Blue-green algae bloom induction factor identification and toxin risk grading evaluation system

Abstract

The invention relates to the technical field of water environment treatment and cyanobacterial bloom early warning, in particular to a cyanobacterial bloom induction factor identification and toxin risk grading evaluation system, which comprises a multi-source data acquisition module used for acquiring environment big data, hydro-meteorological data and historical water bloom monitoring data; and the data preprocessing and fusing module is used for cleaning the multi-source data and aligning the time sequence. According to the invention, environment big data, hydro meteorological data and historical water bloom monitoring data are acquired through the multi-source data acquisition module, and the data are integrated by combining a credibility weighted fusion algorithm of the data preprocessing and fusion module, so that the problem of inaccurate multi-source data integration is solved; and the key induction factor identification and dynamic weight determination module mines and screens potential factors through multivariate statistical analysis and association rules, and then adjusts weights according to real-time and historical data deviation degrees by a sliding window dynamic weight mechanism, so that the problem of disjunction between static weights and an actual water bloom driving mechanism in the prior art is solved.

Description

Cyanobacterial Bloom Inducing Factor Identification and Toxin Risk Classification Assessment System Technical Field

[0001] This invention relates to the field of water environment management and cyanobacterial bloom early warning technology, specifically a system for identifying cyanobacterial bloom inducing factors and assessing toxin risks. Background Technology

[0002] Cyanobacterial blooms are an ecological phenomenon caused by the excessive proliferation of cyanobacteria in freshwater bodies. Their outbreaks not only deplete dissolved oxygen and block sunlight, leading to the death of aquatic organisms and disrupting the aquatic ecological balance, but also pose a serious threat to drinking water safety and human health due to the algal toxins produced by some cyanobacteria (such as Microcystis and Anabaena). Currently, the prediction and early warning of cyanobacterial blooms has become an important research direction in the field of water environment management. The conventional approach is to combine environmental data, hydrological and meteorological data, and historical bloom monitoring data to preliminarily determine the probability of bloom occurrence through statistical analysis or simple models. However, these preliminary attempts have not yet formed a systematic technical framework and are insufficient to meet the needs of precise prevention and control.

[0003] Existing cyanobacterial bloom prediction and early warning technologies, while capable of initial data integration when processing multi-source data such as water quality indicators, water temperature, rainfall, and wind speed, fail to establish a dynamic weight adjustment mechanism during the screening of key inducing factors. This leads to a severe disconnect between factor correlation analysis and timeliness. Specifically, existing technologies typically determine fixed weights for inducing factors based on historical data, failing to consider the impact of environmental conditions on factors over different periods or to adjust factor weights in light of real-time data fluctuations. This static weighting model causes the identification results of key inducing factors to deviate from the actual bloom-driving mechanism, resulting in increased bias in subsequent bloom probability predictions. Furthermore, it fails to accurately quantify the differences in toxicity risks under different scenarios, ultimately affecting the reliability of drinking water safety early warnings and the targeted nature of prevention and control measures. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a system for identifying cyanobacterial bloom inducing factors and assessing toxin risks. This system solves the problems of inaccurate integration of multi-source data, static weights deviating from the actual bloom driving mechanism, large bloom prediction bias, and difficulty in quantifying toxin production risk in existing technologies.

[0005] To achieve the above objectives, this invention provides a system for identifying cyanobacterial bloom inducing factors and assessing toxin risks, comprising: a multi-source data acquisition module for collecting environmental big data, hydrological and meteorological data, and historical bloom monitoring data; a data preprocessing and fusion module for cleaning and aligning the multi-source data, introducing a data credibility weighted fusion algorithm, assigning credibility weights to different types of data based on the stability of the data sources, and fusing the multi-source data into a unified bloom analysis dataset using a weighted average method; and a key inducing factor identification and dynamic weight determination module for preliminary factor screening based on the bloom analysis dataset, using a combination of multivariate statistical analysis and association rule mining methods to identify potential inducing factors, and introducing a sliding window dynamic weight mechanism. Within each sliding window, the contribution of potential inducing factors to the bloom event is calculated using a feature importance assessment method, and real-time and historical data are combined. The system dynamically adjusts the weights of potential inducing factors based on their deviation. A cyanobacterial bloom probability prediction module is used to construct multiple basic prediction models. These models use key inducing factors and dynamic weights as input features, and historical bloom monitoring data is used to train and optimize the parameters. An ensemble learning framework is employed, using the prediction results from multiple basic models as new input features to construct a meta-classifier that fuses the prediction results, thus outputting the probability of cyanobacterial bloom occurrence. A toxin production risk grading module is used to establish a mapping relationship between dominant algal species and toxin production probability based on historical data. It combines real-time monitoring data on algal species composition to assess the dominance of toxin-producing algae and constructs a two-factor risk assessment model based on the dominance of toxin-producing algae and algal toxin concentration data to classify toxin production risk into multiple risk levels. A results output and decision support module is used to output the bloom probability, toxin production risk level, and related assessment results in a visual format, and generate prevention and control recommendations based on the risk level.

[0006] Furthermore, the multi-source data acquisition module is used to collect environmental big data including nutrient salt indicators, dissolved oxygen, pH value and transparency parameters in water bodies; hydrological and meteorological data including daily rainfall, daily average water temperature, daily average wind speed, sunshine duration and water level change data; historical algal bloom monitoring data including records of cyanobacterial bloom occurrence, dominant algal species types and algal toxin concentration detection data.

[0007] Furthermore, the data preprocessing and fusion module has the following functions for data cleaning: outlier removal from environmental big data using outlier identification and processing methods, and missing data completion from hydrological and meteorological data using missing data completion methods; and time-series alignment, which unifies the time dimension of environmental big data, hydrological and meteorological data, and historical algal bloom monitoring data using a unified time unit.

[0008] Furthermore, the key inducing factor identification and dynamic weight determination module has the following functions for preliminary factor screening: first, it analyzes the correlation between each parameter in the algal bloom analysis dataset and the algal bloom occurrence record through correlation analysis method, so as to retain parameters whose correlation reaches a preset threshold; then, it identifies the combined correlation relationship between parameters through association rule mining method, thereby preliminarily determining potential inducing factors.

[0009] Furthermore, the key inducing factor identification and dynamic weight determination module has the following functions for dynamic weight adjustment: if the deviation between real-time data and historical data reaches a preset threshold, the corresponding weight of the potential inducing factor is increased; if the deviation is lower than another preset threshold, the corresponding weight is maintained or decreased.

[0010] Furthermore, the cyanobacterial bloom probability prediction module has the following functions for constructing multiple basic prediction models: establishing multiple machine learning models as basic prediction models, with key inducing factors and dynamic weights as input features and bloom occurrence and non-occurrence labels as output labels, and optimizing the parameters of the basic prediction models through cross-validation.

[0011] Furthermore, the cyanobacterial bloom probability prediction module has the following functions for model integration: the ensemble learning framework is the Stacking ensemble learning framework, the meta-classifier is the logistic regression model, and the meta-classifier integrates the prediction results of the basic prediction model to output the final cyanobacterial bloom probability.

[0012] Furthermore, the toxin-producing risk grading module, which is used for assessing the dominance of toxin-producing algae, includes the following functions: establishing a mapping relationship based on the correlation between dominant algae species and toxin-producing events in historical data; using real-time monitored algal species composition data to provide the cell count of toxin-producing algae, and obtaining the dominance of toxin-producing algae by calculating the proportion of toxin-producing algal cell count to the total number of cyanobacterial cells.

[0013] Furthermore, the toxin production risk classification module has the function of classifying toxin production risk levels by dividing the risk into multiple progressively increasing risk levels based on the combination of indicators such as the probability of algal bloom, the dominance of toxin-producing algal species, and the concentration of algal toxins in the same historical period.

[0014] Furthermore, the result output and decision support module has the following functions for result visualization: displaying the dynamic weight change trend of key inducing factors through line graphs, displaying the spatial distribution of the probability of algal blooms through heat maps, and displaying the toxicity risk level and corresponding assessment indicators through bar charts.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention collects environmental big data, hydrological and meteorological data, and historical algal bloom monitoring data through a multi-source data acquisition module, and integrates the data using a credibility-weighted fusion algorithm in the data preprocessing and fusion module, thus solving the problem of inaccurate multi-source data integration; the key inducing factor identification and dynamic weight determination module screens potential factors through multivariate statistical analysis and association rule mining, and then adjusts the weights based on the deviation between real-time and historical data using a sliding window dynamic weight mechanism, thus solving the problem of the disconnect between static weights and the actual algal bloom driving mechanism in existing technologies; the cyanobacterial bloom probability prediction module constructs a multi-foundation model and fuses the results through a Stacking ensemble learning framework, reducing prediction bias; the toxin production risk classification module calculates the dominance of toxin-producing algal species based on historical mapping relationships and real-time algal species composition, and constructs a two-factor model to classify risk levels based on algal toxin concentration, thus solving the problem of insufficient quantification of toxin production risk; the result output and decision support module visualizes the results and generates prevention and control suggestions, improving the reliability of early warning and the targeting of prevention and control, and meeting the needs of precise prevention and control of the water environment. Attached Figure Description

[0016] Figure 1 is a general diagram of the system architecture of the present invention; Figure 2 is a flowchart of the data preprocessing and fusion process of the present invention; Figure 3 is a diagram of the dynamic weight determination mechanism of the present invention; Figure 4 is a diagram of the Stacking model integration prediction of the present invention; Figure 5 is a diagram of the two-factor risk assessment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please refer to Figures 1-5. This invention provides a system for identifying cyanobacterial bloom inducing factors and assessing toxin risk, including: a multi-source data acquisition module for collecting environmental big data, hydrological and meteorological data, and historical bloom monitoring data; a data preprocessing and fusion module for cleaning and aligning the multi-source data, introducing a data credibility weighted fusion algorithm, assigning credibility weights to different types of data based on the stability of the data sources, and fusing the multi-source data into a unified bloom analysis dataset using a weighted average method; and a key inducing factor identification and dynamic weight determination module for preliminary factor screening based on the bloom analysis dataset, using a combination of multivariate statistical analysis and association rule mining methods to identify potential inducing factors, and introducing a sliding window dynamic weight mechanism. Within each sliding window, the contribution of potential inducing factors to the bloom event is calculated using a feature importance assessment method, and the contribution is dynamically determined by combining the deviation between real-time and historical data. The system dynamically adjusts the weights of potential inducing factors; the cyanobacterial bloom probability prediction module is used to construct multiple basic prediction models, using key inducing factors and dynamic weights as input features, training and optimizing the basic prediction models using historical bloom monitoring data, and employing an ensemble learning framework to use the prediction results of multiple basic prediction models as new input features to construct a meta-classifier to fuse the prediction results, thereby outputting the probability of cyanobacterial bloom occurrence; the toxin production risk classification module is used to establish a mapping relationship between dominant algal species and toxin production probability based on historical data, assess the dominance of toxin-producing algal species by combining real-time monitored algal species composition data, and construct a two-factor risk assessment model based on the dominance of toxin-producing algal species and algal toxin concentration data to classify toxin production risk into multiple risk levels; the result output and decision support module is used to output the bloom probability, toxin production risk level and related assessment results in a visual form, and generate prevention and control recommendations based on the risk level.

[0019] Specifically, the multi-source data acquisition module collects water nutrient indicators in real time through deployed water quality sensors, simultaneously obtains data such as daily rainfall and daily average water temperature from regional hydrological and meteorological stations, and retrieves historical algal bloom monitoring data from the past 5 years to construct a comprehensive data source covering environmental hydrology, meteorology, and historical cases. The data preprocessing and fusion module first uses outlier identification methods to remove outliers from environmental big data, then uses missing data completion methods to complete missing values ​​in hydrological and meteorological data, and finally unifies all data into daily time units. A data credibility weighted fusion algorithm is introduced, and weights are assigned according to the stability of the data source. For example, the weight of data from fixed monitoring stations is set to 0.8, and the weight of temporary sampling data is set to 0.5. The unified algal bloom analysis dataset after fusion is calculated using formula (1): Where D is the unified data value after fusion. Let be the confidence weight of the i-th type of data. is the original value of the i-th type of data, and n is the number of data types involved in the fusion. The key inducing factor identification and dynamic weight determination module first screens parameters with high correlation to algal bloom occurrence through multivariate statistical analysis, and then mines parameter combination relationships using association rules; a 7-day sliding window dynamic weight mechanism is introduced, and the contribution of potential inducing factors is calculated as the initial weight in each window through the feature importance assessment method, and the weight is adjusted by combining the deviation between real-time data and historical data. The deviation is calculated according to formula (2): Where Δ represents the deviation between real-time data and historical data. This represents the real-time monitoring value of potential inducing factors. The historical average value of this factor; weight adjustment is performed according to formula (3): ;in The adjusted weights of potential inducing factors. The initial weights are defined as follows: k is the weight adjustment coefficient, ranging from 0.1 to 0.2. The cyanobacterial bloom probability prediction module constructs three basic models: decision tree, SVM, and neural network. These models are trained using historical bloom monitoring data and optimized through cross-validation. Then, using a stacking ensemble learning framework, the prediction results of the basic models are used as new input features to input into a logistic regression classifier, outputting the probability of cyanobacterial bloom occurrence. The toxin production risk grading module establishes a mapping relationship between dominant algal species and toxin production probability based on historical data. It counts the number of cells of toxin-producing algae and the total number of cyanobacterial cells in real time, and constructs a two-factor risk assessment model based on algal toxin concentration to classify risk levels. The results output and decision support module visualizes the results through line charts, heatmaps, and bar charts, generating prevention and control recommendations for different risk levels. This process addresses the problems of large prediction bias and insufficient quantification of toxin production risk in existing technologies, improving the reliability of early warning and the targeted nature of prevention and control.

[0020] In this embodiment, the multi-source data acquisition module is used to collect environmental big data including nutrient salt indicators, dissolved oxygen, pH value and transparency parameters in water bodies; hydrological and meteorological data including daily rainfall, daily average water temperature, daily average wind speed, sunshine duration and water level change data; historical algal bloom monitoring data including cyanobacterial bloom occurrence records, dominant algal species types and algal toxin concentration detection data.

[0021] Specifically, when collecting environmental big data, the multi-source data acquisition module monitors water nutrient levels in situ using multi-parameter water quality sensors, recording data every 2 hours to ensure timeliness. Dissolved oxygen, pH, and transparency parameters are collected three times daily at fixed points using a portable monitoring instrument, and the average of the three measurements is taken to reduce random errors. Hydrological and meteorological data are retrieved from the regional hydrological and meteorological station database, with daily rainfall accurate to millimeters, daily average water temperature retained to one decimal place, daily average wind speed recorded in meters per second, sunshine duration in hours, and water level change data obtained by comparing the difference between the current day's and the previous day's water levels. Historical algal bloom monitoring data is extracted from environmental monitoring department archives, including the date range and occurrence area of ​​cyanobacterial blooms over the past 10 years. Dominant algal species are identified through microscopic observation, and algal toxin concentrations are detected and recorded using high-performance liquid chromatography. Through comprehensive and accurate multi-source data acquisition, sufficient and reliable basic data are provided for subsequent data fusion and factor identification, avoiding analytical biases caused by single or missing data.

[0022] In this embodiment, the data preprocessing and fusion module has the following functions: data cleaning includes removing outliers from environmental big data using outlier identification and processing methods, and completing missing data in hydrological and meteorological data using missing data completion methods; time-series alignment includes unifying the time dimensions of environmental big data, hydrological and meteorological data, and historical algal bloom monitoring data using a unified time unit.

[0023] Specifically, during data cleaning in the data preprocessing and fusion module, the 3σ rule is used to identify outliers for environmental big data. This method, an existing technology, calculates the mean and standard deviation of the data, and identifies and removes data exceeding the mean plus or minus three times the standard deviation. For missing data in hydrological and meteorological data, linear interpolation is used to fill in missing data if the duration is short, and the mean of data from the same period on three adjacent dates is used to fill in missing data, ensuring data continuity. During time-series alignment, the "day" is used as the unified time unit. Environmental big data is integrated into daily data based on the mean of monitoring values ​​for each time period of the day. Daily statistical data is directly extracted from hydrological and meteorological data, and historical algal bloom monitoring data is mapped to the daily time axis according to the date of occurrence, ensuring complete consistency of the three types of data in the time dimension. When introducing the data credibility weighted fusion algorithm, the weights are assigned according to the stability of the data source. For example, the data credibility of long-term operating automatic monitoring station data is set to 0.9, and the weight of mobile monitoring equipment data is set to 0.6. The multi-source data are fused into a unified dataset through formula (1), which reduces the impact of data noise and time misalignment on subsequent analysis and improves data availability.

[0024] In this embodiment, the key inducing factor identification and dynamic weight determination module has the following functions for preliminary factor screening: first, it analyzes the correlation between each parameter in the algal bloom analysis dataset and the algal bloom occurrence record through correlation analysis method, so as to retain parameters whose correlation reaches a preset threshold; then, it identifies the combined correlation relationship between parameters through association rule mining method, thereby preliminarily determining potential inducing factors.

[0025] Specifically, in the key inducing factor identification and dynamic weight determination module, during the initial factor screening, Pearson correlation coefficient analysis is first used to analyze the correlation between each parameter in the algal bloom analysis dataset and algal bloom records. This method is an existing technology. By calculating the correlation coefficient, it measures the linear correlation between the parameter and algal bloom occurrence. A preset threshold of 0.6 is set, and parameters with correlation coefficients greater than or equal to 0.6, such as water temperature, total phosphorus, and total nitrogen, are retained. Then, the Apriori association rule mining algorithm is used to analyze the retained parameters. This method is also an existing technology. The minimum support is set to 0.2 and the minimum confidence to 0.7 to identify the combined association relationships between parameters. For example, the combination of "water temperature 25-30℃ and total phosphorus > 0.1 mg / L" has a high correlation with algal bloom occurrence, and potential inducing factors are initially determined based on this. Through this two-step screening process, parameters with weak correlation with algal bloom occurrence are eliminated to simplify the analysis dimensions, while the synergistic effects between parameters are also discovered, avoiding the omission of key combined factors and improving the accuracy of potential inducing factor identification.

[0026] In this embodiment, the key inducing factor identification and dynamic weight determination module has the following functions for dynamic weight adjustment: if the deviation between real-time data and historical data reaches a preset threshold, the corresponding weight of the potential inducing factor is increased; if the deviation is lower than another preset threshold, the corresponding weight is maintained or decreased.

[0027] Specifically, when the key inducing factor identification and dynamic weight determination module performs dynamic weight adjustment, it first sets the sliding window duration to 7 days. Within each window, the random forest algorithm is used to calculate the contribution of each potential inducing factor to the algal bloom event, and this contribution is used as the initial weight. Formula (2) is used to calculate the deviation Δ between real-time data and historical data. If the deviation reaches the preset threshold of 0.3, it indicates that the current environmental conditions are significantly different from those in the past, and the weight of the corresponding potential inducing factor needs to be increased. The adjustment is performed according to formula (3). If the deviation is lower than another preset threshold of 0.1, it indicates that the current environment is similar to that in the past, and there is no need to significantly adjust the weight. The original weight is maintained or the weight is reduced by setting k to -0.05 according to formula (3). By dynamically adjusting the weight, the influence of the factor is made to fit the real-time environmental changes, avoiding the problem of static weights being disconnected from the actual algal bloom driving mechanism in the existing technology. This provides more realistic input features for subsequent algal bloom probability prediction and improves prediction accuracy.

[0028] In this embodiment, the cyanobacterial bloom probability prediction module has the function of constructing multiple basic prediction models, including: establishing multiple machine learning models as basic prediction models, with key inducing factors and dynamic weights as input features and bloom occurrence and non-occurrence labels as output labels, and optimizing the parameters of the basic prediction models through cross-validation.

[0029] Specifically, when constructing multiple basic prediction models for the cyanobacterial bloom probability prediction module, three machine learning models—decision tree, support vector machine, and artificial neural network—are established respectively. Key inducing factors and dynamic weights adjusted using formula (3) are used as input features. Historical bloom monitoring data is labeled as follows: "bloom occurred" is marked as 1, and "no bloom occurred" is marked as 0 as the output label. A 5-fold cross-validation method is used to train and optimize the parameters of each basic model. Specifically, the decision tree model optimizes the maximum depth and number of leaf nodes, the support vector machine model optimizes the kernel function parameters, and the artificial neural network model optimizes the number of hidden layer nodes and the learning rate. Through multi-model construction and parameter optimization, each basic model can fully learn the bloom occurrence patterns in historical data, reducing overfitting or underfitting of a single model, improving the predictive ability of a single model for bloom trends, and providing reliable basic prediction results for subsequent ensemble learning.

[0030] In this embodiment, the cyanobacterial bloom probability prediction module has the following functions for model integration: the ensemble learning framework is the Stacking ensemble learning framework, the meta-classifier is the logistic regression model, and the meta-classifier integrates the prediction results of the basic prediction model to output the final cyanobacterial bloom probability.

[0031] Specifically, the cyanobacterial bloom probability prediction module employs a Stacking ensemble learning framework for model integration, which consists of two layers. The first layer is the base model, a decision tree support vector machine artificial neural network model with a weight of 6. Historical data is divided into training and validation sets in a 7:3 ratio. The training set is used to train the base model, and the validation set is used to obtain the prediction results of each base model. The second layer is the meta-classifier, which uses a logistic regression model. The prediction results of the base models are used as new input features, and the meta-classifier is trained using bloom occurrence labels from historical data. After training, the meta-classifier fuses the prediction results of the base models, comprehensively considering the predictive advantages of each model in different scenarios, and outputs the final cyanobacterial bloom probability. Ensemble learning avoids the limitations of a single model, reduces prediction bias caused by insufficient model adaptability, and further improves the accuracy of cyanobacterial bloom probability prediction.

[0032] In this embodiment, the toxin-producing risk classification module has the following functions for assessing the dominance of toxin-producing algae: the mapping relationship is established based on the correlation between dominant algae types and toxin-producing events in historical data; real-time monitored algal composition data is used to provide the cell count of toxin-producing algae, and the dominance of toxin-producing algae is obtained by calculating the proportion of toxin-producing algal cell count to the total number of cyanobacterial cells.

[0033] Specifically, when assessing the dominance of toxin-producing algae in the toxin-producing risk grading module, the system first compiles historical monitoring data from the past 10 years, statistically analyzes the correlation frequency between different dominant algal species and toxin-producing events, and calculates the correlation degree to establish a mapping relationship between dominant algal species and toxin-producing probability. For example, when Microcystis is the dominant algal species, the toxin-producing probability is mapped to high. During real-time monitoring, the number of cells of toxin-producing algae in the water is counted using microscopic counting. At the same time, the total number of cyanobacterial cells was counted. The dominance of the toxin-producing algae was calculated using formula (4). Where P represents the dominance of the toxin-producing algae. The number of cells in the toxin-producing algae species. The total number of cyanobacterial cells is represented. By combining historical mapping relationships with real-time dominance, the proportion of algal species and their numbers is correlated, providing a key quantitative indicator for subsequent toxin production risk classification. This ensures the objectivity and relevance of dominance assessment and avoids the one-sidedness of judging toxin production risk solely based on algal species.

[0034] In this embodiment, the toxin production risk classification module has the function of classifying toxin production risk levels by dividing the risk into multiple progressively increasing risk levels based on the combination of indicators such as the probability of algal bloom, the dominance of toxin-producing algae species, and the concentration of algal toxins in the same historical period.

[0035] Specifically, the toxin production risk grading module comprehensively considers three indicators when classifying toxin production risk levels: the probability of algal blooms, the dominance of toxin-producing algae, and the historical concentration of algal toxins during the same period. Thresholds for low, medium, and high algal bloom probabilities are set at 30% and 70%, respectively, and thresholds for low, medium, and high dominance of toxin-producing algae are set at 20% and 50%, respectively, combined with the criterion of whether the historical concentration of algal toxins during the same period exceeds 0.5 μg / L. Risk levels are classified based on the combination of these three indicators: low probability + low dominance + below 0.5 μg / L = low risk; medium probability + medium dominance + equal to 0.5 μg / L = medium risk; and high probability + high dominance + above 0.5 μg / L = extremely high risk. This multi-indicator combination grading accurately quantifies the differences in toxin production risk under different scenarios, avoiding the limitations of single-indicator assessments and providing a clear basis for the formulation of differentiated prevention and control measures.

[0036] In this embodiment, the result output and decision support module has the following functions for result visualization: displaying the dynamic weight change trend of key inducing factors through a line graph, displaying the spatial distribution of the probability of algal bloom through a heat map, and displaying the toxicity risk level and corresponding assessment indicators through a bar chart.

[0037] Specifically, when visualizing the results in the results output and decision support module, the Matplotlib and Seaborn libraries of Python are used to create charts. Line charts display the dynamic weight changes of key inducing factors over the past 30 days, with the horizontal axis representing the date and the vertical axis representing the weight value, clearly showing the changing patterns of factor influence over time. Heatmaps show the spatial distribution of algal bloom probability, using monitoring areas as grid units and different color gradients to represent probability levels, intuitively reflecting the differences in algal bloom risk in different areas. Bar charts display the toxin production risk level and corresponding assessment indicators, with the horizontal axis representing the risk level and the vertical axis representing the indicator value, clearly comparing the probability of algal bloom occurrence, the dominance of toxin-producing algal species, and the concentration of algal toxins at each level. Simultaneously, prevention and control recommendations are generated based on the risk level, such as monthly regular monitoring for low risk, enhanced nutrient control for medium risk, and daily emergency monitoring and physical algae removal measures for high risk. This ensures that the output results are both intuitive and provide decision-making guidance, offering clear and actionable prevention and control guidelines for water environment management departments.

[0038] In summary, this invention addresses the inaccuracy of multi-source data integration by combining a multi-source data acquisition module that collects environmental big data, hydrological and meteorological data, and historical algal bloom monitoring data with a credibility-weighted fusion algorithm in a data preprocessing and fusion module. A key inducing factor identification and dynamic weight determination module uses multivariate statistical analysis and association rule mining to screen potential factors, then adjusts the weights based on the deviation between real-time and historical data using a sliding window dynamic weighting mechanism, resolving the disconnect between static weights and actual algal bloom driving mechanisms in existing technologies. A cyanobacterial bloom probability prediction module constructs a multi-foundation model and fuses the results using a Stacking ensemble learning framework to reduce prediction bias. A toxin production risk grading module calculates the dominance of toxin-producing algae based on historical mapping relationships and real-time algal species composition, and constructs a two-factor model based on algal toxin concentration to classify risk levels, addressing the insufficient quantification of toxin production risk. A results output and decision support module visualizes the results and generates prevention and control suggestions, improving the reliability of early warnings and the targeting of prevention and control, thus meeting the needs of precise water environment prevention and control.

[0039] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0040] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A system for identifying cyanobacterial bloom inducing factors and assessing toxin risk, characterized in that, include: The multi-source data acquisition module is used to collect environmental big data, hydrological and meteorological data, and historical algal bloom monitoring data; The data preprocessing and fusion module is used to clean and align multi-source data over time, and introduces a data credibility weighted fusion algorithm. It assigns credibility weights to different types of data based on the stability of the data source, and fuses multi-source data into a unified algal bloom analysis dataset through a weighted average method. The key inducing factor identification and dynamic weight determination module is used to perform preliminary factor screening to identify potential inducing factors based on the algal bloom analysis dataset by combining multivariate statistical analysis methods and association rule mining methods. A sliding window dynamic weight mechanism is introduced. Within each sliding window, the contribution of potential inducing factors to algal bloom events is calculated by the feature importance evaluation method. The weights of potential inducing factors are dynamically adjusted by combining the deviation between real-time data and historical data. The cyanobacterial bloom probability prediction module is used to construct multiple basic prediction models. Using key inducing factors and dynamic weights as input features, the basic prediction models are trained and their parameters are optimized using historical algal bloom monitoring data. An ensemble learning framework is used to construct a meta-classifier to fuse the prediction results, thereby outputting the probability of cyanobacterial bloom occurrence. The toxin production risk classification module is used to establish a mapping relationship between dominant algal species and toxin production probability based on historical data, evaluate the dominance of toxin-producing algal species by combining real-time monitored algal species composition data, and construct a two-factor risk assessment model based on the dominance of toxin-producing algal species and algal toxin concentration data to classify toxin production risk into multiple risk levels. The results output and decision support module is used to output the probability of algal blooms, the risk level of toxin production, and related assessment results in a visual form, and generate prevention and control recommendations based on the risk level.

2. The cyanobacterial bloom inducing factor identification and toxin risk classification assessment system according to claim 1, characterized in that, The multi-source data acquisition module is used to collect environmental big data including nutrient indicators, dissolved oxygen, pH value and transparency parameters in water bodies; hydrological and meteorological data including daily rainfall, daily average water temperature, daily average wind speed, sunshine duration and water level change data; historical algal bloom monitoring data including records of cyanobacterial bloom occurrences, dominant algal species types and algal toxin concentration detection data.

3. The cyanobacterial bloom inducing factor identification and toxin risk classification assessment system according to claim 1, characterized in that, The data preprocessing and fusion module has the following functions: data cleaning includes removing outliers from environmental big data using outlier identification and processing methods, and completing missing data in hydrological and meteorological data using missing data completion methods; time-series alignment includes unifying the time dimensions of environmental big data, hydrological and meteorological data, and historical algal bloom monitoring data using a unified time unit.

4. The cyanobacterial bloom inducing factor identification and toxin risk classification assessment system according to claim 1, characterized in that, The key inducing factor identification and dynamic weight determination module has the following functions for preliminary factor screening: first, it analyzes the correlation between each parameter in the algal bloom analysis dataset and the algal bloom occurrence record through correlation analysis method, so as to retain parameters whose correlation reaches a preset threshold; then, it identifies the combined correlation relationship between parameters through association rule mining method, thereby preliminarily determining potential inducing factors.

5. The cyanobacterial bloom inducing factor identification and toxin risk classification assessment system according to claim 4, characterized in that, The key inducing factor identification and dynamic weight determination module has the following functions for dynamic weight adjustment: if the deviation between real-time data and historical data reaches a preset threshold, the corresponding weight of the potential inducing factor is increased; if the deviation is lower than another preset threshold, the corresponding weight is maintained or decreased.

6. The cyanobacterial bloom inducing factor identification and toxin risk classification assessment system according to claim 1, characterized in that, The cyanobacterial bloom probability prediction module has the following functions for constructing multiple basic prediction models: establishing multiple machine learning models as basic prediction models, with key inducing factors and dynamic weights as input features and bloom occurrence and non-occurrence labels as output labels, and optimizing the parameters of the basic prediction models through cross-validation.

7. The cyanobacterial bloom inducing factor identification and toxin risk classification assessment system according to claim 6, characterized in that, The cyanobacterial bloom probability prediction module has the following functions for model integration: the ensemble learning framework is the Stacking ensemble learning framework, the meta-classifier is the logistic regression model, and the meta-classifier integrates the prediction results of the basic prediction model to output the final cyanobacterial bloom probability.

8. The cyanobacterial bloom inducing factor identification and toxin risk classification assessment system according to claim 1, characterized in that, The toxin production risk classification module has the following functions for assessing the dominance of toxin-producing algae: the mapping relationship is established based on the correlation between dominant algae types and toxin production events in historical data; real-time monitored algal composition data is used to provide the cell count of toxin-producing algae, and the dominance of toxin-producing algae is obtained by calculating the proportion of toxin-producing algal cell count to the total number of cyanobacterial cells.

9. The cyanobacterial bloom inducing factor identification and toxin risk classification assessment system according to claim 8, characterized in that, The toxin production risk classification module has the function of classifying toxin production risk levels by dividing the risk into multiple progressively increasing risk levels based on the combination of indicators such as the probability of algal bloom, the dominance of toxin-producing algal species, and the concentration of algal toxins in the same historical period.

10. The system for identifying cyanobacterial bloom inducing factors and assessing toxin risk according to claim 1, characterized in that, The results output and decision support module includes the following functions for results visualization: displaying the dynamic weight change trend of key inducing factors through line graphs, displaying the spatial distribution of the probability of algal blooms through heat maps, and displaying the risk level of toxic production and the corresponding assessment indicators through bar charts.

Images