Circulating water aquaculture management and control method based on cloud platform and cloud platform

Abstract

This invention relates to the field of intelligent management and control technology for recirculating aquaculture systems (RAS), and discloses a cloud-based RAS management and control method and cloud platform. During the operation of the RAS system, behavioral characteristic data of the cultured organisms are collected to construct a behavioral characteristic time series. Based on the behavioral characteristic time series, structural features representing behavioral change patterns are extracted. The behavioral characteristic time series and its corresponding structural features are input into a deep learning model, which outputs the risk evolution rate representing the rate of change of the RAS risk state over time. The risk state is updated recursively over time based on the current risk state and the risk evolution rate. When the updated risk state does not reach a preset alarm threshold and the risk evolution rate shows a continuous upward trend, the system is determined to have entered a potential stress state, triggering early control operations. Through this method, the proactive identification and control of the risk state of the RAS system is achieved.

Description

Technical Field

[0001] This invention relates to the field of intelligent management and control technology for recirculating aquaculture systems, specifically to a cloud-based management and control method for recirculating aquaculture systems and the cloud platform itself. Background Technology

[0002] Recirculating aquaculture systems achieve the reuse of aquaculture water through physical filtration, biochemical treatment, and water circulation. They have advantages such as water conservation and strong controllability and have been widely used in high-density aquaculture scenarios.

[0003] Existing automated management methods for recirculating aquaculture systems typically rely on water quality parameters as the primary basis for monitoring and control, such as dissolved oxygen, ammonia nitrogen, nitrite, and pH. When these water quality parameters exceed preset thresholds, the system then triggers control measures such as oxygenation, water exchange, or biological treatment.

[0004] However, in actual aquaculture practice, it has been found that when farmed organisms are affected by environmental disturbances, changes in metabolic load, or potential diseases, they often first exhibit stress behaviors such as abnormal feeding, changes in population structure, and disordered swimming behavior, while water quality parameters remain within the normal range. Because water quality changes have a significant lag, existing control methods based on water quality thresholds are insufficient to respond promptly, easily leading to a rapid accumulation of risks within a short period, and subsequently causing problems such as sudden changes in water quality and population mortality.

[0005] Some existing technologies attempt to introduce behavior recognition methods, but they mostly remain at the level of behavior classification or anomaly detection. They are usually only used for alarm prompts and have failed to establish a quantitative relationship between behavior changes and the evolution of system risks. In particular, they lack the ability to predict the speed and trend of risk growth, and it is still difficult to achieve true early control.

[0006] Therefore, there is an urgent need for a new method for managing recirculating aquaculture systems that can identify the evolution trend of system risks and quantify the rate of risk growth based on changes in the behavior of aquaculture organisms, even when water quality parameters are not yet abnormal, thereby achieving forward-looking system regulation. Summary of the Invention

[0007] The purpose of this invention is to provide a cloud-based recirculating aquaculture system management method, management system, computer equipment, and computer-readable storage medium to solve the problems mentioned in the background art and further improve the comprehensive benefits of the agricultural-solar complementary system.

[0008] This invention provides a cloud-based method for managing recirculating aquaculture systems, comprising the following steps:

[0009] Step S1: During the operation of the recirculating aquaculture system, collect behavioral characteristic data of the cultured organisms and construct a behavioral characteristic time series based on the behavioral characteristic data;

[0010] Step S2: Based on the behavioral feature time series, extract the behavioral change structural features that characterize the behavioral change pattern;

[0011] Step S3: Input the time series of behavioral features and the corresponding structural features of behavioral changes as input data into the deep learning model in the cloud platform. The deep learning model outputs the risk evolution rate, which represents the rate of change of the risk state of the recirculating aquaculture system over time.

[0012] Step S4: Based on the current risk status and the risk evolution rate, the risk status of the recirculating aquaculture system is updated over time. When the updated risk status does not reach the preset alarm threshold and the risk evolution rate shows a continuous upward trend, the recirculating aquaculture system is determined to have entered a potential stress state, and an early control operation is triggered.

[0013] As an example, prior to step S3, an anomaly-leading analysis of the behavior state of the aquaculture object is performed based on the behavioral characteristic time series, specifically including:

[0014] Perform sliding time window analysis on the time series of the behavioral features to calculate the magnitude, direction, and stability index of the change of at least one behavioral feature within the current time window;

[0015] The changes in the behavioral characteristics within the current time window are compared with the distribution of changes in the behavioral characteristics within the corresponding historical time period. When the changes in the behavioral characteristics deviate from the historical normal distribution and do not correspond to abnormal water quality parameters, it is determined that the current behavioral state has abnormal leading characteristics.

[0016] As an example, the behavioral characteristic data includes one or more of the following: feeding response time of farmed animals, changes in population aggregation, mean swimming speed, fluctuation range of swimming speed, and changes in relative distance between individuals;

[0017] The behavioral change structural features include at least one of the following: behavioral feature change trend features in the time dimension, periodic features, mutation point features, and stability disruption features.

[0018] As an example, in step S3, the behavioral feature time series and the behavioral change structural features are fused in a time-aligned manner and input into the deep learning model in the form of multi-channel time series data.

[0019] As an example, the deep learning model includes:

[0020] A behavioral feature time series encoding subnetwork is used to extract temporal features from the behavioral feature time series to obtain a temporal representation characterizing short-term behavioral perturbations;

[0021] A behavior change structural feature encoding subnetwork is used to model the behavior change structural features to obtain a structural representation that characterizes the medium- and long-term behavior patterns and their stability changes;

[0022] A behavior stability violation perception subnetwork is used to jointly model the temporal representation and the structural representation, and output a stability violation representation vector that represents the degree of collapse of the statistically stable structure in the behavioral feature time series.

[0023] The risk evolution rate output module is used to output the risk evolution rate of the recirculating aquaculture system risk state over time based on the stability destruction characterization vector.

[0024] As an example, the deep learning model also includes an attention bias submodule driven by anomaly-leading features, specifically used for:

[0025] Receive abnormal leading features obtained based on the time series of the behavioral features, and perform time series modeling on the changes of the abnormal leading features within a continuous time window to generate attention bias parameters that are updated over time.

[0026] The attention bias parameter is introduced into the attention calculation process of the behavior stability disruption perception subnetwork, so that the attention bias parameter participates in the calculation or normalization process of attention weights. In order to dynamically modulate the attention weight distribution corresponding to different behavior change structural features during the joint modeling of behavior feature time series and behavior change structural features by the behavior stability disruption perception subnetwork, the attention weight distribution corresponding to different behavior change structural features is dynamically modulated.

[0027] As an example, the advance control operation includes at least one of adjusting the feeding strategy, changing the water circulation flow rate, increasing the oxygenation intensity, and activating the water quality conditioning equipment in advance.

[0028] The present invention also provides a cloud platform, comprising:

[0029] The behavioral feature time series construction module is used to collect behavioral feature data of aquaculture objects during the operation of the recirculating aquaculture system, and construct a behavioral feature time series based on the behavioral feature data.

[0030] The behavior change structural feature extraction module is used to extract behavior change structural features that characterize behavior change patterns based on the behavior feature time series.

[0031] The risk evolution rate calculation module is used to input the time series of the behavioral features and the corresponding structural features of behavioral changes as input data into the deep learning model in the cloud platform, and the deep learning model outputs the risk evolution rate, which represents the rate of change of the risk state of the recirculating aquaculture system over time.

[0032] The risk status recursion and control triggering module is used to recursively update the risk status of the recirculating aquaculture system based on the current risk status and the risk evolution rate. When the updated risk status does not reach the preset alarm threshold and the risk evolution rate shows a continuous upward trend, the recirculating aquaculture system is determined to have entered a potential stress state, and an early control operation is triggered.

[0033] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described in any of the preceding claims.

[0034] The present invention also provides a computer-readable storage medium in which a computer program stored may be executed by a processor to implement the method as described in any of the preceding claims.

[0035] Compared with the prior art, the beneficial effects of the present invention are:

[0036] This invention introduces a joint modeling mechanism based on cloud platform-based behavioral characteristic time series and behavioral change structural characteristics into a recirculating aquaculture system. It uses a deep learning model to output the risk evolution rate that represents the rate of change of risk status over time. By combining the risk evolution trend, it can make early judgments and controls on potential stress states before the risk status reaches the alarm threshold. Thus, it can achieve forward-looking identification and refined management of the operational risks of the aquaculture system without relying on significant changes in water quality parameters. Attached Figure Description

[0037] Figure 1 This is a schematic flowchart of a cloud-based recirculating aquaculture management method disclosed in an embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram of the structure of the deep learning model disclosed in an embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram of another structure of the deep learning model disclosed in an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram of a cloud platform-based structure disclosed in an embodiment of the present invention;

[0041] Figure 5 This is a schematic diagram of the structure of a computer-readable storage medium disclosed in an embodiment of the present invention. Detailed Implementation

[0042] The following specific embodiments illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0043] Furthermore, the technical features involved in the different embodiments of this application described below can be combined with each other as long as they do not conflict with each other.

[0044] This invention provides a cloud-based recirculating aquaculture system (RAS) management solution. Its core lies in using the behavioral state of the cultured organisms as a crucial information source reflecting the system's potential risks. This information is dynamically and quantitatively integrated into the risk assessment and control decision-making process of the RAS system, addressing the problem of existing technologies that primarily rely on water quality parameters and struggle to identify early stress risks in a timely manner. Specifically, this solution jointly models the time series of the cultured organisms' behavioral characteristics and the structural features of their behavioral changes. It utilizes a deep learning model on the cloud platform to output the risk evolution rate of the risk state over time. When the risk state has not yet reached the alarm threshold but the risk evolution rate continues to rise, it preemptively determines that the system has entered a potential stress state and triggers control operations. This achieves proactive risk management of the RAS system, ensuring the stability of aquaculture while improving the overall reliability and refined management level of the system.

[0045] In this embodiment, the recirculating aquaculture system includes, but is not limited to, aquaculture ponds, water circulation pipelines, filtration devices, oxygenation devices, water quality conditioning equipment, and various sensing and perception devices for collecting information on the behavior of aquaculture organisms and water quality.

[0046] The cloud platform connects to the recirculating aquaculture system via a network to receive data collected on-site and perform unified data processing, model inference, and control decisions. It should be noted that the cloud platform described in this embodiment can be deployed in a public cloud, private cloud, or edge cloud environment, and can run deep learning models, time series analysis programs, and strategy decision-making modules internally.

[0047] Please see Figure 1 This invention provides a cloud-based method for managing recirculating aquaculture systems, comprising the following steps:

[0048] Step S1: During the operation of the recirculating aquaculture system, collect behavioral characteristic data of the cultured organisms and construct a behavioral characteristic time series based on the behavioral characteristic data;

[0049] As an example, the behavioral characteristic data includes one or more of the following: feeding response time of farmed animals, changes in population aggregation, mean swimming speed, fluctuation range of swimming speed, and changes in relative distance between individuals;

[0050] In this step, behavioral feature data can be acquired through video acquisition devices, 3D underwater imaging devices, underwater acoustic sensors, or image recognition-based behavioral analysis terminals. The cloud platform or its associated edge nodes preprocess the acquired raw behavioral data to extract quantifiable behavioral feature indicators.

[0051] Behavioral characteristic data includes one or more of the following: feeding response time of farmed animals, such as the length of time required for farmed animals to begin exhibiting obvious feeding behavior after a feeding signal is given; changes in group aggregation, such as changes in the spatial concentration of farmed animals per unit time; mean swimming speed, used to reflect the overall activity level of farmed animals; fluctuation range of swimming speed, used to reflect the stability of behavioral state; and changes in relative distance between individuals, used to characterize changes in the looseness or tightness of group structure. It can be understood that the above-mentioned behavioral characteristic data can reflect the behavioral state and its changing trends of farmed animals from different dimensions.

[0052] After collecting behavioral feature data, the cloud platform timestamps each behavioral feature according to a preset sampling period and arranges them in a unified timeline order to construct a behavioral feature time series. In practical applications, the behavioral feature time series can be sampled at fixed time intervals or updated using an event-driven approach.

[0053] Step S2: Based on the behavioral feature time series, extract the behavioral change structural features that characterize the behavioral change pattern;

[0054] As an example, the behavioral change structural features include at least one of the following: behavioral feature change trend features in the time dimension, periodic features, mutation point features, and stability disruption features.

[0055] In this step, the cloud platform further extracts structural features of behavioral change, representing patterns of behavioral change, based on the aforementioned obtained behavioral feature time series. Unlike behavioral features at a single moment, these structural features focus on characterizing the organization and evolution of behavioral features over time, reflecting whether structural changes have occurred in the behavioral state. It can be understood that structural features of behavioral change are a higher-level representation of behavioral change, distinct from the original behavioral feature time series.

[0056] The behavioral change structural features include at least one of the following: trend features, such as the rising, falling, or slowly drifting trend of behavioral features over time; periodic features, such as behavioral fluctuation patterns related to feeding cycles and diurnal rhythms; abrupt change features, such as a significant jump or break in behavioral features at a certain point in time; and stability disruption features, such as the phenomenon of continuous and intensified fluctuations after behavioral features have remained stable for a long time.

[0057] In this embodiment, the cloud platform can use methods such as sliding window analysis, statistical decomposition, or time series transformation to process the behavioral feature time series, thereby extracting the aforementioned behavioral change structural features.

[0058] As an example, prior to step S3, an anomaly-leading analysis of the behavior state of the aquaculture object is performed based on the behavioral characteristic time series, specifically including:

[0059] Perform sliding time window analysis on the time series of the behavioral features to calculate the magnitude, direction, and stability index of the change of at least one behavioral feature within the current time window;

[0060] The changes in the behavioral characteristics within the current time window are compared with the distribution of changes in the behavioral characteristics within the corresponding historical time period. When the changes in the behavioral characteristics deviate from the historical normal distribution and do not correspond to abnormal water quality parameters, it is determined that the current behavioral state has abnormal leading characteristics.

[0061] In this step, before performing risk evolution rate analysis, an anomaly leading analysis is first performed. This analysis aims to capture weak signals of behavioral characteristics deviating from normal patterns before water quality parameters fluctuate significantly.

[0062] Specifically, the cloud platform first sets a sliding time window (e.g., a window length of 30 minutes and a sliding step of 5 minutes), performs statistical analysis on the behavioral feature time series within each window, and extracts one or more change features, including the magnitude of change (e.g., the average absolute change between adjacent sampling points), the direction of change (e.g., the cumulative slope of an upward or downward trend), and the stability index of change (e.g., the standard deviation or coefficient of variation of sample values ​​within the window).

[0063] Furthermore, the cloud platform compares the calculated change characteristics within the current time window with the distribution of change characteristics during the same historical period (such as the same aquaculture stage or the same day-night cycle). It determines whether the change significantly deviates from the historical normal range by calculating the degree of deviation from the historical distribution mean or by using hypothesis testing methods (such as the Z-test). If the current behavioral change characteristics significantly deviate from the historical normal distribution, and the simultaneously monitored water quality parameters such as dissolved oxygen and ammonia nitrogen are all within the preset normal threshold range, the cloud platform determines that the current behavioral state has exhibited abnormal leading characteristics, indicating that the recirculating aquaculture system may have entered a potential early stress stage, but has not yet triggered a traditional water quality alarm.

[0064] In addition, this anomalous leading feature will serve as an enhanced input to subsequent deep learning models to improve the model's predictive sensitivity to the speed of risk evolution; details will not be elaborated further.

[0065] Step S3: Input the time series of behavioral features and the corresponding structural features of behavioral changes as input data into the deep learning model in the cloud platform. The deep learning model outputs the risk evolution rate, which represents the rate of change of the risk state of the recirculating aquaculture system over time.

[0066] As an example, in step S3, the behavioral feature time series and the behavioral change structural features are fused in a time-aligned manner and input into the deep learning model in the form of multi-channel time series data.

[0067] In this step, the time series of behavioral features and the structural features of behavioral changes are fused according to a time-aligned method and input into the deep learning model in the form of multi-channel time series data. Specifically, the cloud platform constructs the first channel data from the original behavioral features (such as feeding response duration, group aggregation degree, etc.) corresponding to the same timestamp, and constructs the second channel data from the extracted structural features of behavioral changes (such as trend features, mutation point features, etc.), ensuring that the two channels of data are strictly aligned on the time axis. Then, the dual-channel data is input into the deep learning model in the form of a multi-dimensional time series tensor, enabling the deep learning model to simultaneously perceive the instantaneous performance of behavioral features and their structural changes in the time dimension.

[0068] As an example, please refer to Figure 2 The deep learning model includes:

[0069] A behavioral feature time series encoding subnetwork is used to extract temporal features from the behavioral feature time series to obtain a temporal representation characterizing short-term behavioral perturbations;

[0070] Among them, the behavioral feature time series encoding sub-network is used to extract temporal features from the behavioral feature time series to obtain a temporal representation of short-term behavioral disturbances.

[0071] This sub-network can adopt a structure combining a one-dimensional convolutional neural network (1D-CNN) and a gated recurrent unit (GRU): First, multiple one-dimensional convolutional layers extract local temporal patterns of behavioral features. The convolution operation can be represented as:

[0072]

[0073] in, This represents a one-dimensional convolution operation. and The first The weights and biases of convolutional layers. This is the output of the previous layer.

[0074] Subsequently, the GRU layer is used to capture the temporal dependencies in the short time series. The update gate and reset gate of the GRU are calculated as follows: , The final output is a temporal representation vector characterizing short-term (e.g., minutes to hours) behavioral perturbation features. .

[0075] A behavior change structural feature encoding subnetwork is used to model the behavior change structural features to obtain a structural representation that characterizes the medium- and long-term behavior patterns and their stability changes;

[0076] The behavioral change structural feature encoding subnetwork is used to model the structural features of behavioral changes, obtaining a structural representation of medium- and long-term behavioral patterns and their stability changes. This subnetwork can adopt a structure combining fully connected layers and a self-attention mechanism: firstly, various structural features (such as trend slope, periodic intensity, and mutation point location encoding) are mapped to a high-dimensional space through fully connected layers. Then, a multi-head self-attention layer is used to capture the interrelationships between different structural features and their contribution to temporal evolution. The attention is calculated as follows:

[0077] .

[0078] The final output is a structural representation vector that represents the long-term (e.g., hours to days) behavioral patterns and their stability states. .

[0079] A behavior stability violation perception subnetwork is used to jointly model the temporal representation and the structural representation, and output a stability violation representation vector that represents the degree of collapse of the statistically stable structure in the behavioral feature time series.

[0080] The behavioral stability violation perception subnetwork is used to jointly model temporal and structural representations, outputting a stability violation representation vector that characterizes the degree of collapse of statistically stable structures in the behavioral feature time series. This subnetwork can adopt a fusion structure based on cross-attention: using temporal representation vectors... As a query, the structure represents a vector. As the key and value, the information from both is fused through a cross-attention mechanism:

[0081] .

[0082] This enables the network to perceive whether short-term behavioral disturbances have disrupted the structural stability of medium- to long-term behavioral patterns. (Fused features) After several fully connected layers and normalization processing, the final output is a stability violation representation vector. The magnitude or specific dimension of this vector can reflect the degree of disintegration of a behaviorally stable structure.

[0083] The risk evolution rate output module is used to output the risk evolution rate of the recirculating aquaculture system risk state over time based on the stability destruction characterization vector.

[0084] The risk evolution rate output module is used to output the risk evolution rate of the recirculating aquaculture system based on the stability failure representation vector, showing the rate of change of the risk state over time. This module is typically a multilayer perceptron (MLP) that receives the stability failure representation vector. As input, it undergoes a nonlinear transformation through several hidden layers. The final output is a scalar value, namely the risk evolution speed. Understandably, this risk evolution rate value is a positive real number, representing the increment of the risk state per unit time; the larger the value, the faster the risk accumulates.

[0085] As an example, please refer to Figure 3 The deep learning model further includes an attention bias submodule driven by anomaly-leading features, specifically used for:

[0086] Receive abnormal leading features obtained based on the time series of the behavioral features, and perform time series modeling on the changes of the abnormal leading features within a continuous time window to generate attention bias parameters that are updated over time.

[0087] The attention bias submodule receives anomalous leading features derived from behavioral feature time series and performs temporal modeling on the changes of these anomalous leading features within consecutive time windows to generate attention bias parameters that update over time. This submodule may contain a small GRU network that processes the anomalous leading feature sequences from multiple consecutive time windows. As input, the GRU outputs the hidden state at the last time step. Then, it is mapped to an attention bias parameter vector through a fully connected layer: .

[0088] The attention bias parameter is introduced into the attention calculation process of the behavior stability disruption perception subnetwork, so that the attention bias parameter participates in the calculation or normalization process of attention weights. In order to dynamically modulate the attention weight distribution corresponding to different behavior change structural features during the joint modeling of behavior feature time series and behavior change structural features by the behavior stability disruption perception subnetwork, the attention weight distribution corresponding to different behavior change structural features is dynamically modulated.

[0089] In this step, the attention bias parameters generated by the attention bias submodule are introduced into the attention calculation process in the behavior stability disruption perception subnetwork, so that the attention bias parameters participate in the calculation or normalization process of attention weights. In order to dynamically modulate the attention weight distribution corresponding to different behavior change structural features during the joint modeling of behavior feature time series and behavior change structural features by the behavior stability disruption perception subnetwork.

[0090] In specific implementation, the bias parameter vector Cross-attention mechanisms can be introduced in one of the following two ways, for example, by... Add to the key vector or Value vector Above, that is: or Then use or Subsequent attention calculations are performed to guide the query or value passing process with anomalous leading features.

[0091] After calculating the attention score and before Softmax normalization, It is added to the score matrix as a bias term, that is:

[0092] .

[0093] Understandably, this method directly adjusts the attention weight allocation between different feature dimensions, making the model more inclined to focus on structural features that are highly correlated with anomalous leading features.

[0094] In this way, the model can dynamically strengthen the attention weight of behavioral change patterns that are sensitive to abnormal leading features when fusing temporal and structural representations. This enhances the ability to perceive potential unstable signals before risks have accumulated significantly, effectively improving the system's early warning sensitivity to early risk evolution.

[0095] Step S4: Based on the current risk status and the risk evolution rate, the risk status of the recirculating aquaculture system is updated over time. When the updated risk status does not reach the preset alarm threshold and the risk evolution rate shows a continuous upward trend, the recirculating aquaculture system is determined to have entered a potential stress state, and an early control operation is triggered.

[0096] As an example, the advance control operation includes at least one of adjusting the feeding strategy, changing the water circulation flow rate, increasing the oxygenation intensity, and activating the water quality conditioning equipment in advance.

[0097] The risk status of the recirculating aquaculture system is updated over time based on the current risk status and the risk evolution rate. When the updated risk status does not reach the preset alarm threshold and the risk evolution rate shows a continuous upward trend, the recirculating aquaculture system is determined to have entered a potential stress state, and an early control operation is triggered.

[0098] For example, preemptive interventions include at least one of the following: adjusting feeding strategies (e.g., reducing the amount of feed per feeding or extending the feeding interval), altering water circulation flow (e.g., increasing the frequency of circulation pumps to accelerate water renewal), increasing aeration intensity (e.g., increasing the number of aerators or increasing the air supply), or activating water quality conditioning equipment in advance (e.g., activating backwashing of the biological filter in advance or adding microbial agents). It is understood that these interventions aim to mitigate potential instability in the system through slight intervention, preventing it from developing into severe water quality anomalies or mass stress events.

[0099] This invention introduces a joint modeling mechanism based on cloud platform-based behavioral characteristic time series and behavioral change structural characteristics into a recirculating aquaculture system. It uses a deep learning model to output the risk evolution rate that represents the rate of change of risk status over time. By combining the risk evolution trend, it can make early judgments and controls on potential stress states before the risk status reaches the alarm threshold. Thus, it can achieve forward-looking identification and refined management of the operational risks of the aquaculture system without relying on significant changes in water quality parameters.

[0100] Please see Figure 4 This invention also provides a cloud platform 200, comprising:

[0101] The behavioral feature time series construction module 201 is used to collect behavioral feature data of aquaculture objects during the operation of the recirculating aquaculture system, and construct a behavioral feature time series based on the behavioral feature data.

[0102] The behavior change structural feature extraction module 202 is used to extract behavior change structural features that characterize the behavior change pattern based on the behavior feature time series.

[0103] The risk evolution rate calculation module 203 is used to input the behavioral feature time series and its corresponding behavioral change structure features as input data into the deep learning model in the cloud platform, and the deep learning model outputs the risk evolution rate, which represents the rate of change of the risk state of the recirculating aquaculture system over time.

[0104] The risk status recursion and control triggering module 204 is used to recursively update the risk status of the recirculating aquaculture system based on the current risk status and the risk evolution rate. When the updated risk status does not reach the preset alarm threshold and the risk evolution rate shows a continuous upward trend, the recirculating aquaculture system is determined to have entered a potential stress state, and an early control operation is triggered.

[0105] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described in any of the preceding claims.

[0106] Please see Figure 5 The present invention also provides a computer-readable storage medium in which a computer program stored can be executed by a processor to implement the method as described in any of the preceding claims.

[0107] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A cloud platform-based pipe control method for recirculating aquaculture, characterized in that, The method includes the following steps: Step S1: During the operation of the recirculating aquaculture system, collect behavioral characteristic data of the cultured organisms and construct a behavioral characteristic time series based on the behavioral characteristic data; Step S2: Based on the behavioral feature time series, extract the behavioral change structural features that characterize the behavioral change pattern; Step S3: Input the time series of behavioral features and the corresponding structural features of behavioral changes as input data into the deep learning model in the cloud platform. The deep learning model outputs the risk evolution rate, which represents the rate of change of the risk state of the recirculating aquaculture system over time. Step S4: Based on the current risk status and the risk evolution rate, the risk status of the recirculating aquaculture system is updated over time. When the updated risk status does not reach the preset alarm threshold and the risk evolution rate shows a continuous upward trend, the recirculating aquaculture system is determined to have entered a potential stress state, and an early control operation is triggered.

2. The cloud-based recirculating aquaculture system management method according to claim 1, characterized in that: Prior to step S3, an anomaly-leading analysis of the behavior state of the aquaculture object is performed based on the behavioral characteristic time series, specifically including: Perform sliding time window analysis on the time series of the behavioral features to calculate the magnitude, direction, and stability index of the change of at least one behavioral feature within the current time window; The changes in the behavioral characteristics within the current time window are compared with the distribution of changes in the behavioral characteristics within the corresponding historical time period. When the changes in the behavioral characteristics deviate from the historical normal distribution and do not correspond to abnormal water quality parameters, it is determined that the current behavioral state has abnormal leading characteristics.

3. A cloud-based recirculating aquaculture system management method according to claim 1 or 2, characterized in that: The behavioral characteristic data includes one or more of the following: feeding response duration of the farmed animals, changes in group aggregation, mean swimming speed, fluctuation range of swimming speed, and changes in relative distance between individuals; The behavioral change structural features include at least one of the following: behavioral feature change trend features in the time dimension, periodic features, mutation point features, and stability disruption features.

4. The cloud-based recirculating aquaculture system management method according to claim 3, characterized in that: In step S3, the behavioral feature time series and the behavioral change structural features are fused according to the time alignment method, and then input into the deep learning model in the form of multi-channel time series data.

5. The cloud-based recirculating aquaculture system management method according to claim 1, characterized in that: The deep learning model includes: A behavioral feature time series encoding subnetwork is used to extract temporal features from the behavioral feature time series to obtain a temporal representation characterizing short-term behavioral perturbations; A behavior change structural feature encoding subnetwork is used to model the behavior change structural features to obtain a structural representation that characterizes the medium- and long-term behavior patterns and their stability changes; A behavior stability violation perception subnetwork is used to jointly model the temporal representation and the structural representation, and output a stability violation representation vector that represents the degree of collapse of the statistically stable structure in the behavioral feature time series. The risk evolution rate output module is used to output the risk evolution rate of the recirculating aquaculture system risk state over time based on the stability destruction characterization vector.

6. The cloud-based recirculating aquaculture system management method according to claim 5, characterized in that: The deep learning model also includes an attention bias submodule driven by anomaly-leading features, specifically used for: Receive abnormal leading features obtained based on the time series of the behavioral features, and perform time series modeling on the changes of the abnormal leading features within a continuous time window to generate attention bias parameters that are updated over time. The attention bias parameter is introduced into the attention calculation process of the behavior stability disruption perception subnetwork, so that the attention bias parameter participates in the calculation or normalization process of attention weights. In order to dynamically modulate the attention weight distribution corresponding to different behavior change structural features during the joint modeling of behavior feature time series and behavior change structural features by the behavior stability disruption perception subnetwork, the attention weight distribution corresponding to different behavior change structural features is dynamically modulated.

7. The cloud-based recirculating aquaculture system management method according to claim 6, characterized in that: The advance control operations include at least one of the following: adjusting the feeding strategy, changing the water circulation flow rate, increasing the oxygenation intensity, and activating the water quality conditioning equipment in advance.

8. A cloud platform, characterized in that, include: The behavioral feature time series construction module is used to collect behavioral feature data of aquaculture objects during the operation of the recirculating aquaculture system, and construct a behavioral feature time series based on the behavioral feature data. The behavior change structural feature extraction module is used to extract behavior change structural features that characterize behavior change patterns based on the behavior feature time series. The risk evolution rate calculation module is used to input the time series of the behavioral features and the corresponding structural features of behavioral changes as input data into the deep learning model in the cloud platform, and the deep learning model outputs the risk evolution rate, which represents the rate of change of the risk state of the recirculating aquaculture system over time. The risk status recursion and control triggering module is used to recursively update the risk status of the recirculating aquaculture system based on the current risk status and the risk evolution rate. When the updated risk status does not reach the preset alarm threshold and the risk evolution rate shows a continuous upward trend, the recirculating aquaculture system is determined to have entered a potential stress state, and an early control operation is triggered.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer program stored in the computer-readable storage medium can be executed by a processor to implement the method as described in any one of claims 1-7.

Images