Artificial intelligence driven multi-nutrient level integrated aquaculture system

Abstract

The application discloses an artificial intelligence driven multi-nutrition level comprehensive breeding system and relates to the field of intelligent agriculture.The system comprises an ecological perception module, an ecological topology module, a decision control module, an execution control module and a risk response module.The ecological perception module is used for collecting multi-dimensional ecological parameters in a breeding environment and constructing an ecological data set.The ecological topology module is used for constructing a weighted directed graph topology structure.The decision control module is used for constructing an ecological regulation and control model based on a competitive reward function to generate a dynamic adjustment strategy for ecological regulation and control.The execution control module is used for controlling breeding related execution equipment.The risk response module is used for recording current operating state parameters and activating a standby biological reaction chain structure to reconfigure the ecological regulation and control model.The application is supported by the national key research and development plan (2024YFD2401801).

Description

TECHNICAL FIELD

[0001] The present application relates to intelligent agriculture, in particular, to an artificial intelligence driven multi-nutrient level comprehensive breeding system. BACKGROUND

[0002] With the development of intelligent agriculture and artificial intelligence technology, multi-nutrient level water state breeding systems have gradually become an important direction of efficient and sustainable breeding. Such systems introduce multi-level nutrient organisms such as different feeding breeding animals, aquatic plants and benthic animals in the same ecological space, form stable food chain structure and material circulation channel, and have significant advantages in improving resource utilization rate and reducing pollution emissions.

[0003] The multi-nutrient level ecological breeding mode refers to scientifically matching multi-nutrient level aquatic animals and plants in the same breeding water area, utilizing the characteristics of complementary breeding niches to achieve the purpose of stable water quality, recycling of nutrient substances, ecological disease prevention, quality safety improvement, breeding efficiency improvement, and reduction of breeding waste emissions.

[0004] However, the existing multi-nutrient level comprehensive breeding system relies on limited physicochemical indicators when in use, lacks fusion perception of microbial communities, electrochemical signals and multi-spectral characteristics, and is difficult to comprehensively reflect the dynamic changes of the ecological system, making the existing multi-nutrient level comprehensive breeding system have single ecological data acquisition dimension and poor timeliness. At the same time, the multi-nutrient level breeding system often ignores the complex nonlinear interactions between microorganisms, plants and animals at each level, lacks a modeling framework that can express time-varying dependence and feedback cycle structure, resulting in high prediction error of system material conversion efficiency, delayed or even ineffective control strategies.

[0005] At present, there is no effective solution to the problems in the related art. SUMMARY

[0006] In view of the problems in the related art, the present application proposes an artificial intelligence driven multi-nutrient level comprehensive breeding system to overcome the above technical problems existing in the prior art.

[0007] In order to achieve the above purpose, the specific technical solutions adopted by the present application are as follows:

[0008] The artificial intelligence driven multi-nutrient level comprehensive breeding system comprises an ecological perception module, an ecological topology module, a decision control module, an execution control module and a risk response module.

[0009] The ecological perception module is configured to deploy a biological electrode array, a distributed optical fiber sensor and a multi-spectral imaging device, collect multi-dimensional ecological parameters in a breeding environment in real time, and construct an ecological data set with time synchronization characteristics.

[0010] The ecological topology module is configured to construct a weighted directed graph topology according to the ecological data set, optimize the weighted directed graph topology by using a quantum bit encoding parameter, and enhance population diversity by using a Hadamard gate operation.

[0011] The decision control module is configured to set a competitive reward function, construct an ecological regulation and control model based on the competitive reward function, and generate a dynamic adjustment strategy for ecological regulation and control according to the ecological regulation and control model.

[0012] The execution control module is configured to control breeding-related execution equipment according to the dynamic adjustment strategy.

[0013] The risk response module is configured to, when predicting that the collapse probability of the ecological system exceeds a preset threshold, trigger the blockchain storage module to record current operating state parameters, activate a standby biological reaction chain structure, automatically inject microbial preparations, and reconfigure the ecological regulation and control model by using a transfer learning algorithm.

[0014] As a preferred solution, the ecological perception module includes a biological electrode module, an optical fiber sensing module, a spectral imaging module and a parameter data module.

[0015] The biological electrode module is configured to perceive electrochemical response information of a microbial community in water in a breeding farm by using the deployed biological electrode array, and generate corresponding bioactive electrical signal data.

[0016] The optical fiber sensing module is configured to collect breeding physical parameters and breeding chemical parameters of temperature, dissolved oxygen, pH value and turbidity in water in the breeding farm by using the set distributed optical fiber sensor, and construct time series ecological data based on the breeding physical parameters and the breeding chemical parameters.

[0017] The spectral imaging module is configured to collect image information data of the water in the breeding farm by using visible light, near-infrared and ultraviolet bands through the installed multi-spectral imaging device.

[0018] The parameter data module is configured to aggregate the bioactive electrical signal data, the time series ecological data and the image information data to obtain multi-dimensional ecological parameters, and generate the ecological data set based on the time sequence order and the multi-dimensional ecological parameters.

[0019] As a preferred solution, the parameter data module includes a data cleaning module, a feature extraction module and a standard processing module.

[0020] The data cleaning module is configured to detect outliers in the received bioactive electrical signal data, time series ecological data and image information data, and perform interpolation processing on missing values to obtain multi-dimensional ecological parameters.

[0021] The feature extraction module is configured to extract time sequence features from the obtained multi-dimensional ecological parameters using a time sequence sliding window mechanism.

[0022] The standard processing module is configured to normalize the multi-dimensional ecological parameters, and generate an ecological data set based on a unified timestamp and the normalized multi-dimensional ecological parameters.

[0023] As a preferred solution, the ecological topology module includes a topology construction module, an encoding optimization module and a topology update module.

[0024] The topology construction module is configured to divide the multi-dimensional ecological parameters into different trophic level parameters, obtain a hierarchical correlation relationship, and construct a weighted directed graph topology structure according to the trophic level parameters and the hierarchical correlation relationship.

[0025] The encoding optimization module is configured to perform parameterized conversion on the weighted directed graph topology structure through a quantum bit encoding method, optimize the converted weighted directed graph topology structure, and enhance the population diversity of the optimization process through a Hadamard gate operation.

[0026] The topology update module is configured to dynamically update parameters of the collected ecological data, and simultaneously optimize and adjust the node weight and edge weight of the weighted directed graph based on the dynamically updated ecological data parameters, and store the optimized and adjusted characteristic parameters.

[0027] As a preferred solution, the topology construction module includes a hierarchical division module, an association identification module and a structure generation module.

[0028] The hierarchical division module is configured to identify and extract trophic feature parameter sets of farmed animals, microorganisms, plankton, aquatic plants and benthic organisms in the multi-dimensional ecological parameters, and perform trophic level division based on the trophic feature parameter sets.

[0029] The association identification module is configured to analyze time sequence correlation parameters and covariation rule parameters between each level of the trophic levels, and construct logical association parameters between each level of the trophic levels based on the time sequence correlation parameters and the covariation rule parameters.

[0030] The structure generation module is configured to construct a weighted directed graph topology structure according to the trophic level division result and the logical association parameters, and define the ecological factors corresponding to each node, the weight representation parameter impact intensity and directionality of the edges.

[0031] As a preferred solution, the encoding optimization module comprises: a quantum encoding module, a parameter optimization module and a population expansion module.

[0032] The quantum encoding module is configured to encode the weighted directed graph topology by using quantum bits, and map the node attributes and edge weights of the weighted directed graph topology to an adjustable quantum bit sequence.

[0033] The parameter optimization module is configured to iteratively optimize the adjustable quantum bit sequence by using a quantum interference and measurement mechanism, and retain the encoding solution in the adjustable quantum bit sequence after the iterative optimization.

[0034] The population expansion module is configured to introduce a Hadamard gate operation and a quantum state superposition mechanism during the iterative optimization of the adjustable quantum bit sequence, increase the population diversity of candidate solutions, and prevent falling into a local optimum.

[0035] As a preferred solution, the decision control module comprises: a reward function module, an intelligent training module and a strategy generation module.

[0036] The reward function module is configured to set a competitive reward function according to the breeding target, and the reward function comprises a dissolved oxygen stability, a feed conversion rate, a biological growth efficiency and an energy consumption control index.

[0037] The intelligent training module is configured to construct an ecological regulation and control model based on the competitive reward function, and verify and optimize the ecological regulation and control model.

[0038] The strategy generation module is configured to generate a dynamic adjustment strategy according to the verified and optimized ecological regulation and control model, and output a regulation and control instruction for controlling an execution device.

[0039] As a preferred solution, the intelligent training module comprises: a model construction module, a training optimization module and a verification evaluation module.

[0040] The model construction module is configured to construct an ecological regulation and control model based on the competitive reward function, and initialize strategy network parameters and value function network parameters of the ecological regulation and control model.

[0041] The training optimization module is configured to alternately train the ecological regulation and control model by using an ecological data set, and optimize and adjust the strategy network parameters and the value function network parameters of the ecological regulation and control model.

[0042] The verification evaluation module is configured to verify and evaluate the optimized ecological regulation and control model, and generate an evaluation report.

[0043] As a preferred solution, the execution control module comprises: a device interface module, a regulation and control execution module and a feedback receiving module.

[0044] The device interface module is configured to be connected with the aquaculture device and to convert the regulation and control instructions in the dynamic adjustment strategy into a control protocol compatible with the aquaculture device.

[0045] The regulation and control execution module is configured to control the connected aquaculture device to adjust according to the output control protocol.

[0046] The feedback receiving module is configured to compare the multi-dimensional ecological parameter data collected by the ecological perception module with the results of the executed control protocol in real time, and optimize the ecological regulation and control model according to the comparison results.

[0047] As a preferred solution, the risk response module comprises a risk discrimination module, a record storage module and an adaptive repair module.

[0048] The risk discrimination module is configured to evaluate the running state of the ecological system in real time based on the weighted directed graph topology and the dynamic adjustment strategy, and to issue a risk warning signal when predicting that the system collapse probability exceeds a preset threshold.

[0049] The record storage module is configured to receive the risk warning signal and trigger the blockchain storage, and to record the current ecological state parameters, regulation and control history data and system response information through the blockchain storage.

[0050] The adaptive repair module is configured to activate a backup biological reaction chain structure, automatically inject microbial agents and reconstruct the ecological regulation and control model through a transfer learning method when a risk state occurs.

[0051] The present application has the following advantages:

[0052] 1. The present application integrates ecological perception, topology modeling, intelligent decision-making, regulation and control execution and risk response through a modular architecture, builds a full-process intelligent aquaculture system running through data collection, intelligent modeling, dynamic control and emergency disposal, and improves the intelligentization, automation and ecological steady-state regulation and control capability of the multi-nutrient level aquaculture system.

[0053] 2、The present application divides the trophic level based on multi-dimensional ecological parameters through the ecological topology module, mines the time sequence correlation law between levels, describes the dynamic evolution characteristics of the ecological network through the weighted directed graph structure, and then combines quantum bit coding and Hadamard gate operation to optimize the network structure search path and population diversity, so as to build an efficient ecological map model that can be self-adaptively updated, and the decision control module introduces a competitive reward function mechanism, and is oriented to multi-objective collaborative optimization such as dissolved oxygen stability, bait conversion rate, growth efficiency and energy consumption control, trains an ecological regulation model, and outputs a dynamic strategy for precise control of the execution equipment, so as to ensure that the system always runs in a high-efficiency and low-energy consumption state.

[0054] 3、The present application constructs a strategy execution perception feedback closed loop link through the execution control module device interface protocol adaptation, execution feedback collection and regulation result comparison, realizes adaptive intelligent adjustment and real-time dynamic correction of the ecological system, and the risk response module introduces a weighted graph and strategy coupling discrimination mechanism to predict and evaluate the potential collapse risk of the ecological system, and when the trigger threshold is triggered, automatically executes the three response mechanisms of blockchain storage, standby reaction chain activation and model reconstruction, ensures that the system has the ability of continuous operation and self-recovery, and greatly enhances the stability and anti-risk ability of the system in complex environment. BRIEF DESCRIPTION OF DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0056] Fig. 1 is a system block diagram of an artificial intelligence driven multi-trophic level comprehensive aquaculture system according to an embodiment of the present application;

[0057] Fig. 2 is a case mode diagram of an artificial intelligence driven multi-trophic level comprehensive aquaculture system according to an embodiment of the present application.

[0058] In the drawings:

[0059] 1, ecological perception module; 2, ecological topology module; 3, decision control module; 4, execution control module; 5, risk response module. DETAILED DESCRIPTION

[0060] The specific embodiments of the present application will be further described in detail below in combination with the drawings and examples. The following examples are used to illustrate the present application, but not to limit the scope of the present application.

[0061] The following detailed description of embodiments of the application in the drawings provided is not intended to limit the scope of the application claimed, but merely represents selected embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the application.

[0062] The application will be further described in conjunction with the drawings and specific embodiments, as shown, according to the artificial intelligence driven multi-nutrient level comprehensive breeding system, the artificial intelligence driven multi-nutrient level comprehensive breeding system comprises: an ecological perception module 1, an ecological topology module 2, a decision control module 3, an execution control module 4 and a risk response module 5; Figs. 1-2

[0063] The ecological perception module 1 is configured to deploy a biological electrode array, a distributed optical fiber sensor and a multi-spectral imaging device, to collect multi-dimensional ecological parameters in the breeding environment in real time, and to construct an ecological data set with time synchronization characteristics.

[0064] Specifically, the ecological perception module 1 comprises a biological electrode module, an optical fiber sensing module, a spectral imaging module and a parameter data module.

[0065] The biological electrode module is configured to perceive electrochemical response information of microbial communities in the water body of the breeding field through the deployed biological electrode array, and to generate corresponding bioactivity electrical signal data.

[0066] Specifically, the biological electrode array deployed in the breeding water body perceives the electron transfer behavior generated by the microbial community in the metabolic process in situ, the electrode array is composed of conductive films integrated on a flexible substrate, and the micro-current response formed on the electrode surface captures the electron flow signal released by the microorganism in the water body in real time, and the signal reflects the metabolic intensity and community activity state of the microorganism under specific environmental parameters. The electrode array and the matched electrochemical amplification module continuously monitor the metabolic activity in the target area through constant potential control mode, and sample the generated micro-electrical signal at high time resolution, which is input to the parameter data module after analog-digital conversion. The parameter data module is built-in low-noise filter and time synchronization mechanism, which is used for denoising, standardizing and sequencing the collected original electrochemical response data, and finally generates bioactivity electrical signal data with time sequence characteristics, and together with spectral imaging data and physical and chemical parameters, forms an ecological data set in a unified format, providing high-quality input for ecological topology construction and dynamic control strategy generation.

[0067] ​The optical fiber sensing module is configured to collect aquaculture physical parameters and chemical parameters of temperature, dissolved oxygen, pH value and turbidity in the water body of the aquaculture farm through the distributed optical fiber sensor, and construct time series ecological data based on the aquaculture physical parameters and chemical parameters.

[0068] Specifically, the key physical and chemical parameters in the water body are collected at a high frequency through the distributed optical fiber sensor network arranged in the aquaculture area, and the sensor network is composed of an erbium-doped optical fiber or a Bragg grating array and is arranged along the bottom and vertical section of the aquaculture pond to form a multi-dimensional sensing structure with continuous spatial coverage and synchronous node response. The optical fiber sensing node can monitor the typical aquaculture environmental indicators such as temperature, dissolved oxygen, pH value and turbidity in the aquaculture water body in real time. The system converts the reflected spectrum signal into a parameter value through a demodulation module, and completes data preprocessing, time stamp registration and preliminary clustering analysis through an edge computing node. The obtained aquaculture physical parameters and chemical parameters are sent to a parameter data module through a unified interface, and the time series sliding window mechanism is used to structure the continuous data, and the synchronous label is combined to generate complete time series ecological data.

[0069] The spectral imaging module is configured to collect image information data of the water body of the aquaculture farm through a multi-spectral imaging device.

[0070] Specifically, the optical reflection characteristics of target objects such as microbial communities, aquatic plants and plankton in the water body are collected in a targeted manner through a multi-spectral imaging device installed at key ecological sites in the aquaculture area. The multi-spectral imaging device integrates visible light, near-infrared and ultraviolet band sensing modules and cooperates with a light filter group and a multi-band image sensing chip to realize synchronous imaging under different spectra. During device operation, the imaging device periodically collects image information of the surface and middle layer regions of the water body, and the collection frequency can be adjusted dynamically according to the environment. The obtained raw image data is processed by the edge computing module to separate the spectral channels, normalize the reflectivity and perform pseudo-color mapping to obtain ecological image data with spectral distribution characteristics.

[0071] The parameter data module is configured to collect biological active electrical signal data, time series ecological data and image information data to obtain multi-dimensional ecological parameters, and generate an ecological data set based on the time sequence order and the multi-dimensional ecological parameters.

[0072] Specifically, the parameter data module includes a data cleaning module, a feature extraction module and a standard processing module.

[0073] The data cleaning module is configured to detect outliers in the received biological active electrical signal data, time series ecological data and image information data, and perform interpolation processing on missing values to obtain multi-dimensional ecological parameters.

[0074] The feature extraction module is configured to extract time sequence features of the multi-dimensional ecological parameters by using a time sequence sliding window mechanism.

[0075] Specifically, the time sequence sliding window mechanism built in the parameter data module is used to extract time sequence features of the multi-dimensional ecological parameters obtained from various sensors, and the ecological parameters include but are not limited to physical and chemical indexes such as temperature, pH value, dissolved oxygen, turbidity, etc. collected by distributed optical fiber sensors, and electrochemical signals and optical image features obtained by biological electrode arrays and multi-spectral imaging devices. The sliding window mechanism divides the continuously collected data stream into windows in a fixed length and set step length manner, each window being an independent analysis unit that captures the fluctuation trend, periodic change and abnormal disturbance features of the parameters in the time interval. Feature extraction is performed in each time window, and statistical and frequency domain features such as mean, variance, rate of change, Fourier spectrum energy distribution, etc. are calculated, and the results are labeled. The time sequence ecological feature sequence formed after the sliding window processing is stored in a standard time stamp format and provided as input to the ecological topology modeling module and the strategy generation engine, to support modeling of the dynamic behavior of the ecological system and pre-feed prediction of the control strategy.

[0076] The standard processing module is configured to normalize the multi-dimensional ecological parameters, and generate an ecological data set based on the unified time stamp and the normalized multi-dimensional ecological parameters.

[0077] Specifically, the parameter data module receives ecological parameter data collected by various sensing devices such as biological electrode arrays, distributed optical fiber sensors, multi-spectral imaging devices, etc. The data includes electrochemical response signals, spectral image features, temperature, dissolved oxygen, pH value, turbidity, etc. multi-dimensional indexes. Due to the significant differences in numerical scale, physical meaning and change amplitude of various ecological parameters, in order to ensure the uniformity and comparability of data processing, the system performs standardization and normalization processing on all parameters. The normalization processing converts different source parameters by using interval scaling method or Z-score standardization method, so that they are all mapped to [0, 1] or subject to standard normal distribution, to eliminate the influence of dimension difference on subsequent modeling, and through the normalization process, the system dynamically updates the calibration parameters and historical observation range of the collection device, to ensure that the parameter normalization processing and the actual ecological state keep synchronous response. The normalized multi-dimensional ecological parameters are synchronized and aligned with real-time collection through the unified time stamp mechanism, the system fuses and structures the data of different types of parameters according to the sensing time sequence, to form a time sequence ecological data set, and the ecological data set is represented in a two-dimensional time feature structure, which supports dynamic topology modeling, control strategy generation and anomaly detection, etc. intelligent function modules as standardized input, to build a basic data framework for system-level ecological sensing and intelligent control.

[0078] The ecological topology module 2 is configured to construct a weighted directed graph topology according to an ecological data set, optimize the weighted directed graph topology by using a quantum bit encoding parameter, and enhance population diversity by using a Hadamard gate operation.

[0079] Specifically, the ecological topology module 2 includes a topology construction module, an encoding optimization module, and a topology update module.

[0080] The topology construction module is configured to divide multi-dimensional ecological parameters into different trophic level parameters, obtain a hierarchical correlation relationship, and construct a weighted directed graph topology according to the trophic level parameters and the hierarchical correlation relationship.

[0081] Specifically, the topology construction module includes a hierarchical division module, a correlation identification module, and a structure generation module.

[0082] The hierarchical division module is configured to identify and extract trophic characteristic parameter sets of cultured animals, microorganisms, plankton, aquatic plants, and benthic organisms in the multi-dimensional ecological parameters, and perform trophic level division based on the trophic characteristic parameter sets.

[0083] Specifically, the parameter data module receives multi-dimensional ecological parameters from a biological electrode array, a distributed optical fiber sensor, a multi-spectral imaging device, and a physicochemical detection device. The parameters include electrochemical response values, spectral reflection characteristics, metabolic rates, dissolved oxygen concentrations, nitrogen-phosphorus ratios, algal densities, and biological membrane activity indices. In order to realize the identification and classification of different trophic level species, the system is built-in with a trophic characteristic identification sub-module. A feature selection algorithm is used to extract a feature parameter set that is significantly related to the activities of cultured animals, microorganisms, plankton, aquatic plants, and benthic organisms. In the feature extraction process, a combination method based on correlation analysis and information gain ranking is used to calculate the dynamic coupling degree between each ecological parameter and the typical trophic community metabolic index, automatically remove redundant or low sensitivity parameters, and generate a trophic characteristic parameter set corresponding to each type of organism. The obtained parameter set includes high correlation indicators such as population density variation, carbon source utilization rate, oxygen consumption rate, chlorophyll reflection intensity, and sediment disturbance frequency, forming a multi-category and multi-dimensional ecological function feature space. Subsequently, the system constructs a similarity function matrix based on the feature parameter set, and uses a spectral clustering algorithm or a graph-based hierarchical clustering method to divide the ecological units into hierarchical levels. The output includes the microorganism trophic layer, the aquatic plant trophic layer, the plankton trophic layer, and the benthic trophic layer. The boundary conditions of each level are jointly determined by the trophic source structure, the material exchange frequency, and the metabolic time sequence mode, and are adaptively adjusted during the dynamic update process of the ecological topology graph.

[0084] The association recognition module is configured to analyze time-series correlation parameters and covariation law parameters between each level of the nutrition levels, and construct logical association parameters between each level of the nutrition levels based on the time-series correlation parameters and the covariation law parameters.

[0085] Specifically, on the basis of completing the nutrition level division, the nutrition level analysis module is used to model the cross-level dynamic relationship between the historical ecological data of different levels. The module first receives a standardized time-series data set from the parameter data module, including a feature parameter sequence of each nutrition level at different time points, such as a microbial activity index, a plankton density, a water plant photosynthetic rate, and a benthic organism disturbance frequency.

[0086] In order to identify the interaction relationship between the levels, the system uses a joint method based on a delay correlation coefficient and mutual information entropy analysis to extract time-series correlation parameters between different nutrition levels, capture the causal evolution trend between the previous and subsequent time periods, and simultaneously extract covariation laws in multiple parameter dimensions between the nutrition levels through principal component covariation analysis and a sliding covariance matrix extraction strategy, to identify the cooperative change mode of the system during material circulation or ecological disturbance. On this basis, the system constructs a logical association parameter set based on the extracted time-series correlation parameters and covariation law parameters. Each association parameter represents a dynamic response relationship, a signal coupling strength, and a feedback regulation direction between specific levels, to describe the behavior channel between different nodes in the nutrition level network. The constructed logical association parameters are not only used for edge weighting and direction determination in the topological graph structure, but also serve as important prior knowledge for the strategy generated by the ecological regulation engine, to achieve precise modeling and response regulation of the multi-level state of the ecological system.

[0087] The structure generation module is configured to construct a weighted directed graph topology according to the nutrition level division result and the logical association parameters, and define the ecological factors corresponding to each node, and the weight of the edge to represent the parameter influence strength and directionality.

[0088] Specifically, based on the nutrition level division result and the logical correlation parameter between the nutrition levels, a weighted directed graph topology reflecting the multi-level dynamic coupling relationship of the ecosystem is constructed by a graph structure modeling module. The topology takes the nutrition level unit as the graph node and the ecological influence path with significant time sequence dependence or covariation rule between levels as the directed edge, forming a graph expression model of ecological causality flow. In terms of graph node definition, the system takes the microbial layer, the plankton layer, the aquatic plant layer and the benthic organism layer as the basic nodes respectively, each node binds its corresponding set of ecological factor parameters, including representative metabolic indicators, behavior frequency, population density change and nutrition absorption characteristics in the nutrition level, and all ecological factors are vectorized and coded as node feature embedding, supporting subsequent topology evolution modeling and strategy prediction tasks. In terms of graph edge definition, the system constructs the direction and weight of the edge according to the extracted logical correlation parameters. The direction of the edge is determined by the causal relationship (for example: microbial metabolic activity affects plankton density); the weight of the edge is calculated by the correlation strength, mutual information value or dynamic adjustment frequency in the corresponding logical correlation parameter, to represent the cross-layer conduction strength and regulation influence degree of ecological factors. The finally constructed nutrition level weighted directed graph structure has time sequence responsiveness, structure adaptability and semantic expression ability, and can be used as the input basis of the graph neural network model for ecological state prediction, control strategy generation and abnormal source analysis and other intelligent control application scenarios. The coding optimization module is used for parameterized conversion of the weighted directed graph topology through quantum bit coding, and optimization of the converted weighted directed graph topology, and enhancement of the population diversity of the optimization process through Hadamard gate operation;

[0089] Specifically, the coding optimization module includes a quantum coding module, a parameter optimization module and a population expansion module.

[0090] The quantum coding module is used for quantum bit coding of the weighted directed graph topology, and mapping the node attributes and edge weights of the weighted directed graph topology to adjustable quantum bit sequences.

[0091] Specifically, after the construction of the weighted directed graph topology is completed, the graph structure information is quantum mapped based on the graph coding module to adapt to subsequent quantum neural network modeling or quantum optimization strategy calculation. The graph structure takes the nutrient level as the node and the ecological causal relationship as the edge, and is attached with the ecological factor vector of each node and the weight parameter of the edge. In the quantum coding process, the system first allocates a corresponding number of quantum bit groups according to the number of nodes and the connection structure in the graph, and constructs a dedicated quantum state representation for each node through a variational quantum circuit. The ecological factor parameters (such as population density, metabolic rate, photosynthetic activity, etc.) of each node are normalized and mapped as rotation angles for the initial state setting of the quantum gate, thereby embedding the ecological semantics into the quantum bit sequence. In the quantum mapping of the graph edge weight, the system adopts an entanglement structure construction strategy based on edge weight regulation, maps the logical correlation strength as the adjustment factor of the control gate, and reflects the influence strength and directionality between parameters by regulating the entanglement degree between different quantum bits. The forward causal path is mapped as a standard control structure, and the reverse path adopts a controlled anti-spin structure to ensure that the topological directionality is maintained in the quantum state evolution. The regenerated quantum bit sequence can be used for quantum state measurement, quantum state distance calculation, or as input into a quantum neural network model to realize high-dimensional quantum modeling and regulatory decision support for complex correlation structures in ecological systems.

[0092] The parameter optimization module is configured to perform iterative optimization on the adjustable quantum bit sequence through a quantum interference and measurement mechanism, and retain the encoding solution in the adjustable quantum bit sequence after the iterative optimization.

[0093] Specifically, based on the constructed adjustable quantum bit sequence, the generated quantum encoding structure is iteratively optimized through quantum interference and measurement mechanism to improve the expression ability of ecological state modeling or control strategy prediction, and the process is based on variational quantum circuit structure and quantum gradient estimation technology, and the core goal is to search for the optimal encoding solution in the quantum state space. The optimization process first interferes the quantum state in the adjustable quantum bit sequence by running the quantum circuit multiple times, wherein the control gate is used to construct the entanglement channel between nodes, and the rotation gate is used to fine-tune the encoding amplitude angle of each ecological factor, and the system calculates the observation probability distribution and expected value of the current quantum state under the target function, such as classification accuracy, loss function value or ecological topology fitting degree, by sampling the statistical distribution of the measured state. Subsequently, the system updates the control parameters of the corresponding rotation gate in the bit sequence based on the measurement gradient (such as parameter shift rule) in the parameterized quantum circuit, and realizes the quantum optimization of the node ecological factor and the edge weight logical structure. The optimization cycle continues until the measurement convergence condition is met or the maximum iteration number is reached, and finally the converged quantum bit configuration is retained as the encoding solution output, and the optimized adjustable quantum bit sequence not only contains the topological semantics of the original graph structure, but also integrates the potential features such as ecological structure stability, feedback channel responsiveness and state distinguishability extracted in multiple rounds of interference measurement, providing interpretable and high-density quantum embedding representation for subsequent quantum neural network modeling and complex ecological strategy generation.

[0094] The population expansion module is used to introduce Hadamard gate operation and quantum state superposition mechanism during the iterative optimization process of the adjustable quantum bit sequence, increase the population diversity of the candidate solution, and prevent falling into local optimum.

[0095] Specifically, in the process of executing the iterative optimization of the adjustable quantum bit sequence, in order to improve the coverage of the search space and inhibit the local optimal trap, the Hadamard gate operation and the quantum state superposition mechanism are introduced into the variational quantum circuit structure to enhance the population diversity and global exploration ability of the candidate solution, and in the initial optimization stage, the system applies the Hadamard gate operation to part or all of the adjustable quantum bits, so that the quantum state is changed from the classical ground state to the superposition state with equal probability, and in order to avoid the optimization process from falling into a specific path due to weight initialization or structural bias in the early stage, the system introduces a dynamic Hadamard disturbance mechanism in each iteration, that is, at the key iteration point or when the convergence speed slows down, the Hadamard gate or combined interference gate is periodically applied to part of the quantum bits again, artificially increasing the distribution reorganization of the quantum state, so as to break the existing interference mode and activate the potential global optimal structure. After quantum measurement and interference optimization, multiple candidate quantum bit sequences will be sent to the reservation and screening module, and the system will archive the solutions with excellent performance in the superposition state according to the target function value and the quality of the measurement state distribution, and reserve their control parameter configurations as effective encoding solutions after iterative optimization. The mechanism ensures that each round of optimization is not just a local fine-tuning, but continuously introduces generalized disturbance and distribution jump in the encoding space, effectively improving the global solution coverage probability.

[0096] The topology updating module is configured to collect ecological data dynamic updating parameters, and optimize and adjust the node weight and edge weight of the weighted directed graph based on the ecological data dynamic updating parameters, and generate and store the optimization adjustment characteristic parameters.

[0097] Specifically, the ecological perception layer continuously collects dynamic ecological data from devices such as distributed fiber optic sensors, biopotential electrode arrays, and multispectral imaging devices, and constructs an ecological data dynamic updating parameter set in the parameter data module. The parameter set records the change trend, fluctuation amplitude and abnormal disturbance signal of key environmental indicators (such as temperature, dissolved oxygen, pH value, microbial activity, etc.) in time series format, forming real-time updating driving factors reflecting the evolution process of the ecological state.

[0098] During the graph structure optimization phase, the system adjusts the node attributes and edge weights in the weighted directed graph at the trophic level in real time based on this dynamically updated parameter set. During node optimization, the system adjusts the weights of the ecological factors bound to the current node in the graph according to the degree of change in the updated parameters (such as a surge in metabolic rate or a sudden change in plankton density), reflecting the changes in ecological influence at the current level. Edge weight optimization dynamically reassigns the directionality or weight strength of edges based on the parameter linkage between cross-layer nodes (such as increased latency correlation or decreased coupling stability). Simultaneously, the system automatically generates an optimization adjustment characteristic parameter set after each update, recording the state of each node and edge before and after adjustment, including: the magnitude of weight changes, the rate of change of node factors, the direction of change in edge coupling strength, and the probability of path migration. This characteristic parameter set is stored in the ecological graph evolution cache module after standardization encoding, serving as both a basis for tracking historical states and a parameter reference input for training graph neural network models or quantum control modules.

[0099] The decision control module 3 is used to set a competitive reward function, construct an ecological regulation model based on the competitive reward function, and then generate a dynamic adjustment strategy for ecological regulation based on the ecological regulation model.

[0100] Specifically, decision control module 3 includes: a reward function module, an intelligent training module, and a strategy generation module;

[0101] The reward function module is used to set a competitive reward function according to the aquaculture target, and the reward function includes dissolved oxygen stability, feed conversion rate, biological growth efficiency and energy consumption control indicators.

[0102] Specifically, adjustable competitive multi-objective reward functions are set according to different types of aquaculture objectives (such as high-yield propagation, energy conservation and consumption control, and ecological stability maintenance) to drive the multi-agent regulatory system to achieve the optimal generation of ecological regulation strategies. This reward function constructs a set of sub-objective functions based on key process indicators in the ecosystem and integrates the dynamic balance relationships among them to achieve the evolution of control strategies under complex objective trade-offs.

[0103] In the index design, the system introduces four types of core ecological state parameters: dissolved oxygen stability index: through the sliding window statistics of water dissolved oxygen concentration short-term standard deviation and long cycle trend deviation degree, for measuring the adjustment stability of oxygen supply system. The smaller the deviation interval, the higher the reward. The bait conversion rate index: based on the feeding log and the biomass growth data, the net biomass gain rate corresponding to the unit bait input is calculated. When the conversion efficiency is high and the bait is wasted, the system gives positive reinforcement. Biological growth efficiency index: the average length or mass growth rate of the cultured object per unit time is calculated, and the individual health index change trend is combined to give reward weight to the state of high growth rate and no disease burden. Energy consumption control index: according to the running time and real-time power data of water pump, aeration, temperature control and other subsystems, the energy consumption density corresponding to unit ecological improvement index (such as DO improvement) is calculated. The lower the energy consumption, the stronger the reward. And each sub-index is normalized to the [0, 1] interval, and forms a reward function vector input multi-objective optimization module. The system uses dynamic weight method or Pareto frontier sorting mechanism to compete and integrate each sub-reward item, and the relative weight distribution of sub-targets can be dynamically adjusted under different breeding strategies (such as high proliferation priority or energy efficiency priority), to ensure that the control strategy realizes strategic guidance and resource coordination among multiple objectives.

[0104] The intelligent training module is configured to construct an ecological regulation model based on a competitive reward function and to verify and optimize the ecological regulation model.

[0105] Specifically, the intelligent training module includes a model construction module, a training optimization module, and a verification and evaluation module.

[0106] The model construction module is configured to construct an ecological regulation model based on a competitive reward function and to initialize strategy network parameters and value function network parameters of the ecological regulation model.

[0107] Specifically, based on the defined competitive multi-objective reward function, an ecological regulation model is constructed for learning and outputting the optimal control strategy in a dynamic ecological environment. The model is based on a multi-agent reinforcement learning framework, combined with the nonlinearity, time-varying nature and multi-objective conflict characteristics of the breeding environment, and uses a dual-channel structure of strategy network and value function network to realize the decoupling and cooperation of strategy generation and value evaluation. In terms of model structure, the strategy network is responsible for receiving the current ecological state features (such as dissolved oxygen concentration, feeding frequency, growth rate, etc.) as input, and outputs a behavior probability distribution including feeding rate adjustment, aeration time adjustment, water quality circulation triggering and other control actions through a multi-layer perception network. The value function network receives the same state input in parallel and outputs the expected return value of the current state under the given reward function, which is used to evaluate the long-term effectiveness of the current strategy.

[0108] In the parameter initialization phase, the system sets the model weight according to the following mechanism: policy network initialization: all layer weights use the Xavier initialization method to keep the activation of the front and back layers stable variance; the bias term is initialized to zero, and the output layer uses the Softmax function to normalize the generated behavior distribution, ensuring that the policy is derivable and suitable for probability sampling mechanism. Value function network initialization: use the He initialization method to enhance the sensitivity of deep network to gradient, and introduce the BatchNorm layer to stabilize the training process; the output layer is a linear activation function, which is used to express the true numerical expected value, and the initialization process integrates the sample ecological state feature mean and variance statistics to realize the consistency alignment with the real ecological distribution. After initialization, the system updates the policy and value function network parameters by sampling the environment interaction trajectory, and guides the model to focus on high value actions according to the reward feedback signal, and the final model has the ability to adaptively output ecological regulation instructions according to different breeding targets, supporting intelligent collaborative regulation of multiple actions such as feeding, oxygenation, water quality regulation, etc.

[0109] The training optimization module is configured to alternately train the ecological regulation model based on the ecological data set and optimize and adjust the policy network parameters and the value function network parameters of the ecological regulation model.

[0110] Specifically, based on the standardized ecological data set constructed, the policy network and the value function network in the ecological regulation model are executed alternately iterative training to improve the adaptability of the model to the dynamic environment and the optimality expression ability of the regulation instruction, and the ecological data set includes historical state sequence, executed control behavior, corresponding feedback reward value and environmental evolution information, covering dissolved oxygen change, bait conversion performance, biological growth record and energy consumption response and other multi-dimensional ecological factors. The alternately training process: the alternately training adopts a two-stage training mechanism based on the Actor-Critic structure, and the policy network is used as the Actor to output the control behavior distribution under the current ecological state, and the value function network is used as the Critic to evaluate the long-term return of the behavior selected by the Actor as the basis for policy adjustment. The training process is as follows: in the policy evaluation stage, under the condition of fixed policy network parameters, the action generated by the current policy and the environmental feedback are used to train the value function network, and the parameters are updated by minimizing the mean square error loss between the predicted value and the actual cumulative reward to improve the value estimation accuracy. In the policy improvement stage, the updated value function network parameters are fixed, and the policy network is updated by the gradient ascent method to make the probability of high value actions increase and the selection probability of low value actions decrease, thereby strengthening the expected return oriented policy generation process. After each training cycle, the system automatically evaluates the policy improvement effect to determine whether the policy convergence condition is met or further iteration is needed, and the alternation process ensures that the model realizes closed-loop collaborative optimization between behavior generation and value evaluation.

[0111] Parameter optimization details: Strategy network parameter optimization: learning rate adopts adaptive adjustment strategy, and introduces strategy entropy penalty term in initial stage to prevent premature convergence to single action path; strategy regularization is used to enhance strategy diversity. Value function network parameter optimization: sliding target network structure is introduced to relieve gradient shock, and experience replay mechanism is used to disrupt training sample sequence to improve generalization ability. All optimization parameters and gradient information are monitored and recorded during the training period. The system generates model optimization trajectory log and training round weight snapshot to verify the model, trace the ecological strategy, and align the subsequent quantum control mapping module.

[0112] The verification and evaluation module is configured to verify and evaluate the optimized ecological regulation model and generate an evaluation report.

[0113] Specifically, after the training and optimization of the strategy network and value function network of the ecological regulation model are completed, the verification and evaluation process is started to systematically test the regulation effectiveness, ecological adaptability, and multi-objective achievement rate of the model, and a standardized evaluation report is automatically generated for result archiving, model comparison, and strategy optimization support. The model verification process includes evaluation environment construction, deployment of the optimized model in a virtual simulation environment and a historical ecological data driven environment, simulation of ecological disturbances existing in real farming scenarios (such as sudden decrease of dissolved oxygen, high peak of feed waste, abnormal growth rate, etc.), and playback verification of the model regulation response. Multi-dimensional evaluation index setting: the evaluation dimensions cover four types of objective functions: dissolved oxygen stability score, feed conversion efficiency score, growth rate response score, and unit energy consumption control score. Each index is calculated by the deviation function of the actual regulation result and the target value, and the normalized score is output to measure the matching degree of the model to each ecological target. Behavior robustness analysis: environmental parameter disturbance experiments (such as temperature ±3℃ fluctuation, random shading of light intensity, etc.) are introduced to test whether the model can still maintain the stability and interpretability of the strategy output under non-stable conditions of the ecological system, analyze the strategy change amplitude and abnormal action triggering probability, and evaluate the robustness of the model. The evaluation report generation logic automatically summarizes the following information to form a structured report after the verification process is completed: model version and configuration (strategy network structure, optimization strategy, weight snapshot ID); test environment parameters and disturbance setting list (simulation data compared with actual historical data); key indicator score table and trend chart (including the distribution of four types of core objective functions); strategy output change curve (control action changes over time and disturbance response); abnormal response and failure record (such as over-limit adjustment, strategy fluctuation); the report format supports multiple output structures, is automatically archived to the model evaluation log module, and generates a unique verification task number for subsequent model comparison, version tracing, and pre-online audit use.

[0114] The strategy generation module is configured to generate a dynamic adjustment strategy according to the verified and optimized ecological regulation model, and output a regulation instruction for controlling an execution device.

[0115] Specifically, a real-time collected ecological state vector is received, including dissolved oxygen concentration, water temperature, pH value, bait residue, biological activity and other multi-dimensional characteristics, and is transmitted as input into the optimized strategy network, and the strategy network outputs a set of action probability distribution according to the current state distribution, reflecting the optimal ecological intervention behavior selection tendency under the current condition. For different time periods or external constraints (such as power limiting period, before typhoon, early stage of epidemic, etc.), the system can adjust the weights of each sub-goal in the reward function (such as temporarily emphasizing the lowest energy consumption or DO stability priority) to affect the output direction of the strategy, thereby realizing the dynamic adaptation and generation of the goal-oriented strategy, and the system performs action sampling with constraints (such as selecting the optimal ventilation duration within the maximum oxygen supply capacity limit) according to the strategy output distribution, and performs action discretization, amplitude clipping and other processing to ensure that the strategy meets the device physical parameter boundary and ecological safety standards.

[0116] The strategy sequence is converted into a structured control instruction set after analysis, mainly including the following types: oxygenation device instructions: start / stop state, running time (unit: minutes), target DO reference value; feeding device instructions: feeding quality (unit: grams), distribution frequency (unit: times / hour); water pump / circulation device instructions: on / off state, flow level, running period; water quality regulation device instructions (such as pH buffer addition): dosage estimation, duration; device combination instruction schedule: cross-device collaborative instruction packaging, supporting planned batch execution. All control instructions are standardized and packaged into common formats and pushed to the device control interface layer, and the system can be connected to edge computing devices or Internet of Things control nodes to realize fast instruction issuance and feedback confirmation.

[0117] The execution control module 4 is configured to control the aquaculture-related execution devices according to the dynamic adjustment strategy.

[0118] Specifically, the execution control module 4 includes a device interface module, a regulation and control execution module, and a feedback receiving module.

[0119] The device interface module is configured to connect with the aquaculture devices and convert the regulation and control instructions in the dynamic adjustment strategy into control protocols compatible with the aquaculture devices for output.

[0120] Specifically, all aquaculture devices are connected to a unified control network, including oxygenators, automatic feeders, water quality circulation pumps, water temperature regulators, lighting systems, etc., and the connections between devices are through industrial communication protocols such as ModbusRTU, ModbusTCP, MQTT or RS-485 bus, and a centralized device control platform is used to uniformly manage all device communication and instruction scheduling.

[0121] The platform allows the system to access all device states and control options in a visual interface, and obtain operating parameters through the device interface module, such as the current working state of the oxygenation device, the remaining bait amount of the feeder, the current running frequency of the water pump, and other information. These data can be aggregated through standard device communication interfaces or through edge control nodes, supporting synchronous use at the policy generation end, and after receiving the control instructions generated by the policy network (such as oxygenation for 20 minutes or reducing the feeding frequency), the control actions in the policy are automatically mapped and matched with the device instruction template, and are encapsulated into low-level control data packets according to the communication protocol specifications supported by the device. For example: for devices that support ModbusRTU, generate data frames containing function code, register address, write value, and CRC check bits; for devices that use MQTT protocol, encapsulate the control actions into standard JSON format payload; for old serial port devices, the control instructions are converted into ASCII format control commands or hexadecimal instruction sequences. The converted instructions are sent to the device control end through the network or serial communication module, and the system continuously monitors the device feedback state code or ACK response to ensure that the instruction execution is successful, and in the case of failure or timeout, triggers an automatic retry mechanism or generates an alarm event.

[0122] The control execution module is configured to control the connected aquaculture devices to adjust according to the output control protocol;

[0123] Specifically, ensure that the control protocol instruction format received by all connected aquaculture devices is compatible with their hardware control interfaces, including oxygenation devices that support ModbusRTU, intelligent feeders that use MQTT protocol, water quality circulating pumps based on serial communication, and other aquaculture auxiliary devices with digital control input interfaces. After the adjustment instruction is generated, the system pushes the control protocol directly to the receiving interface of the corresponding device through the communication driver module, and monitors the execution result in real time. The execution of the adjustment behavior is completed through the control circuit built-in the device, such as the oxygenation machine automatically enabling the motor module and starting the running timer after receiving the 1200 second command; the intelligent feeder drives the control arm to act after receiving the 100 gram command, and records the feeding period and dose at this time. For devices with feedback capability, the system will synchronously receive the execution state, error code or environmental response data (such as DO value change, current load, etc.), and write it into the device execution state buffer area. The system supports checking whether the feedback instruction execution is completed through the device internal register or remote control frame, and judges whether the adjustment behavior is effective by comparing the collected data before and after execution (such as water oxygen content, power consumption value, etc.), and if necessary, triggers a secondary adjustment or abnormal alarm process.

[0124] The feedback receiving module is configured to compare the multi-dimensional ecological parameter data collected by the ecological perception module with the control protocol execution results in real time, and optimize the ecological regulation model according to the comparison results.

[0125] Specifically, the ecological perception module and the ecological regulation platform are ensured to maintain continuous communication connection, including a dissolved oxygen sensor, a thermometer, a pH sensor, a water flow speed meter, an electrical conductivity probe, a biological activity monitoring device, and the like, and multi-dimensional ecological parameters are collected at a high frequency through a standard communication interface such as MQTT, Modbus or LoRa. Meanwhile, the execution results of the control protocol that have been issued are recorded through device return data, response codes or action logs, including device response time, execution duration, status code feedback and change trend of perception data during execution, and the like.

[0126] The ecological regulation platform receives the data stream from the perception module in real time through the data interface service, and compares the collection results with the baseline values and model output expected values before regulation item by item, for example, the deviation between the current dissolved oxygen value and the strategy target value, the error between the actual feeding amount and the expected feeding amount, the hysteresis between the temperature rise and fall amplitude in the environmental feedback and the control behavior response, and the influence trend of the control behavior on water quality parameters such as pH and EC. The data comparison engine built in the platform automatically performs regression analysis on the actual feedback generated by all regulation behaviors, and evaluates whether the strategy execution achieves the expected ecological benefit based on the deviation direction and amplitude. If the deviation exceeds the dynamic tolerance threshold, the system will mark the round of strategy as a low-efficiency feedback behavior, and push it to the regulation model optimization module. The model optimization execution logic automatically extracts the sample data of the ecological state, control behavior, execution result and the like corresponding to the low-efficiency strategy based on the platform, adds an online incremental training set, updates the TD error sample of the value function network, strengthens the recognition ability of the model to low-return behaviors, adjusts the action probability distribution output by the strategy network, reduces the possibility of similar behaviors being sampled again in similar states, introduces new sampling data for short-period fine-tuning training, and maintains the adaptability and stability of the model in the actual ecological environment.

[0127] All optimization behaviors will record model version update logs in the background, and retain model strategy output difference analysis reports before and after optimization, for expert review and strategy backtracking.

[0128] The risk response module 5 is configured to record the current running state parameters through the blockchain storage module when predicting that the collapse probability of the ecological system exceeds a preset threshold, and to activate a standby biological reaction chain structure, automatically inject microbial preparations, and reconstruct the ecological regulation model through a transfer learning algorithm.

[0129] Specifically, the risk response module 5 includes a risk discrimination module, a storage record module and an adaptive repair module.

[0130] The risk discrimination module is configured to evaluate the running state of the ecological system in real time based on the weighted directed graph topology and the dynamic adjustment strategy, and to send a risk warning signal when predicting that the system collapse probability exceeds a preset threshold.

[0131] Specifically, all key factor nodes and control channels in the ecological system are connected through a weighted directed graph topology, including water body nutrient level nodes, biological population density nodes, energy flow nodes, and control behavior nodes, and the edge weight represents the influence strength and adjustment sensitivity between different factors. The graph structure is dynamically updated to reflect the current ecological evolution trend of the system. The graph model and the dynamic adjustment strategy are used to map the strategy output to the control node in real time, and to receive the environmental state data uploaded by the ecological perception module, including dissolved oxygen, pH, temperature, conductivity, and biological behavior abnormalities. The system calculates the stability index and potential chain reaction path of the whole graph in real time by combining the node state change and edge weight update. The platform has a risk propagation model based on the graph propagation algorithm, which simulates the factor disturbance triggered by each adjustment behavior path, and calculates the collapse probability of the current system by combining the propagation amplitude, duration, and target node vulnerability of each path, and compares it with the safety threshold set in advance.

[0132] The risk warning triggering mechanism includes:

[0133] When the system detects that the warning value at any time point exceeds the set threshold, the platform will immediately generate a risk warning event, including the current node state snapshot, control action record, graph structure weight matrix, propagation path simulation result, and risk index change trend. The risk signal is then pushed to the control platform to notify the ecological control model to pause the current strategy output and automatically switch to a conservative strategy mode, such as reducing the operation frequency and minimizing ecological disturbance. The expert intervention interface is triggered or the backup strategy is enabled, the external expert decision module or rule library model is connected through the interface, the artificial strategy input window is provided, and the safe and stable historical strategy is reserved as a backup to perform strategy rollback or minimum risk strategy replacement.

[0134] The evidence recording module is configured to receive the risk warning signal and trigger the blockchain evidence, and record the current ecological state parameters, control history data, and system response information through the blockchain;

[0135] Specifically, the ecological regulation platform ensures continuous monitoring of the risk index output by the system crash probability evaluation module. When the risk index exceeds the preset threshold and the system enters a high-risk state, the platform immediately triggers the risk warning signal processing module and links the blockchain storage engine to start the full-chain data sealing process. The platform automatically packages the risk identification event into a storage trigger request through the event-driven mechanism and accesses the private blockchain or alliance chain system. Through the smart contract interface, it calls the chain storage contract, generates a unique event hash index, and synchronously seals the following three types of core data: ecological state parameter snapshots, including current water quality data (such as DO, pH, temperature, ammonia nitrogen, nitrate, etc.), biological activity indicators (such as feeding rate, swimming frequency), system load state (such as device running state, power consumption), etc. The regulation history record includes: the output instructions of all strategy actions in the past 24 hours, the execution device, the duration, the device response code and the corresponding timestamp; At the same time, record the strategy model version and the key control parameter snapshot. System response and risk parameter data include: risk index value, warning threshold, propagation path simulation result, strategy interruption record and standby strategy trigger state; If there is microbial preparation injection or expert intervention, the relevant information is synchronously sealed, and the above data is packaged into on-chain transaction data through the storage contract, and the SHA-256 hash digest is used for digital signature, written into the on-chain distributed ledger node, and all storage data can be uniquely identified and tamper-proof verified through the on-chain hash value, supporting subsequent ecological behavior compliance audit, abnormal traceability and security event responsibility definition, and the platform simultaneously generates on-chain index code and storage certificate for each storage record for regulatory systems, scientific research teams or system maintenance parties to review.

[0136] The adaptive repair module is configured to activate a standby biological reaction chain structure, automatically inject microbial preparations, and reconstruct the ecological regulation model through a transfer learning method when a risk state occurs.

[0137] Specifically, the ecosystem has been pre-configured with at least one set of standby biological reaction chain structures during continuous operation, including ammonia-reducing microbial chain, denitrification chain, organic matter degradation chain, and oxygen-increasing collaborative chain, and the types of microbial agents associated with each reaction chain, the dosage range, and the target indicators have been registered in the system rule base. The system monitors the current ecological state and evaluates the collapse risk in real time. When the risk index exceeds the preset safety threshold, the standby biological reaction chain structure will be activated immediately. After identifying the risk type (such as ammonia nitrogen surge, pH anomaly, or plankton out of control), the ecological regulation platform calls the corresponding microbial compensation scheme of the reaction chain and starts the automatic injection module. Through the precise control of the liquid pump or the dissolution injection system, the microbial agents are injected at the set dose to ensure that the reaction chain restores the system function balance in the shortest time. For example, for the ammonia nitrogen rapid rise event, the system will call the nitrite oxidizing bacteria and denitrifying bacteria collaborative chain structure and inject the two types of agents in proportion, while starting the aeration assistance to improve the reaction rate.

[0138] In actual deployment, the present application can adapt to typical multi-nutrient level ecological combinations, such as the model of carnivorous fish (such as mandarin fish, bass, etc.) plus filter-feeding fish (such as silver carp and bighead carp) plus benthic omnivorous fish (such as loach and yellow catfish) plus aquatic plants (such as Elodea canadensis), and the dynamic optimization and regulation of the ecological combination includes: initial species and biomass recommendation: based on regional environmental parameters (temperature, water quality, nutrient salt concentration, etc.) and breeding goals, the system automatically recommends mixed culture combinations and initial proportions, and the system uses the population stability indicators of the nodes and the material exchange efficiency of the edges in the ecological graph as references to generate initial suggestions such as 4:1:2:2 adaptive structure, and Elodea is expressed in terms of biomass density per unit area or unit volume, not in equivalent proportion with fish, which can be adjusted according to actual specific needs at the time of use.

[0139] Dynamic biomass adjustment mechanism: AI continuously analyzes the ammonia nitrogen produced by the metabolism of carnivorous fish, the accumulation of leftover feed, etc.; if the water body nitrogen load increases, the system will suggest: increase the density of aquatic plants; increase the proportion of filter-feeding fish; reduce feeding or control the metabolic rate of carnivorous fish by delaying the feeding frequency.

[0140] Strategy self-optimization based on reinforcement learning: reward function combination: DO stability plus feed conversion rate plus growth efficiency plus energy consumption control; system automatic learning: regulation strategy with maximum mandarin fish growth rate plus total nitrogen concentration not exceeding limit plus sufficient plant photosynthesis; automatically adjust the feeding plan and species proportion strategy in different seasons, water temperatures, and feeding stages.

[0141] Ecological feedback and emergency intervention: if the bottom sensing system predicts that the water body will be eutrophicated or that plankton will explode; the system can start the standby reaction chain (such as injecting denitrifying bacteria) and repair the regulation model through transfer learning.

[0142] The above merely provides the preferred embodiment of the present application, and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. An artificial intelligence-driven multi-trophic level integrated aquaculture system, characterized in that, This AI-driven multi-trophic level integrated aquaculture system includes: an ecological perception module, an ecological topology module, a decision control module, an execution control module, and a risk response module. The ecological sensing module is used to deploy a bioelectrode array, a distributed optical fiber sensor, and a multispectral imaging device to collect multidimensional ecological parameters in the aquaculture environment in real time and construct an ecological dataset with time synchronization characteristics. The ecological topology module is used to construct a weighted directed graph topology based on the ecological dataset, optimize the weighted directed graph topology using qubit encoding parameters, and enhance population diversity through Hadamard gate operations. The decision control module is used to set a competitive reward function, construct an ecological regulation model based on the competitive reward function, and then generate a dynamic adjustment strategy for ecological regulation based on the ecological regulation model. The execution control module is used to control aquaculture-related execution equipment according to a dynamic adjustment strategy; The risk response module is used to trigger the blockchain storage module to record the current operating status parameters and activate the backup biological reaction chain structure when the probability of ecosystem collapse is predicted to exceed a preset threshold. It also automatically injects microbial agents and reconstructs the ecological regulation model through transfer learning algorithms. The ecological topology module includes: a topology construction module, a coding optimization module, and a topology update module; The topology construction module is used to divide multidimensional ecological parameters into different trophic level parameters, obtain hierarchical association relationships, and construct a weighted directed graph topology structure based on the trophic level parameters and hierarchical association relationships. The encoding optimization module is used to perform parameterization transformation of the weighted directed graph topology using qubit encoding, optimize the transformed weighted directed graph topology, and enhance the population diversity of the optimization process through Hadamard gate operations. The topology update module is used to dynamically update parameters based on the collected ecological data, and to optimize and adjust the node weights and edge weights of the weighted directed graph based on the dynamically updated ecological data parameters, while generating and storing the optimized adjustment characteristic parameters. The topology construction module includes: a hierarchy partitioning module, an association identification module, and a structure generation module; The hierarchical division module is used to identify and extract the nutrient characteristic parameter set of microorganisms, plankton, aquatic plants and benthic organisms in the multidimensional ecological parameters, and to perform nutrient hierarchical division based on the nutrient characteristic parameter set. The association identification module is used to analyze the time-series correlation parameters and covariation law parameters between each trophic level, and to construct logical association parameters between each trophic level based on the time-series correlation parameters and covariation law parameters. The structure generation module is used to construct a weighted directed graph topology based on the trophic level division results and logical association parameters, and to define the ecological factors corresponding to each node and the influence intensity and directionality of the edge weights.

2. The AI-driven multi-trophic level integrated aquaculture system according to claim 1, characterized in that, The ecological sensing module includes: a bioelectrode module, a fiber optic sensing module, a spectral imaging module, and a parameter data module; The bioelectrode module is used to sense the electrochemical response information of the microbial community in the aquaculture farm water body through the deployed bioelectrode array, and generate corresponding bioactive electrical signal data. The fiber optic sensing module is used to collect physical and chemical parameters of aquaculture, such as temperature, dissolved oxygen, pH value, and turbidity, in the aquaculture water body through the set distributed fiber optic sensors, and to construct time series ecological data based on the physical and chemical parameters of aquaculture. The spectral imaging module is used to acquire image information data of the aquaculture farm water body using visible light, near infrared and ultraviolet bands through the installed multispectral imaging device. The parameter data module is used to summarize bioactive electrical signal data, time-series ecological data and image information data to obtain multidimensional ecological parameters, and generate an ecological dataset based on the time sequence and multidimensional ecological parameters.

3. The AI-driven multi-trophic level integrated aquaculture system according to claim 2, characterized in that, The parameter data module includes: a data cleaning module, a feature extraction module, and a standard processing module; The data cleaning module is used to detect outliers in the received bioactive electrical signal data, time-series ecological data and image information data, and to impute missing values ​​to obtain multidimensional ecological parameters. The feature extraction module is used to extract temporal features from the obtained multidimensional ecological parameters using a temporal sliding window mechanism; The standard processing module is used to normalize the multidimensional ecological parameters and generate an ecological dataset based on the unified timestamp and the normalized multidimensional ecological parameters.

4. The AI-driven multi-trophic level integrated aquaculture system according to claim 1, characterized in that, The encoding optimization module includes: a quantum encoding module, a parameter optimization module, and a population expansion module; The quantum encoding module is used to encode the weighted directed graph topology into qubits and map the node attributes and edge weights of the weighted directed graph topology into an adjustable qubit sequence. The parameter optimization module is used to iteratively optimize the tunable qubit sequence through quantum interference and measurement mechanisms, and retain the encoded solution in the tunable qubit sequence after iterative optimization. The population expansion module is used to introduce Hadamard gate operation and quantum state superposition mechanism during the iterative optimization of tunable qubit sequences, thereby increasing the population diversity of candidate solutions and preventing them from getting trapped in local optima.

5. The AI-driven multi-trophic level integrated aquaculture system according to claim 1, characterized in that, The decision control module includes: a reward function module, an intelligent training module, and a strategy generation module; The reward function module is used to set a competitive reward function according to the aquaculture target, and the reward function includes dissolved oxygen stability, feed conversion rate, biological growth efficiency and energy consumption control indicators. The intelligent training module is used to construct an ecological regulation model based on a competitive reward function, and to verify and optimize the ecological regulation model. The strategy generation module is used to generate dynamic adjustment strategies based on the verified and optimized ecological regulation model, and output regulation commands for controlling the execution equipment.

6. The AI-driven multi-trophic level integrated aquaculture system according to claim 5, characterized in that, The intelligent training module includes: a model building module, a training optimization module, and a verification and evaluation module; The model building module is used to build an ecological regulation model based on a competitive reward function and initialize the policy network parameters and value function network parameters of the ecological regulation model. The training optimization module is used to alternately train the ecological regulation model using the ecological dataset, and to optimize and adjust the policy network parameters and value function network parameters of the ecological regulation model. The verification and evaluation module is used to verify and evaluate the optimized ecological regulation model and generate an evaluation report.

7. The AI-driven multi-trophic level integrated aquaculture system according to claim 1, characterized in that, The execution control module includes: a device interface module, a control execution module, and a feedback receiving module; The device interface module is used to connect with the aquaculture equipment and convert the control commands in the dynamic adjustment strategy into a control protocol output compatible with the aquaculture equipment. The control execution module is used to control the connected aquaculture equipment to adjust according to the output control protocol; The feedback receiving module is used to receive multi-dimensional ecological parameter data collected by the ecological sensing module in real time, compare it with the results of the executed control protocol, and optimize the ecological regulation model based on the comparison results.

8. The AI-driven multi-trophic level integrated aquaculture system according to claim 1, characterized in that, The risk response module includes: a risk assessment module, an evidence storage and recording module, and an adaptive repair module; The risk discrimination module is used to assess the operating status of the ecosystem in real time based on the weighted directed graph topology and dynamic adjustment strategy, and to issue a risk warning signal when the predicted probability of system collapse exceeds a preset threshold. The evidence storage and recording module is used to receive risk warning signals and trigger blockchain evidence storage, and then record the current ecological status parameters, historical data of regulation and control and system response information through blockchain evidence storage; The adaptive repair module is used to activate the backup biological reaction chain structure, automatically inject microbial agents, and reconstruct the ecological regulation model through transfer learning when a risky state occurs.

Images