Dynamic sampling regulation method and system based on water quality detection linkage feedback

Abstract

The application provides a dynamic sampling regulation method and system based on water quality detection linkage feedback, relates to the technical field of water quality detection, and comprises the following steps: acquiring real-time water quality detection data, residual power and cruising time estimation; inputting the water quality detection data into a forward-looking pre-adaptive ring and a physical constraint emergency ring to generate first and second regulation strategies; determining the tactical freedom degree level through energy grading rules according to the power and cruising time; selecting the second regulation strategy in the sufficient level, weighting the two strategies to obtain the third regulation strategy in the tense level, and adopting the first regulation strategy in the critical level; and controlling the unmanned aerial vehicle to execute a new round of dynamic sampling according to the final issued strategy. The application solves the technical problems that the existing technology has a solidified sampling scheme, causes sampling lag when facing sudden pollution, and is difficult to balance flight safety. The application has the technical effects of generating differentiated sampling strategies in parallel through double rings and combining energy grading scheduling, and balancing water quality monitoring requirements and unmanned aerial vehicle flight safety.

Description

Technical Field

[0001] This invention relates to the field of water quality testing technology, specifically to a dynamic sampling and control method and system based on water quality testing linkage feedback. Background Technology

[0002] In recent years, drones have been gradually applied to water environment monitoring due to their maneuverability and flexibility; however, existing technologies still have significant limitations. Traditional methods often employ a fixed operational mode with pre-defined sampling points, depths, and frequencies. Drones simply perform water sampling tasks along predetermined routes, and water samples need to be transported to laboratories for offline analysis. This creates an open-loop chain of sampling before testing and separation of sampling and testing, making it impossible to adjust sampling parameters based on real-time water quality monitoring results. When pollution events occur, fixed vertical lines are prone to deviating from the core pollution area, fixed depths are insufficient to penetrate the transition layer to capture enriched layers, and fixed frequencies are insufficient to track the rapid evolution of water quality, resulting in severe sampling lags and monitoring blind spots. Furthermore, sampling planning does not take into account the drone's endurance, which can easily lead to flight safety risks. At the same time, existing methods generally lack intelligent data analysis and autonomous decision-making capabilities, are unable to deeply integrate multi-parameter water quality data for forward-looking prediction, and lack rapid response mechanisms for sudden exceedance events, resulting in serious deficiencies in the timeliness and accuracy of emergency monitoring.

[0003] In summary, existing technologies suffer from the problem of using fixed sampling schemes, which cannot provide real-time feedback and result in delayed and incomplete sampling when facing sudden pollution, making it difficult to ensure flight safety. Summary of the Invention

[0004] The purpose of this application is to provide a dynamic sampling and control method and system based on water quality detection linkage feedback, in order to solve the technical problems of existing technologies that use fixed sampling schemes, cannot achieve real-time linkage feedback, resulting in delayed and one-sided sampling when facing sudden pollution, and difficulty in ensuring flight safety.

[0005] In view of the above problems, this application provides a dynamic sampling and control method and system based on water quality detection linkage feedback.

[0006] The first aspect of this application provides a dynamic sampling control method based on water quality detection linkage feedback. This method includes: acquiring real-time water quality detection data, remaining battery power, and estimated flight time for each sampling point in the current sampling round; inputting the real-time water quality detection data into a look-ahead pre-adaptation loop and a physical constraint emergency loop to generate a first control strategy and a second control strategy; calculating the current tactical degree of freedom level based on the remaining battery power and estimated flight time according to a preset energy grading rule; if the tactical degree of freedom level is sufficient, using the second control strategy as the final deployment strategy; if the tactical degree of freedom level is tense, weighted fusion of the first and second control strategies to generate a third control strategy as the final deployment strategy; and if the tactical degree of freedom level is critical, using the first control strategy as the final deployment strategy to control the UAV to execute a new round of dynamic sampling.

[0007] Optionally, the look-ahead pre-adaptation loop incorporates a time-series prediction model; the time-series prediction model predicts the water quality risk probability field at future moments based on real-time water quality monitoring data and historical water quality data; and a first control strategy with the goal of maximizing information gain is generated based on the risk probability field. The first control strategy includes a first sampling vertical position distribution, a first sampling depth distribution, and a first sampling frequency.

[0008] Optionally, the physical constraint emergency loop incorporates water quality exceedance triggering logic and a reverse diffusion solution model. When the water quality exceedance triggering logic determines that any water quality parameter exceeds a preset emergency threshold or the rate of change exceeds a sudden change threshold, the reverse diffusion solution model inverts the diffusion main axis direction based on the current spatial distribution of pollutants. Based on the diffusion main axis direction, an asymmetric densification second control strategy is generated. The second control strategy includes a second sampling vertical line position distribution, a second sampling depth distribution, and a second sampling frequency that are stretched and densified along the diffusion main axis direction.

[0009] Optionally, based on the risk probability field, a primary traction target point and a secondary traction target point are determined. The offset vectors corresponding to the primary traction target point and the secondary traction target point are weighted and superimposed to synthesize the first sampling vertical line position distribution. The risk probability profile curve of the risk probability field along the water depth direction at the current sampling vertical line position is extracted. The inflection point layer and the abrupt change layer of the risk probability change with water depth in the risk probability profile curve are identified. The encrypted sampling interval is allocated at the inflection point layer and the abrupt change layer to obtain the first sampling depth distribution. The risk probability values ​​of the same spatial location in the risk probability field at each future prediction time are extracted to form a risk time series. The cumulative value of the probability fluctuation amplitude between adjacent time moments is calculated as the total risk amplitude index. The first sampling frequency is matched according to the total risk amplitude index.

[0010] Optionally, the risk probability field is divided into planar grid cells, and the peak risk probability and its location within each grid cell are extracted. The center coordinates of the grid cell containing the highest risk probability value are taken as the primary traction target point. A first offset vector pointing from the current UAV projection position to the primary traction target point is calculated. Simultaneously, the spatial gradient distribution of the risk probability field in the horizontal direction is calculated, and the direction with the largest gradient magnitude is identified as the traction direction where the water quality parameters change most drastically. The centroids of several high gradient regions in the spatial gradient distribution whose gradient magnitudes exceed a preset active threshold are weighted to obtain secondary traction target points. A second offset vector pointing from the current UAV projection position to the secondary traction target point is calculated. The first offset vector and the second offset vector are vector-superimposed according to a preset dynamic ratio to synthesize a horizontal offset vector as the first sampling vertical line position distribution. The preset dynamic ratio is inversely proportional to the spatial dispersion of the current risk probability field.

[0011] Optionally, based on the pollutant concentration detection values ​​at each sampling point, taking the sampling point with the highest concentration as the tracking origin, a first sampling vertical line sequence is laid out along the diffusion main axis, and a second sampling vertical line sequence is laid out along the direction perpendicular to the diffusion main axis, forming a second sampling vertical line position distribution; water temperature-depth profile data and / or turbidity-depth profile data of the current water body are acquired, the location of the strata boundary in the water body is identified, and additional sampling depth layers are added at the adjacent positions above and below the strata boundary, respectively, to establish the second sampling depth distribution; the sampling frequency of each sampling point in the first sampling vertical line sequence is set as the first emergency frequency, and the sampling frequency of each sampling point in the second sampling vertical line sequence is set as the second emergency frequency, wherein the first emergency frequency is higher than the second emergency frequency, to obtain the second sampling frequency.

[0012] Optionally, the horizontal spacing between adjacent vertical lines in the first sampling vertical line sequence is smaller than the horizontal spacing between adjacent vertical lines in the second sampling vertical line sequence, and the number of the first sampling vertical line sequence is greater than the number of the second sampling vertical line sequence.

[0013] Optionally, the remaining battery percentage of the drone and the estimated percentage of battery power required for return are obtained; the estimated percentage of battery power required for return and the safety margin are subtracted from the remaining battery percentage to obtain the tactical available margin; when the tactical available margin is greater than or equal to a first preset threshold, it is determined to be in the sufficient level; when the tactical available margin is less than the first preset threshold but greater than or equal to a second preset threshold, it is determined to be in the tense level; when the tactical available margin is less than the second preset threshold, it is determined to be in the critical level.

[0014] Optionally, a dynamic fusion weight coefficient is calculated based on the available tactical margin; the first control strategy and the second control strategy are weighted and fused based on the dynamic fusion weight coefficient to generate the third control strategy.

[0015] A second aspect of this application provides a dynamic sampling and control system based on water quality detection linkage feedback. The system includes: a data acquisition module for acquiring real-time water quality detection data, remaining battery power, and estimated flight time for each sampling point in the current sampling round; a strategy generation module for inputting real-time water quality detection data into a look-ahead pre-adaptation loop and a physical constraint emergency loop to generate a first control strategy and a second control strategy; a level calculation module for calculating the current tactical degree of freedom level based on remaining battery power and estimated flight time, according to a preset energy grading rule; a first deployment strategy determination module for using the second control strategy as the final deployment strategy if the tactical degree of freedom level is sufficient; a second deployment strategy determination module for weighted fusion of the first and second control strategies to generate a third control strategy as the final deployment strategy if the tactical degree of freedom level is stressful; and a third deployment strategy determination module for using the first control strategy as the final deployment strategy if the tactical degree of freedom level is critical, controlling the UAV to execute a new round of dynamic sampling.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: The method provided in this application acquires real-time water quality monitoring data, remaining battery power, and estimated flight time for each sampling point in the current sampling round; inputs the real-time water quality monitoring data into a look-ahead pre-adaptation loop and a physical constraint emergency loop to generate a first control strategy and a second control strategy; calculates the current tactical degree of freedom level based on the remaining battery power and estimated flight time according to preset energy grading rules; if the tactical degree of freedom level is sufficient, the second control strategy is used as the final deployment strategy; if the tactical degree of freedom level is tense, the first control strategy and the second control strategy are weighted and fused to generate a third control strategy as the final deployment strategy; if the tactical degree of freedom level is critical, the first control strategy is used as the final deployment strategy to control the UAV to execute a new round of dynamic sampling. This achieves the technical effect of generating differentiated sampling strategies in parallel through dual loops and combining them with energy grading scheduling, adaptively selecting or fusing strategies based on flight time, balancing water quality monitoring needs with UAV flight safety, and improving the overall efficiency of dynamic sampling.

[0017] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 A flowchart illustrating the dynamic sampling and control method based on water quality detection linkage feedback provided in this application.

[0020] Figure 2 A schematic diagram of the dynamic sampling and control system based on water quality detection linkage feedback provided in this application.

[0021] Explanation of reference numerals in the attached diagram: Data acquisition module 11, strategy generation module 12, level calculation module 13, first policy determination module 14, second policy determination module 15, third policy determination module 16. Detailed Implementation

[0022] This application provides a dynamic sampling and control method and system based on water quality detection linkage feedback. It addresses the technical problems of existing technologies that employ fixed sampling schemes, lack real-time linkage feedback, resulting in delayed and incomplete sampling in the face of sudden pollution, and compromising flight safety. The method achieves the technical effect of improving the overall efficiency of dynamic sampling by generating differentiated sampling strategies in parallel through dual-loop processes and combining them with energy-level scheduling, adaptively selecting or fusing strategies based on endurance, thus balancing water quality monitoring needs with UAV flight safety.

[0023] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0024] Example 1 like Figure 1 As shown, this application provides a dynamic sampling and control method based on water quality detection linkage feedback, the dynamic sampling and control method based on water quality detection linkage feedback includes: Obtain real-time water quality monitoring data, remaining battery power, and estimated battery life for each sampling point in the current sampling round.

[0025] Specifically, when the current water quality sampling cycle begins, the multi-parameter water quality sensor array carried by the UAV, including but not limited to pH electrodes based on electrochemical principles, dissolved oxygen fluorescence quenching sensors, turbidity scattering light sensors, conductivity quadrupole electrodes, temperature sensors, and anodic stripping voltammetry sensors for specific heavy metal detection, completes water quality sampling at preset sampling points. The raw electrical or optical signals output by each sensor are then converted from analog to digital to form water quality parameter values, such as dissolved oxygen, pH, turbidity, temperature, and conductivity. These water quality parameter values ​​are combined with the spatial coordinates of the current sampling point, sampling depth, and sampling timestamp provided by the onboard RTK-GNSS to form real-time water quality detection data. The current sampling cycle refers to the sampling period required for the UAV to complete one full spatial coverage according to the predetermined mission plan.

[0026] The remaining battery power is obtained from the drone's Battery Management System (BMS), and the estimated flight time is obtained by combining the drone's real-time power consumption with the estimated energy consumption required for return. First, the drone's instantaneous total current and total voltage are read in real time by onboard sensors, such as an ammeter, to calculate the current real-time power consumption. Then, the estimated energy consumption required for return is obtained through a dynamic model. The onboard navigation module obtains the drone's current three-dimensional position coordinates and the coordinates of the preset return point, calculating the horizontal return distance D and altitude difference Δh. Simultaneously, the onboard anemometer measures the current ambient wind speed V. wind The equivalent headwind component V along the return path is calculated by taking the wind angle θ between the headwind and the return direction. wind ⋅cosθ. Based on the preset return cruise speed V cruise By combining the aerodynamic drag coefficient Cd of the UAV, the lift-drag characteristics of the wing or rotor, and the motor propulsion efficiency η, the instantaneous power required for each flight stage during the return journey is calculated. The entire return path is then integrated over time, and the estimated energy consumption for the return journey is accumulated. Subsequently, the remaining battery power is subtracted from the estimated energy consumption for the return journey, and then divided by the current real-time power consumption to obtain the estimated remaining operational time. By acquiring real-time water quality monitoring data, remaining battery power, and estimated operational time, comprehensive data support is provided for dynamic sampling and control, thereby improving the effectiveness and safety of control decisions.

[0027] Real-time water quality monitoring data is input into the look-ahead pre-adaptation loop and the physical constraint emergency loop to generate the first and second control strategies.

[0028] Furthermore, the look-ahead pre-adaptive loop incorporates a time-series prediction model; the time-series prediction model predicts the water quality risk probability field at future moments based on real-time water quality monitoring data and historical water quality data; based on the risk probability field, a first control strategy with the goal of maximizing information gain is generated, the first control strategy including a first sampling vertical position distribution, a first sampling depth distribution, and a first sampling frequency.

[0029] Furthermore, the physical constraint emergency loop incorporates water quality exceedance triggering logic and a reverse diffusion solution model. When the water quality exceedance triggering logic determines that any water quality parameter exceeds a preset emergency threshold or the rate of change exceeds a sudden change threshold, the reverse diffusion solution model inverts the diffusion axis direction based on the current spatial distribution of pollutants. Based on the diffusion axis direction, an asymmetric encrypted second control strategy is generated. The second control strategy includes a second sampling vertical line position distribution, a second sampling depth distribution, and a second sampling frequency that are stretched and encrypted along the diffusion axis direction.

[0030] Specifically, after the real-time water quality monitoring data for the current sampling round is collected, the real-time water quality monitoring data is input into the look-ahead pre-adaptation loop and the physical constraint emergency loop, respectively. The look-ahead pre-adaptation loop incorporates a time-series prediction model, such as a Long Short-Term Memory (LSTM) network. First, continuous time-series water quality monitoring data, including dissolved oxygen, pH, turbidity, conductivity, and specific pollutant concentrations, is extracted from the historical water quality monitoring database of the target water area. These data are uniformly organized into equally spaced time-series samples according to the sampling timestamp, with the sampling interval determined based on the actual monitoring frequency. A fixed-time-step sliding window method is used to obtain training samples: the water quality monitoring data from the current moment and the past T consecutive moments are used as the input sequence for the time-series prediction model, where T ranges from 6 to 12, depending on the data sampling period and prediction requirements. The risk probability values ​​corresponding to the actual water quality monitoring data at each sampling point at multiple prediction moments after T moments (e.g., 5 minutes, 10 minutes, 15 minutes, and 20 minutes in the future) are used as output labels to form training sample pairs. For each dimension of water quality parameters in the input sequence, Z-score normalization is used to eliminate numerical differences caused by different units, ensuring that the contribution of each parameter to the model training is relatively balanced. The output labels are compressed to the [0,1] interval using Min-Max normalization to match the range of the output layer's Sigmoid activation function. After preprocessing, all samples are randomly divided into training, validation, and test sets in a 7:2:1 ratio.

[0031] An LSTM network adapted for water quality time-series prediction tasks is constructed, with the following structure: an input layer, the number of nodes of which is equal to the dimension D of the water quality detection data input at a single time step; two LSTM hidden layers, each with a preset number of LSTM neurons, such as 128; each LSTM neuron in the hidden layer has a built-in three-gating structure: an input gate, a forget gate, and an output gate. The forget gate is used to control the retention and discard of historical water quality time-series information in the cell state; the input gate filters the newly added water quality features and updates the cell state; the output gate extracts the effective features in the cell state and outputs them outward. The three sets of gates work together to suppress the gradient vanishing phenomenon during long-term training, capture the evolution law of water quality parameters at different time scales, and introduce Dropout regularization between the two layers with a dropout rate of 0.2~0.3 to reduce the risk of overfitting; and a fully connected output layer, the number of nodes of which is equal to the number of predicted future time steps multiplied by the number of sampling points. The output layer uses the Sigmoid activation function to map the network output to the [0,1] interval, which serves as the predicted water quality risk probability value for each sampling point at each future time step. Training set samples are input into the LSTM network in batches, such as 32 or 64, for forward propagation. Mean squared error is used as the loss function to measure the deviation between the predicted risk probability and the actual risk probability. The Adam optimizer is used with an initial learning rate of 10. -3 The network weights are updated iteratively. After each training cycle, the current loss value is calculated on the validation set. If the loss on the validation set does not decrease for 10 consecutive cycles, an early stopping mechanism is triggered to terminate the training in advance to prevent overfitting. The trained time-series prediction model is then used for real-time water quality monitoring data analysis.

[0032] The time-series prediction model receives real-time water quality monitoring data from the current sampling round and uses Z-score-standardized water quality parameters as input. After forward inference calculations based on historical water quality data, it outputs the risk probability value of each sampling point at future times, forming a water quality risk probability field predicting future times. Each sampling point corresponds to a risk probability value; a higher risk probability value indicates a greater likelihood of water quality exceeding standards at that location in the future. After obtaining the risk probability field, a first control strategy is generated with the goal of maximizing information gain. Maximizing information gain means that the deployment of sampling points in the next round should prioritize covering spatial locations that minimize the uncertainty of water quality prediction. Furthermore, the first control strategy includes the first sampling vertical position distribution, the first sampling depth distribution, and the first sampling frequency.

[0033] The physical constraint emergency loop incorporates water quality exceedance triggering logic and a reverse diffusion solution model. The water quality exceedance triggering logic is a parallel comparator array that receives real-time water quality monitoring data from each sampling point in the current sampling round. Simultaneously, it reads preset emergency thresholds and abrupt change thresholds for various water quality parameters from onboard storage. These thresholds are determined based on specific limits given by surface water environmental quality standards or task assignments, such as dissolved oxygen below 2.0 mg / L, pH outside the 6-9 range, cyanide concentration exceeding 0.005 mg / L, and dissolved oxygen change rate exceeding 1.5 mg / L / min between adjacent rounds. The water quality exceedance triggering logic performs a dual judgment on each water quality parameter in the monitoring data: detecting whether the current water quality parameter exceeds the preset emergency threshold, and calculating the rate of change of the current water quality parameter relative to the same location in the previous sampling round to determine whether it exceeds the preset abrupt change threshold. When any water quality parameter at any sampling point meets any of the above triggering conditions, the water quality exceedance triggering logic outputs a trigger signal, activating the reverse diffusion solution model.

[0034] The reverse diffusion solution model uses the measured pollutant concentration values ​​of each sampling point in the current sampling round as the observation input. The spatial coordinates of these sampling points and the corresponding concentration values ​​constitute a discrete observation point set. A forward model is established based on the steady-state analytical solution of the two-dimensional convection-diffusion equation or the concentration distribution formula of the Gaussian plume model. Through optimization algorithms, parameters such as the pollution source location coordinates (x0, y0), the lateral diffusion coefficient σx, the longitudinal diffusion coefficient σy, and the average velocity direction angle θ are derived to minimize the sum of squared errors between the predicted values ​​of the forward model and the measured concentrations at each observation point. The direction of the diffusion main axis is determined by the velocity direction angle θ combined with the spatial vector pointing from the pollution source to the direction of the maximum concentration gradient. For example, in a water quality monitoring exercise, using a 3×3 grid with a grid spacing of 20m, nine sampling vertical lines measured cyanide concentrations at a bottom water depth of 6.0m. The concentrations were as follows: highest at the center point (0,0) at 15.0 μg / L; 10.2 μg / L at the east side (20m,0); 4.5 μg / L at the west side (-20m,0); 7.8 μg / L at the north side (0,20m); 13.1 μg / L at the south side (0,-20m); and 6.5 μg / L at the northeast side (20m,20m). The concentrations at the remaining points were all below 3.0 μg / L. The reverse diffusion solution model substituted these discrete concentration values ​​into the Gaussian plume steady-state diffusion equation: Where C(x,y) is the predicted concentration at location (x,y), Q is the pollutant release rate, U is the average flow velocity (0.12 m / s in this water area), and σ x and σ y Here, σ represents the diffusion coefficients in the downwind and crosswind directions, respectively. zLet σ be the vertical diffusion coefficient, and (x0, y0) be the location of the pollution source to be inverted. The Levenberg-Marquardt nonlinear optimization algorithm is used iteratively to solve the problem, with the objective function being to minimize the root mean square error between the observed and predicted concentrations. After 15 iterations, the algorithm converges, yielding the inversion result: pollution source location (x0, y0) = (2.3m, -3.1m), and downwind diffusion coefficient σ. x =14.5m, crosswind diffusion coefficient σ y =8.2m, average flow direction angle θ=156°, that is, with true north as 0° and clockwise direction, the flow direction is 24° south of east. The reverse diffusion solution model determines the main diffusion axis direction as 156°. That is, the pollutants diffuse and advance most significantly from the pollution source to the southeast-southeast direction along this direction. Therefore, the sampling vertical lines are stretched and densified along this main axis direction. When the amount of observation data is small, the Tikhonov regularization method is used to add a regularization term to the objective function. The spatial smoothness and boundedness of the constraint solution are ensured, with the regularization parameter λ automatically selected using the L-curve method to guarantee the stability and rationality of the inversion results.

[0035] After obtaining the main diffusion axis direction, the physical constraint emergency loop generates an asymmetric densification second control strategy based on this. Asymmetric densification refers to the non-uniform, anisotropic spatial distribution of sampling points, with dense points along the main diffusion axis and sparse points in the vertical direction, to highlight the core diffusion path of the pollution plume. The second control strategy includes the distribution of the second sampling vertical line position, the second sampling depth distribution, and the second sampling frequency, all of which are based on the densification along the main diffusion axis.

[0036] By employing a dual-loop parallel architecture of a look-ahead pre-adaptive loop and a physical constraint emergency loop, dual-objective differentiated processing of the same set of real-time water quality monitoring data is achieved: the look-ahead pre-adaptive loop uses a time-series prediction model to extrapolate the probability field of future water quality risks and generates prediction-driven strategies guided by maximizing information gain; the physical constraint emergency loop uses exceedance triggering logic and a reverse diffusion solution model to quickly locate the occurrence of sudden pollution and generate event-driven strategies. This achieves an organic unity between two modes: active sampling under normal operating conditions and rapid emergency response under sudden operating conditions, improving the effectiveness and accuracy of dynamic sampling decisions, while also enhancing the precision and reliability of water quality monitoring.

[0037] Furthermore, the first control strategy, which aims to maximize information gain, is generated based on the risk probability field, including: determining the primary traction target point and the secondary traction target point based on the risk probability field; weighting and superimposing the offset vectors corresponding to the primary traction target point and the secondary traction target point to synthesize the first sampling vertical line position distribution; extracting the risk probability profile curve of the risk probability field along the water depth direction at the current sampling vertical line position; identifying the inflection point layer and the abrupt change layer in the risk probability profile curve where the risk probability changes with water depth; allocating a denser sampling interval at the inflection point layer and the abrupt change layer to obtain the first sampling depth distribution; extracting the risk probability values ​​of the same spatial location at each future prediction time in the risk probability field to form a risk time series; calculating the cumulative value of the probability fluctuation amplitude between adjacent time moments as the total risk amplitude index; and matching the first sampling frequency according to the total risk amplitude index.

[0038] Further, based on the risk probability field, the primary traction target point and the secondary traction target point are determined. The offset vectors corresponding to the primary traction target point and the secondary traction target point are weighted and superimposed to synthesize the first sampling vertical line position distribution. This includes: dividing the risk probability field into planar grid cells, extracting the risk probability peak value and its location within each grid cell, taking the center coordinates of the grid cell with the highest risk probability value as the primary traction target point, and calculating the first offset vector pointing from the current UAV projection position to the primary traction target point; simultaneously calculating the spatial gradient distribution of the risk probability field in the horizontal direction, identifying the direction with the largest gradient magnitude as the traction direction with the most drastic changes in water quality parameters; calculating the weighted centroid of several high gradient regions in the spatial gradient distribution whose gradient magnitude exceeds a preset active threshold to obtain the secondary traction target point, and calculating the second offset vector pointing from the current UAV projection position to the secondary traction target point; and superimposing the first offset vector and the second offset vector according to a preset dynamic ratio to synthesize a horizontal offset vector as the first sampling vertical line position distribution; wherein, the preset dynamic ratio is inversely proportional to the spatial dispersion of the current risk probability field.

[0039] Specifically, after the time-series prediction model built into the look-ahead pre-adaptation loop outputs the water quality risk probability field for future moments, the predicted water quality risk probability field is divided into regularly arranged planar grid cells on a horizontal plane at a preset resolution, such as 0.5m × 0.5m. For each grid cell, the peak risk probability and its corresponding spatial coordinates are extracted. The peak risk probability refers to the maximum risk probability of all sampled prediction points within that grid cell. After traversing all grid cells, the center coordinates of the grid cell containing the highest risk probability value are determined as the main traction target point, representing the location with the highest probability of future water quality deterioration and a core area requiring priority attention. Then, the current latitude and longitude coordinates of the UAV are obtained via airborne RTK-GNSS, along with the latitude and longitude coordinates of the main traction target point. Both are converted from the geodetic coordinate system (latitude and longitude) to a local planar coordinate system via Gauss-Kruger projection to obtain the horizontal projection position of the UAV and the planar coordinates of the target point. The coordinate difference between the two points in the X and Y directions is then calculated, and this coordinate difference is used as the horizontal component of the spatial vector, i.e., the first offset vector.

[0040] Simultaneously, the spatial gradient distribution of the risk probability field in the horizontal direction is calculated. For each grid cell, the first-order partial derivatives of its risk probability value in the X and Y directions are calculated. Using the central difference method, the difference in risk probability between this grid cell and its adjacent cells is divided by the cell spacing. Then, the partial derivatives in the two directions are synthesized to obtain the gradient magnitude and gradient direction of this grid cell. The gradient magnitude reflects the degree of spatial variation in risk probability. The larger the gradient magnitude, the more drastic the change in water quality parameters in the horizontal direction, which is usually the location of pollution plume boundaries or hydrological fronts. The direction with the largest gradient magnitude is identified as the direction of most drastic change in water quality parameters. The algorithm identifies high-gradient regions in the spatial gradient distribution whose gradient magnitudes exceed a preset active threshold. This threshold is set based on historical statistical experience, for example, selecting regions with gradient magnitudes 1.5 times the average gradient magnitude in the current risk probability field. The centroids of these high-gradient regions are then calculated using a weighted average based on their respective gradient magnitudes, with regions having higher gradient magnitudes receiving greater weight in the centroid calculation. The calculated centroid coordinates are used as secondary traction target points, representing the center of the region with the most dramatic spatial variations and the richest information regarding water quality parameters. Similar to the first offset vector calculation, the spatial vector pointing from the current UAV's horizontal projection position to this secondary traction target point is calculated and used as the second offset vector.

[0041] After obtaining the first and second offset vectors, they are superimposed according to a preset dynamic ratio. This preset dynamic ratio is inversely proportional to the spatial dispersion of the current risk probability field. When the risk probability field exhibits a unimodal clustering pattern, meaning the risk is highly concentrated in a local area with low dispersion, it indicates that the main controlling factors of future risks are clear, and the highest risk center should be prioritized. In this case, the weight of the first offset vector is higher than that of the second offset vector, with a ratio of, for example, 0.7:0.3. When the risk probability field exhibits a multimodal diffuse pattern, meaning the risk is dispersed across multiple discontinuous areas with high dispersion, it indicates a complex risk distribution and the existence of multiple potential pollution sources or pathways. The boundary areas with drastic changes should be prioritized to obtain more discriminative information, with the weight of the second offset vector higher than that of the first offset vector, with a ratio of, for example, 0.4:0.6. Spatial dispersion is represented by the coefficient of variation, i.e., the ratio of the standard deviation to the mean. The larger the coefficient of variation, the more diffuse the risk probability field, and the higher the weight of the secondary traction target points. According to the preset dynamic ratio, the first offset vector and the second offset vector are multiplied by their respective weight coefficients and then added together to obtain the horizontal offset vector. The calculated horizontal offset vector is then superimposed on the sampling vertical line according to its original layout position to obtain the position distribution of the first sampling vertical line.

[0042] Based on the spatial location of the current sampling vertical line, the risk probability field at that location is extracted along the water depth direction, from the water surface to the bottom, typically with sampling intervals of 0.5m or 1.0m, forming a risk probability profile curve for that sampling vertical line location. The risk probability profile curve is analyzed using a sliding window smoothing algorithm, with a window width of 3-5 depth points to eliminate local spikes caused by prediction noise. Then, the first and second derivatives of the risk probability with respect to water depth at each depth point are calculated. The depth location where the second derivative is zero and the first derivative reaches a local extreme value is defined as the inflection point layer. The rate of change of the risk probability with increasing water depth at the inflection point layer changes from increasing to decreasing or vice versa, indicating a trend reversal in the vertical variation of water quality parameters at this point. The differential detection method is used to calculate the absolute value of the difference between the risk probability values ​​between adjacent depth layers. If the absolute value of the difference continuously exceeds the preset abrupt change threshold in multiple consecutive depth layers, the depth interval is marked as an abrupt change layer. The abrupt change layer is usually the location of the density interface of pollutants or the chemical gradient layer. The preset abrupt change threshold is set to the risk probability value changing by more than 0.15 in every 0.5 meters of water depth.

[0043] After identifying the inflection points and abrupt change zones where risk probability changes with water depth, denser sampling intervals are allocated at these zones. This denser interval, for example, compressing the default 2.0m interval to 0.3m-0.5m, ensures sufficiently dense sample data at these critical depths where water quality parameters exhibit dramatic vertical variations. Simultaneously, sparser sampling intervals, such as 3.0m-4.0m, are allocated in depths where risk probability tends to flatten, to avoid wasting sampling resources in depths with uniform water quality. Furthermore, transitional sampling points are forcibly deployed at the boundary between dense and sparse sampling intervals to ensure data continuity between areas with different sampling densities and prevent vertical profile information breaks caused by abrupt changes in sampling intervals. The same depth distribution calculation is performed on all sampling verticals to obtain the first sampling depth distribution.

[0044] For each spatial location in the risk probability field, the predicted risk probability value of that location at each future prediction time is extracted and arranged in chronological order to form the risk time series of that location. Then, the probability fluctuation amplitude between adjacent times in the risk time series is calculated by the difference, and the fluctuation amplitudes between all adjacent times are summed to obtain the total risk amplitude index of that location. The larger the total amplitude, the more drastic and unstable the water quality parameters at that location will change in the future. The first sampling frequency is matched based on the total risk amplitude index. Preferably, the total risk amplitude index is matched to a specific sampling frequency value according to a preset frequency mapping rule: when the total risk amplitude index is higher than the first threshold, such as >0.6, it is determined that the risk time series at this location fluctuates violently, and a high-density sampling frequency is allocated, such as 4-5 samples per round; when the total risk amplitude index is lower than the second threshold, such as <0.2, it is determined that the risk time series at this location tends to be stable, and a basic sampling frequency is allocated, such as 1 sample per round; when the total risk amplitude index is between the first and second thresholds, a medium sampling frequency is allocated according to the linear interpolation ratio, such as 2-3 samples per round. Simultaneously, the single-step fluctuation amplitude in the risk time series is monitored: if the fluctuation amplitude between adjacent times exceeds a preset jump threshold, such as a single-step fluctuation exceeding 0.35, it indicates that a sudden water quality change may occur at this location within a specific time period, and an additional temporary sampling task is added for this location. The same frequency calculation is performed on all spatial locations to obtain the first sampling frequency.

[0045] By employing a weighted overlay mechanism of primary and secondary target points, the system achieves comprehensive sampling of both high-risk core areas and high-information boundary areas in the horizontal direction. Through intensive identification of inflection points and abrupt change layers, it provides a precise reflection of key interfaces in the vertical distribution of pollutants. Furthermore, by frequency matching of the total risk amplitude index, it enables high-frequency tracking of locations with drastic water quality changes in the temporal dimension. These three mechanisms work together to ensure that the primary control strategy, with the goal of maximizing information gain, achieves optimal allocation of sampling resources simultaneously in both three-dimensional space and time dimensions.

[0046] Furthermore, an asymmetric encryption second control strategy is generated based on the diffusion axis direction. The second control strategy includes a second sampling vertical line position distribution, a second sampling depth distribution, and a second sampling frequency that are stretched and encrypted along the diffusion axis direction. This includes: based on the pollutant concentration detection values ​​at each sampling point, taking the highest concentration sampling point as the tracking origin, a first sampling vertical line sequence is laid out along the diffusion axis direction, and a second sampling vertical line sequence is laid out along a direction perpendicular to the diffusion axis direction to form a second sampling vertical line position distribution; water temperature-depth profile data and / or turbidity-depth profile data of the current water body are acquired, the location of the strata boundary in the water body is identified, and additional sampling depth layers are added at adjacent positions above and below the strata boundary to establish the second sampling depth distribution; the sampling frequency of each sampling point in the first sampling vertical line sequence is set as a first emergency frequency, and the sampling frequency of each sampling point in the second sampling vertical line sequence is set as a second emergency frequency, where the first emergency frequency is higher than the second emergency frequency, thus obtaining the second sampling frequency.

[0047] Furthermore, the horizontal spacing between adjacent vertical lines in the first sampling vertical line sequence is smaller than the horizontal spacing between adjacent vertical lines in the second sampling vertical line sequence, and the number of the first sampling vertical line sequence is greater than the number of the second sampling vertical line sequence.

[0048] Specifically, from the real-time water quality monitoring data of the current sampling round, the pollutant concentration values ​​at each sampling point are traversed. Taking the sampling point with the highest concentration as the tracking origin, a first sampling vertical line sequence is laid out along the main diffusion axis, and a second sampling vertical line sequence is laid out perpendicular to this main diffusion axis, forming the second sampling vertical line position distribution. An asymmetric densification strategy is used for the placement of the sampling vertical line sequences. Specifically, the horizontal spacing between adjacent vertical lines in the first sampling vertical line sequence is smaller than the horizontal spacing between adjacent vertical lines in the second sampling vertical line sequence, and the number of first sampling vertical line sequences is greater than the number of second sampling vertical line sequences. Because the concentration gradient of pollutants is smaller and the distribution range is wider along the main diffusion axis, dense placement is needed to accurately track the extension path and concentration decay trend of pollutants. In the direction perpendicular to the main axis, the pollutant concentration gradient is larger and the distribution range is narrower; sparse placement can effectively delineate the lateral boundary of the pollution plume. The tracking origin itself is also the intersection of the two sampling vertical line sequences, forming the central reference point of the entire second sampling vertical line position distribution. For example, in a pollution event, the main axis of diffusion is 35° east of north. Taking the highest concentration point (10.0m, 12.0m) as the origin, 10 dense vertical lines are laid out at 4-meter intervals along the 35° east of north direction, located at -16m, -12m, -8m, -4m, 0m, +4m, +8m, +12m, +16m, and +20m along the main axis. 4 sparse vertical lines are laid out at 20-meter intervals along the 55° west of north direction, located at -30m, -10m, +10m, and +30m along the vertical direction. A total of 14 sampling vertical lines form the second sampling vertical line distribution. This asymmetric dense layout ensures that limited sampling resources are concentrated on the core path of pollution diffusion.

[0049] Based on water temperature continuously measured along the water depth by a multi-parameter water quality sensor array mounted on a drone, a water temperature-depth profile is generated for the current water area. Simultaneously, based on the collected turbidity, a turbidity-depth profile is generated. Differential analysis is performed on the temperature-depth profile data to identify the temperature gradient boundary where the temperature changes with water depth at the greatest rate. The temperature gradient boundary is defined as a continuous depth range where the temperature gradient exceeds a preset threshold, such as a temperature change rate greater than 0.5℃ / m. Similarly, the turbidity gradient boundary is identified for the turbidity-depth profile data where turbidity rapidly increases or decreases. The turbidity gradient boundary is defined as a continuous depth range where the turbidity change rate exceeds a preset threshold, such as a turbidity change rate greater than 5 NTU / m. Additional sampling depth layers are added at the adjacent locations above and / or below the temperature and / or turbidity gradient boundaries, such as 0.5–1.0 meters above and below the gradient boundary, respectively, while retaining the regular sampling depth layer, forming a second sampling depth distribution.

[0050] The sampling frequency of each sampling point in the first sampling vertical sequence is then set as the first emergency frequency, such as 5 times per round. The sampling frequency of each sampling point in the second sampling vertical sequence is set as the second emergency frequency, such as 2-3 times per round, resulting in the second sampling frequency. Both the first and second emergency frequencies are higher than the normal sampling frequency, which is typically 1-2 times per round under normal conditions, with the first emergency frequency higher than the second emergency frequency, reflecting the enhanced temporal resolution of the entire contaminated area under emergency response conditions. The second sampling frequency, together with the second sampling vertical position distribution and the second sampling depth distribution, constitutes a complete second control strategy.

[0051] By employing an asymmetric density strategy of densely distributing sampling points along the main diffusion axis and sparsely distributing them vertically, limited emergency sampling resources are precisely focused on the core path of pollution diffusion. The addition of layers adjacent to the boundary between different layers ensures accurate capture of concentration differences of pollutants on both sides of the vertical density interface. High-frequency sampling along the main axis ensures the temporal continuity of the dynamic evolution process in the core pollution area. The synergistic effect of these three factors enables the second control strategy to achieve rapid and accurate localization of the spatial range, diffusion path, and vertical distribution of the pollution plume with the fewest sampling points and the shortest sampling time after a sudden pollution event.

[0052] Based on the remaining battery power and estimated flight time, the current tactical freedom level is calculated according to the preset energy classification rules.

[0053] Furthermore, based on the remaining battery power and estimated flight time, the current tactical freedom level is calculated according to preset energy grading rules, including: obtaining the percentage of the drone's remaining battery power and the percentage of power required for the estimated return flight; subtracting the percentage of power required for the estimated return flight and a safety margin from the percentage of the remaining battery power to obtain the tactical available margin; when the tactical available margin is greater than or equal to a first preset threshold, it is determined to be at the sufficient level; when the tactical available margin is less than the first preset threshold but greater than or equal to a second preset threshold, it is determined to be at the tense level; when the tactical available margin is less than the second preset threshold, it is determined to be at the critical level.

[0054] Specifically, the remaining battery percentage E and the estimated return-to-home percentage R are obtained. The remaining battery percentage E is then subtracted from the estimated return-to-home percentage R and the safety margin S to obtain the tactical available margin M. The formula for calculating the tactical available margin M is: M = ERS, where S is the safety margin, such as 10%. The tactical available margin M represents the upper limit of the proportion of battery power that the drone can freely use for non-return-to-home flights after deducting the return-to-home battery power that must be reserved and the safety redundancy when performing the current task. The larger the value of M, the more abundant the drone's current energy reserves are, and the more sampling actions it can support. The smaller the value of M, or even close to zero, the more the drone's energy reserves are approaching the return-to-home critical line, and the safety of the return-to-home must be prioritized.

[0055] The tactical available margin is compared with a first preset threshold and a second preset threshold to classify the UAV's current energy status into three levels: sufficient, strained, and critical. The first preset threshold, such as 40%, and the second preset threshold, such as 15%, are engineering experience values ​​pre-stored in the onboard parameter configuration file. Their specific values ​​are comprehensively calibrated based on the UAV model's endurance characteristics, the area of ​​the mission waterway, and the minimum power requirement for safe return. When the tactical available margin is greater than or equal to the first preset threshold, it is classified as sufficient, indicating that the UAV currently has ample power beyond the power required for return and the safety margin, fully capable of supporting high-energy-consuming emergency encrypted sampling strategies. When the tactical available margin is less than the first preset threshold but greater than or equal to the second preset threshold, it is classified as strained, indicating that the UAV's current power is at a moderate level, sufficient to support continued operation but unable to withstand excessively high-energy-consuming sampling strategies. When the tactical available margin is less than the second preset threshold, it is classified as critical, indicating that the UAV's current power is severely insufficient, only meeting the minimum return requirements; any additional high-energy-consuming sampling operations may prevent the UAV from safely returning.

[0056] By quantifying the state of energy of UAV batteries into tactical degrees of freedom that can be used for strategic scheduling, a direct link is established between water quality sampling strategy planning and UAV flight safety constraints. This avoids generating high-energy-consuming sampling schemes based solely on water quality monitoring targets while ignoring endurance limits, thus achieving a dynamic balance between monitoring tasks and UAV return-to-home safety.

[0057] If the tactical degree of freedom level is sufficient, the second control strategy shall be used as the final issuance strategy.

[0058] If the tactical freedom level is tense, then the first control strategy and the second control strategy are weighted and fused to generate a third control strategy as the final deployment strategy.

[0059] If the tactical degree of freedom level is critical, the first control strategy is used as the final issuing strategy to control the UAV to perform a new round of dynamic sampling.

[0060] Furthermore, the first control strategy and the second control strategy are weighted and fused to generate a third control strategy, including: calculating a dynamic fusion weight coefficient based on the available tactical margin; and weighting and fusing the first control strategy and the second control strategy based on the dynamic fusion weight coefficient to generate the third control strategy.

[0061] Specifically, when the tactical freedom level is determined to be sufficient, it indicates that the drone's current battery power is extremely abundant. In this case, the generated second control strategy is used as the final deployment strategy. When the tactical freedom level is determined to be critical, it indicates that the drone's current battery power is severely insufficient, and the first control strategy is used as the final deployment strategy. When the tactical freedom level is tense, it indicates that the drone's current battery power is at a moderate level. The dynamic fusion weight coefficient is calculated based on the ratio of the available tactical margin to a first preset threshold. The higher the ratio, the closer the current energy is to the sufficient level, and the higher the dynamic fusion weight coefficient is assigned to the second control strategy. The lower the ratio, the closer the current energy is to the critical level, and the higher the dynamic fusion weight coefficient is assigned to the first control strategy. To prevent extreme weights from causing the fusion strategy to be too biased towards one side and losing the compromise effect, the dynamic fusion weight coefficient is limited, and its value range is forcibly limited to between a preset lower limit and upper limit. Preferably, the dynamic fusion weight coefficient is set to 0.2~0.8. When the dynamic fusion weight coefficient is less than 0.2, it is 0.2; when the dynamic fusion weight coefficient is greater than 0.8, it is 0.8. The dynamic fusion weight coefficient after amplitude limiting is the fusion weight α of the first control strategy, and the fusion weight of the second control strategy is 1-α. The amplitude limiting mechanism ensures that even when the available tactical margin is extremely low or high, the fusion strategy can still retain a certain proportion of the opponent's strategy components, avoiding abrupt strategy changes caused by the weight returning to zero.

[0062] Based on the dynamic fusion weight coefficients, weighted fusion is performed on the same type of control parameters in the first and second control strategies. Specifically, for each type of control parameter, the parameter values ​​in the first and second control strategies are first uniformly converted into normalized numerical vectors that can be used for arithmetic operations. For example, for the sampling vertical line position distribution, the planar coordinates of each vertical line are encoded as a position vector; for the sampling depth distribution, each depth layer is encoded as a depth vector; and for the sampling frequency, the frequency of each sampling point is encoded as a frequency vector. Then, weighted fusion is performed to generate the third control strategy, and the scheme parameter P of the third control strategy is P = α·P. A +(1-α)·P B , where P A and P BThe normalized numerical vectors corresponding to the same type of control parameters in the first and second control strategies are respectively included, comprising three types of parameters: sampling vertical position distribution, sampling depth distribution, and sampling frequency. Through fusion, the coverage of high-risk prediction areas by the first control strategy is preserved, while the tracking of the main axis of pollution diffusion by the second control strategy is incorporated, achieving an optimal trade-off under the energy constraint of the tension level. When the tactical degree of freedom level is tension level, the generated third control strategy is used as the final deployment strategy.

[0063] Regardless of whether the final issued strategy is the second control strategy at the sufficient level, the third control strategy at the tense level, or the first control strategy at the critical level, the control parameters in the strategy, including the target plane coordinates of each sampling vertical, the depth layer sequence of each vertical, and the sampling frequency of each sampling point, are converted into flight control commands and sampler action commands for the UAV. The flight control commands include parameters such as waypoint coordinate sequence, hovering time, and climb / fall speed, and are executed by the flight control module. The sampler action commands include parameters such as the water sampling triggering sequence and water sampling volume at each depth layer, and are executed by the onboard sampling controller. After the UAV completes a new round of dynamic sampling according to the final issued strategy, it uses the new water quality data obtained in this round of sampling as the input for the next round of the cycle, and returns to: obtain the real-time water quality detection data of each sampling point in the current sampling round, the remaining power, and the estimated flight time, forming a continuous closed-loop dynamic sampling and control.

[0064] Based on the determination results of the tactical degree of freedom level, differentiated strategy selection is executed, realizing dynamic optimal matching between the UAV's energy state and sampling strategy. This improves the UAV's autonomous survivability, energy utilization efficiency, and emergency response quality in complex water monitoring tasks, avoids extreme situations such as premature power depletion or inadequate emergency response caused by fixed strategies, balances water quality monitoring needs with UAV flight safety, and enhances the overall efficiency of dynamic sampling.

[0065] Example 2 Based on the same inventive concept as the dynamic sampling and control method based on water quality detection linkage feedback in the aforementioned embodiments, such as Figure 2 As shown, this application provides a dynamic sampling and control system based on water quality detection linkage feedback, wherein the dynamic sampling and control system based on water quality detection linkage feedback includes: The data acquisition module 11 is used to acquire real-time water quality detection data, remaining power, and estimated flight time of each sampling point in the current sampling round; the strategy generation module 12 is used to input the real-time water quality detection data into the look-ahead pre-adaptation loop and the physical constraint emergency loop to generate a first control strategy and a second control strategy; the level calculation module 13 is used to calculate the current tactical degree of freedom level based on the remaining power and estimated flight time, according to a preset energy grading rule; the first deployment strategy determination module 14 is used to use the second control strategy as the final deployment strategy if the tactical degree of freedom level is sufficient; the second deployment strategy determination module 15 is used to perform a weighted fusion of the first control strategy and the second control strategy if the tactical degree of freedom level is tense, to generate a third control strategy as the final deployment strategy; the third deployment strategy determination module 16 is used to use the first control strategy as the final deployment strategy if the tactical degree of freedom level is critical, to control the UAV to execute a new round of dynamic sampling.

[0066] Furthermore, the strategy generation module 12 is also used for: embedding a time-series prediction model in the look-ahead pre-adaptation loop; predicting the water quality risk probability field at future moments based on real-time water quality detection data and historical water quality data through the time-series prediction model; and generating a first control strategy with the goal of maximizing information gain based on the risk probability field. The first control strategy includes a first sampling vertical position distribution, a first sampling depth distribution, and a first sampling frequency.

[0067] Furthermore, the strategy generation module 12 is also used for: the physical constraint emergency loop has a built-in water quality exceedance triggering logic and a reverse diffusion solution model; when the water quality exceedance triggering logic determines that any water quality parameter exceeds the preset emergency threshold or the rate of change exceeds the mutation threshold, the reverse diffusion solution model inverts the diffusion main axis direction according to the current spatial distribution of pollutants; and generates an asymmetric encrypted second control strategy based on the diffusion main axis direction, the second control strategy including the second sampling vertical line position distribution, the second sampling depth distribution and the second sampling frequency stretched and encrypted along the diffusion main axis direction.

[0068] Furthermore, the strategy generation module 12 is also used to: determine the primary traction target point and the secondary traction target point based on the risk probability field; weight and superimpose the offset vectors corresponding to the primary traction target point and the secondary traction target point to synthesize the first sampling vertical line position distribution; extract the risk probability profile curve of the risk probability field along the water depth direction at the current sampling vertical line position; identify the inflection point layer and the abrupt change layer of the risk probability change with water depth in the risk probability profile curve; allocate a dense sampling interval at the inflection point layer and the abrupt change layer to obtain the first sampling depth distribution; extract the risk probability values ​​of the same spatial location in the risk probability field at each future prediction time to form a risk time series; calculate the cumulative value of the probability fluctuation amplitude between adjacent time moments as the total risk amplitude index; and match the first sampling frequency according to the total risk amplitude index.

[0069] Furthermore, the strategy generation module 12 is also used to: divide the risk probability field into planar grid cells, extract the risk probability peak value and its location in each grid cell, take the center coordinates of the grid cell where the highest risk probability value is located as the main traction target point, and calculate the first offset vector pointing from the current UAV projection position to the main traction target point; simultaneously calculate the spatial gradient distribution of the risk probability field in the horizontal direction, identify the direction with the largest gradient magnitude as the traction direction where the water quality parameter changes most drastically; perform weighted centroid calculation on several high gradient regions in the spatial gradient distribution where the gradient magnitude exceeds a preset active threshold to obtain the secondary traction target point, and calculate the second offset vector pointing from the current UAV projection position to the secondary traction target point; and vector superimpose the first offset vector and the second offset vector according to a preset dynamic ratio to synthesize a horizontal offset vector as the first sampling vertical line position distribution; wherein, the preset dynamic ratio is inversely proportional to the spatial dispersion of the current risk probability field.

[0070] Furthermore, the strategy generation module 12 is also used to: based on the pollutant concentration detection values ​​of each sampling point, take the sampling point with the highest concentration as the tracking origin, arrange a first sampling vertical line sequence along the diffusion main axis direction, and arrange a second sampling vertical line sequence along the direction perpendicular to the diffusion main axis direction to form a second sampling vertical line position distribution; acquire the water temperature-depth profile data and / or turbidity-depth profile data of the current water body, identify the position of the stratification boundary in the water body, and add additional sampling depth layers at the upper and lower adjacent positions of the stratification boundary to establish the second sampling depth distribution; set the sampling frequency of each sampling point in the first sampling vertical line sequence as a first emergency frequency, set the sampling frequency of each sampling point in the second sampling vertical line sequence as a second emergency frequency, wherein the first emergency frequency is higher than the second emergency frequency, to obtain the second sampling frequency.

[0071] Furthermore, the strategy generation module 12 is also configured to: the horizontal spacing between adjacent vertical lines in the first sampling vertical line sequence is less than the horizontal spacing between adjacent vertical lines in the second sampling vertical line sequence, and the number of the first sampling vertical line sequence is greater than the number of the second sampling vertical line sequence.

[0072] Furthermore, the level calculation module 13 is also used to: obtain the remaining battery power percentage of the drone and the estimated power percentage required for return; subtract the estimated power percentage required for return and the safety margin from the remaining battery power percentage to obtain the tactical available margin; when the tactical available margin is greater than or equal to a first preset threshold, it is determined to be at the sufficient level; when the tactical available margin is less than the first preset threshold but greater than or equal to a second preset threshold, it is determined to be at the tense level; when the tactical available margin is less than the second preset threshold, it is determined to be at the critical level.

[0073] Furthermore, the second distribution strategy determination module 15 is also used to: calculate the dynamic fusion weight coefficient based on the available tactical margin; and generate the third control strategy by weighting and fusing the first control strategy and the second control strategy based on the dynamic fusion weight coefficient.

[0074] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The dynamic sampling and control method and specific examples based on water quality detection linkage feedback in the aforementioned embodiment one are also applicable to the dynamic sampling and control system based on water quality detection linkage feedback in this embodiment. Through the foregoing detailed description of the dynamic sampling and control method based on water quality detection linkage feedback, those skilled in the art can clearly understand the dynamic sampling and control system based on water quality detection linkage feedback in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0075] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0076] Obviously, those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A dynamic sampling and control method based on water quality detection linkage feedback, characterized in that, include: Obtain real-time water quality monitoring data, remaining battery power, and estimated battery life for each sampling point in the current sampling round; Real-time water quality monitoring data is input into the look-ahead pre-adaptation loop and the physical constraint emergency loop to generate the first and second control strategies. Based on the remaining battery power and estimated flight time, the current tactical degree of freedom level is calculated according to the preset energy classification rules. If the tactical degree of freedom level is sufficient, the second control strategy shall be used as the final issuance strategy. If the tactical freedom level is tense, then the first control strategy and the second control strategy are weighted and fused to generate a third control strategy as the final issuance strategy. If the tactical degree of freedom level is critical, the first control strategy is used as the final issuing strategy to control the UAV to perform a new round of dynamic sampling.

2. The dynamic sampling and control method based on water quality detection linkage feedback according to claim 1, characterized in that, The look-ahead pre-adaptive loop incorporates a time-series prediction model; The time-series prediction model predicts the probability field of water quality risk at future moments based on real-time water quality monitoring data and historical water quality data. The first control strategy, which aims to maximize information gain, is generated based on the risk probability field. The first control strategy includes a first sampling vertical position distribution, a first sampling depth distribution, and a first sampling frequency.

3. The dynamic sampling and control method based on water quality detection linkage feedback according to claim 1, characterized in that, The physical constraint emergency loop incorporates water quality exceeding standard triggering logic and a reverse diffusion solution model; When the water quality exceedance trigger logic determines that any water quality parameter exceeds the preset emergency threshold or the rate of change exceeds the mutation threshold, the reverse diffusion solution model inverts the diffusion main axis direction based on the current spatial distribution of pollutants. The second control strategy for generating asymmetric encryption based on the diffusion principal axis direction includes the second sampling vertical line position distribution, the second sampling depth distribution, and the second sampling frequency for stretching encryption along the diffusion principal axis direction.

4. The dynamic sampling and control method based on water quality detection linkage feedback according to claim 2, characterized in that, The first control strategy, which aims to maximize information gain, is generated based on the risk probability field, including: Based on the risk probability field, the main traction target point and the secondary traction target point are determined. The offset vectors corresponding to the main traction target point and the secondary traction target point are weighted and superimposed to synthesize the position distribution of the first sampling vertical line. Extract the risk probability profile curve of the risk probability field along the water depth direction at the current sampling vertical position, identify the inflection point layer and the abrupt change layer of the risk probability change with water depth in the risk probability profile curve, and allocate the densified sampling interval at the inflection point layer and the abrupt change layer to obtain the first sampling depth distribution. The risk probability values ​​of the same spatial location at each future prediction time are extracted from the risk probability field to form a risk time series. The cumulative value of the probability fluctuation amplitude between adjacent time moments is calculated as the total risk amplitude index. The first sampling frequency is matched according to the total risk amplitude index.

5. The dynamic sampling and control method based on water quality detection linkage feedback according to claim 4, characterized in that, Based on the risk probability field, the primary traction target point and the secondary traction target point are determined. The offset vectors corresponding to the primary traction target point and the secondary traction target point are weighted and superimposed to synthesize the first sampling vertical line position distribution, including: The risk probability field is divided into planar grid cells. The risk probability peak and its location in each grid cell are extracted. The center coordinates of the grid cell where the highest risk probability value is located are taken as the main traction target point. The first offset vector pointing the current UAV projection position to the main traction target point is calculated. Simultaneously, the spatial gradient distribution of the risk probability field in the horizontal direction is calculated, and the direction with the largest gradient magnitude is identified as the direction of most dramatic change in water quality parameters. The weighted centroid of several high gradient regions in the spatial gradient distribution whose gradient magnitude exceeds a preset active threshold is calculated to obtain the secondary traction target point, and the second offset vector pointing the current UAV projection position to the secondary traction target point is calculated. The first offset vector and the second offset vector are vector-superimposed according to a preset dynamic ratio to synthesize a horizontal offset vector as the position distribution of the first sampling vertical line. The preset dynamic ratio is inversely proportional to the spatial dispersion of the current risk probability field.

6. The dynamic sampling and control method based on water quality detection linkage feedback according to claim 3, characterized in that, The second control strategy for generating asymmetric encryption based on the diffusion principal axis direction includes stretching the second sampling vertical line position distribution, the second sampling depth distribution, and the second sampling frequency along the diffusion principal axis direction, including: Based on the pollutant concentration detection values ​​at each sampling point, the sampling point with the highest concentration is taken as the tracking origin. A first sampling vertical line sequence is arranged along the diffusion axis, and a second sampling vertical line sequence is arranged along the direction perpendicular to the diffusion axis, forming the second sampling vertical line position distribution. Acquire water temperature-depth profile data and / or turbidity-depth profile data of the current water body, identify the location of the strata boundary in the water body, add additional sampling depth layers at the adjacent locations above and below the strata boundary respectively, and establish the second sampling depth distribution; The sampling frequency of each sampling point in the first sampling vertical sequence is set as the first emergency frequency, and the sampling frequency of each sampling point in the second sampling vertical sequence is set as the second emergency frequency. The first emergency frequency is higher than the second emergency frequency, thus obtaining the second sampling frequency.

7. The dynamic sampling and control method based on water quality detection linkage feedback according to claim 6, characterized in that, The horizontal spacing between adjacent vertical lines in the first sampling vertical line sequence is smaller than the horizontal spacing between adjacent vertical lines in the second sampling vertical line sequence, and the number of the first sampling vertical line sequence is greater than the number of the second sampling vertical line sequence.

8. The dynamic sampling and control method based on water quality detection linkage feedback according to claim 1, characterized in that, Based on remaining battery power and estimated flight time, the current tactical freedom level is calculated according to preset energy grading rules, including: Obtain the remaining battery percentage of the drone and the estimated percentage of battery power required for return; Subtract the estimated percentage of battery power needed for return and the safety margin from the remaining battery power percentage to obtain the tactical available margin; When the available tactical margin is greater than or equal to the first preset threshold, it is determined to be at the sufficient level; When the available tactical margin is less than the first preset threshold but greater than or equal to the second preset threshold, it is determined to be at the tension level; When the available tactical margin is less than the second preset threshold, it is determined to be at the critical level.

9. The dynamic sampling and control method based on water quality detection linkage feedback according to claim 8, characterized in that, The first and second control strategies are weighted and fused to generate a third control strategy, including: Calculate the dynamic fusion weight coefficient based on the available tactical margin; The third control strategy is generated by weighting and fusing the first control strategy and the second control strategy based on the dynamic fusion weight coefficient.

10. A dynamic sampling and control system based on water quality detection and feedback, characterized in that, The step of implementing the dynamic sampling and control method based on water quality detection linkage feedback as described in any one of claims 1 to 9, wherein the dynamic sampling and control system based on water quality detection linkage feedback comprises: The data acquisition module is used to acquire real-time water quality detection data, remaining power, and estimated battery life for each sampling point in the current sampling round. The strategy generation module is used to input real-time water quality monitoring data into the look-ahead pre-adaptation loop and the physical constraint emergency loop to generate the first control strategy and the second control strategy; The level calculation module is used to calculate the current tactical freedom level based on the remaining power and estimated flight time, according to the preset energy level rules. The first distribution strategy determination module is used to use the second control strategy as the final distribution strategy if the tactical degree of freedom level is sufficient. The second distribution strategy determination module is used to weight and fuse the first control strategy and the second control strategy to generate a third control strategy as the final distribution strategy if the tactical degree of freedom level is tense. The third distribution strategy determination module is used to control the UAV to perform a new round of dynamic sampling if the tactical degree of freedom level is critical, using the first control strategy as the final distribution strategy.

Images