A multi-parameter prediction method for water quality in marine ranches

By deploying corrosion-resistant multi-parameter water quality monitor and GIS system in marine ranches, combining machine learning and intelligent data fusion technology, multi-dimensional collection and future prediction of marine ranches water quality are achieved, identifying water quality changes trends and potential abnormalities, and evaluating ecology Risk, dynamically adjust the prediction model, solves functions that cannot be achieved in the existing technology, and improves marine ranch management efficiency and ecological security.

CN118553338BActive Publication Date: 2024-11-15GUANGDONG OCEAN UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411006491.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-25
Publication Date
2024-11-15
Estimated Expiration
2044-07-25

AI Technical Summary

Technical Problem

The existing marine ranch water quality monitoring technology cannot realize the multi-dimensional collection and future prediction of water quality parameters, identify water quality changes trends and potential abnormal situations, analyze the impact of environmental factors on marine ranch ecosystems, and dynamic adjustment and optimization prediction model functions, resulting in the failure of Timely identify and predict abnormal risks and the inability to target the regulation of water quality, resulting in low economic benefits of marine ranches.

Method used

Using corrosion-resistant and waterproof multi-parameter water quality monitor combined with GIS system, marine ranch water quality data is collected in real time through low-power wide area network/marine Internet of Things technology and LoRa data transmission protocol, and using machine learning algorithms for data preprocessing, feature extraction and Forecasting, combine satellite remote sensing and meteorological data to integrate intelligent data, build an ecological risk assessment model, and achieve real-time prediction and dynamic adjustment.

Benefits of technology

Achieve multi-dimensional collection and future prediction of water quality parameters, identify changes in water quality and potential abnormalities, evaluate ecological risks, dynamically adjust the prediction model, improve the management efficiency and ecological safety of marine ranch, reduce environmental risks, and enhance the system Flexibility and reliability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118553338B_ABST
    Figure CN118553338B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for predicting multi-parameter water quality in marine ranches, which relates to the technical field of water quality monitoring, and includes a central server, a data acquisition module, a model building module, a real-time prediction module, and a result analysis module. The data output by the data acquisition module and the model output by the model building module are used as inputs of the real-time prediction module, and the prediction results of the real-time prediction module are used as inputs of the result analysis module. The data acquisition module is used to collect multi-dimensional parameter data of water quality in marine ranches. The present invention realizes multi-dimensional collection and future prediction functions of water quality parameters by designing a data acquisition module and a real-time prediction module, solves the problem that it is difficult to dynamically adjust, collect multi-dimensionally, and predict in real time in traditional prediction methods, improves the accuracy and timeliness of water quality prediction by collecting data at different depths and dimensions, helps managers deal with potential problems in advance, improves the management efficiency and ecological safety of marine ranches, and reduces environmental risks.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of water quality monitoring, and in particular to a multi-parameter prediction method for water quality in an ocean ranch. Background Art

[0002] With the increasing scarcity of global marine resources and the intensification of environmental pollution problems, marine ranching has attracted widespread attention as a sustainable way to utilize marine resources. Marine ranching promotes the reproduction and proliferation of marine biological resources and achieves sustainable development of fishery resources by constructing artificial facilities and regulating the ecology in the ocean. However, the health and sustainable development of marine ranching are affected by many factors, among which water quality is one of the key factors. Water quality parameters such as temperature, salinity, pH, dissolved oxygen, ammonia nitrogen and phosphate are directly related to the survival and reproduction of marine organisms and have an important impact on the stability of marine ecosystems. Traditional water quality monitoring methods mainly rely on manual sampling and laboratory analysis, which have time lags and spatial limitations, making it difficult to grasp the water quality changes of marine ranches in real time and comprehensively.

[0003] With the development of Internet of Things technology, sensor technology and big data technology, real-time monitoring and intelligent prediction of marine ranch water quality have become possible. Through the sensor network deployed in the marine ranch, water quality parameter data can be collected in real time, and combined with advanced machine learning algorithms, multi-parameter water quality prediction can be performed, providing a scientific basis for the management and decision-making of the marine ranch.

[0004] Patent document CN116884523B discloses a method for multi-parameter prediction of marine ranch water quality. The above patent achieves high-precision prediction of multi-parameters of marine ranch water quality, improves the ability to respond to sudden disasters, ensures the aquaculture safety of marine ranches, and reduces the aquaculture risks and economic losses of marine ranches. However, the above patent cannot realize the multi-dimensional collection and future prediction functions of water quality parameters.

[0005] Patent document CN114371267B discloses a comprehensive water quality detection and alarm device for marine ranches. The above patent realizes the positioning of the monitoring device, which is convenient for detecting seawater at a specified location, but the above patent cannot realize the function of identifying water quality change trends and potential abnormal conditions.

[0006] Patent document CN105606404B discloses a marine ranch water quality detection system. The above patent improves the accuracy of water sample detection, but the above patent cannot realize the function of analyzing the impact of environmental factors on the marine ranch ecosystem.

[0007] Patent document CN108519472B discloses a multi-parameter in-situ online monitoring system for multi-water layer water quality environment in a marine ranch. The above patent can prevent the corrosion of detection modules and the like caused by the attachment of organisms to the wall of the sampling tube and the entry of water vapor through the sampling pipeline into the floating body on the water surface; it can also prevent the deviation of some parameter detection values ​​caused by the rapid release of water sample pressure during the sampling process, but the above patent cannot realize the dynamic adjustment and optimization of the prediction model function.

[0008] In summary, the above patents cannot realize the multi-dimensional collection and future prediction functions of water quality parameters, the function of identifying water quality change trends and potential abnormalities, the function of analyzing the impact of environmental factors on the marine ranch ecosystem, and the function of dynamically adjusting and optimizing the prediction model, resulting in the inability to predict water quality in the future, the inability to timely identify prediction risk anomalies, the inability to take targeted measures to regulate water quality, and the low economic benefits of marine ranches due to water quality deterioration;

[0009] To this end, this application proposes a multi-parameter prediction method for marine ranch water quality that can realize multi-dimensional collection and future prediction of water quality parameters, identify water quality change trends and potential abnormal situations, analyze the impact of environmental factors on marine ranch ecosystems, and dynamically adjust and optimize prediction models. Summary of the invention

[0010] The purpose of the present invention is to provide a multi-parameter prediction method for marine ranch water quality, so as to solve the technical problems proposed in the above background technology that the multi-dimensional collection and future prediction functions of water quality parameters, the function of identifying water quality change trends and potential abnormal situations, the impact analysis function of environmental factors on the marine ranch ecosystem and the dynamic adjustment and optimization of the prediction model cannot be realized, resulting in the inability to predict water quality in the future, the inability to timely identify prediction risk anomalies, the inability to take targeted measures to regulate water quality and the low economic benefits of marine ranches due to water quality deterioration.

[0011] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a method for predicting multi-parameter water quality in a marine ranch, comprising a data collection step, a model building step, a real-time prediction step and a result analysis step, wherein the data outputted from the data collection step and the model outputted from the model building step are applied to the real-time prediction step, the prediction result of the real-time prediction step is applied to the result analysis step, and the data collection step is used to collect multi-dimensional parameter data of water quality in the marine ranch;

[0012] The data collection step is to select a corrosion-resistant and waterproof multi-parameter water quality monitor, including a temperature sensor, a salinity sensor, a pH sensor, a dissolved oxygen sensor, an ammonia nitrogen sensor, and a phosphate sensor, and distribute sensor nodes according to the geographical location, water depth, and flow rate of the marine ranch in combination with the GIS system, adopt a grid layout, and regularly distribute node spacing. The sensor data is transmitted to the central server in real time using low-power wide area network / marine Internet of Things technology combined with the LoRa data transmission protocol;

[0013] The real-time prediction step is used to predict water quality changes based on the real-time water quality data collected in the data collection step;

[0014] The real-time prediction step includes: real-time data input and prediction result output;

[0015] The real-time data input is to input the real-time water quality data collected by the sensor into the trained prediction model;

[0016] The prediction result output is the water quality parameter predicted by the prediction model output, and the prediction result is updated in real time to generate water quality change prediction data for a period of time in the future;

[0017] The prediction method is designed with a data preprocessing step, which is used to preprocess the multidimensional parameters of the ocean ranch collected in the data collection step;

[0018] Data preprocessing steps include: removing outliers, data smoothing, and missing data filling;

[0019] Removing outliers means using the 3σ principle statistical method to identify and remove extreme outliers;

[0020] Data smoothing is to use moving average method / Kalman filter to smooth the data and eliminate short-term fluctuation noise;

[0021] Missing data filling is to fill in the missing data using linear interpolation / nearest neighbor interpolation / regression filling method to ensure the continuity and integrity of the data;

[0022] The prediction method is designed with a feature extraction step, which is used to extract key feature parameters from the data processed by the data preprocessing step;

[0023] The feature extraction steps include: feature selection and feature engineering;

[0024] Feature selection is to use correlation analysis / principal component analysis methods to select key characteristic parameters that affect water quality;

[0025] Feature engineering is to standardize and normalize the key characteristic parameters that affect water quality, ensure that data of different dimensions have the same weight in the prediction model, analyze the key characteristic parameters that affect water quality through time series, and extract time characteristics including seasonal and periodic changes;

[0026] The prediction method is also designed with a result analysis step, which is used to analyze the prediction results of the real-time prediction step;

[0027] The result analysis steps include: trend analysis and anomaly detection;

[0028] Trend analysis is to conduct trend analysis on the forecast results, identify the changing trends of water quality parameters, and use time series analysis methods to determine the periodicity and seasonality of water quality changes;

[0029] Anomaly detection is the use of statistical methods and machine learning algorithms, including control chart methods / anomaly detection algorithms, to detect anomalies in the prediction results and promptly warn of existing water quality problems;

[0030] The sensor data collected in the data collection step is applied to an intelligent data fusion step, and the intelligent data fusion step is used to fuse and analyze data from different sources with the sensor data;

[0031] The intelligent data fusion step combines data from different sources, including satellite remote sensing data, meteorological data, and historical water quality data, with sensor data using data fusion technology, and applies spatiotemporal analysis methods to process the temporal and spatial characteristics of the data, and combines GIS to build a spatiotemporal model of water quality parameters;

[0032] The result analysis step also includes an ecological risk assessment step, which is used for risk assessment and early warning;

[0033] The steps of ecological risk assessment include: establishing an ecological risk indicator system, building a risk assessment model and risk early warning response;

[0034] Establishing an ecological risk indicator system means establishing an ecological risk assessment indicator system based on water quality parameters. Ecological risk assessment indicators include toxicity thresholds and ecological health indexes. According to the characteristics of different water quality parameters and ecosystems, corresponding assessment standards and thresholds are set;

[0035] Constructing a risk assessment model means constructing a multi-level ecological risk assessment model, combining the predicted water quality parameters to assess ecological risks at different times and locations;

[0036] Risk warning response is to timely identify and warn of potential ecological risks based on prediction results, and display risk assessment results on the user interface through visualization tools.

[0037] Preferably, the prediction method is further designed with a model building step, and the model building step is used to build a prediction model;

[0038] The model building steps include: algorithm selection and determination of model structure;

[0039] Algorithm selection means selecting appropriate machine learning algorithms according to the characteristics of water quality parameters. Machine learning algorithms include multivariate linear regression, support vector machine, random forest and neural network;

[0040] Determine the model structure, that is, design the model structure, determine the number of neural network layers and the number of neurons in each layer, use LSTM / GRU recurrent neural network for time series data, and use cross-validation and grid search to determine the optimal parameter combination of the model.

[0041] Preferably, the prediction method is further designed with a model training step, and the model training step is used to train and tune the prediction model constructed in the model construction step;

[0042] The model training steps include: data set division and training process;

[0043] Dataset division is to divide the historical water quality data into training set, validation set and test set in a ratio of 8:1:1 for model training, validation and testing;

[0044] The training process is to train the model using the training set and adjust the model parameters through the validation set. The model parameters include learning rate and regularization parameters. The early stopping method / regularization technology is used to prevent the model from overfitting.

[0045] Preferably, the central server is connected to a user interface, and the user interface is used to display the water quality prediction results in real time;

[0046] The user interface provides data visualization and interaction capabilities;

[0047] Data visualization is to visualize the water quality prediction results in the form of charts to facilitate user understanding and analysis;

[0048] The interactive function provides interaction between users and the system. The interactive function includes parameter setting, data query and report generation. Users can adjust parameters, query historical data and generate forecast reports through the interface.

[0049] Preferably, the prediction method is further designed with an adaptive prediction model generation step, which is used for adaptive model selection and optimization update;

[0050] The steps of generating the adaptive prediction model include: adaptive model selection, model automatic optimization and model online update;

[0051] Adaptive model selection is to use ensemble learning methods to combine the advantages of multiple models and dynamically select prediction models based on the characteristics of water quality data;

[0052] Automatic model optimization is to use AutoML technology to automatically adjust and optimize model parameters;

[0053] Online model update means realizing online learning and updating of the model. As new data is continuously added, the model automatically adjusts and optimizes to maintain the real-time and accuracy of the prediction. Incremental learning methods are applied to avoid frequent retraining of the model.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] 1. The present invention realizes the multi-dimensional collection and future prediction functions of water quality parameters by designing data collection steps and real-time prediction steps, solves the problem that traditional prediction methods are difficult to dynamically adjust, collect multi-dimensional data and make real-time predictions, and improves the accuracy and timeliness of water quality predictions by collecting data at different depths and dimensions, helping managers to deal with potential problems in advance, improving the management efficiency and ecological safety of marine ranches, and reducing environmental risks;

[0056] 2. The present invention realizes the function of identifying water quality change trends and potential abnormal situations by designing a result analysis step, provides in-depth analysis and interpretation of water quality prediction results, helps managers understand the trends and laws behind the data, and intuitively displays water quality change trends and abnormal situations, which facilitates managers to respond quickly;

[0057] 3. The present invention realizes the function of analyzing the impact of environmental factors on the marine ranch ecosystem by designing an ecological risk assessment step, systematically assessing ecological risks, helping managers to fully understand the ecological security status of marine ranches, and improving the scientificity, ecological resilience and sustainability of marine ranch ecological management;

[0058] 4. The present invention realizes dynamic adjustment and optimization of the prediction model function by designing an adaptive prediction model generation step, solves the problem of performance degradation of traditional models when the environment changes, maintains the efficiency and accuracy of the prediction model, enhances the flexibility and robustness of the system, and improves the reliability and efficiency of the entire prediction system. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a schematic diagram of the water quality prediction process of the present invention. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Example 1

[0061] See also Figure 1 , an embodiment provided by the present invention: a method for predicting multi-parameter water quality in a marine ranch, comprising a data collection step, a model building step, a real-time prediction step and a result analysis step, wherein the data output by the data collection step and the model output by the model building step are applied to the real-time prediction step, the prediction result of the real-time prediction step is applied to the result analysis step, and the data collection step is used to collect multi-dimensional parameter data of water quality in the marine ranch;

[0062] The data collection step is to select a corrosion-resistant and waterproof multi-parameter water quality monitor, including a temperature sensor, a salinity sensor, a pH sensor, a dissolved oxygen sensor, an ammonia nitrogen sensor, and a phosphate sensor, and distribute sensor nodes according to the geographical location, water depth, and flow rate of the marine ranch in combination with the GIS system, adopt a grid layout, and regularly distribute node spacing. The sensor data is transmitted to the central server in real time using low-power wide area network / marine Internet of Things technology combined with the LoRa data transmission protocol;

[0063] The real-time prediction step is used to predict water quality changes based on the real-time water quality data collected in the data collection step;

[0064] The real-time prediction step includes: real-time data input and prediction result output;

[0065] The real-time data input is to input the real-time water quality data collected by the sensor into the trained prediction model;

[0066] The prediction result output is the water quality parameter predicted by the prediction model output, and the prediction result is updated in real time to generate water quality change prediction data for a period of time in the future;

[0067] Furthermore, using temperature, salinity, pH, dissolved oxygen, ammonia nitrogen and phosphate sensors, the sensors are evenly arranged at different depths (e.g. 10m, 20m, 30m) and areas of the marine ranch to ensure coverage of the entire monitoring area. The sensors automatically collect water quality data every hour and transmit the data to the central data server in real time. Using wireless sensor networks (e.g. LoRa, NB-IoT), the data collected by the sensors are transmitted to the central server in real time. The central server stores and manages the received raw data, creates a timestamp, and marks the data source (sensor number, location, depth, etc.);

[0068] The preprocessed real-time data is input into the trained LSTM model to generate water quality forecast results for a period of time in the future (such as one week). The forecast results are updated in real time to ensure the accuracy and real-time nature of the forecast. The forecast results are analyzed to identify trends and abnormal situations. The forecast results are displayed to users in the form of charts, curves, etc. through the user interface, providing interactive functions to facilitate user query and analysis. Example 2

[0069] See also Figure 1 , an embodiment provided by the present invention: a method for predicting multi-parameter water quality of a marine ranch, the prediction method is designed with a data preprocessing step, the data preprocessing step is used to preprocess the multi-dimensional parameters of the marine ranch collected in the data collection step;

[0070] Data preprocessing steps include: removing outliers, data smoothing, and missing data filling;

[0071] Removing outliers means using the 3σ principle statistical method to identify and remove extreme outliers;

[0072] Data smoothing is to use moving average method / Kalman filter to smooth the data and eliminate short-term fluctuation noise;

[0073] Missing data filling is to fill in the missing data using linear interpolation / nearest neighbor interpolation / regression filling method to ensure the continuity and integrity of the data.

[0074] The prediction method is designed with a feature extraction step, which is used to extract key feature parameters from the data processed by the data preprocessing step;

[0075] The feature extraction steps include: feature selection and feature engineering;

[0076] Feature selection is to use correlation analysis / principal component analysis methods to select key characteristic parameters that affect water quality;

[0077] Feature engineering is to standardize and normalize the key characteristic parameters that affect water quality, ensure that data of different dimensions have the same weight in the prediction model, analyze the key characteristic parameters that affect water quality through time series, and extract time characteristics including seasonal and periodic changes;

[0078] Furthermore, real-time and historical water quality data are obtained from the central server, the acquired data are cleaned, outliers are identified and removed, missing values ​​are filled using appropriate algorithms to ensure data quality and integrity, water quality parameter data are standardized, data of different dimensions are converted to a unified dimension for subsequent analysis and modeling, and smoothing algorithms are applied to smooth the data to reduce noise in the data and improve data stability;

[0079] According to the needs of water quality monitoring and ecological risk assessment, the key water quality parameters required for the prediction model are determined, the time characteristics, statistical characteristics and frequency domain characteristics are extracted, the characteristics of water quality parameters are comprehensively analyzed, and the extracted features are reduced by using PCA and correlation analysis methods. The most important features are selected to reduce the data dimension and improve the model efficiency. New features are created through feature combination, feature interaction and other methods to improve the expression ability and prediction performance of the model. The data after preprocessing and feature extraction are input into the prediction model to predict and analyze water quality parameters. Example 3

[0080] See also Figure 1 , an embodiment provided by the present invention: a multi-parameter prediction method for water quality in an ocean ranch, the prediction method is further designed with a model building step, the model building step is used to build a prediction model;

[0081] The model building steps include: algorithm selection and determination of model structure;

[0082] Algorithm selection means selecting appropriate machine learning algorithms according to the characteristics of water quality parameters. Machine learning algorithms include multivariate linear regression, support vector machine, random forest and neural network;

[0083] Determine the model structure, that is, design the model structure, determine the number of neural network layers and the number of neurons in each layer, use LSTM / GRU recurrent neural network for time series data, and use cross-validation and grid search to determine the optimal parameter combination of the model.

[0084] The prediction method is also designed with a model training step, which is used to train and optimize the prediction model constructed in the model construction step;

[0085] The model training steps include: data set division and training process;

[0086] Dataset division is to divide the historical water quality data into training set, validation set and test set in a ratio of 8:1:1 for model training, validation and testing;

[0087] The training process is to train the model using the training set and adjust the model parameters through the validation set. The model parameters include learning rate and regularization parameters. The early stopping method / regularization technology is used to prevent the model from overfitting.

[0088] Furthermore, we selected the LSTM model according to the characteristics of the data and designed its architecture, including the number of layers, number of neurons and activation function, determined the input layer, hidden layer and output layer structure of the model to ensure that the model can effectively capture the temporal characteristics of the data, selected the mean square error as the loss function to measure the prediction error of the model, selected the Adam optimizer to optimize the model parameters, minimize the loss function and improve the training efficiency of the model, and divided the data set into training set, validation set and test set to ensure the randomness and representativeness of the data segmentation;

[0089] Input the training data into the model and perform iterative training. Set the number of training rounds and batch size to ensure that the model fully learns the data features. In each training round, use the optimizer to adjust the model parameters and minimize the loss function. Use the validation set to validate the model and evaluate the performance of the model on unseen data. By monitoring the loss and evaluation indicators on the validation set, adjust the hyperparameters during the training process to prevent overfitting. After training, use the test set to perform a final evaluation of the model, including the structure and parameters of the model. Deploy the model to the prediction system to achieve real-time water quality prediction and ensure that the system can respond to and handle water quality changes in a timely manner. Example 4

[0090] See also Figure 1 , an embodiment provided by the present invention: a multi-parameter prediction method for water quality in a marine ranch, the prediction method is further designed with a result analysis step, the result analysis step is used to analyze the prediction result of the real-time prediction step;

[0091] The result analysis steps include: trend analysis and anomaly detection;

[0092] Trend analysis is to conduct trend analysis on the forecast results, identify the changing trends of water quality parameters, and use time series analysis methods to determine the periodicity and seasonality of water quality changes;

[0093] Anomaly detection is the use of statistical methods and machine learning algorithms, including control chart methods / anomaly detection algorithms, to detect anomalies in prediction results and provide timely warnings of existing water quality problems.

[0094] The central server is connected to a user interface, and the user interface is used to display the water quality prediction results in real time;

[0095] The user interface provides data visualization and interaction capabilities;

[0096] Data visualization is to visualize the water quality prediction results in the form of charts to facilitate user understanding and analysis;

[0097] The interactive function is to provide the user with the system's interactive function, including parameter setting, data query and report generation. The user can adjust parameters, query historical data and generate forecast reports through the interface;

[0098] Further, receive the predicted water quality parameter results from the model output, including temperature, salinity, pH, dissolved oxygen, ammonia nitrogen, phosphate, etc., preliminarily organize and format the predicted results to ensure data consistency, apply time series analysis methods to perform time series analysis on the predicted results, identify the changing trends of water quality parameters, use moving average, exponential smoothing and other methods to smooth the data, identify short-term and long-term trends, detect outliers and abnormal patterns in the predicted results through statistical methods and machine learning algorithms, apply control charts (such as Shewhart control charts) or anomaly detection algorithms (such as isolated forests, LOF) to identify abnormal situations, interpret and evaluate the predicted results, analyze the relationship and influencing factors between different water quality parameters, combine actual conditions and domain knowledge to conduct in-depth analysis of the predicted results, provide scientific explanations and suggestions, and generate detailed analysis reports based on the analysis results, including data charts, trend analysis, anomaly detection results, etc.;

[0099] The water quality data collected by the sensor and the water quality parameters predicted by the model are displayed in real time on the user interface. Charts (such as line charts, bar charts, pie charts, etc.) and dashboards are used to intuitively display data changes and trends. The abnormal detection results are displayed in real time on the user interface, and an early warning function is provided. Through color identification and pop-up prompts, users are reminded of abnormal situations and corresponding suggestions and measures are provided. Example 5

[0100] See also Figure 1 , an embodiment provided by the present invention: a method for predicting multi-parameter water quality in a marine ranch, wherein the sensor data collected in the data collection step is applied to an intelligent data fusion step, and the intelligent data fusion step is used to fuse and analyze data from different sources with the sensor data;

[0101] The intelligent data fusion step combines data from different sources, including satellite remote sensing data, meteorological data, and historical water quality data, with sensor data using data fusion technology, and applies spatiotemporal analysis methods to process the temporal and spatial characteristics of the data, and combines GIS to build a spatiotemporal model of water quality parameters.

[0102] The prediction method is also designed with an adaptive prediction model generation step, which is used for adaptive model selection and optimization update;

[0103] The steps of generating the adaptive prediction model include: adaptive model selection, model automatic optimization and model online update;

[0104] Adaptive model selection is to use ensemble learning methods to combine the advantages of multiple models and dynamically select prediction models based on the characteristics of water quality data;

[0105] Automatic model optimization is to use AutoML technology to automatically adjust and optimize model parameters;

[0106] Online model update means realizing online learning and updating of the model. As new data is continuously added, the model automatically adjusts and optimizes to maintain the real-time and accuracy of the prediction. Incremental learning methods are applied to avoid frequent retraining of the model.

[0107] Furthermore, select and train the initial model based on historical and real-time data, determine the appropriate algorithm and hyperparameters, the initial model provides basic prediction results, and lays the foundation for subsequent adaptive adjustments. Monitor model performance in real time, record prediction errors and evaluation indicators, set performance thresholds, and regularly add new data to the data set to ensure that the model is always trained with the latest data. When the model performance degrades or new data is available, start adaptive training, use online learning or incremental learning algorithms to adjust the model, apply sliding window technology to ensure the continuity and representativeness of the training data, compare the prediction performance of different models, select the optimal model for adaptive training, apply model integration methods, combine the prediction results of multiple models, improve prediction accuracy and stability, evaluate the model after adaptive training, record and analyze evaluation indicators to ensure model performance improvement, further optimize and adjust the model based on the evaluation results, form a closed-loop feedback mechanism, update the model after adaptive training to the production environment to ensure a smooth transition without affecting the normal operation of the system, and continuously monitor model performance to ensure that the system always maintains efficient and accurate prediction capabilities. Example 6

[0108] See also Figure 1 , an embodiment provided by the present invention: a method for multi-parameter prediction of water quality in a marine ranch, wherein the result analysis step further includes an ecological risk assessment step, which is used for risk assessment and early warning;

[0109] The steps of ecological risk assessment include: establishing an ecological risk indicator system, building a risk assessment model and risk early warning response;

[0110] Establishing an ecological risk indicator system means establishing an ecological risk assessment indicator system based on water quality parameters. Ecological risk assessment indicators include toxicity thresholds and ecological health indexes. According to the characteristics of different water quality parameters and ecosystems, corresponding assessment standards and thresholds are set;

[0111] Constructing a risk assessment model means constructing a multi-level ecological risk assessment model, combining the predicted water quality parameters to assess ecological risks at different times and locations;

[0112] Risk warning response means timely identifying and warning of potential ecological risks based on prediction results, and displaying risk assessment results on the user interface through visualization tools;

[0113] Furthermore, historical water quality data, real-time monitoring data and environmental factor data (such as temperature, salinity, pH, dissolved oxygen, ammonia nitrogen, phosphate, etc.) of the marine ranch are collected, and correlation analysis, regression analysis and other methods are applied to determine the key environmental factors and their interrelationships, and to study the impact of various environmental factors on the marine ranch ecosystem. Based on the results of the environmental factor impact analysis, potential ecological risk factors (such as water quality deterioration, pollutant accumulation, etc.) are identified, a risk indicator system is developed, and indicators and thresholds for various risk factors are clarified;

[0114] Construct a risk assessment model, select appropriate model algorithms (such as analytic hierarchy process (AHP), fuzzy comprehensive evaluation method, Bayesian network, etc.), determine the model input parameters, weights and evaluation criteria, ensure that the model can accurately reflect ecological risks, input the collected environmental data into the risk assessment model, calculate the risk value of each risk indicator, apply a comprehensive evaluation method, conduct a comprehensive evaluation of each risk factor, and determine the overall ecological risk level;

[0115] Set a risk warning threshold. When the risk value exceeds the threshold, the warning mechanism is triggered, and response measures and suggestions are provided, such as strengthening monitoring, adjusting management strategies, etc., to reduce the impact of ecological risks on marine ranches, and display the risk assessment results in the form of charts, reports, etc., to facilitate managers' understanding and decision-making.

[0116] Working principle: First, the data collection step monitors water quality parameters in real time in the ocean ranch and transmits the data to the central server. The collected data is cleaned, denoised and standardized in the data preprocessing step. At the same time, the feature extraction step performs feature extraction and recognition to reduce data dimensions and provide optimized data input for model training.

[0117] Next, through the model building step, model training step and real-time prediction step, the future changes in water quality parameters are predicted. The initial model is built based on historical data, and adaptive training is performed on the basis of real-time data to continuously optimize the model performance. The model training step uses online learning and incremental learning algorithms to ensure that the model can quickly adapt to environmental changes and new data inputs; the real-time prediction step uses the trained model to make real-time predictions of future water quality parameters and provide high-precision prediction results;

[0118] Finally, the result analysis step analyzes the prediction results, applies time series analysis and anomaly detection algorithms to identify water quality change trends and potential anomalies. The analysis results are intuitively displayed in the form of charts and reports through the user interface to facilitate user understanding and decision-making; if an abnormality is detected, the system will trigger an early warning mechanism and provide response measures and suggestions.

[0119] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A multi-parameter prediction method for marine ranch water quality, characterized by: The method comprises a data collection step, a model building step, a real-time prediction step and a result analysis step, wherein the data outputted from the data collection step and the model outputted from the model building step are applied to the real-time prediction step, the prediction result of the real-time prediction step is applied to the result analysis step, and the data collection step is used to collect multi-dimensional parameter data of water quality in the marine ranch; The data collection step is to select a corrosion-resistant and waterproof multi-parameter water quality monitor, including a temperature sensor, a salinity sensor, a pH sensor, a dissolved oxygen sensor, an ammonia nitrogen sensor, and a phosphate sensor, and distribute sensor nodes according to the geographical location, water depth, and flow rate of the marine ranch in combination with the GIS system, adopt a grid layout, and regularly distribute node spacing. The sensor data is transmitted to the central server in real time using low-power wide area network / marine Internet of Things technology combined with the LoRa data transmission protocol; The real-time prediction step is used to predict water quality changes based on the real-time water quality data collected in the data collection step; The real-time prediction step includes: real-time data input and prediction result output; The real-time data input is to input the real-time water quality data collected by the sensor into the trained prediction model; The prediction result output is the water quality parameter predicted by the prediction model output, and the prediction result is updated in real time to generate water quality change prediction data for a period of time in the future; The prediction method is designed with a data preprocessing step, which is used to preprocess the multidimensional parameters of the ocean ranch collected in the data collection step; Data preprocessing steps include: removing outliers, data smoothing, and missing data filling; Removing outliers means using the 3σ principle statistical method to identify and remove extreme outliers; Data smoothing is to use moving average method / Kalman filter to smooth the data and eliminate short-term fluctuation noise; Missing data filling is to fill in the missing data using linear interpolation / nearest neighbor interpolation / regression filling method to ensure the continuity and integrity of the data; The prediction method is designed with a feature extraction step, which is used to extract key feature parameters from the data processed by the data preprocessing step; The feature extraction steps include: feature selection and feature engineering; Feature selection is to use correlation analysis / principal component analysis methods to select key characteristic parameters that affect water quality; Feature engineering is to standardize and normalize the key characteristic parameters that affect water quality, ensure that data of different dimensions have the same weight in the prediction model, analyze the key characteristic parameters that affect water quality through time series, and extract time characteristics including seasonal and periodic changes; The prediction method is also designed with a result analysis step, which is used to analyze the prediction results of the real-time prediction step; The result analysis steps include: trend analysis and anomaly detection; Trend analysis is to conduct trend analysis on the forecast results, identify the changing trends of water quality parameters, and use time series analysis methods to determine the periodicity and seasonality of water quality changes; Anomaly detection is the use of statistical methods and machine learning algorithms, including control chart methods / anomaly detection algorithms, to detect anomalies in the prediction results and promptly warn of existing water quality problems; The sensor data collected in the data collection step is applied to an intelligent data fusion step, and the intelligent data fusion step is used to fuse and analyze data from different sources with the sensor data; The intelligent data fusion step combines data from different sources, including satellite remote sensing data, meteorological data, and historical water quality data, with sensor data using data fusion technology, and applies spatiotemporal analysis methods to process the temporal and spatial characteristics of the data, and combines GIS to build a spatiotemporal model of water quality parameters; The result analysis step also includes an ecological risk assessment step, which is used for risk assessment and early warning; The steps of ecological risk assessment include: establishing an ecological risk indicator system, building a risk assessment model and risk early warning response; Establishing an ecological risk indicator system means establishing an ecological risk assessment indicator system based on water quality parameters. Ecological risk assessment indicators include toxicity thresholds and ecological health indexes. According to the characteristics of different water quality parameters and ecosystems, corresponding assessment standards and thresholds are set; Constructing a risk assessment model means constructing a multi-level ecological risk assessment model, combining the predicted water quality parameters to assess ecological risks at different times and locations; Risk warning response is to timely identify and warn of potential ecological risks based on prediction results, and display risk assessment results on the user interface through visualization tools.

2. A method for multi-parameter prediction of water quality in marine ranches according to claim 1, characterized in that: The prediction method is also designed with a model building step, which is used to build a prediction model; The model building steps include: algorithm selection and determination of model structure; Algorithm selection means selecting appropriate machine learning algorithms according to the characteristics of water quality parameters. Machine learning algorithms include multivariate linear regression, support vector machine, random forest and neural network; Determine the model structure, that is, design the model structure, determine the number of neural network layers and the number of neurons in each layer, use LSTM / GRU recurrent neural network for time series data, and use cross-validation and grid search to determine the optimal parameter combination of the model.

3. The method for multi-parameter prediction of water quality in marine ranches according to claim 1, characterized in that: The prediction method is also designed with a model training step, which is used to train and optimize the prediction model constructed in the model construction step; The model training steps include: data set division and training process; Dataset division is to divide the historical water quality data into training set, validation set and test set in a ratio of 8:1:1 for model training, validation and testing; The training process is to train the model using the training set and adjust the model parameters through the validation set. The model parameters include learning rate and regularization parameters. The early stopping method / regularization technology is used to prevent the model from overfitting.

4. The method for multi-parameter prediction of water quality in marine ranches according to claim 1, characterized in that: The central server is connected to a user interface, and the user interface is used to display the water quality prediction results in real time; The user interface provides data visualization and interaction capabilities; Data visualization is to visualize the water quality prediction results in the form of charts to facilitate user understanding and analysis; The interactive function provides interaction between users and the system. The interactive function includes parameter setting, data query and report generation. Users can adjust parameters, query historical data and generate forecast reports through the interface.

5. The method for multi-parameter prediction of water quality in marine ranches according to claim 1, characterized in that: The prediction method is also designed with an adaptive prediction model generation step, which is used for adaptive model selection and optimization update; The steps of generating the adaptive prediction model include: adaptive model selection, model automatic optimization and model online update; Adaptive model selection is to use ensemble learning methods to combine the advantages of multiple models and dynamically select prediction models based on the characteristics of water quality data; Automatic model optimization is to use AutoML technology to automatically adjust and optimize model parameters; Online model update means realizing online learning and updating of the model. As new data is continuously added, the model automatically adjusts and optimizes to maintain the real-time and accuracy of the prediction. Incremental learning methods are applied to avoid frequent retraining of the model.