Total factor monitoring method for marine ranching

Abstract

The invention discloses a total factor monitoring method for a marine ranch, and belongs to the technical field of ocean engineering and information. The method comprises the following steps: collecting water quality parameters and biological population data in real time; the planktons and pollutants are identified through spectral analysis and chemical sensing, and environmental elements are classified; a sonar and an electronic tag are combined to track a fish trajectory, and dynamic monitoring data are obtained; predicting a red tide probability based on machine learning and outputting risk early warning; continuously updating the model to form multi-element integrated monitoring data; and finally, sorting the key indexes to generate marine ranch key element monitoring framework data. According to the invention, the real-time monitoring capability of the marine ranching environment and the biological state is obviously improved, the ecological risks such as red tide are reduced, and the sustainable development of the marine ranching is ensured.

Description

Technical Field

[0001] This invention belongs to the field of marine engineering and information technology, and in particular relates to a method for comprehensive monitoring of marine ranches. Background Technology

[0002] Marine ranching, as an important form of modern marine fisheries, plays a crucial role in ensuring marine ecological balance and sustainable resource utilization. Real-time monitoring of changes in the marine environment and biological populations can provide a scientific basis for ranch management, ensuring ecological health and economic benefits. However, existing monitoring methods have significant shortcomings in practical applications, struggling to address the complex and diverse environmental and biological elements within marine ranches. Many schemes rely too heavily on single-type sensors, failing to effectively integrate multi-source data, resulting in a lack of comprehensiveness and accuracy in monitoring results. Furthermore, existing technologies are inadequate in dynamic tracking and risk warning, failing to promptly capture rapidly changing ecological characteristics within marine ranches.

[0003] Therefore, the core challenge of comprehensive monitoring of marine ranches lies in achieving effective fusion and dynamic tracking of multi-source data. In the marine environment, factors such as temperature, salinity, and dissolved oxygen interact, and monitoring a single factor cannot comprehensively reflect the ranch's condition. For example, fish movement patterns may be influenced by both ocean currents and water temperature, making it difficult to accurately reconstruct their behavior patterns using only sonar or electronic tags. The complexity of data fusion often leads to information isolation or analysis delays in existing systems when processing multidimensional data. The inadequacy of dynamic tracking further exacerbates this contradiction, as rapid changes in biological populations and environmental factors in marine ranches require real-time, continuous monitoring. For instance, before a red tide occurs, abnormal water quality signals may be scattered across data from different sensors; failure to integrate and analyze these signals in a timely manner could result in missed warnings and ecological losses.

[0004] Therefore, how to effectively integrate and dynamically track multi-source data has become a key issue in the comprehensive monitoring methods for marine ranches. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a comprehensive monitoring method for marine ranches, comprising:

[0006] By deploying sensors to collect environmental parameters and biological population data of the marine ranch, a raw dataset is obtained;

[0007] Based on the original dataset, the species of planktonic organisms and water pollutants are classified and identified to obtain the classification results of environmental elements.

[0008] Based on the classification results of the environmental elements, the activity trajectories and distribution of key fish populations are tracked to obtain biological dynamic monitoring data;

[0009] Based on the biological dynamic monitoring data, multi-source monitoring information is integrated and a comprehensive environmental assessment model is constructed to obtain the multi-source data integration results.

[0010] Based on the results of the multi-source data integration, abnormal fish behavior and the accumulation of harmful substances in the water are identified, and abnormal symptom identification results are obtained.

[0011] Based on the abnormal symptom identification results, a water quality anomaly analysis model is trained and risk warning information is output.

[0012] Based on the aforementioned risk warning information, the dynamic changes in the environment and biological state are continuously monitored to obtain dynamic change trend data;

[0013] Based on the dynamic trend data, update the comprehensive environmental assessment model and generate optimized multi-factor integrated monitoring data;

[0014] Based on the optimized multi-element integrated monitoring data, the key monitoring indicators are prioritized to obtain the key element monitoring framework data.

[0015] Preferably, the process of obtaining the raw dataset by deploying sensors to collect environmental parameters and biological population data of the marine ranch includes:

[0016] Real-time data on water temperature, salinity, dissolved oxygen, pH value, and fish density are obtained through sensors.

[0017] If the data is missing or abnormal, the missing values ​​are interpolated to fill in the missing values ​​and obtain the complete dataset.

[0018] The complete dataset is subjected to dimensionality reduction and the main features are extracted to obtain the feature dataset;

[0019] Based on the feature dataset, the marine ranch area is classified into states to obtain the regional state classification results.

[0020] If the regional status classification result indicates an anomaly, then analyze the data change trend of the anomaly region to obtain the change trend data;

[0021] Based on the aforementioned trend data, the dynamic distribution characteristics of water quality parameters and fish density are calculated to obtain a distribution characteristic dataset;

[0022] Based on the aforementioned distribution feature dataset, a spatial interpolation model is constructed to predict the state of unmonitored areas, thereby obtaining global state prediction data.

[0023] Preferably, the process of classifying and identifying plankton species and water pollutants based on the original dataset to obtain environmental element classification results includes:

[0024] Spectral features of planktonic organisms were extracted using spectral analysis techniques to obtain the first feature set;

[0025] Chemical sensors were used to detect the concentration of pollutants in the water, resulting in a first concentration set.

[0026] The first feature set and the first concentration set are fused to obtain a comprehensive feature set;

[0027] The comprehensive feature set was classified and trained using the support vector machine algorithm to obtain a planktonic species classification model.

[0028] Based on the first classification result output by the planktonic species classification model, the environmental element classification is determined by the decision tree algorithm to obtain the final classification result;

[0029] If the confidence level of the final classification result is lower than a preset threshold, then a second feature extraction is performed on the comprehensive feature set and the final classification result is updated.

[0030] Preferably, the process of tracking the activity trajectory and distribution of key fish populations based on the environmental element classification results to obtain biological dynamic monitoring data includes:

[0031] Sonar signal data of key fish populations were obtained using sonar technology to determine their preliminary distribution locations;

[0032] Key fish species are tagged using electronic tag technology to obtain real-time location and movement data, thus obtaining activity trajectory information;

[0033] If the deviation between the sonar signal data and the electronic tag location data exceeds a preset threshold, the sonar signal data and the electronic tag location data are fused and corrected to obtain the corrected population distribution and trajectory data.

[0034] Based on the corrected data, a time series model of population dynamics is constructed to obtain the dynamic trend.

[0035] Analyze the correlation between the classification of environmental elements and the aforementioned dynamic change trends to identify key environmental impact factors;

[0036] If the key environmental impact factors exceed the preset range, the sonar scanning frequency and the electronic tag data acquisition interval are adjusted to obtain target monitoring data;

[0037] Based on dynamic trends and key environmental influencing factors, predictive data on population dynamics changes in future periods are generated.

[0038] Preferably, the process of integrating multi-source monitoring information and constructing a comprehensive environmental assessment model based on the biological dynamic monitoring data to obtain the multi-source data integration results includes:

[0039] Data on water temperature, salinity, ocean currents, and fish distribution were acquired, and preprocessed to remove noise, resulting in standardized multi-source data.

[0040] The standardized multi-source data is fused using the Kalman filter algorithm to obtain fused data;

[0041] Based on the fused data, the spatial distribution characteristics and temporal variation trends of environmental parameters are calculated to obtain the environmental parameter distribution.

[0042] If the fluctuation of the environmental parameter distribution exceeds a preset threshold, the support vector machine algorithm is used to classify the fish distribution and determine the fish gathering area.

[0043] Based on the fish aggregation area and the distribution of environmental parameters, a Bayesian network model is used to predict the trend of the influence of the environment on the fish distribution, and the environmental impact prediction results are obtained.

[0044] Based on the environmental impact prediction results, calculate the comprehensive environmental assessment indicators and generate a comprehensive environmental assessment model;

[0045] Based on the comprehensive environmental assessment model, the environmental status level of the marine ranch is determined, and the environmental status assessment results are obtained.

[0046] Preferably, the process of identifying abnormal fish behavior and the accumulation of harmful substances in the water based on the multi-source data integration results, and obtaining abnormal symptom identification results, includes:

[0047] Acquire fish behavior videos and aquatic environment data, integrate and process the data to obtain a standardized dataset;

[0048] Image recognition technology was used to analyze fish behavior patterns in video data to identify indicators of abnormal behavior.

[0049] Biomarker detection technology is used to analyze water samples to identify the concentration of harmful substances and obtain the distribution characteristics of the substances;

[0050] If the abnormal behavior index exceeds a preset threshold, then a symptom correlation analysis is performed in conjunction with the substance distribution characteristics to determine the type of abnormal symptom.

[0051] Based on the types of abnormal symptoms, the results of behavioral pattern analysis are combined with environmental parameter monitoring data to generate a comprehensive abnormality assessment result.

[0052] The support vector machine algorithm was used to classify the correlation between abnormal fish behavior and harmful substances, and the classification model output was obtained.

[0053] Based on the output of the classification model, update the environmental parameter monitoring data and determine the final distribution of abnormal symptoms.

[0054] Preferably, the process of training a water quality anomaly analysis model and outputting risk warning information based on the abnormal symptom identification results includes:

[0055] Obtain abnormal symptom data and water quality abnormality data, perform data cleaning and preprocessing, and obtain a standard dataset;

[0056] The key features of the standard dataset are extracted using the principal component analysis algorithm to obtain the key feature set;

[0057] The key feature set was trained using the random forest algorithm to obtain a red tide occurrence probability prediction model.

[0058] If the probability value output by the red tide occurrence probability prediction model exceeds the preset threshold, a logistic regression algorithm is used for secondary verification to obtain the verified probability value.

[0059] Based on the verified probability values, a classification algorithm is used to determine the red tide risk level, and the risk level results are obtained.

[0060] Based on the risk level results and preset warning rules, a risk warning result is generated;

[0061] Data visualization technology is used to dynamically display the risk warning results and output warning information.

[0062] Preferably, the process of continuously monitoring the dynamic changes in the environment and biological state based on the risk warning information to obtain dynamic change trend data includes:

[0063] Based on the acquisition of environmental parameters and biological status data by sensors, if the environmental parameters exceed the preset threshold, an anomaly detection algorithm is used to identify potential risks and generate an abnormal event set.

[0064] Time series analysis algorithms are used to process real-time data streams to determine the changing trends of environmental and biological states.

[0065] Based on the aforementioned trends, a decision tree algorithm is used to classify environmental anomalies and biological behaviors to obtain risk level labels;

[0066] High-risk events are extracted based on the risk level labels, and dynamic trend data is generated by combining them with real-time data streams.

[0067] Based on the dynamic trend data, the parameters of the risk warning algorithm are updated to obtain an optimized warning model;

[0068] The optimized early warning model is used to continuously analyze the subsequently collected data and generate real-time risk warning signals.

[0069] Preferably, the process of updating the comprehensive environmental assessment model and generating optimized multi-factor integrated monitoring data based on the dynamic trend data includes:

[0070] Multi-source monitoring data of marine ranches were acquired and dynamic change characteristics were extracted. The Kalman filter algorithm was used to analyze the trend and obtain time series trend data.

[0071] A weighted data fusion algorithm is used to integrate multi-factor monitoring data to obtain integrated environmental data.

[0072] If the integrated environmental data meets the preset threshold conditions, the parameters of the integrated environmental assessment model are updated using the random forest algorithm.

[0073] Based on the updated model parameters, preliminary monitoring results of multi-element integrated monitoring data for marine ranches are generated.

[0074] Spatial interpolation processing is performed on the preliminary monitoring results, and the data distribution is optimized using the Kriging interpolation method to obtain the final integrated monitoring results;

[0075] Environmental assessment indicators are extracted from the final integrated monitoring results to generate comprehensive environmental status data for the marine ranch.

[0076] Preferably, the process of prioritizing key monitoring indicators based on the optimized multi-factor integrated monitoring data to obtain the key element monitoring framework data includes:

[0077] Environmental parameter data is collected through a sensor network and stored as an initial dataset;

[0078] The initial dataset is cleaned in real time using data stream processing technology to obtain a cleaned dataset.

[0079] If the data points in the cleaned dataset deviate from the preset threshold, outliers are removed to obtain an optimized dataset.

[0080] Principal component analysis algorithm is used to extract key indicators from the optimized dataset to obtain an indicator feature set;

[0081] The weights of each indicator in the feature set are calculated using the entropy method to obtain the indicator weight allocation results;

[0082] Based on the weight allocation results of the indicators, the key indicators are prioritized to obtain a priority sequence;

[0083] Based on the priority sequence and combined with the dynamic monitoring and adjustment strategy, a monitoring framework data for key elements of marine ranching is generated.

[0084] Compared with the prior art, the present invention has the following advantages and technical effects:

[0085] This invention addresses the operational challenges of predicting water quality deterioration, abnormal biological populations, and red tide risks in marine ranching. It utilizes high-sensitivity sensors to collect real-time multi-dimensional data on water temperature, salinity, dissolved oxygen, pH, and fish density. Combining spectral analysis and chemical sensor technology, it classifies plankton and detects pollutants, accurately identifying environmental factors. Subsequently, sonar and electronic tags are used to track fish activity, and a comprehensive environmental assessment model is generated by integrating hydrological and biological data. Abnormal behavior and harmful substance accumulation are identified through image recognition and biomarker detection. Machine learning algorithms analyze water quality anomalies, generating a red tide probability prediction model. This invention continuously updates dynamic trends, optimizes the monitoring model, prioritizes key indicators, and forms a multi-factor integrated monitoring framework. This invention significantly improves the real-time monitoring capabilities of the marine ranching environment and biological status, reduces ecological risks such as red tides, and ensures the sustainable development of marine ranching. Attached Figure Description

[0086] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0087] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0088] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0089] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0090] like Figure 1 As shown, this embodiment provides a method for comprehensive monitoring of marine ranches, including:

[0091] By deploying sensors to collect environmental parameters and biological population data of the marine ranch, a raw dataset is obtained;

[0092] Based on the original dataset, the species of planktonic organisms and water pollutants are classified and identified to obtain the classification results of environmental elements.

[0093] Based on the classification results of environmental elements, the activity trajectory and distribution of key fish populations are tracked to obtain biological dynamic monitoring data;

[0094] Based on biological dynamic monitoring data, multi-source monitoring information is integrated and a comprehensive environmental assessment model is constructed to obtain the multi-source data integration results.

[0095] Based on the results of multi-source data integration, abnormal fish behavior and the accumulation of harmful substances in water bodies are identified, and abnormal symptom identification results are obtained.

[0096] Based on the results of abnormal symptom identification, a water quality anomaly analysis model is trained and risk warning information is output.

[0097] Based on risk warning information, we continuously monitor the dynamic changes in the environment and biological status to obtain dynamic trend data.

[0098] Based on dynamic trend data, update the comprehensive environmental assessment model and generate optimized multi-factor integrated monitoring data;

[0099] Based on the optimized multi-element integrated monitoring data, the key monitoring indicators are prioritized to obtain the key element monitoring framework data.

[0100] Furthermore, the process of obtaining the raw dataset by deploying sensors to collect environmental parameters and biological population data of the marine ranch includes:

[0101] Real-time data on water temperature, salinity, dissolved oxygen, pH value, and fish density are obtained through sensors.

[0102] If the data is missing or abnormal, the missing values ​​are interpolated to fill in the missing values ​​and obtain the complete dataset.

[0103] The complete dataset is subjected to dimensionality reduction and the main features are extracted to obtain the feature dataset.

[0104] Based on the feature dataset, the marine ranch area is classified into states to obtain the state classification results. If the state classification results indicate anomalies, the data change trends of the abnormal areas are analyzed to obtain the change trend data.

[0105] Based on the trend data, the dynamic distribution characteristics of water quality parameters and fish density are calculated to obtain the distribution characteristic dataset;

[0106] Based on the distributed feature dataset, a spatial interpolation model is constructed to predict the state of unmonitored areas, thus obtaining global state prediction data.

[0107] Furthermore, this embodiment uses high-sensitivity sensors to collect data on water temperature, salinity, dissolved oxygen, pH, and fish density, generating a real-time multidimensional dataset. If the collected data is missing or abnormal, a linear interpolation algorithm is used to fill in the missing values, resulting in a complete dataset. Principal component analysis is then used to reduce the dimensionality of the complete dataset, extracting key environmental and biological features to generate a feature dataset. Based on the feature dataset, K-means clustering is applied to classify the marine ranch area into environmental and biological states, determining the state classification results. If the state classification results show abnormal areas, time series analysis is used to detect the data change trends in these abnormal areas, obtaining trend data. Based on the trend data, the dynamic distribution characteristics of water quality parameters and fish density in each area are calculated, generating a distribution feature dataset. Using the distribution feature dataset, a spatial interpolation model is constructed to predict the state of unmonitored areas of the marine ranch, obtaining overall state prediction data.

[0108] For example, when collecting data from a marine ranch using high-sensitivity sensors, suppose water temperature, salinity, dissolved oxygen, pH, and fish density are recorded hourly in a certain area. Water temperature might be 20-25℃, salinity 30-35‰, dissolved oxygen 5-8 mg / L, pH 7.8-8.3, and fish density expressed as the number of fish per cubic meter. The sensor network covers 100 monitoring points, but some points may have missing data due to equipment malfunctions. For instance, if a point has not recorded dissolved oxygen for three consecutive hours, the missing values ​​are filled using a linear interpolation algorithm.

[0109] For example, if the dissolved oxygen level at a certain point is 6 mg / L one hour before and 6.5 mg / L the next hour, the missing hour can be interpolated to 6.25 mg / L. This method is simple and efficient, and can quickly generate a complete dataset, ensuring the continuity of subsequent analyses.

[0110] In one possible implementation, this embodiment uses Principal Component Analysis (PCA) to reduce the dimensionality of the complete dataset. Assuming the dataset contains five variables, PCA can extract the first two principal components, retaining more than 80% of the data variance. For example, water temperature and dissolved oxygen might be identified as key environmental features, and fish density as a key biological feature, generating a low-dimensional feature dataset. This reduces computational complexity while preserving crucial information, facilitating subsequent clustering analysis.

[0111] Specifically, this embodiment uses the K-means clustering algorithm to divide the marine ranch into regions with different states. Assuming K=3, the clustering results may be divided into healthy, sub-healthy, and abnormal regions. For example, a healthy region has a stable water temperature of 22℃, dissolved oxygen above 6mg / L, and moderate fish density; an abnormal region may have dissolved oxygen below 5mg / L and abnormally high fish density. The clustering results intuitively reflect the differences in regional states, which is helpful for precise management.

[0112] For example, in the case of anomaly areas, this embodiment uses time series analysis to detect data trends. Suppose that dissolved oxygen in an anomaly area decreases from 6 mg / L to 4.5 mg / L within 5 days; trend analysis shows a continuous decline, which may be related to pollution or slowed water flow. This trend data provides a basis for analyzing the causes of environmental anomalies.

[0113] In one possible implementation, this embodiment calculates the dynamic distribution characteristics of water quality parameters and fish density based on trend data. For example, if dissolved oxygen decreases at a rate of 0.1 mg / L per day and fish density increases by 10% in an abnormal area, it indicates a potential eutrophication problem. The distribution characteristic dataset can quantify these changes, providing data support for environmental regulation.

[0114] Specifically, spatial interpolation models such as kriging can predict the state of unmonitored areas. Assuming data from 100 monitoring points are known, the model can estimate the water temperature and fish density in the uncovered areas.

[0115] For example, a predicted water temperature of 23℃ and moderate fish density in an unmonitored area indicate that the area is in good health. Comprehensive health prediction data provides a reference for the overall management of marine ranches and optimizes resource allocation.

[0116] Understandably, the methods described above form a complete technology chain, with each step interconnected from data collection to prediction. Missing value imputation ensures data integrity, dimensionality reduction and clustering improve analytical efficiency, while time series and spatial interpolation enhance anomaly detection and global prediction capabilities. These technologies significantly improve the accuracy and management efficiency of marine ranching environmental monitoring.

[0117] Furthermore, based on the original dataset, the process of classifying and identifying plankton species and water pollutants to obtain environmental element classification results includes:

[0118] Spectral features of planktonic organisms were extracted using spectral analysis techniques to obtain the first feature set;

[0119] Chemical sensors were used to detect the concentration of pollutants in the water, resulting in a first concentration set.

[0120] The first feature set and the first concentration set are fused to obtain a comprehensive feature set;

[0121] A support vector machine algorithm was used to classify and train the comprehensive feature set to obtain a planktonic species classification model.

[0122] Based on the first classification result output by the planktonic species classification model, the decision tree algorithm is used to determine the classification of environmental elements and obtain the final classification result.

[0123] If the confidence level of the final classification result is lower than the preset threshold, then a second feature extraction is performed on the comprehensive feature set and the final classification result is updated.

[0124] Furthermore, this embodiment acquires multidimensional environmental data and biological datasets, and uses spectral analysis technology to extract spectral features of planktonic species to obtain a first feature set. Water pollutant concentrations are detected using chemical sensor data, and statistical analysis methods are used to generate pollutant concentration distributions to obtain a first concentration set. If the feature dimensions of the first feature set and the first concentration set are consistent, data fusion processing is used to integrate the spectral features and pollutant concentration data to obtain a comprehensive feature set; if they are inconsistent, features are supplemented for the dataset with lower dimensions to obtain a comprehensive feature set. For the comprehensive feature set, a support vector machine algorithm is used for classification training to obtain a planktonic species classification model. The comprehensive feature set is classified using the planktonic species classification model to obtain a first classification result. Based on the first classification result and environmental data analysis, a decision tree algorithm is used to determine the classification of environmental elements to obtain a final classification result. If the confidence level of the final classification result is lower than a preset threshold, a second spectral feature extraction is performed on the comprehensive feature set to obtain a second feature set, and the classification training and environmental element classification are repeated to obtain an updated final classification result.

[0125] For example, in the field of water quality and biological monitoring in marine ranches, this embodiment utilizes spectral analysis technology to extract the spectral characteristics of planktonic species. Specifically, a high-resolution spectrometer is used to capture the reflectance spectral signals of planktonic organisms. The spectrometer scans in the 400-700 nm wavelength range to obtain the unique spectral curves of planktonic organisms such as diatoms or dinoflagellates, forming a first feature set.

[0126] Specifically, diatoms exhibit a strong absorption peak at 550 nm, while dinoflagellates show significant reflectance at 620 nm. This method distinguishes different planktonic species based on spectral characteristics, providing fundamental data for subsequent classification.

[0127] In one possible implementation, this embodiment detects the concentration of water pollutants, such as ammonia nitrogen and heavy metal ions, using chemical sensor data. Sensors are deployed in key areas of the marine ranch, collecting data hourly, for example, at an ammonia nitrogen concentration of 0.5 mg / L and a lead concentration of 0.02 mg / L. Statistical analysis methods, such as spatial interpolation, are used to generate a pollutant concentration distribution, forming a first concentration set. This distribution map can visually show areas of high pollutant concentration, such as areas near the shore where concentrations may be higher than in the open sea.

[0128] It should be noted that if the dimensions of the first feature set and the first concentration set are inconsistent—for example, the spectral feature set contains 10 feature values ​​while the concentration set only contains 5—then feature completion is required for the concentration set. Specifically, this can be achieved by adding environmental variables such as water temperature and salinity to complete the dimensions and form a comprehensive feature set.

[0129] For example, water temperature of 25℃ and salinity of 30‰ are used as supplementary features to ensure consistency in data dimensions.

[0130] For example, in this embodiment, a plankton species classification model is constructed by training a comprehensive feature set using a support vector machine algorithm. During training, the comprehensive feature set is divided into a training set and a test set in a ratio of 8:2. The model can distinguish between diatoms and dinoflagellates, with a classification accuracy of over 90%. The classification results show that in a certain area, diatoms account for 60%, dinoflagellates account for 30%, and other species account for 10%.

[0131] In one possible implementation, a decision tree algorithm is used in conjunction with environmental data such as pH 7.8 and dissolved oxygen 6 mg / L to analyze the impact of environmental factors on the distribution of plankton and generate the final classification result.

[0132] For example, areas with lower pH levels may be more suitable for dinoflagellate growth. The classification results show that the environmental factors of a certain area were categorized as "suitable for dinoflagellates".

[0133] Understandably, if the confidence level of the final classification result is below 0.8, a second spectral feature extraction is required. A higher resolution spectrometer is used, focusing on the 450-650nm band, to extract more refined spectral features, forming a second feature set.

[0134] For example, the new feature includes the secondary absorption peak of dinoflagellates at 600 nm. After repeated classification training, the model confidence improved to 0.85, and the classification results became more reliable.

[0135] For example, the combined application of the above methods can achieve precise monitoring of plankton and environmental conditions in marine ranches. The combination of spectral analysis and chemical sensors compensates for the limitations of single data sources; feature completion and data fusion improve data integrity; the combination of support vector machines and decision tree algorithms enhances classification accuracy; and secondary feature extraction optimizes low-confidence results. These methods collectively support a comprehensive analysis of the marine ranch environment and biological conditions, providing data support for subsequent management.

[0136] Furthermore, based on the classification results of environmental elements, the process of tracking the activity trajectories and distribution of key fish populations to obtain biological dynamic monitoring data includes:

[0137] Sonar signal data of key fish populations were obtained using sonar technology to determine their preliminary distribution locations;

[0138] Key fish species are tagged using electronic tag technology to obtain real-time location and movement data, thus obtaining activity trajectory information;

[0139] If the deviation between the sonar signal data and the electronic tag location data exceeds a preset threshold, the sonar signal data and the electronic tag location data are fused and corrected to obtain the corrected population distribution and trajectory data.

[0140] Based on the corrected data, a time series model of population dynamics is constructed to obtain the dynamic trend.

[0141] Analyze the correlation between environmental element classification and dynamic change trends to identify key environmental impact factors;

[0142] If the key environmental impact factors exceed the preset range, the sonar scanning frequency and the electronic tag data acquisition interval are adjusted to obtain target monitoring data.

[0143] Based on dynamic trends and key environmental influencing factors, predictive data on population dynamics changes in future periods are generated.

[0144] Furthermore, this embodiment uses sonar technology to scan the aquatic environment, acquire sonar signal data of key fish populations, and determine the preliminary distribution location of the populations. Electronic tagging technology is used to mark individual key fish species, and real-time location and movement data are obtained from the tagged individuals to obtain activity trajectory information. If the deviation between the sonar signal data and the electronic tag location data exceeds a preset threshold, the two types of data are fused using a Kalman filter algorithm to obtain corrected population distribution and trajectory data. Based on the corrected population distribution and trajectory data, a time series model of population dynamic changes is constructed to obtain the dynamic change trend of the population. The correlation between environmental element classification and dynamic change trends is analyzed using a random forest algorithm to determine key environmental impact factors. If the values ​​of key environmental impact factors exceed a preset range, the sonar scanning frequency and electronic tag data acquisition interval are adjusted to obtain more accurate monitoring data. Based on the dynamic change trend and key environmental impact factors, a prediction model of population distribution and activity trajectory is generated to obtain predicted data of population dynamic changes in future periods.

[0145] For example, in this embodiment, when scanning the aquatic environment using sonar technology, a multibeam sonar system can be used to acquire high-resolution underwater sonar signal data. The sonar equipment emits sound waves at a fixed frequency, receives echo signals from fish populations, analyzes the echo intensity and time difference, and generates a preliminary distribution map of the population. Assuming that in a certain sea area, the sonar scan covers 100 square kilometers of water, and detects that a certain fish population is mainly concentrated in an area with a depth of 5-10 meters, the preliminary distribution shows that the core area with higher population density accounts for approximately 20% of the total area. This method can effectively capture the spatial distribution characteristics of the population, providing a basis for subsequent analysis.

[0146] In one possible implementation, the electronic tag technology involved in this embodiment is achieved by implanting small RFID tags on key individual fish. The tag sends a positioning signal every 10 seconds to record the fish's real-time location and movement path.

[0147] For example, after tagging 10 individual fish, monitoring data over 72 consecutive hours showed that a certain fish species exhibited a clear diurnal migration pattern, tending to move closer to the water's edge at night. This activity trajectory information provides a dynamic perspective for analyzing population behavior patterns.

[0148] It should be noted that if the sonar signal data and the electronic tag location data deviate significantly—for example, the sonar positioning accuracy is ±5 meters while the electronic tag positioning accuracy is ±1 meter—then the data is fused using a Kalman filter algorithm. Kalman filtering models the noise characteristics of both types of data to generate corrected population distribution and trajectory data. For example, after fusion, it was found that the core distribution area of ​​a certain fish population was 10% smaller than the results of a single sonar scan, improving the accuracy of the distribution data. This fusion method effectively reduces positioning errors and improves data reliability. For instance, in this embodiment, when constructing a time-series model of population dynamics based on the corrected data, the ARIMA model is used to analyze the daily variation trend of population density. Assuming that monitoring data shows that the population density of a certain fish species exhibits periodic fluctuations over the past 30 days, the model predicts that the population density will increase by 5% in the next 7 days. This time-series analysis helps to understand the dynamic changes in the population and provides a basis for conservation strategies.

[0149] In one possible implementation, the correlation between environmental factors (such as water temperature and dissolved oxygen) and population dynamics is analyzed using a random forest algorithm.

[0150] For example, analysis revealed that fish populations are most active when water temperatures are between 20-25℃, and population density decreases significantly when dissolved oxygen levels fall below 5 mg / L. The identification of these key environmental impact factors provides a targeted basis for environmental management. If the water temperature exceeds 25℃, the sonar scanning frequency can be adjusted to once per hour, and the data collection interval of the electronic tags can be shortened to 5 seconds to obtain more accurate monitoring data.

[0151] For example, based on dynamic trends and key environmental influencing factors, a predictive model is constructed, and a long short-term memory neural network (LSTM) is used to generate predictions of population distribution and activity trajectories for the next 30 days.

[0152] For example, the prediction results show that a certain fish population will migrate to shallower waters in the future, expanding its distribution area by 15%. This predictive model provides a forward-looking reference for aquatic ecological management and helps to formulate scientific population protection measures.

[0153] Furthermore, the process of integrating multi-source monitoring information and constructing a comprehensive environmental assessment model based on biological dynamic monitoring data to obtain the multi-source data integration results includes:

[0154] Data on water temperature, salinity, ocean currents, and fish distribution were acquired, and preprocessed to remove noise, resulting in standardized multi-source data.

[0155] The Kalman filter algorithm is used to fuse standardized multi-source data to obtain fused data;

[0156] Based on the fused data, the spatial distribution characteristics and temporal variation trends of environmental parameters are calculated to obtain the distribution of environmental parameters.

[0157] If the fluctuation of environmental parameter distribution exceeds the preset threshold, the support vector machine algorithm is used to classify the fish distribution and determine the fish gathering area.

[0158] Based on the fish aggregation area and the distribution of environmental parameters, a Bayesian network model is used to predict the trend of the environment's influence on fish distribution, and the environmental impact prediction results are obtained.

[0159] Based on the environmental impact prediction results, calculate the comprehensive environmental assessment indicators and generate the comprehensive environmental assessment model;

[0160] Based on the comprehensive environmental assessment model, the environmental status level of the marine ranch was determined, and the environmental status assessment results were obtained.

[0161] Furthermore, this embodiment acquires water temperature, salinity, ocean current, and fish distribution data from marine ranching monitoring equipment. Noise and outliers are removed through preprocessing to obtain standardized multi-source data. A Kalman filter algorithm is used to fuse the standardized multi-source data, generating fused data containing water temperature, salinity, ocean current, and fish distribution. Based on the fused data, the spatial distribution characteristics and temporal trends of water temperature, salinity, and ocean current are calculated to obtain the environmental parameter distribution. If the fluctuation of the environmental parameter distribution exceeds a preset threshold, a support vector machine algorithm is used to classify the fish distribution and determine the fish aggregation areas. Based on the fish aggregation areas and environmental parameter distribution, a Bayesian network model is used to predict the impact trend of the marine ranching environment on fish distribution, obtaining the environmental impact prediction results. Based on the environmental impact prediction results, a comprehensive environmental assessment index for the marine ranching is calculated, generating a comprehensive environmental assessment model. Based on the comprehensive environmental assessment model, the environmental status level of the marine ranching is determined, obtaining the environmental status assessment results.

[0162] For example, in marine ranching monitoring, data on water temperature, salinity, ocean currents, and fish distribution are acquired in real time through sensor networks. Assume a marine ranch deploys multiple water temperature sensors, recording the water temperature every 10 minutes, with the data ranging from 18.5℃ to 22.3℃. Salinity sensors record salinity values ​​between 30‰ and 35‰, ocean current sensors record current velocities between 0.2 m / s and 0.5 m / s, and fish distribution data is obtained through sonar scanning, displaying fish density distribution. During preprocessing, noise is removed using a moving average method. For example, the water temperature data is averaged over a 5-minute window, and outliers exceeding three standard deviations are discarded; for instance, a recorded temperature of 26℃ is significantly higher and therefore removed. This preprocessing ensures smooth and reliable data.

[0163] Specifically, this embodiment also utilizes the Kalman filter algorithm to fuse multi-source data. Assuming there is a time discrepancy between the data from the water temperature sensor and the ocean current sensor, the Kalman filter generates fused data through prediction and update steps, combined with sensor accuracy weights.

[0164] For example, the combined water temperature in a certain area is 20.1℃, the salinity is 32.5‰, the ocean current velocity is 0.35m / s, and the fish density is 50 fish per cubic meter. This combined data more accurately reflects the environmental conditions.

[0165] In one possible implementation, gridded analysis is used to calculate the spatial distribution and temporal trends of environmental parameters. The marine ranch is divided into 100m × 100m grids, and the average values ​​and rates of change of water temperature, salinity, and ocean currents within each grid are calculated.

[0166] For example, the water temperature in one grid rose from 19.8℃ to 20.5℃ within 24 hours, indicating the influence of a warm current. Fish distribution showed that high-density areas were concentrated in grids with gentler currents and moderate salinity. This analysis reveals the spatial correlation between environmental parameters and fish distribution.

[0167] For example, this embodiment uses a support vector machine algorithm to classify fish aggregation areas. Assuming environmental parameter fluctuations exceed a threshold, such as a salinity change rate greater than 0.5‰ / hour, the fish distribution is divided into high-density, medium-density, and low-density areas. The training data includes fused data from the past week. The classification results show that high-density areas are mostly located in regions with water temperatures between 20°C and 21°C and ocean current velocities below 0.3 m / s. This classification helps to accurately locate fish aggregation hotspots.

[0168] Specifically, Bayesian network models predict the impact of the environment on fish population distribution. The model inputs include water temperature, salinity, and ocean current data, and the output is the probability of fish aggregation. For example, in a certain area with a water temperature of 21℃, salinity of 33‰, and an ocean current of 0.25 m / s, the probability of fish aggregation is 85%. Through training with multiple sets of data, the model can predict the fish distribution trend over the next 24 hours. This prediction supports the dynamic adjustment of ranch management strategies.

[0169] In one possible implementation, the integrated environmental assessment model calculates environmental indicators through weighted calculations.

[0170] For example, water temperature, salinity, and ocean currents are weighted at 0.4, 0.3, and 0.2 respectively, while fish population distribution accounts for 0.1, with the overall score ranging from 0 to 100. A score of 85 for a ranch indicates a good environmental condition. The final environmental condition is categorized into four levels: excellent, good, average, and poor. The assessment results guide ranch management optimization. This method, through multi-dimensional analysis, enhances the comprehensiveness and practicality of monitoring.

[0171] Furthermore, based on the results of multi-source data integration, the process of identifying abnormal fish behavior and the accumulation of harmful substances in the water to obtain abnormal symptom identification results includes:

[0172] Acquire fish behavior videos and aquatic environment data, integrate and process the data to obtain a standardized dataset;

[0173] Image recognition technology was used to analyze fish behavior patterns in video data to identify indicators of abnormal behavior.

[0174] Biomarker detection technology is used to analyze water samples to identify the concentration of harmful substances and obtain the distribution characteristics of the substances;

[0175] If the abnormal behavior indicators exceed the preset threshold, then a symptom correlation analysis is performed in conjunction with the material distribution characteristics to determine the type of abnormal symptoms.

[0176] Based on the type of abnormal symptoms, the results of behavioral pattern analysis are integrated with environmental parameter monitoring data to generate a comprehensive abnormality assessment result;

[0177] The support vector machine algorithm was used to classify the correlation between abnormal fish behavior and harmful substances, and the classification model output was obtained.

[0178] Based on the output of the classification model, update the environmental parameter monitoring data and determine the final distribution of abnormal symptoms.

[0179] Furthermore, this embodiment acquires fish behavior videos and aquatic environment data from multiple sensors, and uses data integration and processing technology to obtain a standardized dataset. Image recognition technology is used to process the video data in the standardized dataset to analyze fish behavior patterns and determine abnormal behavior indicators. Biomarker detection technology is used to analyze water samples in the standardized dataset to identify harmful substance concentrations and obtain substance distribution characteristics. If the abnormal behavior indicators exceed a preset threshold, symptom correlation analysis is performed based on the substance distribution characteristics to determine the type of abnormal symptoms. Based on the abnormal symptom types, the behavioral pattern analysis results and environmental parameter monitoring data are fused to generate a comprehensive anomaly assessment result. Using the comprehensive anomaly assessment result, a support vector machine algorithm is used to classify the correlation between abnormal fish behavior and harmful substances, obtaining a classification model output. Based on the classification model output and the real-time data acquisition results, the environmental parameter monitoring data is updated to determine the final distribution of abnormal symptoms.

[0180] For example, in scenarios involving acquiring fish behavior videos and aquatic environmental data from multiple sensors, a multimodal sensor network deployed within a marine ranch can be used. Sensors include high-definition underwater cameras, temperature, salinity, and depth gauges (TDS meters), and chemical analyzers, used to collect real-time data on fish behavior, water temperature, salinity, and dissolved oxygen. Assuming a marine ranch, cameras record 100 hours of fish activity video daily, water temperature sensors collect data every minute (range 15-25°C), and salinity sensors record data within the range of 30-35‰. Through data cleaning, blurred video frames due to insufficient light or outliers caused by ocean currents are removed, generating a standardized dataset containing timestamps and environmental parameters.

[0181] One possible implementation involves using convolutional neural networks to perform image recognition on video data from a standardized dataset, analyzing fish behavior patterns. The core objective is to extract fish swimming trajectories, group aggregation, and the frequency of abnormal behaviors.

[0182] For example, under normal circumstances, fish swim at a speed of 0.5-1 m / s with stable spacing between groups. If a sudden drop in swimming speed to 0.2 m / s or disorderly aggregation is detected, it is marked as a behavioral abnormality indicator, with thresholds set at a 30% decrease in speed or an abnormal increase in aggregation of 20%. By training the model, abnormal behaviors of fish caused by environmental stress can be identified, such as rapid tail wagging or dispersal of the group.

[0183] Specifically, this embodiment analyzes the concentration of harmful substances in water samples using biomarker detection technology. For example, after collecting water samples, ammonia nitrogen and heavy metal concentrations are detected using a mass spectrometer. Ammonia nitrogen concentrations exceeding 0.5 mg / L or lead concentrations exceeding 0.01 mg / L are considered to be exceeding the harmful substance limit. Combined with spatial gridded sampling, a distribution characteristic map of harmful substances is plotted. It is found that the ammonia nitrogen concentration near the sewage outlet is as high as 0.7 mg / L, indicating a significant impact from the pollution source. This distribution characteristic provides a basis for subsequent symptom correlation analysis.

[0184] In one embodiment, if a behavioral anomaly indicator exceeds a threshold, symptom correlation analysis is performed in conjunction with the distribution characteristics of harmful substances. For example, a decrease in fish swimming speed is correlated with excessive ammonia nitrogen concentration, indicating possible poisoning symptoms. The analysis process, by comparing historical data, confirms that the abnormal symptom type is an acute stress response, rather than behavioral abnormalities caused by disease or predation stress. This correlation analysis helps to accurately pinpoint the root cause of the problem.

[0185] Preferably, behavioral pattern analysis and environmental parameter monitoring data are integrated to generate a comprehensive anomaly assessment result.

[0186] For example, this embodiment uses a weighted scoring model, allocating 40% to abnormal behavior indicators, 40% to harmful substance concentration, and 20% to environmental parameter fluctuations to calculate a comprehensive anomaly index. If the index exceeds 80 points, it is classified as a high-risk state. Based on this, a support vector machine algorithm is used to classify the correlation between abnormal fish behavior and harmful substances.

[0187] For example, the model output shows an 85% correlation between excessive ammonia nitrogen concentration and scattered fish behavior, providing a basis for environmental management decisions. For instance, by combining real-time data with updated environmental parameter monitoring results, assuming newly collected data indicates that ammonia nitrogen concentration has decreased to 0.3 mg / L and fish behavior has returned to normal, the abnormal symptom distribution map is updated to confirm that the pollution impact has weakened. This dynamic update mechanism ensures the timeliness of the assessment results, providing real-time reference for marine ranching management.

[0188] Furthermore, based on the results of abnormal symptom identification, the process of training a water quality anomaly analysis model and outputting risk warning information includes:

[0189] Obtain abnormal symptom data and water quality abnormality data, perform data cleaning and preprocessing, and obtain a standard dataset;

[0190] Principal component analysis (PCA) is used to extract key features from a standard dataset, resulting in a key feature set.

[0191] A red tide occurrence probability prediction model was obtained by training the key feature set using the random forest algorithm.

[0192] If the probability value output by the red tide occurrence probability prediction model exceeds the preset threshold, a logistic regression algorithm is used for secondary verification to obtain the verified probability value.

[0193] Based on the verified probability values, a classification algorithm is used to divide the red tide risk level and obtain the risk level results;

[0194] Based on the risk level results and preset early warning rules, generate risk warning results;

[0195] Data visualization technology is used to dynamically display the risk warning results and output warning information.

[0196] Furthermore, this embodiment acquires abnormal symptom data and water quality anomaly data from monitoring equipment, preprocesses the data using data cleaning techniques to obtain a standard dataset. Based on the standard dataset, principal component analysis (PCA) is used to extract data features, resulting in a key feature set. Using the key feature set, a random forest algorithm is employed to train the model, obtaining a red tide occurrence probability prediction model. If the probability value output by the prediction model exceeds a preset threshold, a logistic regression algorithm is used to perform a secondary verification of the probability value, yielding a verified probability value. Based on the verified probability value, a classification algorithm is used to classify the red tide risk level, resulting in a risk level result. Using the risk level result and preset warning rules, a risk warning result is generated. Based on the risk warning result, data visualization technology is used to generate dynamic trends and output warning information.

[0197] For example, monitoring equipment can acquire data on abnormal fish symptoms, such as slowed swimming speed and abnormal schooling behavior, as well as data on abnormal water quality, such as dissolved oxygen content and ammonia nitrogen concentration. This data is preprocessed using data cleaning techniques.

[0198] In one possible implementation, data cleaning includes removing noisy data, filling in missing values, and standardizing the data format.

[0199] For example, suppose that in a certain body of water, fish swimming speed is monitored to have decreased to 50% of normal, while dissolved oxygen levels have dropped to 4 mg / L, below the normal range. Through data cleaning, outliers are removed, and the data is standardized to a uniform unit, generating a standard dataset.

[0200] Specifically, this embodiment utilizes principal component analysis (PCA) to extract data features. Assuming the dataset contains variables such as fish swimming speed, school density, water temperature, and pH value, PCA can reduce these variables to several key features, such as "behavioral activity" and "comprehensive water quality indicators."

[0201] For example, the analysis results may show that the decline in behavioral activity is highly correlated with the deterioration of comprehensive water quality indicators, contributing more than 80% and constituting a key feature set.

[0202] In one embodiment, this embodiment also relates to a random forest algorithm for model training based on a key feature set. Assume the training data includes 1000 samples, each containing behavioral activity and water quality indicators, labeled as "red tide occurred" or "no red tide". The random forest predicts the probability of red tide occurrence through voting among multiple decision trees.

[0203] For example, a predicted probability of 0.75 indicates a high risk of red tide. This method improves prediction robustness through multi-tree ensemble. For instance, if the predicted probability exceeds a preset threshold of 0.7, a secondary verification is performed using a logistic regression algorithm. Logistic regression can incorporate historical data, such as water quality trends over the past week, to reassess the probability value. For example, if the initial probability of 0.75 is adjusted to 0.72 after verification, it is still higher than the threshold, confirming the existence of risk. This secondary verification reduces the false alarm rate and improves prediction reliability.

[0204] Specifically, the classification algorithm categorizes red tide risk levels based on verified probability values. Assuming a probability of 0.5-0.7 is low risk, 0.7-0.85 is medium risk, and above 0.85 is high risk. For example, a probability of 0.72 is classified as medium risk. This clear threshold classification clearly presents the risk level, facilitating subsequent decision-making.

[0205] In one possible implementation, risk warning results are generated in conjunction with preset rules. For example, the rules stipulate that a yellow warning should be issued for medium-risk areas and a red warning for high-risk areas. Suppose a body of water is classified as medium-risk, the system generates a yellow warning to remind managers to strengthen monitoring. This approach ensures that warnings are timely and intuitive.

[0206] For example, data visualization technology generates dynamic trend charts showing changes in risk levels and environmental parameters. Suppose the chart shows a continuous decline in dissolved oxygen over the past 24 hours, with the red tide probability rising from 0.6 to 0.72, visually illustrating the increasing risk trend. Managers can then take measures, such as increasing the oxygen supply to the water. This visualization method makes complex data easy to understand and supports rapid decision-making.

[0207] Furthermore, based on risk warning information, the process of continuously monitoring the dynamic changes in the environment and biological states to obtain dynamic trend data includes:

[0208] Based on the acquisition of environmental parameters and biological status data by sensors, if the environmental parameters exceed the preset threshold, an anomaly detection algorithm is used to identify potential risks and generate an abnormal event set.

[0209] Time series analysis algorithms are used to process real-time data streams to determine the changing trends of environmental and biological states.

[0210] Based on the changing trends, a decision tree algorithm is used to classify environmental anomalies and biological behaviors to obtain risk level labels;

[0211] High-risk events are extracted based on risk level labels, and dynamic trend data is generated by combining real-time data streams.

[0212] Based on the dynamic trend data, the parameters of the risk warning algorithm are updated to obtain the optimized warning model;

[0213] The optimized early warning model is used to continuously analyze the subsequently collected data and generate real-time risk warning signals.

[0214] Furthermore, this embodiment acquires environmental parameter data and biological status data from marine ranch sensors. If environmental parameters exceed preset thresholds, an anomaly detection algorithm identifies potential risks and generates an anomaly event set. Based on the anomaly event set, a time series analysis algorithm is used to process the real-time data stream to determine the changing trends of the environment and biological status. Using these trends, a decision tree algorithm is employed to classify environmental anomalies and biological behaviors, obtaining risk level labels. High-risk events are extracted from the risk level labels and, combined with the real-time data stream, dynamic trend data is generated. Based on the dynamic trend data, the parameters of the risk warning algorithm are updated to obtain an optimized warning model. Using the optimized warning model, subsequent environmental parameter and biological status data are continuously analyzed to generate real-time risk warning signals.

[0215] Specifically, obtaining environmental parameter data and biological status data from marine ranch sensors involves a variety of monitoring devices, such as temperature sensors, dissolved oxygen sensors, water flow velocity sensors, and biological behavior monitors.

[0216] For example, the temperature sensor records the seawater temperature hourly. Suppose that a sample of 28.5°C is collected, which is outside the normal range of 25-27°C; the dissolved oxygen data is 4.8 mg / L, which is below the safe threshold of 5.0 mg / L; and the fish activity frequency decreases by 20%, indicating an anomaly. If environmental parameters exceed the preset threshold, the anomaly detection algorithm is activated.

[0217] Specifically, this embodiment uses a statistical anomaly detection method to calculate the standard deviation of the data and mark data that deviates from the mean by two times the standard deviation as anomalies.

[0218] For example, if dissolved oxygen levels are below 5.0 mg / L for three consecutive days, an anomaly event set is generated, recording the time, parameter values, and anomaly type. This anomaly event set provides a basis for subsequent analysis and avoids misjudgments. Time series analysis algorithms process the real-time data stream to determine trends.

[0219] For example, analyzing temperature and dissolved oxygen data over 7 days using the moving average method revealed a continuous increase in temperature of 0.2℃ / day, a decrease in dissolved oxygen of 0.1 mg / L / day, and a significant downward trend in fish activity frequency. These trends suggest that environmental degradation may affect the health of organisms, providing a basis for early warning. The decision tree algorithm classifies environmental anomalies and biological behaviors, generating risk level labels.

[0220] In one embodiment, the decision tree categorizes risks into high, medium, and low based on temperature, dissolved oxygen, and fish activity frequency. For example, a temperature >28°C, dissolved oxygen <5.0 mg / L, and a fish activity frequency decrease of >15% are marked as high risk. The classification results are clear, facilitating rapid response. High-risk events are extracted from the risk level labels and combined with real-time data streams to generate dynamic trend data.

[0221] For example, high-risk events may be concentrated in a certain area, with data showing a continuous rise in temperature and a rapid decline in dissolved oxygen. The dynamic trend is displayed through a line graph, facilitating a visual assessment of risk evolution. The risk warning algorithm parameters are updated, and the warning model is optimized. For instance, based on the frequency and intensity of high-risk events, the warning threshold is adjusted, lowering the dissolved oxygen threshold from 5.0 mg / L to 4.9 mg / L to improve sensitivity. The optimized model is more adaptable to dynamic environmental changes. Through the optimized warning model, new data is continuously analyzed to generate real-time risk warning signals.

[0222] For example, when new data indicates that the temperature has risen to 29°C, the model immediately issues a high-risk warning, prompting managers to take cooling or oxygenation measures. Real-time signals ensure timely intervention and reduce losses.

[0223] Furthermore, the process of updating the comprehensive environmental assessment model and generating optimized multi-factor integrated monitoring data based on dynamic trend data includes:

[0224] Multi-source monitoring data of marine ranches were acquired and dynamic change characteristics were extracted. The Kalman filter algorithm was used to analyze the trend and obtain time series trend data.

[0225] A weighted data fusion algorithm is used to integrate multi-factor monitoring data to obtain integrated environmental data.

[0226] If the integrated environmental data meets the preset threshold conditions, the parameters of the integrated environmental assessment model are updated using the random forest algorithm.

[0227] Based on the updated model parameters, preliminary monitoring results of multi-element integrated monitoring data for marine ranches are generated; spatial interpolation processing is performed on the preliminary monitoring results, and the data distribution is optimized using the Kriging interpolation method to obtain the final integrated monitoring results; environmental assessment indicators are extracted from the final integrated monitoring results to generate comprehensive environmental status data for marine ranches.

[0228] Furthermore, for example, in acquiring multi-source monitoring data for marine ranches, environmental data can be collected by deploying various types of sensors, such as temperature, salinity, dissolved oxygen, and pH sensors, while simultaneously using sonar and cameras to acquire biological status data, such as fish density and activity patterns. Suppose a ranch collects 1000 sets of data daily, including parameters such as temperature (25.3℃) and salinity (33.5‰). When extracting dynamic change characteristics, the rate of temperature change over time can be analyzed.

[0229] For example, if the temperature rises from 25.3°C to 26.8°C on a given day, the rate of change indicates potential abnormal water flow. This embodiment uses a Kalman filter algorithm to smooth time-series data and predict the temperature trend for the next hour, generating continuous time-series trend data. Assuming the prediction shows the temperature will rise to 27.5°C within two hours, this indicates a potential risk. For processing the time-series trend data, a weighted data fusion algorithm can integrate multi-factor data. For example, temperature, salinity, and dissolved oxygen data are fused with weights of 0.4, 0.3, and 0.3 respectively to obtain comprehensive environmental data. Assuming the fusion result shows a comprehensive environmental index of 85, exceeding the threshold of 80, this indicates an abnormal environmental condition. A random forest algorithm is then used to update the evaluation model based on historical data, optimizing the model parameters.

[0230] For example, by analyzing data from the past 30 days, the model can identify the correlation between rising temperatures and reduced fish activity, and the updated model can more accurately assess environmental risks. When generating multi-factor integrated monitoring data, preliminary monitoring results can be generated by combining temperature, salinity, and other parameters. For instance, monitoring in a certain area shows dissolved oxygen levels below 5 mg / L, indicating a risk of hypoxia. In spatial interpolation processing, the Kriging interpolation method can optimize data distribution.

[0231] For example, dissolved oxygen data from 10 monitoring points within a 1000-square-meter area of ​​a ranch can be interpolated to generate a continuous distribution map covering the entire area, showing that low-oxygen areas are concentrated in the southeast corner. Finally, integrated monitoring results can be used to extract environmental assessment indicators, such as a comprehensive environmental quality index of 75, which is below the normal range, generating comprehensive environmental status data for the marine ranch.

[0232] Specifically, comprehensive environmental status data is used to guide ranch management.

[0233] For example, in low-oxygen areas, managers can adjust water flow or deploy oxygenation equipment to optimize the ranch environment. These technologies, through multi-source data fusion and dynamic analysis, provide accurate environmental status assessments, helping to identify potential problems in a timely manner and improve the monitoring efficiency and decision support capabilities of marine ranches.

[0234] Furthermore, the process of prioritizing key monitoring indicators based on the optimized multi-factor integrated monitoring data to obtain the key element monitoring framework data includes:

[0235] Environmental parameter data is collected through a sensor network and stored as an initial dataset;

[0236] Data stream processing technology is used to clean the initial dataset in real time to obtain a cleaned dataset.

[0237] If the data points in the cleaned dataset deviate from the preset threshold, outliers are removed to obtain an optimized dataset.

[0238] Principal component analysis algorithm is used to extract key indicators from the optimization dataset to obtain indicator feature set;

[0239] The weights of each indicator in the indicator feature set are calculated using the entropy method to obtain the indicator weight allocation results;

[0240] Based on the result of the indicator weight allocation, the key indicators are prioritized to obtain a priority sequence;

[0241] Based on the priority sequence and combined with dynamic monitoring and adjustment strategies, a monitoring framework data for key elements of marine ranching is generated.

[0242] Furthermore, for example, in marine ranching environmental monitoring, the deployment of sensor networks is a core component in acquiring environmental parameter data. Sensor networks typically include multiple sensors such as temperature, salinity, dissolved oxygen, and pH, deployed at different depths and in different areas of the marine ranch.

[0243] For example, a marine ranch deployed 10 monitoring points in its nearshore area, each equipped with a multi-parameter sensor. Data was collected every 5 minutes, generating an initial dataset containing parameters such as temperature (28.5℃), salinity (33.2‰), and dissolved oxygen (6.8 mg / L). This data was wirelessly transmitted and stored in a cloud database, forming the initial dataset and providing a foundation for subsequent processing. The completeness of the initial dataset directly affects the accuracy of subsequent analyses.

[0244] In one possible implementation, this embodiment cleans the initial dataset in real time using data stream processing technology. Data stream processing uses a streaming computing framework to filter and format the data on the fly.

[0245] For example, to address potential noise data from sensors, such as anomalies like a sudden temperature jump to 50°C, the system filters data within a set reasonable range (e.g., 15-35°C). During the cleaning process, the system also removes missing or duplicate values ​​to ensure data consistency, resulting in a cleaned dataset. This real-time cleaning effectively reduces data processing latency and supports dynamic monitoring.

[0246] Specifically, outlier filtering targets deviations in the cleaned dataset.

[0247] For example, a monitoring session found a dissolved oxygen level of 12.5 mg / L, far exceeding the normal range of 6-9 mg / L. Using an outlier filtering algorithm, the system, based on the statistical distribution of historical data, removed this outlier and retained the optimized dataset. The optimized dataset typically contains more reliable values, such as a dissolved oxygen mean of 7.2 mg / L and a standard deviation of 0.3, ensuring data quality.

[0248] In one embodiment, this embodiment utilizes principal component analysis (PCA) to extract key indicators from the optimized dataset. PCA uses dimensionality reduction techniques to transform multiple variables such as temperature, salinity, and dissolved oxygen into a few principal components. For example, the analysis results show that the combination of temperature and dissolved oxygen accounts for 85% of the data variance and is selected as the key indicator, forming an indicator feature set. This method reduces data redundancy and highlights the factors that have the greatest impact on environmental conditions.

[0249] For example, this embodiment uses the entropy method to calculate the weights of the indicators. Based on the indicator feature set, the system calculates the information entropy of each indicator. For example, the information entropy of dissolved oxygen is 0.92, and that of temperature is 0.87, with weights of 0.55 and 0.45, respectively. This weight allocation reflects the importance of each indicator to the environmental assessment, ensuring that subsequent analysis is more targeted.

[0250] Specifically, the priority ranking is based on the weight allocation results. For example, dissolved oxygen is ranked first due to its higher weight, followed by temperature, generating a priority sequence. The priority sequence guides the allocation of monitoring resources, such as prioritizing the calibration of dissolved oxygen sensors to ensure the monitoring accuracy of key indicators.

[0251] In one possible implementation, a dynamic monitoring and adjustment strategy is combined with a priority sequence to generate monitoring framework data.

[0252] For example, based on the priority of dissolved oxygen, the system dynamically increases the nighttime monitoring frequency to once every 3 minutes, while adjusting the collection frequency of other indicators. This dynamic adjustment optimizes the monitoring framework data, making it more adaptable to the real-time needs of changes in the marine ranch environment.

[0253] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A comprehensive monitoring method for marine ranches, characterized in that, include: By deploying sensors to collect environmental parameters and biological population data of the marine ranch, a raw dataset is obtained; Based on the original dataset, the species of planktonic organisms and water pollutants are classified and identified to obtain the classification results of environmental elements. Based on the classification results of the environmental elements, the activity trajectories and distribution of key fish populations are tracked to obtain biological dynamic monitoring data; Based on the biological dynamic monitoring data, multi-source monitoring information is integrated and a comprehensive environmental assessment model is constructed to obtain the multi-source data integration results. Based on the results of the multi-source data integration, abnormal fish behavior and the accumulation of harmful substances in the water are identified, and abnormal symptom identification results are obtained. Based on the abnormal symptom identification results, a water quality anomaly analysis model is trained and risk warning information is output. Based on the aforementioned risk warning information, the dynamic changes in the environment and biological state are continuously monitored to obtain dynamic change trend data; Based on the dynamic trend data, update the comprehensive environmental assessment model and generate optimized multi-factor integrated monitoring data; Based on the optimized multi-element integrated monitoring data, the key monitoring indicators are prioritized to obtain the key element monitoring framework data.

2. The method according to claim 1, characterized in that, The process of obtaining the raw dataset by deploying sensors to collect environmental parameters and biological population data of marine ranches includes: Real-time data on water temperature, salinity, dissolved oxygen, pH value, and fish density are obtained through sensors. If the data is missing or abnormal, the missing values ​​are interpolated to fill in the missing values ​​and obtain the complete dataset. The complete dataset is subjected to dimensionality reduction and the main features are extracted to obtain the feature dataset; Based on the feature dataset, the marine ranch area is classified into states to obtain the regional state classification results. If the regional status classification result indicates an anomaly, then analyze the data change trend of the anomaly region to obtain the change trend data; Based on the aforementioned trend data, the dynamic distribution characteristics of water quality parameters and fish density are calculated to obtain a distribution characteristic dataset; Based on the aforementioned distribution feature dataset, a spatial interpolation model is constructed to predict the state of unmonitored areas, thereby obtaining global state prediction data.

3. The method according to claim 1, characterized in that, The process of classifying and identifying planktonic species and water pollutants based on the original dataset to obtain environmental element classification results includes: Spectral features of planktonic organisms were extracted using spectral analysis techniques to obtain the first feature set; Chemical sensors were used to detect the concentration of pollutants in the water, resulting in a first concentration set. The first feature set and the first concentration set are fused to obtain a comprehensive feature set; The comprehensive feature set was classified and trained using the support vector machine algorithm to obtain a planktonic species classification model. Based on the first classification result output by the plankton species classification model, the environmental element classification is determined by the decision tree algorithm to obtain the final classification result; If the confidence level of the final classification result is lower than a preset threshold, then a second feature extraction is performed on the comprehensive feature set and the final classification result is updated.

4. The method according to claim 1, characterized in that, Based on the classification results of the environmental elements, the process of tracking the activity trajectories and distribution of key fish populations to obtain biological dynamic monitoring data includes: Sonar signal data of key fish populations were obtained using sonar technology to determine their preliminary distribution locations; Key fish species are tagged using electronic tag technology to obtain real-time location and movement data, thus obtaining activity trajectory information; If the deviation between the sonar signal data and the electronic tag location data exceeds a preset threshold, the sonar signal data and the electronic tag location data are fused and corrected to obtain the corrected population distribution and trajectory data. Based on the corrected data, a time series model of population dynamics is constructed to obtain the dynamic trend. Analyze the correlation between the classification of environmental elements and the aforementioned dynamic change trends to identify key environmental impact factors; If the key environmental impact factors exceed the preset range, the sonar scanning frequency and the electronic tag data acquisition interval are adjusted to obtain target monitoring data; Based on dynamic trends and key environmental influencing factors, predictive data on population dynamics changes in future periods are generated.

5. The method according to claim 1, characterized in that, The process of integrating multi-source monitoring information and constructing a comprehensive environmental assessment model based on the aforementioned biological dynamic monitoring data to obtain the multi-source data integration results includes: Data on water temperature, salinity, ocean currents, and fish distribution were acquired, and preprocessed to remove noise, resulting in standardized multi-source data. The standardized multi-source data is fused using the Kalman filter algorithm to obtain fused data; Based on the fused data, the spatial distribution characteristics and temporal variation trends of environmental parameters are calculated to obtain the environmental parameter distribution. If the fluctuation of the environmental parameter distribution exceeds a preset threshold, the support vector machine algorithm is used to classify the fish distribution and determine the fish gathering area. Based on the fish aggregation area and the distribution of environmental parameters, a Bayesian network model is used to predict the trend of the influence of the environment on the fish distribution, and the environmental impact prediction results are obtained. Based on the environmental impact prediction results, calculate the comprehensive environmental assessment indicators and generate a comprehensive environmental assessment model; Based on the comprehensive environmental assessment model, the environmental status level of the marine ranch is determined, and the environmental status assessment results are obtained.

6. The method according to claim 1, characterized in that, Based on the integrated results of the multi-source data, the process of identifying abnormal fish behavior and the accumulation of harmful substances in the water to obtain abnormal symptom identification results includes: Acquire fish behavior videos and aquatic environment data, integrate and process the data to obtain a standardized dataset; Image recognition technology was used to analyze fish behavior patterns in video data to identify indicators of abnormal behavior. Biomarker detection technology is used to analyze water samples to identify the concentration of harmful substances and obtain the distribution characteristics of the substances; If the abnormal behavior index exceeds a preset threshold, then a symptom correlation analysis is performed in conjunction with the substance distribution characteristics to determine the type of abnormal symptom. Based on the types of abnormal symptoms, the results of behavioral pattern analysis are combined with environmental parameter monitoring data to generate a comprehensive abnormality assessment result. The support vector machine algorithm was used to classify the correlation between abnormal fish behavior and harmful substances, and the classification model output was obtained. Based on the output of the classification model, update the environmental parameter monitoring data and determine the final distribution of abnormal symptoms.

7. The method according to claim 1, characterized in that, The process of training a water quality anomaly analysis model and outputting risk warning information based on the abnormal symptom identification results includes: Obtain abnormal symptom data and water quality abnormality data, perform data cleaning and preprocessing, and obtain a standard dataset; The key features of the standard dataset are extracted using the principal component analysis algorithm to obtain the key feature set; The key feature set was trained using the random forest algorithm to obtain a red tide occurrence probability prediction model. If the probability value output by the red tide occurrence probability prediction model exceeds the preset threshold, a logistic regression algorithm is used for secondary verification to obtain the verified probability value. Based on the verified probability values, a classification algorithm is used to determine the red tide risk level, and the risk level results are obtained. Based on the risk level results and preset warning rules, a risk warning result is generated; Data visualization technology is used to dynamically display the risk warning results and output warning information.

8. The method according to claim 1, characterized in that, The process of continuously monitoring the dynamic changes in the environment and biological states based on the aforementioned risk warning information to obtain dynamic trend data includes: Based on the acquisition of environmental parameters and biological status data by sensors, if the environmental parameters exceed the preset threshold, an anomaly detection algorithm is used to identify potential risks and generate an abnormal event set. Time series analysis algorithms are used to process real-time data streams to determine the changing trends of environmental and biological states. Based on the aforementioned trends, a decision tree algorithm is used to classify environmental anomalies and biological behaviors to obtain risk level labels; High-risk events are extracted based on the risk level labels, and dynamic trend data is generated by combining them with real-time data streams. Based on the dynamic trend data, the parameters of the risk warning algorithm are updated to obtain an optimized warning model; The optimized early warning model is used to continuously analyze the subsequently collected data and generate real-time risk warning signals.

9. The method according to claim 1, characterized in that, The process of updating the comprehensive environmental assessment model and generating optimized multi-factor integrated monitoring data based on the dynamic trend data includes: Multi-source monitoring data of marine ranches were acquired and dynamic change characteristics were extracted. The Kalman filter algorithm was used to analyze the trend and obtain time series trend data. A weighted data fusion algorithm is used to integrate multi-factor monitoring data to obtain integrated environmental data. If the integrated environmental data meets the preset threshold conditions, the parameters of the integrated environmental assessment model are updated using the random forest algorithm. Based on the updated model parameters, preliminary monitoring results of multi-element integrated monitoring data for marine ranches are generated. Spatial interpolation processing is performed on the preliminary monitoring results, and the data distribution is optimized using the Kriging interpolation method to obtain the final integrated monitoring results; Environmental assessment indicators are extracted from the final integrated monitoring results to generate comprehensive environmental status data for the marine ranch.

10. The method according to claim 1, characterized in that, The process of prioritizing key monitoring indicators based on optimized multi-factor integrated monitoring data to obtain key element monitoring framework data includes: Environmental parameter data is collected through a sensor network and stored as an initial dataset; The initial dataset is cleaned in real time using data stream processing technology to obtain a cleaned dataset. If the data points in the cleaned dataset deviate from the preset threshold, outliers are removed to obtain an optimized dataset. Principal component analysis algorithm is used to extract key indicators from the optimized dataset to obtain an indicator feature set; The weights of each indicator in the feature set are calculated using the entropy method to obtain the indicator weight allocation results; Based on the weight allocation results of the indicators, the key indicators are prioritized to obtain a priority sequence; Based on the priority sequence and combined with the dynamic monitoring and adjustment strategy, a monitoring framework data for key elements of marine ranching is generated.

Images