A training system and method of a semi-supervised machine learning model for water quality pollutant detection

Abstract

The application discloses a training system and method of a semi-supervised machine learning model for water quality pollutant detection, and relates to the field of environmental detection. The system comprises a water body sampling module, a pollutant adding module, a data sampling module and a data analysis module. The water body sampling module is used for collecting water body samples in a monitoring area. The pollutant adding module is used for adding pollutants to the water body samples according to randomly obtained different combinations of pollutant types and concentration settings, obtaining the water body samples, and sending the water body samples to the data sampling module. The data sampling module is used for collecting physicochemical properties of the water body samples of the water body sampling module and the pollutant adding module through a sensor. The data analysis module is used for inputting data of the physicochemical properties of the water body samples collected by the water body sampling module into a semi-supervised machine learning model, taking the physicochemical property data as input, and constructing and training the semi-supervised machine learning model. The application is also suitable for the field of real-time training and detection of water quality pollutants.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring technology, specifically to a training system and method for a semi-supervised machine learning model for detecting water pollutants. Background Technology

[0002] Patent publication numbers CN 114778647 A, CN 118549501 A, CN 117390503 A, and CN113820376A disclose that the composition and concentration of water pollutants can be inferred from differences and / or changes in their physicochemical properties, such as conductivity, pH, turbidity, residual chlorine, and color. However, researchers have found it difficult to construct the relationship between these physicochemical properties and water pollutants due to the difficulty in analyzing the complex and subtle differences between data. Machine learning, a process of mining deep-seated patterns through data, has been widely coupled in recent years to the coupling of physicochemical properties and water pollutants. Through in-depth analysis of the response signal relationships of the physicochemical properties of different types and concentrations of water pollutants using machine learning, quantitative and / or qualitative analysis of dissolved substances can ultimately be achieved.

[0003] However, in practical applications, due to the complex composition of the water samples being tested, a relatively large standard curve needs to be constructed in the laboratory, i.e., obtaining a machine learning model trained on pure substances or mixtures of different pure substances. This process is extremely time-consuming and labor-intensive. Furthermore, the physicochemical properties of water are numerous and complex, and the detection methods available in different regions vary, preventing machine learning models from achieving good generalization. In other words, models trained on laboratory data or water samples from other locations may not be applicable to the target water sample. Better methods are needed to combine in-situ water sample testing, allowing for real-time training even if the composition of the target water sample changes, resulting in a model with better generalization capabilities. Summary of the Invention

[0004] This invention addresses the problem that existing methods, due to the complex composition of water samples in practical applications, require the construction of a large standard curve in the laboratory, i.e., obtaining a well-trained machine learning model based on pure substances or mixtures of different pure substances. However, this work is extremely time-consuming and labor-intensive. Furthermore, the physicochemical properties of water are diverse and complex, and the detection methods available in different regions vary, resulting in insufficient versatility of machine learning models.

[0005] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0006] Option 1: This invention proposes a training system for a semi-supervised machine learning model for water pollutant detection. The system includes a water sampling module, a pollutant addition module, a data sampling module, and a data analysis module.

[0007] The water sampling module is used to collect raw water samples from the monitoring area and send pure water samples, raw water samples, and diluted raw water samples to the pollutant dosing module. Then, the pure water samples and raw water samples are sent to the data sampling module.

[0008] Water samples include original water samples, pure water samples, and diluted original water samples;

[0009] The pollutant dosing module is used to add pollutants to water samples based on randomly obtained combinations of pollutant types and concentrations, obtain water samples, and send them to the data sampling module.

[0010] The data sampling module is used to collect the physicochemical properties of water samples from the water sampling module and the pollutant dosing module through detection equipment or sensors;

[0011] The data analysis module is used to input the physicochemical properties data of the water samples collected by the water sampling module into the semi-supervised machine learning model. The semi-supervised machine learning model is constructed and trained using the physicochemical properties data as input.

[0012] Furthermore, a preferred embodiment is provided in which the water sample is transferred to the pollutant dosing module and the data sampling module in the water sampling module by gravity flow, inlet pressure flow or water pump pressure flow.

[0013] Furthermore, a preferred embodiment is provided in which the diluted original water sample obtained by the water sampling module is obtained by randomly selecting the mixing ratio of the original water sample and the pure water sample from 1:4, 2:3, 3:2 and 4:1.

[0014] Furthermore, a preferred embodiment is provided in which the ratio of the training set to the test set in step two is 8:2.

[0015] Furthermore, a preferred embodiment is provided, wherein the physicochemical properties of the water sample in the data acquisition module include pH value, temperature, dissolved oxygen, turbidity, and conductivity.

[0016] Furthermore, a preferred embodiment is provided in which the machine learning training model in the data analysis module is implemented using one or more combinations of greedy algorithms, dynamic programming algorithms, divide-and-conquer algorithms, backtracking algorithms, and branch and bound algorithms.

[0017] Furthermore, a preferred embodiment is provided, wherein the method for constructing the semi-supervised machine learning model includes:

[0018] S1. The physicochemical properties of a pure water sample are used as input, and blank values ​​are used as output. The input and output data are called data a. The physicochemical properties of a pure water sample with added pollutants are used as input, and the types and concentrations of pollutants set for the pure water sample are used as output. The input and output data are called data b. The model is trained in the data analysis module using data a and data b to obtain model M. If a pre-trained model M already exists, training continues on model M.

[0019] S2. Based on model M, the physicochemical properties of the original water sample are used as input to obtain the model output, which is outputT;

[0020] S3. The original water sample after adding pollutants and the diluted original water sample are respectively polluted water sample c and polluted water sample d. The physicochemical properties of polluted water sample c are used as input, and the output is output T plus the sum of the pollutant types and concentrations set for polluted water sample c. The input and output data are called data c. The physicochemical properties of polluted water sample d are used as input, and the output T is calculated as the dilution degree of the diluted original water sample. Then, the result is summed with the pollutant types and concentrations set for polluted water sample d. The input and output data are called data d. Model M is trained in the data analysis module using data c and data d.

[0021] S4, repeat S1.

[0022] Furthermore, in a preferred embodiment, a data transmission module is provided between the data sampling module and the data analysis module to enable remote monitoring.

[0023] Option 2: A training method for a semi-supervised machine learning model for water pollutant detection, characterized in that the training method is implemented based on the system described in any one of Options 1, and the method includes the following steps:

[0024] Step 1: The water sampling module sends pure water samples, original water samples, and diluted original water samples to the pollutant dosing module, and sends water samples and original water samples to the data analysis module.

[0025] Step 2: The pollutant addition module randomly selects a set of pollutant types and concentration settings, and adds the pollutants to the water sample from the water sampling module;

[0026] Step 3: The data sampling module acquires the physicochemical properties of water samples from the water sampling module and the pollutant dosing module;

[0027] Step 4: The data transmission module sends the physicochemical property information collected by the data sampling module to the data analysis module;

[0028] Step 5: The data analysis module inputs the physicochemical property information into the semi-supervised machine learning model and trains the semi-supervised machine learning model using the construction method of the semi-supervised machine learning model.

[0029] Step 6: Repeat steps 1 to 5 to train a semi-supervised machine learning model for detecting water pollutants.

[0030] Option 3: A computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the method described in Option 2.

[0031] Option 4: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described in Option 2.

[0032] The advantages of this invention are:

[0033] The training system and method for a semi-supervised machine learning model for water pollutant detection proposed in this invention constructs a machine learning model by collecting and analyzing the physicochemical properties of different types of water samples. By randomly setting pollutant concentrations and combining them with in-situ detection data, a semi-supervised learning algorithm is used to train the model. This enables the machine learning model to achieve stable training capabilities for different target water bodies, giving the model higher versatility. In other words, the sampling and training methods employed in this invention improve the model's generalization ability to different water conditions, thereby facilitating real-time training and detection of water pollutants.

[0034] This invention is also applicable to the field of real-time training and detection of water pollutants. Attached Figure Description

[0035] Figure 1 This is a schematic diagram illustrating the principle of a training system for a semi-supervised machine learning model for detecting water pollutants, as described in Implementation Method 1.

[0036] Figure 2 This is a flowchart of the model training process in the data analysis module of the training method for a semi-supervised machine learning model for water pollutant detection described in Embodiment 8.

[0037] Among them, 11 is the original water sample, 12 is the diluted original water sample, 13 is the pure water sample, 14 is valve 1, 15 is valve 2, 16 is the sampling port of the original water sample, 17 is the sampling port of the diluted original water sample, and 18 is the sampling port of the pure water sample; 1 is the pollutant dosing module, 21 is the water sample inlet of pollutant dosing module 1, 22 is the CCl4 dosing device, 23 is the CCl3 dosing device, and 24 is the CCl2 dosing device; 2 is the data sampling module, 31 is the pH analyzer, 32 is the turbidity analyzer, 33 is the spectrophotometer, 34 is the voltammetric cycle analyzer, and 35 is the chlorine analyzer; 3 is the data transmission module; and 4 is the data analysis module. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0039] Implementation Method 1: This implementation method proposes a training system for a semi-supervised machine learning model for water pollutant detection. The system includes a water sampling module, a pollutant addition module, a data sampling module, and a data analysis module.

[0040] The water sampling module is used to collect raw water samples from the monitoring area and send pure water samples, raw water samples, and diluted raw water samples to the pollutant dosing module. Then, the pure water samples and raw water samples are sent to the data sampling module.

[0041] Water samples include original water samples, pure water samples, and diluted original water samples;

[0042] The pollutant dosing module is used to add pollutants to water samples based on randomly obtained combinations of pollutant types and concentrations, obtain water samples, and send them to the data sampling module.

[0043] The data sampling module is used to collect the physicochemical properties of water samples from the water sampling module and the pollutant dosing module through detection equipment or sensors;

[0044] The data analysis module is used to input the physicochemical properties data of the water samples collected by the water sampling module into the semi-supervised machine learning model. The semi-supervised machine learning model is constructed and trained using the physicochemical properties data as input.

[0045] Implementation Method 2: This implementation method further defines the training system of a semi-supervised machine learning model for detecting water pollutants as described in Implementation Method 1. In the water sampling module, the water sample is transferred to the pollutant dosing module and the data sampling module by gravity flow, inlet pressure flow or water pump pressure flow.

[0046] Implementation Method 3: This implementation method further defines the training system of the semi-supervised machine learning model for detecting water pollutants described in Implementation Method 1. The diluted original water sample after dilution by the water sampling module is obtained by randomly selecting the mixing ratio of the original water sample and the pure water sample from 1:4, 2:3, 3:2 and 4:1.

[0047] Implementation Method 4: This implementation method further defines the training system of the semi-supervised machine learning model for detecting water pollutants described in Implementation Method 1. The physicochemical properties of the water sample in the data acquisition module include pH value, temperature, dissolved oxygen, turbidity, and conductivity.

[0048] Implementation Method 5: This implementation method further defines the training system for a semi-supervised machine learning model for water pollutant detection described in Implementation Method 1. The machine learning training model in the data analysis module is implemented using one or more combinations of greedy algorithms, dynamic programming algorithms, divide-and-conquer algorithms, backtracking algorithms, and branch and bound algorithms.

[0049] Implementation Method Six: This implementation method further defines the training system for a semi-supervised machine learning model for water pollutant detection described in Implementation Method One. The method for constructing the semi-supervised machine learning model includes:

[0050] S1. The physicochemical properties of a pure water sample are used as input, and blank values ​​are used as output. The input and output data are called data a. The physicochemical properties of a pure water sample with added pollutants are used as input, and the types and concentrations of pollutants set for the pure water sample are used as output. The input and output data are called data b. The model is trained in the data analysis module using data a and data b to obtain model M. If a pre-trained model M already exists, training continues on model M.

[0051] S2. Based on model M, the physicochemical properties of the original water sample are used as input to obtain the model output, which is outputT;

[0052] S3. The original water sample after adding pollutants and the diluted original water sample are respectively polluted water sample c and polluted water sample d. The physicochemical properties of polluted water sample c are used as input, and the output is output T plus the sum of the pollutant types and concentrations set for polluted water sample c. The input and output data are called data c. The physicochemical properties of polluted water sample d are used as input, and the output T is calculated as the dilution degree of the diluted original water sample. Then, the result is summed with the pollutant types and concentrations set for polluted water sample d. The input and output data are called data d. Model M is trained in the data analysis module using data c and data d.

[0053] S4, repeat S1.

[0054] Implementation Method Seven: This implementation method further defines the training system for a semi-supervised machine learning model for detecting water pollutants as described in Implementation Method One. A data transmission module is also provided between the data sampling module and the data analysis module to enable remote monitoring.

[0055] Implementation Method Eight: This implementation method provides a training method for a semi-supervised machine learning model for water pollutant detection. The training method is implemented based on the system described in any one of Implementation Methods One to Seven, and the method includes the following steps:

[0056] Step 1: The water sampling module sends pure water samples, original water samples, and diluted original water samples to the pollutant dosing module, and sends water samples and original water samples to the data analysis module.

[0057] Step 2: The pollutant addition module randomly selects a set of pollutant types and concentration settings, and adds the pollutants to the water sample from the water sampling module;

[0058] Step 3: The data sampling module acquires the physicochemical properties of water samples from the water sampling module and the pollutant dosing module;

[0059] Step 4: The data transmission module sends the physicochemical property information collected by the data sampling module to the data analysis module;

[0060] Step 5: The data analysis module inputs the physicochemical property information into the semi-supervised machine learning model and trains the semi-supervised machine learning model using the construction method of the semi-supervised machine learning model.

[0061] Step 6: Repeat steps 1 to 5 to train a semi-supervised machine learning model for detecting water pollutants.

[0062] Implementation Method Nine: This implementation method provides a computer device, including a memory and a processor. The memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes the method described in Implementation Method Eight.

[0063] Implementation Method 10: This implementation method provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described in Implementation Method 8.

[0064] Implementation Method Eleven: This implementation method provides an example, which is used to explain the above-described implementation methods one through ten. The specific example is as follows:

[0065] See Figure 1 and Figure 2 This implementation method specifically includes the following steps:

[0066] This invention proposes a training method for a semi-supervised machine learning model for water pollutant detection. By combining in-situ target water sample data, a more versatile machine learning model is obtained through a semi-supervised mode.

[0067] A training system for a semi-supervised machine learning model for water pollutant detection includes: a water sampling module, a pollutant dosing module, a data sampling module, and a data analysis module. When remote monitoring is required, a data transmission module can be added between the data sampling module and the data analysis module.

[0068] The water sampling module is used to collect raw water samples from the monitoring area and send pure water samples, raw water samples, and diluted raw water samples to the pollutant dosing module. Then, the pure water samples and raw water samples are sent to the data sampling module.

[0069] Original water samples, pure water samples, and diluted original water samples are collectively referred to as water samples.

[0070] In the water sampling module, water samples are transferred to the pollutant dosing module and the data sampling module via gravity flow, inlet pressure flow, or water pump pressure flow.

[0071] Diluting the original water sample is achieved by mixing the original water sample with a pure water sample in a certain ratio, with the dilution degree being a random value. For example, depending on the setting of the random seed, the mixing ratio of the original water sample and the pure water sample is randomly selected from 1:4, 2:3, 3:2, and 4:1. The choice of dilution degree is not limited to the values ​​mentioned above.

[0072] The pollutant addition module is used to add pollutants to the water samples from the water sampling module. Based on randomly obtained combinations of pollutant types and concentrations, pollutants are added to the water samples to obtain polluted water samples, which are then sent to the data sampling module.

[0073] The data sampling module collects the physicochemical properties of water samples from the water sampling module and the pollutant dosing module through detection equipment and / or sensors, including:

[0074] pH value, temperature, dissolved oxygen, turbidity, conductivity, total organic carbon (TOC), ammonia nitrogen (NH3-N), nitrate nitrogen (NO3-N) - -N), phosphate (PO4) 3- Chemical oxygen demand (COD), biochemical oxygen demand (BOD), dissolved solids (TDS), suspended solids (SS), heavy metal ions (such as lead, mercury, copper, cadmium, etc.), and chloride (Cl) - ), sulfate (SO4) 2-The following parameters are considered: amino acids, volatile organic compounds (VOCs), pesticide residues, total nitrogen, total phosphorus, dissolved gas content (such as nitrogen, oxygen, carbon dioxide), changes in water pH, dissolved oxygen saturation, total dissolved gas pressure, total suspended particulate matter (TSS) concentration, and / or current distribution curves obtained by cyclic voltammetry, and / or peak signals obtained by liquid chromatography, and / or absorbance measured by ultraviolet-visible spectrophotometry, and / or gas composition determined by gas chromatography, and / or ion mass-to-charge ratio data obtained by mass spectrometry, and / or fluorescence intensity data obtained by fluorescence spectroscopy, and / or measurements by atomic absorption spectrometry. The concentration of metal ions, and / or the concentration of ions determined by ion-selective electrode method, and / or the ion components obtained by ion chromatography, and / or the infrared absorption spectrum obtained by Fourier transform infrared spectroscopy (FTIR), and / or the molecular structure data of organic matter in water obtained by nuclear magnetic resonance (NMR), and / or the water pollutant concentration information obtained by electrochemical sensors, and / or the assessment of microbial activity in water by microbial electrode method, and / or the detection of volatile organic compounds (VOCs) in water by electronic nose technology, and / or the multidimensional water quality data obtained by multi-parameter spectroscopy, and / or the particle size distribution of particulate matter in water determined by laser particle size analysis, and / or detect the concentration of suspended solids in water by light scattering, and / or obtain quantitative analysis data of organic pollutants by high performance liquid chromatography-mass spectrometry (HPLC-MS), and / or determine the concentration of trace elements in water by trace element analysis, and / or obtain pollutant molecular information by self-assembled nanosensors, and / or obtain the concentration of harmful gases in water by non-metallic sensors, and / or obtain molecular structure information of pollutants in water by detecting molecular vibrational modes in water by Raman spectroscopy, and / or detect the content of heavy metals in water by laser-induced breakdown spectroscopy (LIBS), and / or monitor the sound wave propagation characteristics of pollutants in water by acoustic sensors, and / or through Non-invasive analysis of pollutant distribution in water bodies can be performed using magnetic resonance imaging (MRI) technology, and / or quantitative assessment of solid pollutants in water can be achieved using thermogravimetric analysis (TGA), and / or simultaneous determination of organic pollutants and heavy metals in water can be achieved using ultraviolet-fluorescence spectroscopy, and / or analysis of pollutant molecular interactions in water bodies can be performed using surface plasmon resonance (SPR) technology, and / or real-time dynamic observation of particles in water can be achieved using holographic microscopy, and / or real-time determination of the reactivity and concentration changes of chemical substances in water can be achieved using capacitance-conductivity methods, and / or high-throughput analysis of trace pollutants in water can be achieved using droplet microfluidic chip technology.And / or monitor the dynamic changes of water temperature, pH and nitrogen and phosphorus pollutants using meteorological satellite data combined with remote sensing technology, and / or detect heavy metal pollutants in water using thermoluminescent spectrophotometry (TLPS), and / or detect micro particulate matter and its properties in water using polarized light scattering technology, and / or sensitively detect trace organic matter in water using quantum dot sensors, and / or simultaneously detect chemical and physical pollutants in water using acousto-optic composite sensors, and / or monitor changes in microbial communities in water and their response to pollutants using gene chip technology, and / or conduct rapid on-site water quality testing using portable water quality analyzers, and / or monitor the spatial distribution of pollutants in water in real time using multispectral imaging technology, and / or detect pollutants in deep water areas using endoscopic underwater sensors, and / or monitor the dynamic changes of suspended particulate matter in water in real time using holographic interferometer technology, and / or monitor the concentration and changes of harmful chemical substances in water using microwave sensing technology.

[0075] The data analysis module is used to input the physicochemical properties data of water samples collected by the water sampling module into the machine learning model. The machine learning model is trained using this physicochemical property data as input. The models used for machine learning training include:

[0076] Exhaustive search, greedy algorithm, dynamic programming algorithm, divide and conquer algorithm, backtracking algorithm, branch and bound algorithm, simulated annealing algorithm, tabu search algorithm, ant colony algorithm, particle swarm optimization algorithm, artificial bee colony algorithm, genetic algorithm, Monte Carlo method, gradient descent algorithm, forgetting learning algorithm, neural network algorithm, genetic programming algorithm, reinforcement genetic algorithm, Bayesian optimization algorithm, Lagrange relaxation algorithm, convex optimization algorithm, Newton's method, quasi-Newton method, interior point method, Lagrange multiplier method, alternating direction multiplier method, graph algorithm, permutation algorithm, probabilistic graphical model, fuzzy logic algorithm, neurofuzzy system, quantum algorithm, linear programming algorithm, nonlinear programming algorithm, multi-objective optimization algorithm, Markov decision process, game theory algorithm, differential evolution algorithm, XGBoost (extreme gradient boosting), LightGBM (lightweight gradient boosting), random forest, support vector machine (SVM), K-nearest neighbor (KNN) algorithm, linear regression, ridge regression and Lasso regression. One or more of the following algorithms: Regression, Self-Attention, Autoencoder, Generative Adversarial Network (GAN), Convolutional Neural Network (CNN), Long Short-Term Memory Network (LSTM), Graph Convolutional Network (GCN), Graph Neural Network (GNN), Reinforcement Learning (Q-learning, Policy Gradient, Deep Q-Network (DQN)), Transfer Learning, Deep Neural Network (DNN), Naive Bayes Classifier, K-means Clustering, DBSCAN (Density-based Spatial Clustering), Laplacian Smoothing, Pseudo-Inverse, Monte Carlo Tree Search (MCTS), Markov Chain Monte Carlo (MCMC), Variational Autoencoder (VAE), and Transformer algorithm.

[0077] The output acquisition and training of a semi-supervised machine learning model are performed according to the following steps, which is the method for building a semi-supervised model in the data analysis module:

[0078] S1. Using the physicochemical properties of a pure water sample as input and blank values ​​as output, this input and output data are referred to as data a. Using the physicochemical properties of a pure water sample with added pollutants as input and the type and concentration of pollutants specified for the pure water sample as output, this input and output data are referred to as data b. The model is trained in the data analysis module using data a and data b to obtain model M. If a pre-trained model M already exists, training continues on model M.

[0079] S2. Based on model M, using the physicochemical properties of the original water sample as input, obtain the model output. The meaning of the model output at this point is the type and concentration of pollutants in the original water sample predicted based on model M; this set of outputs is called output T.

[0080] S3. The original water sample after adding pollutants and the diluted original water sample are respectively polluted water sample c and polluted water sample d. The physicochemical properties of polluted water sample c are used as input, and the output is the sum of output T and the pollutant types and concentrations set for polluted water sample c. This input and output data is called data c. The physicochemical properties of polluted water sample d are used as input, and output T × the dilution rate of the diluted original water sample is calculated. This is then summed with the pollutant types and concentrations set for polluted water sample d to obtain the output. This input and output data is called data d. Model M is trained in the data analysis module using data c and data d.

[0081] S4. Repeat step 1.

[0082] When remote monitoring is required, a data transmission module is added between the data sampling module and the data analysis module.

[0083] The data transmission module sends the collected information from the data sampling module to the data analysis module through a remote signal transmitter. The remote signal transmitter includes one or more of the following: 2G / 3G / 4G / 5G DTU (data transmission unit), Wi-Fi module, Bluetooth module, ZigBee module, LoRa module, NB-IoT module, and satellite communication module.

[0084] The system includes a water sampling module, a pollutant dosing module, a data sampling module, and a data analysis module. For remote monitoring, a data transmission module can be added between the data sampling module and the data analysis module.

[0085] This invention provides the following technical steps:

[0086] S1. The water sampling module sends water samples, original water samples, and diluted original water samples to the pollutant dosing module, and sends water samples and original water samples to the data analysis module.

[0087] S2. The pollutant addition module randomly selects a set of pollutant types and concentration settings, and adds pollutants to the water sample from the water sampling module.

[0088] S3. The data sampling module acquires the physicochemical properties of water samples from the water sampling module and the pollutant dosing module;

[0089] S4. The data transmission module sends the collected information from the data sampling module to the data analysis module;

[0090] S5. The data analysis module inputs the physicochemical properties information into the machine learning model and trains the model based on the semi-supervised model construction method mentioned above.

[0091] S6. Repeat from step 1.

[0092] This invention utilizes in-situ target detection water sample data and employs a semi-supervised learning method to enable the machine learning model to achieve stable training capabilities on different target water bodies, thereby giving the model greater versatility.

[0093] Those skilled in the art will understand that the above description is merely a preferred embodiment of the present invention, and the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. This is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0094] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A training system for a semi-supervised machine learning model for detecting water pollutants, characterized in that, The system includes a water sampling module, a pollutant dosing module, a data sampling module, and a data analysis module; The water sampling module is used to collect raw water samples from the monitoring area and send pure water samples, raw water samples, and diluted raw water samples to the pollutant dosing module. Then, the pure water samples and raw water samples are sent to the data sampling module. Water samples include original water samples, pure water samples, and diluted original water samples; The pollutant dosing module is used to add pollutants to water samples based on randomly obtained combinations of pollutant types and concentrations, obtain water samples, and send them to the data sampling module. The data sampling module is used to collect the physicochemical properties of water samples from the water sampling module and the pollutant dosing module through detection equipment or sensors; The data analysis module is used to input the physicochemical properties data of the water samples collected by the water sampling module into the semi-supervised machine learning model. The semi-supervised machine learning model is constructed and trained using the physicochemical properties data as input. The method for constructing the semi-supervised machine learning model includes: S1. The physicochemical properties of a pure water sample are used as input, and blank values ​​are used as output. The input and output data are called data a. The physicochemical properties of a pure water sample with added pollutants are used as input, and the types and concentrations of pollutants set for the pure water sample are used as output. The input and output data are called data b. The model is trained in the data analysis module using data a and data b to obtain model M. If a pre-trained model M already exists, training continues on model M. S2. Based on model M, the physicochemical properties of the original water sample are used as input to obtain the model output, which is output T. S3. The original water sample after adding pollutants and the diluted original water sample are respectively polluted water sample c and polluted water sample d. The physicochemical properties of polluted water sample c are used as input, and the output is output T plus the sum of the pollutant types and concentrations set for polluted water sample c. The input and output data are called data c. The physicochemical properties of polluted water sample d are used as input, and the output T is calculated as the dilution degree of the diluted original water sample. Then, the result is summed with the pollutant types and concentrations set for polluted water sample d. The input and output data are called data d. Model M is trained in the data analysis module using data c and data d. S4, repeat S1.

2. The training system for the semi-supervised machine learning model for water pollutant detection according to claim 1, characterized in that, The water sample is transferred from the water sampling module to the pollutant dosing module and the data sampling module via gravity flow, inlet pressure flow, or water pump pressure flow.

3. The training system for the semi-supervised machine learning model for water pollutant detection according to claim 1, characterized in that, The diluted original water sample obtained by the water sampling module is obtained by randomly selecting a mixing ratio of the original water sample and the pure water sample from 1:4, 2:3, 3:2 and 4:

1.

4. The training system for the semi-supervised machine learning model for water pollutant detection according to claim 1, characterized in that, The physicochemical properties of the water samples in the data acquisition module include pH value, temperature, dissolved oxygen, turbidity, and conductivity.

5. The training system for the semi-supervised machine learning model for water pollutant detection according to claim 1, characterized in that, The machine learning training models in the data analysis module are implemented using one or more combinations of greedy algorithms, dynamic programming algorithms, divide-and-conquer algorithms, backtracking algorithms, and branch and bound algorithms.

6. The training system for the semi-supervised machine learning model for water pollutant detection according to claim 1, characterized in that, A data transmission module is also provided between the data sampling module and the data analysis module to enable remote monitoring.

7. A training method for a semi-supervised machine learning model for water pollutant detection, characterized in that, The training method is implemented based on the system according to any one of claims 1 to 6, and the method includes the following steps: Step 1: The water sampling module sends pure water samples, original water samples, and diluted original water samples to the pollutant dosing module, and sends water samples and original water samples to the data analysis module. Step 2: The pollutant addition module randomly selects a set of pollutant types and concentration settings, and adds the pollutants to the water sample from the water sampling module; Step 3: The data sampling module acquires the physicochemical properties of water samples from the water sampling module and the pollutant dosing module; Step 4: The data transmission module sends the physicochemical property information collected by the data sampling module to the data analysis module; Step 5: The data analysis module inputs the physicochemical property information into the semi-supervised machine learning model and trains the semi-supervised machine learning model using the construction method of the semi-supervised machine learning model. Step 6: Repeat steps 1 to 5 to train a semi-supervised machine learning model for detecting water pollutants. The method for constructing the semi-supervised machine learning model includes: S1. The physicochemical properties of a pure water sample are used as input, and blank values ​​are used as output. The input and output data are called data a. The physicochemical properties of a pure water sample with added pollutants are used as input, and the types and concentrations of pollutants set for the pure water sample are used as output. The input and output data are called data b. The model is trained in the data analysis module using data a and data b to obtain model M. If a pre-trained model M already exists, training continues on model M. S2. Based on model M, the physicochemical properties of the original water sample are used as input to obtain the model output, which is output T. S3. The original water sample after adding pollutants and the diluted original water sample are respectively polluted water sample c and polluted water sample d. The physicochemical properties of polluted water sample c are used as input, and the output is output T plus the sum of the pollutant types and concentrations set for polluted water sample c. The input and output data are called data c. The physicochemical properties of polluted water sample d are used as input, and the output T is calculated as the dilution degree of the diluted original water sample. Then, the result is summed with the pollutant types and concentrations set for polluted water sample d. The input and output data are called data d. Model M is trained in the data analysis module using data c and data d. S4, repeat S1.

8. A computer device, including a memory and a processor, characterized in that, The memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor performs the method described in claim 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method of claim 7.

Images