A method and system for identifying olfactory characteristics in water based on electronic nose topological fingerprint

Abstract

The present invention discloses a method and system for identifying odor characteristics in water based on electronic nose topological fingerprints, and relates to the technical field of odor characteristics identification in water; the method comprises the following steps: data acquisition: using an electronic nose sensor array to react with odor substances to generate corresponding electrical signal responses, and collecting electrical signal data; data preprocessing; topological fingerprint construction: converting the preprocessed electrical signal data into a topological fingerprint, in which nodes represent sensors, and edges represent the mutual relationship and interaction mode between sensors; graph neural network model processing: inputting the topological fingerprint data into a graph neural network model, which extracts and fuses the features of nodes and their neighboring nodes layer by layer through multi-layer graph convolution operations to generate high-level feature representations; based on the feature representation in the graph neural network model, the odor category recognition and odor threshold prediction of odor substances are performed. The present invention improves the detection capability of complex odors and low-concentration odor substances.

Description

Technical Field

[0001] The present invention belongs to the technical field of olfactory feature recognition in water, and in particular relates to an olfactory feature recognition method and system in water based on electronic nose topological fingerprint. Background Art

[0002] The odor problem in drinking water is one of the environmental issues of widespread concern around the world. Odor substances not only affect the sensory quality of water and reduce user satisfaction, but may also pose potential health risks. Therefore, identifying the odor characteristics of odor substances in water is crucial to ensure the safety of drinking water.

[0003] The olfactory characteristics of odorous substances include odor threshold and odor category. Olfactory threshold refers to the concentration of odorous substances that cause the least stimulation to human sense of smell. When dealing with odor events in water, it is often necessary to quickly and accurately determine the specific odor-causing substances that cause the odor so as to take appropriate measures to deal with and resolve them. By determining the odor threshold of volatile organic compounds in water, potential key odor-causing substances can be quickly located, and targeted treatment measures can be taken. Therefore, determining the odor threshold of odor-causing substances in water is of great significance for solving odor events. It can help quickly locate and identify problems and provide effective guidance and reference for subsequent water quality management and treatment. In addition to dealing with odor events in water, determining odor thresholds is also widely used in food, beverages, cosmetics, spices and other fields. In these fields, odor thresholds are used to evaluate product quality and safety and guide product formula design and improvement.

[0004] In addition, when discussing the odor problem in drinking water, it is also necessary to classify and identify the categories of odor in detail. Odor categories refer to the classification of odor substances into different categories according to their sensory characteristics, such as earthy musty smell, fishy smell, chemical smell, etc. This classification is crucial to understanding the source and nature of odor substances, because different categories of odor usually point to different causes and sources. For example, earthy musty smell is often produced by microorganisms such as Bacillus licheniformis and is the result of the decomposition of organic matter in the water body; while fishy smell may come from high concentrations of algae and certain types of bacterial activities in the water. Identifying these odor categories can not only help to quickly locate the problem, but also more effectively select appropriate water treatment technology or improvement measures.

[0005] By accurately identifying odor characteristics, water quality issues can be monitored and managed more systematically, the pertinence and efficiency of water treatment can be improved, and relevant departments can respond quickly and take appropriate emergency measures when odor events occur to protect public health and satisfaction. Existing methods for analyzing odor substances include the following:

[0006] Chemical analysis method: Common methods include gas chromatography-mass spectrometry (GC-MS), liquid chromatography (HPLC), etc. Among them, gas chromatography-mass spectrometry is a commonly used analytical method that can be used to quantify and identify volatile organic compounds (VOCs) in water. By extracting and concentrating the organic matter in the water sample and analyzing it using GC-MS technology, the odor threshold of different organic matter can be determined, and the correlation between its concentration in water and human perception can be evaluated. Gas chromatography-olfaction technology combines the characteristics of gas chromatography and human perception of smell, and can directly link volatile organic compounds in water with human smell. This technology combines the compounds separated by GC with the nasal cavity in the olfactory chamber, and the odor threshold of each compound is determined by trained identifiers. These methods are highly sensitive and highly selective, but usually require complex pre-treatment steps and expensive equipment, have long operation times and complex operations, and are not suitable for real-time monitoring.

[0007] Sensory evaluation method: Professional assessors conduct olfactory sensory evaluation of water samples. Although this method can directly reflect the intensity and type of odor, it is greatly influenced by subjective factors, is difficult to standardize, and is not suitable for large-scale monitoring.

[0008] Electronic nose technology: The electronic nose (E-nose) is an intelligent sensor system that simulates the human olfactory system and is used to detect and identify complex odors. The electronic nose consists of multiple chemical sensors, each of which responds to specific odor molecules. By analyzing the overall response pattern of the sensor array, different odors can be identified.

[0009] The electronic nose is mainly composed of a sensor array, a signal acquisition system, and a pattern recognition algorithm. The sensor array contains a variety of chemical sensors that are sensitive to different odors. The signal acquisition system is used to collect and process the response signals of the sensors, and the pattern recognition algorithm is used to analyze and classify the odor data. When the odor molecules come into contact with the sensor surface, the sensor will generate an electrical signal response. Different sensors have different response intensities to odor molecules. By analyzing these response signals, the characteristic information of the odor can be obtained and classified. However, the traditional electronic nose has poor adaptability to different environments and background odors, and is easily interfered by environmental factors, resulting in unstable detection results. In addition, the data processing and pattern recognition algorithms of the traditional electronic nose (such as principal component analysis, linear discriminant analysis, etc.) have limited performance when processing high-dimensional and nonlinear data, making it difficult to accurately classify and identify. The traditional electronic nose also has insufficient sensitivity and accuracy in the detection of complex odor mixtures and low-concentration odor substances, making it difficult to effectively identify and classify.

[0010] Graph neural networks are a type of machine learning model used to process graph data. They can effectively capture the complex relationships between nodes in a graph and perform tasks such as classification, link prediction, and attribute prediction on nodes in the graph. Graph neural networks usually build deeper models by stacking multiple graph convolutional layers and fully connected layers to improve the performance and learning ability of the model. At the same time, some advanced graph neural network models may also include attention mechanisms, residual connections, graph attention networks, and other technologies to further improve the performance and generalization ability of the model.

[0011] The present invention proposes a method and system for identifying olfactory features in water based on electronic nose topological fingerprints. By combining the electronic nose and graph neural network (GNN), the sensitivity and accuracy can be improved, real-time online monitoring can be achieved, and environmental adaptability can be enhanced. Summary of the invention

[0012] The purpose of the present invention is to provide a method and system for identifying olfactory characteristics in water based on electronic nose topological fingerprints, so as to solve the problems mentioned in the above background technology that traditional chemical analysis methods in the prior art cannot achieve real-time monitoring, traditional electronic noses have poor environmental adaptability, limitations in data processing and pattern recognition, and insufficient sensitivity and accuracy in detecting odorous substances.

[0013] To achieve the above object, the present invention adopts the following technical solutions:

[0014] The first aspect of the present invention provides a method for identifying olfactory characteristics in water based on an electronic nose topological fingerprint, comprising the following steps:

[0015] S1. Data collection: using the electronic nose sensor array to react with the odor substance to generate a corresponding electrical signal response and collect electrical signal data;

[0016] S2. Data preprocessing: preprocessing the electrical signal data, including denoising, filtering, and normalization operations;

[0017] S3, topological fingerprint construction: converting the preprocessed electrical signal data into a specific topological data structure, namely, a topological fingerprint, in which the nodes represent sensors and the edges represent the relationships between sensors;

[0018] S4, graph neural network model processing: input the topological fingerprint data into the graph neural network model, and the graph neural network model extracts and fuses the features of the node and its neighboring nodes layer by layer through multi-layer graph convolution operations to generate a high-level feature representation;

[0019] S5. Olfactory category recognition and olfactory threshold prediction; olfactory category recognition and olfactory threshold prediction of olfactory substances based on feature representation in the graph neural network model.

[0020] Preferably, the topological fingerprint in S3 is constructed as follows:

[0021] Each sensor in the electronic nose sensor array is regarded as a node in the topological fingerprint, and the response signal of the sensor is regarded as the node feature vector; the edges between the nodes are constructed according to the response patterns between the sensors to represent the association between the sensors.

[0022] Preferably, the graph neural network model processing in S4 specifically includes the following steps:

[0023] S401, constructing a graph neural network model;

[0024] The graph neural network model includes an input layer, a graph convolution layer, a pooling layer, a fully connected layer, and an output layer connected in sequence;

[0025] The graph convolution layer is used to update the representation of nodes by aggregating the features of neighboring nodes of each node to capture the local connection pattern between nodes. The aggregation process includes weighted aggregation. The aggregated features are then processed by a linear transformation (weight matrix multiplication) to adapt to the input format of the next layer. This structure allows the graph convolution layer to effectively integrate the information of the surrounding environment while maintaining the local structural information of each node.

[0026] The pooling layer is used for graph downsampling. It uses a multi-scale pooling strategy to capture topological fingerprint information at different levels, so as to better understand and express various olfactory characteristics, and aggregate node subsets in the topological fingerprint into smaller representations to preserve the overall structure and characteristics of the graph.

[0027] The fully connected layer contains multiple neurons, each of which is connected to the output of the pooling layer. It performs feature transformation and nonlinear mapping by learning weights and biases. At the same time, a random dropout strategy is applied in the fully connected layer to prevent overfitting, and an adaptive learning rate adjustment technique is used to improve learning efficiency and model stability.

[0028] Output layer: Based on the processing of the previous layers, the output layer is responsible for the specific tasks of odor category recognition and odor threshold prediction. In addition, a multi-task learning framework is introduced to share the underlying feature representation and optimize the classification and prediction tasks at the same time to improve the accuracy and efficiency of the prediction.

[0029] S402, training and optimizing the graph neural network model;

[0030] S403: Use the trained graph neural network model to process the topological fingerprint data.

[0031] Preferably, the odor category identification and odor threshold prediction in S5 are specifically as follows:

[0032] The output layer of the graph neural network model is used to identify the odor category and predict the odor threshold of odor substances. The classifier and regression model are used in the output layer, and a multi-task learning framework is introduced.

[0033] The features processed by the fully connected layer are classified by the classifier to identify the odor category of the odorous substances in the water;

[0034] The features processed by the fully connected layer are regressed through a regression model to predict the odor threshold of odorous substances in water.

[0035] The underlying feature representation is shared through a multi-task learning framework to simultaneously optimize the odor category recognition and odor threshold prediction tasks.

[0036] Preferably, the graph neural network model is trained and optimized in S402, specifically comprising the following steps:

[0037] S4021, acquiring the electrical signal data of the electronic nose sensor array and processing it into topological fingerprint data, marking the odor category and odor threshold of the odor substance in the topological fingerprint data, and forming a data set;

[0038] S4022. Divide the data set into a training set, a test set, and a validation set, and train and optimize the neural network model based on the training set.

[0039] Preferably, the topological fingerprint data is processed in S403 as follows:

[0040] The features of each node and its neighboring nodes are convolved through the graph convolution layer to generate new node features;

[0041] The pooling layer is used to aggregate the features of all nodes to form a global feature representation of the entire graph.

[0042] Nonlinear transformation is performed through the fully connected layer, and the aggregated features are transformed through the nonlinear activation function to enhance the feature expression capability.

[0043] The second aspect of the present invention provides an electronic nose topological fingerprint-based water olfactory feature recognition system applied to the above method, comprising:

[0044] An electronic nose sensor array, including ten different types of sensors, is used to detect odor substances in water and generate corresponding electrical signal responses;

[0045] The signal acquisition and preprocessing module is used to collect the electrical signals of the sensor and perform data preprocessing, which includes denoising, filtering, and normalization operations;

[0046] A topological fingerprint building module is used to build the preprocessed sensor data into a topological fingerprint, where nodes represent sensors and edges represent the relationship between sensors;

[0047] The graph neural network model includes an input layer, a graph convolution layer, a pooling layer, a fully connected layer and an output layer; the topological fingerprint data is subjected to feature extraction and information fusion through graph convolution operations to generate a high-level feature representation; and the output layer includes a classifier and a regression model, the classifier is used to identify the olfactory category of the olfactory substance through the feature representation, and the regression model is used to perform regression analysis on the feature representation to predict the olfactory threshold of the olfactory substance.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] (1) The analysis system of the present invention can improve the detection sensitivity and accuracy through feature extraction and data fusion of GNN. Among them, the high-level feature extraction of GNN is to extract high-level features in the sensor response data through multi-layer graph convolution operations, thereby improving the recognition ability of low-concentration and complex mixed odors; data fusion is to use the GNN model to effectively fuse the response data of different sensors, thereby enhancing the detection ability of weak odor signals; therefore, the detection ability of complex odors and low-concentration odor substances is improved.

[0050] (2) The analysis system of the present invention realizes real-time online monitoring, and combines the rapid response characteristics of the electronic nose and the efficient computing power of GNN to realize real-time online monitoring of odor substances.

[0051] (3) The analysis system in the present invention has high robustness, the electronic nose sensor array has high stability and reliability, and can work stably in different environments; the GNN model has strong noise resistance and can extract useful information from noisy data. Through the nonlinear modeling of GNN, the GNN model can process high-dimensional and nonlinear data and capture complex topological fingerprint information, thereby improving the accuracy and robustness of pattern recognition and further improving environmental adaptability.

[0052] (4) The analysis system of the present invention is capable of performing diversified detection. The sensor array includes multiple sensors that can detect multiple types of odor substances, including aromatic components, nitrogen oxides, ammonia, sulfides, etc. The GNN model can effectively process multiple sensor data and comprehensively analyze the odor categories and odor thresholds of odor substances. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a flow chart of a method for identifying olfactory characteristics in water based on electronic nose topological fingerprint in the present invention;

[0054] Figure 2 Schematic diagram of topological fingerprint in the present invention;

[0055] Figure 3 The model training diagram of the graph neural network model in the present invention, wherein (a) is a schematic diagram of the effect of the model's verification indicator R², and (b) is a schematic diagram of the effect of the model's verification indicator RMSE;

[0056] Figure 4 This is a model training effect diagram of the graph neural network model in the present invention (ODT: olfactory threshold);

[0057] Figure 5 This is a model effect diagram of the neural network model in the present invention (ODT: olfactory threshold). DETAILED DESCRIPTION

[0058] 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.

[0059] Embodiment 1:

[0060] The water odor feature recognition system based on the electronic nose topological fingerprint consists of the following parts:

[0061] Electronic nose sensor array: includes multiple different types of chemical sensors that are used to detect odorous substances in water and generate corresponding electrical signal responses.

[0062] In this embodiment, the model of the electronic nose is PEN3 (Airsense Analytics, Schwerin, Germany); the electronic nose sensor array is specifically shown in Table 1.

[0063] Table 1 Names of sensors in the electronic nose sensor array

[0064]

[0065] Through the rational combination and configuration of the above sensor arrays, efficient and accurate detection of multiple odor substances in water can be achieved. These sensors work together to form a comprehensive odor substance detection network that can cover a wide range of pollutant types and provide reliable detection results.

[0066] Signal acquisition and preprocessing module: collects the electrical signals of the sensor and performs data preprocessing, such as filtering and normalization.

[0067] Topological fingerprint construction module: The preprocessed sensor data is constructed into a topological fingerprint, where nodes represent sensors and edges represent the relationships and interaction patterns between sensors.

[0068] Graph Neural Network (GNN) model: It extracts features and fuses information on topological fingerprint data through graph convolution operations to generate high-level feature representations.

[0069] In this embodiment, the graph neural network includes the following layers:

[0070] Input Layer: The input layer accepts the representation of graph data, which is a topological fingerprint composed of nodes and edges. Each node carries a feature vector of node attributes, and each edge carries a feature vector of edge attributes.

[0071] Graph Convolutional Layer: The graph convolutional layer is one of the core layers of the graph neural network, responsible for information transfer and feature extraction on topological fingerprints. It updates the representation of nodes by aggregating the features of neighboring nodes of each node in order to capture the local connection pattern between nodes. The aggregation process includes weighted aggregation, and the aggregated features are then processed through a linear transformation (weight matrix multiplication) to adapt to the input format of the next layer. This structure allows the graph convolution layer to effectively integrate information about the surrounding environment while maintaining the local structural information of each node.

[0072] Pooling Layer: The pooling layer is used to downsample the graph, aggregating a subset of nodes in the graph into a smaller representation. This helps reduce computational complexity and improve model efficiency while preserving the overall structure and features of the graph. A multi-scale pooling strategy is used to capture topological fingerprint information at different levels, thereby better understanding and expressing various olfactory characteristics.

[0073] Fully Connected Layer: The fully connected layer receives the output of the pooling layer and maps it to one or more target variables. It can contain multiple neurons, each of which is connected to the output of the pooling layer. It performs feature transformation and nonlinear mapping by learning weights and biases. At the same time, a random dropout strategy is applied in the fully connected layer to prevent overfitting, and an adaptive learning rate adjustment technique is used to improve learning efficiency and model stability.

[0074] Output Layer: The output layer is responsible for generating the final prediction results. Based on the processing of the previous layers, the output layer is responsible for specific odor category identification and odor threshold prediction tasks. In addition, a multi-task learning framework is introduced to share the underlying feature representation while optimizing the classification and prediction tasks to improve the accuracy and efficiency of the prediction. For different tasks, the output layer can use different activation functions and loss functions. The present invention uses the sigmoid function for binary classification tasks and the softmax function for multi-classification tasks. The softmax function is used to output the probability of each category for odor classification; the linear function is used for regression tasks, etc. The output layer has only one neuron, which outputs a continuous value for odor threshold prediction.

[0075] Specifically, the features of the GNN output are classified through a classifier to identify the odor categories of odorous substances in water; and the features of the GNN output are regressed and analyzed through a regression model to predict the odor thresholds of odorous substances.

[0076] Specifically, the electronic nose's electrical signal is input into the convolutional layer and the fully connected layer. In the convolutional layer, the graph neural network extracts and updates the electronic nose signal to capture important information about the olfactory characteristics. Subsequently, the extracted features are mapped through the fully connected layer, and the output features in the GNN model are used to classify the olfactory substances and predict their concentrations, and finally the predicted olfactory characteristics are output.

[0077] Result output and display module: outputs the detection results of odor substances, including odor category and odor threshold information.

[0078] See also Figure 1 The method for analyzing odorous substances in water using the above system comprises the following steps:

[0079] Step 1: Sensor data collection;

[0080] The electronic nose sensor reacts with the odorous substance and generates corresponding electrical signals, which reflect the characteristic information of the odorous substance.

[0081] Step 2: signal preprocessing;

[0082] The signal acquisition module collects the electrical signal data of the sensor and performs preprocessing, including denoising, filtering, normalization and other operations to obtain clear and consistent data.

[0083] Step 3: Topological fingerprint construction;

[0084] The preprocessed sensor data is converted into a specific topological data structure, namely the “topological fingerprint”, such as Figure 2 The specific method is as follows:

[0085] Node: Each sensor is a node in the graph, and the node feature vector represents the response signal of the sensor.

[0086] Edge: Edges are constructed based on the response patterns between sensors to represent the association between sensors.

[0087] Step 4: Graph neural network processing;

[0088] The topological fingerprint data is input into the graph neural network (GNN) model. GNN extracts and fuses the features of nodes and their neighbor nodes layer by layer through multi-layer graph convolution operations to generate high-level feature representations. The specific processing steps of GNN are as follows:

[0089] Graph convolution: Perform convolution operations on the features of each node and its neighboring nodes to generate new node features.

[0090] Feature aggregation: Aggregate the features of all nodes to form a global feature representation of the entire graph.

[0091] Non-linear transformation: Through non-linear activation functions, the aggregated features are transformed to enhance the feature expression capability.

[0092] Step 5: Classification and prediction;

[0093] The output features of the GNN model are used to classify the odor and predict the concentration of odor substances. The specific method is as follows:

[0094] Odor category classification: The features output by GNN are classified by the classifier to identify the odor category of odorous substances in water.

[0095] Odor threshold prediction: The odor threshold of odor substances is predicted by performing regression analysis on the features output by GNN through a regression model.

[0096] Step 6: Result output

[0097] The classification and prediction results are output to the result display module to display the type, concentration and odor intensity of the detected odor substances.

[0098] Experimental verification:

[0099] The present invention performs training optimization on the graph neural network (GNN) model. The gradual improvement of the training set, test set, and validation set effects with the increase of training rounds reflects the effective learning and optimization of the model during the training process. The improvement of the training set effect shows that the model better fits the training data, while the improvement of the test set and validation set effects shows that the model's generalization ability for unknown data is enhanced. At the same time, the improvement of the validation set effect also implies the effectiveness of the model hyperparameter adjustment. In this case, the model has stronger adaptability and predictive ability, can predict new data more accurately, and achieve better performance on different data sets. Figure 3 shown.

[0100] The R² scores of the model on the training set, test set, and validation set are 0.82, 0.72, and 0.70, respectively. Figure 4 As shown, the model has good performance. Although the R² scores on the test set and validation set are slightly lower than those on the training set, they still reach a high level, indicating that the model has good generalization ability and can make accurate predictions on unknown data.

[0101] During the external validation process, the present invention collected 200 additional olfactory threshold data from the literature. These data are completely different from the data used in the modeling process, in order to verify the generalization ability of the model on new data. By preprocessing these data and inputting them into the trained model, the present invention evaluated the model's prediction performance on new data, and finally R 2 Reached 0.71, such as Figure 5 As shown, the robustness and reliability of the model in practical applications are ensured.

[0102] The water odor feature recognition system based on the electronic nose topological fingerprint of the present invention has the following advantages:

[0103] Efficiency: The system quickly collects data through the electronic nose sensor array and efficiently processes and analyzes it through the GNN model, significantly shortening the detection time. The automated detection process reduces manual operation and intervention and improves work efficiency.

[0104] Convenience: The system is designed to be simple and easy to use. Users only need to perform basic operations to complete the detection of odor substances. The result output and display module provides intuitive test results and detailed analysis reports, which are convenient for users to understand and apply.

[0105] Economical: The combination of the electronic nose sensor array and the GNN model reduces the detection cost and is more economical than traditional laboratory detection methods. The system can be reused, the maintenance cost of the sensor array is low, and the long-term use cost is relatively controllable.

[0106] Innovation: The system combines electronic nose technology with the graph neural network (GNN) model for the first time, proposing a new method for detecting odorous substances in water. The system design is innovative in sensor data preprocessing, topological fingerprint construction, GNN model training and result output, filling the gap in existing technologies.

[0107] The above description is only used to help understand the method of the present invention and its core essence, but the protection scope of the present invention is not limited thereto. For those skilled in the art in the art, equivalent replacement or change according to the technical solution and inventive concept of the present invention within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for identifying olfactory characteristics in water based on electronic nose topological fingerprint, characterized in that: The following steps are involved: S1. Data collection: using the electronic nose sensor array to react with the odor substance to generate a corresponding electrical signal response and collect electrical signal data; S2. Data preprocessing: preprocessing the electrical signal data, including denoising, filtering, and normalization operations; S3, topological fingerprint construction; The preprocessed electrical signal data is converted into a topological fingerprint, in which nodes represent sensors and edges represent the relationships and interaction patterns between sensors; The details are as follows: each sensor in the electronic nose sensor array is taken as a node in the topological fingerprint, and the response signal of the sensor is taken as the node feature vector; the edges between the nodes are constructed according to the response patterns between the sensors to represent the association between the sensors; S4, graph neural network model processing: input the topological fingerprint data into the graph neural network model, and the graph neural network model extracts and fuses the features of the node and its neighboring nodes layer by layer through multi-layer graph convolution operations to generate a high-level feature representation; S401, constructing a graph neural network model; the graph neural network model includes an input layer, a graph convolution layer, a pooling layer, a fully connected layer and an output layer connected in sequence; S402, training and optimizing the graph neural network model; S4021, acquiring the electrical signal data of the electronic nose sensor array and processing it into topological fingerprint data, marking the odor category and odor threshold of the odor substance in the topological fingerprint data, and forming a data set; S4022, dividing the data set into a training set, a test set, and a validation set, and training and optimizing the neural network model based on the training set; S403, using the trained graph neural network model to process the topological fingerprint data; The features of each node and its neighboring nodes are convolved through the graph convolution layer to generate new node features; The pooling layer is used to aggregate the features of all nodes to form a global feature representation of the entire graph. Through the fully connected layer, nonlinear transformation is performed, and the aggregated features are transformed through nonlinear activation functions to enhance the feature expression capability; S5. Odor category recognition and odor threshold prediction: Odor category recognition and odor threshold prediction of odor substances based on feature representation in the graph neural network model; The output layer of the graph neural network model is used to identify the odor category and predict the odor threshold of odor substances. The classifier and regression model are used in the output layer, and a multi-task learning framework is introduced. The features processed by the fully connected layer are classified by the classifier to identify the odor category of the odorous substances in the water; The features processed by the fully connected layer are regressed through a regression model to predict the odor threshold of odorous substances in water. The underlying feature representation is shared through a multi-task learning framework to simultaneously optimize the odor category recognition and odor threshold prediction tasks.

2. The method for identifying olfactory characteristics in water based on electronic nose topological fingerprint according to claim 1 is characterized in that: The graph neural network model includes an input layer, a graph convolution layer, a pooling layer, a fully connected layer and an output layer connected in sequence; Graph convolutional layer, used to update the representation of nodes by aggregating the features of neighboring nodes of each node, thereby capturing the local connection pattern between nodes. The aggregation process includes weighted aggregation; The aggregated features are then processed through a linear transformation, which is a weight matrix multiplication, to adapt to the input format of the next layer; The pooling layer is used for graph downsampling. It adopts a multi-scale pooling strategy to capture topological fingerprint information at different levels and aggregate node subsets in the topological fingerprint into smaller representations. The fully connected layer contains multiple neurons, each of which is connected to the output of the pooling layer. It performs feature transformation and nonlinear mapping by learning weights and biases. At the same time, a random dropout strategy is applied in the fully connected layer to prevent overfitting, and an adaptive learning rate adjustment technique is used.

3. A water olfactory feature recognition system based on electronic nose topological fingerprint applied to the method described in claim 1 or 2, characterized in that: include: An electronic nose sensor array, including ten different types of sensors, is used to detect odor substances in water and generate corresponding electrical signal responses; The signal acquisition and preprocessing module is used to collect the electrical signals of the sensor and perform data preprocessing, which includes denoising, filtering, and normalization operations; A topological fingerprint building module is used to build the preprocessed sensor data into a topological fingerprint, where nodes represent sensors and edges represent the relationship between sensors; The graph neural network model includes an input layer, a graph convolution layer, a pooling layer, a fully connected layer and an output layer; the topological fingerprint data is subjected to feature extraction and information fusion through graph convolution operations to generate a high-level feature representation; and the output layer includes a classifier and a regression model, the classifier is used to identify the olfactory category of the olfactory substance through the feature representation, and the regression model is used to perform regression analysis on the feature representation to predict the olfactory threshold of the olfactory substance.

Images