables multimodal sensor data fusion and anomaly detection, acquiring real-time data from each sensor layer, including parameters such as water level (H), water temperature (T), flow velocity (v), turbidity (S), dissolved oxygen (O), and pollutant concentration (C) (detected via a miniature mass spectrometer). A multimodal neural network is employed to fuse time-series data and spatial distribution characteristics of water temperature, turbidity, and dissolved oxygen, and a lightweight anomaly detection model is used to identify sensor malfunctions or data conflicts (such as logical contradictions between dissolved oxygen and pollutant concentration in a certain layer) in real time, eliminating outliers.

[0095] Multimodal fusion enhances data reliability and prevents erroneous control due to sensor malfunctions. Anomaly detection models are embedded in edge computing devices to achieve millisecond-level fault alarms and data repair.

[0096] The green energy-driven self-powered water intake system module integrates a flexible solar film and a micro-hydrogen generator on the outer wall of the water intake tower, utilizing sunlight and water flow energy to power sensors and water intake valves. Power supply formula:

[0097] P total =η solar •A • I sun +η hydro • ρ • Q • g • H;

[0098] Where: P total η represents the total electricity generated from sunlight and water flow. solar A represents the solar energy conversion efficiency; A represents the heat collection area of ​​the outer wall of the intelligent water intake tower; I represents the solar energy conversion efficiency. sun η is the light intensity; hydro ρ is the solar energy conversion efficiency; Q is the water density; H is the flow rate; g is the head height; and g is the gravitational acceleration.

[0099] An energy management algorithm is adopted to prioritize the use of real-time generated energy;

[0100] ;

[0101] in: This represents the amount of electricity required. Power supply for the water intake valve; Power supply to the sensor.

[0102] During the day when there is sufficient sunlight, the normal operation mode meets the energy consumption constraints and is used. When there is insufficient light at night, the self-powered system switches to a "low-power energy-saving mode": the sampling frequency of non-critical sensors is reduced (e.g., from 1Hz to 0.2Hz); the deep reinforcement learning engine is switched to a lightweight network; and the model prediction control module based on digital twins adopts a reduced-order model (ROM) to accelerate simulation, reducing the solution time by 60%.

[0103] like Figure 3 As shown, the Deep Reinforcement Learning (DRL) dynamic weight allocation module combines the dynamic weight allocation module with deep reinforcement learning to construct a state-action-reward model and adjust the objective function weight coefficients according to the exceedance level.

[0104] The training process involves pre-training a DRL policy network using historical data. During the online phase, the policy is updated every 5 minutes and interacts with the environment (digital twin + physical system). Its reward function includes downstream parameter deviations, energy consumption of the water intake valve operation, and power supply status. Parameter weights are dynamically adjusted based on downstream demand and urgency, as well as water quality standards (e.g., prioritizing pollutant control and increasing w when pollutants exceed standards). C This enables adaptive weighting.

[0105] State space: sensor data from each layer, downstream parameter deviations, and historical control records.

[0106] s t =[T i S i O i C i ΔQ target P consumed ], (i=1, 2, ..., n);

[0107] Among them, s t Let T be the state-space parameter set at time t; i S i O i C i These represent the water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C of the i-th layer, respectively. The value of i is determined by the water level H collected from the water intake. ΔQ target The adjustment range for the target flow rate at the water intake tower outlet; P consumed This represents the amount of electricity required.

[0108] Action space: weighting coefficient w T w S w O w C The adjustment range.

[0109] α t =Δw T Δw S Δw O Δw C (Satisfies ∑w=1);

[0110] Where, α t The action space parameter set at time t; w T w S w O w C Δw is a weighting coefficient for water temperature (T), turbidity (S), dissolved oxygen (O), and pollutant concentration (C), reflecting the importance of each parameter; T Δw S Δw O Δw CThe adjustment range of the weighting coefficients for water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C.

[0111] Reward function: a weighted sum of the rate of reduction in downstream parameter deviation and the rate of reduction in energy consumption.

[0112] r t =−(α1‖y out -y target || 2 +α2∑∣Δu i |+α3P consumed );

[0113] Among them, r t Let be the reward function parameter set at time t; α1 is the action space parameter representing the difference between the downstream outlet parameter set (composed of water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C) and the target parameter set; α2 is the action space parameter representing the total change in valve opening; α3 is the action space parameter representing the required electricity consumption; y out =[T out S out O out C out ] T y out This is a parameter set consisting of downstream outlet water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C; y target =[T target S target O target C target ] T y target This is the set of target parameters to be achieved for the downstream outlet water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C; Δu i P represents the change in the opening degree of the water intake valve. consumed y represents the amount of electricity required; 安全阈值 This is a set of safe threshold parameters for water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C.

[0114] Weighting coefficient w T w S w O w C The system is dynamically adjusted based on the real-time water quality exceedance level. For example, if C... out >C 安全阈值 That is, if the pollutant concentration C at the downstream outlet is greater than the safe value of pollutant concentration C, then w C Automatically boosts to the maximum value, prioritizing pollutant suppression. Weight adjustment rules can be embedded with fuzzy logic or a rule engine. Enables intelligent switching from "fixed priority" to "emergency response".

[0115] This module addresses the issue of incomplete rule base coverage in sudden pollution events, such as automatically identifying the control priority of emerging pollutants. Compared to static rules, DRL reduces control errors by more than 40% in complex scenarios (simulation data).

[0116] The multi-objective optimization solution module uses a constrained quadratic programming algorithm to minimize the downstream parameter deviation and calls the solver to calculate the opening degree of the water intake valve.

[0117] ① Flow model: Assume the flow rate Q at each water intake is... i With respect to the opening degree of the water intake valve u i The relationship is:

[0118] Q i =k i • u i ;

[0119] Where, k i The flow coefficient (related to the characteristics of the water intake valve and the water pressure ΔP) i (related), can be simplified to a constant or through Dynamic calculation, C v Let ΔP be the flow coefficient. i This refers to water pressure. i Let represent the opening degree of the water intake valve for the i-th layer.

[0120] ② Downstream parameter model

[0121] The downstream parameter is a flow-weighted average of the parameters at each layer:

[0122] ;

[0123] ;

[0124] ;

[0125] ;

[0126] in, , , These are the downstream outlet water temperature (T), turbidity (S), dissolved oxygen (O), and pollutant concentration (C), respectively; T i S i O i C i The water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C of the i-th layer are respectively; Q i Let be the water intake volume of the water intake valve at the i-th layer.

[0127] ③ Optimization problem: The goal is to minimize the deviation between downstream parameters and target values ​​while satisfying the total flow constraint.

[0128] Objective function: ;

[0129] Constraints:

[0130] Total flow constraint: ;

[0131] Water intake valve opening limit: 0≤u i ≤1 (i=1,2,…,n);

[0132] Where: w T w S w O w C The weighting coefficients for water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C; , , These are the downstream outlet water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C, respectively. , , , These are the target values ​​for water temperature (T), turbidity (S), dissolved oxygen (O), and pollutant concentration (C), respectively; Q i Q represents the water intake volume of the water intake valve at the i-th layer; target The water intake is for the downstream export target.

[0133] ④ Solution algorithm:

[0134] Quadratic programming: Expand the objective function into a quadratic form and solve it using a solver.

[0135] Objective function standardization: ;

[0136] in: u i Let H be the opening degree of the water intake valve at the i-th layer; H is the Hessian matrix, determined by the weights of each parameter w. T w S w O w C and T i S i O i C i The composition is as follows: f is a linear term, which is related to the deviation between the target value and the current parameter.

[0137] A model predictive control module based on digital twins (a digital twin-driven CFD simulation model) constructs a high-fidelity digital twin model, integrating a CFD solver and a water quality diffusion model. It performs online parameter identification and real-time simulation of downstream parameters under different intake valve openings to predict the effectiveness of control strategies. The simulation process involves generating a flow field mesh based on the current intake valve opening, solving the fluid dynamics and water quality equations in parallel, and outputting predicted values ​​y for downstream water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C. pred It predicts future water quality changes, is used to optimize the generation of the initial solution of the algorithm, continuously optimizes the action of the water intake valve and the control strategy for future periods, and provides timely feedback and correction to correct model errors.

[0138] The digital twin system interacts bidirectionally with the physical system: physical sensor data calibrates the virtual model, and virtual prediction results guide optimization algorithms. Virtual simulation allows for the rehearsal of control strategies, mitigating risky operations in the actual system (such as frequent start-stop of water intake valves). It supports long-term strategy optimization, for example, by combining historical data to predict seasonal stratification patterns and generate contingency plans in advance.

[0139] The cross-layer coupling compensation algorithm module establishes an inter-layer hydraulic interference model, quantifies the impact of the action of a water intake valve in one layer on the flow velocity of other layers (e.g., an increase in the opening of the middle layer will lead to a decrease in the pressure of the bottom layer), and calculates the inter-layer interference coefficient β in real time. ij A coupling compensation term is introduced to correct the flow forecast value.

[0140] Interlayer hydraulic interference model:

[0141] ;

[0142] in, k represents the actual water intake volume of the water intake valve at level i; i The flow coefficient (related to the characteristics of the water intake valve and the water pressure ΔP) i (related), can be simplified to a constant or through Dynamic calculation, C v Let ΔP be the flow coefficient. i For water pressure; u i β represents the opening degree of the water intake valve at the i-th layer; ij The interlayer interference coefficient is the effect of the opening degree of the water intake valve on the flow velocity of other layers; u ij The coupling degree between the opening degree of the water intake valve of a certain floor and the opening degree of the water intake valve of other floors; This represents the probability of this situation occurring.

[0143] This invention incorporates interlayer coupling into the control model, avoiding the negative feedback loop of "opening optimization - actual deviation". The coupling coefficient β is dynamically updated through online parameter identification. ij .

[0144] The intelligent compensation mechanism module incorporates feedback from downstream sensors to correct prediction errors, thereby enhancing the system's anti-interference capabilities and adapting to sudden pollution events.

[0145] ;

[0146] in, The opening degree of the water intake valve at the i-th layer after feedback correction from the downstream sensor; The opening degree of the water intake valve at the i-th layer before feedback correction from the downstream sensor; C is the conversion coefficient between pollutant concentration and water intake valve opening. target The target pollutant concentration to be achieved; C measured This represents the actual measured concentration of pollutants.

[0147] The intelligent control execution module consists of a water intake valve drive device, supporting adjustments to opening degree, angle, and position. It adjusts the water intake valve to the target opening degree and monitors downstream parameter feedback. It can dynamically optimize the opening degree of water intake valves at each level, precisely controlling downstream water quality parameters.

[0148] This invention also provides a control method for a stratified water intake intelligent control system based on multimodal fusion and autonomous evolution, comprising the following steps:

[0149] S1: Complete the initial configuration of the intelligent water intake tower, output the shape of the deformable water intake opening, the processed sensor data, and intelligently determine the power consumption mode of the self-powered water intake system.

[0150] S1.1: The deformable intake optimization module operates, with sensors distributed on the outer wall and rim of the intake collecting parameters such as water level, temperature, flow velocity, and turbidity around the intake. The central controller of the deformable intake operates, calculating the opening shape based on the collected data. Through a combination of shape memory alloy and hydraulic drive, the opening shape is dynamically adjusted to adapt to different flow velocities and turbidity conditions and reduce turbulence. This "structure-control" collaborative optimization reduces the risk of aquatic organism inhalation, reduces sediment deposition by 35% at the same flow rate, improves water intake efficiency and equipment lifespan, and extends the maintenance cycle to over 2 years.

[0151] S1.2: The data acquisition module integrating the multimodal sensor array operates. It acquires real-time data from each layer of sensors, including parameters such as water temperature (T), turbidity (S), dissolved oxygen (O), and pollutant concentration (C). A multimodal neural network is used to fuse the temporal data and spatial distribution characteristics of water temperature (T), turbidity (S), dissolved oxygen (O), and pollutant concentration (C). A lightweight anomaly detection model embedded in an edge computing device is used to identify sensor faults or data conflicts in real-time (millisecond level) (such as logical contradictions between dissolved oxygen and pollutant concentration in a certain layer), and outliers are removed. The filtered and processed data is output, including parameters such as water temperature (T), turbidity (S), dissolved oxygen (O), and pollutant concentration (C). Multimodal fusion improves data reliability and avoids erroneous adjustments due to sensor faults. The anomaly detection model embedded in the edge computing device enables millisecond-level fault alarms and data repair.

[0152] S1.3: The self-powered water intake system module, driven by green energy, is in operation. A flexible solar film and a micro-hydrogen generator are integrated into the outer wall of the water intake tower to collect solar energy and water flow energy. The self-powered water intake system module operates by using water flow energy and sunlight to power sensors and water intake valves. An energy management algorithm, namely P... total =η solar •A• I sun +η hydro •ρ•Q•g•H, prioritize the use of real-time power generation. When energy constraints are met. When normal power consumption is not met, the system operates in normal mode; when energy constraints are not met, it switches to low-power energy-saving mode. For example, during the day when there is sufficient sunlight, the normal operation mode meets the energy consumption constraints and operates in normal mode. At night when there is insufficient sunlight, the self-powered system switches to "low-power energy-saving mode": reducing the sampling frequency of non-critical sensors (e.g., from 1Hz to 0.2Hz); switching the deep reinforcement learning engine to a lightweight network; and using a reduced-order model (ROM) to accelerate simulation in the digital twin-based model prediction control module, reducing the solution time by 60% to save energy.

[0153] S2: The Deep Reinforcement Learning (DRL) dynamic weight allocation module runs. It constructs a state-action-reward model and adjusts the objective function weight coefficients w based on the exceedance level. T w S w O w C The training process involves pre-training a DRL policy network using historical data. During the online phase, the policy is updated every 5 minutes and interacts with the environment (digital twin + physical system). Its reward function includes downstream parameter deviations, energy consumption of the water intake valve, and power supply status. Parameter weights are dynamically adjusted based on downstream demand and urgency, as well as water quality standards (e.g., prioritizing pollutant control and increasing w when pollutants exceed standards). C This enables adaptive weighting.

[0154] Beneficial effect: Weighting coefficient w T w S w O w C The system is dynamically adjusted based on the real-time water quality exceedance level. For example, if C... out >C 安全阈值 Then w C Automatically boosts to the maximum value, prioritizing pollutant suppression. Weight adjustment rules can be embedded with fuzzy logic or a rule engine. Enables intelligent switching from "fixed priority" to "emergency response." This module addresses the issue of incomplete rule base coverage in sudden pollution events, such as automatically identifying the control priority of emerging pollutants. Compared to static rules, DRL reduces control errors by more than 40% in complex scenarios (simulation data).

[0155] S3: Multi-objective optimization solution module operation. Using the Deep Reinforcement Learning (DRL) dynamic weight allocation module to solve for the weight coefficients w. T w S w O w C The optimization objective is to minimize the deviation between downstream parameters and target values. The constraints are the total flow rate and the opening limit of the water intake valve. A constrained quadratic programming algorithm is used to calculate the opening of the water intake valve at each level and accurately control the downstream water quality parameters.

[0156] S4: The model predictive control module based on digital twins is in operation. A high-fidelity digital twin model is constructed, integrating a CFD solver and a water diffusion model to simulate downstream parameters under different intake valve openings in real time, outputting predicted downstream parameter values ​​y. pred It predicts future water quality changes. It continuously optimizes the operation and control strategies of water intake valves for future periods, providing timely feedback and correction to rectify errors in the high-fidelity digital twin model. Through virtual simulation, it rehearses control strategies to avoid risky operations in the actual system (such as frequent start-stop of water intake valves). It supports long-term strategy optimization, such as predicting seasonal stratification patterns based on historical data to generate contingency plans in advance. The speed of control strategy generation is increased by 5 times, and long-term energy consumption is reduced by 18%.

[0157] S5: The cross-layer coupling compensation algorithm module is running. An inter-layer hydraulic interference model is established, the impact of a water intake valve operation on the flow velocity of other layers is quantified, and the inter-layer interference coefficient β is calculated in real time. ij A coupling compensation term is introduced to correct the flow prediction value. This invention incorporates inter-layer coupling into the control model, avoiding the negative feedback loop of "opening optimization - actual deviation". The coupling coefficient β is dynamically updated through online parameter identification. ij The stability of downstream parameters under complex operating conditions is improved by 50%, and the number of water intake valve operations is reduced by 30%.

[0158] S6: The intelligent compensation mechanism module is in operation. It incorporates feedback from downstream sensors, corrects model prediction errors, and adjusts the opening of water intake valves at each level in real time, thereby enhancing the system's anti-interference capability and adapting to sudden pollution events.

[0159] S7: Intelligent control execution module operation. Composed of an electric valve drive device, it supports adjustment of opening degree, angle, and position. It adjusts the valve to the target opening degree and monitors downstream parameter feedback. It can dynamically optimize the valve opening at each level, precisely controlling downstream water quality parameters.

[0160] Example 1: Layered water intake from reservoir

[0161] (1) Scenario: In summer, the water temperature is clearly stratified, and it is necessary to extract the low-temperature water at the bottom to ensure the ecological flow.

[0162] (2) Control process:

[0163] Step 1: The sensor detects that the surface pollutant concentration exceeds the standard (C1=0.5mg / L), and the dynamic weighting module increases w. C .

[0164] Step 2: The optimization module calculates that the surface valve should be closed (u1=0), and the opening of the middle layer should be increased (u2=0.8).

[0165] Step 3: The model prediction and control module predicts the pollutant diffusion trend in the next 3 hours and adjusts the opening of the water intake valve in advance.

[0166] Example 2: Low-power mode operation

[0167] (1) Scenario: Insufficient light at night, the self-powered system switches to energy-saving mode.

[0168] (2) Control process:

[0169] Step 1: The energy management algorithm reduces the sampling frequency of non-critical sensors (from 1 Hz to 0.2 Hz).

[0170] Step 2: Switch the Deep Reinforcement Learning (DRL) engine to a lightweight network, and compress the input dimension st′ = [Ti, Ci, E]. battery ].

[0171] Step 3: The digital twin system uses a reduced-order model to accelerate simulation, reducing the solution time by 60%.

[0172] Example 3: Emergency Control of Sudden Pollutants

[0173] (1) Scenario: The reservoir water source is suddenly polluted, and the system automatically controls the emergency.

[0174] (2) Control process:

[0175] Step 1: The surface sensor detects pesticide contamination (C1=1.2mg / L), and the deep reinforcement learning (DRL) engine automatically increases the efficiency of the sensor. C Up to 0.7.

[0176] Step 2: The digital twin system simulates different opening combinations to predict whether water intake from the middle layer can dilute pollutants.

[0177] Step 3: Solve the optimization problem with coupling compensation:

[0178] ;

[0179] ;

[0180] Top layer opening u1=0 (contamination layer forcibly closed);

[0181] Step 4: Adjust the middle layer opening to u2=0.85 and the bottom layer opening to u3=1.0. The shape controller will change the middle layer opening to a flat slit shape to reduce turbulence.

[0182] Example 4: Wastewater Influent Distribution for Wastewater Treatment Plant

[0183] (1) Scenario: The concentration of pollutants in the influent fluctuates, and the load of the downstream treatment unit needs to be stabilized.

[0184] (2) Control process:

[0185] Step 1: Based on the turbidity (S) of each layer i ) and dissolved oxygen (O i Generate the objective function.

[0186] Step 2: Solve for the opening degree using the intelligent compensation mechanism module to satisfy Q. target =10m 3 / s.

[0187] Step 3: The downstream pH sensor feedback triggers the compensation mechanism, which finely adjusts the opening of the water intake valve.

[0188] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Contents not described in detail in this specification belong to prior art known to those skilled in the art.

Claims

1. A layered water intake intelligent control system based on multimodal fusion and autonomous evolution, characterized in that: Includes intelligent water intake towers and intelligent decision-making units; The intelligent water intake tower is a vertical multi-layer structure. Each layer is equipped with a movable and deformable water intake, a multi-modal sensor array, a water intake valve located at the deformable water intake, a deformable water intake optimization module, a data acquisition module of the multi-modal sensor array, and a green energy-driven self-powered module. The deformable water intake adopts a biomimetic flexible structure; the deformable water intake optimization module is driven by both hydraulic and shape memory alloys to dynamically adjust the shape of the deformable water intake opening; the data acquisition module of the multimodal sensor array controls the multimodal sensor array to collect data in real time, including at least water level, water temperature, flow rate, turbidity, dissolved oxygen, and pollutant concentration, and transmits the collected data to the intelligent decision-making unit; the green energy-driven self-powered module utilizes solar and hydropower to provide the energy required for the operation of the intelligent water intake tower. The intelligent decision-making unit includes a deep reinforcement learning dynamic weight allocation module, a multi-objective optimization solution module, a digital twin-based model prediction and control module, a cross-layer coupling compensation algorithm module, an intelligent compensation mechanism module, and an intelligent control execution module. The system comprises the following modules: a deep reinforcement learning dynamic weight allocation module, which integrates data from a multimodal sensor array data acquisition module and dynamically adjusts the weights of each parameter based on downstream demand, urgency, and water quality standards; a multi-objective optimization solution module, which uses a constrained quadratic programming algorithm to minimize downstream parameter deviations and calculate the intake valve opening; a digital twin-based model prediction control module, which simulates downstream parameters under different intake valve openings and outputs predicted downstream parameter values ​​to predict future water quality changes and optimize intake valve actions and control strategies for future periods; a cross-layer coupling compensation algorithm module, which adjusts the control strategy based on the coupling effect during water intake from different water layers in the intelligent water intake tower and compensates for the mutual influence between water intake operations at each water layer; an intelligent compensation mechanism module, which dynamically compensates for sudden pollution events reported downstream and adjusts the intake valve openings of each layer in real time; and an intelligent control execution module, which sends control commands to the intelligent water intake tower according to the optimized control strategy to control the intake valve openings of each layer, thereby achieving precise layered water intake and accurate control of downstream water quality parameters.

2. The stratified water intake intelligent control system based on multimodal fusion and autonomous evolution as described in claim 1, characterized in that: Formula for matching the shape of the water intake opening with the flow field: ； Constraint: Q(s) ≥ Q min The opening stress σ≤σ SMA ; Where s is the opening shape parameter; v is the flow velocity; and M is the turbulence suppression weight matrix; is the friction factor; sedimentation rate (s) is the sedimentation rate when the opening shape parameter is s; Q(s) is the intake flow rate when the opening shape parameter is s; Q min σ is the minimum water intake flow rate; σ is the intake opening stress; σ SMA The allowable stress for shape memory alloys.

3. The stratified water intake intelligent control system based on multimodal fusion and autonomous evolution as described in claim 1, characterized in that: The power supply formula for the green energy-driven self-powered module is as follows: P total =h solar •A • I sun +n hydro • p • Q • g • H; Where: P total η represents the total electricity generated from sunlight and water flow. solar A represents the solar energy conversion efficiency; A represents the heat collection area of ​​the outer wall of the intelligent water intake tower; I represents the solar energy conversion efficiency. sun η is the light intensity; hydro ρ is the solar energy conversion efficiency; Q is the water density; H is the flow rate; g is the head height; and g is the gravitational acceleration. An energy management algorithm is adopted to prioritize the use of real-time generated energy; ； in: This represents the amount of electricity required. Power supply for the water intake valve; Power supply to the sensor.

4. The stratified water intake intelligent control system based on multimodal fusion and autonomous evolution according to claim 1, characterized in that: The deep reinforcement learning dynamic weight allocation module combines dynamic weight allocation with deep reinforcement learning to construct a state-action-reward model, defining its state space, action space, and reward function, including... State space: s t =[T i S i O i C i ΔQ target P consumed ], i = 1, 2, ..., n; Among them, s t Let T be the state-space parameter set at time t; i S i O i C i These represent the water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C of the i-th layer, respectively. The value of i is determined by the water level H collected from the water intake. ΔQ target The adjustment range for the target flow rate at the water intake tower outlet; P consumed This represents the amount of electricity required. Action space: α t =Δw T Δw S Δw O Δw C , satisfying ∑w=1; Where, α t The action space parameter set at time t; w T w S w O w C Δw is a weighting coefficient for water temperature (T), turbidity (S), dissolved oxygen (O), and pollutant concentration (C), reflecting the importance of each parameter; T Δw S Δw O Δw C The adjustment range of the weighting coefficients for water temperature T, turbidity S, dissolved oxygen O and pollutant concentration C; Reward function: a weighted sum of the rate of reduction in downstream parameter deviation and the rate of reduction in energy consumption. r t =−(α1‖y out −y target ‖ 2 +α2∑∣Δu i ∣+α3P consumed ); Where, r t Let be the reward function parameter set at time t; α1 is the action space parameter representing the difference between the downstream outlet parameter set (composed of water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C) and the target parameter set; α2 is the action space parameter representing the total change in valve opening; α3 is the action space parameter representing the required electricity consumption; y out =[T out S out O out C out ] T y out This is a parameter set consisting of downstream outlet water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C; y target =[T target S target O target C target ] T y target This is the set of target parameters to be achieved for the downstream outlet water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C; Δu i P represents the change in the opening degree of the water intake valve. consumed This represents the amount of electricity required.

5. The stratified water intake intelligent control system based on multimodal fusion and autonomous evolution according to claim 1, characterized in that: Calculating the opening degree of the water intake valve includes Assume the flow rate Q at each water intake point i With respect to the opening degree of the water intake valve u i The relationship is: Q i =k i • u i ; Where, k i For flow coefficient; The downstream parameter is a flow-weighted average of the parameters at each layer: ； ； ； ； in, , , These are the downstream outlet water temperature (T), turbidity (S), dissolved oxygen (O), and pollutant concentration (C), respectively; T i S i O i C i The water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C of the i-th layer are respectively; Q i Let be the water intake volume of the water intake valve at the i-th layer; The goal is to minimize the deviation of downstream parameters from the target value while satisfying the total flow constraint. Objective function: ； Constraints: Total flow constraint: ; Water intake valve opening limit: 0≤u i ≤1 (i=1,2,…,n); Where: w T w S w O w C The weighting coefficients for water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C; , , These are the downstream outlet water temperature T, turbidity S, dissolved oxygen O, and pollutant concentration C, respectively. , , , These are the target values ​​for water temperature (T), turbidity (S), dissolved oxygen (O), and pollutant concentration (C), respectively; Q i Q represents the water intake volume of the water intake valve at the i-th layer; target The target water intake for downstream export; Quadratic programming: Expand the objective function into a quadratic form and solve it using a solver; Objective function standardization: ; in: u i Let H be the opening degree of the water intake valve at the i-th layer; H is the Hessian matrix, determined by the weights of each parameter w. T w S w O w C and k i T i S i O i C i The composition is as follows: f is a linear term, which is related to the deviation between the target value and the current parameter.

6. The stratified water intake intelligent control system based on multimodal fusion and autonomous evolution according to claim 1, characterized in that: Compensation for the mutual influence between water intake operations in different water layers includes establishing an interlayer hydraulic interference model: ； in, k represents the actual water intake volume of the water intake valve at level i; i For flow coefficient; u i β represents the opening degree of the water intake valve at the i-th layer; ij The interlayer interference coefficient is the effect of the opening degree of the water intake valve on the flow velocity of other layers; u ij The coupling degree between the opening degree of the water intake valve of a certain floor and the opening degree of the water intake valve of other floors; This represents the probability of this situation occurring.

7. The stratified water intake intelligent control system based on multimodal fusion and autonomous evolution according to claim 1, characterized in that: Dynamic compensation for sudden pollution events reported from downstream sources is achieved through the following formula: ； in, The opening degree of the water intake valve at the i-th layer after feedback correction from the downstream sensor; The opening degree of the water intake valve at the i-th layer before feedback correction from the downstream sensor; C is the conversion coefficient between pollutant concentration and water intake valve opening. target The target pollutant concentration to be achieved; C measured This represents the actual measured concentration of pollutants.

8. A stratified water intake intelligent control method based on multimodal fusion and autonomous evolution, characterized in that: include The system collects real-time data including at least water level, water temperature, flow rate, turbidity, dissolved oxygen, and pollutant concentration. It integrates this data and dynamically adjusts the weights of each parameter based on downstream demand, urgency, and water quality standards. The control strategy is then adaptively adjusted according to these weights. A constrained quadratic programming algorithm is used to calculate the intake valve opening, aiming to minimize downstream parameter deviations. Downstream parameters under different intake valve openings are simulated, and predicted downstream parameter values ​​are output to predict future water quality changes, optimizing intake valve actions and control strategies for future periods. The control strategy is adjusted based on the coupling effects during water intake from different water layers in the intelligent water intake tower, compensating for the mutual influence between water intake operations at each layer. Dynamic compensation is provided for sudden pollution events reported downstream, and the intake valve openings at each layer are adjusted in real-time. Control commands are sent to the intelligent water intake tower based on the optimized control strategy to control the intake valve openings at each layer, achieving precise stratified water intake and accurate control of downstream water quality parameters.

Images