CN-122024889-A - Method, system, medium and terminal for predicting occurrence form of heavy metal element in waste residue

CN122024889ACN 122024889 ACN122024889 ACN 122024889ACN-122024889-A

Abstract

The invention discloses a method, a system, a medium and a terminal for predicting occurrence forms of heavy metal elements in waste residues, wherein the method comprises the steps of obtaining physicochemical attribute data in the waste residues to be predicted, preprocessing the data to obtain characteristic data, wherein the physicochemical attribute data comprises one or more of pH value, conductivity, element content, waste residue type, organic matter content, phase content and a detection method, inputting the characteristic data into a self-adaptive model for constructing and completing training target heavy metal occurrence form prediction to predict, and outputting to obtain occurrence forms of the target heavy metal elements. The method introduces similar element occurrence form data with similar physicochemical properties in the same group, same period and diagonal positions of the target heavy metal elements in the periodic table, and successfully establishes a prediction model of the occurrence form of the heavy metal in the waste residue with high prediction precision and strong generalization capability through a migration learning method.

Inventors

WANG HAN
LIN ZHANG
DU CHENGYUAN
LI YIN
QI CHONGCHONG
LIN LE

Assignees

中南大学

Dates

Publication Date: 20260512
Application Date: 20251224

Claims (10)

1. The method for predicting the occurrence form of the heavy metal element in the waste residue is characterized by comprising the following steps of: Obtaining physical and chemical attribute data in waste residues to be predicted, and carrying out data preprocessing to obtain characteristic data, wherein the physical and chemical attribute data comprise one or more of pH value, conductivity, element content, waste residue type, organic matter content, phase content and detection method; The characteristic data is used as input, the characteristic data is input into a self-adaptive model for predicting the occurrence form of the target heavy metal, which is constructed and trained, the occurrence form of the target heavy metal is obtained through prediction, the occurrence form comprises four occurrence forms, namely a weak acid extraction form, a reducible form, an oxidizable form and a residue form, and the self-adaptive model for predicting the occurrence form of the target heavy metal is obtained through training of a deep neural network model by utilizing physical and chemical attribute data of the target heavy metal and similar elements in the same group, same period and diagonal positions of the target heavy metal and a data set constructed by the corresponding occurrence form.
2. The method of claim 1, wherein the data preprocessing includes filling in empty values in the data set, numerical converting non-numerical features, data normalization, and rejecting redundant features using correlation analysis methods.
3. The method of claim 1, wherein the adaptive model of target heavy metal occurrence morphology prediction comprises a shared encoder, a domain-specific encoder, a feature decoupling module, an output predictor; The shared encoder is used for extracting the feature data obtained after the pretreatment and converting the feature data into high-dimensional domain invariant features, wherein the domain invariant features are common rule features of target heavy metal elements and similar elements in the same family, same period and diagonal positions with the target heavy metal elements; The domain specific encoder comprises a target domain encoder and a source domain encoder, wherein the target domain encoder is used for extracting the specific characteristics of target heavy metal elements; the characteristic decoupling module is used for separating the domain invariant characteristic extracted by the shared encoder and the specific characteristic extracted by the domain specific encoder through orthogonal constraint, and fusing to obtain fused characteristics; the domain discriminator is used for performing countermeasure training with the shared encoder by using the gradient inversion layer during model training; The output predictor is used for mapping the fusion characteristics obtained by the characteristic decoupling module into a duty ratio prediction result of the occurrence form.
4. The method of claim 3 wherein the shared encoder is comprised of an input layer and 3 fully connected layers, each fully connected layer followed by a ReLU activation function and Dropout layer, wherein the number of neurons in the input layer is the same as the characteristic dimension of the characteristic data, the domain-specific encoder is a two-layer fully connected layer followed by a ReLU activation function, the domain arbiter is comprised of a three-layer bi-classified neural network, and the output predictor is comprised of a fully connected layer and an output layer.
5. The method according to claim 4, wherein the loss function of the orthogonal constraint in the feature decoupling module is specifically: ; wherein, among them, For shared encoder output; Is the domain specific encoder output.
6. The method of claim 1, wherein the training process of the target heavy metal element occurrence form prediction model is specifically: Obtaining physical and chemical attribute data and corresponding occurrence form data of waste residues of target heavy metal elements and similar elements in the same group, same period and diagonal positions of the target heavy metal elements in the periodic table of chemical elements; Filling the blank value in the data set by using a KNN blank value filling algorithm, converting non-numerical characteristics by adopting single-heat coding, further carrying out standardization processing on all data, and carrying out oversampling processing on standardized target heavy metal elements and corresponding occurrence forms to expand the sample size to a preset proportion number; Freezing a target domain specific encoder and a domain discriminator, inputting physical and chemical attribute data of similar elements into a network, and training only a shared encoder, a source domain specific encoder and an output predictor to obtain a pre-training model, wherein an optimizer is an Adam optimizer in the training process, and a loss function is cross entropy loss of a predicted value; In the fine tuning stage, the shared encoder can be trained to support alignment of the opposite domains, the target domain specific encoder and the output predictor both participate in updating, the physical and chemical attribute data of the target heavy metal element and the physical and chemical attribute data of the similar element are input into the model in a combined way, and the self-adaptive model suitable for prediction of the heavy metal occurrence form of the target domain is finally obtained by optimizing only task loss, opposite domain discrimination loss and characteristic orthogonal constraint of the source domain; Evaluating the self-adaptive model for predicting the occurrence form of the target heavy metal based on the test set data, and finishing training when the evaluation result meets the preset condition, otherwise continuing training, wherein the decision coefficient, the average absolute error and the root mean square error are adopted as evaluation indexes; And analyzing and identifying key influence factors through marginal contribution average values, and determining the key influence degree of each feature on model prediction, wherein the key influence factors are the input features of the preset quantity of marginal contribution average values which are ordered in descending order.
7. The method of claim 6, wherein a dynamic weight self-adaptive mechanism is adopted in the training process, the weight of the similar element and the target heavy metal element in the total loss is dynamically adjusted according to the loss ratio of the previous epoch, the key parameters of the model are optimized through a Bayesian optimization super-parameter search strategy, wherein the key parameters comprise the learning rate, the network layer number and the batch size, and the stability of the model is ensured by using five-fold cross-validation.
8. A system for predicting the occurrence of heavy metal elements in waste residues, the system being adapted to perform the method of any one of claims 1 to 7, comprising: the data acquisition module is used for acquiring physical and chemical attribute data in the waste residue to be predicted and carrying out data preprocessing to obtain characteristic data, wherein the physical and chemical attribute data comprises one or more of pH value, conductivity, element content, waste residue type, organic matter content, phase content and detection method; The result prediction module is used for inputting the characteristic data into a self-adaptive model for predicting the occurrence form of the target heavy metal, which is constructed and trained, and outputting the characteristic data to obtain the occurrence form of the target heavy metal element, wherein the occurrence form comprises a weak acid extraction state, a reducible state, an oxidizable state and a residue state, and the self-adaptive model for predicting the occurrence form of the target heavy metal is obtained by training a deep neural network model by utilizing physical and chemical attribute data of the target heavy metal and similar elements in the same group, same period and diagonal positions of the target heavy metal element and a data set constructed by the corresponding occurrence form.
9. A readable storage medium, characterized in that a computer program is stored, which computer program, when being called by a processor, performs the steps of the method according to any of the claims 1-7.
10. An electronic terminal comprising a processor and a memory, said memory storing a computer program, said processor invoking said computer program to perform the steps of the method according to any of claims 1-7.

Description

Method, system, medium and terminal for predicting occurrence form of heavy metal element in waste residue Technical Field The invention relates to the technical field of intersection of environmental protection and machine learning, in particular to a method, a system, a medium and a terminal for predicting occurrence forms of heavy metal elements in waste residues. Background Heavy metals are a class of potentially biotoxic metals or metalloid elements having a density greater than 5 g/cm3, such As lead (Pb), cadmium (Cd), mercury (Hg), arsenic (As), chromium (Cr) and the like. The background content of the heavy metals in natural environment is low, but with the aggravation of the activities such as industrialization, mining, smelting, agricultural chemical fertilizers, sewage irrigation and the like, a large amount of heavy metals are released into soil, water and atmosphere, so that the environment pollution is widely and permanently caused. Heavy metals are difficult to degrade, are easy to enrich through food chains and finally enter human bodies, and can cause serious health risks such as nervous system injury, liver and kidney dysfunction, carcinogenesis, teratogenesis and the like. Of particular concern, the morphology of heavy metals in the environment (e.g., weak acid extractable, reducible, oxidizable, residual, etc.) directly determines their mobility, bioavailability, and toxicity. Therefore, the method accurately identifies and predicts the occurrence form of the heavy metal, and has important significance for evaluating ecological risks, formulating repair strategies and guaranteeing environmental safety. In recent years, an intelligent modeling method combining machine learning and physical and chemical properties provides a new approach for predicting the heavy metal morphology under trans-regional and multi-source heterogeneous conditions. The prior art is difficult to meet the urgent need of heavy metal pollution treatment, and needs to be further improved. And the occurrence form research is significant for the development of processing technology. The proportion of different occurrence forms of heavy metals in waste residues determines the selection and combination of different separation and recovery methods. Accurate prediction of occurrence forms is a precondition for scientific treatment of heavy metals in waste residues. The traditional experimental method aims at researching the occurrence form of heavy metals in waste residues, only can analyze the occurrence form under specific conditions, and systematic research of the overall occurrence form rule of the heavy metals in the waste residues is difficult to guide. The traditional experimental method has obvious limitation in the research of heavy metal occurrence morphology system in waste residues. With the continuous development of machine learning technology, machine learning is also a powerful tool for researching the occurrence form of heavy metals in waste residues. The machine learning has strong data processing capability, can establish the action relation between each influence factor of the system and the target object, and even explore new influence or action mechanism. Provides a scientific method for the systematic research of the occurrence form of heavy metals in waste residues. However, the quantity of heavy metal occurrence morphology research documents in waste residues is small, data acquisition is difficult, the existing data quantity is small, machine learning modeling is difficult for predicting the heavy metal occurrence morphology in conventional waste residues, model precision is not ideal, and model application difficulty is high. Accordingly, in order to solve the problems of small amount of heavy metal data, difficult modeling and insufficient prediction stability in the above waste residues, there is a need for a method for predicting the occurrence form of heavy metal in waste residues, which can improve the accuracy and application degree of the model. Disclosure of Invention Aiming at the technical problems in the background technology, the invention provides a method, a system, a medium and a terminal for predicting the occurrence form of heavy metal elements in waste residues, wherein the method introduces similar element occurrence form data with similar physicochemical properties with the same group, same period and diagonal positions of target heavy metal elements in the periodic table of elements, and successfully establishes a prediction model of the occurrence form of the heavy metal elements in the waste residues with high prediction precision and strong generalization capability through a migration learning method. In a first aspect, the invention provides a method for predicting occurrence forms of heavy metal elements in waste residues, which comprises the following steps: Obtaining physical and chemical attribute data in waste residues to be predicted, and carrying out data preprocess