Unmanned water body ecological detection and treatment system

Abstract

The invention provides an unmanned water body ecology detection and treatment system, which comprises an unmanned water surface carrier platform, a multi-dimensional space-time sensing module, a predictive decision processing module, a preventive treatment execution module, a data link module and a shore-based control center, the invention provides a multi-factor fusion decision-making mechanism based on a prediction result and real-time perception, a differentiated and prospective treatment strategy can be automatically generated and executed according to the potential risk level of algae outbreak, and intelligentization and precision of treatment actions are realized. By constructing a model iterative optimization module of the shore-based control center, a complete closed-loop intelligent system capable of self-learning and sustainable evolution is formed, the accuracy and adaptability of long-term operation of the system are ensured, the manual operation and maintenance cost is remarkably reduced, and the system is suitable for large-scale popularization and application. And a systematic technical solution is provided for refined and automatic ecological management of a large-range water area.

Description

Technical Field

[0001] This invention relates to the field of water ecological treatment technology, and in particular to an unmanned water body ecological detection and treatment system. Background Technology

[0002] Current water ecological monitoring and management technologies are mainly divided into three categories: traditional manual methods, semi-automated monitoring equipment, and basic unmanned surface vessel systems. These technologies have the following core shortcomings, making it difficult to meet the needs of precise and efficient management of complex water bodies: Monitoring aspects: low accuracy, narrow coverage, poor adaptability: Traditional manual monitoring relies on manual sampling and laboratory analysis, with low monitoring frequency (once a day to once a week) and limited coverage (≤1km). 2 The data lag is significant; existing automated monitoring equipment is mostly deployed at fixed points, unable to achieve full-area mobile monitoring, and the sensors are easily affected by water turbidity (data error ≥20% in high turbidity waters >50 NTU), lacking self-cleaning and automatic calibration mechanisms, resulting in poor long-term operational stability. Basic unmanned surface vessel systems only carry a single GPS positioning system (accuracy ≥1 meter), with fixed trajectory planning, unable to dynamically adjust for pollution hotspots, relying on a single sensor (such as infrared) for obstacle avoidance, with a response time ≥1 second, and an obstacle avoidance success rate ≤95% in complex waters (including buoys and bridge piers).

[0003] At the decision-making and governance level: Lagging, extensive, and high pollution risk: Existing technologies lack a "prediction-governance" linkage mechanism, relying solely on passive governance based on real-time monitoring data, unable to predict algal bloom trends in advance (most systems lack prediction functions, and a few prediction models have errors ≥30%); governance methods primarily rely on chemical spraying, posing a risk of secondary pollution, and exhibiting poor positioning accuracy (error ≥5 meters), resulting in a chemical waste rate ≥30%; some ultrasonic treatment equipment uses a single frequency output, failing to dynamically adjust parameters based on algal density, leading to an algal nucleus inactivation rate ≤85% and limited treatment effectiveness. Furthermore, existing systems do not consider aquatic life protection, resulting in a fish survival rate ≤90% during treatment, indicating poor ecological compatibility.

[0004] System collaboration and scalability: fragmentation, poor compatibility, and complex operation and maintenance: The navigation, detection, and governance modules of existing unmanned vessels are independently designed, with data interaction latency ≥50ms, no standardized interfaces (most are dedicated interfaces), and third-party platform integration cycles ≥15 days; the energy storage system mainly uses traditional lithium batteries (energy density ≤150Wh / kg, cycle life ≤2000 times), with short endurance (≤6 hours) and a lack of flexible solar panel collaborative compensation schemes; the system lacks digital twin modeling capabilities, cannot simulate pollution diffusion trends, the shore-based control center can only achieve basic data display, lacks model iteration and dynamic optimization capabilities, and operation and maintenance rely on manual calibration and fault diagnosis, resulting in high labor costs. Summary of the Invention

[0005] The purpose of this invention is to provide an unmanned water body ecological detection and management system, which aims to solve the problems of insufficient accuracy, delayed decision-making, extensive management and poor system coordination in the existing technology.

[0006] To achieve this objective, the present invention provides an unmanned water body ecological detection and management system, which includes: an unmanned water surface vehicle platform, a multi-dimensional spatiotemporal perception module, a predictive decision processing module, a preventive management execution module, a data link module, and a shore-based control center.

[0007] The multi-dimensional spatiotemporal perception module is installed on an unmanned water surface vehicle platform to collect multi-dimensional ecological parameter data of the target water body in real time.

[0008] The predictive decision processing module establishes a data connection with the multi-dimensional spatiotemporal perception module and is configured to receive multi-dimensional ecological parameter data. Based on the preset time-series prediction model and decision logic, it generates navigation control commands for the unmanned surface vehicle platform and governance commands for the target water area.

[0009] The preventive governance execution module, mounted on an unmanned surface vehicle platform and controlled by a predictive decision processing module, is used to perform preventive or inhibitory governance operations on the target water area according to governance instructions. The data link module is used to enable two-way data communication between the unmanned surface vehicle platform and the shore-based control center; The shore-based control center is used for remote monitoring and mission planning of unmanned surface vehicles, as well as iterative optimization of the models within the predictive decision processing module.

[0010] Furthermore, the hardware carrier of the predictive decision processing module is an embedded industrial computer, and its software consists of the following four interlocking sub-modules: The Long Short-Term Memory Network Temporal Prediction Submodule is used to receive historical and real-time ecological and environmental parameters and predict the evolution trend of key ecological indicators in the target water area within a specific time period in the future. The convolutional neural network image recognition submodule is used to analyze underwater images acquired by the multidimensional spatiotemporal perception module, identify the dominant algal populations in the water body and assess their spatial distribution density. The multi-factor fusion decision submodule is used to fuse the prediction results of the time series prediction submodule, the recognition results of the image recognition submodule, and the real-time data of the multi-dimensional spatiotemporal perception module, and generate specific governance instructions based on a set of preset decision matrices; and the digital twin modeling submodule.

[0011] Furthermore, the network structure of the Long Short-Term Memory Network Temporal Prediction Submodule includes an input layer, three hidden layers, and an output layer. The number of neurons in the three hidden layers is set to 64, 32, and 16 respectively, and a Dropout layer with a dropout rate of 0.2 is set between the hidden layers. The input to the Long Short-Term Memory Network (LSTM) time-series prediction submodule is a multi-dimensional time-series feature vector containing daily average water quality monitoring data for the past 12 months and corresponding historical meteorological data obtained from the shore-based control center. The water quality monitoring data includes water temperature, total nitrogen concentration, total phosphorus concentration, and chlorophyll a concentration, while the historical meteorological data includes daily average sunshine duration, air temperature, rainfall, and wind speed. The output of the LTM time-series prediction submodule is a sequence of hourly predicted chlorophyll a concentration values ​​for the target water area over the next 72 hours. The Long Short-Term Memory Network (LSTM) time series prediction submodule also integrates a meteorological data linkage mechanism. This mechanism acquires meteorological forecast data for the next 24 hours in real time and quantifies meteorological factors into a weight coefficient through a weighted fusion algorithm to dynamically correct the model's original output.

[0012] Furthermore, the input to the multi-factor fusion decision submodule is a real-time state vector, which includes: the algae concentration prediction curve for the next 72 hours output by the long short-term memory network time-series prediction submodule, the current dominant algae species and density output by the convolutional neural network image recognition submodule, the current real-time water quality parameters output by the multi-dimensional spatiotemporal perception module, and real-time meteorological information obtained from the shore-based control center. The multi-factor fusion decision-making submodule internally contains a three-level response decision matrix, and its decision logic is defined as follows: When the predicted peak algae concentration is below the first threshold within the next 72 hours, and the real-time concentration is also below the first threshold, the decision status is "Level 1 Monitoring," and the system only performs patrol monitoring tasks. When the predicted peak concentration is between the first and second thresholds, or the real-time concentration has exceeded the first threshold, the decision status switches to "Level 2 Prevention," and the system generates a preventive treatment instruction. When the predicted peak concentration is higher than the second threshold, or when the real-time concentration has exceeded the second threshold, the decision state switches to "Level 3 Suppression", and the system generates an enhanced suppression control instruction. The first threshold was set at 30 μg / L, and the second threshold was set at 100 μg / L.

[0013] Furthermore, the multi-dimensional spatiotemporal perception module includes: A water quality parameter sensor array, which consists of an ion-selective electrode pH sensor, a fluorescence quenching dissolved oxygen sensor, a total nitrogen sensor, a total phosphorus sensor, and an optical fluorescence chlorophyll a sensor; The underwater image acquisition unit consists of an industrial-grade CMOS image sensor with a resolution of 1920×1080 pixels and a matching wide-angle lens, and is equipped with a 6W LED fill light with adjustable brightness and sonar equipment.

[0014] Furthermore, the preventive governance execution module includes a quantum wave and ultrasonic wave synergistic generator, which is installed at the bottom of the unmanned surface vehicle platform; The synergistic effect generator contains an ultrasonic transducer array and a low-frequency electromagnetic field generating coil. The ultrasonic transducer array is controlled by a power driver and can emit ultrasonic waves with a center frequency of 25kHz and a total acoustic power of 50W-80W that is adjustable. These waves are used to generate cavitation effects in water to increase the permeability of algal cell walls. The low-frequency electromagnetic field generating coil can generate pulsed electromagnetic fields of a specific frequency, which can be used to penetrate the cell wall to interfere with the transmembrane transport of key ions and the cell division process inside algal cells; in addition, the synergistic generator uses phase synergistic technology to ensure that the electromagnetic field pulse and the ultrasonic wave peak are superimposed in phase in the target action area.

[0015] Furthermore, the unmanned surface vehicle platform integrates an autonomous navigation and obstacle avoidance subsystem, which includes: Real-time dynamic differential positioning unit is used to receive multi-frequency satellite signals to achieve centimeter-level positioning; An inertial measurement unit with a built-in three-axis gyroscope, three-axis accelerometer and three-axis magnetometer; The multi-source data fusion processor executes an extended Kalman filter algorithm to fuse the positioning data from the real-time dynamic differential positioning unit with the attitude data from the inertial measurement unit in real time, and outputs high-frequency vehicle pose information. Millimeter-wave radar uses the millimeter-wave band for ranging and velocity detection. It has strong penetration capabilities and is suitable for obstacle detection in adverse weather conditions such as rain and fog. Solid-state LiDAR uses solid-state technology to achieve high-speed rotating scanning and generate high-resolution 3D point cloud data for environmental modeling and obstacle recognition. Infrared proximity sensors detect nearby objects by emitting and receiving infrared rays, providing short-range obstacle warnings and enhancing safety.

[0016] Furthermore, the predictive decision processing module internally incorporates a hierarchical obstacle avoidance decision-making logic, which is specifically implemented as follows: when the millimeter-wave radar detects an obstacle within a range of 20 meters to 10 meters ahead, the system triggers a first-level warning and calls an improved A* path planning algorithm to recalculate and generate a local optimal obstacle avoidance route under global path constraints; when the obstacle enters a range of 10 meters to 3 meters, the system triggers a second-level warning and controls the unmanned surface vehicle platform's propulsion system to reduce the speed to 30% of the preset safety threshold; when the obstacle is less than 3 meters away, the system triggers a third-level emergency command to execute full-power reverse thrust braking or maximum angle evasive maneuvering.

[0017] Furthermore, the shore-based control center includes a model iterative optimization module; The model iteration optimization module uses an incremental learning algorithm and is configured to automatically call up newly added valid data pairs in the time-series historical database of the shore-based control center every quarter. The valid data pairs include water quality parameter comparison data before and after the governance intervention, as well as the corresponding environmental parameters. The model iteration and optimization module uses effective data pairs to retrain and fine-tune the long short-term memory network temporal prediction submodule and the convolutional neural network image recognition submodule that have been deployed in the predictive decision processing module, thereby forming an intelligent closed loop of "perception-prediction-decision-execution-feedback-optimization".

[0018] Compared with the prior art, the advantages of the present invention are: By employing high-frequency, gridded autonomous cruise monitoring, combined with multi-dimensional water quality sensing and underwater imaging technology, this invention significantly improves the spatiotemporal resolution and coverage of aquatic ecological parameters, overcoming the limitations of traditional fixed-point monitoring or manual sampling.

[0019] This invention innovatively introduces a time-series prediction model based on Long Short-Term Memory (LSTM) networks and integrates real-time meteorological data to achieve accurate prediction of future trends of key ecological indicators (such as algae concentration), thus elevating water ecological management from "post-event response" to a new dimension of "pre-event early warning".

[0020] This invention proposes a multi-factor fusion decision-making mechanism based on prediction results and real-time perception, which can automatically generate and execute differentiated and forward-looking governance strategies according to the potential risk level of algal outbreaks, thus realizing intelligent and precise governance actions.

[0021] This invention employs a non-chemical treatment method that combines quantum waves and ultrasound, directly targeting the root cause of algal cell reproduction. It prioritizes prevention, fundamentally avoiding large-scale algal blooms while eliminating the risk of secondary pollution from chemical agents, thus balancing treatment effectiveness with ecological safety.

[0022] By constructing a model iteration and optimization module for the shore-based control center, this invention forms a complete closed-loop intelligent system capable of self-learning and continuous evolution, ensuring the accuracy and adaptability of the system's long-term operation, significantly reducing manpower maintenance costs, and providing a systematic technical solution for the refined and automated ecological management of large-scale water areas. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the operation of the system in an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be further described below.

[0025] This invention aims to provide an unmanned aquatic ecosystem monitoring and management system. This system deeply integrates multiple technological dimensions, including autonomous navigation, multi-dimensional perception, temporal prediction, intelligent decision-making, and precise management, constructing a closed-loop intelligent system capable of proactive, preventative, and autonomous management of aquatic ecosystems. The following will provide a detailed engineering description of each module constituting the system and their collaborative working methods. The workflow of each module is as follows: Figure 1 As shown.

[0026] The unmanned surface vehicle platform involved is the physical foundation and mobile operation basis of the entire system. In this embodiment, the platform adopts a catamaran design to balance navigation stability and maneuverability, enabling it to adapt to various aquatic environments, including lakes, reservoirs, and rivers.

[0027] The core of the unmanned surface vehicle platform's propulsion system consists of two 1.5kW permanent magnet synchronous brushless DC motors, each driving a ducted thruster mounted below the stern of the catamaran hull. The ducted thruster has a 150mm inner diameter fairing and houses a five-bladed, high-angle propeller made of high-strength nylon composite material. The duct design effectively regulates water flow, improves propulsion efficiency, and protects the propeller from entanglement by weeds and other debris. The two thrusters are controlled by independent electronic speed controllers. A predictive decision processing module outputs two independent pulse-width modulation (PWM) signals to precisely control the motor speeds, enabling highly maneuverable navigation maneuvers such as forward, backward, differential steering, and U-turns.

[0028] The unmanned surface vehicle platform integrates a high-precision autonomous navigation and obstacle avoidance subsystem. The core of this subsystem's positioning is a real-time dynamic differential positioning unit (RTK) with multi-constellation, multi-frequency reception capabilities. Specific models can support GPS L1 / L2 / L5, BDS B1I / B2I / B3I, and GLONASS G1 / G2 frequencies. By receiving differential correction data broadcast from ground base stations, this unit can achieve a positioning accuracy of less than or equal to 0.2 meters, with a positioning data update frequency of 10Hz. Attitude perception relies on an industrial-grade inertial measurement unit (IMU) that integrates a three-axis MEMS gyroscope, a three-axis accelerometer, and a three-axis magnetometer. The aforementioned RTK positioning data and IMU attitude data are simultaneously input into a multi-source data fusion processor. This processor is a high-performance ARM Cortex-A series chip with a floating-point unit, running an optimized extended Kalman filter algorithm. The algorithm's state vector includes the vehicle's three-dimensional position, three-dimensional velocity, three-dimensional attitude angles (roll, pitch, and heading), and the IMU's zero-bias error. By optimally estimating and fusing the RTK position observations and the IMU's kinematic model predictions, it ultimately outputs continuous and smooth vehicle attitude information, providing high-quality input for path tracking control. Specifically, when the hull roll angle is detected to be >15° or pitch angle to be >10°, the power distribution and speed of the dual thrusters are dynamically adjusted to generate a counter-compensation torque, achieving a roll / pitch suppression rate of ≥85%. Equipped with an automatic anchoring system or dynamic positioning system, it can achieve high-precision monitoring of the stationary state of the detection setpoint. During motion state detection, the hull is kept stable through GPS navigation and the attitude control system.

[0029] At the obstacle avoidance perception level, the subsystem of this invention employs a multi-sensor redundancy configuration. A solid-state millimeter-wave radar is installed in the forward main detection direction, with a maximum effective detection range of 50 meters and a range resolution of 0.1 meters, enabling reliable detection of obstacles such as ships, bridge piers, and navigation marks on the water surface ahead, 24 / 7. A solid-state lidar is installed on the top of the vehicle's mast to construct a 360-degree point cloud map, accurately perceiving low-lying obstacles or shoreline contours at close range. Simultaneously, four sets of infrared proximity sensors are deployed around the hull to detect obstacles within a range of less than 1.5 meters. The layered obstacle avoidance decision logic embedded in the predictive decision processing module is specifically implemented as follows: when the millimeter-wave radar detects an obstacle within a range of 20 to 10 meters on the route, the system enters a first-level warning state, and the navigation controller invokes an improved A* path planning algorithm. The algorithm's cost function not only considers the geometric distance to the target point but also incorporates a heading change penalty and an obstacle threat assessment term. Under global flight path constraints, it can quickly generate an energy-optimal and safe local obstacle avoidance trajectory. When an obstacle enters the range of 5 to 10 meters, the system triggers a level two warning, and the control command limits the total output power of the propulsion system to 30% of a preset safety threshold (e.g., normal cruise power) to achieve smooth deceleration. Once an obstacle intrudes into the danger zone within 5 meters, a level three emergency command is triggered. The system will, based on the obstacle's relative position and speed, execute actions such as full-power reverse rotation of both thrusters for emergency braking, or maximum differential speed command for extreme cornering avoidance.

[0030] The energy supply system is crucial for ensuring the platform's long-term autonomous operation. The main component of this system is a 24V battery pack composed of lithium iron phosphate (LiFePO4) cells connected in series and parallel, with a nominal capacity of 800AH and a total energy storage of approximately 19.2kWh. This battery pack is monitored by an advanced battery management system (BMS), which can monitor the voltage and temperature of each cell string in real time, perform active balancing management, and provide overcharge, over-discharge, overcurrent, and short-circuit protection. To extend the range, the vehicle's deck is equipped with a flexible monocrystalline silicon solar panel array with a total peak power of 300W. Under typical summer sunlight conditions, the light intensity is no less than 800W / m², enabling the vehicle to maintain a stable continuous operating range of over 8 hours when performing low-speed cruise monitoring missions.

[0031] Next, the multi-dimensional spatiotemporal sensing module will be described in detail. This module is the direct source for the system to acquire parameters of the aquatic ecosystem state. To ensure measurement accuracy, the entire module is integrated into a specially designed shock-absorbing chamber. This chamber is integrally molded from carbon fiber composite material and flexibly connected to the main frame of the hull through eight sets of specially designed metal-rubber dampers. These dampers can effectively absorb and dissipate high-frequency vibrations (50-200Hz band) generated by hull navigation and wave impact, ensuring that the measurements of the internal precision sensors are not interfered with by mechanical noise.

[0032] Inside the shock-absorbing chamber, the core component is a water quality parameter sensor array. This array includes: a pH sensor employing ion-selective electrode technology, with a built-in reference and measuring electrode in its electrode head, covering a pH range of 0 to 14, and achieving an accuracy of ±0.01 pH after calibration with a standard buffer solution; a dissolved oxygen sensor using fluorescence quenching, which works by irradiating a fluorescent substance with excitation light of a specific wavelength, the fluorescence intensity of which is inversely proportional to the dissolved oxygen concentration in the water. This method offers fast response, consumes no electrolyte, and has a measurement range of 0 to 20 mg / L with an accuracy of ±0.1 mg / L; and a total nitrogen sensor using ultraviolet absorption spectroscopy, calculating nitrate nitrogen concentration by measuring the difference in absorbance between 220 nm and 275 nm in the water sample, and then converting it to total nitrogen, with a measurement range of 0 to 50 mg / L and an accuracy of ±0.05 mg / L. One total phosphorus sensor employs a molybdenum blue colorimetric method. A built-in micropump mixes the water sample with a colorimetric reagent (ammonium molybdate and potassium antimony tartrate), which reacts under specific conditions to form a blue complex. The total phosphorus concentration is determined by measuring the absorbance at a specific wavelength, with a measurement range of 0 to 10 mg / L and an accuracy of ±0.05 mg / L. The other sensor is an optical fluorescence chlorophyll a sensor. It emits blue light of a specific wavelength to excite chlorophyll a in the water to produce red fluorescence. The chlorophyll a concentration is quantified by measuring the fluorescence intensity, with a measurement range of 0 to 500 μg / L and an accuracy of ±2 μg / L.

[0033] In addition to water chemistry parameters, the module also includes an underwater image acquisition unit. This unit uses a 1 / 2.8-inch industrial-grade CMOS image sensor with an effective pixel count of 1920×1080. The entire imaging component is encapsulated within a spherical transparent housing made of high-transmittance polycarbonate injection molding, with an IP68 protection rating. To cope with different lighting conditions, eight 6W white LEDs are arranged in a ring around it, with their brightness linearly adjustable from 0% to 100% via a PWM signal, and a constant color temperature of 5500K to ensure accurate color reproduction of the image.

[0034] Signals from all sensors and image acquisition units are ultimately converged onto a data acquisition and preprocessing board. The core processor of this board is a high-performance 32-bit microcontroller, specifically the STMicroelectronics STM32H7 series, with a clock speed of up to 480MHz and powerful data processing capabilities. The firmware running on this microcontroller synchronously samples all sensor channels at a frequency of 1Hz. For each set of raw data acquired, the program first applies a median filter algorithm with a window size of 5 to effectively filter out abnormal spike pulses caused by circuit noise or transient interference from tiny particles in the water. Subsequently, the filtered data is normalized using a min-max scaling method to linearly map values ​​of different physical units and dimensions (such as pH, mg / L, μg / L, etc.) to the dimensionless interval [0, 1], facilitating subsequent processing by the neural network model. Finally, the microcontroller encapsulates the precise timestamps (from the RTK module), GPS coordinates, vehicle attitude data (pitch and roll angles), and preprocessed multidimensional perception data into a JSON (JavaScript Object Notation) data frame according to a predefined format, and sends it stably to the predictive decision processing module via the CAN bus at a frequency of 1Hz.

[0035] The predictive decision processing module is the core hub of this invention, enabling a shift from passive response to proactive prediction and from data acquisition to intelligent decision-making. Its hardware carrier is an embedded industrial computer installed in the drying chamber inside the vehicle. The computer operates on mainstream operating systems such as Windows, Linux, and Android.

[0036] At the software level, the core of this module is a Long Short-Term Memory (LSTM) time series prediction submodule. As a specific implementation, the network structure of this LSTM model has been carefully designed and optimized, comprising an input layer, three hidden layers, and an output layer. The input layer receives a multi-dimensional time series consisting of daily average water quality and meteorological data from the past 365 days (forming annual periodic characteristics). Specific feature vectors include: water temperature, total nitrogen concentration, total phosphorus concentration, chlorophyll a concentration, and corresponding historical meteorological data synchronized in real-time from the shore-based control center via a data link, such as daily average sunshine duration, average temperature, daily rainfall, and average wind speed. The number of neurons in the three hidden layers decreases sequentially, to 64, 32, and 16 respectively. The activation function for all hidden layer neurons is the Modified Linear Unit (ReLU), as it effectively alleviates the vanishing gradient problem and accelerates model convergence. To prevent overfitting during training, a Dropout layer with a dropout rate of 0.2 is set after each hidden layer. The output layer contains 72 neurons, corresponding to hourly predictions for the next 72 hours. Its activation function is the Sigmoid function to ensure that the predicted chlorophyll a concentration is normalized to the [0, 1] interval. The model is trained using the Adam optimizer with mean squared error (MSE) as the loss function. After training on a comprehensive dataset containing five years of historical data from more than five different reservoirs, its root mean square error on the independent validation set is strictly controlled within 10%. Furthermore, this submodule integrates a meteorological data linkage mechanism. It obtains authoritative weather forecast data for the next 24 hours in real time via an API interface and uses a pre-defined weighted fusion algorithm to quantify the promoting or inhibiting effects of meteorological factors (such as upcoming consecutive hot and sunny days or heavy rain) on algal growth into a dynamic weight coefficient (e.g., a weight coefficient of 1.2 for hot and sunny days and 0.8 for heavy rain, accounting for 30% of the total influencing factors). This coefficient is used to dynamically correct the original prediction sequence output by the LSTM model, thereby significantly improving the prediction accuracy and robustness under scenarios of drastic weather changes.

[0037] Parallel to the LSTM is a Convolutional Neural Network (CNN) image recognition submodule. This submodule is used for real-time analysis of underwater images, identifying dominant algal populations and assessing their spatial density. Its network structure is determined to be a lightweight sequence model containing three convolutional layers, two max-pooling layers, and two fully connected layers. All convolutional kernels are set to 3×3 with a stride of 1, padding of 'same', and the activation function is ReLU. Before being fed into the network, the input image undergoes a series of preprocessing steps to enhance image contrast in turbid water. The model's output consists of two parts: a classification probability vector and a spatial density matrix. The former, output through a softmax-activated fully connected layer, represents the identification probability of the main algal populations in the image, for example (Cyanobacteria: 0.92, Chlorophyta: 0.08, Diatoms: 0.00); the latter is obtained by upsampling and visualizing the feature map of the last convolutional layer, intuitively representing the heatmap of algal aggregate distribution in the image pixel space.

[0038] The future trend predictions of LSTM and the current state recognition results of CNN, along with real-time water quality parameters, vehicle location, and meteorological information, are all integrated into a multi-factor fusion decision submodule. This submodule is essentially an expert system built on deterministic rules and state transition logic. Internally, it has a three-level response decision matrix with the following decision logic: State 1, Level 1 Monitoring. When the chlorophyll a concentration peak predicted by LSTM for the next 72 hours is lower than the first threshold (e.g., set to 30 μg / L based on the eutrophication level of the water body), and the concentration measured by the real-time water quality parameter sensor is also lower than this value, the system determines that the aquatic ecosystem is in a safe and stable state. The decision module only instructs the unmanned vehicle to perform routine patrol monitoring tasks according to a preset gridded path, while the governance module remains on standby. State 2, Level 2 Prevention. When the predicted concentration peak is between the first threshold (30 μg / L) and the second threshold (e.g., set as the critical outbreak value for algal blooms of 100 μg / L), or when the real-time concentration has exceeded the first threshold, the system determines that there is a potential risk of algal bloom. The decision-making module immediately switches to Level 2 prevention mode and generates a preventative control instruction. Level 3, Level 3 Suppression: Once the predicted peak concentration exceeds the second threshold (100 μg / L), or the real-time concentration has already exceeded this value, it indicates that algal blooms have initially formed or are about to erupt on a large scale, and the system classifies it as a high-risk event. The decision-making module switches to Level 3 suppression mode and generates an enhanced suppression control instruction.

[0039] The preventative governance execution module receives and executes the aforementioned governance instructions. At its core is an innovative quantum wave and ultrasonic synergistic generator. This generator is precisely mounted at the center bottom between the catamaran hulls of the unmanned surface vehicle platform. Its outer shell measures 30 cm × 20 cm × 15 cm and is encapsulated in 316L stainless steel to ensure corrosion resistance. Internally, it integrates an ultrasonic transducer array consisting of 12 piezoelectric ceramic transducers and a low-frequency electromagnetic field generating coil composed of a multi-turn solenoid. The ultrasonic transducer array can emit an ultrasonic beam with a continuously adjustable total acoustic power ranging from 50 W to 80 W near a center frequency of 25 kHz. The low-frequency electromagnetic field generating coil can generate pulsed electromagnetic fields with specific waveforms (such as pulsed square waves) and frequencies (such as 15 Hz). The synergistic mechanism lies in the fact that when 25kHz ultrasound propagates in water, it utilizes the acoustic cavitation effect to generate numerous tiny cavitation bubbles around algal cells. The instantaneous collapse of these bubbles generates localized high temperature and pressure, as well as powerful microjets, applying continuous mechanical stress to the algal cell walls, disrupting their integrity and significantly increasing their permeability. Simultaneously, the synergistically applied specific low-frequency pulsed electromagnetic field can penetrate the damaged cell walls in a non-thermal manner, directly interfering with the transmembrane transport balance of key ions (such as Ca²⁺) within the algal cells, thereby affecting key electrophysiological activities such as tubulin polymerization during cell division. This synergistic effect, particularly through precise phase coordination technology, ensures that the peak value of the electromagnetic field pulse and the pressure peak of the ultrasound are superimposed in phase within the target area, greatly enhancing the inhibitory effect on algal cell reproduction. It can effectively induce programmed cell death or render algae unable to reproduce before they enter the exponential growth phase, i.e., in the early stages of cell division, thus achieving a preventative effect of "treating disease before it occurs."

[0040] Information exchange between the unmanned surface vehicle platform and the shore-based control center is ensured by the data link module. This module adopts a multi-mode communication architecture to guarantee seamless and reliable connectivity. Its core integrates a cellular communication module supporting 4G and 5G networks, as well as a satellite communication module supporting BeiDou-3 short message communication. In areas with good cellular network coverage, especially 5G areas, the system defaults to using this link for high-speed data transmission, with an uplink bandwidth of at least 1Mbps, sufficient to push 1080P video streams from underwater cameras to the shore-based control center in real time via the RTSP protocol for operator monitoring. When the vehicle navigates to areas with weak network signals, such as deep reservoirs or remote river sections, the link management unit detects increased network latency and packet loss rate; when the communication quality falls below a preset threshold, the system will automatically and seamlessly switch to BeiDou satellite communication mode within 0.5 seconds.

[0041] The shore-based control center, acting as the "brain" of the entire system, has its software deployed on a cloud server or a server cluster in a local data center. This center comprises four main functional modules: an integrated control platform providing operators with a web-based graphical user interface, allowing for remote planning, editing, and distribution of flight routes and tasks, real-time monitoring of various vehicle operating parameters (such as battery level, speed, and location), and the ability to manually control the vehicle in emergencies; a visualized geographic information monitoring interface; a historical database, using a database specifically optimized for time-series data, storing all historical monitoring data and management operation logs long-term and at high density; and a model iteration and optimization module, trained based on daily historical data from the past 12 months (including water temperature, total nitrogen / total phosphorus concentration, chlorophyll a concentration, average daily sunshine duration, and meteorological data), with ≥1000 iterations and a loss function value ≤0.05, achieving 72-hour algae concentration prediction with a prediction error ≤10% and a confidence level ≥95%.

[0042] The technical effects of the present invention will be further illustrated below through specific embodiments.

[0043] Example 1: Algae Control in an Eutrophic Lake System Configuration: Sensors: Chlorophyll a sensor (accuracy ±2μg / L), underwater camera (IP68 waterproof).

[0044] LSTM model: The training data consists of daily water quality data (water temperature, total phosphorus, chlorophyll) of the lake over the past 12 months, with 64 / 32 / 16 neurons in the hidden layer.

[0045] Governance parameters: Quantum wave / ultrasound energy generator and release device.

[0046] Operation process: The unmanned vessel cruises at a speed of 2 m / s, collecting data every second and uploading it to the AI ​​module; LSTM predicts that the algal concentration will reach 220 μg / L in 3 days, and CNN identifies it as a cyanobacterial bloom; The strategy module activates the quantum wave / ultrasound energy generator to perform targeted inactivation treatment on algae. Results: After 72 hours, the algae concentration dropped to 75 μg / L, which is 40% more efficient than traditional treatment methods.

[0047] Example 2: Urban River Water Ecological Monitoring System Configuration: Sensors: pH (0-14), dissolved oxygen (0-20 mg / L), total nitrogen (0-60 mg / L); LSTM model: Input parameters are water temperature and dissolved oxygen, predicting algae growth trends; Operation process: The unmanned boat cruises along a 5km river channel, monitors water quality parameters in real time, and stores the data in a digitized manner at the shore-based center. LSTM predicts that the algal concentration is stable at 20 μg / L (low risk), and the system will only continuously monitor without initiating treatment. Results: Enables visualized monitoring of water quality across the entire river area, with data update frequency increased to once per second, expanding the coverage area by 5 times compared to traditional monitoring stations.

[0048] Example 3: Early Warning and Prevention of Algae in Reservoirs System Configuration: Sensors: Chlorophyll a, light intensity sensor; Governance Module: Quantum Wave / Ultrasound Energy Generator and Releaser; Operation process: LSTM predicts that the algal concentration will rise from 40 μg / L to 65 μg / L after 72 hours (medium risk). The decision module activates the quantum wave / ultrasound energy generator to inactivate the algal nuclei. Results: Algae concentration was maintained below 30 μg / L, preventing outbreaks, reducing pesticide dosage by 100%, and lowering treatment costs by 30%.

[0049] The above embodiments demonstrate that this system can dynamically adjust monitoring and management strategies according to different water scenarios, and achieve precise water ecology management through data-driven and intelligent means, thus having broad application value.

[0050] The above are merely preferred embodiments of the present invention and do not constitute any limitation on the present invention. Any equivalent substitutions or modifications made by those skilled in the art to the technical solutions and content disclosed in the present invention without departing from the scope of the present invention shall be deemed to have remained within the protection scope of the present invention.

Claims

1. An unmanned water body ecological monitoring and management system, characterized in that, include: Unmanned surface vehicle platform, multi-dimensional spatiotemporal perception module, predictive decision processing module, preventive governance execution module, data link module and shore-based control center; The multi-dimensional spatiotemporal perception module is mounted on the unmanned water surface vehicle platform and is used to collect multi-dimensional ecological parameter data of the target water body in real time. The predictive decision processing module establishes a data connection with the multi-dimensional spatiotemporal perception module. It is configured to receive the multi-dimensional ecological parameter data and generate navigation control commands for the unmanned water vehicle platform and governance commands for the target water area based on a preset time-series prediction model and decision logic. The preventive governance execution module is mounted on the unmanned surface vehicle platform and controlled by the predictive decision processing module. It is used to perform preventive or inhibitory governance operations on the target water area according to the governance instructions. The data link module is used to realize bidirectional data communication between the unmanned surface vehicle platform and the shore-based control center; The shore-based control center is used for remote monitoring and mission planning of the unmanned surface vehicle platform, and for iterative optimization of the model within the predictive decision processing module.

2. The unmanned water body ecological monitoring and management system according to claim 1, characterized in that, The hardware carrier of the predictive decision processing module is an embedded industrial computer, and its software consists of the following four mutually coupled sub-modules: The Long Short-Term Memory Network Temporal Prediction Submodule is used to receive historical and real-time ecological and environmental parameters and predict the evolution trend of key ecological indicators in the target water area within a specific time period in the future. The convolutional neural network image recognition submodule is used to analyze underwater images acquired by the multidimensional spatiotemporal perception module, identify the dominant algal populations in the water body and assess their spatial distribution density. The multi-factor fusion decision submodule is used to fuse the prediction results of the time-series prediction submodule, the recognition results of the image recognition submodule, and the real-time data of the multi-dimensional spatiotemporal perception module, and generate specific governance instructions based on a preset decision matrix; and the digital twin modeling submodule.

3. The unmanned water body ecological monitoring and management system according to claim 2, characterized in that, The network structure of the Long Short-Term Memory Network Temporal Prediction Submodule includes an input layer, three hidden layers, and an output layer. The number of neurons in the three hidden layers is set to 64, 32, and 16 respectively, and a Dropout layer with a dropout rate of 0.2 is set between the hidden layers. The input to the Long Short-Term Memory Network time series prediction submodule is a multi-dimensional time series feature vector, which contains the daily average water quality monitoring data for the past 12 months and the corresponding historical meteorological data obtained from the shore-based control center. The water quality monitoring data includes water temperature, total nitrogen concentration, total phosphorus concentration and chlorophyll a concentration, and the historical meteorological data includes daily average sunshine duration, temperature, rainfall and wind speed. The output of the Long Short-Term Memory Network temporal prediction submodule is a sequence of hourly predicted values ​​of chlorophyll a concentration in the target water area over the next 72 hours; The Long Short-Term Memory Network (LSTM) time-series prediction submodule also integrates a meteorological data linkage mechanism. This mechanism acquires meteorological forecast data for the next 24 hours in real time and quantifies meteorological factors into a weighted coefficient through a weighted fusion algorithm to dynamically correct the model's original output.

4. The unmanned water body ecological monitoring and management system according to claim 2, characterized in that, The input to the multi-factor fusion decision submodule is a real-time state vector, which includes: the algae concentration prediction curve for the next 72 hours output by the long short-term memory network time-series prediction submodule, the current dominant algae species and density output by the convolutional neural network image recognition submodule, the current real-time water quality parameters output by the multi-dimensional spatiotemporal perception module, and the real-time meteorological information obtained from the shore-based control center. The multi-factor fusion decision submodule internally contains a three-level response decision matrix, and its decision logic is determined as follows: When the predicted peak algae concentration is below the first threshold within the next 72 hours, and the real-time concentration is also below the first threshold, the decision status is "Level 1 Monitoring", and the system only performs patrol monitoring tasks. When the predicted concentration peak is between the first threshold and the second threshold, or when the real-time concentration has exceeded the first threshold, the decision state switches to "secondary prevention" and the system generates a preventive treatment instruction. When the predicted concentration peak is higher than the second threshold, or the real-time concentration has exceeded the second threshold, the decision state switches to "Level 3 Suppression", and the system generates an enhanced suppression control instruction; The first threshold was set to 30 μg / L, and the second threshold was set to 100 μg / L.

5. The unmanned water body ecological monitoring and management system according to claim 1, characterized in that, The multidimensional spatiotemporal sensing module includes: A water quality parameter sensor array, which consists of an ion-selective electrode pH sensor, a fluorescence quenching dissolved oxygen sensor, a total nitrogen sensor, a total phosphorus sensor, and an optical fluorescence chlorophyll a sensor; The underwater image acquisition unit consists of an industrial-grade CMOS image sensor with a resolution of 1920×1080 pixels and a matching wide-angle lens, and is equipped with a 6W LED fill light with adjustable brightness and sonar equipment.

6. The unmanned water body ecological monitoring and management system according to claim 5, characterized in that, The preventive governance execution module includes a quantum wave and ultrasonic wave synergistic generator, which is installed at the bottom of the unmanned surface vehicle platform; The synergistic effect generator contains an ultrasonic transducer array and a low-frequency electromagnetic field generating coil. The ultrasonic transducer array is controlled by a power driver and can emit ultrasonic waves with a center frequency of 25kHz and a total acoustic power of 50W-80W that is adjustable. These waves are used to generate cavitation effects in water to increase the permeability of algal cell walls. The low-frequency electromagnetic field generating coil can generate pulsed electromagnetic fields of a specific frequency, which are used to penetrate the cell wall to interfere with the transmembrane transport of key ions and the cell division process inside the algal cell; and the synergistic generator adopts phase synergistic technology to ensure that the electromagnetic field pulse and the ultrasonic wave peak are superimposed in phase in the target action area.

7. The unmanned water body ecological monitoring and management system according to claim 1, characterized in that, The unmanned surface vehicle platform integrates an autonomous navigation and obstacle avoidance subsystem, which includes: The real-time dynamic differential positioning unit adopts multi-frequency satellite signal reception technology and supports GPS, GLONASS and Beidou global navigation satellite systems. It achieves centimeter-level high-precision positioning through real-time differential correction and is widely used in autonomous driving and precision agriculture. The inertial measurement unit has a built-in three-axis gyroscope, three-axis accelerometer and three-axis magnetometer. The gyroscope is used to measure the change of angular velocity, the accelerometer detects the linear acceleration and the magnetometer provides the orientation reference, which together realize the real-time monitoring of the carrier's attitude. The multi-source data fusion processor executes an extended Kalman filter algorithm to fuse the positioning data of the real-time dynamic differential positioning unit and the attitude data of the inertial measurement unit in real time, eliminating noise and errors, and outputting high-frequency, high-precision vehicle attitude information, supporting hundreds of updates per second, and ensuring stable performance in dynamic environments. Millimeter-wave radar uses the millimeter-wave band for ranging and velocity detection. It has strong penetration capabilities and is suitable for obstacle detection in adverse weather conditions such as rain and fog. Solid-state LiDAR uses solid-state technology to achieve high-speed rotating scanning and generate high-resolution 3D point cloud data for environmental modeling and obstacle recognition. Infrared proximity sensors detect nearby objects by emitting and receiving infrared rays, providing short-range obstacle warnings and enhancing safety.

8. The unmanned water body ecological monitoring and management system according to claim 7, characterized in that, The predictive decision processing module internally incorporates a hierarchical obstacle avoidance decision logic, which is specifically implemented as follows: when the millimeter-wave radar detects an obstacle within a range of 20 meters to 10 meters ahead, the system triggers a first-level warning and invokes an improved A* path planning algorithm to recalculate and generate a locally optimal obstacle avoidance route under global path constraints; when the obstacle enters a range of 10 meters to 3 meters, the system triggers a second-level warning and controls the unmanned surface vehicle platform's propulsion system to reduce its speed to 30% of a preset safety threshold; when the obstacle is less than 3 meters away, the system triggers a third-level emergency command, executing full-power reverse thrust braking or maximum angle evasive maneuvering.

9. The unmanned water body ecological monitoring and management system according to claim 7, characterized in that, The shore-based control center includes a model iterative optimization module; The model iterative optimization module adopts an incremental learning algorithm, which is set to automatically call the newly added valid data pairs in the time-series historical database of the shore-based control center every quarter. These data pairs include water quality parameter comparison data before and after the governance intervention and their related environmental parameters. The model iterative optimization module uses these effective data pairs to retrain and fine-tune the long short-term memory network temporal prediction submodule and the convolutional neural network image recognition submodule deployed in the predictive decision processing module, thereby constructing an intelligent closed loop of "perception-prediction-decision-execution-feedback-optimization".

Images