Method for early warning of algal bloom levels based on Ordinal Forests model

Patent metadata

Field	Value
Publication number	US-12504418-B2
Application number	US-202519026492-A
Country	US
Kind code	B2
Filing date	Jan 17, 2025
Priority date	Apr 10, 2024
Publication date	Dec 23, 2025
Grant date	Dec 23, 2025

Abstract

A method for early warning of algal bloom levels based on an Ordinal Forests model includes the following steps: S1, preprocessing water quality data from a system for online monitoring of water quality and water ecology; S2, determining an algal bloom level according to a chlorophyll a value of the pre-processed water quality data; S3, using a resampling method to solve the problem of imbalanced algal bloom level data, and synthesizing a dataset of balanced algal bloom levels; and S4, taking the newly synthesized dataset in the S3 as an input variable, constructing a model for early warning of algal bloom levels based on the Ordinal Forests model, and performing early warning of algal bloom levels through the trained model for early warning of algal bloom levels.

First claim

What is claimed is:

1 . A method for an early warning of algal bloom levels based on an Ordinal Forests model, comprising the following steps:
S1, preprocessing water quality data from a system for online monitoring of water quality and water ecology; wherein the preprocessing the water quality data in the S1 comprises the following steps:
S11, performing quality control on the water quality data, comprising deleting abnormal type data, deleting duplicate data, and resampling data;
S12, processing missing values of the water quality data, using a Kalman filtering method to fill the missing values, gradually improving an accuracy of state estimation through continuous measurement value integration and state estimation, and completing filling of the missing values;
S13, selecting features of the water quality data through a principal component analysis method; and
S14, standardizing the water quality data, with a calculation formula as follows:

y
i

=

x
i

-

x
_

σ

,

wherein in the calculation formula, y i is standardized data; x i is measured data; x is a mean of a dataset; and σ is a standard deviation of the dataset;

S2, determining an algal bloom level according to a chlorophyll a value of pre-processed water quality data;
S3, using a resampling method to solve a problem of imbalanced algal bloom level data, and synthesizing a dataset of balanced algal bloom levels; and
S4, taking a newly synthesized dataset from the S3 as an input variable, constructing a model for the early warning of algal bloom levels based on the Ordinal Forests model, and performing the early warning of algal bloom levels through a trained model for the early warning of algal bloom levels;
wherein the S4 comprises the following steps:
S41, dividing the newly synthesized dataset from the S3 through a 5-fold cross-validation method, i.e., dividing the newly synthesized dataset into five folds, selecting a fold of the newly synthesized dataset as a test set, with other folds of the newly synthesized dataset as training sets in each round, and repeating five rounds; and
S42, training the Ordinal Forests model using data of the training sets with given hyperparameters, using algal bloom level data and water quality data of a previous day as input variables of the Ordinal Forests model, and using algal bloom level data of a current day as an output variable of the Ordinal Forests model to construct the model for the early warning of algal bloom levels based on the Ordinal Forests model;
S43, inputting test set data into the model for the early warning of algal bloom levels, evaluating a performance of the trained model for the early warning of algal bloom levels in accuracy and consistency, and adjusting hyperparameters of the model for the early warning of algal bloom levels; and

S44, applying the trained model for the early warning of algal bloom levels to a water quality monitoring site, repeating steps S1, S2, and S3 to preprocess water quality data and determine algal bloom levels every day, and inputting the algal bloom level data and the water quality data of each day as input variables into the model for the early warning of algal bloom levels, to obtain a local daily algal bloom level forecast.

2 . The method for the early warning of algal bloom levels based on the Ordinal Forests model according to claim 1 , wherein the water quality data in the S1 comprises concentration data, pH data, water temperature data, conductivity data, turbidity data, dissolved inorganic nitrogen data, and dissolved inorganic phosphorus data of chlorophyll a.

3 . The method for the early warning of algal bloom levels based on the Ordinal Forests model according to claim 1 , wherein the deleting the abnormal type data in the S11 means deleting non-numeric data, wherein the non-numeric data comprises characters and null values; the deleting duplicate data means deleting data with duplicate timestamps; and the resampling data means unified data resampling on a daily basis by using a median method.

4 . The method for the early warning of algal bloom levels based on the Ordinal Forests model according to claim 3 , wherein the S13 comprises the following steps:
S131, centralizing and standardizing processing of raw water quality data such that each feature has a mean of 0 and a variance of 1; S132, calculating a covariance matrix of standardized data, and then performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues and corresponding eigenvectors; and S133, selecting a number of principal components to be retained according to the eigenvalues, and selecting features required for modeling.

5 . The method for the early warning of algal bloom levels based on the Ordinal Forests model according to claim 1 , wherein the determining the algal bloom level in the S2 comprises the following steps:
S21, observing relations between chlorophyll a concentration changes and various physical indicators according to data of field observations of water transparency changes in a reservoir, wherein the physical indicators comprise water body colors and transparency; and S22, with reference to relevant standards, dividing algal bloom risks into five levels according to a total chlorophyll a concentration: level I: chlorophyll a value≤10 μg/L, indicating no algal bloom and good water quality; level II: 10 μg/L<chlorophyll a value≤15 μg/L, early warning, indicating a potential risk of algal bloom; level III: 15 μg/L<chlorophyll a value≤50 μg/L, indicating mild algal bloom, and an obvious algal bloom event in a water body; level IV: 50 μg/L<chlorophyll a value≤100 μg/L, indicating severe algal bloom, and an algal bloom event posing a serious threat to an ecosystem; and level V: chlorophyll a value >100 μg/L, indicating an algal bloom disaster easily causing extremely serious ecological, social, and economic impacts.

6 . The method for the early warning of algal bloom levels based on the Ordinal Forests model according to claim 1 , wherein the S3 comprises the following steps:
S31, counting a total number of categories of each algal bloom level according to algal bloom level data obtained after determination based on standards in the S2, and determining a data imbalance ratio before adjustment; and S32, applying adaptive synthetic sampling of the algal bloom level data to synthesize a new dataset, and performing a data balance calculation according to differences in a number of samples of different categories of the algal bloom level data, wherein the data balance calculation comprises the following steps: S321, calculating a proportion p of majority class samples around each minority class sample; S322, for each minority class sample, calculating a number of new samples that need to be generated β=p×α, wherein α is an adjustable parameter used to control the number of new samples; and S323, randomly selecting n samples from k majority class samples nearest to each minority class sample, and inserting the n samples between the minority class sample and the k majority class samples nearest thereto to synthesize a balanced dataset of algal bloom levels.

Assignees

• Univ Xiamen

Inventors

• Liang Zhongyao
• Zhao Xinye
• Chen Nengwang
• Chen Jixin
• Qu Fan
• Yu Yiqi
• Li Shaobin

Classifications

• G06N20/20

  Ensemble learning · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N20/00

  Machine learning · CPC title

• Y02A20/152

  Water filtration · CPC title

• G06F18/213

  Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods · CPC title