A fish-light complementary unmanned aerial vehicle cooperative intelligent feeding system

Abstract

The application relates to the technical field of intelligent feeding of aquaculture, and discloses a fish-light complementary unmanned aerial vehicle cooperative intelligent feeding system. The system divides a pond into sub-regions for feeding, sets a feeding model, and outputs feeding parameters corresponding to each sub-region based on the characteristics of each sub-region, the environment of the breeding area, and the characteristics of the breeding objects, so that the optimal feeding path is planned according to the feeding parameters and the real-time state of the current unmanned aerial vehicle group, the intelligent system of the multiple unmanned aerial vehicles is intelligently scheduled, the breeding environment data is real-timely linked, and the automatic optimization feeding strategy is realized, the feeding efficiency and precision are improved, the labor cost is reduced, and the breeding demand of the fish-light complementary scene is adapted.

Description

Technical Field

[0001] This invention relates to the field of intelligent feeding technology in aquaculture, and in particular to an intelligent feeding system for aquaculture-solar hybrid drone collaboration. Background Technology

[0002] In existing aquaculture, intelligent equipment feeding technology has been gradually used to replace simple manual feeding. The most common feeding method is a combination of timed and quantitative feeding by a single drone and feeding by a fixed manual device. However, this feeding method still has the following drawbacks: First, the coverage area of ​​a single drone is limited, and when multiple drones are operating, there is a lack of coordinated scheduling, which easily leads to repeated feeding or missed feeding; second, it cannot be linked with real-time environmental data such as light and water quality in fish-solar complementary ponds, resulting in inaccurate feeding timing and quantity; third, the high degree of human intervention results in low feeding efficiency and is greatly affected by weather and other environmental factors. Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide a collaborative intelligent feeding system for fishery-solar complementary drones, in order to solve the technical problems of poor coordination and insufficient adaptation to the fishery-solar complementary environment of existing drone feeding systems, which result in low feeding accuracy, low feeding efficiency, and waste of feeding materials.

[0004] This invention discloses a collaborative intelligent feeding system for fishery-solar hybrid drones, which includes a data acquisition module, a data transmission module, a collaborative control center, and a drone swarm; wherein,

[0005] The data acquisition module is used to collect real-time data on the aquaculture environment and the aquaculture objects, serving as the primary data to be collected.

[0006] The data transmission module is used to upload the first collected data to the collaborative control center;

[0007] The collaborative control center also includes:

[0008] The sub-unit division is used to divide the pond into multiple dynamic sub-regions based on the first collected data, historical aquaculture environment and aquaculture object collected data, and historical feeding records;

[0009] The extraction sub-unit is used to extract the first feature of each sub-region based on the dynamic sub-region division and the first acquired data;

[0010] The feeding parameter determination subunit is used to input the first feature into the feeding model and determine the feeding parameters for each sub-region through the feeding model;

[0011] The planning sub-unit is used to generate feeding tasks and path planning instructions with different priorities based on the characteristics of each sub-region, the corresponding feeding parameters, and the current real-time status of the UAV.

[0012] The drone swarm includes a feeding execution unit, which receives and executes the feeding tasks and path planning instructions from the collaborative control center, and provides real-time feedback on the operation status.

[0013] Furthermore, the process of the sub-region division operation performed by the sub-unit specifically includes:

[0014] Based on the first collected data, historical aquaculture environment and aquaculture object data, and historical feeding records, a time-series dynamic feature is constructed; the time-series dynamic feature includes a time-series heat map of the distribution of aquaculture objects, periodic statistics of light intensity and dissolved oxygen value, and the correspondence between historical feeding parameters and the aggregation response of aquaculture objects;

[0015] The pond region is divided using a clustering algorithm based on the aforementioned temporal dynamic characteristics to determine the boundaries of the sub-regions.

[0016] Furthermore, the process of performing a clustering algorithm to divide the pond region based on the temporal dynamic features and determining the boundaries of the sub-regions includes:

[0017] A multidimensional feature vector space is constructed based on the aforementioned time-series dynamic features; the multidimensional feature vector includes the distribution density index of the cultured objects, the light intensity fluctuation index, the dissolved oxygen value change index, and the historical feeding response index;

[0018] Clustering operations are performed in the feature vector space to form a preliminary set of sub-regions;

[0019] During the clustering iterative update process, constraints on drone accessibility, energy consumption cost, and environmental risk factors are introduced. When the sub-region boundary violates the constraints, the corresponding cluster center position is adjusted and / or adjacent sub-regions are merged. Under the premise of satisfying the constraints, the dynamically adjusted final sub-region boundary is output.

[0020] Furthermore, the process of the extraction subunit performing the first feature extraction operation specifically includes:

[0021] Determine the historical region-related data for each sub-region, and extract the first feature based on the first collected data and historical region-related data for each sub-region;

[0022] The first feature includes regional attribute features, environmental parameter statistical features, environmental parameter trend features, spatial uniformity features, aquaculture object features, historical feeding response features, task feasibility and operation and maintenance cost features, and aquaculture-solar complementarity features; wherein, the spatial uniformity feature represents the degree of consistency in the distribution of each monitoring parameter within the sub-region.

[0023] Furthermore, the feeding model includes a feature fusion layer, a hierarchical constraint layer, and a two-stage output layer; wherein,

[0024] The feature fusion layer is used to perform channelization processing on each feature in the first feature, and fuse the processing results into a regional feeding demand vector.

[0025] The hierarchical constraint layer is used to perform threshold truncation and time window masking on the regional feeding demand vector based on environmental parameter thresholds, time window constraints, and stability constraints to obtain the first regional feeding demand vector.

[0026] Furthermore, when the dual-stage output layer receives the first region feeding demand vector processed by the hierarchical constraint layer, it generates local feeding demand candidates for each dynamic sub-region based on the first region feeding demand vector and the region attribute features in the first stage.

[0027] In the second stage, under the constraint of the sustainability of UAV mission execution, the local feeding demand candidates are globally coordinated and allocated, and the final feeding parameters corresponding to each sub-region are output.

[0028] Furthermore, the process of generating feeding task priorities for the planning subunit specifically includes:

[0029] Based on the regional attribute characteristics, feeding parameters, characteristics of aquaculture objects, and stability characteristics of environmental parameters of each dynamic sub-region, a set of task evaluation indicators is constructed.

[0030] Based on the set of task evaluation indicators, a task priority value is generated for each sub-region. During the task priority value generation process, threshold constraints are set on the characteristics of the aquaculture object and the stability characteristics of environmental parameters. When the threshold constraints are violated, the task priority value of the corresponding sub-region is reduced.

[0031] A task priority queue is established based on the task priority value of each sub-region, and a sorted list of feeding tasks is output for the drone cluster to execute.

[0032] Furthermore, the path planning process specifically includes:

[0033] Based on the task priority queue and the boundary range and geometric complexity of each dynamic sub-region, a coverage constraint graph is generated.

[0034] The path generation operation is performed based on the coverage constraint graph, and the reachability and energy consumption cost of the path are verified based on the real-time status information of the UAV.

[0035] After the path is generated, overlap detection and time conflict detection are performed on different UAV paths. When a conflict is detected, the path adjustment operation is performed until the mission path of each UAV meets the coverage constraint, energy consumption constraint and conflict resolution condition.

[0036] Output the adjusted path planning instructions.

[0037] Furthermore, the division sub-unit is also used to set a buffer zone at the boundary of each divided dynamic sub-region; the width of the buffer zone is set based on the statistical results of the historical activity range of the aquaculture objects.

[0038] During the task execution phase, when it is detected that the aquaculture object is moving across the boundary of the buffer zone, the feeding parameters are flexibly adjusted between adjacent sub-regions according to the crossing ratio of the aquaculture object through the collaborative control center.

[0039] Furthermore, during the task execution phase, the data acquisition module is also used to collect surface distribution data and underwater aggregation distribution data of the aquaculture objects;

[0040] The collaborative control center then performs real-time corrections on the location of the aquaculture objects within the sub-region based on the surface distribution data and underwater aggregation distribution data of the aquaculture objects.

[0041] Based on the real-time correction of the location of the aquaculture object, fine-tuning is performed on the current feeding path and feeding release location of the drone.

[0042] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0043] This invention dynamically divides a pond into feeding sub-regions and sets up a feeding model. Based on the characteristics of each sub-region, the aquaculture environment, and the characteristics of the aquaculture species, the feeding model outputs feeding parameters for each sub-region. Based on the feeding parameters and the real-time status of the drone swarm, the optimal feeding path is planned. This enables an intelligent system that achieves intelligent scheduling of multiple drones, real-time linkage with aquaculture environment data, and automatic optimization of feeding strategies. This improves feeding efficiency and accuracy, reduces labor costs, and adapts to the aquaculture needs of fishery-solar complementary scenarios. Attached Figure Description

[0044] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and constitute a part of this application, do not limit the scope of the invention. In the drawings:

[0045] Figure 1 This is a schematic diagram of a fishery-solar hybrid drone-assisted intelligent feeding system disclosed in Embodiment 1 of the present invention. Detailed Implementation

[0046] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Example

[0047] This invention discloses a collaborative intelligent feeding system for fishery-solar hybrid unmanned aerial vehicles (UAVs). Please refer to [link / reference]. Figure 1 , Figure 1 This is a schematic diagram of a collaborative intelligent feeding system for fisheries and solar power using unmanned aerial vehicles (UAVs) according to an embodiment of the present invention. The system includes a data acquisition module, a data transmission module, a collaborative control center, and a UAV cluster; wherein,

[0048] The data acquisition module is used to collect real-time data on the aquaculture environment and the aquaculture objects, serving as the primary data to be collected.

[0049] The data transmission module is used to upload the first collected data to the collaborative control center;

[0050] The collaborative control center also includes:

[0051] The sub-unit division is used to divide the pond into multiple dynamic sub-regions based on the first collected data, historical aquaculture environment and aquaculture object collected data, and historical feeding records;

[0052] The extraction sub-unit is used to extract the first feature of each sub-region based on the dynamic sub-region division and the first acquired data;

[0053] The feeding parameter determination subunit is used to input the first feature into the feeding model and determine the feeding parameters for each sub-region through the feeding model;

[0054] The planning sub-unit is used to generate feeding tasks and path planning instructions with different priorities based on the characteristics of each sub-region, the corresponding feeding parameters, and the current real-time status of the UAV.

[0055] The drone swarm includes a feeding execution unit, which receives and executes the feeding tasks and path planning instructions from the collaborative control center, and provides real-time feedback on the operation status.

[0056] Furthermore, the process of the sub-region division operation performed by the sub-unit specifically includes:

[0057] Based on the first collected data, historical aquaculture environment and aquaculture object data, and historical feeding records, a time-series dynamic feature is constructed; the time-series dynamic feature includes a time-series heat map of the distribution of aquaculture objects, periodic statistics of light intensity and dissolved oxygen value, and the correspondence between historical feeding parameters and the aggregation response of aquaculture objects;

[0058] The pond region is divided using a clustering algorithm based on the aforementioned temporal dynamic characteristics to determine the boundaries of the sub-regions.

[0059] Furthermore, the process of determining the boundaries of sub-regions by performing a clustering algorithm to divide the pond region based on the aforementioned temporal dynamic features includes:

[0060] A multidimensional feature vector space is constructed based on the aforementioned time-series dynamic features; the multidimensional feature vector includes the distribution density index of the cultured objects, the light intensity fluctuation index, the dissolved oxygen value change index, and the historical feeding response index;

[0061] Clustering operations are performed in the feature vector space to form a preliminary set of sub-regions;

[0062] During the clustering iterative update process, constraints on drone accessibility, energy consumption cost, and environmental risk factors are introduced. When the sub-region boundary violates the constraints, the corresponding cluster center position is adjusted and / or adjacent sub-regions are merged. Under the premise of satisfying the constraints, the dynamically adjusted final sub-region boundary is output.

[0063] In this embodiment of the invention, the purpose of setting up sub-units is to divide the entire fishery-solar hybrid pond into multiple dynamic sub-regions, so as to achieve higher accuracy and differentiated scheduling in subsequent feeding model inference and UAV task allocation.

[0064] Specifically, firstly, multi-source data is collected and integrated to form input information for segmentation. This input includes, but is not limited to, parameters that dynamically change over time, such as dissolved oxygen, pH, water temperature, and light intensity. For example, when the cultured organism is fish, the data can include indicators reflecting the behavior of the cultured organisms, such as fish population density, activity frequency, and location heatmaps. Historical feeding records include past feeding times, amounts, and corresponding fish aggregation responses during feeding, aiming to reveal the causal relationship between feeding and fish feeding behavior. After fusion analysis, the above multi-source information is used to construct time-series dynamic features.

[0065] After obtaining the time-series dynamic features, a multi-dimensional feature vector space is constructed based on these features. The dimensions of the feature vector space include, but are not limited to, the distribution density index of the cultured organisms, the light intensity fluctuation index, the dissolved oxygen value change index, and the historical feeding response index.

[0066] To ensure comparability across different dimensions, each indicator is first normalized, and local extrema are selected in the feature vector space as initial cluster centers. Then, the regions are divided based on a clustering algorithm. Specifically, by calculating the similarity metric between sampling points and cluster centers, each sampling point is assigned to a corresponding cluster center, thus gradually forming a preliminary set of sub-regions. The similarity metric considers not only the Euclidean distance of the distribution density of farmed objects but also the difference metric and cosine similarity between environmental parameters, enabling a more accurate characterization of the coupling relationship between environmental factors and farming behavior. As the iteration progresses, the cluster centers are continuously updated, and the boundaries of the sub-regions gradually converge, thus achieving the initial division of the ponds.

[0067] To avoid unexecutable or unreasonable partitioning results during the clustering iteration process, this invention introduces multiple constraints to dynamically correct the boundaries. Specifically, a correction mechanism is triggered when the boundary of a sub-region violates preset constraints. These preset constraints include: 1) UAV reachability constraints: By calculating the UAV's flight radius and operational trajectory, it ensures that all sub-regions are within the UAV's reachable range. If an inaccessible region is found, it is corrected by adjusting the cluster center position or boundary. 2) Energy consumption cost constraints: By combining the UAV's energy consumption model, the cost of the flight path to each candidate sub-region is calculated. If a boundary partitioning causes energy consumption to exceed a preset threshold, the cluster center is automatically adjusted, and / or high-cost sub-regions are merged with neighboring low-cost regions, thereby ensuring that overall energy consumption remains within a reasonable range. 3) Environmental risk factor constraints: When a local area experiences high-risk situations such as extreme sunlight, drastic water quality fluctuations, or sudden drops in dissolved oxygen, the independence of that region is reduced during the iteration process, shrinking its boundary towards low-risk regions to avoid generating feeding tasks in such areas. By combining the above constraints with the clustering iteration process, dynamic correction of the sub-region boundaries can be achieved, so that the final output partitioning result can simultaneously satisfy environmental rationality and UAV operation feasibility.

[0068] After multiple iterations and constraint corrections, a dynamically adjusted sub-region boundary is obtained. This dynamic sub-region not only reflects the behavioral characteristics of the farmed organisms and the state of the farming environment, but also ensures the operability and energy efficiency of the UAV's mission execution, thus providing an accurate and robust regional foundation for subsequent feeding parameter determination and path planning.

[0069] Furthermore, the process of the extraction subunit performing the first feature extraction operation specifically includes:

[0070] Determine the historical region-related data for each sub-region, and extract the first feature based on the first collected data and historical region-related data for each sub-region;

[0071] The first feature includes regional attribute features, environmental parameter statistical features, environmental parameter trend features, spatial uniformity features, aquaculture object features, historical feeding response features, task feasibility and operation and maintenance cost features, and aquaculture-solar complementarity features; wherein, the spatial uniformity feature represents the degree of consistency in the distribution of each monitoring parameter within the sub-region.

[0072] Specifically, in this embodiment of the invention, the purpose of setting up the extraction sub-unit is to generate a first feature for each sub-region to be used by the model after the dynamic sub-region division is completed. To ensure the stability and comparability of the features, the extraction sub-unit first performs unified timestamp alignment and spatial attribution mapping on the time axis between the real-time collected first data and the historical regional related data. Spatial attribution mapping refers to mapping historical data back to the currently defined sub-region range according to geographic coordinates or raster indexes, so that historical data corresponds one-to-one with the current sub-region. After data alignment is completed, missing data completion and outlier identification are performed. Priority is given to using adjacent time-preservation or interpolation strategies for completion, and the confidence level generated during the completion process is labeled. The confidence level will be output along with the features for subsequent model use.

[0073] The historical data related to the region in this embodiment includes not only the historical trajectory of the first collected data in the past period, but also multi-source historical data directly related to aquaculture and operations, including but not limited to historical environmental parameter time series, historical records of aquaculture objects such as stocking time and density, fish species ratio, body length and weight distribution, seasonal activity patterns and feeding periods, historical feeding records such as feeding time, feeding amount, pellet size and feeding response and uneaten feed records, historical equipment and operation and maintenance records such as aeration, oxygenation, inspection, no-fly or restricted flight periods, historical spatial bottom sediment and morphological information such as pond bottom topography, water depth distribution, obstacles and shoreline morphology, and aquaculture-solar complementary operation data such as photovoltaic array arrangement, historical power generation curves, shading time series and maintenance shutdown periods. To facilitate trend and periodic analysis, historical data is summarized into three scales: short window (last 1-3 days), medium window (last 1-2 weeks), and long window (last 1-3 months), so that subsequent features can stably represent the state of the sub-region at different time scales.

[0074] Based on the above data preparation, sub-units are extracted to generate the first feature. Among these, regional attribute features reflect the geometric and static environmental constraints of the sub-region, including but not limited to sub-region area, statistical distribution of water depth, boundary shape complexity, and shoreline tortuosity. Complexity can be quantified by the change in turning density and concavity / convexity per unit boundary length, thus characterizing the curve cost basis of path planning. Environmental parameter statistical features describe the current state of environmental monitoring quantities within the sub-region, expressed as the average, variance, extreme values, and quantiles within the same timeframe or short window to reduce the impact of instantaneous noise. These features are directly calculated from the real-time first-collection data and compared with historical reference intervals to facilitate subsequent model identification of abnormal deviations. Environmental parameter trend features characterize the direction and speed of environmental change, derived from first-order rate of change, second-order rate of change, and multi-scale slip differential constructed from historical regional data. Furthermore, for parameters with distinct diurnal and weekly rhythms, such as illumination and dissolved oxygen, seasonal decomposition and periodic amplitude estimation are performed to obtain trend strength and periodicity indicators. Spatial uniformity features are used to describe the spatial consistency of monitoring parameters within a sub-region. Specifically, sensor point values ​​are first interpolated into a regular grid within the sub-region, and then the dispersion coefficient, maximum spatial gradient amplitude, and spatial correlation coefficient of the grid values ​​are calculated. When spatial uniformity is poor, it means that there are significant environmental differences within the same sub-region, and subsequent feeding needs to be matched with coverage constraints and path refinement.

[0075] The characteristics of farmed animals are used to characterize the animals themselves and their behavioral states. The data comes from visual and acoustic perception as well as historical farming records. These characteristics include at least current and recent population distribution density, population activity, body length and weight range ratios, and species composition. Historical feeding time preferences and diurnal activity differences are also incorporated to express the temporal tendency of feeding intentions.

[0076] Historical feeding response characteristics are mainly used to reflect the empirical relationship between feeding and ingestion. By fitting historical feeding amounts with ingestion response records, parameters such as saturation threshold, response slope, and lag time are obtained. At the same time, the threshold range for the appearance of uneaten food and the minimum safe interval duration are recorded, thus providing a basis for subsequent feeding upper limits and time interval constraints.

[0077] The mission feasibility and operation and maintenance cost features are used to characterize the accessibility and cost of sub-regions at the execution level. Combining take-off and landing point locations, no-fly / restricted-fly zones, water surface obstacles, historical inspection paths and charging turnover records, the arrival cost, path risk level and estimated energy consumption are calculated, and an execution feasibility score is output.

[0078] The solar-aquaculture complementary feature is used to take into account the temporal impact of photovoltaics on aquaculture and operations. Based on historical and predicted power generation curves and shading ratio sequences, features such as average power, fluctuation range and shading period proportion are extracted. When shading and high irradiance alternate significantly, this feature can help determine the appropriate feeding time window and drone operation window.

[0079] To ensure that the features can be stably utilized by the subsequent feeding model, this embodiment performs uniform dimension elimination and interval normalization on each feature, and retains confidence flags for missing data completion and anomaly removal in the feature vector. It is important to emphasize that the generation of environmental parameter trend features must rely on historical environmental parameter time series from relevant historical regional data; otherwise, stable first- and second-order rates of change and periodic indicators cannot be formed. Similarly, historical feeding response features must rely on historical feeding records and feeding feedback; otherwise, key parameters such as saturation and lag cannot be established. These historical and real-time terms together constitute the explanatory input for the feeding model.

[0080] In the feature output stage, the extraction sub-unit concatenates various features in a fixed order into a first feature vector of fixed length, and outputs it along with the confidence flag for the feature fusion layer of the feeding model to receive and process. Through the above operations, the first feature not only fully reflects the geometric and environmental state of the sub-region, the aquaculture species and feeding patterns, the feasibility of operation and maintenance, and the temporal impact of fishery-solar complementarity, but also has traceability and stability in both time and space dimensions, thus providing a sufficient and reproducible data foundation for subsequent determination of feeding parameters and coordinated scheduling.

[0081] Furthermore, the feeding model includes a feature fusion layer, a hierarchical constraint layer, and a two-stage output layer; wherein,

[0082] The feature fusion layer is used to perform channelization processing on each feature in the first feature, and fuse the processing results into a regional feeding demand vector.

[0083] The hierarchical constraint layer is used to perform threshold truncation and time window masking on the regional feeding demand vector based on environmental parameter thresholds, time window constraints, and stability constraints to obtain the first regional feeding demand vector.

[0084] Furthermore, when the dual-stage output layer receives the first region feeding demand vector processed by the hierarchical constraint layer, it generates local feeding demand candidates for each dynamic sub-region based on the first region feeding demand vector and the region attribute features in the first stage.

[0085] In the second stage, under the constraint of the sustainability of UAV mission execution, the local feeding demand candidates are globally coordinated and allocated, and the final feeding parameters corresponding to each sub-region are output.

[0086] Specifically, in this embodiment of the invention, the feeding model is configured to consist of a feature fusion layer, a hierarchical constraint layer, and a two-stage output layer in sequence. The purpose is to progressively transform the data-driven feeding demand representation into executable feeding parameters at the dynamic sub-region level, and to coordinate with the UAV's execution capabilities globally to form a closed-loop control command. The feeding parameters of this invention include, but are not limited to, feeding amount, feeding timing, and feeding location.

[0087] In the feature fusion layer, channelization refers to mapping each feature in the first feature set into an independent numerical channel. Before entering the fusion operation, each channel undergoes unified time reference alignment, missing data labeling, and dimensionless processing. Specifically, to prevent trend information from being overwhelmed by instantaneous noise, the environmental parameter trend channel uses a time series segment composed of a fixed window's first / second-order rate of change and multi-scale slip as the input segment.

[0088] The channelized multi-source vectors are appended with periodic positional codes for intraday time and weekly location to explicitly express diurnal and weekly rhythms. The fusion operation adopts a local-to-global approach, first performing small-scale linear projection and nonlinear interaction within a channel to compress redundancy and extract key patterns. Then, interactive mapping is performed between channels to capture cross-channel coupling relationships, such as combinations of low dissolved oxygen rising trends, high fish activity, and weakening light intensity. After the above processing, a regional feeding demand vector is output. This vector is a fixed-length real number representation, organized according to semantic sub-segments such as basic demand intensity, temporal elasticity, feeding saturation risk, spatial distribution sensitivity, and execution feasibility confidence, to summarize the feeding tendency of the sub-region at the current time and in the near-future window. It should be noted that the regional feeding demand vector is only a demand representation within the feasible solution space, not a direct feeding parameter. It will be constrained, pruned, and projected onto executable discrete time periods and magnitudes in subsequent layers.

[0089] The hierarchical constraint layer is used to restrict the above-mentioned feeding demand representation within the feasible domain jointly defined by biosafety and operational safety. Environmental parameter thresholds represent biological or managerial hard limits, such as dissolved oxygen below the safety lower limit, pH exceeding the allowable range, or water temperature within the feeding inhibition range. Once a threshold is triggered, the corresponding demand component is immediately truncated to zero or limited to the minimum safe value, ensuring that feeding is not triggered under unsuitable conditions. Time window constraints represent when execution is permitted or more suitable. These constraints are derived from periods of shading during fish-solar complementary farming, equipment maintenance and no-fly periods, historical peak feeding periods, and management-preset windows. In implementation, a masking function is applied to the time segments of the demand vector to zero out infeasible windows and suppress low-optimal windows. Stability constraints suppress drastic changes in feeding recommendations between adjacent time periods, avoiding frequent and repeated switching. In practice, differential and second-order rate-of-change limits are applied to the recommendation strength of adjacent time slices, and a hysteresis mechanism is introduced. When a recommendation fluctuates near a threshold, several consecutive time slices must be satisfied before cross-threshold switching is allowed. After processing by the hierarchical constraint layer, the first region feeding demand vector is obtained. Compared with the original demand, this vector has eliminated the components of environmental insecurity and time infeasibility, and forms a smooth and executable demand profile on the time axis.

[0090] After receiving the feeding demand vector for the first region, the dual-stage output layer first generates local feeding demand candidates for each dynamic sub-region in the first stage. This stage combines the feeding demand vector of the first region with regional attribute features, projecting the continuous demand intensity onto discrete candidate time periods and candidate magnitudes. Specifically, several candidate time slices corresponding to the demand peak are selected on a preset discrete time grid, and a reasonable magnitude range for a single feeding is estimated based on the region area, water depth statistical distribution, and aquaculture population density. Simultaneously, the saturation threshold and minimum safe interval from historical feeding responses are used as upper limits and interval constraints, forming candidate pairs of time and magnitude for each candidate time slice. To avoid feeding offset caused by uneven spatial distribution, the candidate pairs also include a coverage sensitivity index, which indicates whether subsequent paths within the sub-region need to be densified to achieve uniform delivery. This stage is solved independently only within the sub-region and does not involve resource coordination between UAVs.

[0091] In the second phase, under the constraints of drone mission sustainability, candidate pairs in all sub-regions are globally coordinated, and the final delivery parameters are output. Drone mission sustainability constraints characterize the execution boundaries at both the individual drone and cluster levels, including at least closed-loop constraints on current and predicted power and payload, availability and turnaround time of takeoff and landing points and charging stations, estimated flight energy consumption and operation duration, and accessibility and trajectory conflict limitations. The global coordination process treats candidate pairs in each sub-region as job units to be assigned, and performs joint solutions on the time axis and spatial topology, ensuring that each drone mission chain simultaneously satisfies the combined constraints of power and payload, operation path coverage, and time window. When constraints cannot be met, candidate pairs are preferentially reduced in magnitude or postponed to a suboptimal window, triggering cross-drone reassignment or split execution if necessary. After this phase, the final delivery parameters for each sub-region are obtained, i.e., the specific delivery timing and corresponding delivery quantity, and are distributed along with the corresponding drone mission chains and operation paths.

[0092] The reason why the hierarchical structure of the above-mentioned feeding model can derive reliable feeding parameters from multi-source features lies in the decoupling of abstract demand representation, feasible domain pruning, and resource coordination into three stages. The feature fusion layer, under the premise of spatiotemporal consistency, performs intra-channel purification and inter-channel interaction on multiple types of features, ensuring that the generated regional feeding demand vector carries essential information about needs and when is more suitable, without being mixed with execution-side constraints. The hierarchical constraint layer restricts demands within the biological-operation dual safety boundary, preventing unsuitable periods or unstable suggestions from entering the scheduling. The two-stage output layer first converts continuous demands into discrete candidates within the region, and then uses sustainability constraints to filter and rearrange them at the cluster level, ultimately obtaining time-scale pairs that satisfy executability and resource boundaries. This allows for the stable generation of final feeding parameters for UAV collaborative feeding.

[0093] In this embodiment of the invention, the feeding model is modeled in a differentiable manner by combining multi-source historical data and constraints, and is updated adaptively online. Specifically, training samples are constructed using collected historical environmental data, data related to the farmed organisms, and historical feeding records, and organized using sub-regions and time slices to ensure that the training samples cover the differences in different regions and time periods. In the early stage of training, the model is pre-trained under self-supervised supervision, using mask reconstruction and cross-modal contrastive learning methods, enabling the model to automatically capture the inherent patterns of environmental changes and farming behavior.

[0094] The supervised training phase then begins, using the feeding response, uneaten feed amount, and dissolved oxygen disturbance amplitude of the cultured organisms after feeding as supervisory labels to guide the model in learning the correspondence between regional feeding demands and actual responses. During this process, the model's hierarchical constraints are embedded in the training in a differentiable manner, allowing the constraints to participate in backpropagation and guiding the model to gradually converge to a solution space that conforms to environmental safety thresholds and time-related patterns. To address the drift issues caused by seasonal variations and the growth cycle of the cultured organisms, this invention deploys online fine-tuning based on a sliding time window. When overfeeding or underfeeding rates deviate from preset thresholds, some model parameters are updated in small steps to maintain the adaptability and stability of the predictions.

[0095] In practical implementation, the functional layers of the feeding model can be based on various deep neural network architectures to adapt to different feature types and computational requirements. For example, when the feature fusion layer performs channelization processing on spatial features, it can use a Convolutional Neural Network (CNN) structure to extract local statistical patterns and spatial correlations within a region through convolutional kernels. When modeling temporal dynamic features, it can use structures such as Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), or Gated Recurrent Unit (GRU) to capture the nonlinear laws governing the changes in environmental parameters and the behavior of the aquaculture population over time. When it is necessary to consider both cross-regional feature interactions and global dependencies, a Transformer structure based on an attention mechanism can be used to achieve global interaction and weighted combination between features through a multi-head attention mechanism, thereby improving the feeding model's ability to express complex scenarios. For the final mapping and nonlinear combination of feeding parameters, a multi-layer perceptron (MLP) can be used as the output layer to achieve a joint mapping of local feeding requirements and global scheduling constraints. It should be noted that the choice of the above network architecture is not the only limitation; this invention is not limited to a specific neural network form. Any network structure capable of fusing and modeling input features and outputting feeding parameters that meet the requirements of this invention falls within the scope of protection of this invention.

[0096] Furthermore, the process of generating feeding task priorities for planning sub-units specifically includes:

[0097] Based on the regional attribute characteristics, feeding parameters, characteristics of aquaculture objects, and stability characteristics of environmental parameters of each dynamic sub-region, a set of task evaluation indicators is constructed.

[0098] Based on the set of task evaluation indicators, a task priority value is generated for each sub-region. During the task priority value generation process, threshold constraints are set on the characteristics of the aquaculture object and the stability characteristics of environmental parameters. When the threshold constraints are violated, the task priority value of the corresponding sub-region is reduced.

[0099] A task priority queue is established based on the task priority value of each sub-region, and a sorted list of feeding tasks is output for the drone cluster to execute.

[0100] Furthermore, the path planning process specifically includes:

[0101] Based on the task priority queue and the boundary range and geometric complexity of each dynamic sub-region, a coverage constraint graph is generated.

[0102] The path generation operation is performed based on the coverage constraint graph, and the reachability and energy consumption cost of the path are verified based on the real-time status information of the UAV.

[0103] After the path is generated, overlap detection and time conflict detection are performed on different UAV paths. When a conflict is detected, the path adjustment operation is performed until the mission path of each UAV meets the coverage constraint, energy consumption constraint and conflict resolution condition.

[0104] Output the adjusted path planning instructions.

[0105] Specifically, in this embodiment of the invention, it is preferable to construct a task evaluation index set using features such as regional attribute characteristics, aquaculture object characteristics, and environmental parameter stability characteristics, along with feeding parameters. This is because regional attribute characteristics characterize the objective conditions of a sub-region, such as area, water depth, and geometric complexity, which can affect the coverage and path length of the drone feeding operation. Aquaculture object characteristics directly reflect the necessity and priority of feeding. Environmental parameter stability characteristics describe the short-term fluctuations in parameters such as dissolved oxygen, pH, and water temperature. When the environment is unstable, even if the aquaculture object population density is high, immediate feeding is not advisable to prevent stress reactions. By incorporating the above characteristics along with the corresponding sub-region feeding parameters into the index set, a comprehensive evaluation that considers environmental conditions, aquaculture needs, and feeding resources can be achieved.

[0106] During the generation of task priority values, the planning sub-unit converts each indicator into a comparable score according to preset measurement standards and combines them to form the task priority value for each sub-region. In this process, threshold constraints are set based on the characteristics of the cultured organisms and the stability of environmental parameters. For example, when the fish density falls below the minimum feeding threshold, or dissolved oxygen fluctuations exceed the safe range, the task priority value for that sub-region is automatically reduced. This threshold constraint mechanism prevents the erroneous increase in feeding priority for certain areas due to abnormal environments or unsuitable conditions, thereby improving the rationality of task scheduling.

[0107] During the path planning phase, the planning sub-unit first generates a coverage constraint graph based on the task priority queue and the boundary range and geometric complexity of each dynamic sub-region. Specifically, the coverage constraint graph is constructed by discretizing the sub-region boundaries into network polygonal units and marking each unit with a task requirement weight. This task requirement weight is positively correlated with the task priority value of the sub-region, ensuring that high-priority areas are covered first during path generation. In this constraint graph, different units are connected by edges, forming a weighted graph structure, where the edge weight represents the spatial cost of the UAV moving between two units. By performing shortest path or heuristic search on this graph model, a set of candidate paths covering all target units is obtained.

[0108] Simultaneously with path generation, accessibility verification and energy cost assessment are performed using the UAV's real-time status information. Accessibility verification includes two aspects: first, whether the path length exceeds the flight radius supported by the UAV's current remaining battery power; and second, whether any path segments cross areas with geographical obstacles or beyond visual line of sight. Energy cost assessment uses the UAV's battery level, current payload weight, and path length as inputs, calculating the predicted energy consumption for a single path based on an empirical energy consumption function, and comparing it with the UAV's available energy limit. If a candidate path does not meet the accessibility or energy consumption requirements, the planning subunit adjusts the path, including shortening the path range, segmenting the task, or reassigning the task to other UAVs, to ensure the final path is feasible.

[0109] After path generation is complete, the planning sub-unit further performs overlap detection and temporal conflict detection on multiple UAV paths. Overlap detection is performed by calculating the spatial intersection and temporal overlap of different UAV paths. If two paths are found to spatially overlap in the same area and simultaneously cover the same area, it is considered a potential risk of duplicate deployment. Temporal conflict detection is based on the path timeline, checking whether different UAVs will enter adjacent or identical sub-areas at the same time. If such a situation exists, it is considered a conflict.

[0110] When a conflict is detected, the planning subunit performs corrections based on a path adjustment algorithm. Specifically, the corrections resolve time conflicts by delaying the start time of some tasks, replanning local paths in the coverage constraint graph to eliminate spatial overlap, and dynamically splitting tasks to transfer tasks from some sub-regions to other paths or drones. This correction process is iterative; each adjustment requires re-verification of coverage, energy consumption, and conflict status until all drone task paths meet the coverage constraints, energy consumption constraints, and conflict resolution conditions. Finally, the planning subunit outputs optimized path planning instructions, ensuring that the drone swarm can efficiently and collaboratively complete the delivery task under limited energy consumption and dynamic environmental conditions.

[0111] Through the above operations, the planning sub-unit can organically combine task priority generation with path planning. The former ensures the rationality and urgency of task allocation, while the latter guarantees the feasibility and safety of the UAV at the execution level, thereby achieving efficient and collaborative operation of the entire system in a dynamic environment.

[0112] Furthermore, the division sub-unit is also used to set a buffer zone at the boundary of each divided dynamic sub-region; the width of the buffer zone is set based on the statistical results of the historical activity range of the aquaculture objects.

[0113] During the task execution phase, when it is detected that the aquaculture object is moving across the boundary of the buffer zone, the feeding parameters are flexibly adjusted between adjacent sub-regions according to the crossing ratio of the aquaculture object through the collaborative control center.

[0114] In this embodiment of the invention, the sub-unit division not only divides the pond into multiple dynamic sub-regions based on temporal dynamic characteristics, but also further sets buffer zones at the boundaries of each dynamic sub-region to accommodate the actual swimming behavior of the farmed organisms. Specifically, the width of the buffer zone is not fixed, but dynamically determined based on the statistical results of the historical activity range of the farmed organisms. For example, by collecting long-term data on the location distribution of fish schools, the activity frequency distribution curves of fish schools in the vicinity of each boundary are obtained, and the width of the buffer zone is set according to the 95% coverage of the distribution, thereby ensuring that the buffer zone can cover most common cross-boundary swimming behaviors.

[0115] During the mission execution phase, the collaborative control center continuously receives real-time location data of the aquaculture species population from visual monitoring and / or underwater acoustic detection equipment. When aquaculture species is detected moving across the buffer zone boundary, the crossing ratio is calculated, which is the ratio of the number of aquaculture species entering the adjacent sub-region to the total number within the buffer zone. This crossing ratio reflects the dynamic distribution trend of the aquaculture species between the two sub-regions.

[0116] Once the crossing ratio is determined, the collaborative control center triggers a flexible adjustment mechanism for feeding parameters. Specifically, if α% of the cultured organisms in a certain sub-region are detected crossing into a neighboring sub-region, the feeding requirement value of that sub-region is reduced by a ratio of 1-α, while the feeding requirement value of the neighboring sub-region is increased by a ratio of α. This adjustment process is based on real-time statistical results within the buffer zone, ensuring that the feeding amount can dynamically change with the actual location distribution of the fish population, and will not cause feeding misalignment or omissions due to the static nature of boundary demarcation.

[0117] During the flexible adjustment process, the collaborative control center will also perform secondary optimization based on the sustainability constraints of the drone's mission. If the adjacent sub-area after the replenishment exceeds the drone's maximum load capacity, the replenishment amount will be split among multiple drones for shared execution, or some replenishment tasks will be delayed to the next time window to ensure the feasibility of the overall delivery plan.

[0118] Through the aforementioned buffer zone and flexible adjustment operations, this invention achieves flexibility in sub-region division, ensuring both the spatial accuracy of region boundary division and avoiding boundary effects caused by the natural movement of aquatic organisms. This ensures that the feeding strategy can dynamically adapt to the actual distribution of aquatic organisms, thereby improving feeding efficiency and resource utilization.

[0119] Furthermore, during the task execution phase, the data acquisition module is also used to collect surface distribution data and underwater aggregation distribution data of the aquaculture objects;

[0120] The collaborative control center then performs real-time corrections on the location of the aquaculture objects within the sub-region based on the surface distribution data and underwater aggregation distribution data of the aquaculture objects.

[0121] Based on the real-time correction of the location of the aquaculture object, fine-tuning is performed on the current feeding path and feeding release location of the drone.

[0122] In this embodiment of the invention, the data acquisition module is not only used to collect conventional environmental parameters and data related to the aquaculture objects, but also further used to collect surface distribution data and underwater aggregation distribution data of the aquaculture objects during the task execution phase. Specifically, the surface distribution data mainly reflects the activity density and location distribution of the aquaculture object population at the water surface, while the underwater aggregation distribution data reflects the aggregation area and depth distribution characteristics of the underwater aquaculture object population.

[0123] After receiving the aforementioned data, the collaborative control center uses a data fusion algorithm to perform real-time correction of the locations of aquaculture objects within the sub-region. Specifically, it first aligns the spatial coordinates of surface distribution data and underwater distribution data, and then obtains the overall three-dimensional distribution of the aquaculture object population through depth-weighted matching. If a significant difference is detected between the surface and underwater distributions, the underwater clustered distribution is prioritized, with the surface distribution used as a secondary factor, to form a weighted correction result, thereby avoiding biases caused by a single data source.

[0124] After real-time correction of the location of the farmed animals, the collaborative control center fine-tunes the drone's feeding path and release location based on the corrected location. Specifically, during path generation, the drone's flight path nodes are locally adjusted to ensure that the drone's feeding release point coincides as closely as possible with the corrected location of the farmed animal population's gathering center. When the entire population submerges or rises, the feeding release altitude and timing are fine-tuned to ensure that the feed is effectively delivered to the area where the farmed animals are gathered.

[0125] Through the above operations, the present invention can realize real-time correction of the location of fish groups in sub-regions when the distribution of aquaculture objects changes dynamically with time and environment, so that the feeding path and release point of the drone are always consistent with the actual distribution of the aquaculture object group, thereby improving feeding accuracy and reducing feed waste.

[0126] As a further preferred embodiment, the collaborative control center is not only used for task allocation and path planning, but also for performing return and re-deployment operations based on feedback information from the UAV swarm. Specifically, during the task execution phase, the UAV swarm will periodically report task feedback information to the collaborative control center, including but not limited to the current remaining payload capacity, remaining power level, task completion progress, and environmental monitoring data.

[0127] When the collaborative control center detects that a drone's remaining payload is insufficient to complete unfinished feeding tasks, or that its remaining battery power is below a preset threshold, a return-to-home decision is triggered. At this point, the collaborative control center calculates the optimal return path based on the drone's current location and the geographical locations of various take-off and landing points. The path generation process considers flight distance, power consumption models, and obstacle avoidance rules simultaneously to ensure that the drone can safely return to the nearest take-off and landing point to complete resupply or charging operations before its battery is completely depleted.

[0128] Meanwhile, the collaborative control center also monitors the task execution results of sub-regions. When it detects that the actual feeding amount in a sub-region is lower than the preset target, or when airborne visual and acoustic sensors identify that the aquaculture population has not yet reached the target feeding state, the collaborative control center will trigger a supplementary feeding operation. Specifically, a supplementary feeding task is generated based on the unfinished task amount, and then available drones are rescheduled globally. During this scheduling process, the collaborative control center comprehensively considers the drone's remaining payload capacity, flight accessibility, and pond environmental safety factors. For example, if a certain area is identified as a high-risk area due to abnormal water quality or excessive sunlight, drones in that area will be restricted from performing supplementary feeding, and the task will be performed by drones in other areas or by standby drones.

[0129] After completing the aforementioned secondary scheduling, the collaborative control center outputs a list of replenishment tasks and generates new path planning and delivery instructions for each task. The new path planning must not only meet coverage and energy consumption constraints but also coordinate with the original task paths to avoid path or time conflicts introduced by the replenishment operation. The execution results of the replenishment tasks are then fed back to the collaborative control center, forming a closed-loop monitoring and scheduling process.

[0130] Through the above operations, this invention enables dynamic return and refeeding scheduling during UAV missions, ensuring the continuity and effectiveness of mission execution even in the face of unforeseen circumstances such as insufficient power, insufficient payload, or unsatisfactory feeding results. This operation significantly improves the mission completion rate of UAV swarms in solar-aquaculture complementary feeding scenarios, reducing insufficient feeding and feed waste caused by mid-mission failures or environmental anomalies.

[0131] Finally, it should be noted that the above-described embodiments include multiple parallel implementations of the present invention. Deleting or otherwise adjusting one or more of these implementations will not affect the implementation of the solution. Furthermore, the fishery-solar complementary UAV collaborative intelligent feeding system disclosed in the embodiments of the present invention is merely a preferred embodiment of the present invention and is only used to illustrate the technical solution of the present invention, not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A collaborative intelligent feeding system for fishery-solar hybrid unmanned aerial vehicles, characterized in that, The system includes a data acquisition module, a data transmission module, a collaborative control center, and a drone swarm; wherein, The data acquisition module is used to collect real-time data on the aquaculture environment and the aquaculture objects, serving as the primary data to be collected. The data transmission module is used to upload the first collected data to the collaborative control center; The collaborative control center also includes: The sub-unit division is used to divide the pond into multiple dynamic sub-regions based on the first collected data, historical aquaculture environment and aquaculture object collected data, and historical feeding records; The extraction sub-unit is used to extract the first feature of each sub-region based on the dynamic sub-region division and the first acquired data; The feeding parameter determination subunit is used to input the first feature into the feeding model and determine the feeding parameters for each sub-region through the feeding model; The planning sub-unit is used to generate feeding tasks and path planning instructions with different priorities based on the characteristics of each sub-region, the corresponding feeding parameters, and the current real-time status of the UAV. The drone swarm includes a feeding execution unit, which receives and executes the feeding tasks and path planning instructions from the collaborative control center, and provides real-time feedback on the operation status. The process of the extraction subunit performing the first feature extraction operation specifically includes: Determine the historical region-related data for each sub-region, and extract the first feature based on the first collected data and historical region-related data for each sub-region; The first feature includes regional attribute features, environmental parameter statistical features, environmental parameter trend features, spatial uniformity features, aquaculture object features, historical feeding response features, task feasibility and operation and maintenance cost features, and fishery-solar complementarity features; wherein, the spatial uniformity feature represents the degree of consistency in the distribution of each monitoring parameter within the sub-region; The feeding model includes a feature fusion layer, a hierarchical constraint layer, and a two-stage output layer; wherein... The feature fusion layer is used to perform channelization processing on each feature in the first feature, and fuse the processing results into a regional feeding demand vector. The hierarchical constraint layer is used to perform threshold truncation and time window masking on the regional feeding demand vector based on environmental parameter thresholds, time window constraints, and stability constraints to obtain the first regional feeding demand vector. When the dual-stage output layer receives the first region feeding demand vector after it has been processed by the hierarchical constraint layer, it generates local feeding demand candidates for each dynamic sub-region based on the first region feeding demand vector and the region attribute features in the first stage. In the second stage, under the constraint of the sustainability of UAV mission execution, the local feeding demand candidates are globally coordinated and allocated, and the final feeding parameters corresponding to each sub-region are output.

2. The fishery-solar hybrid drone-assisted intelligent feeding system according to claim 1, characterized in that, The process of performing sub-region division operation in the sub-unit specifically includes: Based on the first collected data, historical aquaculture environment and aquaculture object data, and historical feeding records, a time-series dynamic feature is constructed; the time-series dynamic feature includes a time-series heat map of the distribution of aquaculture objects, periodic statistics of light intensity and dissolved oxygen value, and the correspondence between historical feeding parameters and the aggregation response of aquaculture objects; The pond region is divided using a clustering algorithm based on the aforementioned temporal dynamic characteristics to determine the boundaries of the sub-regions.

3. The fishery-solar hybrid drone-assisted intelligent feeding system according to claim 2, characterized in that, The process of dividing the pond region according to the time-series dynamic features using a clustering algorithm and determining the boundaries of the sub-regions includes: A multidimensional feature vector space is constructed based on the aforementioned time-series dynamic features; the multidimensional feature vector includes the distribution density index of the cultured objects, the light intensity fluctuation index, the dissolved oxygen value change index, and the historical feeding response index; Clustering operations are performed in the feature vector space to form a preliminary set of sub-regions; During the clustering iterative update process, constraints on drone accessibility, energy consumption cost, and environmental risk factors are introduced. When the sub-region boundary violates the constraints, the corresponding cluster center position is adjusted and / or adjacent sub-regions are merged. Under the premise of satisfying the constraints, the final sub-region boundary after dynamic adjustment is output.

4. The fishery-solar hybrid drone-assisted intelligent feeding system according to claim 3, characterized in that, The process of generating feeding task priorities in the planning sub-unit specifically includes: Based on the regional attribute characteristics, feeding parameters, characteristics of aquaculture objects, and stability characteristics of environmental parameters of each dynamic sub-region, a set of task evaluation indicators is constructed. Based on the set of task evaluation indicators, a task priority value is generated for each sub-region. During the task priority value generation process, threshold constraints are set on the characteristics of the aquaculture object and the stability characteristics of environmental parameters. When the threshold constraints are violated, the task priority value of the corresponding sub-region is reduced. A task priority queue is established based on the task priority value of each sub-region, and a sorted list of feeding tasks is output for the drone cluster to execute.

5. The fishery-solar hybrid drone-assisted intelligent feeding system according to claim 4, characterized in that, The path planning process specifically includes: Based on the task priority queue and the boundary range and geometric complexity of each dynamic sub-region, a coverage constraint graph is generated. The path generation operation is performed based on the coverage constraint graph, and the reachability and energy consumption cost of the path are verified based on the real-time status information of the UAV. After the path is generated, overlap detection and time conflict detection are performed on different UAV paths. When a conflict is detected, the path adjustment operation is performed until the mission path of each UAV meets the coverage constraint, energy consumption constraint and conflict resolution condition. Output the adjusted path planning instructions.

6. The fishery-solar hybrid drone-assisted intelligent feeding system according to any one of claims 1-5, characterized in that, The division sub-unit is also used to set buffer zones at the boundaries of each divided dynamic sub-region; the width of the buffer zone is set based on the statistical results of the historical activity range of the aquaculture objects. During the task execution phase, when it is detected that the aquaculture object is moving across the boundary of the buffer zone, the feeding parameters are flexibly adjusted between adjacent sub-regions according to the crossing ratio of the aquaculture object through the collaborative control center.

7. The fishery-solar hybrid drone-assisted intelligent feeding system according to claim 6, characterized in that, During the task execution phase, the data acquisition module is also used to collect surface distribution data and underwater aggregation distribution data of the aquaculture objects. The collaborative control center then performs real-time corrections on the location of the aquaculture objects within the sub-region based on the surface distribution data and underwater aggregation distribution data of the aquaculture objects. Based on the real-time correction of the location of the aquaculture object, fine-tuning is performed on the current feeding path and feeding release location of the drone.

Images