Water quality monitoring buoy

Abstract

The water quality monitoring buoy comprises a sensor mounting part, a floater and a balance weight, the sensor mounting part is arranged in a gravity center hole of the floater and can float up and down along with buoyancy of the floater, the balance weight is mounted at the bottom of the floater to lower the overall gravity center, and the water quality monitoring buoy further comprises an active buoyancy adjusting unit and a control system; the active buoyancy adjusting unit comprises a water pump, a high-pressure gas cylinder, a high-pressure gas pump, an electromagnetic valve and an exhaust valve, the water pump is used for pumping water into or from the interior of the floater, the high-pressure gas cylinder is communicated with the interior of the floater through the electromagnetic valve, the high-pressure gas pump is used for supplementing compressed air into the high-pressure gas cylinder, and the exhaust valve is used for exhausting air in the floater; the control system is used for controlling the active buoyancy adjusting unit to achieve floating, diving and hovering of the buoy and controlling the self-cleaning structure in the sensor mounting part to work. According to the invention, the probability that the buoy is trapped can be reduced, so that the accuracy of monitoring data and the continuity of equipment operation are improved.

Description

Technical Field

[0001] This application relates to the field of water quality monitoring, and more particularly to a water quality monitoring buoy. Background Technology

[0002] With increasingly stringent requirements for environmental protection and water resource management, long-term, continuous, and multi-point real-time monitoring of water quality in rivers, lakes, and other water bodies has become an important need. In practical applications, water levels often fluctuate significantly due to seasonal, climatic, and human-induced changes. Furthermore, issues such as silt accumulation at the bottom of water bodies and the attachment of suspended solids can affect the normal operation of monitoring equipment. Therefore, there is an urgent need for water quality monitoring equipment that can adapt to dynamic changes in water levels and effectively prevent sensor contamination.

[0003] Currently, there is a type of water quality monitoring buoy that combines a fixed float with a retractable sensing arm. This type of buoy typically uses a mechanical telescopic rod located beneath the float to carry the sensor. A motor drives the telescopic rod to raise and lower, allowing the sensor to extend into different water depths for measurement. This approach achieves multi-depth monitoring through the longitudinal movement of the mechanical structure and relies on the raising of the telescopic rod to move the sensor away from the riverbed, avoiding direct contact with silt.

[0004] However, the above-mentioned solutions rely on complex mechanical transmission mechanisms to achieve depth adjustment. Long-term underwater operation is prone to corrosion, biological adhesion, or silt intrusion, which can cause the mechanism to jam, wear, or even fail, resulting in poor reliability. At the same time, simple mechanical lifting cannot provide effective adaptive adjustment capabilities when the overall attitude of the buoy becomes unstable or encounters sudden sludge adsorption. In low water levels or environments with a lot of silt, there is still a risk of sensor contamination or equipment getting trapped, which affects the accuracy of monitoring data and the continuity of equipment operation. Summary of the Invention

[0005] This application provides a water quality monitoring buoy, which aims to improve the buoy's adaptive adjustment ability when the overall attitude is unstable or encounters sudden sludge adsorption, reduce the probability of the buoy getting trapped, and thus improve the accuracy of monitoring data and the continuity of equipment operation.

[0006] To address the aforementioned technical problems, this application provides the following technical solutions:

[0007] A water quality monitoring buoy includes: a sensor mounting part, a float and a counterweight. The sensor mounting part is disposed in the center-of-gravity cavity of the float and can float up and down with the buoyancy of the float. The counterweight is installed at the bottom of the float to lower the overall center of gravity. The water quality monitoring buoy also includes an active buoyancy adjustment unit and a control system. The active buoyancy adjustment unit includes a water pump, a high-pressure gas cylinder, a high-pressure air pump, a solenoid valve, and an exhaust valve. The water pump is used to pump water into or out of the float. The high-pressure gas cylinder is connected to the inside of the float through the solenoid valve. The high-pressure air pump is used to replenish compressed air into the high-pressure gas cylinder. The exhaust valve is used to discharge air from the inside of the float. The control system is used to control the active buoyancy adjustment unit to enable the buoy to rise, dive, and hover, and to control the operation of the self-cleaning structure in the sensor mounting section.

[0008] The beneficial effects of this application are as follows: 1. This application controls the air and water volume inside the float through the active buoyancy adjustment unit (including a water pump, a high-pressure air cylinder, a high-pressure air pump, a solenoid valve, and an exhaust valve), thereby changing the overall buoyancy and draft of the buoy, achieving ascent, descent, and hovering. This fluid displacement-based adjustment method reduces the long-term exposure and movement of complex mechanical transmission components underwater, and to a certain extent reduces the risk of mechanism jamming due to corrosion and sediment intrusion, thus relatively improving the operational reliability of the equipment.

[0009] 2. In the event of a sudden sludge adsorption risk or the need to adapt to rapid water level changes, the control system can coordinate the water pump and air cylinder to quickly change buoyancy, allowing the buoy to detach from its restraints or adjust to a safe depth. Compared to mechanical solutions that can only move a portion of the sensor arm, this provides a more fundamental and effective attitude and position adjustment capability, thereby relatively enhancing adaptability and risk resistance in complex hydrological environments.

[0010] 3. The integrated control system of this application, which integrates the active buoyancy adjustment unit and the self-cleaning structure, enables the equipment to automatically perform depth adjustment and cleaning maintenance actions based on preset strategies or sensor feedback. This integrated autonomous management function reduces reliance on fixed mechanical structures and shifts to maintaining monitoring efficiency through systematic state control, providing a more adaptable technical path for solving the dynamic pollution and sedimentation problems encountered in long-term continuous monitoring. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of the overall structure of a water quality monitoring buoy provided in an embodiment of this application; Figure 2A cross-sectional structural diagram of a water quality monitoring buoy provided in this application embodiment; Figure 3 This is a schematic diagram of the exploded structure of a water quality monitoring buoy provided in an embodiment of this application; Figure 4 This application provides a schematic diagram of the structure of a sensor mounting section in a water quality monitoring buoy. Figure 5 This is a schematic diagram of a self-cleaning structure in a water quality monitoring buoy provided in an embodiment of this application.

[0013] Reference numerals: 1. Anchor fixing hole; 2. Sensor probe; 3. Water surface diagram; 4. Silt layer diagram; 5. Sensor mounting part; 6. Float; 7. Base plate; 8. Counterweight; 9. Wiring harness protective cover; 10. Handle; 11. Sealed cavity; 12. Turbidity sensor; 13. Self-cleaning structure; 14. Outer anti-fouling net; 15. Sensor mounting base; 16. Servo motor; 17. Rocker arm; 18. Brush; 19. High-pressure air pump; 20. Water pump; 21. Tuning fork level sensor; 22. High-pressure air cylinder; 23. Solenoid valve; 24. Exhaust valve; 25. Image acquisition module. Detailed Implementation

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] Reference Figures 1 to 5 This application provides a water quality monitoring buoy, including: a sensor mounting part 5, a float 6 and a counterweight 8. The sensor mounting part 5 is disposed in the center of gravity cavity of the float 6 and can float up and down with the buoyancy of the float 6. The counterweight 8 is installed at the bottom of the float 6 to lower the overall center of gravity. The water quality monitoring buoy also includes an active buoyancy adjustment unit and a control system. The active buoyancy adjustment unit includes a water pump 20, a high-pressure gas cylinder 22, a high-pressure air pump 19, a solenoid valve 23, and an exhaust valve 24. The water pump 20 is used to pump water into or out of the float 6. The high-pressure gas cylinder 22 is connected to the inside of the float 6 through the solenoid valve 23. The high-pressure air pump 19 is used to replenish compressed air into the high-pressure gas cylinder 22. The exhaust valve 24 is used to discharge the air inside the float 6. The control system is used to control the active buoyancy adjustment unit to realize the buoy's rising, diving and hovering, and to control the operation of the self-cleaning structure 13 in the sensor mounting part 5.

[0016] Reference Figure 1 and Figure 2 The water quality monitoring buoy includes a sensor mounting section 5, a float 6, and a counterweight 8. The sensor mounting section 5 houses multiple sensors, which can be of the same type, different types, or a combination of both. A center-of-gravity cavity, penetrating both the upper and lower surfaces of the float 6, is located in the center of the float 6. The sensor mounting section 5 is installed within this cavity, with its upper portion extending beyond the upper surface of the float 6. To ensure better sinking and stability, a counterweight 8 is installed below the float 6, lowering the overall center of gravity of the buoy. The specific number and installation location of the counterweights 8 can be determined based on actual conditions.

[0017] To enable the buoy to better adapt to changes in the water, the buoy also includes an active buoyancy adjustment unit and a control system. The active buoyancy adjustment unit includes a water pump 20, a high-pressure air cylinder 22, a high-pressure air pump 19, a solenoid valve 23, and an exhaust valve 24. The water pump 20 pumps water into or out of the buoy 6. The high-pressure air cylinder 22 is connected to the inside of the buoy 6 via the solenoid valve 23. The high-pressure air pump 19 replenishes compressed air into the high-pressure air cylinder 22. The exhaust valve 24 expels air from inside the buoy 6. The control system controls the active buoyancy adjustment unit to achieve the buoy's rising, sinking, and hovering, and controls the operation of the self-cleaning structure 13 within the sensor mounting section 5.

[0018] In one embodiment, refer to Figure 1 and Figure 2 The float 6 is a hollow annular cylinder. The float 6 includes an upper surface, a lower surface, an inner surface, an outer surface, and a buoyancy adjustment cavity composed of the four surfaces. The buoyancy adjustment cavity is a closed structure and can be connected to the outside world through the active buoyancy adjustment unit. The water pump 20 is located inside the center of gravity cavity and is fixedly connected to the inner surface of the float 6. For example, the water pump 20 is fixed to the inner surface of the float 6 through a connecting frame (not shown in the figure). The water pump 20 is connected to the buoyancy adjustment cavity through a pipe. The water pump 20 can be rotated to draw in or discharge water in the buoyancy adjustment cavity. Alternatively, two water pumps 20 can be set, namely an inlet pump 20 and a drain pump 20. They are also connected to the buoyancy adjustment cavity through pipes. The outlet of the inlet pump 20 is connected to the pipe, and the inlet of the drain pump 20 is connected to the pipe. This allows the inlet pump 20 to draw water from an external water source into the buoyancy adjustment cavity, and the drain pump 20 to discharge the water from the buoyancy adjustment cavity to the outside. This allows the weight of the float to be changed by adjusting the amount of water in the buoyancy adjustment cavity, thereby achieving the raising and lowering of the float.

[0019] The high-pressure gas cylinder 22, high-pressure gas pump 19, solenoid valve 23, and exhaust valve 24 can all be located inside the buoyancy adjustment cavity, or they can be located outside the buoyancy adjustment cavity. Of course, they can also be partially located inside the buoyancy adjustment cavity and partially located outside the buoyancy adjustment cavity. For example, the high-pressure gas pump 19 can be located outside the buoyancy adjustment cavity, while the high-pressure gas cylinder 22, solenoid valve 23, and exhaust valve 24 can be located inside the buoyancy adjustment cavity.

[0020] This embodiment is illustrated using the example where the high-pressure gas cylinder 22, high-pressure gas pump 19, solenoid valve 23, and exhaust valve 24 are all located inside the buoyancy adjustment cavity. (Refer to...) Figure 2 The air inlet of the high-pressure air pump 19 is connected to the outside through an air inlet pipe, and the air outlet of the high-pressure air pump 19 is connected to the air inlet of the high-pressure gas cylinder 22 through a connecting air pipe. The air outlet of the high-pressure gas cylinder 22 is connected to the solenoid valve 23. Thus, the high-pressure air pump 19 can pump outside air into the high-pressure gas cylinder 22, which can store gas. At the same time, when it is necessary to adjust the buoyancy of the buoy, the air in the high-pressure gas cylinder 22 can enter the buoyancy adjustment cavity by adjusting the switch of the solenoid valve 23. Meanwhile, the buoyancy adjustment cavity is connected to the outside through an exhaust valve 24. The gas in the buoyancy adjustment cavity can be discharged by adjusting the switch of the exhaust valve 24, thereby reducing the buoyancy of the buoy.

[0021] During the process of adjusting the buoyancy of the buoy, the water pump 20, high-pressure gas cylinder 22, high-pressure air pump 19, solenoid valve 23 and exhaust valve 24 can work together. For example, when it is necessary to reduce the buoyancy of the buoy, the exhaust valve 24 is in the open state to expel the air inside the buoyancy adjustment cavity. At the same time, the water pump 20 pumps water from the external water source into the buoyancy adjustment cavity, which can accelerate the expulsion of air from the buoyancy adjustment cavity to the outside, thereby reducing the buoyancy of the buoy.

[0022] When the buoyancy of the buoy needs to be increased, the exhaust valve 24 is closed, preventing air from escaping from the buoyancy adjustment cavity. Simultaneously, the solenoid valve 23 opens, pumping air from the high-pressure gas cylinder 22 into the buoyancy adjustment cavity. When the air in the high-pressure gas cylinder 22 is insufficient—meaning the pressure difference between the high-pressure gas cylinder 22 and the buoyancy adjustment cavity cannot allow air to enter—the high-pressure air pump 19 activates, drawing outside air into the high-pressure gas cylinder 22 to increase its internal pressure, thus enabling air to enter the buoyancy adjustment cavity. When the required buoyancy is met, the solenoid valve 23 closes. Simultaneously, when air enters the buoyancy adjustment cavity, the water pump 20 activates, pumping water out of the cavity, thereby reducing the weight of the buoy and facilitating a rapid increase in buoyancy.

[0023] The buoy also includes a control system, which is electrically connected to the water pump 20, high-pressure gas cylinder 22, high-pressure air pump 19, solenoid valve 23 and vent valve 24 via wired and / or wireless means. The control system can receive signals sent by the water pump 20, high-pressure gas cylinder 22, high-pressure air pump 19, solenoid valve 23 and vent valve 24 or send signals to the water pump 20, high-pressure gas cylinder 22, high-pressure air pump 19, solenoid valve 23 and vent valve 24, thereby controlling the active buoyancy adjustment unit to achieve the buoy's ascent, descent and hovering.

[0024] The buoy also includes a power supply system (not shown in the figure), which provides power to the control system, water pump 20, high-pressure gas cylinder 22, high-pressure gas pump 19, solenoid valve 23, and exhaust valve 24. The power supply system can be any known existing power supply system, such as solar power or battery power. Since power supply systems are widely known, this embodiment will not elaborate on them.

[0025] Optional, refer to Figures 2 to 4 The sensor mounting part 5 includes a sealed cavity 11, a turbidity sensor 12, a self-cleaning structure 13, and an outer anti-fouling net 14; the sealed cavity 11 is used to install the control circuit of the control system; the lowest point of the sensor probe 2 of the turbidity sensor 12 is higher than the bottom surface of the counterweight 8; the self-cleaning structure 13 is located between the counterweight 8 and the outer anti-fouling net 14, and is used to clean the sensor probe 2 of the turbidity sensor 12; the outer anti-fouling net 14 is disposed outside the sensor mounting part 5.

[0026] Reference Figure 4 The sensor mounting section 5 includes a sealed cavity 11, a turbidity sensor 12, a self-cleaning structure 13, and an outer anti-fouling mesh 14. The sealed cavity 11 consists of a disc-shaped lower part, a cylindrical middle part, and an annular upper part. The upper, lower, and middle parts are coaxially arranged, with the diameter of the middle part equal to the inner diameter of the upper part and the outer diameter of the upper part equal to the diameter of the lower part. The central cavity formed by the upper, middle, and lower parts is the sealed cavity 11. The control circuitry of the control system, such as circuit boards, chips, and other components, is installed inside the sealed cavity 11.

[0027] A top cover is fixed to the upper part by bolts. The top cover is disc-shaped and its diameter is the same as the outer diameter of the upper part. A sealing ring is also provided between the top cover and the upper part. The sealing cavity 11 is sealed to the outside through the top cover and the sealing ring, thereby minimizing the probability of corrosion of the control circuit and extending its service life.

[0028] Meanwhile, for ease of operation, a handle and a wire harness protection cover 9 are fixed on the upper surface of the cover. The handle makes it easy for the operator to lift the buoy, and the wire harness protection cover 9 can protect the connecting wires extending from the sealed cavity 11. An opening is provided on the surface of the cover or the middle part to facilitate the wire harness to extend from the sealed cavity 11. At the same time, in order to improve the sealing performance, the wire harness and the opening can be sealed by adding sealant.

[0029] The outer anti-fouling net 14 is cylindrical, and its surface, located outside the float 6 / center of gravity cavity, has several evenly distributed through holes. This allows water to enter the outer anti-fouling net 14 and contact the sensor while simultaneously blocking impurities from entering as much as possible. A connecting ring is also fixed to the upper surface of the outer anti-fouling net 14. The connecting ring is circular, with an outer diameter the same as the lower part's diameter and an inner diameter the same as the outer anti-fouling net 14's diameter. The connecting ring is fixed to the lower part by bolts.

[0030] The turbidity sensor 12 and other sensors are fixedly connected to the lower part through the sensor mounting base 15. The sensor mounting base 15 is fixed to the lower surface of the lower part by bolts. The sensors are fixed by the fixing ring on the sensor mounting base 15, so that various sensors can be placed inside the outer anti-pollution net 14. At the same time, the lowest point of the sensor probe 2 of the turbidity sensor 12 is higher than the bottom surface of the counterweight 8.

[0031] The self-cleaning structure 13 is located between the counterweight 8 and the outer anti-fouling net 14, and is used to clean the sensor probe 2 of the turbidity sensor 12.

[0032] Optional, refer to Figure 5 The self-cleaning structure 13 includes a sensor mounting base 15, a servo motor 16, a rocker arm 17, and a brush 18. The servo motor 16 drives the rocker arm 17 to move the brush 18 back and forth to clean the surface of the sensor probe 2 of the turbidity sensor 12.

[0033] The sensor mounting base 15 is fixed to the lower part by bolts. The servo motor 16 is installed at the bottom of the sensor mounting base 15. The output shaft of the servo motor 16 is fixedly connected to the rocker arm 17. The rocker arm 17 can rotate synchronously with the rotation of the output shaft of the servo motor 16. A brush 18 is fixedly connected to the rocker arm 17. During operation, the servo motor 16 can rotate back and forth, thereby driving the rocker arm 17 to swing back and forth, so that the brush 18 swings back and forth to clean the sensor probe 2.

[0034] The sensor is detachably connected to the sensor mounting base 15, and the lowest point of the sensor probe 2 is lower than the highest point of the bristles of the brush 18, but higher than the body of the brush 18. When the servo motor 16 is started, the bristles of the brush 18 can swing to the lowest point of the sensor probe 2 and clean the probe by reciprocating motion.

[0035] Optionally, it also includes a base plate 7 and an anchor fixing hole 1. The base plate 7 cooperates with the sensor mounting part 5 to clamp and fix the float 6. The anchor fixing hole 1 is used to connect to the pre-buried anchor point on the bank or riverbed through a steel wire rope.

[0036] A base plate 7 is provided below the float 6. The basic diameter is greater than or equal to the outer diameter of the float 6. The edge of the base plate 7 is provided with a protrusion that extends out of the base plate 7. An anchor fixing hole 1 is provided on the surface of the protrusion. The anchor fixing hole 1 is used to connect to the pre-buried anchor point on the bank or riverbed through a steel wire rope to prevent the float from being washed away by the water flow, thereby ensuring the stability of the float.

[0037] The bottom of the outer anti-fouling net 14 is fixed with a base plate. The center of the base plate has an opening that can accommodate the sensor to pass through. Several bolt fixing holes are opened around the edge of the base plate. A fixing plate with the same shape and size as the base plate is set below the base plate. The base plate and the fixing plate are fixed together by bolts.

[0038] The substrate 7 has several through holes on its surface. A fixing ear is fixed on the lower surface of the substrate 7 near the through holes. The fixing plate and the fixing ear are fixed together by a connecting plate. The connecting plate is fixed to the fixing plate and the fixing ear by bolts, so that the substrate 7 can fix the sensor mounting part 5 by the connecting plate.

[0039] Meanwhile, an annular positioning ring is provided on the inner surface of the float 6 along the circumference of the inner surface. The outer diameter of the positioning ring is equal to the diameter of the inner ring of the float 6, and the inner diameter of the positioning ring is smaller than the outer diameter of the fixed plate. The positioning ring is located between the fixed plate and the bottom plate, and it abuts against the upper surface of the fixed plate and the lower surface of the bottom plate. Figure 1 (The bolts between the fixing plate and the base plate are not tightened), thus allowing the fixing plate and the base plate to fix and position the positioning ring, thereby fixing the float 6. Furthermore, through the interconnection of the base plate 7, fixing lugs, connecting plate, fixing plate, positioning ring, and base plate, the sensor mounting part 5, the float 6, and the base plate are mutually fixed. Furthermore, in one embodiment, the counterweight 8 is located on the lower surface of the base plate and is fixed to the base plate 7 by bolts. The surface of the base plate 7 is provided with several threaded holes. When in use, a suitable threaded hole can be selected to fix the counterweight 8, and the mass and quantity of the configuration can be adjusted according to the requirements, thereby minimizing the probability of the buoy tilting and improving the stability of the buoy.

[0040] The lowest point of the sensor probe 2 of the turbidity sensor 12 is located above the substrate 7 and inside the center of gravity cavity. This ensures that the sensor can perform water quality monitoring normally while facilitating the cleaning of the sensor probe 2 by the self-cleaning structure 13. At the same time, when the counterweight 8 comes into contact with the silt on the riverbed, the sensor probe 2 is prevented from being covered by silt as much as possible because the lowest point of the sensor probe 2 is located above the substrate 7, thus ensuring normal monitoring by the sensor as much as possible.

[0041] Optionally, S1 also includes a tuning fork level sensor 21, which is installed at the bottom of the buoy and is used to determine whether the buoy is covered by silt by judging the amplitude and vibration frequency of the tuning fork level sensor 21. S2. When it is detected that the float is covered by silt, a signal is sent to the control system. The control system controls the water pump 20 to drain the water inside the float 6 and controls the solenoid valve 23 to release the compressed air in the high-pressure gas cylinder 22 into the float 6, so as to increase the buoyancy of the float 6 and make the float break free from the silt.

[0042] S3. Optionally, the control system is further configured to: based on the monitoring data of the turbidity sensor 12, automatically adjust and suspend the buoy at a water depth that meets the preset water quality conditions by controlling the active buoyancy adjustment unit.

[0043] Preferably, the specific implementation process of S1 is as follows: A tuning fork level sensor 21 installed at the bottom of the buoy has a core component of a mechanical resonator driven by piezoelectric ceramic. The lowest end of the tuning fork level sensor 21 extends beyond the bottom surface of the base, and the height of the lowest end of the tuning fork level sensor 21 is less than or equal to the lowest point of the counterweight 8. In clean water, the control system applies a specific frequency drive signal to the tuning fork level sensor 21, causing it to generate stable mechanical vibration. At this time, the sensor will feedback a characteristic fundamental frequency and a characteristic amplitude. When the buoy sits on the bottom or sinks, causing the tuning fork level sensor 21 to contact or sink into the riverbed silt, the physical properties of the medium surrounding the sensor oscillator change significantly. Silt, as a viscous non-Newtonian fluid, will have a strong damping effect on the vibration of the oscillator. This application uses the sensor's built-in measurement circuit to collect the vibration feedback signal of the tuning fork level sensor 21 under drive in real time. This vibration feedback signal contains real-time vibration frequency and vibration amplitude information. The control system compares the collected real-time vibration frequency with the pre-calibrated characteristic fundamental frequency in clear water to calculate the first frequency offset. Simultaneously, it compares the collected real-time vibration amplitude with the pre-calibrated characteristic amplitude to calculate the first amplitude attenuation. Subsequently, the control system jointly judges the first frequency offset and the first amplitude attenuation with a set of experimentally predetermined frequency offset thresholds and amplitude attenuation thresholds corresponding to the "mud-covered" state. If the first frequency offset exceeds the frequency offset threshold, and the first amplitude attenuation simultaneously exceeds the amplitude attenuation threshold, the vibration state of the tuning fork level sensor 21 is determined to be consistent with the characteristics of being enveloped by a high-viscosity medium, thereby generating a "mud-covered confirmation" status signal. This embodiment differs from traditional methods that rely solely on a single parameter (such as hydrostatic pressure) to determine whether a buoy is stuck on the bottom. By jointly analyzing the two interrelated dynamic physical quantities of vibration frequency and amplitude, it can more reliably distinguish whether the buoy is in contact with a hard riverbed or stuck in soft mud, thus improving the reliability of identifying the "trapped" risk.

[0044] Preferably, in the specific implementation of S2, when the control system receives the aforementioned "sludge coverage confirmation" status signal, it triggers a preset emergency buoyancy boosting and escape procedure. This procedure first sends a control command to the water pump 20, driving it to operate at maximum power to discharge the water inside the float 6 through the drainage pipe. This step directly reduces the total mass of the buoy system. Immediately afterwards, the control system sends an opening command to the solenoid valve 23 connected to the high-pressure gas cylinder 22. The compressed gas stored in the high-pressure gas cylinder 22 is rapidly released into the cavity inside the float 6 under the action of the pressure difference. This process achieves two key physical effects: first, the continuously injected gas further displaces and replaces the residual water inside the float 6; second, the injection of a large amount of gas significantly increases the volume of water displaced by the float 6. The combined effect of these two actions—drainage and inflation—directly results in a rapid increase in the net buoyancy of the entire buoy in a short period. The increase in net buoyancy is designed to overcome the sum of the viscous adhesion and static friction forces exerted by the silt on the bottom of the buoy (especially the counterweight 8 and the tuning fork level sensor 21). During the escape procedure, the control system continuously monitors the real-time vibration frequency and amplitude fed back by the tuning fork level sensor 21. Once the real-time vibration frequency is detected to return to the fundamental frequency characteristic of clear water, and the real-time vibration amplitude recovers to a level close to that of clear water, it indicates that the tuning fork has escaped the silt environment. The control system then generates an "escape successful" signal, stops the water pump 20, and closes the solenoid valve 23. The buoy will then rise to a safe water depth using the increased net buoyancy it gains at this time. This embodiment actively and rapidly changes the overall buoyancy of the system to counteract the adhesion force, unlike traditional designs that passively wait for the water level to rise or rely on additional mechanical lifting mechanisms. This provides the buoy with active survival capabilities in low-water-level silt environments.

[0045] Preferably, in one scenario, when S3 is specifically implemented, this application aims to enable the buoy to autonomously find and maintain itself in a water layer with a low risk of sensor contamination. After a complete water quality profile monitoring operation, the control system retrieves a series of turbidity monitoring data collected by the turbidity sensor 12 at different depth points during the process. These turbidity monitoring data, together with the corresponding depth information at the time of collection, constitute a "depth-turbidity" relationship dataset. The control system presets an "acceptable turbidity threshold," which is a parameter used to characterize the relative clarity of the water body, set based on historical data and the sensor's contamination tolerance. Subsequently, the control system analyzes the "depth-turbidity" relationship dataset. Its core logic is: starting from the shallowest water depth, it traverses downwards to find the first depth point whose turbidity monitoring data is lower than the "acceptable turbidity threshold," and records this depth point as a "candidate clean depth." If no point below the threshold is found after traversing all data, the depth point with the lowest turbidity monitoring data is taken as the "candidate clean depth." After determining the "candidate clean depth," the control system calculates the depth difference between the current buoy depth and the "candidate clean depth." Based on the sign and magnitude of the depth difference, the control system sends instructions to the active buoyancy adjustment unit: if surfacing is required, the solenoid valve 23 is opened to inject gas and the water pump 20 may be activated to assist in drainage; if submersion is required, the water pump 20 is controlled to pump in an appropriate amount of water and the vent valve 24 may be opened. Through closed-loop feedback control, the actual depth of the buoy gradually approaches and stabilizes at the "candidate clean depth." Thereafter, the buoy will hover at this depth for long-term monitoring until the next preset profile monitoring cycle is triggered. This embodiment differs from traditional fixed-depth or simply drift-with-the-wave buoys, and also from depth adjustments solely for collision avoidance. By using water quality parameters (turbidity) as the direct decision-making basis for depth adjustment, it couples the two objectives of "monitoring" and "sensor self-maintenance," enabling the device to actively avoid high-turbidity, high-pollution-risk water layers, reducing the rate at which sensors are rapidly contaminated from the environmental source, thereby reducing clean energy consumption and improving the effectiveness of data quality in long-term deployment.

[0046] Optionally, S4, the sensor mounting part 5 also integrates an image acquisition module 25, and the image acquisition module 25 and the turbidity sensor 12 constitute a dual-mode detection system; S5. The image acquisition module 25 is used to acquire image information of the surface of the sensor probe 2 at a preset frequency. The control system has a built-in image recognition module for identifying and classifying dirt in the acquired image information. S6. The turbidity sensor 12 is used to monitor the optical parameters of the water body within a preset range of the sensor probe 2. The control system fuses the optical parameter monitoring data with the image recognition results to generate dirt status judgment information.

[0047] Preferably, the specific implementation process of S4 to S6 is as follows: This application integrates a miniature image acquisition module 25 (such as an image sensor or other camera) and a laser scattering turbidity sensor 12 in the sensor mounting part 5. The two are physically close to each other and their fields of view / monitoring range both point towards the surface of the sensor probe 2 and the water body adjacent to it, thus forming a dual-mode detection system. Specifically, in one embodiment, the miniature image acquisition module 25 is fixed on the upper surface of the base and faces the sensor probe 2. This embodiment differs from the single sensing mode and aims to improve the reliability of dirt state judgment through multi-dimensional information complementarity.

[0048] In one scenario, specifically in the implementation of S5, the image acquisition module 25 acquires digital images of the surface of the sensor probe 2 according to a preset periodic instruction (e.g., at regular intervals), obtaining the raw probe surface image. Due to the complex underwater ambient light, the raw probe surface image is first sent to an image preprocessing unit. This image preprocessing unit performs a series of operations, including dehazing based on dark channel priors to improve clarity, and using edge enhancement algorithms (such as the Sobel operator) to enhance the boundary contrast between the dirt and the probe body, outputting the preprocessed probe surface image. Subsequently, the preprocessed probe surface image is input into a lightweight convolutional neural network model, which serves as the core of the image recognition module. This convolutional neural network model, after training, is able to extract and recognize visual features representing different types of dirt from the image (such as the viscous texture of biological slime, the discrete dots of inorganic particles, and the flocculent green clumps of algae). The final layer of the convolutional neural network model outputs a probability distribution vector of dirt type, where each dimension of the vector corresponds to a preset dirt type (such as "biological slime", "inorganic particles", "algae attachment", "clean"), and its value represents the confidence that the image belongs to that category. This vector is the image recognition result.

[0049] In one embodiment, the lightweight convolutional neural network model can be trained in the following manner: Construction of the training dataset: First, a dedicated image dataset for dirt identification of sensor probe 2 needs to be constructed. This dataset is obtained as follows: In long-term deployments in laboratories and different field waters, the image acquisition module 25 is used to collect a large number of raw sample images under various known conditions such as "clean," "covered with biological slime," "attached with inorganic particles," and "proliferating with algae." Each raw sample image is labeled with its corresponding real dirt type label manually or through auxiliary detection equipment (such as microscopic observation), forming an "image-label" pair.

[0050] Data Preprocessing and Augmentation: The original sample images in the aforementioned dataset undergo the same preprocessing procedures as described in S5, including dehazing based on dark channel priors and edge enhancement, to obtain preprocessed training images. To improve the model's generalization ability, further data augmentation operations are performed on the preprocessed training images, including random rotation (e.g., within ±10 degrees), brightness and contrast fine-tuning, and adding noise simulating suspended matter in water, generating an augmented training image set.

[0051] Network Architecture Selection and Lightweight Design: A lightweight convolutional neural network architecture suitable for embedded devices (such as MobileNetV2 or a variant of ShuffleNet) was selected as the initial model. For the dirt recognition task, the last classification layer of this architecture was modified to match its output dimension with the number of dirt type categories (e.g., 4 categories: clean, bio-sludge, inorganic particles, and algae attachment). The lightweight feature was primarily achieved by replacing standard convolutions with depthwise separable convolutions and introducing channel pruning constraints early in the network, aiming to reduce the number of parameters and computational cost while maintaining a high recognition rate.

[0052] Model Training and Optimization: Loss Function: The Cross-Entropy Loss function is used as the training objective. This function measures the difference between the model's predicted dirt type probability distribution vector and the true label. Optimizer: The Adam Optimizer is used to update the model parameters. Training Process: The augmented training image set is proportionally divided into training and validation sets. The images from the training set are input into the network model described above. Forward propagation is used to obtain the predicted probability distribution; the cross-entropy loss between the prediction and the true label is calculated; the gradient of the loss function with respect to the parameters of each layer of the model is calculated using the backpropagation algorithm; the model parameters are updated based on the gradient using the Adam optimizer. This process is iterated, and the accuracy is monitored on the validation set. Training is stopped when the accuracy on the validation set no longer improves to prevent overfitting, resulting in the initial trained model.

[0053] Model fine-tuning and deployment preparation: Using small batches of image data collected on real buoy devices, which more closely resemble the actual application scenario, the initial trained model was fine-tuned to adapt to actual lighting and water quality conditions. After fine-tuning, the model underwent quantization operations, such as converting the weights from 32-bit floating-point numbers to 8-bit integers, and further pruning was performed using the channel importance information recorded during training. The final result is a lightweight convolutional neural network model file that can run efficiently in the buoy embedded control system.

[0054] Channel importance information refers to a quantitative assessment of the contribution or sensitivity of each output channel (also called a feature map) of the intermediate layers (usually convolutional layers) of the network to the final task (such as classification). It is used to identify and eliminate redundant channels that contribute little to the task, thereby achieving model compression (pruning) and acceleration, and is a key basis in the process of model lightweighting. For example, suppose there is a simplified CNN for dirt recognition, where a certain convolutional layer is as follows: Network layer structure: This convolutional layer has 4 input channels and 3 output channels (i.e., it uses 3 different convolutional kernels). Convolutional kernel parameters: The size of each convolutional kernel is 3x3x4 (height x width x number of input channels). Let K_i represent the i-th convolutional kernel, where i takes values ​​of 1, 2, or 3.

[0055] K1 (convolution kernel 1): The absolute values ​​of all its weights are generally large (e.g., the sum is 15.2). This means that the kernel will produce a strong response to specific combinations of input features, which may correspond to the key pattern for identifying "algal flocculent texture".

[0056] K2 (kernel 2): ​​Its weights have a moderate sum of absolute values ​​(e.g., 8.7). It may correspond to an auxiliary feature.

[0057] K3 (convolutional kernel 3): The sum of the absolute values ​​of its weights is very small (e.g., 0.9). This indicates that the kernel is not sufficiently activated after training, and the features it extracts have little impact on the final classification decision.

[0058] Calculate the importance of channels: Importance of channel 1 = L1 norm (K1) = 15.2, Importance of channel 2 = L1 norm (K2) = 8.7, Importance of channel 3 = L1 norm (K3) = 0.9.

[0059] Generate "Channel Importance Information": Based on the above calculations, an importance ranking list or vector can be obtained, for example: [Importance (Channel 1): 15.2, Importance (Channel 2): ​​8.7, Importance (Channel 3): 0.9]. This is the channel importance information for this layer. It clearly tells us that channel 1 is the most important, channel 2 is next, and channel 3 is the least important and is likely redundant.

[0060] Simultaneously, in S6, the laser scattering turbidity sensor 12 continuously monitors the optical scattering intensity of water within a preset short distance (e.g., a range of several centimeters) in front of the probe, obtaining a real-time optical scattering intensity sequence. The control system maintains a reference scattering intensity value for the sensor in clean water under clean conditions. By calculating the deviation between the real-time optical scattering intensity sequence and the reference scattering intensity value (e.g., calculating the mean difference or variance change within a sliding window), an optical pollution trend index is obtained, which reflects the potential coverage or interference trend of suspended matter in the water on the probe's optical window.

[0061] The control system employs a fusion decision logic that jointly analyzes the probability distribution vector of dirt types from image recognition results with optical contamination trend indicators. For example, when the probability value of the "biological slime" or "algae attachment" category exceeds its first probability threshold, and simultaneously the optical contamination trend indicator exceeds its first trend threshold, the fusion decision logic determines the presence of attached biological dirt. When the probability value of the "inorganic particles" category exceeds its second probability threshold, and the optical contamination trend indicator exhibits rapid fluctuations or a sustained high level, the fusion decision logic determines the presence of suspended particle interference. If the image recognition result indicates the highest probability of "clean," but the optical contamination trend indicator remains abnormal, it may indicate the presence of non-visually visible colloidal contamination outside the sensor's optical window, and the fusion decision logic generates a potential colloidal contamination warning. Any of the above decision outputs constitutes the final dirt status assessment information. This dual-mode information fusion method overcomes the limitations of single visual detection failing in turbid water or single optical detection being unable to distinguish dirt types.

[0062] Optionally, in step S7, the control system is configured to execute a corresponding cleaning strategy based on the type of dirt indicated by the dirt state judgment information; the cleaning strategy includes a mechanical scraping mode executed by the self-cleaning structure 13, a backwashing mode generated by water flow disturbance, and a combination of the mechanical scraping mode and the backwashing mode.

[0063] Preferably, in the specific technical implementation of S7, the control system internally presets a "dirt type-cleaning strategy" mapping table. When the control system receives the dirt status judgment information generated in step S6, it immediately queries this mapping table and triggers the corresponding cleaning strategy execution program.

[0064] If the dirt condition assessment indicates adherent biological dirt, the control system activates a "mechanical scraping and backwashing combination mode." This mode first sends a command to the micro-motor (e.g., servo motor 16) in the self-cleaning structure 13, driving it to move the scraping element (e.g., ceramic brush 18) to mechanically scrape the probe surface at a relatively high frequency (e.g., a higher number of reciprocating strokes than usual) for a first time period (e.g., several minutes) to physically peel off the adhered layer. During the intervals or after the mechanical scraping, the control system then uses the active buoyancy adjustment unit to control the buoy to perform several small, rapid ascents and descents at its current position. These rapid undulations generate significant water pressure changes and flow disturbances in the water surrounding the probe, creating an active backwashing flow on the probe surface, washing away the loosened dirt. This combination of mechanical scraping and backwashing, targeting the firmly attached nature of biological dirt, achieves both peeling and removal.

[0065] If the dirt condition assessment indicates suspended particle interference, the control system activates the "backwash mode." In this mode, the control system does not activate mechanical scraping but instead performs the aforementioned action of generating active backwash water flow through the buoy's movement. This effectively flushes away loose inorganic particles from the probe surface while avoiding unnecessary wear on the probe that might be caused by mechanical scraping.

[0066] If the dirt condition assessment indicates potential colloidal contamination, the control system may activate a gentle "maintenance backwash mode," which reduces the amplitude and frequency of fluctuations, only making periodic disturbances. This differentiated strategy based on the physical characteristics of the dirt, unlike traditional timed or single-intensity cleaning methods, optimizes energy consumption and extends the service life of mechanical components while ensuring cleaning effectiveness.

[0067] Optionally, S8, the control system further includes a prediction module, which is used to predict the dirt accumulation trend and generate cleaning scheduling instructions based on historical monitoring data and environmental parameters; the control system dynamically adjusts the timing and intensity of cleaning operations according to the cleaning scheduling instructions.

[0068] Optionally, in step S9, the prediction module is further configured to receive external environmental data, including weather forecast information; the prediction module modifies the cleaning scheduling instruction by analyzing the correlation between the weather forecast information and historical dirt accumulation data.

[0069] Preferably, in a specific implementation of S8 in a scenario, the control system of this application includes a prediction module, the core of which is a time-series prediction model. This prediction module continuously collects and stores multi-dimensional historical monitoring data and environmental parameters to form a historical dataset. The historical dataset includes at least: a sequence of fouling status judgment information arranged by timestamps (reflecting the type and intensity of fouling occurrence), a sequence of turbidity sensor readings, a sequence of water temperature sensor readings, a sequence of dissolved oxygen sensor readings for the same time period, and time context labels such as season and day / night.

[0070] The prediction module uses a time-series prediction model (e.g., a lightweight long short-term memory network restructured for embedded devices, i.e., a Lightweight LSTM) to learn from historical datasets. This Lightweight LSTM network, through its input, forget, and output gates, learns dynamic patterns of correlation between fouling accumulation rates and water quality parameters (such as turbidity and temperature) and temporal context (such as season, whether it's summer, or whether it's flood season). The trained model takes historical data and environmental parameters over a recent time window as input, and its output is a predicted fouling accumulation curve for a future time period (e.g., 12 hours). This curve represents the trend of the predicted fouling index or cleaning demand over time. Based on this predicted fouling accumulation curve, the prediction module pre-calculates the future time point when the fouling level is expected to exceed the cleaning threshold and generates a pre-generated cleaning schedule instruction, which includes the suggested cleaning operation time points and suggested cleaning intensity levels.

[0071] In one embodiment, the time-series prediction model can be trained through the following process: Construction of the time-series training dataset: Historical monitoring data and environmental parameters generated during long-term deployment are collected to form a long-term series dataset. Each data record includes: timestamp (T), fouling state index (F, which can be quantified from historical dual-mode detection results), turbidity (NTU), water temperature (Temp), dissolved oxygen (DO), and season (e.g., spring, summer, autumn, winter coding) and day / night (day / night) as labels for environmental parameters. These series data are arranged in chronological order.

[0072] Training Sample Generation (Sliding Window Method): The sliding window method is used to construct training samples. A historical observation window length L (e.g., the past 24 hours, L=24 if sampling hourly) and a future prediction window length P (e.g., the next 12 hours) are set. For each time point t in the time series, the data from the previous L time steps [t-L+1, t] are taken as input features. This input feature is a multi-dimensional matrix, where rows represent time steps and columns represent different features (e.g., F, NTU, Temp, DO, Season, DayNight, etc.). The dirt state index [t+1, t+P] for the next P time steps is used as the prediction target. This process is repeated throughout the entire time series, generating a large number of "input sequence-target sequence" pairs, forming the time series training sample set.

[0073] Network structure and training: Model Architecture: A lightweight Long Short-Term Memory (LSTM) network is constructed. This network consists of an input layer (accepting a matrix of dimension L × number of features), one or more LSTM layers (each LSTM unit contains an input gate, a forget gate, a cell state, and an output gate, used to capture long-term dependencies), an optional attention mechanism layer (used to allow the model to focus on the historical periods most relevant to future predictions), and a fully connected output layer (output dimension P, corresponding to the predicted dirt index at P future time steps).

[0074] Lightweighting strategy: Lightweighting is achieved by limiting the number of hidden units in the LSTM layer and the number of network layers. For example, a single-layer LSTM can be used with the number of hidden units set to 32 or 64.

[0075] Loss Function and Training: The Mean Squared Error (MSE) loss function is used to measure the difference between the model's predicted sequence for the next P steps and the true target sequence. During training, the input sequence of the time-series training sample set is fed into the LSTM network, forward propagation is performed to obtain the predicted sequence, the MSE loss is calculated, the gradient is calculated using the Backpropagation Through Time (BPTT) algorithm, and the network weights are updated using the Adam optimizer. Early stopping is used during training to prevent overfitting.

[0076] Optimization incorporating domain knowledge (rule-based penalty term): To make predictions more consistent with physical laws, a rule-based penalty term based on domain knowledge can be added to the loss function. For example, a soft constraint can be set such as "when historical water temperatures are consistently above a certain threshold, the future trend of biofouling growth should be more significant." During training, if the model's predictions violate such constraints, a penalty term is added to the loss function to guide the model to learn more reasonable prediction patterns.

[0077] Model Consolidation and Integration: After training, the time-series prediction model is integrated into the prediction module along with the weather-fouling association rule base. The rule base is predefined in the form of "if-then" rules and is not involved in training. In actual deployment, the model receives historical data from the most recent L time steps, outputs a preliminary predicted fouling accumulation curve, and then this curve is corrected based on real-time weather forecast information by querying and applying the weather-fouling association rule base, ultimately generating the final cleaning scheduling instruction.

[0078] Preferably, the specific implementation of S9 further optimizes the above prediction. The prediction module receives external environmental data, especially weather forecast information for a future period (e.g., 24 hours), including precipitation, temperature changes, and wind speed, through a wireless communication module. The prediction module maintains a weather-fouling association rule base, which defines the typical impact of specific weather patterns on fouling accumulation (e.g., "moderate to heavy rain" events usually cause a sharp increase in water turbidity in the short term, associated with a surge in the risk of inorganic particulate fouling; "continuous high temperatures and sunny days" are associated with accelerated algal proliferation and an increased risk of biological fouling). Upon receiving weather forecast information, the prediction module performs pattern matching on the current forecast according to the weather-fouling association rule base. If a relevant rule is matched, the predicted fouling accumulation curve generated by the Lightweight LSTM model in S8 is dynamically corrected. For example, if "moderate to heavy rain" is predicted in the next 3 hours, an increment of turbidity and particulate fouling risk defined by the rule base is superimposed on the corresponding time period of the predicted fouling accumulation curve, thereby generating a corrected predicted fouling accumulation curve. Based on this corrected curve, the prediction module updates the pre-generated cleaning schedule instructions, producing final cleaning schedule instructions. For example, it might move a cleaning scheduled for four hours later to before the rain, or increase the backwashing frequency after the rain. The control system then dynamically adjusts the timing and intensity of cleaning operations based on these final cleaning schedule instructions. This mechanism, which incorporates external weather forecast information for proactive correction, upgrades cleaning scheduling from reactive prediction relying solely on historical data to a more proactive scheduling driven by the external environment, thereby maintaining high cleaning efficiency and energy efficiency in changing environments.

[0079] Optionally, S9 includes: Step 9A: Receive external environmental data and correct the dirt accumulation trend curve according to the predefined weather-dirt association rules to obtain the corrected dirt accumulation trend curve; Step 9B: Update the pre-generated cleaning schedule instruction according to the corrected dirt accumulation trend curve, and generate a final cleaning schedule instruction that includes the final execution time and the final cleaning intensity level. Step 9C: Before executing the final cleaning scheduling instruction, collect a set of real-time water quality parameters for the current water body, including: Step 9C1: Perform feature extraction processing on the real-time water quality parameter set to generate a current environmental feature vector, including: Step 1: Normalize each parameter in the real-time water quality parameter set to obtain normalized real-time water quality parameters; Step 2: Input the normalized real-time water quality parameters into an environmental feature extraction network. The environmental feature extraction network performs nonlinear transformation and dimensionality reduction on the input through a fully connected layer and an activation function, and outputs an environmental feature vector to characterize the overall state of the water body. Step 9C2: Input the current environmental feature vector and the final cleaning intensity level in the final cleaning scheduling instruction into a cleaning parameter dynamic adjustment network to generate a dynamically optimized cleaning parameter set, including: Step 3: The cleaning parameter dynamic adjustment network receives the current environmental feature vector and the final cleaning intensity level as joint input; Step 4: The first fully connected layer of the cleaning parameter dynamic adjustment network fuses and performs preliminary transformation on the joint input to generate the first hidden layer features; Step 5: The second fully connected layer of the cleaning parameter dynamic adjustment network maps the features of the first hidden layer and outputs an adjustment weight vector. Each element in the adjustment weight vector corresponds to the adjustment coefficient of a specific cleaning parameter. Step 6: Multiply the adjusted weight vector element-wise with a baseline cleaning parameter set corresponding to the final cleaning intensity level to obtain the dynamic optimized cleaning parameter set, which includes mechanical scraping frequency adjustment coefficient, backwash water flow intensity adjustment coefficient, and cleaning duration adjustment coefficient. Step 9D: Control the self-cleaning structure 13 to perform cleaning operations according to the final execution timing and the dynamically optimized cleaning parameter set, including: Step 7: When the system time reaches the final execution time, the control system reads the dynamically optimized cleaning parameter set; Step 8: Calculate and set the actual operating frequency of the drive motor in the self-cleaning structure 13 based on the mechanical scraping frequency adjustment coefficient in the dynamically optimized cleaning parameter set. Step 9: Calculate and set the amplitude and speed of the undulating motion executed by the active buoyancy adjustment unit based on the backwash water flow intensity adjustment coefficient in the dynamic optimization cleaning parameter set. Step 10: Calculate and set the total duration of this cleaning operation based on the cleaning duration adjustment coefficient in the dynamically optimized cleaning parameter set; Step 11: Drive the self-cleaning structure 13 and the active buoyancy adjustment unit to work together in sequence according to the set actual working frequency, amplitude and speed of the undulating motion and the total duration to complete the cleaning operation.

[0080] Preferably, the specific implementation process of step 9A is as follows: The prediction module receives structured external environmental data through the wireless communication module. This external environmental data includes weather forecast information for a future time window, such as the time-series forecast values ​​of precipitation probability, precipitation amount, temperature, wind speed, and solar radiation intensity. The prediction module pre-stores a weather-fouling association rule base, which defines the direction and magnitude of the influence of various typical weather patterns on the fouling accumulation rate in the form of "condition-impact" pairs. For example, a rule might define: "Condition: Precipitation forecast value is greater than threshold R1 within the next 6 hours; Impact: Within 3 hours after the start of precipitation, the inorganic particulate fouling accumulation risk index increases by ΔI1, and then linearly decays within the following 6 hours." After receiving the external environmental data, the prediction module matches the future weather forecast information with each rule condition in the weather-fouling association rule base. For each successfully matched rule, the prediction module applies a corresponding offset or scaling factor to the corresponding time period of the fouling accumulation trend curve generated in step 8A, based on the "impact" defined in the rule. After superimposing the effects of all matching rules, the original dirt accumulation trend curve is updated to form a corrected dirt accumulation trend curve that more closely reflects future actual weather conditions. This technology differs from predictions that rely solely on historical data. By incorporating forward-looking external environmental information, it enables predictions to respond to impending weather events (such as a sudden increase in turbidity due to heavy rain), thereby adjusting cleaning expectations in advance.

[0081] Preferably, in the specific implementation of step 9B, the prediction module analyzes the corrected dirt accumulation trend curve. This curve has time on the horizontal axis and the predicted dirt index on the vertical axis. The prediction module presets a cleaning trigger threshold. It iterates through the corrected dirt accumulation trend curve, identifies the future time point when the dirt index first exceeds the cleaning trigger threshold, and determines this time point as the final execution opportunity. Simultaneously, the prediction module maps the rising slope of the corrected dirt accumulation trend curve near the trigger point and the predicted dirt type distribution (which can be obtained from historical dual-mode detection results of the same period) to a preset cleaning intensity level table, determining a final cleaning intensity level. This level can be, for example, divided into "light," "standard," and "enhanced" levels. Finally, the prediction module encapsulates the final execution opportunity and the final cleaning intensity level to generate a final cleaning scheduling instruction. This step transforms the continuous prediction curve into a specific, executable scheduling command.

[0082] Preferably, the specific implementation process of step 1 is as follows: The control system synchronously collects a set of the latest readings from multiple water quality sensors (e.g., turbidity sensor 12, temperature sensor, dissolved oxygen sensor, pH sensor) mounted on the buoy. These readings constitute a set of real-time water quality parameters. Each parameter in this set has a different physical meaning and dimension, and the numerical ranges vary considerably. To eliminate the dimensional differences and ensure that all parameters are within the same order of magnitude range, facilitating subsequent neural network processing, the control system calls the historical statistical ranges of each parameter pre-stored in memory. This historical statistical range includes the minimum and maximum values ​​recorded for each parameter during long-term monitoring. For each parameter value in the real-time water quality parameter set, the control system performs the following calculation: subtract the historical minimum value of the parameter from the current parameter value, and then divide by the difference between the historical maximum and historical minimum values. This linear transformation process is the normalization process. After processing, each parameter value is transformed into a closed interval between 0 and 1, resulting in a set of normalized real-time water quality parameters. This step provides a standardized data foundation for subsequent feature fusion and model calculation.

[0083] Preferably, in the specific implementation of step 2, the control system organizes the normalized real-time water quality parameters into a one-dimensional vector according to a predetermined order (e.g., turbidity, temperature, dissolved oxygen, pH), which serves as the input to the environmental feature extraction network. This environmental feature extraction network is a specially trained lightweight feedforward neural network, its structure optimized for water quality feature extraction tasks. The network structure includes an input layer, a hidden layer (i.e., a fully connected layer) containing multiple nodes, and an output layer. The number of nodes in the input layer is equal to the dimension of the normalized real-time water quality parameter vector. When the normalized real-time water quality parameter vector is input into the network, it first enters the hidden layer. In the hidden layer, the input vector is multiplied by a weight matrix, and a bias vector is added. The number of rows in this weight matrix is ​​equal to the number of nodes in the hidden layer, and the number of columns is equal to the dimension of the input vector; the elements at the intersection of rows and columns represent the influence weight of an input parameter on a hidden layer node. The bias vector provides a learnable offset for each hidden layer node. The results of the linear operations described above are then processed by a nonlinear activation function (e.g., ReLU, or Rectified Linear Unit), which sets all negative values ​​to zero and leaves positive values ​​unchanged. This process achieves nonlinear transformation of the original parameters and interactive learning of features. The transformed features are passed from the hidden layer to the output layer. The output layer is also a fully connected layer with fewer nodes than the hidden layer, serving to reduce dimensionality and condense information. After linearly combining the inputs, the output layer outputs a vector with lower dimensionality that more comprehensively represents the current complex state of the water body (such as abstract states like "eutrophication and high water temperature" or "turbidity but sufficient dissolved oxygen"), i.e., the current environmental feature vector. This network is obtained through supervised training using a large amount of historical water quality data and sensor fouling data from the corresponding periods, enabling it to learn to extract key environmental features potentially related to fouling formation from multidimensional parameters.

[0084] Preferably, in a scenario, when steps 3 to 6 are specifically implemented, the control system initiates a dynamic parameter optimization process.

[0085] In step 3, the control system preprocesses the current environmental feature vector output from step 2 and the final cleaning intensity level determined in step 9B. The final cleaning intensity level is a discrete category label (such as "mild," "standard," or "enhanced"). The control system converts this into a one-bit valid encoding vector; for example, the "standard" level might be encoded as [0,1,0]. Subsequently, this one-bit valid encoding vector is concatenated with the current environmental feature vector along the vector's dimensional direction to form a longer joint input vector that integrates the real-time environmental state and the preset cleaning intensity intention.

[0086] In step 4, the joint input vector is fed into a cleaning parameter dynamic adjustment network. This network is a lightweight feedforward neural network with a first fully connected layer. The first fully connected layer performs matrix multiplication (using its weight matrix) and vector addition (adding its bias vector) on the joint input vector, and then passes it through a ReLU activation function. This series of operations completes the fusion of environmental features and intensity level, and performs preliminary nonlinear transformation and abstraction on the fused information, outputting an intermediate representation, namely the first hidden layer feature. The first hidden layer feature captures "the comprehensive factors that need to be considered when performing a certain intensity level of cleaning under a specific environmental state".

[0087] In step 5, the first hidden layer features are passed to the network's second fully connected layer. The number of nodes in the second fully connected layer is set to the number of cleaning parameters that need to be dynamically adjusted (e.g., 3, corresponding to mechanical scraping, backwash intensity, and duration, respectively). This layer performs a linear mapping (matrix multiplication plus bias) of the first hidden layer features. To limit the output value to a reasonable adjustment range (e.g., 0.5 to 1.5 times the baseline value), the output of the second fully connected layer is scaled and translated using a sigmoid activation function. Specifically, this can be achieved using the formula f(x) = 1.0 + 0.5. (Sigmoid(x) - 0.5) A variant of 2 ensures the output value falls within the range of [0.5, 1.5]. Finally, the second fully connected layer outputs an adjustment weight vector, where each element is an adjustment coefficient for the corresponding cleaning parameter. For example, the first element of the vector is the mechanical scraping frequency adjustment coefficient, the second element is the backwash water flow intensity adjustment coefficient, and the third element is the cleaning duration adjustment coefficient.

[0088] In step 6, the control system maintains a baseline cleaning parameter lookup table. This table, indexed by the final cleaning intensity level, stores the baseline parameter values ​​for each cleaning operation at each level (e.g., at the "Standard" level: baseline scraping frequency is Fb, baseline fluctuation amplitude is Ab, and baseline duration is Tb). The control system retrieves the corresponding baseline cleaning parameter set from this table based on the final cleaning intensity level. Finally, the adjustment weight vector obtained in step 5 is multiplied element-wise with the retrieved baseline cleaning parameter set. That is, the mechanical scraping frequency adjustment coefficient is multiplied by the baseline scraping frequency Fb to obtain the optimized frequency parameter; the backwash water flow intensity adjustment coefficient is multiplied by the baseline fluctuation amplitude Ab to obtain the optimized amplitude parameter; and the cleaning duration adjustment coefficient is multiplied by the baseline duration Tb to obtain the optimized time parameter. All these optimized parameters together constitute the dynamically optimized cleaning parameter set. This step transforms the abstract adjustment intentions inferred by the neural network into specific, executable equipment control parameters.

[0089] Preferably, the specific implementation process of steps 7 to 11 is as follows: Step 7: The system's task scheduler continuously monitors the internal clock. When the system time reaches the final execution time specified in the final cleaning schedule instruction, the scheduler triggers the cleaning execution task. This task first reads the previously calculated and stored set of dynamically optimized cleaning parameters from the system's shared memory or non-volatile memory.

[0090] Step 8: The cleaning task analyzes and dynamically optimizes the cleaning parameter set, extracting the mechanical scraping frequency adjustment coefficient (denoted as K_f). The task internally stores the baseline motor operating frequency F_base for the current cleaning mode (related to the final cleaning intensity level). This is achieved by calculating F_actual = F_base. K_f is used to obtain the actual operating frequency. Subsequently, the task sends a pulse width modulation (PWM) command to the drive motor (such as servo 16 or micro DC motor) in the self-cleaning structure 13 through the motor drive interface to set the motor's operating frequency to the calculated actual operating frequency.

[0091] Step 9: The cleaning task analyzes and dynamically optimizes the cleaning parameter set, extracting the backwash water flow intensity adjustment coefficient (denoted as K_s). The task pre-stores the baseline fluctuation amplitude A_base and baseline fluctuation velocity V_base. This is achieved by calculating A_target = A_base. K_s and V_target = V_base K_s is used to obtain the target undulation amplitude and target undulation velocity. Next, the task calls the control function of the active buoyancy adjustment unit and sends a motion command to it. This command requires the active buoyancy adjustment unit (by controlling the water pump 20 and the air valve) to drive the buoy to perform several complete up-and-down undulation movements, and to precisely control the vertical displacement of each sinking and rising to the target undulation amplitude. The cycle time of completing one undulation movement is determined by the target undulation velocity, thereby generating a backwash water flow with controlled intensity.

[0092] Step 10: The cleaning task is parsed and dynamically optimized, extracting the cleaning duration adjustment coefficient (denoted as K_t). The task internally stores the baseline cleaning duration T_base. This is achieved by calculating T_total = T_base. K_t represents the total duration of this operation. This total duration will be used as the timer threshold for the entire cleaning sequence.

[0093] Step 11: The cleaning task coordinates the two actuators to initiate a timed collaborative workflow. It first sets a total timer with a time limit equal to the total duration. Simultaneously with the start of the total timer, it commands the drive motor of the self-cleaning structure 13 to begin operating at the actual working frequency. At the same time, within the total duration, it commands the active buoyancy adjustment unit to perform a undulating motion with the target undulation amplitude and target undulation speed at preset time points (e.g., 2 seconds or 10 seconds after start) or at fixed intervals. These two processes are performed in parallel or alternately. The task continuously monitors the total timer, and when the total duration expires, the task sends a stop command to the drive motor and the active buoyancy adjustment unit, thereby ending the cleaning operation. In this way, the cleaning operation is precisely executed within a predetermined time, using a dynamically optimized combination of parameters based on the real-time environment. Among these, " " indicates calculating the product, for example, T_base K_t represents the product between T_base and K_t.

[0094] It should be understood that any connection between devices and components that require circuit control in this application is made via wired and / or wireless means. The fact that they are not described as being electrically connected does not mean that they are not connected.

[0095] It should be understood that this application is not limited to the processes and structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A water quality monitoring buoy, comprising: The water quality monitoring buoy includes a sensor mounting part (5), a float (6), and a counterweight (8). The sensor mounting part (5) is located in the center-of-gravity cavity of the float (6) and can float up and down with the buoyancy of the float (6). The counterweight (8) is installed at the bottom of the float (6) to lower the overall center of gravity. The water quality monitoring buoy also includes an active buoyancy adjustment unit and a control system. The active buoyancy adjustment unit includes a water pump (20), a high-pressure gas cylinder (22), a high-pressure air pump (19), a solenoid valve (23), and an exhaust valve (24). The water pump (20) is used to pump water into or draw water from the float (6). The high-pressure gas cylinder (22) is connected to the inside of the float (6) through the solenoid valve (23). The high-pressure air pump (19) is used to replenish compressed air into the high-pressure gas cylinder (22). The exhaust valve (24) is used to discharge the air inside the float (6). The control system is used to control the active buoyancy adjustment unit to realize the buoy's rising, diving and hovering, and to control the operation of the self-cleaning structure (13) in the sensor mounting part (5).

2. The water quality monitoring buoy according to claim 1, characterized in that, The sensor mounting part (5) includes a sealed cavity (11), a turbidity sensor (12), a self-cleaning structure (13), and an outer anti-fouling net (14); The sealed cavity (11) is used to install the control circuit of the control system; The lowest point of the sensor probe (2) of the turbidity sensor (12) is higher than the bottom surface of the counterweight (8); The self-cleaning structure (13) is located between the counterweight (8) and the outer anti-fouling net (14) and is used to clean the sensor probe (2) of the turbidity sensor (12); The outer anti-fouling net (14) is set outside the sensor mounting part (5).

3. The water quality monitoring buoy according to claim 1, characterized in that, The self-cleaning structure (13) includes a sensor mounting base (15), a servo motor (16), a rocker arm (17), and a brush (18). The servo motor (16) drives the rocker arm (17) to move the brush (18) back and forth to clean the surface of the sensor probe (2) of the turbidity sensor (12).

4. The water quality monitoring buoy according to claim 1, characterized in that, It also includes a tuning fork level sensor (21), which is installed at the bottom of the buoy and is used to determine whether the buoy is covered by silt by judging the amplitude and vibration frequency of the tuning fork level sensor (21). When the silt cover is detected, a signal is sent to the control system, which controls the water pump (20) to drain the water inside the float (6) and controls the solenoid valve (23) to release the compressed air in the high-pressure gas cylinder (22) into the float (6) to increase the buoyancy of the float (6) so that the float can break free from the silt.

5. The water quality monitoring buoy according to claim 1, characterized in that, The sensor mounting part (5) also integrates an image acquisition module (25), and the image acquisition module (25) and the turbidity sensor (12) constitute a dual-mode detection system; The image acquisition module (25) is used to acquire image information of the surface of the sensor probe (2) at a preset frequency. The control system has a built-in image recognition module for identifying and classifying dirt in the acquired image information. The turbidity sensor (12) is used to monitor the optical parameters of the water body within a preset range of the sensor probe (2). The control system fuses the optical parameter monitoring data with the image recognition results to generate dirt status judgment information.

6. The water quality monitoring buoy according to claim 5, characterized in that, The control system is configured to execute a corresponding cleaning strategy based on the type of dirt indicated by the dirt state judgment information; the cleaning strategy includes a mechanical scraping mode executed by the self-cleaning structure (13), a backwashing mode generated by water flow disturbance, and a combination mode of the mechanical scraping mode and the backwashing mode.

7. The water quality monitoring buoy according to claim 5, characterized in that, The control system also includes a prediction module, which is used to predict the dirt accumulation trend and generate cleaning scheduling instructions based on historical monitoring data and environmental parameters; the control system dynamically adjusts the timing and intensity of cleaning operations according to the cleaning scheduling instructions.

8. The water quality monitoring buoy according to claim 7, characterized in that, The prediction module is also used to receive external environmental data, including weather forecast information; the prediction module corrects the cleaning scheduling instructions by analyzing the correlation between the weather forecast information and historical dirt accumulation data.

9. The water quality monitoring buoy according to claim 1, characterized in that, The control system is also configured to: based on the monitoring data of the turbidity sensor (12), automatically adjust the buoy and suspend it at a water depth that meets the preset water quality conditions by controlling the active buoyancy adjustment unit.

10. The water quality monitoring buoy according to claim 1, characterized in that, It also includes a base plate (7) and an anchor fixing hole (1). The base plate (7) cooperates with the sensor mounting part (5) to clamp and fix the float (6). The anchor fixing hole (1) is used to connect to the pre-buried anchor point on the bank or riverbed through a steel wire rope.

Images