Integrated water quality monitoring and processing system for aquaculture

Abstract

The invention discloses an integrated water quality monitoring and processing system for aquaculture, and the system comprises a data collection module which is used for obtaining dissolved oxygen, ammonia nitrogen, pH value, temperature and flow data from a plurality of spatial distributed sensing nodes; the data preprocessing module is used for completing noise removal, correction and time sequence synchronization; the water quality state estimation module is used for outputting real-time multi-point water quality states based on a distributed unscented Kalman filtering algorithm; the space-time fluid modeling module is used for constructing a Lagrange model to extract a flow velocity field and a diffusion coefficient; the scheduling optimization module is used for generating an aeration and dosing scheme by adopting an improved multi-target Pareto optimization algorithm based on a space-time sensitive boundary; the execution module is used for controlling an air blower, an aeration pipe and a dosing pump to complete oxygenation and dosing operations; and the self-adaptive updating module is used for collecting feedback data to adjust filtering gain and optimizing parameters so as to realize closed-loop control of the system.

Description

Technical Field

[0001] The present invention relates to the technical field of water quality monitoring, and in particular to an integrated water quality monitoring and treatment system for aquaculture. Background Art

[0002] As the aquaculture industry develops toward high-density, high-efficiency, and intensive operations, real-time dynamic monitoring and precise regulation of water quality in aquaculture waters have become critical factors in ensuring aquaculture profitability and ecological stability. This is especially true in models like recirculating aquaculture and industrial pond aquaculture, where water quality fluctuates frequently and pollution accumulates rapidly. Without precise monitoring and response mechanisms, this can easily trigger stress reactions, disease spread, and even mortality in fish and shrimp, resulting in severe economic losses. Therefore, establishing a scientific, real-time, and highly accurate water quality monitoring and treatment system has become a crucial component of modern aquaculture technology.

[0003] Currently, aquaculture water quality management relies heavily on the following traditional methods: First, manual sampling and testing are used to obtain water parameters such as dissolved oxygen, ammonia nitrogen, pH, and temperature through chemical analysis. However, this method has a long cycle, poor timeliness, and high costs, making it difficult to adapt to the needs of dynamic regulation. Second, a single-point monitoring system based on fixed sensors obtains some water quality parameters through a single or small number of sensors and uploads the data to a backend system for recording and display via wired or wireless means. Although this method has achieved digital management of water quality data to a certain extent, due to sparse distribution points, isolated data, and lagging algorithms, it cannot accurately reflect the spatial distribution changes of the entire aquaculture water body, and its ability to identify abnormal changes and potential risks is relatively weak.

[0004] In addition, some existing water quality control systems are based on set thresholds to trigger operations such as oxygenation or drug administration. Their core relies on static parameter judgment logic and lacks the ability to model complex hydrodynamic characteristics and multi-parameter coupling mechanisms. Especially when faced with situations such as multi-point spatial disturbances, sudden pollution, and complex flow field distribution, such systems are unable to achieve precise intervention and dynamic scheduling. Some studies have attempted to introduce data-driven algorithms such as neural networks to improve prediction and control capabilities, but most methods rely on historical static data training and are difficult to adapt to the current water environment status in real time. At the same time, existing optimization algorithms are mostly general heuristic algorithms that lack modeling support for the evolution of water flow fields and the boundary-sensitive characteristics of control variables, resulting in slow convergence of control solutions and poor accuracy, which cannot meet the "fast response, low volatility, and high stability" control requirements in aquaculture systems.

[0005] More critically, most existing technologies lack adaptive feedback mechanisms. While some systems possess closed-loop control capabilities at the execution layer, their ability to correct errors in sensor data is limited. They are unable to dynamically compensate for factors such as equipment aging, sensor drift, and data noise, leading to significant deviations between monitored data and actual water quality. Furthermore, optimization algorithm parameters are mostly fixed and lack a dynamic update mechanism based on feedback data. This prevents self-correction based on the temporal and spatial variations of aquaculture water, thus compromising the overall stability and robustness of the system.

[0006] Therefore, how to provide an integrated water quality monitoring and treatment system for aquaculture is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0007] One purpose of the present invention is to propose an integrated water quality monitoring and treatment system and method for aquaculture. The present invention fully combines spatially distributed sensing technology, distributed unscented Kalman filtering algorithm, Lagrangian space-time fluid modeling and a multi-objective optimization mechanism based on space-time sensitive boundaries, and systematically constructs a closed-loop path from raw water quality data collection to execution feedback control. It has the advantages of high water quality monitoring accuracy, strong scheduling and control intelligence, and outstanding adaptability, and can effectively improve the stability and regulation efficiency of the aquaculture water environment.

[0008] The integrated water quality monitoring and treatment system for aquaculture according to an embodiment of the present invention includes the following steps:

[0009] The data acquisition module is used to automatically obtain raw data of dissolved oxygen, ammonia nitrogen, pH value, temperature and flow from multiple spatially distributed sensor nodes, and synchronously transmit the collected raw data to the data preprocessing module;

[0010] The data preprocessing module is used to perform noise removal, numerical correction and time series synchronization on the collected raw data to form a usable data set;

[0011] Water quality state estimation module, which is used to fuse available data sets based on the distributed unscented Kalman filter algorithm and output real-time multi-point water quality status;

[0012] The spatiotemporal fluid modeling module is used to construct a Lagrangian spatiotemporal fluid distribution model based on real-time multi-point water quality status and extract the velocity field and diffusion coefficient of each monitoring point;

[0013] The scheduling optimization module uses an improved multi-objective Pareto optimization algorithm based on time-space sensitive boundaries to process the velocity field, diffusion coefficient and water quality state output by the time-space fluid modeling module to generate aeration flow and dosing rate plans;

[0014] The execution module is used to send the aeration flow and dosing rate plans output by the scheduling optimization module to the blower, microporous aeration pipe and quantitative dosing pump to implement the oxygenation and dosing operations respectively;

[0015] The adaptive update module is used to collect multi-point water quality feedback data after execution, and adjust the filter gain and search parameters for the unscented Kalman filter algorithm and the improved multi-objective Pareto optimization algorithm based on time-space sensitive boundaries respectively to achieve closed-loop adaptive control.

[0016] Optionally, modules can be connected using the following methods:

[0017] S1, collects dissolved oxygen data, ammonia nitrogen data, pH data, temperature data and flow data output by multiple spatially distributed sensor nodes to construct the original water quality data set;

[0018] S2, performing noise filtering, numerical correction and time series alignment on the original water quality dataset to generate a structured water quality dataset;

[0019] S3, input the structured water quality data set into the distributed unscented Kalman filter algorithm for data fusion, and output real-time multi-point water quality status information;

[0020] S4. Construct a Lagrangian spatiotemporal fluid distribution model based on real-time multi-point water quality status information, and calculate the water velocity field and diffusion coefficient distribution matrix corresponding to each monitoring point;

[0021] S5. Inputting water quality status information, water velocity field, and diffusion coefficient distribution matrix into an improved multi-objective Pareto optimization algorithm model based on time-space sensitive boundaries to generate corresponding aeration flow scheduling matrix and agent delivery rate matrix;

[0022] S6. According to the aeration flow scheduling matrix and the drug delivery rate matrix, the blower, the microporous aeration tube and the quantitative drug delivery pump are driven to perform the oxygenation operation and the drug delivery operation;

[0023] S7. Collect multi-point water quality feedback data after the oxygenation and dosing operations are completed, input it into the control feedback mechanism, iteratively update the distributed unscented Kalman filter parameters and Pareto optimization search parameters, and generate a new round of scheduling matrix and control execution instructions.

[0024] The present invention constructs an integrated water quality monitoring and treatment system including distributed unscented Kalman filtering, Lagrangian space-time fluid modeling, and an improved multi-objective Pareto optimization algorithm based on space-time sensitive boundaries. It achieves real-time perception of water quality status at multiple points in aquaculture farms, dynamic modeling of fluid diffusion characteristics, and intelligent scheduling integration of aeration / medication operations. It can dynamically respond to water quality fluctuations in complex environments, accurately generate regulation strategies, and optimize control parameters in a closed loop. It effectively improves the system's response speed and regulation accuracy to water quality anomalies, and ensures ecological stability and growth benefits under high-density aquaculture conditions.

[0025] Optionally, S1 includes the following specific steps:

[0026] S11. Deploy multiple spatially distributed sensor nodes in the aquaculture area, each sensor node having a unique number and being equipped with dissolved oxygen, ammonia nitrogen, pH value, temperature, and flow collection sensors;

[0027] S12. Setting a unified sampling period and a synchronous clock, each sensor node synchronously obtains the water quality parameter value of the current location in each sampling period, forming a five-item indicator data record including a timestamp and a node number;

[0028] S13, structurally encapsulating the water quality parameters collected by each sensor node according to the time series, and constructing a data unit including the sensor node number, sampling time, dissolved oxygen value, ammonia nitrogen value, pH value, temperature value and flow value;

[0029] S14, aggregate the data units collected by all sensor nodes in the same sampling period into a multi-point observation sample, and continuously record them in ascending order of sampling time to form a preliminary sampling data set;

[0030] S15. All sampled data are spatially indexed and time-synchronized, and stored in a three-dimensional structure according to sensor node number, sampling time, and parameter type;

[0031] S16. After completing the periodic data collection tasks of all sensor nodes, an original water quality data set containing multiple time periods, multiple parameters, and multiple locations is generated.

[0032] This method deploys spatially distributed sensor nodes with unique numbering identifiers within aquaculture areas, combining a unified sampling cycle with a synchronized clock mechanism to achieve the simultaneous acquisition of key water quality parameters such as dissolved oxygen, ammonia nitrogen, pH, temperature, and flow at multiple points. It then constructs structured water quality data units using timestamps and node numbers, further organizing them into raw water quality datasets with a three-dimensional time-space-parameter structure. This method not only enables multi-point collaborative perception of water quality changes, but also exhibits excellent temporal continuity and spatial coverage, providing a high-quality, time-consistent data foundation for subsequent filtering fusion, fluid modeling, and control optimization, effectively improving the system's water quality monitoring accuracy and response efficiency in complex aquaculture environments.

[0033] Optionally, S2 includes the following specific steps:

[0034] S21, perform threshold removal and sliding window smoothing processing on each parameter sequence in the original water quality data set to remove abnormal mutation values ​​and sensor noise interference;

[0035] S22, performing dimensional normalization on the noise-removed data, uniformly mapping the five parameters of dissolved oxygen, ammonia nitrogen, pH value, temperature, and flow rate to the normalization interval [0,1], and constructing a normalized matrix;

[0036] S23. Set a unified reference time axis, complete and align the data uploaded by different sensor nodes according to the minimum common time interval, and form a synchronized sample sequence with consistent time.

[0037] S24, concatenating the normalized parameters of different nodes at each time step in the order of node numbers to construct a unified observation vector;

[0038] S25. Arrange all observation vectors in chronological order to construct a three-dimensional structured water quality dataset, where the first dimension is the time index, the second dimension is the spatial node index, and the third dimension is the parameter category index;

[0039] S26. Take the structured water quality dataset as the output result.

[0040] The present invention introduces a multi-stage data preprocessing mechanism to perform outlier removal and sliding window smoothing operations on the five types of parameter sequences in the original water quality data set, effectively filtering out mutation interference and sensor noise, and improving data stability and availability. It further adopts a unified normalization strategy to map water quality parameters of different dimensions to the same numerical interval, enhancing the compatibility of parameter fusion processing. Through reference time axis setting and time completion alignment operations, strict time synchronization of multi-node asynchronous data is achieved, and a unified observation vector is constructed according to the node number. Ultimately, the system organizes data according to time, space, and parameter dimensions to form a three-dimensional structured water quality data set, which not only ensures data integrity and standardization, but also provides highly consistent and timely data input for subsequent filtering fusion and modeling optimization, thereby significantly improving the data processing capability and practical effect of the overall system.

[0041] Optionally, S3 includes the following specific steps:

[0042] S31, receiving the structured water quality data set, constructing the five indicators of dissolved oxygen, ammonia nitrogen, pH, water temperature and flow observed by each monitoring node i at sampling time t into an observation vector

[0043] S32, each node initializes the state estimation mean vector and the covariance matrix Generate 2n+1 sigma points based on unscented transformation Defined as:

[0044]

[0045] Where λ is the expansion parameter, [·] m represents the mth column component of the Cholesky expansion;

[0046] S33, perform nonlinear state propagation on each sigma point to obtain the predicted state Calculate the weighted mean and forecast covariance as follows:

[0047]

[0048]

[0049] in, are the mean and covariance weights respectively;

[0050] S34. Map the sigma point to the observation space and calculate the observation estimate Constructing the predicted observed mean Observation covariance S i(t) and the state-observation cross covariance C xy，i (t);

[0051] S35. Update the local state estimate based on the following gain formula:

[0052]

[0053] Among them, K i (t) is the Kalman gain matrix;

[0054] S36, each node to the adjacent set broadcast With P i (t), according to the predefined fusion weight coefficient α ij , complete state vector fusion:

[0055] S37, all nodes output The water quality state estimation set at time t is composed of

[0056] The present invention realizes high-precision fusion and dynamic tracking of water quality parameters of multiple monitoring nodes by constructing a distributed state estimation mechanism based on untraceable transformation. The system takes a structured water quality data set as input, first generates an observation vector corresponding to the sampling time at each node, and initializes the prior estimate of the state mean and covariance, then generates a set of sigma points through untraceable sampling, uses nonlinear propagation to obtain state prediction, and combines weights to calculate the state prior mean and covariance, thereby maximizing information retention under nonlinear modeling. Subsequently, the system maps the sigma points to the observation space, estimates the observation expectation, covariance and co-covariance relationship between state and observation, and uses the Kalman gain to locally update the node state to improve the observation response capability and noise resistance. On this basis, a collaborative fusion mechanism of adjacent node states is introduced, and the final state estimate is updated by weighted average, effectively suppressing local anomalies and enhancing system consistency, thereby outputting a real-time water quality state estimation set of multiple nodes at the time, providing a dynamic and reliable data foundation for subsequent fluid modeling and optimization regulation.

[0057] Optionally, S4 includes the following specific steps:

[0058] S41. Obtain output real-time multi-point water quality status information set in Represents the state vector of the monitoring node numbered i at time t; let the position of each node in three-dimensional space be p i =(x i ,y i , z i );

[0059] S42. Set the time window length Δt and obtain the historical position sequence of node i in M+1 consecutive time steps. where t k =tk·Δt, construct the trajectory matrix of each node Each row is the three-dimensional position of a node at a certain historical moment;

[0060] S43, the trajectory matrix R i and the node's current state vector Perform association and construct a Lagrangian space-time fluid distribution model structure set in It represents the joint modeling result of multi-point fluid path and state of the system at time t;

[0061] S44, trajectory matrix R i The position difference of the last two lines is forward-differentiated to calculate the water velocity vector of node i at the current moment:

[0062]

[0063] in, is the water flow velocity at the current location of the node;

[0064] S45, trajectory matrix R i Perform covariance calculation and construct the diffusion coefficient distribution matrix of node i:

[0065]

[0066] in, is the diffusion tensor of the node’s current position;

[0067] S46, all the water velocity vectors corresponding to the nodes As the water velocity field output of the system at time t, all diffusion coefficient matrices The output is a diffusion coefficient distribution matrix.

[0068] The present invention introduces the Lagrangian spatiotemporal modeling mechanism, combines multi-point water quality status information with the historical position trajectory of the node, and constructs a dynamic fluid distribution model to achieve accurate characterization of the water velocity field and diffusion behavior. Specifically, the system generates a trajectory matrix based on the three-dimensional spatial position of each node and the position sequence of continuous moments, and combines it with the current state vector to construct a spatiotemporal coupled Lagrangian modeling structure; then, the current water velocity vector is derived using the position difference at the last two moments of the trajectory, and the node diffusion tensor is constructed based on the trajectory covariance, and the water velocity field and diffusion coefficient distribution matrix are output respectively. This method not only depicts the transmission trend of the local water body, but also reflects the differences in diffusion capacity in different regions, providing a high-precision input basis for subsequent optimization and regulation, and effectively enhancing the system's fluid perception ability and response efficiency in complex aquaculture environments.

[0069] Optionally, S5 includes the following specific steps:

[0070] S51. Construct a node set at time t The water quality state vector of the i-th node is recorded as The velocity vector is The diffusion coefficient matrix is Construct the joint input vector for each node: Where vec(·) indicates that the matrix columns are expanded into vectors first, so that all z i,t Constructing a set of tensor inputs

[0071] S52. Define the boundary perturbation function of node i as:

[0072]

[0073] in, is the spatial coordinate of node i, Represents the divergence of the local velocity field at the node, tr(D i,t ) represents the trace of the diffusion matrix, is the gradient of the water quality state vector in the spatial dimension, is a preset non-negative weight coefficient;

[0074] S53, take the input tensor Z t and the perturbation function set Enter the improved multi-objective Pareto optimization algorithm based on time-space sensitive boundaries, and assume that the scheduling variable vector is:

[0075] in is the aeration flow control quantity of node i, is the drug delivery rate of node i; define three objective functions:

[0076]

[0077] in, are the target dissolved oxygen and ammonia nitrogen concentration values ​​for node i, is the control input u t The model prediction value under ;

[0078] S54, construct boundary priority relationship < Φ , for any two scheduling solutions If satisfied: It is judged as And add it to the frontier solution set;

[0079] S55, for each scheduling solution Compute the set of participating nodes on the boundary Construct response weights: Calculate the search compression ratio factor control The search step range is scaled;

[0080] S56, from all satisfied dominance relations Φ The optimal solution that satisfies the following two conditions is selected from the scheduling solutions It is a non-inferior frontier solution on f1, f2, and f3; its boundary response index Minimum; vector The first N components constitute the aeration flow scheduling matrix: The last N components form the drug delivery rate matrix:

[0081] This method constructs a unified tensor input based on water quality state, fluid velocity, and diffusion structure. It introduces a perturbation term based on node location and defines a boundary sensitivity function to establish a multi-objective Pareto optimization model driven by spatiotemporal and temporal sensitive boundaries. First, the water quality state vector, velocity vector, and diffusion matrix of each node are concatenated into a column-major vector to construct an input tensor. Then, based on the divergence, diffusion trace, and state gradient of the node's local velocity field, a perturbation function is defined to reflect the sensitivity of boundary effects to water body regulation. Subsequently, the target difference between dissolved oxygen and ammonia nitrogen, and the energy consumption of aeration and dosing are set as three objective functions. By constructing a boundary-priority dominance relationship, the frontier scheduling solution set is screened, and the compression ratio of the node set participating in the boundary response is calculated, dynamically shrinking the search step size. Finally, the unique scheduling solution with the minimum boundary response is selected from the non-inferior solutions, and the aeration flow scheduling matrix and the drug delivery rate matrix are extracted and generated. This method effectively improves the accuracy and control sensitivity of water body responses in boundary areas, and possesses strong regulation adaptability and local response capabilities.

[0082] Optionally, S6 includes the following specific steps:

[0083] S61, receiving aeration flow scheduling matrix and the drug delivery rate matrix

[0084] S62, the matrix The flow control value corresponding to each node in Input blower control unit, which controls the air output power through pulse width modulation signal;

[0085] S63, introducing the blower output airflow into the microporous aeration pipe through the gas distribution pipeline, and then introducing it into the bottom of the aquaculture water body according to the layout position of each node, forming a bubble-type oxygen diffusion structure;

[0086] S64, the matrix The rate value corresponding to each node in Input to the quantitative dosing pump control channel to control the solution delivery speed according to the preset drug concentration coefficient;

[0087] S65, connecting the output liquid flow of the dosing pump to the delivery points around each monitoring node through the partition pipeline, and the liquid spray nozzle releases the liquid in a distributed manner in different areas;

[0088] S66. The controller periodically records the blower output status and the dosing pump flow status, and compares them with the execution instruction matrix corresponding to the current time t in real time to complete the control task execution closed loop in the current cycle.

[0089] After receiving the aeration flow scheduling matrix and the drug delivery rate matrix, the present invention completes the oxygenation and drug delivery operations based on the node-level control method. First, the flow control value of each node in the matrix is ​​input into the corresponding blower control channel, and the air output power is adjusted by pulse width modulation. The air flow is driven through the distribution pipeline to be introduced into the microporous aeration pipe arranged at the bottom of the culture pond, forming a partitioned bubble diffusion structure to increase the dissolved oxygen concentration of the water body. Subsequently, the drug delivery rate of each node is input into the corresponding control port of the quantitative drug delivery pump, and the drug delivery rate is adjusted according to the set drug concentration parameter. The drug delivery point around the node is introduced through the partitioned pipeline, and the distributed release is completed by the nozzle. The system records the operating status of the blower and the drug delivery pump within the control cycle, and compares it with the scheduling matrix of the current cycle to complete the distributed execution closed loop under each node, ensuring that the control result is highly consistent with the target solution.

[0090] Optionally, S7 includes the following specific steps:

[0091] S71. After each scheduling cycle, water quality feedback vectors are collected node by node after the oxygenation and dosing operations are completed.

[0092] S72. Extract the mean of the distributed unscented Kalman filter prediction observations and the corresponding prediction covariance P i (t|t-1), calculate the residual vector:

[0093] S73, construct the observation noise covariance matrix R with the residual vector i (t) = ε i (t)ε i (t) T , and based on P i (t|t-1) and the observation matrix Recalculate the Kalman filter gain matrix:

[0094]

[0095] S74, using K i (t) Prediction of previous state Update and obtain the posterior state estimate vector: And synchronously update the state covariance: P i (t|t)=(IK i (t)H i )P i (t|t-1), where is the identity matrix;

[0096] S75, construct the boundary response function value based on the square of the residual two norm of each node And perform a linear update on the search weights in the Pareto algorithm:

[0097] ω i (t+1)=(1-β)ω i (t)+βφ i (t);

[0098] in, is the weight of the previous cycle, β∈(0,1) is the update coefficient;

[0099] S76, the updated state estimate The perturbation function set {φ i (t)} and the search weight set {ω i (t+1)} is based on the improved multi-objective Pareto optimization algorithm of time-space sensitive boundaries, which performs objective function evaluation, boundary priority dominance relationship construction, non-dominated solution screening and search compression ratio factor calculation to generate a new round of scheduling variable vector Ψ * (t+1)=[a1,…,a N ,r1,…,rN ] T , where a i is the aeration flow rate of node i, r i is the drug delivery rate of node i;

[0100] S77, will * (t+1) is split into aeration flow scheduling matrix and the drug delivery rate matrix As the input of the execution module in the next cycle.

[0101] This method combines the residuals from the distributed unscented Kalman filter with the state covariance to construct a feedback evaluation path, and introduces a node-level boundary response function to update the Pareto optimization search weights, achieving adaptive adjustment of the scheduling strategy within the feedback closed loop. This approach possesses a clear evolutionary structure and high-frequency update capabilities. Furthermore, the observation residuals are used to drive the dynamic adjustment of the dominance relationship and compression factor in the optimization model, ensuring that each round of scheduling variable generation is based on the latest system state and response performance. This improves the response speed and accuracy of the scheduling strategy to sudden changes in water quality conditions, effectively enhancing the robustness, adaptability, and energy efficiency of the aquaculture environmental control system, and enabling refined and dynamic iterative optimization of control strategies in complex multi-objective scenarios.

[0102] The beneficial effects of the present invention are:

[0103] The present invention systematically introduces a distributed unscented Kalman filter algorithm and a Lagrangian space-time fluid modeling method by constructing an integrated water quality monitoring and treatment system, effectively solving the problems of existing water quality monitoring technologies such as insufficient multi-point data fusion, low real-time estimation accuracy, and weak flow field dynamic modeling capabilities. The proposed distributed unscented Kalman filter algorithm can achieve real-time fusion estimation of water quality status under the condition of asynchronous sampling of multiple sensor nodes, and has the advantages of strong noise resistance and accurate state prediction. Combined with the Lagrangian space-time modeling method, the system can dynamically construct the flow velocity vector and diffusion tensor of any monitoring point in the water body, significantly improving the ability to characterize complex water body behaviors and providing a data basis for precise regulation.

[0104] This paper innovatively proposes an improved multi-objective Pareto optimization algorithm based on spatiotemporal sensitive boundaries. This algorithm incorporates a water boundary perturbation function into the scheduling control process, enabling the scheduling model to prioritize areas sensitive to boundary conditions, thereby more rationally allocating aeration flow and agent delivery rates. This method not only fully considers the interactions between water quality, velocity field, and diffusion coefficient, but also dynamically adjusts scheduling variables by optimizing dominating relationships and response compression factors. This overcomes the limitations of traditional scheduling algorithms, which often focus on a single optimization objective and lack control precision, achieving a more targeted and locally responsive control solution.

[0105] The present invention introduces a feedback-driven adaptive update mechanism, using multi-point water quality feedback data to iteratively update the observation noise covariance matrix of the unscented Kalman filter and the search weight factor of the Pareto optimization algorithm, thereby achieving dynamic self-optimization of the monitoring and control strategy. The system can automatically adjust the filter gain and scheduling search path according to the current control execution effect, and has significant environmental adaptability and control robustness. Compared with existing methods, the present invention not only improves the intelligence level of water quality monitoring and control, but also significantly reduces the response lag and miscontrol risks caused by data drift and model rigidity, providing a stable, efficient, low-intervention human-machine collaborative water quality management solution for high-density aquaculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0106] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0107] Figure 1 This is a flow chart of the integrated water quality monitoring and treatment system for aquaculture proposed by the present invention;

[0108] Figure 2 This is a workflow diagram of water quality state estimation based on distributed unscented Kalman filtering proposed by the present invention;

[0109] Figure 3 This is the scheduling flowchart of the improved multi-objective Pareto optimization algorithm based on time-space sensitive boundaries proposed in this invention. DETAILED DESCRIPTION

[0110] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0111] refer to Figure 1-3 , an integrated water quality monitoring and treatment system for aquaculture, including the following steps:

[0112] The data acquisition module is used to automatically obtain raw data of dissolved oxygen, ammonia nitrogen, pH value, temperature and flow from multiple spatially distributed sensor nodes, and synchronously transmit the collected raw data to the data preprocessing module;

[0113] The data preprocessing module is used to perform noise removal, numerical correction and time series synchronization on the collected raw data to form a usable data set;

[0114] Water quality state estimation module, which is used to fuse available data sets based on the distributed unscented Kalman filter algorithm and output real-time multi-point water quality status;

[0115] The spatiotemporal fluid modeling module is used to construct a Lagrangian spatiotemporal fluid distribution model based on real-time multi-point water quality status and extract the velocity field and diffusion coefficient of each monitoring point;

[0116] The scheduling optimization module uses an improved multi-objective Pareto optimization algorithm based on time-space sensitive boundaries to process the velocity field, diffusion coefficient and water quality state output by the time-space fluid modeling module to generate aeration flow and dosing rate plans;

[0117] The execution module is used to send the aeration flow and dosing rate plans output by the scheduling optimization module to the blower, microporous aeration pipe and quantitative dosing pump to implement the oxygenation and dosing operations respectively;

[0118] The adaptive update module is used to collect multi-point water quality feedback data after execution, and adjust the filter gain and search parameters for the unscented Kalman filter algorithm and the improved multi-objective Pareto optimization algorithm based on time-space sensitive boundaries respectively to achieve closed-loop adaptive control.

[0119] In this embodiment, the modules are connected through the following methods:

[0120] S1, collects dissolved oxygen data, ammonia nitrogen data, pH data, temperature data and flow data output by multiple spatially distributed sensor nodes to construct the original water quality data set;

[0121] S2, performing noise filtering, numerical correction and time series alignment on the original water quality dataset to generate a structured water quality dataset;

[0122] S3, input the structured water quality data set into the distributed unscented Kalman filter algorithm for data fusion, and output real-time multi-point water quality status information;

[0123] S4. Construct a Lagrangian spatiotemporal fluid distribution model based on real-time multi-point water quality status information, and calculate the water velocity field and diffusion coefficient distribution matrix corresponding to each monitoring point;

[0124] S5. Inputting water quality status information, water velocity field, and diffusion coefficient distribution matrix into an improved multi-objective Pareto optimization algorithm model based on time-space sensitive boundaries to generate corresponding aeration flow scheduling matrix and agent delivery rate matrix;

[0125] S6. According to the aeration flow scheduling matrix and the drug delivery rate matrix, the blower, the microporous aeration tube and the quantitative drug delivery pump are driven to perform the oxygenation operation and the drug delivery operation;

[0126] S7. Collect multi-point water quality feedback data after the oxygenation and dosing operations are completed, input it into the control feedback mechanism, iteratively update the distributed unscented Kalman filter parameters and Pareto optimization search parameters, and generate a new round of scheduling matrix and control execution instructions.

[0127] In this embodiment, S1 includes the following specific steps:

[0128] S11. Deploy multiple spatially distributed sensor nodes in the aquaculture area, each sensor node having a unique number and being equipped with dissolved oxygen, ammonia nitrogen, pH value, temperature, and flow collection sensors;

[0129] S12. Setting a unified sampling period and a synchronous clock, each sensor node synchronously obtains the water quality parameter value of the current location in each sampling period, forming a five-item indicator data record including a timestamp and a node number;

[0130] S13, structurally encapsulating the water quality parameters collected by each sensor node according to the time series, and constructing a data unit including the sensor node number, sampling time, dissolved oxygen value, ammonia nitrogen value, pH value, temperature value and flow value;

[0131] S14, aggregate the data units collected by all sensor nodes in the same sampling period into a multi-point observation sample, and continuously record them in ascending order of sampling time to form a preliminary sampling data set;

[0132] S15. All sampled data are spatially indexed and time-synchronized, and stored in a three-dimensional structure according to sensor node number, sampling time, and parameter type;

[0133] S16. After completing the periodic data collection tasks of all sensor nodes, an original water quality data set containing multiple time periods, multiple parameters, and multiple locations is generated.

[0134] In this embodiment, S2 includes the following specific steps:

[0135] S21, perform threshold removal and sliding window smoothing processing on each parameter sequence in the original water quality data set to remove abnormal mutation values ​​and sensor noise interference;

[0136] S22, performing dimensional normalization on the noise-removed data, uniformly mapping the five parameters of dissolved oxygen, ammonia nitrogen, pH value, temperature, and flow rate to the normalization interval [0,1], and constructing a normalized matrix;

[0137] S23. Set a unified reference time axis, complete and align the data uploaded by different sensor nodes according to the minimum common time interval, and form a synchronized sample sequence with consistent time.

[0138] S24, concatenating the normalized parameters of different nodes at each time step in the order of node numbers to construct a unified observation vector;

[0139] S25. Arrange all observation vectors in chronological order to construct a three-dimensional structured water quality dataset, where the first dimension is the time index, the second dimension is the spatial node index, and the third dimension is the parameter category index;

[0140] S26. Take the structured water quality dataset as the output result.

[0141] In this embodiment, S3 includes the following specific steps:

[0142] S31, receiving the structured water quality data set, constructing the five indicators of dissolved oxygen, ammonia nitrogen, pH, water temperature and flow observed by each monitoring node i at sampling time t into an observation vector

[0143] S32, each node initializes the state estimation mean vector and the covariance matrix Generate 2n+1 sigma points based on unscented transformation Defined as:

[0144]

[0145] Where λ is the expansion parameter, [·] m represents the mth column component of the Cholesky expansion;

[0146] S33, perform nonlinear state propagation on each sigma point to obtain the predicted state Calculate the weighted mean and forecast covariance as follows:

[0147]

[0148] in, are the mean and covariance weights respectively;

[0149] S34. Map the sigma point to the observation space and calculate the observation estimate Constructing the predicted observed mean Observation covariance S i (t) and the state-observation cross covariance C xy，i (t);

[0150] S35. Update the local state estimate based on the following gain formula:

[0151]

[0152] Among them, K i (t) is the Kalman gain matrix;

[0153] S36, each node to the adjacent set broadcast With P i (t), according to the predefined fusion weight coefficient α ij , complete state vector fusion:

[0154] S37, all nodes output The water quality state estimation set at time t is composed of

[0155] In this embodiment, S4 includes the following specific steps:

[0156] S41. Obtain output real-time multi-point water quality status information set in Represents the state vector of the monitoring node numbered i at time t; let the position of each node in three-dimensional space be p i =(x i ,y i , z i );

[0157] S42. Set the time window length Δt and obtain the historical position sequence of node i in M+1 consecutive time steps. where t k =tk·Δt, construct the trajectory matrix of each node Each row is the three-dimensional position of a node at a certain historical moment;

[0158] S43, the trajectory matrix R i and the node's current state vector Perform association and construct a Lagrangian space-time fluid distribution model structure set in It represents the joint modeling result of multi-point fluid path and state of the system at time t;

[0159] S44, trajectory matrix R i The position difference of the last two lines is forward-differentiated to calculate the water velocity vector of node i at the current moment:

[0160]

[0161] in, is the water flow velocity at the current location of the node;

[0162] S45, trajectory matrix R i Perform covariance calculation and construct the diffusion coefficient distribution matrix of node i:

[0163]

[0164] in, is the diffusion tensor of the node’s current position;

[0165] S46, all the water velocity vectors corresponding to the nodes As the water velocity field output of the system at time t, all diffusion coefficient matrices The output is a diffusion coefficient distribution matrix.

[0166] In this embodiment, S5 includes the following specific steps:

[0167] S51. Construct a node set at time t The water quality state vector of the i-th node is recorded as The velocity vector is The diffusion coefficient matrix is Construct the joint input vector for each node: Where vec(·) indicates that the matrix columns are expanded into vectors first; let all z i，t Constructing a set of tensor inputs

[0168] S52. Define the boundary perturbation function of node i as:

[0169]

[0170] in, is the spatial coordinate of node i, Represents the divergence of the local velocity field at the node, tr(D i，t ) represents the trace of the diffusion matrix, is the gradient of the water quality state vector in the spatial dimension, is a preset non-negative weight coefficient;

[0171] S53, take the input tensor Z t and the perturbation function set Enter the improved multi-objective Pareto optimization algorithm based on time-space sensitive boundaries, and assume that the scheduling variable vector is:

[0172] in is the aeration flow control quantity of node i, is the drug delivery rate of node i; define three objective functions:

[0173]

[0174] in, are the target dissolved oxygen and ammonia nitrogen concentration values ​​for node i, is the control input u tThe model prediction value under ;

[0175] S54, construct boundary priority relationship < Φ , for any two scheduling solutions If satisfied: It is judged as And add it to the frontier solution set;

[0176] S55, for each scheduling solution Compute the set of participating nodes on the boundary Construct response weights: Calculate the search compression ratio factor control The search step range is scaled;

[0177] S56, from all satisfied dominance relations Φ The optimal solution that satisfies the following two conditions is selected from the scheduling solutions It is a non-inferior frontier solution on f1, f2, and f3; its boundary response index Minimum; vector The first N components constitute the aeration flow scheduling matrix: The last N components form the drug delivery rate matrix:

[0178] In this embodiment, S6 includes the following specific steps:

[0179] S61, receiving aeration flow scheduling matrix and the drug delivery rate matrix

[0180] S62, the matrix The flow control value corresponding to each node in Input blower control unit, which controls the air output power through pulse width modulation signal;

[0181] S63, introducing the blower output airflow into the microporous aeration pipe through the gas distribution pipeline, and then introducing it into the bottom of the aquaculture water body according to the layout position of each node, forming a bubble-type oxygen diffusion structure;

[0182] S64, the matrix The rate value corresponding to each node in Input to the quantitative dosing pump control channel to control the solution delivery speed according to the preset drug concentration coefficient;

[0183] S65, connecting the output liquid flow of the dosing pump to the delivery points around each monitoring node through the partition pipeline, and the liquid spray nozzle releases the liquid in a distributed manner in different areas;

[0184] S66. The controller periodically records the blower output status and the dosing pump flow status, and compares them with the execution instruction matrix corresponding to the current time t in real time to complete the control task execution closed loop in the current cycle.

[0185] In this embodiment, S7 includes the following specific steps:

[0186] S71. After each scheduling cycle, water quality feedback vectors are collected node by node after the oxygenation and dosing operations are completed.

[0187] S72. Extract the mean of the distributed unscented Kalman filter prediction observations and the corresponding prediction covariance P i (t|t-1), calculate the residual vector:

[0188] S73, construct the observation noise covariance matrix R with the residual vector i (t) = ε i (t)ε i (t) T , and based on P i (t|t-1) and the observation matrix Recalculate the Kalman filter gain matrix:

[0189]

[0190] S74, using K i (t) Prediction of previous state Update and obtain the posterior state estimate vector: And synchronously update the state covariance: P i (t|t)=(IK i (t)H i )P i (t|t-1), where is the identity matrix;

[0191] S75, construct the boundary response function value based on the square of the residual two norm of each node And perform a linear update on the search weights in the Pareto algorithm:

[0192] ω i (t+1)=(1-β)ω i (t)+βφ i (t);

[0193] in, is the weight of the previous cycle, β∈(0,1) is the update coefficient;

[0194] S76, the updated state estimate The perturbation function set {φ i (t)} and the search weight set {ω i (t+1)} is based on the improved multi-objective Pareto optimization algorithm of time-space sensitive boundaries, which performs objective function evaluation, boundary priority dominance relationship construction, non-dominated solution screening and search compression ratio factor calculation to generate a new round of scheduling variable vector Ψ * (t+1)=[a1,…,a N , r1,…,r N ] T , where a i is the aeration flow rate of node i, r i is the drug delivery rate of node i;

[0195] S77, will * (t+1) is split into aeration flow scheduling matrix and the drug delivery rate matrix As the input of the execution module in the next cycle.

[0196] Example 1:

[0197] In order to verify the feasibility of the present invention in practice, the present invention was applied to a newly built high-density recirculating aquaculture demonstration base in an inland area for a two-month field test. The breeding base adopts a three-layer three-dimensional breeding pond system equipped with a fully enclosed circulating water treatment unit, and conducts intensive winter breeding operations from October to December 2024. The test subjects are large-sized high-protein fish with an average size of 22 cm in length and 380 g in weight, and the breeding density is as high as 50 kg / m 3 The total aquaculture water capacity is about 1200m 3 , 16 monitoring nodes are deployed in the pool to collect water quality and flow data in real time and upload them to the control platform.

[0198] In the traditional aquaculture control mode, the execution cycles and frequencies of oxygenation, filtration and feeding instructions are preset based on manual experience, resulting in delayed response, frequent fluctuations in water quality, and high aquaculture risks. In particular, when the density changes or the temperature suddenly changes, the system is difficult to adapt quickly, and a sudden drop in dissolved oxygen or accumulation of ammonia nitrogen often occurs, resulting in reduced fish food intake, slow growth, and even inducing stress diseases.

[0199] To address these issues, the testing team deployed the proposed factory-scale high-density aquaculture method based on circulating water treatment within the system and integrated its control instructions into the edge controller, enabling real-time control of the aerator, filter pump, and automatic feeder. The system continuously collects key parameters such as pH, DO, ammonia nitrogen, water temperature, and flow rate over a five-minute period. A dual-branch feature extraction module extracts joint topological and physical features, which are then integrated into the environmental state modeling network to generate a state vector, which in turn drives the output of the control strategy within the dual-layer meta-reinforcement learning collaborative control framework.

[0200] During testing, the system executed over 4,000 continuous control cycles, significantly improving its regulation efficiency and water quality stability under low-temperature aquaculture conditions of 10-14°C. Compared to a control group of aquaculture ponds not equipped with the present invention during the same period, the group implementing the present invention demonstrated significant advantages in key indicators such as oxygenation efficiency, water quality stabilization rate, and feed conversion rate.

[0201] The following is a typical data statistics table during the test:

[0202] Table 1 Comparative performance data of the method of the present invention and the traditional method in high-density aquaculture

[0203]

[0204] As can be seen from the above table, the method of the present invention is superior to traditional control methods in many aspects. First, in terms of dissolved oxygen control, the present invention significantly reduces the DO fluctuation range, with the average daily fluctuation amplitude reduced from 3.4 mg / L in the traditional method to 1.2 mg / L, and the stability improved by 64.7%. In terms of ammonia nitrogen control, under traditional methods, a strong water change is usually required every 6 to 7 days to eliminate accumulation. However, the present invention extends this period to more than 10 days through intelligent control of filtration and oxygenation, effectively reducing the frequency of water changes and water resource waste.

[0205] In terms of pH control, the method proposed in this paper dynamically adjusts the buffering capacity of the circulatory system, reducing the diurnal pH fluctuation by 58.5%, significantly reducing the risk of alkali and acid toxicity. In terms of control response, the state-based command reasoning proposed in this paper significantly shortens feeding response time from the traditional 18 minutes to less than 5 minutes, significantly improving fish feeding initiative and feed utilization efficiency.

[0206] In terms of feed conversion rate, through the system's optimized dynamic feeding strategy, the FCR was reduced from 1.68 to 1.43, reducing feed waste by over 15%, directly lowering aquaculture costs. In terms of health performance, the system achieves precise oxygenation and ammonia nitrogen regulation, significantly reducing the probability of stress-induced disease outbreaks and reducing fish mortality to 1.6%, a decrease of over 50% compared to traditional methods.

[0207] In summary, the present invention significantly improves the intelligent control capability and water quality stability of the aquaculture system under high-density recirculating aquaculture conditions, while reducing energy consumption, improving feed efficiency and survival rate, verifying that the present invention has good practicality and promotion value.

[0208] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. An integrated water quality monitoring and treatment system for aquaculture, characterized by: include: The data acquisition module is used to automatically obtain raw data of dissolved oxygen, ammonia nitrogen, pH value, temperature and flow from multiple spatially distributed sensor nodes, and synchronously transmit the collected raw data to the data preprocessing module; The data preprocessing module is used to perform noise removal, numerical correction and time series synchronization on the collected raw data to form a usable data set; Water quality state estimation module, which is used to fuse available data sets based on the distributed unscented Kalman filter algorithm and output real-time multi-point water quality status; The spatiotemporal fluid modeling module is used to construct a Lagrangian spatiotemporal fluid distribution model based on real-time multi-point water quality status and extract the velocity field and diffusion coefficient of each monitoring point; The scheduling optimization module uses an improved multi-objective Pareto optimization algorithm based on time-space sensitive boundaries to process the velocity field, diffusion coefficient and water quality state output by the time-space fluid modeling module to generate aeration flow and dosing rate plans; The execution module is used to send the aeration flow and dosing rate plan output by the scheduling optimization module to the blower, microporous aeration pipe and quantitative dosing pump to implement the oxygenation and dosing operations respectively; The adaptive update module is used to collect multi-point water quality feedback data after execution, and adjust the filter gain and search parameters for the unscented Kalman filter algorithm and the improved multi-objective Pareto optimization algorithm based on time-space sensitive boundaries respectively to achieve closed-loop adaptive control.

2. The integrated water quality monitoring and treatment system for aquaculture according to claim 2 is characterized in that: The modules are implemented as follows: S1, collects dissolved oxygen data, ammonia nitrogen data, pH data, temperature data and flow data output by multiple spatially distributed sensor nodes to construct the original water quality data set; S2, performing noise filtering, numerical correction and time series alignment on the original water quality dataset to generate a structured water quality dataset; S3, input the structured water quality data set into the distributed unscented Kalman filter algorithm for data fusion, and output real-time multi-point water quality status information; S4. Construct a Lagrangian spatiotemporal fluid distribution model based on real-time multi-point water quality status information, and calculate the water velocity field and diffusion coefficient distribution matrix corresponding to each monitoring point; S5. Inputting water quality status information, water velocity field, and diffusion coefficient distribution matrix into an improved multi-objective Pareto optimization algorithm model based on time-space sensitive boundaries to generate corresponding aeration flow scheduling matrix and agent delivery rate matrix; S6. According to the aeration flow scheduling matrix and the drug delivery rate matrix, the blower, the microporous aeration tube and the quantitative drug delivery pump are driven to perform the oxygenation operation and the drug delivery operation; S7. Collect multi-point water quality feedback data after the oxygenation and dosing operations are completed, input it into the control feedback mechanism, iteratively update the distributed unscented Kalman filter parameters and Pareto optimization search parameters, and generate a new round of scheduling matrix and control execution instructions.

3. The integrated water quality monitoring and treatment system for aquaculture according to claim 2 is characterized in that: Said S1 specifically includes: S11. Deploy multiple spatially distributed sensor nodes in the aquaculture area, each sensor node having a unique number and being equipped with dissolved oxygen, ammonia nitrogen, pH value, temperature and flow collection sensors; S12. Setting a unified sampling period and a synchronous clock, each sensor node synchronously obtains the water quality parameter value of the current location in each sampling period, forming a five-item indicator data record including a timestamp and a node number; S13, structurally encapsulating the water quality parameters collected by each sensor node according to the time series, and constructing a data unit including the sensor node number, sampling time, dissolved oxygen value, ammonia nitrogen value, pH value, temperature value and flow value; S14, aggregate the data units collected by all sensor nodes in the same sampling period into a multi-point observation sample, and continuously record them in ascending order of sampling time to form a preliminary sampling data set; S15. All sampled data are spatially indexed and time-synchronized, and stored in a three-dimensional structure according to sensor node number, sampling time, and parameter type; S16. After completing the periodic data collection tasks of all sensor nodes, an original water quality data set containing multiple time periods, multiple parameters, and multiple locations is generated.

4. The integrated water quality monitoring and treatment system for aquaculture according to claim 2 is characterized in that: The S2 specifically includes: S21, perform threshold removal and sliding window smoothing processing on each parameter sequence in the original water quality data set to remove abnormal mutation values ​​and sensor noise interference; S22, performing dimensional normalization on the noise-removed data, uniformly mapping the five parameters of dissolved oxygen, ammonia nitrogen, pH value, temperature, and flow rate to the normalization interval [0,1], and constructing a normalized matrix; S23. Set a unified reference time axis, complete and align the data uploaded by different sensor nodes according to the minimum common time interval, and form a synchronized sample sequence with consistent time. S24, concatenating the normalized parameters of different nodes at each time step in the order of node numbers to construct a unified observation vector; S25. Arrange all observation vectors in chronological order to construct a three-dimensional structured water quality dataset, where the first dimension is the time index, the second dimension is the spatial node index, and the third dimension is the parameter category index; S26. Take the structured water quality dataset as the output result.

5. The integrated water quality monitoring and treatment system for aquaculture according to claim 2 is characterized in that: The S3 specifically includes: S31, receiving the structured water quality data set, constructing the five indicators of dissolved oxygen, ammonia nitrogen, pH, water temperature and flow observed by each monitoring node i at sampling time t into an observation vector S32, each node initializes the state estimation mean vector and the covariance matrix Generate 2n+1 sigma points based on unscented transformation Defined as: Where λ is the expansion parameter, [·] m represents the mth column component of the Cholesky expansion; S33, perform nonlinear state propagation on each sigma point to obtain the predicted state Calculate the weighted mean and forecast covariance as follows: in, are the mean and covariance weights respectively; S34. Map the sigma point to the observation space and calculate the observation estimate Constructing the predicted observed mean Observation covariance S i (t) and the state-observation cross covariance C xy,i (t); S35. Update the local state estimate based on the following gain formula: Among them, K i (t) is the Kalman gain matrix; S36, each node to the adjacent set broadcast With P i (t), according to the predefined fusion weight coefficient α ij , complete state vector fusion: S37, all nodes output The water quality state estimation set at time t is composed of 6. The integrated water quality monitoring and treatment system for aquaculture according to claim 2 is characterized in that: The S4 specifically includes: S41. Obtain output real-time multi-point water quality status information set in Represents the state vector of the monitoring node numbered i at time t, and assumes that the position of each node in three-dimensional space is p i =(x i ,y i , z i ); S42. Set the time window length Δt and obtain the historical position sequence of node i in M+1 consecutive time steps. where t k =tk·Δt, construct the trajectory matrix of each node Each row is the three-dimensional position of a node at a certain historical moment; S43, the trajectory matrix R i and the node's current state vector Perform association and construct a Lagrangian space-time fluid distribution model structure set in represents the joint modeling result of multi-point fluid path and state at time t; S44, trajectory matrix R i The position difference of the last two lines is forward-differentiated to calculate the water velocity vector of node i at the current moment: in, is the water flow velocity at the current location of the node; S45, trajectory matrix R i Perform covariance calculation and construct the diffusion coefficient distribution matrix of node i: in, is the diffusion tensor of the node’s current position; S46, all the water velocity vectors corresponding to the nodes As the water velocity field output of the system at time t, all diffusion coefficient matrices The output is a diffusion coefficient distribution matrix.

7. The integrated water quality monitoring and treatment system for aquaculture according to claim 2 is characterized in that: The S5 comprises the following steps: S51. Construct a node set at time t The water quality state vector of the i-th node is recorded as The velocity vector is The diffusion coefficient matrix is Construct the joint input vector for each node: Where vec(·) indicates that the matrix columns are expanded into vectors first, so that all z i,t Constructing a set of tensor inputs S52. Define the boundary perturbation function of node i as: in, is the spatial coordinate of node i, Represents the divergence of the local velocity field at the node, tr(D i,t ) represents the trace of the diffusion matrix, is the gradient of the water quality state vector in the spatial dimension, is a preset non-negative weight coefficient; S53, take the input tensor Z t and the perturbation function set Enter the improved multi-objective Pareto optimization algorithm based on time-space sensitive boundaries, and assume that the scheduling variable vector is: in is the aeration flow control quantity of node i, For the drug delivery rate of node i, three objective functions are defined: in, are the target dissolved oxygen and ammonia nitrogen concentration values ​​for node i, is the control input u t The model prediction value under ; S54, construct boundary priority relationship < Φ , for any two scheduling solutions If satisfied: It is judged as And add it to the frontier solution set; S55, for each scheduling solution Compute the set of participating nodes on the boundary Construct response weights: Calculate the search compression ratio factor control The search step range is scaled; S56. Select the only optimal solution that meets the following two conditions from all scheduling solutions that satisfy the dominance relationship < Φ It is a non-inferior frontier solution on f1, f2, and f3; its boundary response index Minimum; vector The first N components constitute the aeration flow scheduling matrix: The last N components form the drug delivery rate matrix:

8. The integrated water quality monitoring and treatment system for aquaculture according to claim 2 is characterized in that: The S6 comprises the following steps: S61, receiving aeration flow scheduling matrix and the drug delivery rate matrix S62, the matrix The flow control value corresponding to each node in Input blower control unit, which controls the air output power through pulse width modulation signal; S63, introducing the blower output airflow into the microporous aeration pipe through the gas distribution pipeline, and then introducing it into the bottom of the aquaculture water body according to the layout position of each node, forming a bubble-type oxygen diffusion structure; S64, the matrix The rate value corresponding to each node in Input to the quantitative dosing pump control channel to control the solution delivery speed according to the preset drug concentration coefficient; S65, connecting the output liquid flow of the dosing pump to the delivery points around each monitoring node through the partition pipeline, and the liquid spray nozzle releases the liquid in a distributed manner in different areas; S66. The controller periodically records the blower output status and the dosing pump flow status, and compares them with the execution instruction matrix corresponding to the current time t in real time to complete the control task execution closed loop in the current cycle.

9. The integrated water quality monitoring and treatment system for aquaculture according to claim 2, characterized in that: The S7 comprises the following steps: S71. After each scheduling cycle, water quality feedback vectors are collected node by node after the oxygenation and dosing operations are completed. S72. Extract the mean of the distributed unscented Kalman filter prediction observations and the corresponding prediction covariance P i (t|t-1), calculate the residual vector: S73. Construct the observation noise covariance matrix using the residual vector Based on P i (t|t-1) and the observation matrix Recalculate the Kalman filter gain matrix: S74, using K i (t) Prediction of previous state Update and obtain the posterior state estimate vector: And synchronously update the state covariance: P i (t|t)=(IK i (t)H i )P i (t|t-1), where is the identity matrix; S75, construct the boundary response function value based on the square of the residual norm of each node And perform a linear update on the search weights in the Pareto algorithm: oh i (t+1)=(1-β)ω i (t)+βφ i (t); in, is the weight of the previous cycle, β∈(0,1) is the update coefficient; S76, the updated state estimate The perturbation function set {φ i (t)} and the search weight set {ω i (t+1)} is based on the improved multi-objective Pareto optimization algorithm of time-space sensitive boundaries, which performs objective function evaluation, boundary priority dominance relationship construction, non-dominated solution screening and search compression ratio factor calculation to generate a new round of scheduling variable vector Ψ * (t+1)=[a1,…,a N , r1,…,r N ] T , where a i is the aeration flow rate of node i, r i is the drug delivery rate of node i; S77, will * (t+1) is split into the aeration flow scheduling matrix and the drug delivery rate matrix As the input of the execution module in the next cycle.

Images