Intelligent breeding method and system based on dynamic balance of water quality

Abstract

The invention discloses an intelligent breeding method and system based on water quality dynamic balance, and belongs to the technical field of aquaculture. The method comprises the steps that multi-source breeding data of aquaculture is collected; constructing a water quality dynamic association model based on the multi-source breeding data; collecting current aquaculture data of aquaculture in real time, inputting the current aquaculture data into the water quality dynamic correlation model, and resolving a water quality parameter evolution trend and a imbalance risk index in a future time period; when the imbalance risk index is greater than a preset value, generating a cooperative regulation and control strategy by using the imbalance risk index; and generating a control instruction according to the cooperative control strategy, and determining the execution time of the control instruction based on the water quality dynamic correlation model. According to the invention, prospective and precise cooperative regulation and control of water quality are realized, the unbalance risk is effectively avoided, and the intelligent level and benefit of breeding are improved.

Description

Technical Field

[0001] This invention relates to the field of aquaculture technology, and more specifically to an intelligent aquaculture method and system based on dynamic water quality balance. Background Technology

[0002] Aquaculture is an important component of the agricultural economy, and its sustainable development highly depends on the stability and health of the aquatic environment. Water quality is a key factor determining the success or failure of aquaculture. Core parameters such as dissolved oxygen, ammonia nitrogen, nitrite, pH, and temperature exhibit complex dynamic coupling relationships and are influenced by multiple factors, including the metabolism of farmed organisms, feed intake, microbial activity, and environmental disturbances. Maintaining a dynamic balance in water quality is crucial for ensuring the healthy growth of farmed organisms, improving feed utilization, and reducing disease risks.

[0003] Currently, water quality management and control strategies in large-scale aquaculture are mostly based on single-parameter independent control, neglecting the complex interactions between water quality parameters. For example, oxygenation may affect pH and microbial nitrification processes, and a single regulatory action may trigger new imbalances. Furthermore, there is a time delay from detection to equipment execution, and the equipment's effectiveness in the water body also requires time to produce results. This system response lag makes it difficult to accurately match regulation with the real-time dynamic evolution of water quality. In addition, daily production operations (especially feeding) are one of the largest sources of water quality disturbance, but in existing management systems, production planning and water quality control are usually separated, making it impossible to proactively assess and optimize the potential impact of production activities on water quality balance.

[0004] In summary, how to provide an aquaculture management method and system that can deeply integrate multi-source information, dynamically simulate the correlation and evolution between water quality parameters, and intelligently coordinate and proactively regulate based on future risks is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides an intelligent aquaculture method and system based on dynamic water quality balance, which is an intelligent closed-loop management system capable of simulating, predicting and actively maintaining the dynamic water quality balance of an aquaculture system, thereby significantly improving the scientific nature, safety and production efficiency of the aquaculture process.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: On the one hand, this invention provides an intelligent aquaculture method based on dynamic water quality balance, comprising: Collect multi-source aquaculture data; A dynamic correlation model of water quality is constructed based on the aforementioned multi-source aquaculture data; Real-time aquaculture data is collected and input into the water quality dynamic correlation model to calculate the evolution trend of water quality parameters and the risk index of imbalance in future time periods. When the imbalance risk index is greater than the preset value, a coordinated control strategy is generated based on the imbalance risk index. Control commands are generated based on the collaborative control strategy, and the execution time of the control commands is determined based on the water quality dynamic correlation model.

[0007] Preferably, the method further includes inputting the feeding plan into a water quality dynamic correlation model to simulate water quality evolution, and optimizing the feeding plan when the imbalance risk index of the simulation results exceeds the standard.

[0008] Preferably, the water quality dynamic correlation model includes: establishing a metabolic load prediction sub-model, a multi-parameter coupling sub-model, and an equipment response delay sub-model based on the multi-source aquaculture data.

[0009] Preferably, a metabolic load prediction sub-model is constructed based on the multi-source aquaculture data, including: The metabolic load prediction sub-model includes a feature extraction part, a feature concatenation part, and a load prediction part. The feature extraction part consists of at least two different network models, with the input of each network model serving as the input port of the load prediction model, used to extract metabolic load prediction features from different dimensions. The feature concatenation part is used to concatenate the metabolic load prediction features extracted by different network models to obtain concatenated metabolic load prediction features, which are then input into the load prediction part. The load prediction part is used to output the final metabolic load prediction result.

[0010] Preferably, a multi-parameter coupled sub-model is constructed based on the multi-source aquaculture data, including: Using water quality parameters as nodes in a graph structure, denoted as node set V; based on historical aquaculture data, the dynamic influence relationships between various water quality parameters are identified, and an adjacency matrix A is constructed, where elements A... ij This represents the influence intensity coefficient of the i-th parameter on the j-th parameter, when A ij When ≠0, a directed edge is established between the corresponding nodes to form a water quality parameter association graph G=(V,E); For each node v i ∈V, integrating current monitoring values, rate of change, historical time-series statistical characteristics, and correlation with the metabolic load prediction sub-model, to construct a multi-dimensional node initial feature vector. ; Using a graph attention network for multi-layer message passing, the feature update formula for the (l+1)th layer node is:

[0011] Where N(i) is the node vi The neighborhood group, W represents the weight coefficients dynamically calculated based on the attention mechanism. l Let σ be the trainable parameter matrix, and σ be the activation function. After L layers of aggregation, the node features are aggregated into a deep-coupled perceptual feature vector through the readout layer:

[0012] in, For mean aggregation, For attention-weighted aggregation, W a This is the attention weight matrix; Based on real-time data on aquaculture species, growth stage, and fish carrying capacity, the edge weights of the adjacency matrix A are dynamically updated through a gating mechanism. The deep coupling sensing feature vector H is input into the fully connected layer, and the multi-parameter coupling state index CCS is output.

[0013] Preferably, real-time aquaculture data is collected and input into the water quality dynamic correlation model to calculate the evolution trend of water quality parameters and the risk index of imbalance in future time periods, specifically including: Simultaneously acquire current aquaculture data, and unify aquaculture data with different sampling frequencies to a preset fixed time step Δt through interpolation or resampling methods, constructing a unified input feature vector for the current moment. ; The predicted trajectory of water quality evolution trend in the future period is obtained by multi-step iterative solution based on the metabolic load prediction sub-model and the multi-parameter coupled sub-model. Based on the predicted water quality evolution trajectory over future periods, the risk index R is calculated:

[0014] in, It is the predicted value of parameter i at time t. is the preset safety threshold range; Penalty is the penalty function when the predicted value exceeds the range; Var(ΔDO) is the variance of the dissolved oxygen change rate during the prediction period; Trend(TAN) is the intensity of the cumulative upward trend of total ammonia nitrogen during the entire prediction period; w1, w2, w3 are the weight coefficients of each sub-item.

[0015] Preferably, the predicted trajectory of water quality evolution trend in the future period is obtained through multi-step iterative calculation based on the metabolic load prediction sub-model and the multi-parameter coupled sub-model, including: Set the total prediction duration T pred Let the current solution time t=T0, and initialize the future water quality parameter evolution trajectory sequence as empty; For the solution time t: a) The eigenvector X corresponding to time t t Input the metabolic load prediction sub-model to predict the metabolic waste production rate MP within the time period [t, t+Δt]. t ; b) The eigenvector X corresponding to time t t and generation rate MP t Input the multi-parameter coupled sub-model to obtain the multi-parameter coupled state exponent CCS at time t. t Based on the dynamic relationships built into the multi-parameter coupled sub-model, the changes ΔP of each core water quality parameter within the time period [t, t+Δt] are deduced. t ; c) Based on the change ΔP t The predicted water quality parameters at time t+Δt are updated, and these predicted water quality parameters, along with the predicted equipment status data, are then updated to the feature vector X. t+Δt ; Let t = t + Δt, then update the feature vector X. t As the new input, repeat the above steps until t = T0 + T. pred Thus, we obtain the transition from T0 to T0+T. pred The complete water quality parameter evolution prediction trajectory.

[0016] On the other hand, the present invention provides an intelligent aquaculture system based on dynamic water quality balance, comprising: The data acquisition module is used to collect multi-source aquaculture data. The model building module is used to build a dynamic correlation model of water quality based on the multi-source aquaculture data; The evolution module is used to collect current aquaculture data in real time, input the current aquaculture data into the water quality dynamic correlation model, and calculate the evolution trend of water quality parameters and the risk index of imbalance in future time periods. The judgment module is used to generate a coordinated control strategy based on the imbalance risk index when the imbalance risk index is greater than a preset value. The control module is used to generate control commands according to the collaborative control strategy and determine the execution time of the control commands based on the water quality dynamic correlation model.

[0017] Preferably, the system further includes: inputting the feeding plan into a water quality dynamic correlation model to simulate water quality evolution; and optimizing the feeding plan when the imbalance risk index of the simulation results exceeds the standard.

[0018] As can be seen from the above technical solution, compared with the prior art, the present invention discloses an intelligent aquaculture method and system based on dynamic water quality balance. By constructing a dynamic water quality correlation model that integrates multiple sub-models, it can accurately calculate the evolution trend of water quality parameters and the risk index of imbalance in future periods. It can generate targeted collaborative control strategies when the risk of water quality imbalance exceeds the standard, and determine the advance of control instructions based on the model. At the same time, the feeding plan can be incorporated into the water quality evolution simulation system and optimized and adjusted to achieve forward-looking, precise, and intelligent collaborative control of aquaculture water quality, effectively avoid the risk of water quality imbalance, continuously maintain the dynamic balance of aquaculture water quality, improve the level of refined and intelligent management of aquaculture, ensure the stable and efficient operation of the aquaculture system, reduce aquaculture production losses, and improve the overall benefits of aquaculture. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the process provided by the present invention.

[0021] Figure 2 This is a schematic diagram of the structure provided by the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] This invention discloses an intelligent aquaculture method based on dynamic water quality balance, such as... Figure 1 As shown, it includes: Collect multi-source aquaculture data; the multi-source aquaculture data includes: water quality monitoring data, aquaculture species data, feeding data, and equipment status data; A dynamic correlation model of water quality is constructed based on the aforementioned multi-source aquaculture data; Real-time aquaculture data is collected and input into the water quality dynamic correlation model to calculate the evolution trend of water quality parameters and the risk index of imbalance in future time periods. When the imbalance risk index is greater than the preset value, a coordinated control strategy is generated based on the imbalance risk index. Control commands are generated based on the collaborative control strategy, and the execution time of the control commands is determined based on the water quality dynamic correlation model.

[0024] In another embodiment, the method further includes inputting the feeding plan into a water quality dynamic correlation model to simulate water quality evolution, and optimizing the feeding plan when the imbalance risk index of the simulation results exceeds the standard.

[0025] Specifically, the water quality dynamic correlation model includes: a metabolic load prediction sub-model, a multi-parameter coupling sub-model, and an equipment response delay sub-model established based on the multi-source aquaculture data.

[0026] Furthermore, a metabolic load prediction sub-model is constructed based on the aforementioned multi-source aquaculture data, including: The metabolic load prediction sub-model includes a feature extraction part, a feature concatenation part, and a load prediction part. The feature extraction part consists of at least two different network models, with the input of each network model serving as the input port of the load prediction model, used to extract metabolic load prediction features from different dimensions. The feature concatenation part is used to concatenate the metabolic load prediction features extracted by different network models to obtain concatenated metabolic load prediction features, which are then input into the load prediction part. The load prediction part is used to output the final metabolic load prediction result.

[0027] The specific process of training the load prediction sub-model is as follows: Feature extraction training: Different types of historical aquaculture data are input into the network models of the feature extraction part for training, and feature expressions of each model related to metabolic load prediction are obtained; The training data input into the first network model consists of local time window sequence data about feeding events, feed types, and aquaculture operations. The training data input to the second network model consists of long-term time-series data on water temperature, dissolved oxygen, and ammonia nitrogen concentration, as well as the growth stage codes of the aquaculture species. Feature concatenation: The feature representations extracted from the first network model and the second network model are input into the feature concatenation part, and the concatenation method is used to concatenate them to obtain concatenated features that integrate local operational features and long-term water quality dynamic features; Load prediction training: The spliced ​​features are input into the load prediction part for training. The load prediction part is a feedforward neural network. Through training, it can output a predicted value of the metabolic waste generation rate in the future period based on the fused features. In actual aquaculture, the real-time collected current feeding data and aquaculture operation data are input into the trained first network model, and the current water quality time series data and aquaculture status are input into the trained second network model. The load prediction part of the load prediction hybrid model outputs the real-time metabolic load prediction result as the output of the metabolic load prediction sub-model.

[0028] Furthermore, a multi-parameter coupled sub-model is constructed based on the aforementioned multi-source aquaculture data, including: Using water quality parameters as nodes in a graph structure, denoted as node set V; based on historical aquaculture data, the dynamic influence relationships between various water quality parameters are identified, and an adjacency matrix A is constructed, where elements A... ij This represents the influence intensity coefficient of the i-th parameter on the j-th parameter, when A ij When the value is not equal to 0, directed edges are established between the corresponding nodes to form a water quality parameter association graph G=(V,E); where the water quality parameters include pH value, dissolved oxygen, total ammonia nitrogen, nitrite, nitrate, total nitrogen, biochemical oxygen demand, and chemical oxygen demand; For each node v i ∈V, integrating current monitoring values, rate of change, historical time-series statistical characteristics, and correlation with the metabolic load prediction sub-model, to construct a multi-dimensional node initial feature vector. ; Using a graph attention network for multi-layer message passing, the feature update formula for the (l+1)th layer node is:

[0029] Where N(i) is the node v i The neighborhood group, W represents the weight coefficients dynamically calculated based on the attention mechanism. l Let σ be the trainable parameter matrix, and σ be the activation function. After L layers of aggregation, the node features are aggregated into a deep-coupled perceptual feature vector through the readout layer:

[0030] in, For mean aggregation, For attention-weighted aggregation, W a This is the attention weight matrix; Based on real-time data on aquaculture species, growth stage, and fish carrying capacity, the edge weights of the adjacency matrix A are dynamically updated through a gating mechanism. The deep coupling sensing feature vector H is input into the fully connected layer, and the multi-parameter coupling state index CCS is output.

[0031] In another embodiment, current aquaculture data is collected in real time, and this data is input into the water quality dynamic correlation model to calculate the evolution trend of water quality parameters and the risk index of imbalance in future time periods. Specifically, this includes: Simultaneously acquire current aquaculture data, and unify aquaculture data with different sampling frequencies to a preset fixed time step Δt through interpolation or resampling methods, constructing a unified input feature vector for the current moment. ; The predicted trajectory of water quality evolution trend in the future period is obtained by multi-step iterative solution based on the metabolic load prediction sub-model and the multi-parameter coupled sub-model. Based on the predicted water quality evolution trajectory over future periods, the risk index R is calculated:

[0032] in, It is the predicted value of parameter i at time t. is the preset safety threshold range; Penalty is the penalty function when the predicted value exceeds the range; Var(ΔDO) is the variance of the dissolved oxygen change rate during the prediction period, used to assess system stability; Trend(TAN) is the cumulative upward trend strength of total ammonia nitrogen throughout the prediction period; w1, w2, w3 are the weight coefficients of each sub-item. Among them, the imbalance risk index R comprehensively reflects the degree of risk of water quality exceeding standards, system instability, and pollution accumulation during the forecast period.

[0033] In another embodiment, a multi-step iterative solution based on a metabolic load prediction sub-model and a multi-parameter coupled sub-model is used to obtain the predicted trajectory of water quality evolution trend in future periods, including: Set the total prediction duration T pred Let the current solution time t=T0, and initialize the future water quality parameter evolution trajectory sequence as empty; For the solution time t: a) The eigenvector X corresponding to time t t Input the metabolic load prediction sub-model to predict the metabolic waste production rate MP within the time period [t, t+Δt]. t ; b) The eigenvector X corresponding to time t t and generation rate MP t Input the multi-parameter coupled sub-model to obtain the multi-parameter coupled state exponent CCS at time t. t Based on the dynamic relationships built into the multi-parameter coupled sub-model, the changes ΔP of each core water quality parameter within the time period [t, t+Δt] are deduced. t ; c) Based on the change ΔP tThe predicted water quality parameters at time t+Δt are updated, and these predicted water quality parameters, along with the predicted equipment status data, are then updated to the feature vector X. t+Δt ; Let t = t + Δt, then update the feature vector X. t As the new input, repeat the above steps until t = T0 + T. pred Thus, we obtain the transition from T0 to T0+T. pred The complete water quality parameter evolution prediction trajectory.

[0034] In another embodiment, when the imbalance risk index is greater than a preset value, a coordinated control strategy is generated based on the imbalance risk index, specifically including: Based on the magnitude and composition of the imbalance risk index R, the risk level is divided into three intervals: mild imbalance (R∈[R1,R2)), moderate imbalance (R∈[R2,R3)), and severe imbalance (R≥R3), where R1, R2, and R3 are preset risk thresholds. Simultaneously, the contribution of each component in the risk index R is analyzed to identify the dominant risk factors. like The largest contribution indicates the risk of one or more parameters exceeding the standard, and the control target is to reduce specific water quality parameters to a safe range; like The largest contribution indicates a decrease in the system's buffering capacity, and the control objective is to enhance water body stability; like The largest contribution indicates a trend of metabolic waste accumulation, and the control target is to accelerate pollutant removal or reduce source load.

[0035] Based on the equipment response delay sub-model in the water quality dynamic correlation model, a strategy space S for available controllable equipment is established. This strategy space includes aeration equipment (micro-pore aerators, paddlewheel aerators, pure oxygen injection devices), circulating water treatment equipment (biological filters, protein skimmers, ultraviolet sterilizers), water exchange equipment (inlet and outlet pumps, emergency water exchange systems), and feeding control (pause feeding, reduce feeding amount, adjust feeding time). Each control strategy s∈S is associated with its equipment response characteristic function D. s (t) represents the time delay characteristics and intensity curve from the execution of the instruction to the generation of a stable control effect.

[0036] With the optimization objectives of minimizing control costs, minimizing stress disturbances to cultured organisms, and maximizing the efficiency of risk index reduction, a multi-objective optimization function is constructed:

[0037] Where C(s) represents the energy consumption and operating cost of strategy s, Δt exec(s) represents the equipment startup and stabilization time required for strategy execution, ΔR(s) represents the predicted decrease in risk index after strategy s is implemented, which is obtained by strategy simulation and deduction using the water quality dynamic correlation model; λ1, λ2, λ3 are weighting coefficients; Using heuristic search or reinforcement learning policy networks, Pareto optimal solutions are searched in the policy space S to generate a set of collaborative control policies that include the primary control policy and alternative control policies. ; For the selected main control strategy The response delay feature function is obtained based on the device response delay sub-model. Determine the policy effective delay time τ d With the regulation intensity reaching its peak time τ p ; By combining the predicted trajectory of water quality parameter evolution over future periods, the key time point t is identified as the first time the risk index R exceeds the preset threshold. c Calculate the command advance Δt a :

[0038] Among them, t t For the current moment, τ r k represents the inherent time overhead of system communication and device startup. s ≥1 is a safety factor used to compensate for model prediction errors and equipment response fluctuations; Determine the absolute execution time t of the control command e =t c -Δt a To ensure that the effects of regulation are fully established before the critical moment of risk; After the control command is executed, the actual water quality parameters and equipment operating status are continuously monitored, and the deviation between the actual control effect and the predicted effect is calculated. If | |> t This triggers a strategy reassessment, corrects the parameters of the equipment response delay sub-model based on real-time deviations, and dynamically adjusts the collaborative control strategy for subsequent periods.

[0039] In another embodiment, the construction of the device response latency sub-model includes: Historical control operation logs and corresponding water quality response data were collected to establish an equipment-response dataset. For different types of control equipment, a system identification method or a Long Short-Term Memory (LSTM) network was used to establish an input-output dynamic model to characterize the transfer function relationship between the control command intensity u(t) and the water quality parameter change response y(t). This includes a pure delay element e -τs With inertial elements The model parameters are continuously updated through online learning to adapt to the drift in response characteristics caused by equipment aging, changes in biofilm activity, etc.

[0040] In another embodiment, the feeding plan is input into a water quality dynamic correlation model for water quality evolution simulation. When the equilibrium imbalance risk index of the simulation results exceeds the standard, the feeding plan is optimized, specifically including: Obtain the feeding plan to be evaluated , where t j To plan the baiting time, q j For feeding amount, f j Code the feed type; The feeding plan F is used as an external input event sequence and input into the metabolic load prediction sub-model to predict the metabolic load pulse characteristics triggered by each feeding event; Based on a multi-step iterative calculation process, the future water quality evolution trend and the equilibrium imbalance risk index R are extrapolated after the implementation of the feeding plan. F ; If R F ≤R s The feeding plan was deemed feasible and was carried out as originally planned. If R F >R s This triggers a feeding plan optimization process: with the dual objectives of maximizing the total feeding amount (ensuring growth needs) and minimizing the risk index (ensuring water quality safety), a genetic algorithm or particle swarm optimization algorithm is used to optimize the feeding time t. j Feeding amount q j Optimize the feeding plan to generate an optimized feeding plan. , making Maximize; optimize the process by taking into account the feeding rhythm and digestion cycle constraints of farmed organisms, and avoid excessive scattered feeding that leads to a decline in growth efficiency.

[0041] On the other hand, the present invention provides an intelligent aquaculture system based on dynamic water quality balance, such as... Figure 2 As shown, it includes: The data acquisition module is used to collect multi-source aquaculture data. The model building module is used to build a dynamic correlation model of water quality based on the multi-source aquaculture data; The evolution module is used to collect current aquaculture data in real time, input the current aquaculture data into the water quality dynamic correlation model, and calculate the evolution trend of water quality parameters and the risk index of imbalance in future time periods. The judgment module is used to generate a coordinated control strategy based on the imbalance risk index when the imbalance risk index is greater than a preset value. The control module is used to generate control commands according to the collaborative control strategy and determine the execution time of the control commands based on the water quality dynamic correlation model.

[0042] Furthermore, the system provided by this invention also includes: inputting the feeding plan into a water quality dynamic correlation model to simulate water quality evolution; and optimizing the feeding plan when the imbalance risk index of the simulation results exceeds the standard.

[0043] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0044] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A smart aquaculture method based on dynamic water quality balance, characterized in that, include: Collect multi-source aquaculture data; A dynamic correlation model of water quality is constructed based on the aforementioned multi-source aquaculture data; Real-time aquaculture data is collected and input into the water quality dynamic correlation model to calculate the evolution trend of water quality parameters and the risk index of imbalance in future time periods. When the imbalance risk index is greater than the preset value, a coordinated control strategy is generated based on the imbalance risk index. Control commands are generated based on the collaborative control strategy, and the execution time of the control commands is determined based on the water quality dynamic correlation model.

2. The intelligent aquaculture method based on dynamic water quality balance according to claim 1, characterized in that, The method also includes inputting the feeding plan into a water quality dynamic correlation model to simulate water quality evolution. When the imbalance risk index of the simulation results exceeds the standard, the feeding plan is optimized.

3. The intelligent aquaculture method based on dynamic water quality balance according to claim 1, characterized in that, The water quality dynamic correlation model includes: a metabolic load prediction sub-model, a multi-parameter coupling sub-model, and an equipment response delay sub-model based on the multi-source aquaculture data.

4. The intelligent aquaculture method based on dynamic water quality balance according to claim 3, characterized in that, A metabolic load prediction sub-model is constructed based on the aforementioned multi-source aquaculture data, including: The metabolic load prediction sub-model includes a feature extraction part, a feature concatenation part, and a load prediction part. The feature extraction part consists of at least two different network models, with the input of each network model serving as the input port of the load prediction model, used to extract metabolic load prediction features from different dimensions. The feature concatenation part is used to concatenate the metabolic load prediction features extracted by different network models to obtain concatenated metabolic load prediction features, which are then input into the load prediction part. The load prediction part is used to output the final metabolic load prediction result.

5. The intelligent aquaculture method based on dynamic water quality balance according to claim 3, characterized in that, A multi-parameter coupled sub-model is constructed based on the aforementioned multi-source aquaculture data, including: Using water quality parameters as nodes in a graph structure, denoted as node set V; based on historical aquaculture data, the dynamic influence relationships between various water quality parameters are identified, and an adjacency matrix A is constructed, where elements A... ij This represents the influence intensity coefficient of the i-th parameter on the j-th parameter, when A ij When ≠0, a directed edge is established between the corresponding nodes to form a water quality parameter association graph G=(V,E); For each node v i ∈V, integrating current monitoring values, rate of change, historical time-series statistical characteristics, and correlation with the metabolic load prediction sub-model, to construct a multi-dimensional node initial feature vector. ; Using a graph attention network for multi-layer message passing, the feature update formula for the (l+1)th layer node is: Where N(i) is the node v i The neighborhood group, W represents the weight coefficients dynamically calculated based on the attention mechanism. l Let σ be the trainable parameter matrix, and σ be the activation function. After L layers of aggregation, the node features are aggregated into a deep-coupled perceptual feature vector through the readout layer: in, For mean aggregation, For attention-weighted aggregation, W a This is the attention weight matrix; Based on real-time data on aquaculture species, growth stage, and fish carrying capacity, the edge weights of the adjacency matrix A are dynamically updated through a gating mechanism. The deep coupling sensing feature vector H is input into the fully connected layer, and the multi-parameter coupling state index CCS is output.

6. The intelligent aquaculture method based on dynamic water quality balance according to claim 2, characterized in that, Real-time collection of current aquaculture data, input of this data into the aforementioned water quality dynamic correlation model, and calculation of future water quality parameter evolution trends and balance imbalance risk indices, specifically including: Simultaneously acquire current aquaculture data, and unify aquaculture data with different sampling frequencies to a preset fixed time step Δt through interpolation or resampling methods, constructing a unified input feature vector for the current moment. ; The predicted trajectory of water quality evolution trend in the future period is obtained by multi-step iterative solution based on the metabolic load prediction sub-model and the multi-parameter coupled sub-model. Based on the predicted water quality evolution trajectory over future periods, the risk index R is calculated: in, It is the predicted value of parameter i at time t. is the preset safety threshold range; Penalty is the penalty function when the predicted value exceeds the range; Var(ΔDO) is the variance of the dissolved oxygen change rate during the prediction period; Trend(TAN) is the intensity of the cumulative upward trend of total ammonia nitrogen during the entire prediction period; w1, w2, w3 are the weight coefficients of each sub-item.

7. The intelligent aquaculture method based on dynamic water quality balance according to claim 6, characterized in that, Based on the metabolic load prediction sub-model and the multi-parameter coupled sub-model, a multi-step iterative solution is performed to obtain the predicted trajectory of water quality evolution trend in future periods, including: Set the total prediction duration T pred Let the current solution time t=T0, and initialize the future water quality parameter evolution trajectory sequence as empty; For the solution time t: a) The eigenvector X corresponding to time t t Input the metabolic load prediction sub-model to predict the metabolic waste production rate MP within the time period [t, t+Δt]. t ; b) The eigenvector X corresponding to time t t and generation rate MP t Input the multi-parameter coupled sub-model to obtain the multi-parameter coupled state exponent CCS at time t. t Based on the dynamic relationships built into the multi-parameter coupled sub-model, the changes ΔP of each core water quality parameter within the time period [t, t+Δt] are deduced. t ; c) Based on the change ΔP t The predicted water quality parameters at time t+Δt are updated, and these predicted water quality parameters, along with the predicted equipment status data, are then updated to the feature vector X. t+Δt ; Let t = t + Δt, then update the feature vector X. t As the new input, repeat the above steps until t = T0 + T. pred Thus, we obtain the transition from T0 to T0+T. pred The complete water quality parameter evolution prediction trajectory.

8. An intelligent aquaculture system based on dynamic water quality balance, characterized in that, include: The data acquisition module is used to collect multi-source aquaculture data. The model building module is used to build a dynamic correlation model of water quality based on the multi-source aquaculture data; The evolution module is used to collect current aquaculture data in real time, input the current aquaculture data into the water quality dynamic correlation model, and calculate the evolution trend of water quality parameters and the risk index of imbalance in future time periods. The judgment module is used to generate a coordinated control strategy based on the imbalance risk index when the imbalance risk index is greater than a preset value. The control module is used to generate control commands according to the collaborative control strategy and determine the execution time of the control commands based on the water quality dynamic correlation model.

9. The intelligent aquaculture system based on dynamic water quality balance according to claim 8, characterized in that, Also includes: The feeding plan is input into the water quality dynamic correlation model to simulate water quality evolution. When the imbalance risk index of the simulation results exceeds the standard, the feeding plan is optimized.

Images