CN-121981526-A - Heavy metal ecological risk evaluation and treatment technology recommendation method, equipment and medium based on machine learning

CN121981526ACN 121981526 ACN121981526 ACN 121981526ACN-121981526-A

Abstract

The invention discloses a machine learning-based heavy metal ecological risk evaluation and treatment technology recommendation method, equipment and medium, wherein the ecological risk evaluation method comprises the steps of constructing a sample data set, constructing a machine learning-based occurrence form prediction model, taking characteristic vectors formed by heavy metal types and physicochemical properties as input and four occurrence form duty ratios as output, training the occurrence form prediction model by using the sample data set, acquiring the heavy metal types and the physicochemical properties of a sample to be evaluated, constructing the characteristic vectors, inputting the trained occurrence form prediction model, predicting to obtain four occurrence form duty ratios of the sample to be evaluated, calculating a risk evaluation coding method evaluation index RAC based on the four occurrence form duty ratios of the sample to be evaluated, and determining an ecological risk grade according to the evaluation index RAC. The invention improves the efficiency and the intelligent level of the heavy metal pollution ecological risk assessment and treatment decision.

Inventors

WANG HAN
LIN ZHANG
DU CHENGYUAN
SHI YAN
FENG CHENCHEN
REN CHANGHAI

Assignees

中南大学

Dates

Publication Date: 20260505
Application Date: 20251230

Claims (10)

1. The heavy metal ecological risk evaluation method based on machine learning is characterized by comprising the following steps of: Acquiring heavy metal types, physicochemical properties and four occurrence form proportion data, and constructing a sample data set; Constructing an occurrence form prediction model based on machine learning, taking characteristic vectors formed by heavy metal types and physicochemical properties as input and taking the occupancy rate of four occurrence forms as output; Training an occurrence form prediction model by using a sample data set; Obtaining heavy metal types and physicochemical properties of a sample to be evaluated, constructing feature vectors, inputting a trained occurrence form prediction model, and predicting to obtain four occurrence form duty ratios of the sample to be evaluated; And calculating a risk evaluation coding method evaluation index RAC based on the four occurrence form occupation ratios of the sample to be evaluated, and further determining the ecological risk level according to the RAC value.
2. The machine learning-based heavy metal ecological risk assessment method according to claim 1, further comprising, prior to constructing the sample dataset: And screening the characteristic variables by adopting a recursive characteristic elimination method and combining with the pearson correlation coefficient, and reserving the optimal characteristic combination for predicting the occupancy rate of the four occurrence forms to form a characteristic vector.
3. The machine learning-based heavy metal ecological risk assessment method according to claim 1, wherein the physicochemical properties include pH, EC, TOC and total concentration of heavy metals.
4. The heavy metal ecological risk evaluation method based on machine learning according to claim 1, wherein the sample data set comprises a general sample data set and a special sample data set, the general sample data set comprises different types of heavy metal sample data, the special sample data set comprises target heavy metal sample data, when training the occurrence form prediction model, the general sample data set is used for pre-training the occurrence form prediction model based on machine learning, and then the special sample data set is used for fine tuning the pre-trained occurrence form prediction model to obtain a final trained occurrence form prediction model.
5. The heavy metal ecological risk evaluation method based on machine learning according to claim 1 is characterized in that the occurrence form prediction model adopts a potential attention mechanism combined with a Transformer-BiGRU parallel architecture, an input feature vector is divided into two paths of processing, one path is subjected to low-rank compression and feature enhancement through the potential attention mechanism, then a Transformer encoder is input to extract long-term dependency among features, the other path is directly input to BiGRU network to capture time sequence association features, then the two paths of extracted features are fused through a cross attention mechanism and input to an output layer, and finally four occurrence form duty ratios are output through the output layer.
6. The heavy metal ecological risk assessment method based on machine learning according to any one of claims 1 to 5, wherein the calculation method of the risk assessment coding method assessment index RAC is as follows: ; Wherein C F1 is the target heavy metal concentration in the weak acid extractable state, and C T is the target total heavy metal concentration.
7. The heavy metal ecological risk evaluation and treatment technology recommendation method based on machine learning is characterized by comprising the following steps of: acquiring heavy metal types, physicochemical properties, four occurrence form proportion, ecological risk level and optimal treatment technology data, and constructing a sample data set; Constructing an occurrence form prediction model based on machine learning, taking characteristic vectors formed by heavy metal types and physicochemical properties as input and taking the occupancy rate of four occurrence forms as output; constructing a multi-task learning model, constructing the model based on a multi-task learning structure, inputting the occupancy rate of four occupancy modes output by the occupancy mode prediction model, and outputting a physiological risk level and an optimal processing technology; Training an occurrence form prediction model and a multi-task learning model by using a sample data set; The method comprises the steps of obtaining heavy metal types and physicochemical properties of a sample to be evaluated, constructing feature vectors, inputting a trained occurrence form prediction model, predicting to obtain four occurrence form ratios of the sample to be evaluated, inputting the four occurrence form ratios of the sample to be evaluated into a trained multi-task learning model, and outputting ecological risk levels and optimal processing technologies.
8. The machine learning-based heavy metal ecological risk assessment and processing technology recommendation method according to claim 7, wherein the multi-task learning model introduces a multi-task weight self-adaptive learning mechanism, and a loss function is as follows: ; In the formula, Is the total loss; And Classifying the loss and the corresponding weight for the ecological risk level, For the level of true ecological risk, To predict ecological risk level; And The multi-label loss and corresponding weights are predicted for the best processing technique, In order to be truly optimal for the processing technique, To predict the optimal processing technique; The multi-label loss weight is respectively predicted for the ecological risk level classification loss weight and the optimal processing technology, ; Wherein, the For the learning weight of the multi-task learning model, the learning weight is subjected to the Softmax function Normalizing to obtain Ensure that The sum is 1.
9. An electronic device, comprising: a memory having a computer program stored thereon; A processor for loading and executing the computer program to implement the machine learning-based heavy metal ecological risk assessment method according to any one of claims 1 to 6 or the machine learning-based heavy metal ecological risk assessment and processing technology recommendation method according to any one of claims 7 to 8.
10. A computer-readable storage medium, on which a computer program is stored, characterized in that the computer program, when executed by a processor, implements the machine learning-based heavy metal ecological risk assessment method according to any one of claims 1 to 6 or the machine learning-based heavy metal ecological risk assessment and processing technology recommendation method according to any one of claims 7 to 8.

Description

Heavy metal ecological risk evaluation and treatment technology recommendation method, equipment and medium based on machine learning Technical Field The invention relates to the field of intersection of environmental pollution control and machine learning technologies, in particular to a heavy metal ecological risk evaluation and treatment technology recommendation method, equipment and medium based on machine learning. Background With the acceleration of industrialization and urbanization, heavy metal pollution is increasingly serious, and the heavy metal pollution becomes one of the important environmental problems affecting ecological environment safety and human health. Heavy metal pollutants have the characteristics of high toxicity, difficult degradation, easy enrichment and the like, can migrate through a food chain after being accumulated in soil, water and sediment for a long time, and cause far-reaching harm to an ecological system. Especially in the industries of mining, metallurgy, electroplating, electronic manufacturing and the like, a large amount of industrial waste residues, tailings and polluted soil containing heavy metals are generated, and a potential environmental pollution source is formed. Therefore, how to scientifically evaluate the ecological risk of heavy metals and reasonably select applicable treatment technologies becomes a key problem to be solved urgently in the current environmental engineering field. Meanwhile, the traditional heavy metal ecological risk assessment method mainly comprises a single factor index method, an internal Mei Luo comprehensive pollution index method, a Hakanson potential ecological risk index method and the like. The method generally depends on the measurement of total heavy metal, combines an empirical formula to carry out qualitative or semi-quantitative risk classification, lacks deep consideration of heavy metal occurrence forms, biological effectiveness and migration and transformation behaviors, and is difficult to accurately reflect the actual ecological hazard degree. In addition, the existing risk assessment flow often needs a large amount of manual participation and experience judgment, is complex in operation and high in subjectivity, and is difficult to adapt to complex and changeable actual pollution scenes. In addition, in the aspect of governance technology selection, the method mainly depends on expert experience, literature data and analog analysis of engineering cases, and lacks a systematic and quantifiable risk-governance integrated decision mechanism. In the face of different pollution types, pollution degrees and site characteristics, the selection process of the treatment technology often faces the problems of high uncertainty, low efficiency and the like, so that the repair cost is high, the repair effect is unstable, and even secondary pollution can be possibly caused. In recent years, with the development of artificial intelligence and big data technology, the application of machine learning methods in the field of environmental science is gradually increasing. Particularly, the method has good application prospect in the aspects of pollutant identification, pollution source analysis, environment quality prediction and the like. However, the application of the machine learning technology to the research recommended by the heavy metal ecological risk evaluation and treatment technology is still in the preliminary exploration stage. Most of the existing researches lack unified data sources, standardized modeling flows and extensive applicability verification. Meanwhile, most methods do not fully consider the influence of heavy metal occurrence forms on ecological risks and treatment strategies, and the practicality and popularization value of the model are limited. Disclosure of Invention The invention provides a machine learning-based heavy metal ecological risk evaluation and treatment technology recommendation method, equipment and medium, and aims to solve the problems of complex flow, strong subjectivity and low treatment technology selection efficiency of the traditional heavy metal ecological risk evaluation. In a first aspect, a heavy metal ecological risk evaluation method based on machine learning is provided, which comprises the following steps: Acquiring heavy metal types, physicochemical properties and four occurrence form proportion data, and constructing a sample data set; Constructing an occurrence form prediction model based on machine learning, taking characteristic vectors formed by heavy metal types and physicochemical properties as input and taking the occupancy rate of four occurrence forms as output; Training an occurrence form prediction model by using a sample data set; Obtaining heavy metal types and physicochemical properties of a sample to be evaluated, constructing feature vectors, inputting a trained occurrence form prediction model, and predicting to obtain four occurrence form duty ratios of