A multimodal fusion intelligent closed-loop system for monitoring, evaluating, and regulating aquaculture water bodies

Abstract

This invention relates to the field of aquaculture, specifically to a multimodal fusion intelligent closed-loop system for monitoring, evaluating, and regulating aquaculture water bodies. The system includes a data processing module, a mapping module, a water quality analysis module, a scoring module, an evaluation module, and a regulation module. The data processing module acquires historical monitoring data and constructs training samples. The mapping module builds parameter correlation graphs. The water quality analysis module includes a water quality analysis model to obtain predicted values ​​of various parameters for aquaculture water bodies during a preset period. The scoring module calculates seasonal scores. The evaluation module queries an evaluation level table to obtain the evaluation level corresponding to the seasonal score. If the evaluation level is excellent or good, routine maintenance is performed; otherwise, a regulation command is sent to the regulation module. The regulation module makes regulatory decisions for the aquaculture water bodies. Accurate evaluation based on seasonal characteristics helps improve the accuracy of risk warnings.

Description

Technical Field

[0001] This invention relates to the field of aquaculture, specifically to an intelligent closed-loop system for monitoring, evaluating, and regulating aquaculture water bodies using a multimodal fusion approach. Background Technology

[0002] In aquaculture, the management of aquaculture water bodies has long relied on single-modal monitoring and manual experience-based decision-making. Conventional techniques include: single-sensor monitoring: using independent dissolved oxygen meters, ammonia nitrogen reagent kits, etc., to manually detect key water quality parameters at different times; static threshold early warning: triggering alarms based on fixed water quality standards, such as DO≤5 mg / L, with the alarm standard being fixed and ignoring seasonal changes and species differences; and segmented control: operations such as oxygenation and water changes rely on observation and judgment by aquaculture personnel based on past experience. All of these methods require aquaculture personnel to monitor and operate the aquaculture water bodies regularly. Manual monitoring and operation lead to high monitoring costs. Single-modal monitoring or personnel relying on experience lacks real-time data-driven closed-loop control and cannot automatically adjust the aquaculture water bodies. Problems in the aquaculture water bodies may only be discovered when fish and shrimp "surface," which can easily cause significant economic losses. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent closed-loop system for monitoring, evaluating, and regulating aquaculture water bodies using multimodal fusion to improve the accuracy of risk early warning.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A multimodal integrated intelligent closed-loop system for monitoring, evaluating, and regulating aquaculture water bodies includes a data processing module, a mapping module, a water quality analysis module, a scoring module, an evaluation module, and a regulation module.

[0006] The data processing module is used to acquire historical monitoring data of aquaculture water bodies, preprocess the historical monitoring data, and construct training samples based on the preprocessed historical monitoring data. The aquaculture water body parameters in the training samples include temperature, conductivity, pH, nitrite, dissolved oxygen, and ammonia nitrogen.

[0007] The mapping module is used to construct a parameter association graph of the aquaculture water parameters, and aggregate neighborhood information through the parameter association graph;

[0008] The water quality analysis module includes a water quality analysis model. The water quality analysis model is trained using the training samples and the parameter correlation diagram. The trained water quality analysis model is analyzed based on the measured values ​​of the aquaculture water parameters obtained in real time and the parameter correlation diagram to obtain the predicted values ​​of each aquaculture water parameter for a preset period.

[0009] The scoring module is used to perform seasonal scoring on the predicted value based on preset seasonal characteristics and preset core parameters. The calculations, including the core parameters of ammonia nitrogen, dissolved oxygen, and nitrite, and the seasonality score, are as follows. The following formula is used for calculation:

[0010] ;

[0011] in, Indicates the first Dynamic weights of each parameter, , Seasonal weighting, Water temperature The optimal temperature for the species' survival. Temperature range For adjustment coefficients, This is a pH correction term. For temperature correction, Indicates the first The score of each parameter, The preset seasonal thresholds are obtained by querying the parameter threshold table, and the predicted values ​​of the core parameters are derived from the predicted values ​​of the aquaculture water parameters. This indicates the number of core parameters involved in the calculation;

[0012] The evaluation module is used to query the evaluation level table to obtain the seasonal score. The corresponding evaluation level is assigned. If the evaluation level is excellent or good, routine maintenance is performed; otherwise, a control command is sent to the control module. The control module is used to make control decisions on the aquaculture water body. After control, the scoring module re-performs the seasonal scoring. Earth calculation.

[0013] Preferably, the data processing module uses a sliding window to construct training samples, and the specific steps are as follows:

[0014] Set the window size of the sliding window to T, where T represents the total number of samples taken in the past time period. The label represents the target parameters for the future set time period. The target parameters include temperature, conductivity, pH, nitrite, dissolved oxygen, and ammonia nitrogen.

[0015] Each sliding window generates a T×6 sample matrix and its corresponding labels.

[0016] Preferably, the steps for constructing the parameter association graph by the graphing module are as follows:

[0017] Each of the aquaculture water parameters is taken as a node, and the attribute of the node is the measured value of the corresponding aquaculture water parameter;

[0018] Based on neighborhood knowledge, the physical and chemical relationships between the aquaculture water parameters are pre-defined, static edges are constructed, and the weights of the static edges are adjusted using pre-processed historical monitoring data or physical formulas.

[0019] Generating dynamic edges based on a time-sliding window:

[0020] The preset window size, sliding step size, and number of history windows for the time-based sliding window;

[0021] The lag correlation and lag step size are calculated for the aquaculture water parameters at both ends of the dynamic edge.

[0022] Dynamic edge weights are calculated using the calculated lag correlation and lag step size;

[0023] The parameter association graph is drawn based on the constructed static edges and the generated dynamic edges.

[0024] Preferably, a preset probability is used. Virtual edges are randomly added to the parameter correlation graph, using a preset probability. Delete existing edges and add noise to the node's attributes.

[0025] Preferably, the water quality analysis model constructed by the water quality analysis module includes a graph convolutional network and an improved graph recurrent neural network;

[0026] A graph convolutional network is used for feature extraction to extract the spatial dependency features between the parameters of the aquaculture water body.

[0027] The improved graph recurrent neural network includes introducing the parameter association graph into the gated recurrent unit to capture the correlation between temporal dynamics and aquaculture water parameters, establishing a temporal model, introducing neighborhood information aggregation in state updates, and calculating the hidden state using the following formula. :

[0028]

[0029] in, To update the door, This is the hidden state from the previous moment. Here is the weight matrix for the graph convolution operation. The graph features output by the gated loop unit. For graph convolution operations;

[0030] A fully connected layer is used to map the output of the graph recurrent neural network to the target parameters, with each parameter corresponding to an output head.

[0031] Preferably, the driving mechanism between the preset seasonal features and the preset core parameters is as follows:

[0032] Spring: Water temperature rises, microbial activity increases, ammonia nitrogen production increases, nitrification has not fully recovered, and there is a risk of nitrite accumulation. The preset core parameters for spring are ammonia nitrogen and dissolved oxygen.

[0033] Summer: High temperatures reduce dissolved oxygen saturation, pH increases leading to a rise in the proportion of ammonia nitrogen, and algal blooms cause a sharp drop in dissolved oxygen at night. The key parameters for summer are dissolved oxygen and ammonia nitrogen.

[0034] Autumn: Temperature fluctuations inhibit nitrifying bacteria, hindering nitrite conversion. The death of uneaten food and algae increases the organic matter load. The preset core parameters for autumn are nitrite and organic matter.

[0035] Winter: Low temperatures cause nitrification to stop, nitrite to accumulate, and biological oxygen consumption to decrease. However, ice blockage hinders gas exchange and poses a risk of hypoxia at the bottom layer. The core parameters for winter are nitrite and dissolved oxygen.

[0036] Preferably, the scoring module optimizes the dissolved oxygen score using the following formula:

[0037] ;

[0038] in, This represents the predicted dissolved oxygen value derived from parameters of aquaculture water bodies. , , , These are the levels corresponding to the preset seasonal thresholds of dissolved oxygen in the parameter threshold table;

[0039] The scores for ammonia nitrogen and nitrate were optimized using the following formula:

[0040] ;

[0041] in, This indicates the predicted values ​​of ammonia nitrogen or nitrate from parameters derived from aquaculture water. , , , These are the levels corresponding to the preset seasonal thresholds for ammonia nitrogen or nitrate in the parameter threshold table.

[0042] Preferably, the scoring module uses pH and temperature T to dynamically optimize the scoring of ammonia nitrogen.

[0043] ;

[0044] ;

[0045] The ammonia nitrogen score is calculated based on the calculated ammonia nitrogen solubility and the parameter threshold table.

[0046] Preferably, the control decision of the control module is specifically operated as follows:

[0047] If the water quality assessment level is excellent or good, then routine maintenance should be carried out.

[0048] If the water quality assessment level is poor, emergency treatment will be carried out;

[0049] If the water quality assessment level is medium, then it is necessary to determine the dominant risk parameter. If the risk parameter is insufficient dissolved oxygen, then it is necessary to start the aerator and reduce feeding. If the risk parameter is excessive ammonia nitrogen, then water exchange operation should be carried out according to the preset ratio and nitrifying bacteria should be added. If the risk parameter is nitrite accumulation, then the pH should be increased to 7.5 and denitrifying bacteria should be added.

[0050] Preferably, the scoring module uses the following formula to score... Optimize:

[0051] ;

[0052] Use the following formula to Optimize:

[0053] .

[0054] By adopting the aforementioned design scheme, the beneficial effects of this invention are as follows: the data processing module is used to acquire historical monitoring data of aquaculture water bodies and construct training samples; the mapping module is used to construct parameter correlation diagrams; the water quality analysis module uses the training samples and parameter correlation diagrams to train the water quality analysis model; the trained water quality analysis model is used to predict the parameters of aquaculture water bodies; the scoring module performs seasonal scoring calculations on the predicted data; the evaluation module queries the evaluation level table to obtain the evaluation level corresponding to the seasonal score; and the control module makes control decisions based on the evaluation level. This application acquires multimodal aquaculture water body parameters in real time and performs accurate evaluations based on seasonal characteristics, which helps to improve the accuracy of risk warning. Attached Figure Description

[0055] Figure 1 This is a parameter correlation diagram of the present invention;

[0056] Figure 2 This is a schematic diagram comparing the performance of the water quality analysis model of the present invention;

[0057] Figure 3 This is a flowchart of the seasonal dynamic evaluation system for aquaculture water bodies according to the present invention;

[0058] Figure 4 This is a logic tree diagram of the control decision-making process of the present invention;

[0059] Figure 5 This is a flowchart of the algorithm of the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0061] The terms "first," "second," "third," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0062] like Figure 5 As shown, a multimodal fusion intelligent closed-loop system for monitoring, evaluating, and regulating aquaculture water bodies includes a data processing module, a mapping module, a water quality analysis module, a scoring module, an evaluation module, and a regulation module.

[0063] The data processing module is used to acquire historical monitoring data of aquaculture water bodies, preprocess the historical monitoring data, and construct training samples using a sliding window based on the preprocessed historical monitoring data. The parameters of aquaculture water bodies in the training samples include temperature, conductivity, pH, nitrite, dissolved oxygen, and ammonia nitrogen.

[0064] In this embodiment, the historical monitoring data includes temperature, conductivity, pH, nitrite, dissolved oxygen, and ammonia nitrogen. The preprocessing of the historical monitoring data is normalization, such as Min-Max or Z-Score, to eliminate the influence of dimensions.

[0065] The data processing module uses a sliding window to construct training samples, and the specific steps are as follows:

[0066] Set the window size of the sliding window to T, where T represents the total number of samples taken in the past time period. The label represents the target parameters for the future set time period. The target parameters include temperature, conductivity, pH, nitrite, dissolved oxygen, and ammonia nitrogen. For example, if we obtain data for 30 days and sample once per hour, then T is 720.

[0067] Each sliding window generates a T×6 sample matrix and its corresponding label. The label here is the target parameter within a preset future time period, such as the target parameter for the next day.

[0068] The composition module is used to construct, such as Figure 1 The parameter correlation diagram of the aquaculture water parameters shown aggregates neighborhood information through the parameter correlation diagram;

[0069] The specific steps for constructing the parameter association graph by the graph composition module are as follows:

[0070] Each aquaculture water body parameter is treated as a node, and the attribute of the node is the measured value of the corresponding aquaculture water body parameter;

[0071] Based on neighborhood knowledge, the physical and chemical relationships between aquaculture water parameters are pre-defined, and static edges are constructed, such as the negative correlation edge between temperature and dissolved oxygen, and the chemical balance edge between pH and ammonia nitrogen; the weights of the static edges are adjusted using pre-processed historical monitoring data or physical formulas.

[0072] This application uses the Pearson correlation coefficient to calculate the weights of static edges:

[0073] ;

[0074] in Indicates aquaculture water parameters and aquaculture water parameters The weights of the static edges between them. , For aquaculture water parameters and aquaculture water parameters The monitoring data sequence, To calculate the Pearson correlation coefficient, which is used to measure the degree of linear association between the two from a historical data statistical perspective, These are physicochemical constants, representing quantitative values ​​of the correlation strength between parameters based on physicochemical laws. Different physicochemical correlations correspond to different... For example, the relationship between temperature and dissolved oxygen, pH and ammonia nitrogen, etc., can be determined by using physicochemical formulas to determine the weight of static edges, or the physicochemical constants of temperature and dissolved oxygen can be determined by using Henry's Law coefficients.

[0075] Interactions between water quality parameters often exhibit time lags. For example, after an increase in ammonia nitrogen concentration, nitrite concentration may begin to rise 1-2 hours later; after wastewater discharge causes an increase in nitrite concentration, dissolved oxygen may decrease 4-8 hours later; and the effect of temperature changes on pH may have a delay of several hours. Therefore, a time-sliding window is introduced, and edges are dynamically generated through time-series correlations to capture the lag effects between parameters.

[0076] Generating dynamic edges based on a time-sliding window:

[0077] The window size, sliding step size, and number of historical windows are preset for the time sliding window. In this embodiment, the window size can be set to 24-72 hours according to the characteristics of water quality changes, and the sliding step size can be set to 1-4 hours according to the water quality change cycle. This depends on the monitoring frequency and system response speed, balancing computational efficiency and real-time performance. The number of historical windows is set to 3-5 to cover the lag effect at different time scales.

[0078] The lag correlation and lag step length of the aquaculture water parameters at both ends of the dynamic edge are calculated.

[0079] Assume the aquaculture water parameters at both ends of the dynamic edge are as follows: and The following formula is used to calculate different lag step sizes. The following correlation:

[0080] ;

[0081] in, For the lagging step size, (1, 2, ..., ), For covariance, Standard deviation A time step is a specific moment in time.

[0082] Aquaculture water parameters at both ends of the dynamic edge and Select the lag step size corresponding to the maximum correlation, and calculate the lag step size using the following formula:

[0083] ;

[0084] in, express and The correlation between the two parameters, Indicates the lag step size Find the maximum value. This indicates taking the absolute value.

[0085] Dynamic edge weights are calculated using the obtained lag correlation and lag step size.

[0086] ;

[0087] Draw a parametric graph based on the constructed static edges and the generated dynamic edges. The parametric graph can be drawn manually or using conventional software, such as Python.

[0088] In this embodiment, a preset probability is used. Virtual edges are randomly added to the parametric correlation graph to simulate unknown correlations, using preset probabilities. Remove existing edges and test the model's robustness to actual associations. and all In this embodiment, and All values ​​are set to 0.1, but other settings can be made as needed. Gaussian noise can be randomly added to the node attributes to simulate sensor measurement errors. The following formula can be used to randomly add Gaussian noise:

[0089] ;

[0090] in, The original node attribute value. These are the node attribute values ​​after adding noise. With a mean of 0 and a standard deviation of Gaussian noise distribution Adjust the settings according to the sensor accuracy; the higher the sensor accuracy, the better. The smaller the size, the lower the sensor accuracy. The larger.

[0091] The water quality analysis module includes a water quality analysis model. The water quality analysis model is trained using training samples and parameter correlation diagrams. The trained water quality analysis model is analyzed based on the measured values ​​of aquaculture water parameters and parameter correlation diagrams obtained in real time to obtain the predicted values ​​of each aquaculture water parameter for a preset period.

[0092] The water quality analysis model constructed by the water quality analysis module includes a graph convolutional network and an improved graph recurrent neural network;

[0093] Graph Convolutional Networks (GCNs) are neural network models that process data using graph structures. In aquaculture water monitoring tasks, aquaculture water parameters can be viewed as nodes in a parameter association graph, with relationships between nodes represented by edges in the graph. The main purpose of GCNs is to aggregate neighborhood information for each node from the adjacency matrix. The negative correlation between temperature and dissolved oxygen is aggregated into the features of the temperature node through graph convolution. In this way, GCNs can provide richer upstream and downstream dependency features for each aquaculture water parameter node, thereby helping the model understand the interactions between aquaculture water parameters. By aggregating the neighborhood information of nodes through the parameter association graph structure and using GCNs for feature extraction, spatial dependency features between aquaculture water parameters can be extracted.

[0094] The feature extraction process for graph convolutional networks is as follows:

[0095] Input the original water quality parameter matrix ;

[0096] For the original water quality parameter matrix The extraction is performed iteratively layer by layer, and the calculation formula is as follows:

[0097] ;

[0098] in, For the first Layer node features, initial input node features These are raw water quality parameters, such as temperature, conductivity, and pH. To represent the dependencies between nodes in the dynamic adjacency matrix obtained from the parameter association graph, the adjacency matrix is... Enhancements are made by introducing self-loops to ensure that nodes retain their own characteristics and avoid information loss; Let be a degree matrix, where To perform symmetric normalization, address the uneven degree distribution of the adjacency matrix, and prevent node features from being excessively amplified or reduced due to differences in the number of adjacencies, The weight matrix is ​​trainable. This is the activation function.

[0099] Features from all layer outputs are combined:

[0100] ;

[0101] in, Node features of each layer The feature matrix is ​​obtained by fusion. That is, the node features at each time step t.

[0102] The improved graph recurrent neural network introduces a parameter correlation graph into the gated recurrent unit (GRU) to capture the correlation between temporal dynamics and aquaculture water parameters, establish a time-series model, and, after introducing the parameter correlation graph, the core formula of the gated recurrent control unit is as follows:

[0103] ;

[0104] ;

[0105] ;

[0106] in: To update the door, To reset the door, In hidden state, In the candidate hidden state, The weight matrix from the features of the association graph to the update gate. The state was hidden in the previous moment. To update the gate's bias term, The weight matrix from the features of the association graph to the reset gate. This is the weight matrix from the hidden state in the previous time step to the reset gate. To reset the door's bias, This is the weight matrix from the features of the association graph to the candidate hidden states. This is the weight matrix from historical hidden states to candidate hidden states after the reset gate filtering. The bias term for the candidate hidden state.

[0107] In the state update, neighborhood information aggregation is introduced, and the hidden state is calculated using the following formula. :

[0108]

[0109] in, To update the door, The state was hidden in the previous moment. Here is the weight matrix for the graph convolution operation. The graph features output by the gated loop unit. For graph convolution operations;

[0110] A fully connected layer is used to map the output of the graph recurrent neural network to the target parameters, with each parameter corresponding to an output header. The calculation formula is as follows:

[0111] ;

[0112] in: The parameters are any one of the following: temperature, pH, nitrite, dissolved oxygen, and ammonia nitrogen. For predicted values, the time step needs to be adjusted according to the sampling frequency (e.g., once a day). , For parameters The weight matrix, For parameters The bias term.

[0113] Time step t This refers to a specific point in time corresponding to the output of the prediction model, i.e., a particular moment in time. It represents the future moment corresponding to the predicted value. Time step t The essence of prediction is the position of the prediction result on the time axis, which is determined by the sampling frequency and the prediction duration. For example, if we need to predict water quality data for the next 24 hours, the sampling frequency is once per hour, the total time step T is 24, and the time steps t = 1, 2, 3, ..., 24.

[0114] For example, when inputting data on parameters such as temperature, conductivity, pH, nitrite, dissolved oxygen, and ammonia nitrogen collected over the past 30 days from April 1st to April 30th, the predicted data for the next day is obtained through feature fusion and time-series prediction: dissolved oxygen = 4.2 mg / L, ammonia nitrogen = 0.02 mg / L, nitrite = 0.1 mg / L, pH = 8.6, and water temperature = 30℃.

[0115] The performance of the water quality analysis model constructed in this application is compared with that of LSTM, GRU, and RandomForest algorithms. Figure 2 As shown in the figure. Experimental results show that the water quality analysis model constructed in this application is significantly better than other algorithms in terms of R2, RMSE, and MAE. The algorithm has an R2 of 0.95, an RMSE of 1.25, and an MAE of 0.89, which is excellent. This indicates that the algorithm can perform good feature fusion and time series prediction on the collected water quality data, and the prediction effect is excellent.

[0116] like Figure 3 As shown, the scoring module is used to perform seasonal scoring on the predicted values ​​based on the preset seasonal characteristics and preset core parameters in Table 1. Calculations were performed using core parameters including ammonia nitrogen, dissolved oxygen, and nitrite, along with seasonality scoring. The following formula is used for calculation:

[0117] ;

[0118] in, Indicates the first Dynamic weights of each parameter, , Seasonal weighting, for example, dissolved oxygen weighting is 45% in summer. Water temperature This refers to the optimal survival temperature for a species; for example, the optimal survival temperature for tilapia is 28℃. Temperature range For adjustment coefficients, dissolved oxygen weights ammonia nitrogen weight , This is a pH correction term. For temperature correction, Indicates the first The score of each parameter, The preset seasonal threshold is obtained by querying the parameter threshold table in Table 2. This indicates the number of core parameters involved in the calculation. In this embodiment, the core parameters include three core parameters: ammonia nitrogen, dissolved oxygen, and nitrite, i.e., n is 3.

[0119] In this embodiment, the adjustment of dissolved oxygen weight in the water body for tilapia farming during summer is used as an example:

[0120] Seasonal weight water temperature , ,

[0121] ;

[0122] In this embodiment, the dissolved oxygen score is optimized using the following formula:

[0123] ;

[0124] in, This represents the predicted dissolved oxygen value derived from parameters of aquaculture water bodies. , , , These are the levels corresponding to the preset seasonal thresholds of dissolved oxygen in the parameter threshold table;

[0125] The scores for ammonia nitrogen and nitrate were optimized using the following formula:

[0126] ;

[0127] in, This indicates the predicted values ​​of ammonia nitrogen or nitrate from parameters derived from aquaculture water. , , , These are the levels corresponding to the preset seasonal thresholds for ammonia nitrogen or nitrate in the parameter threshold table.

[0128] Ammonia nitrogen was dynamically scored and optimized using pH and temperature T.

[0129] ;

[0130] ;

[0131] The ammonia nitrogen score is calculated based on the calculated ammonia nitrogen solubility and the parameter threshold table in Table 2.

[0132] In this embodiment, the spring data parameter score calculation example is as follows:

[0133] Dissolved oxygen score (positive effect parameter), measured value V: 5.2 mg / L;

[0134] Parameter thresholds: Excellent ≥ 6.0, Good 5.0-5.9, Average 4.0-4.9;

[0135] Calculation: If the measured value 5.2 ∈ [5.0, 5.9], then use the good grade formula:

[0136] ;

[0137] Ammonia nitrogen score (negative effect parameter + NH3 correction), measured total ammonia nitrogen: 0.028 mg / L, pH=8.2, water temperature T=15℃;

[0138] ;

[0139] NH3 toxicity concentration: 0.028 × 0.62 = 0.017 mg / L;

[0140] Parameter thresholds: Excellent ≤ 0.010, Good 0.011-0.030, 0.017∈[0.011, 0.030], then use the Good grade formula:

[0141] ;

[0142] Nitrite score (negative effect), measured value: 0.09 mg / L;

[0143] Parameter thresholds: Excellent ≤ 0.05, Good 0.06-0.10, 0.09 ∈ (0.05, 0.10], then use the Good grade formula:

[0144] .

[0145] Use the following formula to Optimize:

[0146] ;

[0147] Use the following formula to Optimize:

[0148] ;

[0149] in, The median of the species' suitable temperature range; if the species' suitable temperature range is 25-30℃, then... .

[0150] Table 1 shows the analysis of the influence of seasonal characteristics and core parameters.

[0151]

[0152] In this embodiment, the driving mechanism between preset seasonal characteristics and preset core parameters is as follows:

[0153] Spring: Water temperature rises, microbial activity increases, ammonia nitrogen production increases, nitrification has not fully recovered, and there is a risk of nitrite accumulation. The preset core parameters for spring are ammonia nitrogen and dissolved oxygen.

[0154] Summer: High temperatures reduce dissolved oxygen saturation, pH increases leading to a rise in the proportion of ammonia nitrogen, and algal blooms cause a sharp drop in dissolved oxygen at night. The key parameters for summer are dissolved oxygen and ammonia nitrogen.

[0155] Autumn: Temperature fluctuations inhibit nitrifying bacteria, hindering nitrite conversion. The death of uneaten food and algae increases the organic matter load. The preset core parameters for autumn are nitrite and organic matter.

[0156] Winter: Low temperatures cause nitrification to stop, nitrite to accumulate, and biological oxygen consumption to decrease. However, ice blockage hinders gas exchange and poses a risk of hypoxia at the bottom layer. The core parameters for winter are nitrite and dissolved oxygen.

[0157] Table 2 Parameter Threshold Table

[0158]

[0159] The evaluation module is used to query the evaluation level table in Table 3 to obtain seasonal scores. The corresponding evaluation level is determined by the rating. If the evaluation level is excellent or good, routine maintenance is performed; otherwise, a control command is sent to the control module. The control module is used to make control decisions for the aquaculture water body. After control, the scoring module re-performs the seasonal scoring. Earth calculation.

[0160] Table 3 Evaluation Level Table

[0161]

[0162] In this embodiment, as Figure 4 As shown, the specific operation of the control decision in step S5 is as follows:

[0163] If the water quality assessment level is excellent or good, then routine maintenance should be carried out.

[0164] If the water quality assessment level is poor, emergency treatment will be carried out;

[0165] If the water quality assessment level is medium, then it is necessary to determine the dominant risk parameter. If the risk parameter is insufficient dissolved oxygen, then it is necessary to start the aerator and reduce feeding. If the risk parameter is excessive ammonia nitrogen, then water exchange operation should be carried out according to the preset ratio and nitrifying bacteria should be added. If the risk parameter is nitrite accumulation, then the pH should be increased to 7.5 and denitrifying bacteria should be added. This control decision can be set according to the season and the actual needs of the aquaculture species.

[0166] To better illustrate the regulatory decisions, this embodiment provides the following operational examples for summer and winter:

[0167] The core problems in aquaculture water bodies during the high-temperature summer period are: insufficient dissolved oxygen and high NH3 toxicity, and the corresponding control decisions.

[0168] (1) Oxygenation strategy: Turn on the aerator from 14:00 to 16:00 in the afternoon to prevent surface supersaturation; continuously increase oxygen from 3:00 to 5:00 in the morning to avoid the lowest dissolved oxygen level before dawn.

[0169] (2) Feeding management: When the water temperature is >30℃, reduce the amount of feed by 30%; use low-protein feed to reduce ammonia nitrogen production;

[0170] (3) Microbial regulation: Add thermotolerant nitrifying bacteria, such as Bacillus subtilis, weekly. Combined with carbon-nitrogen ratio regulation, C:N=15:1, it promotes the reproduction of microbial communities;

[0171] The core issues in aquaculture water during the low-temperature winter period are: nitrite accumulation and bottom hypoxia, and the corresponding control decisions.

[0172] (1) Nitrite control: Add low-temperature denitrifying bacteria, such as Pseudomonas denitrificans.

[0173] Add brown sugar at 5 ppm as a carbon source to promote denitrification.

[0174] (2) Dissolved oxygen management: Ventilation holes are opened in the frozen pond, with one 20cm diameter hole per 100㎡; microporous aeration discs are used instead of traditional impeller aerators to reduce water temperature disturbance;

[0175] (3) Ammonia nitrogen adsorption: Sprinkle slaked lime. Once a week.

[0176] To better illustrate the intelligent closed-loop system for monitoring, evaluating, and regulating aquaculture water bodies as described in this application, this embodiment uses summer Litopenaeus vannamei farming as an example.

[0177] The data processing module acquires the aquaculture water parameters collected every hour over the past 30 days, including dissolved oxygen, ammonia nitrogen, nitrite, pH, water temperature, and conductivity data. The data is preprocessed and used to construct training samples.

[0178] The mapping module constructs a parameter correlation diagram of aquaculture water body parameters;

[0179] The water quality analysis module uses a water quality analysis model to analyze and obtain the predicted values ​​of various aquaculture water parameters for the next day: dissolved oxygen = 4.2 mg / L, ammonia nitrogen = 0.05 mg / L, nitrite = 0.1 mg / L, pH = 8.6, water temperature = 30℃;

[0180] The rating module performs seasonal ratings. calculate:

[0181] Dissolved oxygen was 4.2 mg / L, which is "average" (3.5-4.4). Dissolved oxygen score:

[0182] ;

[0183] The ammonia nitrogen level is 0.05 mg / L. Calculate the ammonia nitrogen percentage:

[0184]

[0185] Ammonia nitrogen concentration = 0.05 × 0.009 = 0.0045 mg / L, which is "Excellent" (≤0.008). Ammonia nitrogen score:

[0186] ;

[0187] The nitrite level is 0.1, which is considered "average" (0.09-0.12). Nitrite score:

[0188] ;

[0189] Dynamic weight adjustment

[0190] water temperature , (The optimal survival temperature for shrimp is 28℃); according to the weighted adjustment formula The weights of dissolved oxygen, ammonia nitrogen, and nitrate after adjustment can be obtained by calculation.

[0191] ;

[0192] ;

[0193] ;

[0194] Correction term calculation:

[0195] ;

[0196] From pH=8.6, we can obtain... .

[0197] ;

[0198] The optimal temperature range for shrimp is 28-32℃. , .

[0199] WQI calculation:

[0200] ;

[0201] The evaluation module determined the water quality level to be "medium" based on the WQI calculation result of 6.67, triggering an orange alert and initiating oxygenation and water exchange operations.

[0202] The control module's control decisions and control operations:

[0203] In the calculation of WQI, the scores and adjusted weights of each parameter in the water body are as follows:

[0204]

[0205] Dissolved oxygen has the largest weight but a low score, therefore insufficient dissolved oxygen is the main influencing factor. Based on the control decision tree, a decision is made to start aerators to increase oxygen levels, while simultaneously reducing feed and appropriately replacing water to regulate the oxygen concentration in the water. The specific aeration strategy is as follows: turn on aerators from 14:00 to 16:00 in the afternoon to prevent surface supersaturation; and continuously aerate from 3:00 to 5:00 in the morning to avoid the lowest dissolved oxygen level before dawn.

[0206] In summary, the data processing module acquires historical monitoring data of aquaculture water bodies and constructs training samples; the mapping module constructs parameter correlation diagrams; the water quality analysis module trains the water quality analysis model using the training samples and parameter correlation diagrams, and uses the trained water quality analysis model to predict aquaculture water body parameters; the scoring module calculates seasonal scores on the predicted data; the evaluation module queries the evaluation level table to obtain the evaluation level corresponding to the seasonal score; and the control module makes control decisions based on the evaluation level. This application acquires multimodal aquaculture water body parameters in real time and performs accurate evaluations based on seasonal characteristics, which helps improve the accuracy of risk warning.

[0207] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent closed-loop system for multi-modal fusion of aquaculture water body monitoring, evaluation, regulation and control, characterized by: comprising a data processing module, a graph construction module, a water quality analysis module, a scoring module, an evaluation module and a regulation module; the data processing module is used to obtain historical monitoring data of the aquaculture water body, preprocess the historical monitoring data, and construct training samples according to the preprocessed historical monitoring data, wherein the aquaculture water body parameters in the training samples include temperature, conductivity, pH, nitrite, dissolved oxygen and ammonia nitrogen; the graph construction module is used to construct a parameter correlation graph of the aquaculture water body parameters, and aggregate neighborhood information through the parameter correlation graph; the water quality analysis module comprises a water quality analysis model, which is trained using the training samples and the parameter correlation graph, and the trained water quality analysis model analyzes the real-time acquired measured values of the aquaculture water body parameters and the parameter correlation graph to obtain predicted values of each of the aquaculture water body parameters in a preset period; The scoring module is configured to seasonally score the predicted value according to a preset seasonal feature and a preset core parameter The core parameter includes ammonia nitrogen, dissolved oxygen and nitrite, and the seasonal score The following formula is used for calculation: ； wherein, represents the dynamic weight of the th parameter, , is the seasonal weight, is the water temperature, is the optimal survival temperature of the species, is the temperature range, is the adjustment coefficient, is the PH correction term, is the temperature correction term, represents the score of the th parameter, , the preset seasonal threshold is obtained by querying a parameter threshold table, and the core parameter prediction value comes from a prediction value of the aquaculture water body parameter, represents the number of core parameters participating in the calculation; The evaluation module is configured to query an evaluation grade table to obtain the seasonal score corresponding evaluation grade. If the evaluation grade is excellent or good, regular maintenance is performed. Otherwise, a control command is sent to the control module. The control module is configured to make a control decision for the aquaculture water body. After the control, the scoring module re-performs the seasonal score ground calculation.

2. The multi-modal fusion intelligent closed-loop system for monitoring, evaluating, regulating and controlling aquaculture water bodies according to claim 1, characterized in that: the data processing module uses a sliding window to construct training samples, and the specific steps are as follows: set the window size of the sliding window as N, N represents the total number of samples in the past time period, and the label is the target parameter in the future set time period, wherein the target parameter includes temperature, conductivity, pH, nitrite, dissolved oxygen and ammonia nitrogen; each sliding window generates an N×6 sample matrix and its corresponding label.

3. The multi-modal fused aquaculture water monitoring, evaluation, regulation, and control intelligent closed loop system of claim 2, wherein: The graph construction module constructs the parameter correlation graph as follows: each of the aquaculture water body parameters is taken as a node, and the attribute of the node is the measured value of the corresponding aquaculture water body parameter; based on neighborhood knowledge, the physical and chemical correlations between the aquaculture water body parameters are preset, a static edge is constructed, and the weight of the static edge is adjusted using preprocessed historical monitoring data or physical formulas; based on a time sliding window, a dynamic edge is generated: preset the window size, sliding step and history window number of the time sliding window; perform lag correlation calculation and lag step calculation on the aquaculture water body parameters at both ends of the dynamic edge; perform dynamic edge weight calculation using the calculated lag correlation and lag step; draw the parameter correlation graph according to the constructed static edge and generated dynamic edge.

4. The multi-modal fused aquaculture water monitoring, evaluation, regulation, and control intelligent closed loop system of claim 3, wherein: with a preset probability adding a virtual edge in the parameter correlation graph randomly, with a preset probability deleting an existing edge and adding noise to the attribute of a node 5. The multi-modal fused aquaculture water monitoring, evaluation, regulation, and control intelligent closed loop system of claim 4, wherein: The water quality analysis model constructed by the water quality analysis module comprises a graph convolution network and an improved graph recurrent neural network; feature extraction is performed using the graph convolution network to extract spatial dependency features between the aquaculture water body parameters; The improved graph recurrent neural network comprises introducing the parameter correlation graph in a gated recurrent unit, capturing the correlation between the timing dynamics and the aquaculture water body parameters, establishing a timing model, introducing neighborhood information aggregation in state updating, and calculating a hidden state by using the following formula : wherein, is an update gate, is a hidden state of a previous time, is a weight matrix of a graph convolution operation, is a graph feature output by a gated recurrent unit, is a graph convolution operation; the outputs of the graph recurrent neural network are mapped to target parameters using a fully connected layer, and each parameter corresponds to an output head.

6. The multi-modal fused aquaculture water monitoring, evaluation, regulation, and control intelligent closed loop system of claim 5, wherein: The driving mechanism between the preset seasonal characteristics and the preset core parameters is as follows: spring: water temperature rises, microbial activity increases, ammonia nitrogen generation increases, nitrification does not fully recover, and there is a risk of nitrite accumulation, the preset core parameters in spring are ammonia nitrogen and dissolved oxygen; summer: high temperature reduces dissolved oxygen saturation, pH rise leads to increase of ammonia nitrogen proportion, and algae outbreak leads to sudden drop of dissolved oxygen at night, the preset core parameters in summer are dissolved oxygen and ammonia nitrogen; Autumn: Temperature fluctuation inhibits nitrifying bacteria, making it difficult for nitrite conversion, and the increase in organic matter load caused by the death of leftover food and algae. The preset core parameters in autumn are nitrite and organic matter; Winter: Low temperature leads to nitrification stagnation, nitrite accumulation, and biological oxygen consumption decline, but ice cover hinders gas exchange, and there is a risk of bottom anoxia. The preset core parameters in winter are nitrite and dissolved oxygen.

7. The intelligent closed-loop system for monitoring, evaluating, regulating and controlling of multi-modal fusion of aquaculture water according to claim 6, wherein: The scoring module optimizes the score of dissolved oxygen by using the following formula: ； wherein, represents a predicted value of dissolved oxygen from the aquaculture water body parameters, , , , are respectively the grades corresponding to the preset seasonal threshold values of dissolved oxygen in the parameter threshold table. The score of ammonia nitrogen is optimized by using the following formula: ； wherein, represents the predicted value of ammonia nitrogen from the aquaculture water body parameters, , , , are respectively the grades corresponding to the preset seasonal threshold values of ammonia nitrogen in the parameter threshold table. The score of nitrate is optimized by using the following formula: ； wherein, represents a predicted value of nitrate from the aquaculture water body parameters, , , , are the grades corresponding to the preset seasonal threshold values of nitrate in the parameter threshold table, respectively.

8. The intelligent closed-loop system for monitoring, evaluating, regulating and controlling of multi-modal fusion of aquaculture water according to claim 7, wherein: The scoring module dynamically scores and optimizes ammonia nitrogen by using pH and water temperature T. ； ； The score of ammonia nitrogen is calculated according to the calculated ammonia nitrogen solubility and the parameter threshold table.

9. The multi-modal fused aquaculture water monitoring, evaluation, regulation, and control intelligent closed loop system of claim 8, wherein: The control module performs the following specific operations for the control decision: If the evaluation level of water quality is excellent or good, routine maintenance is performed; If the evaluation level of water quality is poor, emergency treatment is performed; If the evaluation level of water quality is medium, the dominant risk parameter needs to be determined. If the risk parameter is insufficient dissolved oxygen, the oxygenation machine needs to be started and the feeding needs to be reduced. If the risk parameter is excessive ammonia nitrogen, water needs to be replaced at a preset ratio and nitrifying bacteria need to be added. If the risk parameter is nitrite accumulation, the pH value needs to be increased to 7.5 and denitrifying bacteria need to be supplemented.

10. The multi-modal fused aquaculture water monitoring, evaluation, regulation, and control intelligent closed loop system of claim 8, wherein: The scoring module adopts the following formula to score optimization: ； The following formula is used to optimize the parameters: : 。

Images