A method for assessing the pollution risk level of microplastics treatment

Abstract

The present invention discloses a method for assessing the pollution risk level of microplastics, which belongs to the technical field of pollution assessment, and specifically comprises: collecting water samples of microplastic pollutants from a water area to be tested, recording parameters of dangerous pollutants, and when it is detected that the parameters of the pollutants are greater than the set standard, marking the water area to be tested as a high-risk water body; when the pollution parameters are lower than the set standard, collecting the parameters of the polluted water body and a text description of the state of the polluted water body at fixed intervals; inputting the text description into an entity recognition model, and outputting keywords related to microplastic pollution; constructing a pollution level judgment model, extracting dangerous pollutant parameters and keywords on a time series and inputting them into the pollution level judgment model, and outputting the corresponding pollution level; generating a curve of pollution level change over time, and if the growth trend is significant, the water area to be tested is also marked as a high-risk water body; the present invention provides strong data support for environmental protection.

Description

Technical Field

[0001] The present invention relates to the technical field of pollution assessment, and in particular to a method for assessing the pollution risk level of microplastic processing. Background Art

[0002] In the current context of environmental protection and water pollution control, the problem of microplastic pollution has increasingly attracted widespread attention. Microplastics refer to plastic particles with a diameter of less than 5 mm. Due to their tiny size and widespread existence, they pose a serious threat to the water environment and ecosystem. Therefore, accurately assessing the risk level of microplastic pollution is of great significance for formulating corresponding control measures and protecting the ecological environment.

[0003] At present, the assessment methods for microplastic pollution mainly include the steps of collecting samples, detecting the presence and parameters of microplastics, and judging the types and quantities of pollutants. However, the existing assessment methods have certain limitations. First, the parameter detection of dangerous pollutants usually only considers a single factor, which cannot fully reflect the complexity and diversity of microplastic pollution. Secondly, the judgment of pollutant levels mainly relies on manual experience and subjective judgment, which lacks objectivity and accuracy. In addition, the existing assessment methods cannot realize dynamic monitoring and trend analysis of pollution changes. When the pollution of water bodies is in the development stage, it is difficult to timely discover water bodies with risks and take corresponding treatment measures. Therefore, the present invention aims to provide a method for assessing the pollution risk level of microplastic treatment to quantify the development level trend of water pollution. Summary of the invention

[0004] The purpose of the present invention is to provide a method for assessing the pollution risk level of microplastics treatment to solve the following technical problems:

[0005] The existing technology mainly relies on manual experience and subjective judgment to judge the level of pollutants, which lacks objectivity and accuracy. It is impossible to realize dynamic monitoring and trend analysis of pollution changes. When the pollution of water bodies is in the development stage, it is difficult to timely discover water bodies with risks and take corresponding treatment measures.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A method for assessing the pollution risk level of microplastics treatment comprises the following steps:

[0008] Collect water samples of microplastic pollutants from the waters to be tested, and detect whether there are microplastic samples that belong to the list of dangerous pollutants. If so, record the parameters of the dangerous pollutants, including type, area and quantity. When the parameters of the detected pollutants are greater than the set standards, mark the waters to be tested as high-risk waters;

[0009] When the pollution parameter is lower than the set standard, the parameters of the polluted water body and the text description of the state of the polluted water body are collected at fixed intervals; the text description is input into the entity recognition model, and keywords related to microplastic pollution are output;

[0010] A pollution level judgment model is constructed based on a convolutional neural network and a recurrent neural network, the hazardous pollutant parameters and the keywords on the time series are extracted and input into the pollution level judgment model, and the corresponding pollution level is output;

[0011] When the pollution level is greater than the preset level, a curve showing the change of pollution level over time is generated. When the curve is monotonically increasing, the increasing slope is calculated. If the slope is greater than the preset threshold, the water area to be tested is also marked as a high-risk water body.

[0012] As a further solution of the present invention: the pollution level is divided into 10 levels according to the parameters of microplastics and the impact on the ecological environment.

[0013] As a further solution of the present invention: the process of the entity model includes:

[0014] The text description is segmented through a standard dictionary, the text description is split into several words, the words are converted into vectors and then input into an entity recognition model, the entity recognition model includes a BERT layer, a BiLSTM layer and a CRF layer, the words are mapped into word vectors through the BERT layer, the word vectors are input into the BiLSTM layer, the context feature information of any word vector is deeply learned, the probability of each word vector corresponding to the entity is output, an entity sequence is generated, the entity sequence is input into the CRF layer for sorting, and an entity sequence that optimizes the objective function is obtained, and the optimized entity sequence is screened based on a preset microplastic pollution dictionary to screen out valid entities that are consistent with microplastic pollution.

[0015] As a further solution of the present invention: the process of outputting keywords is:

[0016] The keyword entities in the pollution level standard and the valid entities in the text description are taken as nodes of the knowledge graph, and the relationships between entities are taken as edges. The spatial relationships and feature dependencies of the knowledge graph are captured through the learning of graph neural networks to obtain the embedded representation vector of each node. The embedding representation vectors of the entities in the level standard and the text description are taken as the input of the model. The matching degree between the level standard and the text description is measured by calculating the similarity between the vectors, and the reachability and correlation between entities through different paths are obtained based on the path information. The additional information of the direct association between entities is obtained based on the associated entity information, and the matching scores of the valid entities and the keyword entities are output. The keyword entities with matching scores greater than the preset value are output.

[0017] As a further solution of the present invention: the training process of the convolutional neural network is:

[0018] The keywords are converted into feature vectors, the time series corresponding to the feature vectors are obtained, and the keyword matrix formed on the time series is regarded as an image, which is respectively input into the convolutional neural network. The convolutional neural network includes a plurality of convolutional neural layers, each of which includes a plurality of filters. Whenever the matrix image passes through a convolutional neural layer, the filter will slide and scan the matrix and generate a multidimensional array, reduce the size of the matrix image, and increase the dimension of the matrix image. The increased dimension is the same as the number of filters in the convolutional neural network. Each convolutional neural layer reduces the image size and increases the dimension of the matrix image in turn. The multidimensional array finally generated by the convolutional neural network is flattened and converted into a vector, and the vector is used as the input of the recurrent neural network.

[0019] As a further solution of the present invention: the training process of the recurrent neural network is:

[0020] The dimension of the obtained matrix image is kept unchanged, the matrix image is flattened in the time dimension and the eigenvector, a new matrix is ​​generated from the final multidimensional arrays of all convolutional neural layers, and the new matrix vector is used as the input of the long short-term memory neural network in the recurrent neural network as the time dimension of the long short-term memory neural network. The recurrent neural network trains the new matrix by the time dimension and the eigenvector, and outputs the corresponding pollutant level.

[0021] As a further solution of the present invention: a pollution description template is preset, and the text description is filled in according to the pollution description template.

[0022] Beneficial effects of the present invention:

[0023] Through integrated detection and evaluation technology, the present invention can quickly and accurately identify and quantify microplastic pollutants in water bodies, significantly improving the efficiency and accuracy of analysis compared to traditional manual detection and evaluation methods; when the pollution level of the waters to be tested has not yet reached the significant standard, through periodic sample collection and real-time data analysis, the changes in pollution levels are dynamically monitored and pollution trends are discovered in a timely manner. By using advanced deep learning models and graph neural networks, it is possible to deeply mine and understand complex data, extract key information, and enhance the understanding and judgment of pollution data; the combined use of keywords and knowledge graphs strengthens the semantic understanding of pollution description information and increases the dimension and depth of pollution analysis; the new model architecture that combines convolutional neural networks and recurrent neural networks improves the model's ability to process time series data and is suitable for complex environmental data analysis; thereby providing strong data support for environmental protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The present invention will be further described below in conjunction with the accompanying drawings.

[0025] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] See also Figure 1 As shown, the present invention is a method for assessing the pollution risk level of microplastics treatment, comprising the following steps:

[0028] 1. Sample collection and preliminary testing: First, water samples are systematically collected from the waters to be tested. These samples may contain microplastic particles of various sizes and types. In the laboratory, the samples are preliminarily tested using microscopes and other equipment to identify microplastic pollutants and record key parameters such as type, area, and quantity.

[0029] 2. Identification of hazardous pollutants: Compare the detected microplastic samples with the known list of hazardous pollutants to confirm whether there are high-risk microplastic pollutants. For pollutants in the list, record their relevant parameters in detail to provide data support for subsequent risk assessment.

[0030] 3. Preliminary risk assessment: Compare the recorded pollutant parameters, such as concentration, type and toxicity, with the preset safety standards. If the standards are exceeded, the water area will be directly identified as a high-risk water body and corresponding treatment measures need to be taken immediately.

[0031] 4. Regular monitoring and data collection: For waters that do not meet high-risk standards, a regular monitoring plan is developed to continuously track changes in microplastic pollution. In addition to quantitative parameter measurements, text descriptions of the state of polluted water bodies are also collected, such as color, odor, biological activity, etc.

[0032] 5. Text analysis and keyword extraction: Natural language processing technology, especially entity recognition models, are used to analyze the collected text descriptions and extract keywords related to microplastic pollution, such as "plastic fragments" and "fibers". These keywords help to further understand the nature and trends of pollution.

[0033] 6. Construct a pollution level judgment model: Based on convolutional neural networks (CNN) and recurrent neural networks (RNN), especially long short-term memory networks (LSTM), a pollution level judgment model that can process time series data is constructed. CNN is used to extract spatial features, while RNN / LSTM is responsible for capturing dynamic changes in time series.

[0034] 7. Feature extraction and model input: The hazardous pollutant parameters and extracted keywords in the time series are integrated and input into the pollution level judgment model as feature vectors. In this way, the model not only considers the quantitative information of pollutants, but also incorporates the contextual information in the text description.

[0035] 8. Output pollution level: After training, the model can predict the current pollution level based on the input features and provide managers with intuitive risk assessment results.

[0036] 9. Trend analysis and early warning: When the pollution level exceeds the preset threshold, the pollution level can be plotted over time to intuitively show the development trend of the pollution. If the curve shows a monotonically increasing trend with a large slope, it means that the pollution situation is rapidly deteriorating. At this time, an early warning signal should be issued and the water area should be considered to be upgraded to a high-risk area.

[0037] The present invention detects the parameters of dangerous pollutants and inputs keywords related to microplastic pollution into the entity recognition model output, and constructs a pollution level judgment model based on a convolutional neural network and a recurrent neural network, which can accurately assess the pollution risk level of the water area to be tested. When the pollution parameters are lower than the set standard, the parameters of the polluted water body and the text description of the state of the polluted water body are collected at fixed intervals, and a curve of the change of the pollution level over time is generated. When the curve is monotonically increasing, the slope of the increase is calculated. If the slope is greater than the preset threshold, the water area to be tested is marked as a high-risk water body, thereby realizing dynamic monitoring of pollution changes.

[0038] In another preferred embodiment of the present invention, the pollution level is divided into 10 levels according to the parameters of microplastics and the impact on the ecological environment. Subdividing the pollution level into multiple levels helps to understand the degree of pollution in more detail and provide a basis for formulating targeted pollution control strategies and governance measures.

[0039] In another preferred embodiment of the present invention, the process of the entity model includes:

[0040] 1. Text preprocessing and word segmentation

[0041] Standard dictionary construction: First, a standard dictionary containing various microplastic-related vocabulary is constructed. These vocabulary may include different types of microplastics (such as polyethylene, polypropylene, etc.), pollution-related terms (such as "pollution" and "particles", etc.), and vocabulary related to environmental conditions (such as "turbidity" and "odor", etc.).

[0042] Word segmentation: Use natural language processing technology to perform word segmentation on the text description of the polluted water status collected. Split complete sentences or paragraphs into separate words or phrases for subsequent processing and analysis.

[0043] 2. Vector Representation

[0044] Vectorization conversion: Convert the segmented words into numerical vectors. This step is usually implemented through an embedding layer, such as using a pre-trained word vector model (such as Word2Vec, GloVe) or a context-aware embedding model (such as BERT).

[0045] 3. Entity Recognition Model Architecture

[0046] BERT layer: Input the word vector into the BERT layer. BERT is a powerful transformer model that can effectively capture the contextual information of words. It uses a self-attention mechanism to understand the meaning of each word in its context.

[0047] BiLSTM layer: Pass the output of the BERT layer to a bidirectional long short-term memory network layer (BiLSTM). BiLSTM can learn the temporal dependencies of sequence data from two directions (forward and backward) to better capture contextual features in the text.

[0048] CRF layer: Finally, the output of the BiLSTM layer is fed into the Conditional Random Field (CRF) layer. CRF is a statistical modeling method used to label and segment sequence data. It takes into account the transition probability between labels and can achieve global optimality while ensuring local optimality.

[0049] 4. Entity sequence generation and optimization

[0050] Entity probability calculation: Through the above model structure, the probability of each word in the text belonging to a specific entity is calculated.

[0051] Entity sequence generation: Based on these probabilities, a preliminary entity sequence is generated, which contains potential microplastic pollution-related entities.

[0052] Sequence optimization: The characteristics of the CRF layer are used to sort and optimize the entity sequence to ensure that the final entity sequence is semantically coherent and consistent with the actual context.

[0053] 5. Screening of microplastic pollution entities

[0054] Screening based on preset dictionaries: Match the optimized entity sequence with the pre-defined microplastic pollution dictionary to screen out valid entities directly related to microplastic pollution.

[0055] Result output: Output these valid microplastic pollution entities to provide key basis for subsequent pollution level judgment.

[0056] The present invention performs word segmentation on text descriptions through a standard dictionary, converts words into vectors and inputs them into an entity recognition model. It uses a BERT layer, a BiLSTM layer, and a CRF layer for deep learning, outputs the probability that each word vector corresponds to an entity, and performs optimized sorting and screening, thereby improving the accuracy of entity recognition.

[0057] In another preferred embodiment of the present invention, the process of outputting keywords is:

[0058] 1. Constructing the knowledge graph: First, the keyword entities in the pollution level standard and the valid entities in the text description are used as nodes of the knowledge graph. These nodes represent the key concepts and terms related to microplastic pollution. Next, the relationships between entities are determined and added to the knowledge graph as edges. These edges reflect the connections between different entities, such as causal relationships, attribute associations, etc.

[0059] 2. Graph Neural Network Learning: Use Graph Neural Network (GNN) to conduct in-depth learning of knowledge graphs. GNN can capture the spatial relationships and feature dependencies of knowledge graphs, that is, the mutual influence and interaction between different entities. Through message passing and aggregation mechanisms, GNN updates the state of each node and obtains the embedding representation vector of each node. These embedding vectors contain rich information about the entity and can reflect its context and role in the knowledge graph.

[0060] 3. Model input and matching calculation: The embedding representation vectors of the entities in the grade standard and text description are used as the input of the model. By calculating the similarity between the vectors, the matching degree between the grade standard and the text description can be measured. This matching not only takes into account the similarity of the entities themselves, but also their structure and position relationship in the knowledge graph.

[0061] 4. Path information analysis: Based on path information, we can obtain the accessibility and relevance between entities through different paths. In the knowledge graph, there may be multiple paths between two entities, and each path reflects a specific association method or relationship type. By analyzing these paths, we can have a more comprehensive understanding of the complex relationships between entities.

[0062] 5. Related entity information acquisition: Based on the related entity information, additional information about the direct association between entities is obtained. In addition to path information, the related entities themselves may also contain valuable information. For example, if two entities often appear together or have similar attributes, there may be some direct association between them. By mining this related entity information, the understanding of entity relationships can be further enriched.

[0063] 6. Matching score calculation and output: Based on the above analysis results, the matching scores of valid entities and keyword entities are calculated. This score comprehensively considers the similarity of the entities themselves, the structure and position relationship in the knowledge graph, as well as the path information and associated entity information between them. Finally, the keyword entities with matching scores greater than the preset value are output. These keyword entities are key concepts or terms that are highly related to microplastic pollution and are clearly reflected in the text description.

[0064] The present invention uses keyword entities in the pollution level standard and valid entities in the text description as nodes of the knowledge graph, captures the spatial relationships and feature dependencies of the knowledge graph through the learning of the graph neural network, obtains the embedded representation vector of each node, and measures the matching degree between the level standard and the text description by calculating the similarity between the vectors, thereby enhancing the keyword matching capability.

[0065] In another preferred embodiment of the present invention, the training process of the convolutional neural network and the recurrent neural network is:

[0066] Keyword feature vector conversion: Before starting training, the keywords in the text description need to be converted into feature vectors in numerical form. These feature vectors can capture the semantic information of each keyword and provide input for subsequent deep learning models. Through natural language processing technology (such as word embedding methods), each keyword is mapped to a point in a high-dimensional space, which represents the semantic features of the keyword.

[0067] Constructing a time series matrix: In order to capture the changing rules in the time dimension, all feature vectors in the same time period are arranged in chronological order to form a two-dimensional matrix. This matrix not only contains the pollution status information at each time point, but also implies the dynamic characteristics of pollution changes over time.

[0068] Convolutional neural network structure design: Design a convolutional neural network with multiple convolutional layers. Each convolutional layer contains several filters, which are responsible for extracting local features from the input matrix. As the matrix image passes through each convolutional layer, the filter will slide and scan the entire matrix to generate a new multidimensional array. This new array not only reduces the size of the original matrix, but also increases the depth (i.e., the number of dimensions) of the matrix. This operation helps to extract more abstract feature representations from the original data.

[0069] Layer-by-layer feature extraction and dimensionality reduction: Each convolutional layer performs a feature extraction and dimensionality reduction operation on the input matrix. By stacking multiple such convolutional layers, more and more complex features can be gradually extracted from the original data. At the same time, since each layer increases the depth of the matrix, the resulting multidimensional array will contain rich hierarchical feature information.

[0070] Flatten and convert to a one-dimensional vector: In order to connect the output of the convolutional neural network with the subsequent recurrent neural network, the final multidimensional array needs to be flattened into a one-dimensional vector. This vector will be used as the input of the recurrent neural network for further time series analysis.

[0071] Combining recurrent neural networks with long short-term memory networks: Long short-term memory networks (LSTMs) are used to process time series data. The flattened vectors are used as input to the LSTM, while the dimensions of the matrix image remain unchanged, and the new matrix is ​​trained in the time dimension. LSTM can effectively capture long-term dependencies in time series data, thereby more accurately predicting future pollution levels.

[0072] Fusion of time dimension and feature vector: During the LSTM training process, the time dimension and feature vector are fused together to form a new matrix. This matrix not only contains the feature information of each time point, but also takes into account the continuity and change trend in the time dimension.

[0073] Output pollution level: After LSTM training, the model will be able to output the corresponding pollutant level based on the input time series data. This level reflects the severity of the current pollution situation and provides an important reference for environmental protection and management.

[0074] Model optimization and adjustment: In practical applications, it may be necessary to optimize and adjust the model according to specific scenarios and requirements. For example, the accuracy and generalization ability of the model can be improved by adjusting the hyperparameters of the convolutional neural network and LSTM; or other types of neural network structures can be introduced to further improve the model performance.

[0075] Real-time monitoring and early warning: Once the model is deployed and running, real-time monitoring and early warning of microplastic pollution in the waters to be tested can be achieved. When the pollution level exceeds the preset threshold, the system can automatically issue an alarm and take corresponding countermeasures, thereby effectively protecting the safety of the water environment.

[0076] In another preferred embodiment of the present invention, in order to improve the standardization and consistency of the text description, a pollution description template is preset. The text description will be filled in according to the pollution description template. The pollution description template contains multiple fields, such as pollutant type, quantity, color, odor, biological activity, etc. Each field has a clear filling guide and example. The template uses standardized terms and units to ensure that the descriptions between different observers are comparable.

[0077] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for assessing the pollution risk level of microplastics, characterized in that: The following steps are involved: Collect water samples of microplastic pollutants from the waters to be tested, and detect whether there are microplastic samples that belong to the list of dangerous pollutants. If so, record the parameters of the dangerous pollutants, including type, area and quantity. When the parameters of the detected pollutants are greater than the set standards, mark the waters to be tested as high-risk waters; When the pollution parameter is lower than the set standard, the parameters of the polluted water body and the text description of the state of the polluted water body are collected at fixed intervals; the text description is input into the entity recognition model, and keywords related to microplastic pollution are output; A pollution level judgment model is constructed based on a convolutional neural network and a recurrent neural network, the hazardous pollutant parameters and the keywords on the time series are extracted and input into the pollution level judgment model, and the corresponding pollution level is output; When the pollution level is greater than the preset level, a curve showing the change of pollution level over time is generated. When the curve is monotonically increasing, the increasing slope is calculated. If the slope is greater than the preset threshold, the water area to be tested is also marked as a high-risk water body.

2. A method for assessing the risk level of microplastics processing pollution according to claim 1, characterized in that: The pollution level is divided into 10 levels according to the parameters of microplastics and their impact on the ecological environment.

3. A method for assessing the risk level of microplastics processing pollution according to claim 1, characterized in that: The process of the entity recognition model includes: The text description is segmented through a standard dictionary, the text description is split into several words, the words are converted into vectors and then input into an entity recognition model, the entity recognition model includes a BERT layer, a BiLSTM layer and a CRF layer, the words are mapped into word vectors through the BERT layer, the word vectors are input into the BiLSTM layer, the context feature information of any word vector is deeply learned, the probability of each word vector corresponding to the entity is output, an entity sequence is generated, the entity sequence is input into the CRF layer for sorting, and an entity sequence that optimizes the objective function is obtained, and the optimized entity sequence is screened based on a preset microplastic pollution dictionary to screen out valid entities that are consistent with microplastic pollution.

4. A method for assessing the risk level of microplastics processing pollution according to claim 1, characterized in that: The process of outputting keywords is: The keyword entities in the pollution level standard and the valid entities in the text description are taken as nodes of the knowledge graph, and the relationships between entities are taken as edges. The spatial relationships and feature dependencies of the knowledge graph are captured through the learning of graph neural networks to obtain the embedded representation vector of each node. The embedding representation vectors of the entities in the level standard and the text description are taken as the input of the model. The matching degree between the level standard and the text description is measured by calculating the similarity between the vectors, and the reachability and correlation between entities through different paths are obtained based on the path information. The additional information of the direct association between entities is obtained based on the associated entity information, and the matching scores of the valid entities and the keyword entities are output. The keyword entities with matching scores greater than the preset value are output.

5. A method for assessing the risk level of microplastics processing pollution according to claim 1, characterized in that: The training process of the convolutional neural network is: The keywords are converted into feature vectors, the time series corresponding to the feature vectors are obtained, and the keyword matrix formed on the time series is regarded as an image, which is respectively input into the convolutional neural network. The convolutional neural network includes a plurality of convolutional neural layers, each of which includes a plurality of filters. Whenever the matrix image passes through a convolutional neural layer, the filter will slide and scan the matrix and generate a multidimensional array, reduce the size of the matrix image, and increase the dimension of the matrix image. The increased dimension is the same as the number of filters in the convolutional neural network. Each convolutional neural layer reduces the image size and increases the dimension of the matrix image in turn. The multidimensional array finally generated by the convolutional neural network is flattened and converted into a vector, and the vector is used as the input of the recurrent neural network.

6. A method for assessing the risk level of microplastics processing pollution according to claim 5, characterized in that: The training process of the recurrent neural network is: The dimension of the obtained matrix image is kept unchanged, the matrix image is flattened in the time dimension and the eigenvector, a new matrix is ​​generated from the final multidimensional arrays of all convolutional neural layers, and the new matrix vector is used as the input of the long short-term memory neural network in the recurrent neural network as the time dimension of the long short-term memory neural network. The recurrent neural network trains the new matrix by the time dimension and the eigenvector, and outputs the corresponding pollutant level.

7. A method for assessing the risk level of microplastics processing pollution according to claim 1, characterized in that: A pollution description template is preset, and the text description is filled in according to the pollution description template.

Images