Red tide early warning method and system based on buoy data

Abstract

The invention relates to a red tide early warning method and system based on buoy data. The early warning method comprises the steps that historical buoy data samples obtained in advance are preprocessed, a buoy data set is constructed, and the historical buoy data samples comprise red tide state samples and non-red tide state samples; performing feature engineering processing on the buoy data set, and screening to obtain a red tide key factor; constructing a key feature data set according to the red tide key factors, importing the key feature data set into a pre-configured deep learning network, and training to obtain a red tide prediction model; real-time buoy data are obtained and imported into the red tide prediction model for prediction, and real-time red tide early warning is carried out. According to the technical scheme, red tide real-time early warning can be achieved, the early warning accuracy is high, and the real-time early warning and forecasting requirement is met.

Description

Technical Field

[0001] This application relates to the field of marine red tide early warning, and in particular to a red tide early warning method and system based on buoy data. Background Art

[0002] A red tide is a devastating marine ecological phenomenon caused by the explosive proliferation or high concentration of certain phytoplankton, protozoa, or bacteria in seawater under specific environmental conditions, discoloring the water and affecting and endangering the normal survival of other marine life. Therefore, red tide warnings are crucial for mitigating marine environmental disasters and preventing significant economic losses in marine industries, particularly fisheries.

[0003] Numerous buoys collect vast amounts of hydrological monitoring data, but this data suffers from excessive noise and processing overhead, making it impractical for direct application in red tide early warning systems. Identifying the key factors influencing red tide early warnings from the numerous influencing factors collected by buoys and accurately predicting them is crucial for guiding real-time red tide monitoring and early warning. Summary of the Invention

[0004] This application provides a red tide early warning method and system based on buoy data, which at least solves the problem that the limitations of related technologies in determining the trunk line coordinated control scheme lead to poor control effect.

[0005] Firstly, this application provides a red tide early warning method based on buoy data, the method comprising:

[0006] The previously acquired historical buoy data samples are preprocessed to construct a buoy dataset, which includes red tide state samples and non-red tide state samples.

[0007] Feature engineering was performed on the buoy dataset to screen and obtain key red tide factors;

[0008] A key feature dataset is constructed based on the red tide key factors, and the key feature dataset is imported into a pre-configured deep learning network to train and obtain a red tide prediction model.

[0009] Real-time data from buoys is acquired and imported into the red tide prediction model for prediction, enabling real-time red tide early warning.

[0010] Optionally, the preprocessing of the pre-acquired historical buoy data samples includes:

[0011] Initial screening of error values ​​in the historical buoy data sample;

[0012] Error values ​​and missing values ​​in the historical buoy data samples are interpolated and supplemented.

[0013] Optionally, the buoy data set contains several monitoring factors, which at least include:

[0014] Wind speed, wind direction, air temperature, air pressure, air humidity, total suspended solids, fluorescent soluble organic matter, turbidity, pH, water temperature, salinity, chlorophyll a, electrical conductivity, total dissolved solids, dissolved oxygen, and blue-green algae density.

[0015] Optionally, feature engineering is performed on the buoy dataset to filter and obtain key red tide factors, including:

[0016] The features of the buoy dataset are further processed to obtain daily ordinary feature samples and daily composite feature samples;

[0017] Screening and processing the composite characteristic sample and the common characteristic sample to obtain red tide characteristic factors;

[0018] The red tide characteristic factors were normalized to obtain the key red tide factors.

[0019] Optionally, the step of adding processing to the features of the buoy dataset to obtain daily ordinary feature samples and daily composite feature samples includes:

[0020] The monitoring factors for each preset time period are obtained and defined as ordinary feature samples;

[0021] Some of the monitoring factors are defined as key factors, and the daily discrete changes of the key factors are processed to obtain daily composite feature samples.

[0022] Optionally, the composite feature sample includes maximum value data of the key factor, range value data of the key factor, and mean value data of the key factor.

[0023] Optionally, the feature filtering process of the composite feature sample and the ordinary feature sample to obtain red tide feature factors includes:

[0024] Importing the composite characteristic sample and the common characteristic sample into a preset characteristic screening model to screen and obtain initial red tide characteristic factors;

[0025] The initial red tide characteristic factors are subjected to correlation analysis, and the red tide characteristic factors are screened again to obtain the red tide characteristic factors.

[0026] Optionally, the training to obtain a red tide prediction model includes:

[0027] Based on the aforementioned key red tide factors, training and testing datasets are constructed according to a preset ratio.

[0028] The training dataset is imported into a pre-configured deep learning network to train and obtain a red tide prediction model.

[0029] The test dataset is imported into the red tide prediction model to obtain training prediction results, and the training prediction results are compared with the red tide days of the test dataset to obtain the prediction accuracy.

[0030] Based on the predicted accuracy, the red tide prediction model is adjusted and optimized.

[0031] Secondly, this application provides a red tide early warning system based on buoy data, the system comprising:

[0032] The data acquisition module is used to preprocess the pre-acquired historical buoy data samples to construct a buoy dataset. The historical buoy data samples include red tide state samples and non-red tide state samples.

[0033] The feature engineering module is used to perform feature engineering processing on the buoy dataset and filter out key red tide factors.

[0034] The model training module is used to construct a key feature dataset based on the red tide key factors, and import the key feature dataset into a pre-configured deep learning network to train and obtain a red tide prediction model.

[0035] The red tide early warning module is used to acquire real-time data from buoys and import it into the red tide prediction model for prediction and real-time red tide early warning.

[0036] Thirdly, this application provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the red tide early warning method described in the first aspect.

[0037] Compared with related technologies, the red tide early warning method and system based on buoy data provided in this application have at least the following technical advantages:

[0038] By performing feature engineering on pre-acquired historical buoy data, key red tide factors are selected; a key feature dataset is constructed based on the key red tide factors, and the key feature dataset is imported into a pre-configured deep learning network to train a red tide prediction model; real-time buoy data is acquired and imported into the red tide prediction model for prediction, resulting in real-time red tide warnings with high accuracy, meeting the requirements for real-time early warning and forecasting.

[0039] The details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. Attached Figure Description

[0040] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0041] Figure 1 This is a flowchart illustrating a red tide early warning method based on buoy data, according to an exemplary embodiment.

[0042] Figure 2 This is a flowchart illustrating step S20 according to an exemplary embodiment.

[0043] Figure 3 This is a flowchart illustrating step S201 according to an exemplary embodiment.

[0044] Figure 4 This is a flowchart illustrating step S202 according to an exemplary embodiment.

[0045] Figure 5 This is a schematic diagram of the LightGBM model according to an exemplary embodiment.

[0046] Figures 6A to 6I This is a feature engineering screening diagram based on an LGBM model, as illustrated in an exemplary embodiment.

[0047] Figure 7 This is a correlation analysis diagram of initial red tide characteristic factors according to an exemplary embodiment.

[0048] Figure 8A This is a visualization of the TSNE dimensionality reduction and clustering results of the data before feature engineering.

[0049] Figure 8B This is a visualization of the TSNE dimensionality reduction and clustering results of the data after feature engineering.

[0050] Figure 9 This is a flowchart illustrating the red tide prediction model training process according to an exemplary embodiment.

[0051] Figure 10 This is an ROC curve of a pre-engineered red tide prediction model according to an exemplary embodiment.

[0052] Figure 11 This is an ROC curve of a red tide prediction model after feature engineering, as illustrated in an exemplary embodiment.

[0053] Figure 12 This is a block diagram of a red tide early warning system based on buoy data, according to an exemplary embodiment. DETAILED DESCRIPTION

[0054] To make the objectives, technical solutions, and advantages of this application clearer, the application is described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0055] Obviously, the accompanying drawings described below are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, any changes to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0056] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0057] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects are in an "or" relationship. The terms "first," "second," and "third" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0058] In related technologies, numerous buoys can collect massive amounts of hydrological monitoring data. However, the noise and processing volume of this massive data are too high, making it unsuitable for direct application in red tide early warning. Therefore, identifying key factors influencing red tide early warning from the numerous influencing factors collected by buoys and achieving accurate red tide prediction based on these key factors is of great significance for guiding real-time monitoring and early warning of red tides.

[0059] Based on the above situation, an embodiment of the present invention provides a red tide early warning method and system based on buoy data, which is described in detail below with reference to specific embodiments and drawings.

[0060] Example 1

[0061] An embodiment of the present invention provides a red tide early warning method based on buoy data. Figure 1 FIG. 1 is a flow chart of a red tide early warning method based on buoy data according to an exemplary embodiment. Figure 1 As shown, the method includes: This application provides a red tide early warning method based on buoy data, the method includes:

[0062] S10, preprocessing the historical buoy data samples acquired in advance to construct a buoy data set, wherein the historical buoy data samples include red tide state samples and non-red tide state samples;

[0063] In this embodiment, the historical buoy data samples are based on the results of field investigation and laboratory monitoring, and correspond one-to-one with the buoy monitoring data to obtain red tide state samples and non-red tide state samples. In this embodiment, 104 days of historical buoy data samples are collected, including 34 days of red tide and 70 days of non-red tide. The historical buoy data samples corresponding to the red tide date and the non-red tide date are obtained from the buoy database according to the corresponding date.

[0064] Furthermore, the existing buoy update frequency is 0.5 hours / time. Combined with the above-mentioned number of collection days, a total of 24h / (0.5h / time)*34=1632 samples of buoy data were obtained in red tide conditions, and 24h / (0.5h / time)*70=3360 samples in non-red tide conditions. Each sample in the buoy dataset includes at least 16 monitoring factors: wind speed, wind direction, air temperature, air pressure, air humidity, total suspended solids, fluorescent soluble organic matter, turbidity, pH, water temperature, salinity, chlorophyll a, conductivity, total dissolved solids, dissolved oxygen, and cyanobacteria density.

[0065] In this embodiment, the pre-acquired historical buoy data samples are preprocessed, including:

[0066] Initial screening of error values ​​in historical buoy data samples;

[0067] Interpolate and fill in erroneous and missing values ​​in historical buoy data samples;

[0068] In this embodiment, the initial screening of error values ​​in historical buoy data samples includes: first, performing non-negativity screening on the historical buoy data samples, after which the vast majority of monitoring factors in the historical buoy data samples are non-negative;

[0069] Secondly, screening is conducted based on the general range of common meteorological indicators and commonly used water quality indicators, such as the normal range of air pressure and wind speed, and the normal range of water temperature in different seasons.

[0070] Secondly, error values ​​are screened based on the normal range of the monitoring factors mentioned above. For example, the upper and lower limit thresholds of pH, chlorophyll a, dissolved oxygen, etc. under red tide and non-red tide conditions. When the index is significantly lower than or higher than the threshold of each condition, it can be judged as abnormal / erroneous data.

[0071] In this embodiment, the interpolation and completion of erroneous and missing values ​​in historical buoy data samples includes: uniformly setting erroneous and missing values ​​to null values, and interpolating the null value data using an interpolation method.

[0072] First, obtain the time that needs to be interpolated, and determine the type of break in the time point that needs to be interpolated: the first type of data is the time point that is broken at the beginning and end of a certain continuous time period; the second type of data is the point that is broken in the middle of a certain continuous time period.

[0073] Secondly, depending on the type of disconnection, the mean method is used to interpolate the mean forward or backward for the first type of data. For example, if two data points are missing at the beginning, the first interpolation is the mean of the first two existing data points, and the second interpolation is the mean of the first three existing data points. For the second type of data, cubic spline interpolation is used to interpolate and fill in the missing data, and finally all errors and missing data are filled in and corrected.

[0074] S20. Perform feature engineering on the buoy dataset and filter to obtain key factors of red tide;

[0075] Figure 2 This is a flowchart illustrating step S20 according to an exemplary embodiment. (Refer to the attached document.) Figure 2 In this embodiment, step S20 specifically includes:

[0076] S201. Add processing to the features of the buoy dataset to obtain daily ordinary feature samples and daily composite feature samples;

[0077] S202. Perform feature screening and processing on composite feature samples and ordinary feature samples to obtain red tide characteristic factors;

[0078] S203. Normalize the red tide characteristic factors to obtain the key red tide factors.

[0079] Optionally, Figure 3 This is a flowchart illustrating step S201 according to an exemplary embodiment. (Refer to the attached document.) Figure 3 In one exemplary embodiment, step S201 specifically includes:

[0080] S2011. Obtain the monitoring factors for the preset time period each day and define them as ordinary feature samples;

[0081] Among them, the preset time period each day is the time period with sufficient sunlight. In this embodiment, the data collected at 10:00 am every day is used as an example. A total of 16 monitoring factors are collected at 10:00 am every day: wind speed, wind direction, temperature, air pressure, air humidity, total suspended solids, fluorescent soluble organic matter, turbidity, pH, water temperature, salinity, chlorophyll a, conductivity, total dissolved solids, dissolved oxygen, and blue-green algae density. These are defined as ordinary feature samples.

[0082] S2012. Define some of the monitoring factors as key factors, and process the discrete changes of the key factors daily to obtain daily composite feature samples; further, the composite feature samples include the maximum value data of the key factors, the range value data of the key factors, and the mean value data of the key factors.

[0083] In this embodiment, chlorophyll a value, pH value and dissolved oxygen value are defined as key factors among the monitoring factors. Discrete changes are made to these three key factors to obtain a total of nine composite feature samples: maximum value of chlorophyll a, range of chlorophyll a, mean value of chlorophyll a, maximum value of pH, range of pH, mean value of pH, maximum value of dissolved oxygen, range of dissolved oxygen and mean value of dissolved oxygen.

[0084] In this way, the daily monitoring data sample of the buoy was reduced to 1 case per day, resulting in a total of 34 red tide samples and 70 non-red tide samples. Based on these 25 characteristic factors (16 ordinary characteristic samples + 9 composite characteristic samples), a characteristic sample table as shown in Appendix 1 was formed:

[0085] Table 1 Feature Sample Table

[0086]

[0087]

[0088] Optionally, Figure 4 This is a flowchart illustrating step S202 according to an exemplary embodiment. Since unnecessary features reduce training speed and model interpretability, we need to further filter the feature samples in step S201 to remove unimportant feature samples and retain important feature samples; refer to the attached... Figure 4 In an exemplary embodiment, step S202 specifically includes:

[0089] S2021. Importing the composite characteristic samples and the common characteristic samples into a preset characteristic screening model to screen and obtain initial red tide characteristic factors;

[0090] S2022. Correlation analysis was performed on the initial red tide characteristic factors, and red tide characteristic factors were obtained again through screening.

[0091] Continue to refer to the appendix Figure 4In step S2021, we selected the LightGBM model as the feature selection model. GBDT (Gradient Boosting Decision Tree) is a long-standing and popular model in machine learning. Its main idea is to use weak classifiers (decision trees) for iterative training to obtain the optimal model. GBDT's concept gives it a natural advantage in discovering various discriminative features and feature combinations. LightGBM, developed by Microsoft in 2016, is a Gradient Boosting Decision Tree model. Its basic idea is to first discretize continuous floating-point feature values ​​into k integers, simultaneously constructing a histogram of width k as shown in the diagram. Initially, the discretized values ​​are used as indices to accumulate statistics in the histogram. After traversing the data once, the histogram accumulates the statistics needed for discretization. When splitting nodes, the optimal split point can be found from these k buckets based on the discrete values ​​in the histogram. Compared to other GBDT models, LightGBM's biggest advantages are faster training efficiency and higher accuracy, making it suitable for ranking, classification, regression, and many other machine learning tasks.

[0092] Figure 5 This is a schematic diagram of the LightGBM model according to an exemplary embodiment. Figure 5 By adjusting various aspects of the LightGBM model, such as model depth, number of leaf nodes, and learning rate, the optimal configuration suitable for the technical solution of this embodiment is found as the feature selection model. Then, the built-in feature importance evaluation is used to further optimize feature selection. Finally, the cumulative sum of importance is calculated and ranked by feature importance.

[0093] Figures 6A to 6I This is a feature engineering screening diagram based on an LGBM model, illustrated according to an exemplary embodiment. See attached diagram. Figures 6A to 6I The daily sample of 25 feature factors (16 ordinary feature samples + 9 composite feature samples) was imported into the feature screening model based on the LGBM model. Based on the statistical characteristics of the feature screening model, the top 15 features with the highest importance were selected. Appendix 2 is a feature importance ranking analysis table based on the statistical characteristics of the feature screening model.

[0094] Table 2. Feature Importance Ranking Analysis Based on Statistical Properties of Feature Screening Model

[0095]

[0096]

[0097] Combined with appendix Figures 6A to 6IAccording to Table 2, the top 15 important characteristic factors in this embodiment are obtained as the initial red tide characteristic factors, namely: chlorophyll a maximum value, chlorophyll a value, chlorophyll a range value, dissolved oxygen value, chlorophyll a mean value, pH mean, dissolved oxygen mean, dissolved oxygen maximum value, salinity value, total suspended solids value, dissolved oxygen range value, pH maximum value, pH range value, conductivity value, and wind speed value.

[0098] Understandable, Figures 6A to 6I In Appendix 2, chla represents chlorophyll a value, max(chla) represents the maximum value of chlorophyll a, diff(chla) represents the range of chlorophyll a values, and avg(chla) represents the mean value of chlorophyll a; PH represents the pH value, max(PH) represents the maximum value of pH, diff(PH) represents the range of pH values, and avg(PH) represents the mean value of pH; rjy represents the dissolved oxygen value, max(rjy) represents the maximum value of dissolved oxygen, diff(rjy) represents the range of dissolved oxygen values, and avg(rjy) represents the mean value of dissolved oxygen; ddl represents the electrical conductivity; salt represents the salinity value; wet represents the air humidity value; shuiwen represents the water temperature value; wind represents the wind speed value; zxfgt represents the total suspended solids value; zrjxgt represents the total dissolved solids value; and ygkrxyjw represents fluorescent soluble organic matter.

[0099] Figure 7 This is a correlation analysis diagram of initial red tide characteristic factors according to an exemplary embodiment. See Appendix [Figure Number]. Figure 4 and 7 In step S2022, among the top 15 important features selected in step S2021, too many highly correlated features will also amplify the effect of noise. Multiple features actually "half-divide" the contribution of such features to the model, which makes the model more sensitive to data changes, increases the generalization error, and leads to poor final classification prediction results. Figure 6A ~Attached Figure 6I And attached Figure 7 The correlation between the five pairs of feature factors—the maximum value of chlorophyll a and the range of chlorophyll a, the value of dissolved oxygen and the maximum value of dissolved oxygen, the mean value of pH and the value of pH, the conductivity and the salinity, and the total dissolved solids and the salinity—was relatively high. Referring to the previously selected feature ranking, five redundant features—the range of chlorophyll a, the maximum and minimum values ​​of dissolved oxygen, the real-time value of pH, the conductivity, and the total dissolved solids—were removed from the feature importance ranking based on the statistical characteristics of the feature selection model. This resulted in the final ten red tide feature factors used for prediction and classification: the maximum and minimum values ​​of chlorophyll a, chlorophyll a, dissolved oxygen, the mean value of pH, the mean value of dissolved oxygen, the mean value of chlorophyll a, the salinity value, the water temperature value, the range of dissolved oxygen, the range of pH, and the total suspended solids value.

[0100] Furthermore, efficient feature engineering is beneficial to improving the accuracy of red tide prediction models. In order to verify the effectiveness of feature engineering in steps S201 and S202, in step S203, the red tide feature factors are normalized to obtain the key red tide factors. In this embodiment, the feature engineering effect of steps S201 and S202 is verified by TSNE-based dimensionality reduction clustering and visualized.

[0101] It is well known that for clustering, the input feature dimension is high-dimensional (greater than three dimensions), and it is generally difficult to directly display the clustering results with the original features. TSNE provides an effective data dimensionality reduction mode, which uses a nonlinear dimensionality reduction algorithm to allow us to display the clustering results in two-dimensional or three-dimensional space. Since the numerical values ​​of the red tide characteristic factors vary, in order to reduce the influence of the large weights of some data due to their large numerical values ​​and make the calculation more reasonable, it is necessary to normalize the red tide characteristic factor data and map them to the same scale. In this embodiment, the training data and the measured data are mapped by the TSNE algorithm to obtain the overall distribution of the measured mapping points and the training data mapping points, thereby judging the clustering quality of the mapping points.

[0102] Table 3 TSNE Dimensionality Reduction Clustering Characteristics

[0103] Dimensionality reduction feature X0 Dimensionality reduction feature X1 Red Tide Type -1.58816004 9.74566364 0. -0.5707761 9.44970894 0 -1.52651644 10.23952675 0. -7.41239357 -2.55537724 1. -6.02822685 -2.19920301 0. -6.89476728 -2.66878223 1. ... ... ...

[0104] Table 3 is the TSNE dimensionality reduction clustering feature table. After two-dimensional dimensionality reduction using the TSNE algorithm, two features, X0 and X1, are formed in the table. The red tide type is 0 for non-red tide days and 1 for red tide days. Figure 8A This is a visualization of the TSNE dimensionality reduction and clustering results of the data before feature engineering. Figure 8B The image shows the TSNE dimensionality reduction and clustering visualization of the data after feature engineering. See Table 3 and Appendix. Figure 8A and attached Figure 8B The non-red tide state samples are 0; the red tide state samples are 1. According to the dimensionality reduction clustering effect, the non-red tide state samples and red tide state samples in the data without feature engineering are randomly distributed and difficult to distinguish. However, after feature engineering, the samples can be clearly distinguished, and the clustering effect is better.

[0105] S30. Construct a key feature dataset based on the key factors of red tide, and import the key feature dataset into a pre-configured deep learning network to train and obtain a red tide prediction model.

[0106] S40. Acquire real-time buoy data and import it into the red tide prediction model for prediction, providing real-time red tide warnings.

[0107] In one exemplary embodiment, deep learning, with its powerful deep data mining and feature extraction capabilities, has developed rapidly as an emerging direction in the field of machine learning and has attracted widespread attention from all sectors of society. Deep Belief Networks (DBNs) combine the characteristics of supervised and unsupervised classification. They consist of several layers of neurons, with Restricted Boltzmann Machines (RBMs) as their constituent elements. Connecting several RBMs in series forms a DBN. DBNs combine forward unsupervised learning with backward supervised learning, effectively suppressing the overfitting phenomenon that easily occurs in neural networks during classification, thereby improving the accuracy of DBN model classification or prediction.

[0108] In step S30 of this application embodiment, by constructing a DBN model, applying the DBN to red tide disaster monitoring and early warning, and training a red tide prediction model, it is expected to improve the accuracy of red tide early warning and prediction. Figure 9 This is a flowchart illustrating the red tide prediction model training process according to an exemplary embodiment. (See attached diagram.) Figure 9 ,The red tide prediction model training process includes:

[0109] S301. Construct training and test datasets according to preset proportions based on key red tide factors;

[0110] S302. Import the training dataset into the pre-configured deep learning network and train it to obtain the red tide prediction model.

[0111] S303. Import the test dataset into the red tide prediction model to obtain the training prediction results, and compare the training prediction results with the red tide days of the test dataset to obtain the prediction accuracy.

[0112] For further information, please refer to the attached Figure 9 , the red tide prediction model training process also includes:

[0113] S304. Adjust and optimize the red tide prediction model based on the prediction accuracy; improve the accuracy of the red tide prediction model in classification or prediction.

[0114] Figure 10 This is an ROC curve of a pre-engineered red tide prediction model according to an exemplary embodiment. Figure 11 This is an ROC curve of a red tide prediction model after feature engineering, according to an exemplary embodiment. (See attached diagram.) Figure 10-11It can be seen that the supervised DBN model after feature engineering has higher accuracy and a better ROC curve. The red tide prediction model before feature engineering has an accuracy of 87.5%, while the red tide prediction model after feature engineering has an accuracy of over 97%. The prediction method provided in this embodiment has been applied to red tide monitoring in the Gulf. Since 2024, it has effectively monitored the Gulf online for more than 180 days, issued 19 red tide warnings, and achieved an accuracy of 97%. The high accuracy of the red tide warnings meets the requirements for real-time early warning and forecasting.

[0115] In summary, the red tide early warning method based on buoy data provided in this embodiment of the invention performs feature engineering on pre-acquired historical buoy data to screen and obtain key red tide factors; constructs a key feature dataset based on the key red tide factors, and imports the key feature dataset into a pre-configured deep learning network to train a red tide prediction model; acquires real-time buoy data and imports it into the red tide prediction model for prediction, providing real-time red tide early warning with high accuracy, thus meeting the requirements for real-time early warning and forecasting.

[0116] Example 2

[0117] Example 2 of the present application provides a red tide early warning system based on buoy data. Figure 12 FIG is a block diagram of a red tide early warning system based on buoy data according to an exemplary embodiment. Figure 12 , the system comprises:

[0118] The data acquisition module is used to preprocess the pre-acquired historical buoy data samples to construct a buoy dataset. The historical buoy data samples include red tide state samples and non-red tide state samples.

[0119] Feature engineering module, used to perform feature engineering on buoy datasets and screen out key factors of red tide;

[0120] The model training module is used to construct a key feature dataset based on the key factors of red tides, and import the key feature dataset into a pre-configured deep learning network to train a red tide prediction model;

[0121] The red tide warning module is used to obtain real-time data from buoys and import it into the red tide prediction model for prediction, providing real-time red tide warnings.

[0122] In summary, the red tide early warning method and system based on buoy data provided in this invention performs feature engineering on pre-acquired historical buoy data to screen and obtain key red tide factors; constructs a key feature dataset based on the key red tide factors, and imports the key feature dataset into a pre-configured deep learning network to train a red tide prediction model; acquires real-time buoy data and imports it into the red tide prediction model for prediction, providing real-time red tide early warning with high accuracy, thus meeting the requirements for real-time early warning and forecasting.

[0123] Example 3

[0124] Embodiment 3 of this application provides a computer-readable storage medium storing a program thereon. When the program is executed by a processor, it implements the red tide early warning method in Embodiment 1 above.

[0125] The readable storage medium may include, but is not limited to, a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0126] In a possible implementation, the present invention can also be implemented as a program product comprising program code, which, when the program product is run on a terminal device, causes the terminal device to perform the steps of implementing the red tide warning method in Embodiment 1.

[0127] The program code for executing the present invention can be written in any combination of one or more programming languages. The program code can be executed entirely on the user device, partially on the user device, as a standalone software package, partially on the user device and partially on a remote device, or entirely on a remote device.

[0128] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0129] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A red tide early warning method based on buoy data, characterized in that: The method comprises: Preprocessing the previously acquired historical buoy data samples to construct a buoy data set, wherein the historical buoy data samples include red tide state samples and non-red tide state samples; Performing feature engineering processing on the buoy data set to screen and obtain key factors of red tide; Constructing a key feature data set based on the key factors of red tide, and importing the key feature data set into a pre-configured deep learning network to train and obtain a red tide prediction model; The real-time data of the buoy is obtained and imported into the red tide prediction model for prediction, and a real-time red tide warning is issued.

2. The red tide early warning method according to claim 1, characterized in that: The preprocessing of the pre-acquired historical buoy data samples includes: Preliminary screening of error values ​​of the historical buoy data samples; The erroneous values ​​and missing values ​​of the historical buoy data samples are supplemented by interpolation.

3. The red tide early warning method according to claim 1, characterized in that: The buoy data set contains several monitoring factors, which at least include: Wind speed, wind direction, air temperature, air pressure, air humidity, total suspended solids, fluorescent soluble organic matter, turbidity, pH, water temperature, salinity, chlorophyll a, electrical conductivity, total dissolved solids, dissolved oxygen and blue-green algae density.

4. The red tide early warning method according to claim 3, characterized in that: The buoy data set was processed by feature engineering to screen out key factors of red tide, including: Adding processing to the features of the buoy data set to obtain daily common feature samples and daily composite feature samples; Screening and processing the composite characteristic samples and the common characteristic samples to obtain red tide characteristic factors; The red tide characteristic factors are normalized to obtain the key factors of red tide.

5. The red tide early warning method according to claim 4, characterized in that: The step of adding features to the buoy data set to obtain daily common feature samples and daily composite feature samples includes: Obtain the monitoring factors during a preset period of time each day and define them as common feature samples; Some of the monitoring factors are defined as key factors, and the daily discrete changes of the key factors are processed to obtain daily composite feature samples.

6. The red tide early warning method according to claim 4, characterized in that: The composite feature sample includes the maximum value data of the key factor, the extreme value data of the key factor and the mean value data of the key factor.

7. The red tide early warning method according to claim 4, characterized in that: The method of screening and processing the composite characteristic samples and the common characteristic samples to obtain red tide characteristic factors includes: Importing the composite characteristic sample and the common characteristic sample into a preset characteristic screening model to screen and obtain initial red tide characteristic factors; A correlation analysis is performed on the initial red tide characteristic factors, and the red tide characteristic factors are screened again to obtain the red tide characteristic factors.

8. The red tide early warning method according to claim 1, characterized in that: The training to obtain the red tide prediction model includes: Constructing a training data set and a test data set according to the key factors of red tide in a preset ratio; Importing the training data set into a pre-configured deep learning network to train and obtain a red tide prediction model; The test data set is imported into the red tide prediction model to obtain a training prediction result, and the training prediction result is compared with the red tide day of the test data set to obtain the prediction accuracy. According to the prediction accuracy, the red tide prediction model is adjusted and optimized.

9. A red tide early warning system based on buoy data, characterized in that: The system comprises: A data acquisition module is used to pre-process the pre-acquired historical buoy data samples to construct a buoy data set, wherein the historical buoy data samples include red tide state samples and non-red tide state samples; A feature engineering module is used to perform feature engineering processing on the buoy data set to screen and obtain key factors of red tide; A model training module is used to construct a key feature data set based on the key factors of red tide, and import the key feature data set into a pre-configured deep learning network to train and obtain a red tide prediction model; The red tide warning module is used to obtain real-time data of the buoy and import it into the red tide prediction model for prediction, and to provide real-time red tide warning.

10. A computer-readable storage medium, characterized in that: A program is stored thereon, and when the program is executed by a processor, the red tide early warning method described in any one of claims 1 to 8 is implemented.

Images