Method and system for automatically adjusting feeding amount based on fish feeding behavior

Abstract

The present application relates to the technical field of feeding management, and particularly relates to a method and system for automatically adjusting feeding amount based on fish feeding behavior, historical feeding state parameters of a target breeding pond are acquired, the historical feeding state parameters include historical environment data, historical feeding data, historical fish feeding behavior data and historical residual material rate; target feeding behavior characteristics of the target breeding pond are constructed according to the historical feeding state parameters; feeding behavior similarity between the target breeding pond and each reference breeding pond is determined based on the target feeding behavior characteristics; the reference breeding ponds with feeding behavior similarity greater than a preset threshold are marked as candidate breeding ponds; and the target feeding amount of the target breeding pond is determined based on the historical feeding amount and the feeding behavior similarity of the candidate breeding ponds. The present application can analyze the historical feeding law of the fish population and match the experience data of similar breeding scenes, so as to realize intelligent decision of the feeding amount which is accurate, self-adaptive and cross-scene.

Description

Technical Field

[0001] This invention relates to the field of feeding management technology, and in particular to a method and system for automatically adjusting the amount of feed based on the feeding behavior of fish. Background Technology

[0002] In intensive aquaculture, precise feeding is a key step in controlling costs, ensuring quality, and preventing water quality deterioration. Insufficient feeding affects fish growth, while overfeeding leads to feed waste and water pollution.

[0003] Traditional feeding decisions rely heavily on the experience of fish farmers, estimating feed amounts by visually observing fish feeding behavior. This approach has several drawbacks: first, it is highly subjective, making standardized management difficult; second, the data is isolated, failing to integrate environmental changes and historical feeding records for comprehensive analysis; and third, the response is delayed, often only being adjusted passively after significant uneaten feed appears, lacking foresight.

[0004] With the development of IoT technology, existing solutions can collect fish activity data, but they mostly remain at the level of simple threshold alarms, triggering adjustments only when the feeding intensity falls below a fixed value. This approach severs the inherent connections between the feeding behaviors of fish in different ponds and at different times, failing to fully utilize historical data to form an intelligent decision-making model. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method and system for automatically adjusting the amount of feed based on the feeding behavior of fish. It can achieve accurate, adaptive and cross-scenario intelligent decision-making on the amount of feed by analyzing the historical feeding patterns of fish and matching experience data from similar aquaculture scenarios.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0007] A method for automatically adjusting feeding amount based on fish feeding behavior includes:

[0008] Obtain historical feeding status parameters of the target aquaculture pond, including historical environmental data, historical feeding data, historical fish feeding behavior data, and historical feed residue rate;

[0009] Based on the historical feeding data, extract the temporal features of the target aquaculture pond to characterize the feeding time distribution pattern of the fish population;

[0010] Based on the historical environmental data, the historical feeding data, and the historical feed residue rate, environmental sensitivity characteristics of the target aquaculture pond are extracted to characterize the fish population's response to environmental factors.

[0011] Based on the historical feeding data and the historical fish feeding behavior data, physiological cycle characteristics of the target aquaculture pond are extracted to characterize the dynamic recovery of fish feeding demand.

[0012] The target feeding behavior characteristics of the target aquaculture pond are constructed based on the time sequence characteristics, the environmental sensitivity characteristics, and the physiological cycle characteristics.

[0013] The similarity of the feeding behavior between the target culture pond and each reference culture pond is determined based on the target feeding behavior characteristics.

[0014] Reference aquaculture ponds with a feeding behavior similarity greater than a preset threshold are marked as candidate aquaculture ponds;

[0015] Obtain the current environmental parameters of the target culture pond, and determine the recommended feeding amount per pond for each candidate culture pond based on the current environmental parameters and the historical feeding amount of each candidate culture pond;

[0016] The recommended feeding amounts for each individual pond are weighted and fused based on the similarity of the feeding behavior to generate the target feeding amount for the target aquaculture pond.

[0017] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is as follows:

[0018] A system for automatically adjusting the amount of feed based on the feeding behavior of fish schools includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the method for automatically adjusting the amount of feed based on the feeding behavior of fish schools described above.

[0019] The beneficial effects of this invention are as follows: It acquires historical feeding status parameters of the target culture pond, including historical environmental data, historical feeding data, historical fish feeding behavior data, and historical feed residue rate, to collect multi-dimensional historical information, providing a comprehensive data foundation for subsequent analysis and avoiding data isolation; it constructs the target feeding behavior of the target culture pond based on the historical feeding status parameters, integrating environmental, feeding, behavioral, and feed residue information to form comprehensive features to capture fish feeding patterns; it determines the feeding behavior similarity between the target culture pond and each reference culture pond based on the target feeding behavior features, quantifying the similarity between different ponds through feature comparison and avoiding subjective judgment using historical data; it marks reference culture ponds with feeding behavior similarity greater than a preset threshold as candidate culture ponds to screen out highly similar reference ponds, ensuring recommendations are based on reliable matching; and it determines the target feeding amount of the target culture pond based on the historical feeding amount and feeding behavior similarity of the candidate culture ponds to generate intelligent target values ​​and achieve proactive adjustment. This application constructs the feeding behavior characteristics of fish groups by comprehensively analyzing historical data, and intelligently determines the feeding amount based on similarity matching with reference breeding ponds, thus solving the problems of subjectivity, data isolation and response lag in traditional decision-making. Attached Figure Description

[0020] Figure 1 A flowchart illustrating a method for automatically adjusting feeding amount based on fish feeding behavior, provided in an embodiment of the present invention;

[0021] Figure 2 This is a schematic diagram of the architecture of a system for automatically adjusting the amount of feed based on the feeding behavior of fish, provided in an embodiment of the present invention.

[0022] Label Explanation:

[0023] 1. A system for automatically adjusting the amount of food fed based on the feeding behavior of fish; 101. Memory; 102. Processor. Detailed Implementation

[0024] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0025] Embodiments of the present invention provide a method for automatically adjusting the amount of feed based on the feeding behavior of fish, comprising:

[0026] Obtain historical feeding status parameters of the target aquaculture pond, including historical environmental data, historical feeding data, historical fish feeding behavior data, and historical feed residue rate;

[0027] The target feeding behavior characteristics of the target aquaculture pond are constructed based on the historical feeding state parameters.

[0028] The similarity of the feeding behavior between the target culture pond and each reference culture pond is determined based on the target feeding behavior characteristics.

[0029] Reference aquaculture ponds with a feeding behavior similarity greater than a preset threshold are marked as candidate aquaculture ponds;

[0030] The target feeding amount for the target culture pond is determined based on the historical feeding amount of the candidate culture ponds and the similarity of the feeding behavior.

[0031] As can be seen from the above description, the beneficial effects of the present invention are as follows: It acquires historical feeding status parameters of the target culture pond, including historical environmental data, historical feeding data, historical fish feeding behavior data, and historical feed residue rate, to collect multi-dimensional historical information, providing a comprehensive data foundation for subsequent analysis and avoiding data isolation; it constructs the target feeding behavior of the target culture pond based on the historical feeding status parameters, integrating environmental, feeding, behavior, and feed residue information to form comprehensive features to capture fish feeding patterns; it determines the feeding behavior similarity between the target culture pond and each reference culture pond based on the target feeding behavior features, quantifying the similarity between different ponds through feature comparison and avoiding subjective judgment using historical data; it marks reference culture ponds with feeding behavior similarity greater than a preset threshold as candidate culture ponds to screen out highly similar reference ponds, ensuring recommendations are based on reliable matching; and it determines the target feeding amount of the target culture pond based on the historical feeding amount and feeding behavior similarity of the candidate culture ponds to generate intelligent target values ​​and achieve proactive adjustment. This application constructs the feeding behavior characteristics of fish groups by comprehensively analyzing historical data, and intelligently determines the feeding amount based on similarity matching with reference breeding ponds, thus solving the problems of subjectivity, data isolation and response lag in traditional decision-making.

[0032] Furthermore, constructing the target feeding behavior characteristics of the target aquaculture pond based on the feeding state parameters includes:

[0033] Based on the historical feeding data, extract the temporal features of the target aquaculture pond to characterize the feeding time distribution pattern of the fish population;

[0034] Based on the historical environmental data, the historical feeding data, and the historical feed residue rate, environmental sensitivity characteristics of the target aquaculture pond are extracted to characterize the fish population's response to environmental factors.

[0035] Based on the historical feeding data and the historical fish feeding behavior data, physiological cycle characteristics of the target aquaculture pond are extracted to characterize the dynamic recovery of fish feeding demand.

[0036] The target feeding behavior characteristics of the target aquaculture pond are constructed based on the time sequence characteristics, the environmental sensitivity characteristics, and the physiological cycle characteristics.

[0037] As described above, extracting temporal features from historical feeding data helps capture the temporal distribution patterns of fish feeding, such as identifying peak feeding periods and avoiding ignoring the impact of time patterns on similarity. Extracting environmentally sensitive features from historical environmental data, historical feeding data, and historical uneaten feed rates considers changes in environmental factors such as water temperature or dissolved oxygen levels, as well as uneaten feed feedback, ensuring the features are sensitive to environmental responses and enhancing the model's adaptability to dynamic conditions. Extracting physiological cycle features from historical feeding data and historical fish feeding behavior data focuses on the dynamic recovery of fish feeding needs, such as changes in hunger levels over time, reflecting physiological states and making the features more closely match the actual needs of the fish. Finally, constructing target feeding behavior features based on temporal features, environmentally sensitive features, and physiological cycle features, and integrating these multi-dimensional features to form a comprehensive behavioral representation, provides a more reliable basis for subsequent similarity comparisons.

[0038] Furthermore, the historical feeding data includes historical feeding times and the historical feed intake per unit body weight of fish at those historical feeding times;

[0039] The temporal features extracted from the historical feeding data to characterize the feeding time distribution of fish in the target aquaculture pond include:

[0040] Based on the preset time division granularity, the daytime is divided into multiple consecutive time intervals;

[0041] Based on the historical feeding times, the historical food intake was categorized into the corresponding time intervals.

[0042] Calculate the average historical food intake within each of the time intervals;

[0043] The target aquaculture pond is constructed based on the average historical feeding amount in each of the time intervals to characterize the temporal characteristics of the feeding time distribution pattern of the fish population.

[0044] As described above, firstly, historical feeding data includes historical feeding times and historical feed intake per unit body weight of the fish, ensuring data standardization and avoiding distortion in feed intake comparisons due to differences in fish body weight, thereby improving feature comparability. Secondly, daily time intervals are divided into continuous time intervals based on a preset time granularity. This preset time granularity achieves objective time segmentation, eliminating subjective arbitrariness and ensuring the systematic nature of time pattern capture. Next, feed intake for each feeding time is categorized into corresponding intervals based on historical feeding times. The temporal correlation of feeding times is used to organize data, avoiding the influence of isolated data points and enhancing the coherence of time distribution. Then, the average historical feed intake within each interval is calculated. Averaging smooths data fluctuations and noise, extracts representative values, and improves feature stability and reliability. Finally, time-series features are constructed based on the average feed intake of each interval. The quantified results are integrated into structured features, allowing time distribution patterns to be directly used for similarity calculations, supporting the accuracy of intelligent feeding decisions.

[0045] Furthermore, based on the historical environmental data, the historical feeding data, and the historical feed residue rate, the environmental sensitivity features extracted from the target aquaculture pond to characterize the fish population's response to environmental factors include:

[0046] Based on the historical feed surplus rate, standard feeding data that meets the preset successful feeding conditions are selected from the historical feeding data;

[0047] A standard dataset is constructed using the historical environmental factors in the historical environmental data corresponding to the standard feeding data as independent variables and the historical food intake corresponding to the standard feeding data as the dependent variable.

[0048] The response function of the standard dataset is fitted by nonlinear regression, and environmental sensitivity features of the target aquaculture pond are extracted from the response function to characterize the fish population's response to environmental factors.

[0049] As described above, firstly, feeding data is screened based on historical feed retention rates. Invalid records with high feed retention rates are excluded by pre-setting successful feeding conditions, ensuring data source reliability and avoiding noise interference, thereby improving the accuracy of feature extraction. Secondly, based on the screened standard feeding data, a standard dataset is constructed by setting historical environmental factors as independent variables and historical feed intake as dependent variables. This establishes a direct correlation model between environment and feeding behavior, strengthening the causal relationship between the data. Then, a nonlinear regression is used to fit the response function, handling the complex nonlinear relationship between environmental factors and feed intake, enhancing the model's ability to simulate actual fish behavior. Finally, environmentally sensitive features are extracted from the response function to quantify the fish's response to environmental changes, providing a scientific basis for feeding decisions. The entire process emphasizes the synergistic effect of data screening and model optimization, ensuring the authenticity and practicality of environmentally sensitive features.

[0050] Furthermore, based on the historical feeding data and the historical fish feeding behavior data, physiological cycle characteristics are extracted from the target aquaculture pond to characterize the dynamic recovery of fish feeding demand, including:

[0051] Based on the historical feeding data from two consecutive feedings, obtain the historical feed amount at the previous feeding time and the time interval between the two consecutive feeding times;

[0052] The intensity of fish feeding competition before the next feeding time is obtained based on the historical fish feeding behavior data.

[0053] Based on the correlation between the historical feeding amount at the previous feeding time and the time required for the fish population's feeding intensity to recover to a preset standard value, the physiological cycle characteristics of the target aquaculture pond used to characterize the dynamic recovery of the fish population's feeding demand are determined.

[0054] As described above, obtaining the historical feeding amount and time interval of the previous feeding time based on two consecutive historical feeding data ensures the accurate capture of the continuity and temporal sequence of feeding events, providing fundamental data for subsequent analysis and avoiding errors caused by isolated events. Then, the feeding intensity of the fish before the next feeding time is obtained based on historical fish feeding behavior data. This directly uses behavioral data to reflect the fish's hunger state as an objective indicator of recovery, replacing subjective observation. Finally, the physiological cycle characteristics are determined based on the correlation between the historical feeding amount at the previous feeding time and the time required for the fish's feeding intensity to recover to a preset standard value. This establishes a quantitative model between feeding amount and recovery time, enabling the characteristics to dynamically represent the demand recovery process and improving accuracy.

[0055] Furthermore, the target feeding behavior characteristics of the target aquaculture pond are constructed based on the temporal characteristics, the environmental sensitivity characteristics, and the physiological cycle characteristics, including:

[0056] The baseline value of feeding activity of fish in the target aquaculture pond is extracted based on the historical fish feeding behavior data;

[0057] The baseline value of the feed surplus rate of the fish population in the target aquaculture pond is extracted based on the historical feed surplus rate.

[0058] The time-series features, environmental sensitivity features, physiological cycle features, feeding activity benchmark values, and feed residue rate benchmark values ​​are spliced ​​together to generate the target feeding behavior features of the target aquaculture pond.

[0059] As described above, extracting a baseline value for feeding activity based on historical fish feeding behavior data allows for the direct quantification of normal activity levels using historical fish behavior records. This provides a benchmark reference for the intensity of behavior, thus compensating for the shortcomings of relying solely on temporal, environmental, and physiological characteristics, which may overlook dynamic fish behavior. Extracting a baseline value for feed surplus based on historical feed surplus rates establishes a standard value for feed remaining after feeding, introducing a benchmark for feed surplus status and resolving the lack of quantitative indicators for feed surplus in existing features. By concatenating temporal features, environmentally sensitive features, physiological cycle features, feeding activity baseline values, and feed surplus rate baseline values, multi-dimensional data fusion is achieved. The generated target feeding behavior features not only cover temporal distribution patterns, environmental response mechanisms, and dynamic physiological needs, but also add benchmark information on activity levels and feed surplus status. This allows the features to more completely depict fish feeding behavior, providing a more accurate basis for subsequent similarity calculations and ultimately improving the intelligence level of feed quantity adjustment.

[0060] Further, determining the similarity of feeding behavior between the target culture pond and each reference culture pond based on the target feeding behavior characteristics includes:

[0061] Obtain reference feeding behavior characteristics from each reference aquaculture pond in the preset feeding behavior pattern library;

[0062] Calculate the weighted distance between the target feeding behavior feature and each of the reference feeding behavior features in each feature dimension;

[0063] The feeding behavior similarity between the target culture pond and each of the reference culture ponds is determined based on the weighted distance.

[0064] As described above, the reference feeding behavior characteristics of each reference breeding pond in the preset feeding behavior pattern library are obtained. Historical data is used to establish a reference library, providing a reliable basis for comparison and avoiding data isolation. The weighted distance between the target feeding behavior characteristics and each reference feeding behavior characteristics in each feature dimension is calculated. The weight allocation takes into account the differences in importance of different feature dimensions, such as the different contributions of time series, environmental sensitivity, and physiological cycle characteristics, to prevent bias caused by simple averaging. The feeding behavior similarity between the target breeding pond and each reference breeding pond is determined based on the weighted distance. The distance is converted into a similarity index to achieve objective quantification and improve matching accuracy.

[0065] Further, determining the target feeding amount for the target culture pond based on the historical feeding amounts of the candidate culture ponds and the similarity of the feeding behaviors includes:

[0066] Obtain the current environmental parameters of the target aquaculture pond;

[0067] The recommended feeding amount per pond for each candidate culture pond is determined based on the current environmental parameters and the historical feeding amount of each candidate culture pond.

[0068] The recommended feeding amounts for each individual pond are weighted and fused based on the similarity of the feeding behavior to generate the target feeding amount for the target aquaculture pond.

[0069] As described above, obtaining the current environmental parameters of the target aquaculture pond directly captures the real-time environmental status, ensuring that the recommendations are based on the latest data. The recommended feeding amount for each candidate aquaculture pond is determined based on the current environmental parameters and the historical feeding amounts of each candidate pond. By combining the current environment with historical feeding amounts, personalized recommendations are generated for each candidate pond, taking into account environmental matching and making the recommendations more aligned with the dynamic needs of the current scenario. The recommended feeding amounts for each pond are weighted and fused based on the similarity of feeding behavior to generate the target feeding amount for the target aquaculture pond. Similarity is used as a weight to fuse multiple recommendations; ponds with higher similarity contribute more, thus integrating the experience of similar ponds to improve the overall accuracy and robustness of the decision-making process.

[0070] Further, determining the recommended feed amount per pond for each candidate culture pond based on the current environmental parameters and the historical feed amounts of each candidate culture pond includes:

[0071] In each of the candidate culture ponds, the environmental matching degree between the current environmental parameters and the historical environmental data of the candidate culture pond is calculated, and effective feeding data is selected from the historical feeding data of the candidate culture pond based on the environmental matching degree.

[0072] The recommended feeding amount per pond is obtained by weighting and averaging all valid feeding data in the candidate breeding ponds based on the environmental matching degree.

[0073] As described above, firstly, the environmental matching degree is calculated for each candidate aquaculture pond. This feature quantifies the similarity between the current environment and historical environments, providing an objective basis for subsequent screening and avoiding data bias caused by subjective judgment. Secondly, effective feeding data is filtered based on the environmental matching degree. This feature only retains historical records with high matching degrees, excluding irrelevant or low-similarity data, ensuring that the recommendations are based on reliable historical information. Finally, a weighted average is applied to the effective feeding data based on the environmental matching degree. This feature assigns higher weights to data with high matching degrees, making the recommendations more closely reflect the current environmental dynamics and improving overall accuracy.

[0074] Another embodiment of the present invention provides a system for automatically adjusting the amount of feed based on the feeding behavior of fish schools, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the various steps of the above-described method for automatically adjusting the amount of feed based on the feeding behavior of fish schools.

[0075] As can be seen from the above description, the beneficial effects of the present invention are as follows: It acquires historical feeding status parameters of the target culture pond, including historical environmental data, historical feeding data, historical fish feeding behavior data, and historical feed residue rate, to collect multi-dimensional historical information, providing a comprehensive data foundation for subsequent analysis and avoiding data isolation; it constructs the target feeding behavior of the target culture pond based on the historical feeding status parameters, integrating environmental, feeding, behavior, and feed residue information to form comprehensive features to capture fish feeding patterns; it determines the feeding behavior similarity between the target culture pond and each reference culture pond based on the target feeding behavior features, quantifying the similarity between different ponds through feature comparison and avoiding subjective judgment using historical data; it marks reference culture ponds with feeding behavior similarity greater than a preset threshold as candidate culture ponds to screen out highly similar reference ponds, ensuring recommendations are based on reliable matching; and it determines the target feeding amount of the target culture pond based on the historical feeding amount and feeding behavior similarity of the candidate culture ponds to generate intelligent target values ​​and achieve proactive adjustment. This application constructs the feeding behavior characteristics of fish groups by comprehensively analyzing historical data, and intelligently determines the feeding amount based on similarity matching with reference breeding ponds, thus solving the problems of subjectivity, data isolation and response lag in traditional decision-making.

[0076] Embodiments of the present invention provide a method and system for automatically adjusting feeding amount based on fish feeding behavior. This method can achieve precise, adaptive, and cross-scenario intelligent decision-making regarding feeding amount by analyzing historical feeding patterns of fish and matching them with experience data from similar aquaculture scenarios. Specific embodiments are described below:

[0077] Please refer to Figure 1 Embodiment 1 of the present invention is as follows:

[0078] A method for automatically adjusting the amount of feed based on the feeding behavior of fish groups includes the following steps S1-S5:

[0079] S1. Obtain historical feeding status parameters of the target aquaculture pond. These parameters include historical environmental data, historical feeding data, historical fish feeding behavior data, and historical feed waste rate.

[0080] These data can be collected through regular manual recording, deployment of various sensors (such as water quality sensors and underwater cameras), or manual observation and recording. Historical environmental data can include water temperature, dissolved oxygen, pH value, etc.; historical feeding data can include the time, amount, type of feed, and weight of the fish each time; historical fish feeding behavior data can include the activity level and feeding intensity of the fish during feeding; historical feed retention rate can be estimated by manual visual inspection or underwater image recognition technology.

[0081] In other equivalent embodiments, multimodal feature extraction and fusion can be performed on historical fish feeding behavior data, including:

[0082] Acquire historical fish feeding behavior data of the target aquaculture pond, including acoustic feeding intensity signals, underwater video image sequences, and water quality disturbance sensor data;

[0083] Spectral analysis was performed on the acoustic feeding intensity signal to extract time-frequency domain features characterizing feeding activity.

[0084] Optical flow field analysis was performed on the underwater video image sequence to extract visual motion features characterizing the intensity and density of fish movement.

[0085] Time-series analysis was performed on the water quality disturbance sensing data to extract disturbance features characterizing water quality changes during feeding.

[0086] An attention fusion mechanism is used to weight and fuse the time-frequency domain features, visual motion features, and perturbation features to generate fused feeding behavior features, which serve as an enhanced representation of the historical fish feeding behavior data.

[0087] S2. Construct the target feeding behavior characteristics of the target aquaculture pond based on historical feeding status parameters.

[0088] Specifically, step S2 includes the following steps S21-S24:

[0089] S21. Extract the temporal characteristics of the target aquaculture pond based on historical feeding data to characterize the feeding time distribution pattern of fish.

[0090] This involves extracting temporal characteristics from historical feeding data to characterize the feeding time distribution patterns of fish in the target aquaculture pond, aiming to capture the changing patterns of feeding activity throughout the day. This can be achieved in various ways, such as by statistically analyzing the feeding amount or frequency at different time periods (e.g., morning, noon, afternoon, and evening) in historical feeding data, or by periodically analyzing historical feeding times and extracting their main periodic components.

[0091] S22. Based on historical environmental data, historical feeding data, and historical feed surplus rate, extract the target aquaculture pond to characterize the environmental sensitivity characteristics of the fish population in response to environmental factors.

[0092] This process involves extracting environmental sensitivity characteristics from target aquaculture ponds based on historical environmental data, historical feeding data, and historical feed residue rates. These characteristics characterize the fish population's response to environmental factors, describing how feeding behavior is influenced by environmental factors (such as water temperature, dissolved oxygen, and light intensity), and then adjusting these characteristics based on actual feeding conditions and feed residue feedback. This can be achieved by establishing a multiple regression or nonlinear model between environmental factors and feed intake, using the model's coefficients or specific parameters as environmental sensitivity features; or by analyzing the trends in feed intake and feed residue rates under different environmental conditions, extracting key environmental factor thresholds or the slope of the response curve as features.

[0093] S23. Extract physiological cycle characteristics of the target aquaculture pond based on historical feeding data and historical fish feeding behavior data to characterize the dynamic recovery of fish feeding demand.

[0094] This involves extracting physiological cycle characteristics from target aquaculture ponds based on historical feeding data and historical fish feeding behavior data to characterize the dynamic recovery of fish feeding demand. The aim is to reflect the dynamic process of the fish's hunger or feeding demand recovering over time between feedings. This can be achieved by analyzing the relationship between the previous feeding amount, the interval between feedings, and the intensity of competition for food at the next feeding, constructing a function or parameter reflecting the rate of recovery of feeding demand; or by using a machine learning model, inputting historical feeding amounts and intervals, to predict the feeding demand level of the fish at a specific time point, with the model's internal parameters or specific output values ​​serving as physiological cycle characteristics.

[0095] S24. Construct target feeding behavior characteristics for the target aquaculture pond based on temporal characteristics, environmental sensitivity characteristics, and physiological cycle characteristics.

[0096] Specifically, the target feeding behavior features for the target aquaculture pond are constructed based on temporal characteristics, environmental sensitivity characteristics, and physiological cycle characteristics. The aim is to integrate the extracted multi-dimensional features to form a comprehensive, multi-dimensional representation of fish feeding behavior. This can be achieved by concatenating feature vectors to combine different types of features into a high-dimensional vector; or by using dimensionality reduction techniques such as principal component analysis (PCA) or factor analysis to fuse these features into fewer but more representative features.

[0097] In a specific application scenario, when extracting temporal features from a target aquaculture pond to characterize the feeding time distribution pattern of fish, a preset time division granularity can be set to hourly, dividing a 24-hour day into 24 consecutive time intervals. Then, the average historical feeding amount within each time interval is calculated, forming a 24-dimensional temporal feature vector. When extracting environmental sensitivity features from a target aquaculture pond to characterize the fish's response to environmental factors, historical data can be used to fit the relationship curves between environmental factors such as water temperature, dissolved oxygen, and light intensity and the fish's feeding amount using a nonlinear regression model (e.g., multinomial regression or support vector regression). Specific parameters of the curve (such as inflection points and slopes) or predicted feeding amounts under specific environmental conditions can be used as environmental sensitivity features. When extracting physiological cycle features from a target aquaculture pond to characterize the dynamic recovery of fish feeding demand, the time required for the fish's feeding intensity to drop from its peak to a preset baseline value after the last feeding can be analyzed. Combined with the previous feeding amount, an exponential or function reflecting the rate of hunger recovery can be constructed. Finally, when constructing the target feeding behavior features of the target aquaculture pond, the extracted time-series feature vectors, environmentally sensitive feature parameter sets, and physiological cycle feature indices can be concatenated. For example, by connecting them into a long vector, a unified, multi-dimensional target feeding behavior feature vector can be formed.

[0098] Specifically, historical feeding data includes historical feeding times and the historical feed intake per unit weight of fish at those times. Historical feeding data is crucial information for recording feeding behavior during the aquaculture process. Historical feeding times refer to the exact moments of each feeding, accurate to the hour, minute, or even second, used to capture the temporal patterns of fish feeding behavior. Historical feed intake per unit weight of fish at those feeding times refers to the amount of feed consumed per unit weight of fish at a specific feeding time. This standardization process eliminates the influence of differences in individual fish size across different ponds or growth stages on feed intake, making feeding data comparable across different time points or ponds. This data can be recorded using automated feeding equipment; for example, a feeder automatically records the feeding time and amount each time, and then standardizes it using a pre-set total fish weight or estimated weight. Alternatively, aquaculture personnel can manually record feeding times and amounts, and combine this with periodic weighing data for calculation.

[0099] Step S21 includes the following steps S211-S214:

[0100] S211. Based on the preset time division granularity, the daily time is divided into multiple continuous time intervals.

[0101] The preset time granularity refers to the smallest unit of time used to divide a 24-hour day into continuous time intervals. This granularity determines the precision of observing the feeding time distribution patterns of fish. For example, a granularity of 1 hour divides the day into 24 time intervals; a granularity of 30 minutes divides the day into 48 time intervals. Choosing an appropriate granularity balances the complexity of data processing with the accuracy of capturing feeding patterns. This granularity can be set based on aquaculture experience, fish species characteristics, or data analysis needs. By dividing the 24 hours of a day according to the preset time granularity, a series of continuous and non-overlapping time periods can be obtained. For example, if the time granularity is 1 hour, the time intervals are [00:00-01:00), [01:00-02:00), ..., [23:00-24:00]. This division method helps to categorize discrete historical feeding events into specific time periods, thus facilitating the statistical analysis of the feeding intensity of fish at different times.

[0102] S212. Classify the historical feeding amount into the corresponding time interval according to the historical feeding time.

[0103] For each historical feeding record, the system determines the time interval to which the feeding time belongs and then includes the corresponding historical feed intake within that time interval for statistical analysis. This allows for independent examination of fish feeding activity within each time interval in subsequent analyses. For example, one can iterate through all historical feeding records, determine the feeding time for each record, and add the corresponding feed intake to the cumulative list for the relevant time interval; alternatively, database query functions can be used to group data based on time range conditions.

[0104] S213. Calculate the average historical food intake for each time interval.

[0105] For each time interval, the historical food intake of all categories within it is summed and then divided by the number of feedings to obtain the average historical food intake. Alternatively, when the amount of data is large, statistical methods such as weighted average or median can be used to further optimize representativeness.

[0106] S214. Construct target culture ponds based on the average historical feeding amount in each time interval to characterize the temporal characteristics of the feeding time distribution pattern of fish populations.

[0107] By arranging the average historical feeding amounts calculated for all time intervals throughout the day in chronological order, a sequence or vector is formed that intuitively reflects the changing patterns of the fish's feeding intensity throughout the day. This sequence is the temporal feature, which quantifies the feeding preferences and activity levels of the fish at different times. For example, the average historical feeding amounts for all time intervals can be combined into a vector, with the dimension equal to the number of time intervals, and each element representing the average feeding amount for one time interval; alternatively, these average feeding amounts can be used as input, and their periodic components can be extracted using methods such as Fourier transform to form a more abstract temporal feature.

[0108] Step S22 includes the following steps S221-S223:

[0109] S221. Based on the historical feed surplus rate, select standard feeding data that meets the preset successful feeding conditions from the historical feeding data.

[0110] One approach is to set a feed waste rate threshold. For example, if the historical feed waste rate exceeds a certain preset percentage (such as 5% or 10%), the feeding is considered invalid and not included in the standard feeding data. Alternatively, data on the fish's feeding behavior can be combined. For instance, if the fish's feeding intensity drops rapidly after feeding and the feed waste rate is high, this can also be marked as invalid feeding. This method effectively filters out data points that do not accurately reflect the fish's true feeding behavior, laying a reliable data foundation for subsequent extraction of environmentally sensitive features.

[0111] S222. Construct a standard dataset using historical environmental factors from the historical environmental data corresponding to the standard feeding data as independent variables and historical feed intake corresponding to the standard feeding data as the dependent variable.

[0112] For each selected standard feeding data point, various environmental factors (such as water temperature, dissolved oxygen, pH, and light intensity) from its corresponding historical environmental data are used as independent variables, while the historical feed intake corresponding to that feeding is used as the dependent variable. The combination of these independent and dependent variables constitutes a data sample within the standard dataset. In this way, the feeding behavior of fish can be quantitatively correlated with their environmental conditions, providing data support for revealing the response patterns of fish to environmental changes.

[0113] S223. Fit the response function of the standard dataset using nonlinear regression, and extract the environmental sensitivity characteristics of the target aquaculture pond from the response function to characterize the fish population's response to environmental factors.

[0114] Multinomial regression models can be used to describe the trend of food intake as a function of environmental factors; alternatively, support vector regression (SVR) models can be used, with kernel functions handling nonlinear relationships; or neural network models, such as multilayer perceptrons (MLPs), can be utilized, which have powerful nonlinear fitting capabilities and can learn and represent the complex mapping relationship between environmental factors and food intake. Through nonlinear regression, a response function that can predict the food intake of fish populations under different environmental conditions can be obtained. The sensitivity of the fish population to changes in environmental factors can be quantified from the fitted response function. Specifically, key parameters can be extracted from the response function as environmental sensitivity features. For example, if the response function is polynomial, the coefficients of the polynomial can be extracted to characterize the weights of different environmental factors on food intake; if a neural network model is used, the weights from each input layer to the hidden layer and from the hidden layer to the output layer can be analyzed, or sensitivity analysis can be performed on the model to calculate the rate of change of the predicted food intake value when a certain environmental factor changes slightly, and these rates of change can be used as environmental sensitivity features. These features can intuitively reflect the activity level of the fish population's feeding behavior and its adaptability to environmental changes under specific environmental conditions.

[0115] Step S23 includes the following steps S231-S233:

[0116] S231. Obtain the historical feed amount at the previous feeding time and the time interval between the two consecutive feeding times based on the historical feeding data of two consecutive feeding times.

[0117] The historical feed amount at the previous feeding time refers to the amount of feed given when the feeding time occurs earlier in two consecutive feedings. This reflects the fish's satiety level after the previous feeding. The time interval between two consecutive feeding times directly affects the fish's digestion and the speed at which their hunger recovers.

[0118] S232. Obtain the feeding intensity of the fish population before the next feeding time based on historical fish feeding behavior data.

[0119] Historical fish feeding behavior data can be obtained from various sensors, such as the intensity of feeding sounds detected by acoustic sensors, fish aggregation density analyzed by image recognition, and the frequency of fish activity captured by underwater cameras. Fish feeding intensity is an indicator of the speed and activity of fish in response to feed, and is typically monitored at the start of feeding or for a period before feeding. The intensity of feeding can be quantified by analyzing acoustic data from a period before feeding (e.g., 5 minutes before feeding) to calculate sound pressure levels or energy within a specific frequency range. Alternatively, image recognition technology can be used to monitor fish in the pond before feeding, statistically analyzing their aggregation density or swimming speed in the feeding area to characterize feeding intensity.

[0120] S233. Based on the correlation between the historical feeding amount at the previous feeding time and the time required for the fish population's feeding intensity to recover to the preset standard value, determine the physiological cycle characteristics of the target aquaculture pond used to characterize the dynamic recovery of the fish population's feeding demand.

[0121] The preset standard value refers to a threshold at which the fish's feeding intensity reaches a certain level, indicating that the fish have recovered from a state of satiation after the last feeding to a state of hunger with a certain feeding demand. This value can be set empirically or calibrated using experimental data. The correlation describes the time required for the fish's feeding intensity to recover from the low point after the last feeding to the preset standard value, given the previous feeding amount. This relationship reflects the periodic changes in the fish's digestion rate, metabolism, and hunger. Physiological cycle characteristics are a quantitative indicator or model that can predict when the fish will again have a strong feeding demand under different feeding amounts and time intervals. A regression model (such as multinomial regression or nonlinear regression) can be constructed, using "previous feeding amount" and "time interval" as inputs and "time required for the fish's feeding intensity to recover to the preset standard value" as output, to fit the functional relationship between them. Alternatively, machine learning algorithms (such as support vector machines or neural networks) can be used to train on a large amount of historical data to learn the complex nonlinear mapping relationship between the previous feeding amount, time interval, and the recovery time of the fish's feeding intensity, thereby obtaining physiological cycle characteristics.

[0122] Step S24 includes the following steps S241-S243:

[0123] S241. Extract the baseline value of feeding activity of fish in the target aquaculture pond based on historical fish feeding behavior data.

[0124] The feeding activity benchmark refers to the average or typical feeding activity level exhibited by fish in a target aquaculture pond under normal or ideal feeding conditions. This benchmark reflects the health status, hunger level, and responsiveness to feed of the fish. This benchmark can be extracted by statistically analyzing historical fish feeding behavior data, such as calculating the average or median intensity of feeding competition within a specific time period (e.g., daily peak feeding times); or by using a machine learning model to train historical data, identifying behavioral patterns representing a "normally active" state, and quantifying them as benchmark values.

[0125] S242. Extract the baseline value of the feed surplus rate of the target fish population in the aquaculture pond based on the historical feed surplus rate.

[0126] Historical feed surplus rate refers to the proportion of feed remaining after each feeding in past feeding processes. This data is typically obtained through underwater camera observation, uneaten feed collection devices, or water quality sensors (monitoring feed decomposition products). Historical feed surplus rate is an important indicator for evaluating feeding effectiveness and fish feeding efficiency; an excessively high rate indicates overfeeding, while a low rate may indicate underfeeding. The baseline feed surplus rate refers to the proportion of feed remaining in the target aquaculture pond at an acceptable or ideal level after sufficient feeding. This baseline reflects the balance between meeting the fish's feeding needs and minimizing feed waste and water pollution. This baseline can be extracted through statistical analysis of historical feed surplus rate data, such as calculating the average feed surplus rate under successful feeding scenarios (i.e., good fish feeding with moderate feed surplus); or by combining expert experience and aquaculture goals to set an optimal range for feed surplus rate, using its center or upper limit as a baseline.

[0127] S243. The time-series characteristics, environmental sensitivity characteristics, physiological cycle characteristics, feeding activity benchmark values ​​and feed residue rate benchmark values ​​are spliced ​​together to generate the target feeding behavior characteristics of the target aquaculture pond.

[0128] Feature concatenation refers to combining multiple independent feature vectors or scalar data in a predetermined order to form a longer, more comprehensive feature vector. The purpose of this operation is to integrate feeding behavior information from different dimensions and sources into a unified representation, facilitating subsequent similarity calculations and model training. Feature concatenation can be achieved through simple vector joins, such as arranging temporal feature vectors, environmentally sensitive feature vectors, physiological cycle feature vectors, feeding activity baseline values ​​(scalars), and feed retention rate baseline values ​​(scalars) sequentially to form a high-dimensional feature vector.

[0129] In other equivalent embodiments, the steps further include:

[0130] The feeding efficiency evaluation index of the target aquaculture pond in the historical feeding cycle is obtained. The feeding efficiency evaluation index includes feed consumption per unit weight gain, feeding response delay time and feeding intensity decay rate.

[0131] A feeding efficiency scoring function is constructed based on the feeding efficiency evaluation index, and the feeding efficiency score for each historical feeding cycle is calculated.

[0132] Using the feeding efficiency score as a weight, the feeding behavior feature components corresponding to each historical feeding cycle are updated in a weighted manner to generate the optimized target feeding behavior features;

[0133] The weight coefficients of the weighted update are positively correlated with the feeding efficiency score, and the higher the historical period with the feeding efficiency score, the greater its contribution to the optimized feature.

[0134] S3. Determine the similarity of feeding behavior between the target culture pond and each reference culture pond based on the characteristics of the target feeding behavior.

[0135] Specifically, step S3 includes the following steps S31-S33:

[0136] S31. Obtain the reference feeding behavior characteristics of each reference breeding pond in the preset feeding behavior pattern library.

[0137] The acquisition of reference feeding behavior characteristics for each reference culture pond in a pre-established feeding behavior pattern library refers to retrieving a dataset representing the feeding behavior patterns of different reference culture ponds from a pre-established and stored database or knowledge base. This pre-established feeding behavior pattern library can be a structured database containing a large amount of historical aquaculture data and analysis results, where the feeding behavior characteristics of each reference culture pond have been preprocessed and standardized to facilitate subsequent comparisons. For example, the library can store typical feeding behavior characteristics for different fish species, different growth stages, different geographical regions, or different aquaculture management models. Alternatively, the library can be a dynamically updated system that continuously monitors and analyzes real-time data from a set of well-performing reference culture ponds, periodically generating and updating their reference feeding behavior characteristics.

[0138] S32. Calculate the weighted distance between the target feeding behavior feature and each reference feeding behavior feature in each feature dimension.

[0139] The calculation can be performed using mathematical methods such as weighted Euclidean distance, weighted Mahalanobis distance, or weighted cosine distance. The weights can be set based on the importance of each feature dimension (such as temporal features, environmental sensitivity features, physiological cycle features, feeding activity baseline values, and feed retention rate baseline values) to the feeding behavior of the fish population. These weights can be determined through expert experience, historical data analysis (e.g., determining the degree of influence of each feature on feed amount or feed retention rate through regression analysis or machine learning models), or adaptive learning algorithms.

[0140] S33. Determine the similarity of feeding behavior between the target culture pond and each reference culture pond based on the weighted distance.

[0141] The distance value can be mapped to a similarity score between 0 and 1 using an inverse function (such as 1 / (1+distance)) or an exponential decay function (such as exp(-k*distance)), where 0 represents complete dissimilarity and 1 represents complete similarity.

[0142] S4. Mark reference aquaculture ponds with feeding behavior similarity greater than a preset threshold as candidate aquaculture ponds.

[0143] S5. Determine the target feeding amount for the target culture pond based on the historical feeding amount and feeding behavior similarity of the candidate culture ponds.

[0144] Specifically, step S5 includes the following steps S51-S53:

[0145] S51. Obtain the current environmental parameters of the target aquaculture pond.

[0146] Obtaining the current environmental parameters of the target aquaculture pond refers to the real-time collection of various environmental data to reflect the current aquaculture conditions. This can be achieved by deploying various sensors in the pond, such as temperature sensors, dissolved oxygen sensors, pH sensors, and ammonia nitrogen sensors, to monitor and collect environmental parameters such as water temperature, dissolved oxygen content, pH value, and ammonia nitrogen concentration in real time. These sensors can periodically upload data to a data processing unit. Alternatively, remote sensing technology or drones equipped with environmental monitoring equipment can be used to inspect the aquaculture pond, acquiring macroscopic environmental data such as water quality and color, and extracting current environmental parameters through image processing or spectral analysis techniques.

[0147] S52. Determine the recommended feeding amount per pond for each candidate culture pond based on the current environmental parameters and the historical feeding amount of each candidate culture pond.

[0148] This involves constructing a predictive model based on environmental parameters and historical feeding data, such as a regression model or a neural network model. This model takes current environmental parameters as input and is trained using historical feeding data from candidate rearing ponds. Using this model, a recommended feeding amount for each candidate rearing pond under the current environment can be predicted.

[0149] Step S52 includes the following steps S521-S522:

[0150] S521. In each candidate aquaculture pond, calculate the environmental matching degree between the current environmental parameters and the historical environmental data of the candidate aquaculture pond, and select valid feeding data from the historical feeding data of the candidate aquaculture pond based on the environmental matching degree.

[0151] The environmental matching degree can be calculated in several ways. For example, it can be calculated using the Euclidean distance, cosine similarity, or Manhattan distance between the current environmental parameters (such as water temperature, dissolved oxygen, pH, ammonia nitrogen content, etc.) and the corresponding parameters in historical environmental data. The smaller the distance or the higher the similarity, the higher the environmental matching degree. Alternatively, fuzzy logic reasoning or a pre-trained machine learning model can be used, taking the current environmental parameters as input and outputting a matching degree score between 0 and 1, where 1 indicates highly similar environmental conditions and 0 indicates significantly different environmental conditions. A preset environmental matching degree threshold is set, and only historical feeding data with a matching degree higher than this threshold are retained as valid feeding data. Furthermore, a dynamic threshold adjustment strategy can be used, or historical data can be sorted according to matching degree, selecting the top N% of data with the highest matching degree as valid feeding data.

[0152] S522. Based on the environmental matching degree, the recommended feeding amount per pond is obtained by weighted averaging of all effective feeding data in the candidate breeding ponds.

[0153] One approach is to calculate a weighted average of the selected effective feeding data, using their corresponding environmental matching scores as weights. For example, the recommended feeding amount for a single pool can be represented as the sum of the products of all effective historical feeding amounts and their corresponding environmental matching scores, divided by the sum of all environmental matching scores. Alternatively, a non-linear weighting function can be designed so that the impact of data with higher matching scores on the final recommendation amount increases exponentially, further strengthening the role of high-matching-score data.

[0154] S53. Based on the similarity of feeding behavior, the recommended feeding amount for each individual pond is weighted and fused to generate the target feeding amount for the target aquaculture pond.

[0155] One approach is to use the similarity of feeding behavior among each candidate pond as a weight to weight the corresponding recommended feeding amount per pond. Assuming the similarity of candidate pond A is S_A and the recommended feeding amount per pond is T_A; and the similarity of candidate pond B is S_B and the recommended feeding amount per pond is T_B, then the target feeding amount can be calculated as (S_A*T_A+S_B*T_B) / (S_A+S_B). Alternatively, ensemble learning methods from machine learning, such as weighted voting or stacking, can be used. The recommended feeding amount per pond for each candidate pond can be used as the output of a base learner, and the feeding behavior similarity can be used as the fusion weight to train a meta-learner to generate the final target feeding amount.

[0156] Please refer to Figure 2 Embodiment two of the present invention is as follows:

[0157] A system 1 for automatically adjusting the amount of feed based on the feeding behavior of fish groups includes a memory 101, a processor 102, and a computer program stored in the memory 101 and running on the processor 102. When the processor 102 executes the computer program, it implements the various steps of the method for automatically adjusting the amount of feed based on the feeding behavior of fish groups according to Embodiment 1.

[0158] This invention achieves significant technological advancements by constructing a multi-dimensional feeding behavior feature system, introducing a cross-pond similarity matching mechanism, and designing a two-layer fusion decision-making strategy. First, the solution comprehensively integrates historical environmental data, feeding data, fish feeding behavior data, and feed residue rates, extracting temporal features, environmental sensitivity features, and physiological cycle features to form standardized target feeding behavior features. This effectively solves the problems of isolated data and single feature dimensions in traditional methods, laying a reliable foundation for accurate decision-making. Second, it innovatively establishes a feeding behavior similarity calculation and candidate pond selection mechanism, enabling target ponds to learn from the historical experience of reference ponds that highly match their feeding patterns. This achieves cross-scenario transfer of aquaculture knowledge, particularly suitable for new ponds or situations with sparse historical data, significantly improving the adaptability and robustness of decision-making. Finally, it calculates environmental matching degree based on current environmental parameters to screen effective feeding data, and weights and fuses the recommendations for each candidate pond based on feeding behavior similarity, forming a feedforward intelligent decision-making approach. Compared to traditional lagging adjustment methods that rely on manual experience or simple threshold alarms, this approach has stronger foresight and accuracy. Overall, this invention effectively reduces the risk of feed waste and water pollution, and improves the intelligence and standardization of feeding management.

[0159] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for automatically adjusting the amount of feed based on the feeding behavior of fish, characterized in that, include: Obtain historical feeding status parameters of the target aquaculture pond, including historical environmental data, historical feeding data, historical fish feeding behavior data, and historical feed residue rate; Based on the historical feeding data, extract the temporal features of the target aquaculture pond to characterize the feeding time distribution pattern of the fish population; Based on the historical environmental data, the historical feeding data, and the historical feed residue rate, environmental sensitivity characteristics of the target aquaculture pond are extracted to characterize the fish population's response to environmental factors. Based on the historical feeding data and the historical fish feeding behavior data, physiological cycle characteristics of the target aquaculture pond are extracted to characterize the dynamic recovery of fish feeding demand. Based on the temporal characteristics, environmental sensitivity characteristics, and physiological cycle characteristics, the target feeding behavior characteristics of the target aquaculture pond are constructed, including: The baseline value of feeding activity of fish in the target aquaculture pond is extracted based on the historical fish feeding behavior data; The baseline value of the feed surplus rate of the fish population in the target aquaculture pond is extracted based on the historical feed surplus rate. The time-series features, environmental sensitivity features, physiological cycle features, feeding activity benchmark values, and feed residue rate benchmark values ​​are spliced ​​together to generate the target feeding behavior features of the target aquaculture pond. The similarity of the feeding behavior between the target culture pond and each reference culture pond is determined based on the target feeding behavior characteristics. Reference aquaculture ponds with a feeding behavior similarity greater than a preset threshold are marked as candidate aquaculture ponds; Obtain the current environmental parameters of the target culture pond, and determine the recommended feeding amount per pond for each candidate culture pond based on the current environmental parameters and the historical feeding amount of each candidate culture pond; The recommended feeding amounts for each individual pond are weighted and fused based on the similarity of the feeding behavior to generate the target feeding amount for the target aquaculture pond.

2. The method for automatically adjusting the feeding amount based on the feeding behavior of fish groups according to claim 1, characterized in that, The historical feeding data includes historical feeding times and the historical feed intake per unit body weight of fish at those historical feeding times; The temporal features extracted from the historical feeding data to characterize the feeding time distribution of fish in the target aquaculture pond include: Based on the preset time division granularity, the daytime is divided into multiple consecutive time intervals; Based on the historical feeding times, the historical food intake was categorized into the corresponding time intervals. Calculate the average historical food intake within each of the time intervals; The target aquaculture pond is constructed based on the average historical feeding amount in each of the time intervals to characterize the temporal characteristics of the feeding time distribution pattern of the fish population.

3. The method for automatically adjusting the feeding amount based on the feeding behavior of fish groups according to claim 2, characterized in that, Based on the historical environmental data, the historical feeding data, and the historical feed residue rate, the environmental sensitivity features extracted from the target aquaculture pond to characterize the fish population's response to environmental factors include: Based on the historical feed surplus rate, standard feeding data that meets the preset successful feeding conditions are selected from the historical feeding data; A standard dataset is constructed using the historical environmental factors in the historical environmental data corresponding to the standard feeding data as independent variables and the historical food intake corresponding to the standard feeding data as the dependent variable. The response function of the standard dataset is fitted by nonlinear regression, and environmental sensitivity features of the target aquaculture pond are extracted from the response function to characterize the fish population's response to environmental factors.

4. The method for automatically adjusting the feeding amount based on the feeding behavior of fish groups according to claim 1, characterized in that, Based on the historical feeding data and the historical fish feeding behavior data, the physiological cycle characteristics extracted from the target aquaculture pond to characterize the dynamic recovery of fish feeding demand include: Based on the historical feeding data from two consecutive feedings, obtain the historical feed amount at the previous feeding time and the time interval between the two consecutive feeding times; The intensity of fish feeding competition before the next feeding time is obtained based on the historical fish feeding behavior data. Based on the correlation between the historical feeding amount at the previous feeding time and the time required for the fish population's feeding intensity to recover to a preset standard value, the physiological cycle characteristics of the target aquaculture pond used to characterize the dynamic recovery of the fish population's feeding demand are determined.

5. The method for automatically adjusting the feeding amount based on the feeding behavior of fish groups according to claim 1, characterized in that, Determining the similarity of feeding behavior between the target culture pond and each reference culture pond based on the target feeding behavior characteristics includes: Obtain reference feeding behavior characteristics from each reference aquaculture pond in the preset feeding behavior pattern library; Calculate the weighted distance between the target feeding behavior feature and each of the reference feeding behavior features in each feature dimension; The feeding behavior similarity between the target culture pond and each of the reference culture ponds is determined based on the weighted distance.

6. The method for automatically adjusting the feeding amount based on the feeding behavior of fish groups according to claim 1, characterized in that, The recommended feeding amount per pond for each candidate culture pond is determined based on the current environmental parameters and the historical feeding amount of each candidate culture pond, including: In each of the candidate culture ponds, the environmental matching degree between the current environmental parameters and the historical environmental data of the candidate culture pond is calculated, and effective feeding data is selected from the historical feeding data of the candidate culture pond based on the environmental matching degree. The recommended feeding amount per pond is obtained by weighting and averaging all valid feeding data in the candidate breeding ponds based on the environmental matching degree.

7. A system for automatically adjusting the amount of feed based on the feeding behavior of fish, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement each step of the method for automatically adjusting the amount of feed based on the feeding behavior of fish groups according to any one of claims 1-6.

Images