Seedling culture growth condition monitoring method

Abstract

The invention provides a seedling cultivation growth condition monitoring method, which comprises the steps of performing data processing on a time series data set of hydrological conditions, obtaining a dynamic association rule among different hydrological parameters, constructing an association model of the hydrological parameters, and quantitatively describing mutual influence and restriction relationships among water temperature, dissolved oxygen and salinity parameters by the association model; performing data processing on the time sequence data set of seedling growth, establishing a seedling growth process model, determining key parameters of a seedling growth curve by analyzing growth rate change rules of seedlings in different growth stages, and correcting the model according to initial quality and stocking density factors of the seedlings; a seedling growth prediction model is combined with an intelligent optimization algorithm, key control parameters in the breeding process are dynamically adjusted, and according to the seedling growth trend predicted by the model, hydrological conditions and feeding strategies are adaptively optimized, so that intelligent management of the breeding process is realized.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method for monitoring the growth status of seedling cultivation. Background Art

[0002] In the monitoring of seedling cultivation growth, in order to establish a seedling growth prediction model based on hydrological conditions, a large amount of hydrological data and seedling growth data need to be collected. However, in actual breeding scenarios, the changes in hydrological parameters are very complex, and there is a relationship of mutual influence and restriction between different parameters. For example, the increase in water temperature will lead to a decrease in dissolved oxygen, which in turn will inhibit the growth of seedlings. At the same time, different varieties of seedlings have different adaptability to hydrological conditions. Some seedlings are very sensitive to changes in water temperature, while others are more sensitive to salinity fluctuations. In addition, the growth of seedlings is not only affected by hydrological conditions, but also closely related to factors such as the initial quality of seedlings, stocking density, and feeding. The interaction between these factors further increases the difficulty of establishing an accurate prediction model. Therefore, how to collect and process a large amount of hydrological and growth data in a complex breeding environment and discover hidden associations is the main technical challenge faced in establishing a seedling growth prediction model. At the same time, how to apply the established model to actual breeding production and adjust the breeding strategy in a timely manner according to the prediction results of the model is also a key issue to consider. Summary of the invention

[0003] The present invention provides a method for monitoring the growth status of seedlings, which mainly includes:

[0004] Obtain real-time monitoring data on water temperature, dissolved oxygen, and salinity in the farm, and construct a time series data set of hydrological conditions based on water temperature, dissolved oxygen, and salinity. At the same time, obtain key growth indicator data on body length and weight during the growth of seedlings of different varieties to construct a time series data set of seedling growth;

[0005] Process the time series data set of hydrological conditions to obtain the dynamic correlation between different hydrological parameters and construct a correlation model of hydrological parameters. The correlation model quantitatively describes the mutual influence and constraint relationship between water temperature, dissolved oxygen and salinity parameters.

[0006] Data processing was performed on the time series data set of seedling growth to establish a seedling growth process model. By analyzing the growth rate variation patterns of seedlings at different growth stages, the key parameters of the seedling growth curve were determined, and the model was calibrated based on the initial quality of the seedlings and stocking density factors.

[0007] The hydrological condition parameters are used as inputs to the established hydrological parameter association model and seedling growth process model, and seedling growth indicators are output. Through training with a large number of historical data of seedling growth time series data sets, the complex nonlinear mapping relationship between hydrological conditions and seedling growth is obtained, and seedling growth prediction based on hydrological conditions is realized.

[0008] Introduce other influencing factors in the breeding process, including feed feeding amount and breeding density, construct a multi-source heterogeneous data set of other influencing factors in the breeding process, unify the representation and correlation analysis of the multi-source data set of other influencing factors, obtain the implicit relationship between each factor and seedling growth, and form a growth prediction model;

[0009] Combine the seedling growth prediction model with the intelligent optimization algorithm to dynamically adjust the key control parameters in the breeding process. According to the seedling growth trend predicted by the model, the hydrological conditions and feeding strategies are adaptively optimized to achieve intelligent management of the breeding process.

[0010] Develop a model-based breeding decision support system, which obtains monitoring data of the breeding environment in real time, calls the growth prediction model for online prediction, and gives optimization suggestions for breeding strategies based on the prediction results. At the same time, the breeding decision support system tracks the historical data of the breeding process, continuously evaluates and improves the model, and provides data-driven solutions for refined breeding management.

[0011] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0012] The present invention discloses a method for monitoring the growth status of seedling cultivation. The method constructs a hydrological condition time series data set by real-time monitoring of hydrological parameters such as water temperature, dissolved oxygen, and salinity in a breeding farm; at the same time, key indicators of seedling growth are obtained to establish a seedling growth time series data set. The two data sets are processed to establish a hydrological parameter association model and a seedling growth process model. Combined with factors such as feed feeding amount and breeding density, a multivariate growth prediction model is constructed to reveal the complex nonlinear mapping relationship between hydrological conditions and seedling growth. The present invention combines the growth prediction model with an intelligent optimization algorithm to dynamically adjust key breeding parameters to achieve intelligent management of the breeding process. By developing a model-based breeding decision support system, a data-driven solution is provided for refined breeding management, which effectively improves the scientificity and accuracy of aquaculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 The present invention is a flow chart of a method for monitoring the growth status of seedling cultivation.

[0014] Figure 2 It is a schematic diagram of a method for monitoring the growth status of seedling cultivation of the present invention. DETAILED DESCRIPTION

[0015] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments. The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are only used to explain the relevant inventions, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0016] like Figure 1-2 In this embodiment, a method for monitoring the growth status of seedlings culture may specifically include:

[0017] Step S101, obtain real-time monitoring data of water temperature, dissolved oxygen content, and salinity in the farm, and construct a time series data set of hydrological conditions based on water temperature, dissolved oxygen content, and salinity. At the same time, obtain key growth indicator data of body length and weight during the growth of seedlings of different varieties to construct a time series data set of seedling growth.

[0018] Real-time data from multiple monitoring points are obtained, the real-time data including water temperature, dissolved oxygen content, and salinity, and the real-time data are cleaned, denoised, and detected for outliers to obtain a hydrological environment parameter data set; the morphological characteristics of the seedlings, including body length and body shape, are extracted based on the seedling images collected by the underwater camera system; weight data of the seedlings is obtained when they pass through an automatic weighing system set in a specific area of ​​the breeding pond; individual seedlings are separated from the collected images using an image segmentation tool, and variety matching and identification are performed in combination with a pre-established seedling feature database to obtain variety information of the seedlings in the breeding pond; the correlation coefficient between the hydrological conditions and the seedling growth indicators is calculated, and if the correlation coefficient exceeds a preset threshold, the hydrological conditions are determined to be the hydrological factors that have the greatest impact on the growth of the seedlings; and a correlation model between the hydrological environment parameters and the seedling growth indicators is established based on the calculation results.

[0019] Specifically, the water quality monitoring sensor network deployed in the farm is used to obtain real-time data such as water temperature, dissolved oxygen, and salinity from multiple monitoring points. The collected data is transmitted to the central data processing unit through the data collector, and the raw data is cleaned, denoised, and outliers are detected to obtain an effective hydrological environment parameter data set. The seedling images are collected using an underwater camera system, and the morphological characteristics of the seedlings, including body length and body shape, are extracted using image processing software. An automatic weighing system is set up in a specific area of ​​the breeding pond, and the seedlings are quickly weighed to obtain weight data when they pass through. The extracted body length, body shape, and weight data are associated with timestamps to construct a time series data set of seedling growth. For the identified seedlings of different varieties, the image segmentation tool is used to separate individual seedlings from the collected images, and the seedling feature database established in advance is combined to perform variety matching and identification to obtain the variety information of the seedlings in each breeding pond. The time series processing tool is used to time align and clean the hydrological environment parameter data set and the seedling growth time series data set to ensure that the two data sets match in the time dimension. The correlation calculation function was used to calculate the correlation coefficient between hydrological conditions and seedling growth indicators, and the hydrological factors with the greatest impact on seedling growth were identified. Based on the calculation results, a correlation model between hydrological environmental parameters and seedling growth indicators was established to predict and optimize the breeding environment. In the farm, the water quality monitoring sensor network includes 20 monitoring points, each of which is equipped with water temperature, dissolved oxygen and salinity sensors. The data collector obtains data from these sensors every 5 minutes and transmits it to the central data processing unit via wireless. The raw data is denoised by median filtering, and then outliers are detected using the 3σ criterion. The processed data forms a hydrological environmental parameter data set, which includes timestamp, monitoring point ID, water temperature, dissolved oxygen and salinity. The underwater camera system consists of 10 high-definition cameras, which shoot 5 frames per second. The image processing software uses an edge detection algorithm to extract the outline of the seedlings and calculate the body length and body size ratio. The automatic weighing system is set at the outlet of the breeding pond. When the seedlings swim by, the infrared sensor triggers rapid weighing with an accuracy of 0.1g. The weighing data is matched with the most recent image data to form a complete growth record. The seedling growth time series dataset contains timestamp, seedling ID, body length, weight and body size ratio. The watershed algorithm is used for image segmentation to segment the group image into individual seedling images. The pre-trained ResNet50 model is used to identify seedling varieties with an accuracy of 95%. Time series processing uses the resample function of the pandas library to unify hydrological data and growth data to hourly granularity. The Pearson correlation coefficient is used for correlation analysis to calculate the correlation between water temperature, dissolved oxygen, salinity and body length and weight. For example, it was found that the correlation coefficient between water temperature and body length growth rate was 0.85, indicating a strong correlation. Based on the correlation results, a multivariate linear regression model was constructed to predict the growth rate of seedlings under different hydrological conditions. The determination coefficient R of the model is 2The correlation coefficient reached 0.78, indicating that the model explained 78% of the growth variation. This association model is used to optimize the breeding environment and automatically adjust the aeration equipment or temperature control system when the predicted growth rate is lower than the threshold.

[0020] Step S102, data processing is performed on the time series data set of hydrological conditions to obtain the dynamic correlation law between different hydrological parameters and construct a correlation model of hydrological parameters, wherein the correlation model quantitatively describes the mutual influence and constraint relationship between water temperature, dissolved oxygen content and salinity parameters.

[0021] The data series of three parameters, water temperature, dissolved oxygen and salinity, were extracted from the hydrological condition time series data set. The data series were cleaned and outliers were detected. The processed hydrological parameter time series data were obtained by the moving average method of the 24-hour window. The processed hydrological parameter time series data were standardized to eliminate the dimensional differences between different parameters. The linear and nonlinear correlations between water temperature, dissolved oxygen and salinity were calculated using the Pearson correlation coefficient and the Spearman rank correlation coefficient. The standardized hydrological parameter data were segmented using the sliding window technology, and the window size was set to 7 days. Multiple linear regression analysis and generalized additive model fitting were performed in each time window to establish the mathematical relationship between water temperature, dissolved oxygen and salinity. According to the results of the multiple linear regression analysis and generalized additive model fitting, the fitting results of each time window were analyzed in time series to identify the periodic changes and long-term trends of the parameter relationship. Based on the results of regression analysis and generalized additive model, a dynamic association diagram of hydrological parameters was constructed, in which the nodes represent the three parameters of water temperature, dissolved oxygen and salinity, the size of the nodes represents the degree of variation of the parameters, the edges represent the influence relationship between the parameters, the thickness of the edges represents the intensity of the influence, and the direction of the edges represents the dominant direction of the influence. The association diagram quantitatively describes the mutual influence and constraint relationship between the water temperature, dissolved oxygen and salinity parameters, and displays the dynamic change process of the parameter relationship on the timeline.

[0022] Specifically, the data series of three parameters, water temperature, dissolved oxygen, and salinity, were extracted from the hydrological condition time series data set. Data cleaning and outlier detection were performed on each series. The data was smoothed using the moving average method of a 24-hour window to obtain the processed hydrological parameter time series data. The processed hydrological parameter time series data were standardized to eliminate the dimensional differences between different parameters. The Pearson correlation coefficient and Spearman rank correlation coefficient were used to calculate the linear and nonlinear correlations between water temperature, dissolved oxygen, and salinity to determine the preliminary correlation between the parameters. The standardized hydrological parameter data were segmented using the sliding window technology, and the window size was set to 7 days. Multiple linear regression analysis and generalized additive model fitting were performed in each time window to establish the mathematical relationship between water temperature, dissolved oxygen, and salinity, and obtain the dynamic correlation between parameters in different time periods. The fitting results of each time window were analyzed in time series to identify the periodic changes and long-term trends of the parameter relationship. Based on the results of regression analysis and generalized additive model, a dynamic association diagram of hydrological parameters was constructed. The nodes in the diagram represent the three parameters of water temperature, dissolved oxygen and salinity. The size of the nodes represents the degree of variation of the parameters. The edges represent the influence relationship between the parameters. The thickness of the edges represents the intensity of the influence. The direction of the edges represents the dominant direction of the influence. The association diagram quantitatively describes the mutual influence and constraint relationship between the water temperature, dissolved oxygen and salinity parameters, and displays the dynamic change process of the parameter relationship on the time axis. In aquaculture farms, water temperature, dissolved oxygen and salinity data within one year were extracted from the hydrological condition time series data set. The median absolute deviation method was used to detect outliers, and data points exceeding 3 times the standard deviation were marked as abnormal. The moving average method with a 24-hour window was used to smooth the data to obtain the processed hydrological parameter time series data. The processed data were z-score standardized to eliminate dimensional differences. The Pearson correlation coefficient and Spearman rank correlation coefficient were calculated, and it was found that the Pearson correlation coefficient of water temperature and dissolved oxygen was -0.75, and the Spearman rank correlation coefficient was -0.82, indicating a strong nonlinear negative correlation. The data were segmented using a sliding window of 7 days, and multiple linear regression and generalized additive model fitting were performed within each window. The multiple linear regression equation was obtained: dissolved oxygen = 12.5-0.3 water temperature + 0.1 salinity, R 2is 0.68. The generalized additive model captures the nonlinear relationship between water temperature and dissolved oxygen, with a goodness of fit of 0.85. A time series analysis of the fitting results within the window showed that the negative correlation between dissolved oxygen and water temperature was strongest in summer, with a correlation coefficient of -0.9, and weakened to -0.6 in winter. A dynamic association graph was constructed, with the node size proportional to the parameter standard deviation. The water temperature node is the largest, with a diameter of 15 pixels. The thickness of the edge indicates the intensity of the influence, and the edge from water temperature to dissolved oxygen is the thickest, with a width of 5 pixels. The direction of the edge points from water temperature to dissolved oxygen, indicating that changes in water temperature dominate changes in dissolved oxygen. On the timeline, the association graph is updated every hour, clearly showing the daily and seasonal variation patterns of the parameter relationship.

[0023] Step S103, data processing is performed on the time series data set of seedling growth, and a seedling growth process model is established. By analyzing the growth rate variation pattern of seedlings at different growth stages, the key parameters of the seedling growth curve are determined, and the model is calibrated according to the initial quality of the seedlings and the stocking density factors.

[0024] A seedling growth time series data set is obtained, and key growth indicators are extracted from the seedling growth time series data set, wherein the key growth indicators include body length and weight; according to the key growth indicators, outliers are processed by using a moving median method and a method based on an interquartile range to obtain processed seedling growth time series data; according to the processed seedling growth time series data, the growth rate of the seedlings at different time points is calculated, and the growth rate curve is fitted by using a spline interpolation method to determine the key turning point of seedling growth; according to the growth stage divided by the key turning point, the growth curve of the seedling is fitted by using a Gompertz growth model, and the Levenberg growth model is used to fit the growth curve of the seedling. -Marquardt algorithm for nonlinear least squares estimation to obtain key parameters of the growth curve; according to the key parameters, a multilayer perceptron regression model is constructed based on initial quality and stocking density data to analyze the influence of the initial quality and stocking density on the key parameters of the growth curve; the analysis results of the multilayer perceptron regression model are integrated into the Gompertz model to obtain a corrected seedling growth process model; the corrected model is subjected to k-fold cross validation to evaluate its prediction accuracy under different initial conditions; if the prediction accuracy does not meet the preset threshold, the structure and parameters of the multilayer perceptron model are adjusted until the required prediction accuracy is achieved.

[0025] Specifically, key growth indicators such as body length and weight were extracted from the seedling growth time series data set, the data was cleaned and outliers were detected, and the moving median method and the interquartile range-based method were used to process the outliers to obtain the processed seedling growth time series data. According to the processed seedling growth time series data, the growth rate of the seedlings at different time points was calculated, and the growth rate curve was fitted by spline interpolation method to determine the key turning points of seedling growth, and the growth process was divided into different growth stages, including rapid growth period, deceleration growth period and stable growth period. According to the divided growth stages, the Gompertz growth model was used to fit the seedling growth curve, and the Levenberg-Marquardt algorithm was used for nonlinear least squares estimation to obtain the key parameters of the growth curve, including the asymptotic maximum weight, growth rate coefficient and growth curve inflection point. Based on the initial mass and stocking density data, a multi-layer perceptron regression model was constructed to analyze the influence of these factors on the key parameters of the growth curve, and the regression results were integrated into the Gompertz model to obtain the corrected seedling growth process model. The corrected model was subjected to k-fold cross validation to evaluate its prediction accuracy under different initial conditions. The model performance was quantified by the root mean square error and the coefficient of determination. If the prediction accuracy did not meet the preset threshold, the structure and parameters of the multi-layer perceptron model were adjusted until the required prediction accuracy was achieved. In an aquaculture farm, the body length and weight data of 10,000 fry were extracted from the time series data set of seedling growth, spanning 180 days. The moving median method and the interquartile range-based method were used to process the outliers, and a total of 213 abnormal data points were identified and corrected. The processed data showed that the average initial body length of the fry was 2.5 cm and the weight was 0.5 g. The growth rate curve was fitted using the cubic spline interpolation method, and R 2 The growth rate of the 1.5-day-old fry reached 0.98, and two key turning points were identified: body length of 5 cm, weight of 2 g on the 30th day and body length of 12 cm, weight of 25 g on the 90th day. The growth process was divided into a rapid growth period of 0-30 days, a deceleration growth period of 31-90 days, and a stable growth period of 91-180 days. The Gompertz growth model was used for fitting, and the Levenberg-Marquardt algorithm was used for parameter estimation. After 50 iterations, the convergence was obtained, and the asymptotic maximum weight was 500 g, the growth rate coefficient was 0.015 / day, and the inflection point was on the 60th day. A three-layer multilayer perceptron regression model was constructed, with 2 nodes in the input layer, 10 nodes in the hidden layer, and 3 nodes in the output layer, to analyze the effects of initial mass and stocking density on growth parameters. 80% of the data was used for model training, 20% for validation, and the learning rate was set to 0.01. After 1000 rounds of training, the mean square error dropped below 0.05. The regression results were integrated into the Gompertz model to obtain the corrected growth model. A 5-fold cross validation was performed, with an average root mean square error of 15 g and a coefficient of determination R2 It is 0.92, which meets the preset threshold requirement of 0.9 and no further adjustment of the model structure is required.

[0026] Step S104, using the hydrological condition parameters as the input of the established hydrological parameter association model and seedling growth process model, outputting the seedling growth index, training through a large amount of historical data of the seedling growth time series data set, obtaining the complex nonlinear mapping relationship between the hydrological conditions and the seedling growth, and realizing the seedling growth prediction based on the hydrological conditions.

[0027] The hydrological condition parameters and seedling growth indicators in the historical database were obtained, and data cleaning was performed on the hydrological condition parameters and seedling growth indicators, including filling missing values ​​with the median and removing outliers based on the interquartile range method. According to the cleaned hydrological condition parameters, the adaptive window method was used to extract time series features, and the window size with the strongest autocorrelation was selected by calculating the autocorrelation coefficients under different window sizes. Based on the extracted time series features, a long short-term memory network enhanced by the attention mechanism was constructed. The number of nodes in the input layer of the long short-term memory network was the same as the number of extracted features, and the attention layer was used to capture the importance of different time steps. The trained attention-enhanced long short-term memory network was integrated with the pre-established hydrological parameter association model and seedling growth process model. The stacked generalization method was used to fuse the output results of the three models using logistic regression as a meta-learner. A Monte Carlo simulation test was performed on the integrated model after fusion. By adding different degrees of random perturbations to the hydrological condition parameters, the prediction stability and reliability of the integrated model under different hydrological conditions were evaluated.

[0028] Specifically, hydrological condition parameters and corresponding seedling growth indicators were extracted from the historical database, and the data were cleaned, including filling missing values ​​with the median and removing outliers based on the interquartile range method, and then z-score standardization was performed. The data set was divided into a training set and a test set in chronological order, with the training set accounting for 80% of the total data volume and the test set accounting for 20%. The adaptive window method was used to extract time series features of hydrological condition parameters. By calculating the autocorrelation coefficient under different window sizes, the window size with the strongest autocorrelation was selected, and statistical features were extracted within the selected window, including the mean, standard deviation, maximum, minimum and median, as input features. A long short-term memory network enhanced by the attention mechanism was constructed. The number of nodes in the input layer was the same as the number of extracted features. The attention layer was used to capture the importance of different time steps. The hidden layer was set to two layers, each containing 128 neurons. The number of nodes in the output layer was the same as the number of seedling growth indicators. The Adam optimizer was used to train the model, and the early stopping method was used to prevent overfitting. The trained attention-enhanced long short-term memory network was integrated with the previously established hydrological parameter association model and seedling growth process model. The stacked generalization method was used to fuse the output results of the three models using logistic regression as a meta-learner to obtain the final seedling growth prediction results. The final integrated model was tested by Monte Carlo simulation. By adding different degrees of random perturbations to the hydrological condition parameters, the prediction stability and reliability of the model under different hydrological conditions were evaluated, and the confidence interval of the prediction results was calculated. In a large aquaculture farm, the historical database contained 5 years of hydrological condition parameters and seedling growth indicators, totaling 1825 days of data. During the data cleaning process, it was found that 3% of the data were missing, and the median of the previous and next 7 days was used to fill it; the outlier detection used the 1.5 times interquartile range method to identify and process 2.5% of the abnormal data points. The z-score standardization normalized all features to a distribution with a mean of 0 and a standard deviation of 1. The adaptive window method determined the optimal window size of 12 days by calculating the autocorrelation coefficient of different window sizes from 1 to 30 days, and 60 statistical features were extracted within this window. The constructed attention-enhanced long short-term memory network contains 256 input nodes, 2 hidden layers with 128 neurons each, and 8 output nodes corresponding to different growth indicators. The model training uses the Adam optimizer with a batch size of 64 and a learning rate of 0.001. After 500 iterations, the mean absolute error on the validation set drops below 0.05. In the stacked generalization integration process, 5-fold cross-validation is used to generate meta-features, and the regularization parameter of the logistic regression meta-learner is determined to be 0.1 through grid search. The Monte Carlo simulation test performs 1000 random perturbations, adding random fluctuations of ±5%, ±10%, and ±2% to water temperature, dissolved oxygen, and salinity, respectively. The 95% confidence interval width of the predicted results is calculated to be ±7.5% of the predicted value on average.

[0029] Aiming at the complex nonlinear mapping relationship between hydrological conditions and seedling growth, we realize seedling growth prediction based on hydrological conditions, analyze the photosynthesis and respiration of seedlings in different water temperature ranges, the root health and nutrient absorption of seedlings in different dissolved oxygen ranges, and the osmotic pressure of seedlings in different salinity ranges, and determine the growth patterns of seedlings under different water temperatures, dissolved oxygen levels and salinities.

[0030] The dynamic time warping algorithm was used to cluster the time series of water temperature, dissolved oxygen and salinity, and the low, medium and high ranges of the water temperature, dissolved oxygen and salinity parameters were obtained; according to the water temperature range, the net photosynthetic rate, dark respiration rate, chlorophyll fluorescence parameters and oxygen consumption rate of the seedlings were measured, and the relationship curve between water temperature and photosynthesis and respiration was constructed; for the dissolved oxygen range, the root length, root surface area and branch number parameters of the seedlings were measured, the nitrogen absorption rate was determined by the 15N tracer method, and the association model between dissolved oxygen and root development and nutrient absorption was established; within the salinity range, the osmotic pressure changes in the seedlings were measured, and the relationship curve between salinity and osmotic pressure was fitted by polynomial regression; the multivariate adaptive regression spline method was used to integrate the comprehensive effects of water temperature, dissolved oxygen and salinity on seedling growth, and a nonlinear mapping model between hydrological conditions and seedling growth was constructed; the influence of water temperature, dissolved oxygen and salinity factors was analyzed by partial dependence diagrams, and the growth law of seedlings under different hydrological conditions was determined.

[0031] Specifically, based on historical data, the dynamic time warping algorithm was used to cluster the time series of water temperature, dissolved oxygen, and salinity, and each parameter was divided into three intervals: low, medium, and high, to obtain the interval range of each parameter. For different water temperature intervals, the net photosynthetic rate, dark respiration rate, chlorophyll fluorescence parameters, and oxygen consumption rate of the seedlings were measured using a photosynthesis meter, and the relationship curve between water temperature and photosynthesis and respiration was established. For different dissolved oxygen intervals, the root health status and nutrient absorption capacity of the seedlings were evaluated, and the morphological parameters such as root length, root surface area, and number of branches were measured by a root scanner. The nitrogen absorption rate was determined by the 15N tracer method, and the association model between dissolved oxygen and root development and nutrient absorption was constructed. In different salinity intervals, the osmotic pressure meter was used to measure the osmotic pressure changes in the seedlings, and the relationship curve between salinity and osmotic pressure was fitted by polynomial regression. The multivariate adaptive regression spline method was used to integrate the comprehensive effects of water temperature, dissolved oxygen, and salinity on seedling growth, and a nonlinear mapping model between hydrological conditions and seedling growth was constructed. The influence of each factor was analyzed by partial dependence diagram, and the growth law of seedlings under different hydrological conditions was comprehensively determined. In a large aquaculture farm, 1000 fish fry were monitored for 90 days. The dynamic time warping algorithm was used to cluster the time series of water temperature, dissolved oxygen, and salinity, and the water temperature range was 15-20℃, 20-25℃, 25-30℃, the dissolved oxygen range was 4-6mg / L, 6-8mg / L, 8-10mg / L, and the salinity range was 15-20‰, 20-25‰, 25-30‰. The physiological indicators of seedlings were measured in different water temperature ranges using LI-6400XT portable photosynthesis meter. The results showed that in the range of 25-30℃, the net photosynthetic rate was the highest, reaching 12μmol CO2·m-2·s-1, and the dark respiration rate was 2.5μmol CO2·m-2·s-1. The WinRH I ZO root analysis system was used to evaluate the root health in different dissolved oxygen ranges. In the range of 8-10 mg / L, the average root length was 15 cm and the root surface area was 25 cm 2 , with 35 branches. 15N tracer determination showed that the nitrogen absorption rate was the highest in this range, at 0.8mgN·g-1 dry weight·h-1. The osmotic pressure meter measurement results showed that in the salinity range of 20-25‰, the osmotic pressure in the seedlings was closest to the external environment, at 320mOsm / kg. The integrated analysis of the multivariate adaptive regression spline method showed that the influence weights of water temperature on seedling growth were 0.5, dissolved oxygen content was 0.3, and salinity was 0.2. Partial dependence diagram analysis further confirmed that under the combined conditions of water temperature of 25-28℃, dissolved oxygen content of 8-9mg / L, and salinity of 22-24‰, the seedling growth rate was the fastest, with a daily weight gain rate of 3.5%.

[0032] Step S105, introduce other influencing factors in the breeding process, including feed feeding amount and breeding density, construct a multi-source heterogeneous data set of other influencing factors in the breeding process, unify the representation and association analysis of the multi-source data set of other influencing factors, obtain the implicit relationship between each factor and seedling growth, and form a growth prediction model.

[0033] The data of feed feeding amount and stocking density in the aquaculture management system are obtained, and a multi-source heterogeneous data set is constructed in combination with the hydrological condition data; the z-score standardization method is used to process the multi-source heterogeneous data set to obtain standardized data with a mean of 0 and a standard deviation of 1; based on the standardized data, the kernel principal component analysis method is used to reduce the data dimension, and if the explained variance ratio reaches a preset threshold, the reduced-dimensional feature data set is determined; for the feature data set, the mutual information analysis method is used to calculate the nonlinear time dependence relationship between each influencing factor and the seedling growth index, and the optimal time window is determined; based on the feature data set and the optimal time window, a long short-term memory network growth prediction model is constructed, wherein the reduced-dimensional feature data and time delay data are used as input, and the seedling growth index is used as output; the parameters of the long short-term memory network growth prediction model are optimized by cross-validation to obtain the final growth prediction model; the SHAP value analysis method is used to interpret the prediction results of the growth prediction model, and the contribution of each influencing factor to seedling growth is quantified.

[0034] Specifically, the feed feeding amount and stocking density data were extracted from the aquaculture management system, combined with the existing hydrological condition data, to construct a multi-source heterogeneous data set, and the z-score standardization method was used to unify the data from different sources and units to a mean of 0 and a standard deviation of 1. The standardized multi-source data was subjected to feature extraction, and the kernel principal component analysis method was used to reduce the data dimension, retain the principal components with an explained variance ratio of 95%, obtain the reduced-dimensional feature data set, and reconstruct it into a time series format. The mutual information analysis method was used to calculate the nonlinear time dependence between each influencing factor and the seedling growth index, determine the optimal time window, and construct a data set considering the time delay effect. The growth prediction model was constructed using a long short-term memory network, with the reduced-dimensional feature data and time delay data as input, and the seedling growth index as output. The model parameters were optimized through cross-validation to obtain the final growth prediction model. The SHAP value analysis method was used to interpret the model prediction results, quantify the contribution of each influencing factor to seedling growth, and reveal the implicit relationship between factors such as feed feeding amount and stocking density and seedling growth. In a large aquaculture farm, 180 days of data were extracted from the aquaculture management system, including daily feed intake, ranging from 0.5-2.5 kg / m 3 and stocking density, ranging from 50-200 fish / m 3, combined with the existing water temperature, dissolved oxygen and salinity data, a multi-source heterogeneous data set containing 5 variables was constructed. The z-score standardization method was used to process the data, and all variables were adjusted to a range of mean 0 and standard deviation 1. Kernel principal component analysis with RBF kernel function was used for dimensionality reduction, and the first three principal components were selected, with a cumulative explained variance of 96.7%. The reconstructed time series data contained 180 time points, each of which contained 3 eigenvalues. Through mutual information analysis, it was found that the effect of feed amount on seedling growth had a lag effect of 3 days, and the effect of stocking density had a lag period of 7 days. The constructed long short-term memory network contained 64 hidden units, and the data of the past 14 days were used to predict the growth of the next 7 days. After 5-fold cross-validation, the root mean square error of the model on the test set was 0.15g, relative to the average body weight. SHAP value analysis showed that the importance of feed amount, stocking density, water temperature, dissolved oxygen, and salinity in predicting seedling growth rate was 35%, 25%, 20%, 15%, and 5%, respectively. It was found that the feed rate was 1.8-2.2kg / m 3 The growth promoting effect is most obvious within the range, and the stocking density exceeds 150 tails / m 3 It will significantly inhibit growth.

[0035] By combining the data of feed feeding amount and stocking density for correlation analysis, a multivariate growth prediction model was constructed, and the combinations of feed feeding amount and stocking density within the target range were identified. The seedling growth curves under the combinations within the target range were monitored separately, and the implicit relationship between feed feeding amount and stocking density and seedling growth was determined based on the two sets of growth curves.

[0036] Obtain historical data on feed amount and stocking density and seedling growth index data in the breeding management system. Use Pearson correlation coefficient and mutual information analysis methods to calculate linear and nonlinear correlations between variables, and select intervals with correlation coefficients greater than a preset threshold or mutual information values ​​ranked in the top 20% as preliminary target intervals. According to the preliminary target interval, different combinations of the feed amount and the stocking density are generated by the grid search method to construct a multivariate growth prediction model. The multivariate growth prediction model is combined with the NSGA-II multi-objective optimization algorithm to predict and optimize the feed amount and stocking density combination within the target interval, and a non-dominated solution set is obtained as the optimized target interval. For the feed amount and stocking density combination within the optimized target interval, a functional data analysis method is used to process the seedling growth curve, and the asymptotic weight, growth rate and inflection time of the seedling growth curve are extracted. The effects of the feed amount and the stocking density on seedling growth were quantitatively evaluated through partial dependence diagrams and SHAP value analysis methods, and an interpretable implicit relationship model was constructed to reveal the nonlinear interaction between the feed amount and the stocking density and seedling growth.

[0037] Specifically, the historical data of feed amount and stocking density were extracted from the breeding management system, combined with the seedling growth index data, and the Pearson correlation coefficient and mutual information analysis methods were used to calculate the linear and nonlinear correlations between variables, and the intervals with correlation coefficients greater than 0.6 or mutual information values ​​ranked in the top 20% were selected as the preliminary target intervals. Based on the preliminary target intervals, the grid search method was used to generate different combinations of feed amount and stocking density, and a multivariate growth prediction model was constructed. The random forest algorithm was used for model training, and the model parameters were optimized by cross-validation. Using the trained multivariate growth prediction model, combined with the NSGA-II multi-objective optimization algorithm, while considering the growth rate and feed conversion rate, the feed amount and stocking density combinations within the target interval were predicted and optimized, and the non-dominated solution set was screened out as the optimized target interval. For the feed amount and stocking density combinations within the optimized target interval, the functional data analysis method was used to process the seedling growth curve, and the key characteristic parameters of the growth curve, such as asymptotic weight, growth rate and inflection time, were extracted. The partial dependence diagram and SHAP value analysis method were used to quantitatively evaluate the effects of feed amount and stocking density on seedling growth, and an interpretable implicit relationship model was constructed to reveal the nonlinear interaction between feed amount and stocking density and seedling growth. In a large aquaculture farm, the system extracted 180 days of historical data from the aquaculture management database, including daily feed amount of 0.5-3kg / m 3 , and the stocking density is 50-250 / m 3 Records, and the corresponding average daily growth rate of seedlings. Through Pearson correlation coefficient analysis, it was found that the correlation coefficient between feed amount and growth rate was 0.72, and the correlation coefficient between stocking density and growth rate was -0.65. Mutual information analysis showed that the mutual information values ​​of feed amount and stocking density ranked second and third among all factors, respectively. Based on these results, the system determined the initial target range to be 1.5-2.5kg / m 3 , stocking density 100-200 fish / m 3 Subsequently, the system used grid search to generate 100 different combinations, built a random forest model for training, set 100 decision trees, and set the minimum number of leaf node samples to 5. After 5-fold cross validation, the model's R on the test set was 2 The NSGA-II algorithm takes the average daily growth rate and feed conversion rate as the optimization targets, the population size is set to 50, and 20 non-dominated solutions are obtained after 200 generations of evolution. Functional data analysis extracts features from these 20 groups of growth curves, including asymptotic weight, average 565g, maximum growth rate, average 3.8g / day), inflection point time, average 45 days. Partial dependence plot analysis shows that when the feed amount is 2.0-2.3kg / m 3 The stocking density is 120-150 fish / m3 The growth effect was the best when . SHAP value analysis further quantified the effects of feed amount (importance 38%) and stocking density (importance 31%) on growth, revealing the nonlinear interaction between them and providing data support for optimizing the farming strategy.

[0038] Step S106, combining the seedling growth prediction model with the intelligent optimization algorithm, dynamically adjusting the key control parameters in the breeding process, adaptively optimizing the hydrological conditions and feeding strategies according to the seedling growth trend predicted by the model, and realizing intelligent management of the breeding process.

[0039] The hydrological parameter data collected in real time by the aquaculture environment monitoring system are obtained. The hydrological parameter data include water temperature, dissolved oxygen content and salinity. The hydrological parameter data are combined with the feed feeding amount and aquaculture density information, and multi-source data fusion is performed through the Kalman filter to obtain an input data set in a unified format. The input data set is passed into the pre-trained seedling growth prediction model, and the long short-term memory network algorithm is used to predict the future growth trend of the seedlings, and the growth curve and key growth indicators are determined according to the prediction results. According to the growth trend, the NSGA-II multi-objective optimization algorithm is used to dynamically optimize the hydrological conditions and feeding strategies. The NSGA-II multi-objective optimization algorithm sets the objective function including maximizing growth rate, minimizing feed coefficient, maintaining environmental stability and improving economic benefits, and judging the optimal control parameter combination. The control parameters are passed to the aquaculture environment control system, and the oxygenation equipment, temperature control equipment and automatic feeding system are adjusted through a closed-loop feedback mechanism. The control parameters are updated every six hours, and the alarm function is triggered if the parameters exceed the preset range. Acquire the newly collected aquaculture data, regularly update the prediction model and optimization algorithm, adjust the model parameters through online learning, and judge the long-term adaptability and prediction accuracy of the system.

[0040] Specifically, the hydrological parameter data such as water temperature, dissolved oxygen, and salinity are obtained from the aquaculture environment monitoring system in real time. Combined with the information of feed feeding amount and aquaculture density, the Kalman filter is used to perform multi-source data fusion, and the heterogeneous data are integrated into a unified format input data set. The integrated input data set is passed to the pre-trained seedling growth prediction model, and the long short-term memory network algorithm is used to predict the seedling growth trend in the next 30 days. The trend, seasonality and random components are analyzed using time series decomposition technology to obtain the predicted growth curve and key growth indicators. According to the predicted growth trend, the NSGA-II multi-objective optimization algorithm is used to dynamically optimize the hydrological conditions and feeding strategies. The objective function is set to maximize the growth rate, minimize the feed coefficient, maintain environmental stability and improve economic benefits, and solve the optimal control parameter combination. The optimized control parameters are passed to the aquaculture environment control system, and the oxygenation equipment, temperature control equipment and automatic feeding system are adjusted in real time through the closed-loop feedback mechanism. The control parameters are updated every 6 hours, and the abnormal detection threshold is set. When the parameters exceed the preset range, the alarm function is triggered. The prediction model and optimization algorithm are updated regularly with newly collected data, and the model parameters are adjusted through online learning to improve the long-term adaptability and prediction accuracy of the system. In an intelligent aquaculture farm, the environmental monitoring system collects data every 5 minutes, including water temperature, dissolved oxygen and salinity. The Kalman filter uses a 15-minute time window to fuse these data with the daily recorded feed feeding amount and stocking density. The long short-term memory network model, which contains 3 hidden layers and 128 neurons in each layer, uses the data of the past 90 days to predict the growth trend of the next 30 days. The time series decomposition uses the STL method to extract the seasonal components and long-term trends with a period of 7 days. The NSGA-II algorithm sets the population size to 100 and the evolutionary generations to 200. The optimization objectives include the average daily weight gain rate, feed coefficient, dissolved oxygen fluctuation range and profit rate. The closed-loop control system updates parameters every 6 hours, such as adjusting the aerator power, ranging from 0-100%, the heater temperature, ranging from 18-32℃, and the automatic feeding amount, ranging from 0.5-3kg / m 3 / day. The abnormal detection setting triggers an alarm when the dissolved oxygen content is lower than 4mg / L or the temperature changes by more than 3℃ / day. The system updates the model at 2 a.m. every Sunday, using the sliding window method to incorporate the latest 7 days of data, and online learning for 100 rounds, reducing the prediction error by 12% on average. Through this intelligent management system, the farm increased the average daily weight gain rate by 15%, feed utilization by 8%, and reduced energy consumption by 10% within 3 months.

[0041] Step S107, develop a model-based breeding decision support system, which obtains monitoring data of the breeding environment in real time, calls the growth prediction model for online prediction, and gives optimization suggestions for the breeding strategy based on the prediction results. At the same time, the breeding decision support system tracks the historical data of the breeding process, continuously evaluates and improves the model, and provides data-driven solutions for refined breeding management.

[0042] The environmental parameter data collected in real time by the aquaculture environment monitoring system are obtained. The environmental parameter data include water temperature, dissolved oxygen content and pH value. The 3σ rule is used to detect abnormal values ​​and perform data standardization on the environmental parameter data to obtain standardized environmental parameter data. According to the standardized environmental parameter data, the pre-trained growth prediction model is called, and the growth prediction results of the seedlings in the next 7 days and 30 days are calculated by the long short-term memory network algorithm to obtain the predicted growth curve and key growth indicators. According to the growth prediction results, the NSGA-II multi-objective optimization algorithm is used to generate aquaculture strategy recommendations, and the aquaculture strategy recommendations include the optimal feed feeding amount, water quality adjustment plan and aquaculture density adjustment recommendations. The historical data of the aquaculture process is obtained, and the growth prediction results are compared and analyzed with the actual situation, and the prediction deviation indicators of the root mean square error and the mean absolute error are calculated. If the prediction deviation index exceeds the preset threshold, the model update process is triggered. A digital twin model for different aquaculture units is established, and the digital twin model is used to realize the refined management of sub-pools and batches, and determine personalized aquaculture decision support plans.

[0043] Specifically, the water temperature, dissolved oxygen, pH value, ammonia nitrogen, nitrite and other environmental parameter data are obtained in real time from the aquaculture environment monitoring system. The 3σ rule is used in the data preprocessing module to detect outliers and standardize the data, and the processed data is transmitted to the core computing unit of the decision support system. The core computing unit calls the pre-trained growth prediction model and uses the long short-term memory network algorithm to predict the growth of seedlings in the next 7 days and 30 days, obtains the predicted growth curve and key growth indicators, and calculates the confidence interval of the prediction results. The interpretability of the prediction results is provided through feature importance analysis. Based on the prediction results and the current aquaculture status, the decision support system uses the NSGA-II multi-objective optimization algorithm to generate aquaculture strategy recommendations, including the optimal feed feeding amount, water quality regulation plan and aquaculture density adjustment recommendations, and simulates and predicts the effects of the optimization recommendations after implementation, and presents the optimization recommendations and prediction results on the visual interface. The decision support system continuously tracks the historical data of the breeding process, regularly compares and analyzes the model prediction results with the actual situation, calculates the prediction deviation indicators such as the root mean square error and the mean absolute error, and automatically triggers the model update process when the deviation exceeds the preset threshold. The model is incrementally learned and optimized using the online gradient descent method. For different breeding units, a digital twin model is established to achieve refined management of sub-pools and batches, and provide personalized breeding decision support. In a modern aquaculture farm, the environmental monitoring system collects data every 5 minutes, including water temperature, dissolved oxygen, pH value, ammonia nitrogen and nitrite. The 3σ rule detects 2% of outliers and replaces them with the median of the previous 30 minutes. The long short-term memory network model contains 3 layers, each with 128 neurons. It uses the data of the past 90 days to predict the growth of the next 7 and 30 days. The average relative error of the 7-day prediction is 3.5%, and the 30-day prediction is 6.8%. Feature importance analysis shows that dissolved oxygen has the highest impact weight on growth, which is 0.35. The NSGA-II algorithm sets the population size to 100 and evolves for 200 generations. The optimization targets include daily weight gain rate, feed coefficient and water quality stability. The optimization results suggest that the feed amount should be adjusted to 2.3kg / m 3 / day, and the dissolved oxygen content is maintained at 6.5-7.5mg / L. Simulation prediction shows that the daily weight gain rate will increase by 12% within 7 days after the implementation of the optimization plan. The system performs model evaluation at 2 am every day, and when the root mean square error for 7 consecutive days exceeds 5%, incremental learning is triggered. The online gradient descent uses a learning rate of 0.01, and each iteration uses the 1000 data points of the latest 7 days, which reduces the prediction error by 15% on average. The digital twin model establishes a virtual mapping for the 10 independent breeding ponds in the farm, reflecting the status of each pond in real time, and formulating personalized management plans for each pond according to the differences in growth stages, such as adjusting the water temperature range and pH value.

[0044] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for monitoring the growth status of seedlings, characterized in that: The method comprises: Obtain real-time monitoring data on water temperature, dissolved oxygen, and salinity in the farm, and construct a time series data set of hydrological conditions based on water temperature, dissolved oxygen, and salinity. At the same time, obtain key growth indicator data on body length and weight during the growth of seedlings of different varieties to construct a time series data set of seedling growth; Process the time series data set of hydrological conditions to obtain the dynamic correlation between different hydrological parameters and construct a correlation model of hydrological parameters. The correlation model quantitatively describes the mutual influence and constraint relationship between water temperature, dissolved oxygen and salinity parameters. Data processing was performed on the time series data set of seedling growth to establish a seedling growth process model. By analyzing the growth rate variation patterns of seedlings at different growth stages, the key parameters of the seedling growth curve were determined, and the model was calibrated based on the initial quality of the seedlings and stocking density factors. The hydrological condition parameters are used as inputs to the established hydrological parameter association model and seedling growth process model, and seedling growth indicators are output. Through training with a large number of historical data of seedling growth time series data sets, the complex nonlinear mapping relationship between hydrological conditions and seedling growth is obtained, and seedling growth prediction based on hydrological conditions is realized. Introduce other influencing factors in the breeding process, including feed feeding amount and breeding density, construct a multi-source heterogeneous data set of other influencing factors in the breeding process, unify the representation and correlation analysis of the multi-source data set of other influencing factors, obtain the implicit relationship between each factor and seedling growth, and form a growth prediction model; Combine the seedling growth prediction model with the intelligent optimization algorithm to dynamically adjust the key control parameters in the breeding process. According to the seedling growth trend predicted by the model, the hydrological conditions and feeding strategies are adaptively optimized to achieve intelligent management of the breeding process. Develop a model-based breeding decision support system, which obtains monitoring data of the breeding environment in real time, calls the growth prediction model for online prediction, and gives optimization suggestions for breeding strategies based on the prediction results. At the same time, the breeding decision support system tracks the historical data of the breeding process, continuously evaluates and improves the model, and provides data-driven solutions for refined breeding management.

2. The method according to claim 1, characterized in that The real-time monitoring data of water temperature, dissolved oxygen and salinity in the farm are obtained, and a time series data set of hydrological conditions is constructed according to the water temperature, dissolved oxygen and salinity. At the same time, key growth indicator data of body length and weight during the growth of seedlings of different varieties are obtained to construct a time series data set of seedling growth, including: Acquire real-time data from multiple monitoring points, the real-time data including water temperature, dissolved oxygen content and salinity, clean, denoise and detect outliers on the real-time data to obtain a hydrological environment parameter data set; Extracting the morphological characteristics of the seedlings, including body length and body shape, based on the seedling images collected by the underwater camera system; By using an automatic weighing system provided in a specific area of ​​the culture pond, weight data of the seedlings are obtained when the seedlings pass through; Using an image segmentation tool to separate individual seedlings from the collected images, combining with a pre-established seedling feature database to perform variety matching and identification, and obtaining variety information of the seedlings in the breeding pond; Calculating the correlation coefficient between the hydrological condition and the seedling growth index, and if the correlation coefficient exceeds a preset threshold, determining that the hydrological condition is the hydrological factor that has the greatest impact on the growth of the seedling; According to the calculation results, a correlation model between the hydrological environment parameters and the seedling growth indicators is established.

3. The method according to claim 1, characterized in that The data processing of the time series data set of hydrological conditions is performed to obtain the dynamic correlation law between different hydrological parameters, and to construct a correlation model of hydrological parameters. The correlation model quantitatively describes the mutual influence and constraint relationship between water temperature, dissolved oxygen, and salinity parameters, including: Extracting data series of three parameters, namely, water temperature, dissolved oxygen and salinity, from the hydrological condition time series data set, performing data cleaning and outlier detection on the data series, and obtaining processed hydrological parameter time series data using a 24-hour window moving average method; The processed hydrological parameter time series data were standardized to eliminate the dimensional differences between different parameters, and the linear and nonlinear correlations between water temperature, dissolved oxygen, and salinity were calculated using the Pearson correlation coefficient and the Spearman rank correlation coefficient. The standardized hydrological parameter data were segmented using sliding window technology, with the window size set to 7 days. Multiple linear regression analysis and generalized additive model fitting were performed in each time window to establish the mathematical relationship between water temperature, dissolved oxygen and salinity. Based on the results of the multivariate linear regression analysis and the generalized additive model fitting, a time series analysis is performed on the fitting results of each time window to identify the periodic changes and long-term trends of the parameter relationship; Based on the results of regression analysis and generalized additive model, a dynamic association diagram of hydrological parameters was constructed, in which nodes represent three parameters: water temperature, dissolved oxygen and salinity. The size of the nodes represents the degree of variation of the parameters. The edges represent the influence relationship between the parameters. The thickness of the edges represents the intensity of the influence. The direction of the edges represents the dominant direction of the influence. The association diagram quantitatively describes the mutual influence and constraint relationship between the water temperature, dissolved oxygen and salinity parameters, and displays the dynamic change process of the parameter relationship on the timeline.

4. The method according to claim 1, characterized in that The data processing of the seedling growth time series data set is performed to establish a seedling growth process model, and the key parameters of the seedling growth curve are determined by analyzing the growth rate change law of the seedlings at different growth stages, and the model is corrected according to the initial quality of the seedlings and the stocking density factors, including: Acquire a seedling growth time series data set, and extract key growth indicators from the seedling growth time series data set, wherein the key growth indicators include body length and weight; According to the key growth indicators, outliers are processed using a moving median method and an interquartile range-based method to obtain processed seedling growth time series data; Calculating the growth rate of the seedlings at different time points according to the seedling growth time series data after the treatment, fitting the growth rate curve using the spline interpolation method, and determining the key turning point of the seedling growth; According to the growth stages divided by key turning points, the Gompertz growth model was used to fit the growth curve of the seedlings, and the nonlinear least squares estimation was performed using the Levenberg-Marquardt algorithm to obtain the key parameters of the growth curve. According to the key parameters, a multi-layer perceptron regression model is constructed based on the initial mass and stocking density data to analyze the effects of the initial mass and stocking density on the key parameters of the growth curve; Integrating the analysis results of the multi-layer perceptron regression model into the Gompertz model to obtain a corrected seedling growth process model; Performing k-fold cross validation on the corrected model to evaluate its prediction accuracy under different initial conditions; If the prediction accuracy does not meet the preset threshold, the structure and parameters of the multi-layer perceptron model are adjusted until the required prediction accuracy is achieved.

5. The method according to claim 1, characterized in that The method uses hydrological condition parameters as inputs of the established hydrological parameter association model and seedling growth process model, outputs seedling growth indicators, and trains with a large amount of historical data of a time series data set of seedling growth to obtain a complex nonlinear mapping relationship between hydrological conditions and seedling growth, thereby realizing seedling growth prediction based on hydrological conditions, including: Obtaining hydrological condition parameters and seedling growth indicators from a historical database, and performing data cleaning on the hydrological condition parameters and seedling growth indicators, including filling missing values ​​with the median and removing outliers based on the interquartile range method; According to the cleaned hydrological condition parameters, the adaptive window method is used to extract time series features. The window size with the strongest autocorrelation is selected by calculating the autocorrelation coefficients under different window sizes. Based on the extracted temporal features, a long short-term memory network enhanced by an attention mechanism is constructed, wherein the number of input layer nodes of the long short-term memory network is the same as the number of extracted features, and the attention layer is used to capture the importance of different time steps; The trained attention-enhanced long short-term memory network was integrated with the pre-established hydrological parameter association model and seedling growth process model, and the output results of the three models were fused using the stacked generalization method and logistic regression as a meta-learner. A Monte Carlo simulation test is conducted on the integrated model after fusion, by adding different degrees of random disturbances to the hydrological condition parameters to evaluate the prediction stability and reliability of the integrated model under different hydrological conditions; It also includes: targeting the complex nonlinear mapping relationship between hydrological conditions and seedling growth, realizing seedling growth prediction based on hydrological conditions, analyzing the photosynthesis and respiration of seedlings in different water temperature ranges, the root health and nutrient absorption of seedlings in different dissolved oxygen ranges, and the osmotic pressure of seedlings in different salinity ranges, and determining the growth patterns of seedlings under different water temperatures, dissolved oxygen levels, and salinities.

6. The method according to claim 5, characterized in that The complex nonlinear mapping relationship between hydrological conditions and seedling growth is used to predict seedling growth based on hydrological conditions, analyze the photosynthesis and respiration of seedlings in different water temperature ranges, the root health and nutrient absorption of seedlings in different dissolved oxygen ranges, and the osmotic pressure of seedlings in different salinity ranges, and determine the growth rules of seedlings under different water temperatures, dissolved oxygen, and salinity, respectively, including: The dynamic time warping algorithm is used to cluster the time series of water temperature, dissolved oxygen, and salinity to obtain the low, medium, and high ranges of the water temperature, dissolved oxygen, and salinity parameters; According to the range of water temperature, the net photosynthetic rate, dark respiration rate, chlorophyll fluorescence parameters and oxygen consumption rate of the seedlings were measured, and the relationship curve between water temperature and photosynthesis and respiration was constructed; In the range of dissolved oxygen, the root length, root surface area and branch number of the seedlings were measured, the nitrogen absorption rate was determined by the 15N tracer method, and a correlation model between dissolved oxygen, root development and nutrient absorption was established. Within the salinity range, the osmotic pressure changes in the seedlings were measured, and the relationship curve between salinity and osmotic pressure was fitted by polynomial regression. The multivariate adaptive regression spline method was used to integrate the comprehensive effects of water temperature, dissolved oxygen and salinity on seedling growth, and a nonlinear mapping model between hydrological conditions and seedling growth was constructed. The influence of water temperature, dissolved oxygen and salinity was analyzed through partial dependence diagram to determine the growth pattern of seedlings under different hydrological conditions.

7. The method according to claim 1, characterized in that The other influencing factors in the breeding process are introduced, including feed feeding amount and breeding density, to construct a multi-source heterogeneous data set of other influencing factors in the breeding process, to perform unified representation and association analysis on the multi-source data set of other influencing factors, to obtain the implicit relationship between each factor and seedling growth, and to form a growth prediction model, including: Obtain feed amount and stocking density data in the aquaculture management system, and build a multi-source heterogeneous data set in combination with hydrological condition data; The multi-source heterogeneous data set is processed using a z-score standardization method to obtain standardized data with a mean of 0 and a standard deviation of 1; According to the standardized data, the data dimension is reduced by using the kernel principal component analysis method, and if the explained variance ratio reaches a preset threshold, the feature data set after the dimension reduction is determined; For the characteristic data set, a mutual information analysis method is used to calculate the nonlinear time dependence relationship between each influencing factor and the seedling growth index, and determine the optimal time window; According to the feature data set and the optimal time window, a long short-term memory network growth prediction model is constructed, wherein the dimensionally reduced feature data and time delay data are used as input, and the seedling growth index is used as output; Optimizing the parameters of the long short-term memory network growth prediction model through cross-validation to obtain a final growth prediction model; The SHAP value analysis method is used to interpret the prediction results of the growth prediction model and quantify the contribution of each influencing factor to seedling growth; It also includes: conducting correlation analysis by combining data on feed amount and stocking density, constructing a multivariate growth prediction model, identifying combinations of feed amount and stocking density within a target range, monitoring the seedling growth curves under the combinations within the target range, and determining the implicit relationship between feed amount and stocking density and seedling growth based on the two sets of growth curves.

8. The method according to claim 7, characterized in that The method comprises the following steps: performing association analysis on the data of feed amount and stocking density, constructing a multivariate growth prediction model, identifying the combination of feed amount and stocking density within the target interval, monitoring the seedling growth curves under the combination within the target interval, and determining the implicit relationship between feed amount and stocking density and seedling growth according to the two groups of growth curves, including: Obtain historical data on feed feeding amount and stocking density and seedling growth index data in the aquaculture management system; Pearson correlation coefficient and mutual information analysis methods were used to calculate linear and nonlinear correlations between variables, and intervals with correlation coefficients greater than a preset threshold or mutual information values ​​ranked in the top 20 percent were selected as preliminary target intervals; According to the preliminary target interval, different combinations of the feed feeding amount and the breeding density are generated by a grid search method to construct a multivariate growth prediction model; The multivariate growth prediction model is combined with the NSGA-II multi-objective optimization algorithm to predict and optimize the feed amount and stocking density combination within the target interval, and a non-dominated solution set is obtained as the optimized target interval; For the combination of feed feeding amount and stocking density within the optimized target interval, a functional data analysis method is used to process the seedling growth curve, and the asymptotic weight, growth rate and inflection point time of the seedling growth curve are extracted; The effects of the feed amount and the stocking density on seedling growth were quantitatively evaluated through partial dependence diagrams and SHAP value analysis methods, and an interpretable implicit relationship model was constructed to reveal the nonlinear interaction between the feed amount and the stocking density and seedling growth.

9. The method according to claim 1, characterized in that: The seedling growth prediction model is combined with the intelligent optimization algorithm to dynamically adjust the key control parameters in the breeding process, adaptively optimize the hydrological conditions and feeding strategies according to the seedling growth trend predicted by the model, and realize the intelligent management of the breeding process, including: Acquire the hydrological parameter data collected in real time by the aquaculture environment monitoring system, the hydrological parameter data including water temperature, dissolved oxygen content and salinity, combine the hydrological parameter data with the feed amount and aquaculture density information, perform multi-source data fusion through the Kalman filter, and obtain an input data set in a unified format; The input data set is passed into a pre-trained seedling growth prediction model, a long short-term memory network algorithm is used to predict the future growth trend of the seedling, and a growth curve and key growth indicators are determined according to the prediction results; According to the growth trend, the hydrological conditions and feeding strategies are dynamically optimized using the NSGA-II multi-objective optimization algorithm. The NSGA-II multi-objective optimization algorithm sets objective functions including maximizing growth rate, minimizing feed coefficient, maintaining environmental stability and improving economic benefits, and determines the optimal control parameter combination; The control parameters are transmitted to the aquaculture environment control system, and the oxygenation equipment, temperature control equipment and automatic feeding system are adjusted through a closed-loop feedback mechanism. The control parameters are updated every six hours, and an alarm function is triggered if the parameters exceed the preset range; Acquire newly collected breeding data, regularly update prediction models and optimization algorithms, adjust model parameters through online learning, and determine the system's long-term adaptability and prediction accuracy.

10. The method according to claim 1, characterized in that The model-based aquaculture decision support system is developed. The aquaculture decision support system obtains monitoring data of the aquaculture environment in real time, calls the growth prediction model for online prediction, and gives optimization suggestions for aquaculture strategies based on the prediction results. At the same time, the aquaculture decision support system tracks the historical data of the aquaculture process, continuously evaluates and improves the model, and provides data-driven solutions for refined aquaculture management, including: Acquire environmental parameter data collected in real time by the aquaculture environment monitoring system, the environmental parameter data including water temperature, dissolved oxygen content and pH value, perform abnormal value detection and data standardization on the environmental parameter data using the 3σ rule, and obtain standardized environmental parameter data; According to the standardized environmental parameter data, a pre-trained growth prediction model is called, and the growth prediction results of the seedlings in the next 7 days and 30 days are calculated by a long short-term memory network algorithm to obtain a predicted growth curve and key growth indicators; Based on the growth prediction results, the NSGA-II multi-objective optimization algorithm is used to generate aquaculture strategy recommendations, which include optimal feed feeding amount, water quality regulation scheme and aquaculture density adjustment recommendations; Obtain historical data of the breeding process, compare and analyze the growth prediction results with the actual situation, calculate the prediction deviation indicators of the root mean square error and the mean absolute error, and trigger the model update process if the prediction deviation indicator exceeds the preset threshold; A digital twin model is established for different breeding units, and the digital twin model is used to achieve refined management of pools and batches and determine personalized breeding decision support solutions.

Images