Space-time adaptive intelligent ultrasonic algae control method

Abstract

The invention discloses a space-time self-adaptive intelligent ultrasonic algae control method, which aims at solving the problems of frequent algae blooms, difficult accurate control and the like of slow-flow water bodies such as lakes, reservoirs and the like, comprehensively applies numerical simulation and deep learning algorithms, and realizes water body algae bloom hot spot pixel-level identification and week / hour-level accurate prediction. A neighborhood suppression iterative selection strategy based on a service radius is adopted, space adaptive deployment of the ultrasonic algae control unmanned ship can be driven, and real-time closed-loop operation of the ultrasonic algae control unmanned ship can be realized by further hour-level shipborne data and deep learning prediction and fusion of a dynamic planning optimization algorithm.

Description

Technical Field

[0001] This invention relates to the field of comprehensive water environment management, specifically to a spatiotemporally adaptive intelligent ultrasonic algae control method. Background Technology

[0002] Frequent algal blooms pose a serious threat to aquatic ecosystem health and drinking water safety, necessitating efficient and low-ecological-impact algae control technologies. Among these technologies, ultrasonic algae control unmanned surface vessels (USVs) utilize USV systems equipped with ultrasonic algae control devices. These devices release elastic ultrasonic waves to disrupt the buoyancy regulation mechanisms of algae without interfering with the existing aquatic ecosystem. This results in impaired photosynthesis, metabolic disorders, and cell rupture, ultimately causing the algae to settle as particles or be decomposed and mineralized by microorganisms.

[0003] The patent specification with publication number CN117985811A discloses a composite wave algae suppressor and its usage method. The composite wave algae suppressor includes a body and a solar power module, an ultrasonic module, an oxygen supply module, a wave module, an infrared laser detection module, and a control system installed on the body. The solar power module provides power to the device. The ultrasonic module destroys the algae structure by generating ultrasonic waves of a certain frequency. The oxygen supply module supplies oxygen to the bottom of the unmanned vessel through a DLB centrifugal air pump, increasing the dissolved oxygen content of the water and destroying the algae growth environment. The wave module consists of a fan and a water circulation pump, which stirs the water to generate waves to suppress algae. The infrared laser detection module is an infrared laser rangefinder. The infrared laser detection module measures the algae density by emitting infrared lasers in all directions and selects a suitable treatment scheme.

[0004] The patent specification with publication number CN119898851A discloses an intelligent algae control method and device based on target recognition. The method identifies the severity of algal blooms based on the average concentration of chlorophyll a, identifies algae based on images from different locations, obtains algal body identification results, statistically analyzes the algal body identification results from images from different locations, obtains the number of different types of algae in the target algae control area, and calculates the growth rate by combining the number of different types of algae obtained continuously, thereby formulating an algae control plan.

[0005] However, algal blooms in aquatic bodies exhibit significant spatiotemporal heterogeneity and sudden occurrence. Most existing ultrasonic algae control unmanned surface vessels (USVs) are based on static deployments and lack the ability to adapt to dynamic spatiotemporal changes. In recent years, with the development of emerging technologies such as the Internet of Things (IoT), big data, and artificial intelligence (AI), algae bloom prediction based on multi-source data acquisition through remote sensing forecasting and online sensing, as well as combining numerical simulation with AI, can provide ultrasonic algae control USVs with accurate spatiotemporal prediction, operational deployment, and real-time scheduling support. Summary of the Invention

[0006] This invention provides a spatiotemporally adaptive intelligent ultrasonic algae control method. Addressing the challenges of frequent algal blooms and precise control in slow-flowing water bodies such as lakes and reservoirs, it comprehensively applies numerical simulation and deep learning algorithms to achieve pixel-level identification and weekly / hourly-level accurate prediction of algal bloom hotspots. Employing a service radius-based neighborhood inhibition iterative selection strategy, it drives the spatially adaptive deployment of an ultrasonic algae control unmanned surface vessel (USV). Furthermore, by integrating hourly-level onboard data with deep learning predictions and a dynamic programming optimization algorithm, it enables real-time closed-loop operation of the ultrasonic algae control USV.

[0007] The specific technical solution is as follows: A spatiotemporally adaptive intelligent ultrasonic algae control method includes the following steps: S1: Grid the target water area, construct a hydrodynamic-water quality numerical model, and realize the time series prediction of hydrodynamics and water quality for each grid in the future (e.g., one week); S2: Calculate the floating algae index (FAI) of pixels based on satellite remote sensing images of algal blooms in the target water area, fuse the time series of numerical model input and output with historical FAI data and rolling updated FAI time series data, construct deep learning model I, realize the time series prediction of FAI for each pixel in the future (e.g., one week) and calculate the time series mean. S3: Based on the time-series mean of FAI of each pixel, a neighborhood suppression iterative selection strategy based on service radius is adopted to determine the optimal operating position of the unmanned vessel in the target water area, so as to realize the spatial adaptive deployment of the unmanned vessel. S4: Based on the time-series data output by the numerical model and deep learning model I, as well as the shipborne water quality sensor and operating parameters of each unmanned vessel and its neighboring pixels, deep learning model II is constructed to predict the chlorophyll a (Chl a) concentration of each unmanned vessel's pixels in the future (e.g., 24 hours). Dynamic programming (DP) optimization algorithm is integrated to solve the optimal control quantity in real time under the given objective and physical constraints and implement it, so as to realize the spatiotemporal adaptive ultrasonic algae control operation of the unmanned vessel.

[0008] In some embodiments, the hydrodynamic-water quality numerical model in step S1 is constructed based on Delft3D+WAQ (Delft 3D hydrodynamic simulation system + water quality simulation module).

[0009] In some embodiments, the horizontal grid cells in step S1 are aligned with the satellite remote sensing image pixels in step S2.

[0010] In some embodiments, the vertical mesh in step S1 is divided into equal-thickness sections.

[0011] In some embodiments, in step S2, the calculated pixel FAI spatiotemporal sequence is interpolated and reconstructed based on the spatiotemporal kriging method to obtain a completed FAI field with temporal resolution consistent with the numerical model output.

[0012] In some embodiments, pixels with a time-series mean FAI greater than the floating algae index at the beginning of an algal bloom (e.g., 0.02) in step S3 are candidates for deploying unmanned vessels.

[0013] In some embodiments, in step S3, the pixels are sorted in descending order according to the FAI time-series mean to obtain a candidate set; Initialize the selected set; select the cell with the largest FAI time-series mean from the candidate set as the unmanned surface vessel (USV) operation site and add it to the selected set, and delete the cell and all cells within the USV service radius (e.g., 250 m) of the cell from the candidate set to avoid spatial overlap; repeat the above selection-addition-deletion operation until the number of elements in the selected set reaches the number of USVs to be deployed in the target waters; Each unmanned vessel autonomously navigates to a selected pixel in the set to perform ultrasonic algae control.

[0014] In some embodiments, step S3 is performed every certain period of time according to the ultrasonic algae control cycle of the unmanned vessel, and the number of unmanned vessels deployed in the target water area is adjusted in combination with on-site feedback to achieve closed-loop optimization of the spatial adaptive deployment of unmanned vessels.

[0015] In some embodiments, in step S4, the time series data of each unmanned vessel and its neighboring pixels, output by the numerical model and deep learning model I, as well as the shipborne water quality sensing and operating parameters, are aligned and completed according to the same time scale (e.g., by hour) and then used as the input of deep learning model II.

[0016] In some embodiments, the sampling time of the unmanned vessel ultrasonic algae control operation in step S4 is consistent with the time accuracy of the output of the deep learning model II (e.g., 1 hour), and the control time domain is consistent with the prediction time domain of the deep learning model II (e.g., 24 hours).

[0017] In some embodiments, the optimal control quantities solved in step S4 include the ultrasonic generator power and frequency.

[0018] Compared with the prior art, the beneficial effects of this invention are as follows: (1) By integrating images and multi-source sensor data, and applying numerical simulation and BiLSTM-Attention (bidirectional long short-term memory network + attention mechanism) deep learning algorithm, we can achieve pixel-level identification and weekly accurate prediction of algal bloom hotspots in water bodies, which can help preventive deployment and reduce the probability and ecological risks of peak outbreaks and large-scale post-event disposal.

[0019] (2) The optimal operating site of the ultrasonic algae control unmanned vessel is determined by the neighborhood inhibition iterative selection strategy based on the service radius, so as to realize spatial adaptive deployment.

[0020] (3) By combining dynamic programming optimization algorithm, the optimal control is solved in real time under the given objective and physical constraints, realizing the spatiotemporal adaptive algae control operation of unmanned ships, improving the unit governance efficiency of single ship operation, and thus maximizing ecological benefits under limited transport capacity. Attached Figure Description

[0021] Figure 1 This is a schematic flowchart of a spatiotemporally adaptive intelligent ultrasonic algae control method according to the present invention.

[0022] Figure 2 It is a structured grid map of the target water area.

[0023] Figure 3 This is a graph showing the predicted water quality concentrations generated by the hydrodynamic-water quality numerical model.

[0024] Figure 4 This is a scatter plot of the FAI prediction data for a certain pixel on the 7th day from the future using Deep Learning Model I.

[0025] Figure 5 This is a scatter plot of Chl a concentration prediction data for a certain pixel in the next 24 hours using Deep Learning Model II.

[0026] Figure 6 This is a graph showing the changes in Chl a concentration in the target water area over 5 days of unmanned vessel operation. Detailed Implementation

[0027] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0028] See Figure 1 A spatiotemporally adaptive intelligent ultrasonic algae control method includes the following steps: S1: Grid the target water area, construct a hydrodynamic-water quality numerical model, and realize the time series prediction of hydrodynamics and water quality for each grid in the future (e.g., one week); S2: Calculate the floating algae index (FAI) of pixels based on satellite remote sensing images of algal blooms in the target water area, fuse the time series of numerical model input and output with historical FAI data and rolling updated FAI time series data, construct deep learning model I, realize the time series prediction of FAI for each pixel in the future (e.g., one week) and calculate the time series mean. S3: Based on the time-series mean of FAI of each pixel, a neighborhood suppression iterative selection strategy based on service radius is adopted to determine the optimal operating position of the unmanned vessel in the target water area, so as to realize the spatial adaptive deployment of the unmanned vessel. S4: Based on the time-series data output by the numerical model and deep learning model I, as well as the shipborne water quality sensor and operating parameters of each unmanned vessel and its neighboring pixels, deep learning model II is constructed to predict the chlorophyll a (Chl a) concentration of each unmanned vessel's pixels in the future (e.g., 24 hours). Dynamic programming (DP) optimization algorithm is integrated to solve the optimal control quantity in real time under the given objective and physical constraints and implement it, so as to realize the spatiotemporal adaptive ultrasonic algae control operation of the unmanned vessel.

[0029] Step S1, the hydrodynamic-water quality numerical model, is built based on Delft3D+WAQ and includes the following steps: S11: The target water area is divided into structured grids, with horizontal grid cells aligned with the satellite remote sensing image pixels from step S2, and the grid horizontal resolution is [not specified]. l 网格 Should be related to pixel size l 像元 Consistent or maintains a compatible resolution that can be directly registered ( l 像元 / l 网格 =Integer); at the same time, the vertical grid adopts equal thickness division, and the number of layers is denoted as z.

[0030] S12: Obtain static basic data, boundary condition data, meteorological driving data, hydrodynamic and water quality observation data, and relevant process parameters of the target water area to construct a hydrodynamic-water quality numerical model, and apply the Sobol-guided SDCV (Spatial Distributed Cross-Validation Guided by Sensitivity Analysis) method for calibration and validation.

[0031] S13: The constructed hydrodynamic-water quality numerical model can predict the hydrodynamic and water quality time series of the target grid over a future period (e.g., one week) based on the current time, with a time step ≤ 24 h, as shown below:

[0032] In the formula, and These represent the state vectors of the target grid cell for hydrodynamics and water quality, respectively, including water level, horizontal velocity (including flow direction), flux, temperature, turbulence parameters, dissolved oxygen (DO), pH, chemical oxygen demand (COD), suspended solids (SS), and ammonia nitrogen (NH4). + -N), nitrate (NO3) - -N), total nitrogen (TN), phosphate ions (PO4) 3- ), total phosphorus (TP), Chl a, etc.; For boundary conditions and meteorological driving data, including point source and non-point source hydrodynamic and water quality prediction data for the target water area; Indicates the current time; Represents the time step; The number of time steps, rounded to the nearest integer; The constructed hydrodynamic-water quality numerical model.

[0033] Step S2 specifically includes the following steps: S21: Collect satellite remote sensing images of algal blooms in the target water area, perform atmospheric correction and remove interference such as clouds, cloud shadows and solar flares, select high-quality observation data to calculate FAI pixel by pixel; further, interpolate and reconstruct the calculated FAI spatiotemporal sequence based on the spatiotemporal kriging method to obtain a complete FAI field with temporal resolution consistent with the numerical model output.

[0034] S22: For each pixel, apply the BiLSTM-Attention deep learning algorithm to construct a high-precision prediction model (i.e., deep learning model I) for the FAI time series of the pixel over a future period (e.g., one week). The model output time step is consistent with the numerical model above (≤24 h). The model input consists of boundary condition data and meteorological driving data for each target water area grid point within the pixel, time series data output from the hydrodynamic-water quality numerical model for each target water area grid point, and historical FAI data and rolling updated FAI time series data based on the pixel at the current time. The time step of all inputs is the previous 50 steps from the target prediction time point.

[0035] S23: Apply Bayesian optimization to tune the hyperparameters of deep learning model I, including the number of BiLSTM layers and hidden units per layer, the encoder / decoder input feature dimension and input sequence length, the attention dimension or number of heads, the dropout rate, and the learning rate. The resulting optimal model can perform the accurate predictions mentioned in S22, as shown below:

[0036] In the formula, For the target pixel's FAI, Indicates the current time; Represents the time step; The number of time steps, rounded to the nearest integer; The constructed BiLSTM-Attention model (Deep Learning Model I).

[0037] Step S3 specifically includes the following steps: S31: Based on the current time, calculate the mean of the FAI time series for each pixel in the target water area over a future period (e.g., one week), denoted as... If any pixel exists If the value exceeds the initial accumulation value of algae bloom (floating algae index = 0.02), the deployment procedure of the unmanned vessel will be initiated; otherwise, it will remain in standby mode.

[0038] S32: Press all pixels Sort the values ​​in descending order to construct a candidate set. G At the same time, initialize the selected set. S Select candidate set G middle The cell with the largest value As an operational site for the unmanned vessel, it is added to the selected set. S Meanwhile, based on the service radius of the unmanned vessel (e.g., 250 m), the pixels will be... All surrounding pixels from the set G Remove from the middle to avoid spatial overlap; repeat the above selection, addition, and removal steps until a selected set is reached. S The number of elements reaches a predetermined value N, where N is the number of unmanned vessels deployed in the target waters.

[0039] S33: Based on the selected set S At each of the designated algae delivery points, the unmanned vessel autonomously navigates to the point and activates its ultrasonic algae control function.

[0040] S34: Further periodic re-evaluation and adjustment. Based on the algae control cycle of the ultrasonic algae control unmanned vessel, repeat steps S31~S32 every 3 days or as needed. The deployment quantity N can be adjusted based on on-site feedback to achieve closed-loop optimization.

[0041] Step S4 specifically includes the following steps: S41: For each unmanned surface vessel (USV), collect multi-source data, including time-series data of the USV's location and its eight adjacent pixels (forming a nine-square grid centered on the USV), output by the numerical model and deep learning model I, as well as the USV's own water quality sensor observation sequences and operating parameter sequences during operation. The USV's own water quality sensor observation indicators include DO, pH, SS, and NH4. + -N, NO3 - -N, Chl a, operating parameters include the power of the ultrasonic generator (set to 5W / L, 20 W / L, 30 W / L) and frequency (30 kHz, 60 kHz, 100 kHz, 150 kHz).

[0042] S42: Align the multi-source data collected in S41 by hourly resolution and imput missing values, and use it as input to the deep learning model II. The time step of the input is the previous 50 steps of the current time. Apply the BiLSTM-Attention deep learning algorithm to construct an accurate prediction model of the hourly Chl a concentration of the pixel where the unmanned ship is located for the next 24 hours.

[0043] S43: Applying Bayesian optimization to tune the relevant hyperparameters of the deep learning model II, the resulting optimal model can perform the accurate predictions mentioned in S42, as shown below:

[0044] In the formula, For the water quality indicators observed by the unmanned vessel's own water quality sensors, For unmanned surface vessel operating parameters, Indicates the current time; This represents the time step, which is 1 hour. The number of time steps, rounded to the nearest integer; This is for the constructed hour-level BiLSTM-Attention model (Deep Learning Model II).

[0045] It should be noted that the spatiotemporal adaptive ultrasonic algae control operation of the unmanned vessel mentioned in step S4 is configured with a sampling time of 1 hour, consistent with the time accuracy of the deep learning model II output; and a control time domain of 24 hours, consistent with the prediction time domain. Its immediate objective is to minimize the Chl a concentration in the pixel where the unmanned vessel is located within each control cycle, with the ultimate goal of reducing Chl a to below the safe threshold of 10 μg / L, as specifically stated below:

[0046] In the formula, The objective function is denoted as .

[0047] Meanwhile, in the optimization problem of the unmanned vessel spatiotemporally adaptive ultrasonic algae control operation, the control action is set as the power and frequency of the ultrasonic generator, and physical constraints are set for the control action, which are the selectable power and frequency of the ultrasonic generator in step S41, as specifically expressed as follows:

[0048]

[0049] In the formula, The operating power of the ultrasonic generator on the unmanned surface vessel. Its selectable power set; The operating frequency of the ultrasonic generator on the unmanned surface vessel. It is a set of selectable frequencies.

[0050] It should be noted that the optimal dynamic programming algorithm is used for rolling optimization in the spatiotemporal adaptive ultrasonic algae control operation of the unmanned vessel.

[0051] At time t, based on the deep learning model II, and under physical constraints, the dynamic programming algorithm is applied to perform a rolling search that satisfies the objective function. The decision variables, in Implement decision variables within the first sampling time after time step [time]. Repeat the above process continuously.

[0052] Example 1: This case study focuses on a city lake, a high-risk area for algal blooms, which have historically experienced seasonal or intermittent outbreaks. Through prior implementation of a smart water environment project, comprehensive underwater topographic data was collected. First, a structured grid was created for the target algal-controlling water body. Considering the pixel size of the subsequently acquired remote sensing images is 150 m, the horizontal resolution was set to 50 m × 50 m. Figure 2 The two are resolution compatible. At the same time, the vertical direction is divided into equal thicknesses of 0.3 m, and the number of layers is denoted as z.

[0053] Daily meteorological data on temperature, relative humidity, and rainfall for the lake were collected. Daily water level, flow rate, and water quality monitoring data were also collected from the lake's hydrological and water quality monitoring stations. A hydrodynamic-water quality numerical model was constructed using Delft3D+WAQ and calibrated and validated using the Sobol-guided SDCV method. The model can predict the time series of hydrodynamic and water quality indicators for the target grid over the next week, with a specific time step of 24 hours. Figure 3 The figure shows the ammonia nitrogen prediction data generated by the hydrodynamic-water quality numerical model.

[0054] Further, Sentinel-2 MSI satellite remote sensing images of the algal bloom in the lake were acquired. After atmospheric correction and removal of interference from clouds, cloud shadows, and solar flares, high-quality observation data were selected to calculate the Facial Index (FAI) pixel-by-pixel. The obtained FAI spatiotemporal sequence was then interpolated and reconstructed using the spatiotemporal kriging method to obtain a complete FAI field with temporal resolution consistent with the numerical model output. For each remote sensing pixel, a high-precision prediction model of the FAI index time series for the next week was constructed using the BiLSTM-Attention deep learning algorithm. Bayesian optimization was applied for hyperparameter tuning, and the optimal model hyperparameters are shown in Table 1 below.

[0055] Table 1

[0056] like Figure 4 As shown, a pixel in the southeast corner of the target water area is selected. Based on the current time, the FAI (Fast Index) of this grid for the next 7 days is predicted. Figure 4 As shown, the validation set R of the deep learning model I's FAI prediction for this pixel on day 7 is... 2 The value reached 0.85.

[0057] The algorithm was executed on July 28, 2025, and found that the average FAI time series value for the next week was 0.032. Therefore, the deployment procedure of the unmanned vessel was initiated, and the ultrasonic algae control operation was started.

[0058] Furthermore, for the pixel where the unmanned vessel is located, the output sequences of the aforementioned numerical model and deep learning model I for the pixel and its eight neighboring pixels, as well as the observation sequences and operating parameter sequences of the vessel's own water quality sensor during operation, are collected. The BiLSTM-Attention deep learning algorithm is then applied again to construct an accurate prediction model (deep learning model II, hyperparameters shown in Table 2) for the hourly Chl a concentration of the unmanned vessel in its pixel over the next 24 hours. Figure 5 As shown, the validation set of the deep learning model II for predicting Chl a concentration of this pixel at 24 h reached 0.81.

[0059] Table 2

[0060] Furthermore, based on the Chl a concentration accurate prediction model (deep learning model II), and under the physical constraints of the power and frequency of the unmanned surface vessel's ultrasonic generator, a dynamic programming algorithm is integrated to solve and implement optimal control in real time, thereby achieving spatiotemporal adaptive algae control operations for the unmanned surface vessel. Figure 6 The figure shows the changes in Chl a concentration in the regional water body over 5 days of unmanned vessel operation.

[0061] In summary, this invention first grids the target water area and constructs a hydrodynamic-water quality numerical model to simulate the hydrodynamic and water quality time series for each grid over the next week. Satellite remote sensing imagery is acquired and pixel FAI is calculated. The numerical model's input-output time series and historical FAI data are fused to construct a deep learning model that accurately predicts the FAI sequence for each pixel over the next week. Based on the weekly FAI mean for each pixel, a neighborhood suppression iterative selection strategy based on service radius is used to determine the optimal operating location for the ultrasonic algae control unmanned surface vessel (USV), achieving spatially adaptive deployment. Furthermore, time series data from the numerical and deep models are collected for each USV's pixel and its neighborhood, along with onboard water quality sensors and operational parameters. Multi-source data is aligned and completed hourly, and a deep learning model is constructed to accurately predict the hourly Chl a for the pixel where the USV is located over the next 24 hours. Finally, a dynamic programming optimization algorithm is integrated to solve for and implement optimal control in real time under predetermined objectives and physical constraints, achieving spatiotemporally adaptive algae control operations for the USV.

[0062] Furthermore, it should be understood that after reading the above description of the present invention, those skilled in the art can make various alterations or modifications to the present invention, and these equivalent forms also fall within the scope defined by the appended claims.

Claims

1. A spatiotemporally adaptive intelligent ultrasonic algae control method, characterized in that, Including the following steps: S1: Grid the target water area, construct a hydrodynamic-water quality numerical model, and realize the time series prediction of hydrodynamics and water quality for each grid in the future. S2: Calculate the FAI of pixels based on satellite remote sensing images of algal blooms in the target water area, fuse the time series of numerical model input and output with historical FAI data and rolling updated FAI time series data, construct deep learning model I, realize the time series prediction of FAI of each pixel for a future period and calculate the time series mean. S3: Based on the time-series mean of FAI of each pixel, a neighborhood suppression iterative selection strategy based on service radius is adopted to determine the optimal operating position of the unmanned vessel in the target water area, so as to realize the spatial adaptive deployment of the unmanned vessel. S4: Based on the time-series data output by the numerical model and deep learning model I, as well as the shipborne water quality sensor and operating parameters of each unmanned vessel and its neighboring pixels, deep learning model II is constructed to predict the chlorophyll a concentration of each unmanned vessel's pixels in the future. By integrating dynamic programming optimization algorithm, the optimal control quantity is solved in real time under the given objective and physical constraints and implemented to realize the spatiotemporal adaptive ultrasonic algae control operation of the unmanned vessel.

2. The spatiotemporally adaptive intelligent ultrasonic algae control method according to claim 1, characterized in that, Step S1: The hydrodynamic-water quality numerical model is constructed based on Delft3D+WAQ.

3. The spatiotemporally adaptive intelligent ultrasonic algae control method according to claim 1, characterized in that, In step S1, the horizontal grid cells are aligned with the satellite remote sensing image pixels in step S2, and the vertical grid is divided using equal thickness.

4. The spatiotemporally adaptive intelligent ultrasonic algae control method according to claim 1, characterized in that, In step S2, the calculated pixel FAI spatiotemporal sequence is interpolated and reconstructed based on the spatiotemporal kriging method to obtain a completed FAI field with the same temporal resolution as the numerical model output.

5. The spatiotemporally adaptive intelligent ultrasonic algae control method according to claim 1, characterized in that, In step S3, pixels with a time-series mean FAI greater than the floating algae index at the beginning of algal bloom are candidates for deploying unmanned vessels.

6. The spatiotemporally adaptive intelligent ultrasonic algae control method according to claim 1, characterized in that, In step S3, the pixels are sorted in descending order according to the FAI time series mean to obtain the candidate set; Initialize the selected set; select the cell with the largest FAI time-series mean from the candidate set as the unmanned vessel operation site and add it to the selected set, and delete the cell and all cells within the service radius of the unmanned vessel from the candidate set to avoid spatial overlap; repeat the above selection-addition-deletion operation until the number of elements in the selected set reaches the number of unmanned vessels to be deployed in the target waters; Each unmanned vessel autonomously navigates to a selected pixel in the set to perform ultrasonic algae control.

7. The spatiotemporally adaptive intelligent ultrasonic algae control method according to claim 1, characterized in that, Based on the ultrasonic algae control cycle of the unmanned vessel, step S3 is performed every certain period of time, and the number of unmanned vessels deployed in the target water area is adjusted in combination with on-site feedback, thus optimizing the spatial adaptive deployment of unmanned vessels in a closed loop.

8. The spatiotemporally adaptive intelligent ultrasonic algae control method according to claim 1, characterized in that, In step S4, the time-series data of each unmanned vessel and its neighboring pixels, output by the numerical model and deep learning model I, along with the shipborne water quality sensing and operating parameters, are aligned and completed according to the same time scale and then used as the input of deep learning model II.

9. The spatiotemporally adaptive intelligent ultrasonic algae control method according to claim 1, characterized in that, In step S4, the sampling time of the unmanned vessel's ultrasonic algae control operation is consistent with the time accuracy of the output of the deep learning model II, and the control time domain is consistent with the prediction time domain of the deep learning model II.

10. The spatiotemporally adaptive intelligent ultrasonic algae control method according to claim 1, characterized in that, The optimal control variables to be solved in step S4 include the ultrasonic generator power and frequency.

Images