A sewage treatment water quality intelligent prediction method and system based on discharge characteristics

Abstract

The application relates to the sewage treatment technical field, in particular to a sewage treatment water quality intelligent prediction method and system based on discharge characteristics, which comprises the following steps: acquiring a multidimensional data set and performing preliminary discharge characteristic analysis, extracting a discharge characteristic key mode, calculating a time correlation distribution of pollutant concentration variation and deeply mining a time correlation characteristic, fusing the two types of characteristics in multiple levels to obtain comprehensive prediction characteristics, and realizing intelligent prediction by using a pre-trained water quality prediction model. Through multidimensional data analysis and characteristic fusion, the water quality prediction precision and adaptability under a complex discharge scene are significantly improved.

Description

Technical Field

[0001] This invention belongs to the field of wastewater treatment technology, specifically a method and system for intelligent prediction of wastewater quality based on discharge characteristics. Background Technology

[0002] With the intelligent development of wastewater treatment technology, data-driven and machine learning-based water quality prediction methods have gradually become a research hotspot. These methods provide important support for the precise control of effluent water quality by analyzing multidimensional data in the wastewater treatment process. However, existing technologies still have shortcomings in feature extraction, model generalization ability, and targeted processing of wastewater discharge characteristics, affecting prediction accuracy and practical application effects. For example, the patent with publication number CN117273540B uses a CNN-LSTM network to perform water quality prediction by fusing multimodal data from wastewater surface images and water quality sensors. Although this improves prediction accuracy and reduces the number of sensors used, it does not fully consider the impact of wastewater discharge characteristics (such as pollutant types, concentration changes, and discharge patterns) on the prediction, which may lead to insufficient accuracy in complex discharge scenarios, and the cost of multimodal data acquisition and processing is high. Furthermore, while the patent with publication number CN111553468B, which uses an online monitoring system and a temporal convolutional network for water quality prediction, is innovative in data processing and model optimization, it lacks specialized modeling of wastewater discharge characteristics. This makes it difficult to accurately capture the dynamic changes of pollutants in scenarios with significant variations in discharge characteristics, leading to decreased prediction reliability. These issues indicate that existing technologies still need improvement in feature extraction, discharge characteristic modeling, and adaptability to complex scenarios. Therefore, there is an urgent need for an intelligent prediction method that can deeply mine wastewater discharge characteristics and combine them with advanced machine learning algorithms to improve the accuracy, robustness, and practical application value of water quality prediction. Summary of the Invention

[0003] This invention provides a method and system for intelligent prediction of wastewater treatment water quality based on emission characteristics. Its main purpose is to address the problems of insufficient accuracy and poor adaptability of existing water quality prediction methods in complex emission scenarios. To achieve the above objective, this invention provides an intelligent prediction method for wastewater treatment water quality based on emission characteristics, comprising: acquiring a multidimensional dataset related to emission characteristics during the wastewater treatment process; performing preliminary emission characteristic analysis on the multidimensional dataset to obtain preliminary emission characteristic analysis results; extracting key patterns from the preliminary emission characteristic analysis results to obtain key emission characteristic patterns; calculating the temporal correlation of pollutant concentration changes in the multidimensional dataset to obtain a temporal correlation distribution; performing deep information mining on the temporal correlation distribution to obtain temporal correlation features; fusing the temporal correlation features with the key emission characteristic patterns at multiple levels to obtain comprehensive prediction features; and using a pre-trained water quality prediction model to intelligently predict the effluent water quality of the wastewater treatment process based on the comprehensive prediction features to obtain water quality prediction results.

[0004] Optionally, a preliminary emission characteristic analysis is performed on the multidimensional dataset to obtain preliminary emission characteristic analysis results, including: identifying pollutant types in the multidimensional dataset to obtain the pollutant type distribution; detecting concentration fluctuations in the multidimensional dataset to obtain the concentration fluctuation distribution; and integrating the pollutant type distribution and concentration fluctuation distribution to obtain preliminary emission characteristic analysis results.

[0005] Optionally, pollutant type identification is performed on the multidimensional dataset to obtain the pollutant type distribution, including: establishing a pollutant type classification rule base; and labeling the pollutant types in the multidimensional dataset according to the classification rule base to obtain the pollutant type distribution.

[0006] Optionally, key patterns are extracted from the preliminary analysis results of emission characteristics to obtain key emission characteristic patterns, including: dividing the preliminary analysis results of emission characteristics into multiple sub-intervals according to the time series; statistically analyzing the pollutant concentration change trend in each sub-interval to obtain trend statistics; constructing a heat map based on the trend statistics, and extracting significant patterns from the heat map to obtain key emission characteristic patterns.

[0007] Optionally, deep information mining is performed on the time correlation distribution to obtain time correlation features, including: performing hierarchical analysis on the time correlation distribution to extract high-contribution segments in the time correlation distribution; performing pattern clustering on the high-contribution segments to obtain time correlation pattern clusters; and reconstructing the features of the time correlation pattern clusters to obtain time correlation features.

[0008] Optionally, the time-related features and key emission characteristic patterns are fused at multiple levels to obtain comprehensive prediction features, including: mapping the time-related features and key emission characteristic patterns to the same feature space to obtain a fused feature matrix; performing structured decomposition on the fused feature matrix to obtain low-dimensional sub-features; and performing weighted superposition processing on the low-dimensional sub-features to obtain comprehensive prediction features.

[0009] Optionally, before using a pre-trained water quality prediction model to intelligently predict the effluent quality of wastewater treatment based on comprehensive prediction features, the method further includes: collecting a large amount of historical datasets with water quality prediction labels, calculating the temporal correlation of the concentration changes of each pollutant in the historical datasets, and obtaining the historical temporal correlation distribution; performing deep information mining on the historical temporal correlation distribution to obtain historical temporal correlation features, and extracting key emission characteristic patterns from the historical datasets; fusing the historical temporal correlation features and the key emission characteristic patterns at multiple levels to obtain historical comprehensive prediction features; dividing the historical comprehensive prediction features into training and validation sets, using a machine learning model to predict water quality on the training set, and obtaining training prediction results; evaluating the error value between the training prediction results and the actual water quality labels; if the error value is greater than or equal to a preset threshold, adjusting the model parameters and re-executing the above steps of using a machine learning model to predict water quality on the training set; if the error value is less than the preset threshold, using a validation set to evaluate the model's performance; if the model fails the performance evaluation, optimizing the model structure and re-executing the above steps of using a validation set to evaluate the model's performance; if the model passes the performance evaluation, confirming that the model training is complete, and obtaining the water quality prediction model.

[0010] Optionally, the error value between the training prediction results and the actual water quality labels is evaluated, including: quantifying the error value based on the degree of deviation between the training prediction results and the actual water quality labels, wherein the error value reflects the prediction accuracy.

[0011] Optionally, a pre-trained water quality prediction model is used to intelligently predict the effluent quality of wastewater treatment based on comprehensive prediction features, thereby obtaining water quality prediction results. This includes: using the water quality prediction model to perform feature enhancement processing on the comprehensive prediction features to obtain enhanced features; using the water quality prediction model to dynamically model the enhanced features through a recursive network to obtain dynamic prediction results; and using the water quality prediction model to output the dynamic prediction results through a decision layer to obtain the final water quality prediction results.

[0012] To address the aforementioned issues, this invention also provides an intelligent wastewater quality prediction system based on emission characteristics. The system includes: a data acquisition module, which acquires a multidimensional dataset related to emission characteristics during wastewater treatment, performs preliminary emission characteristic analysis on the multidimensional dataset, and obtains preliminary emission characteristic analysis results; a pattern extraction module, which extracts key patterns from the preliminary emission characteristic analysis results to obtain key emission characteristic patterns; a time analysis module, which calculates the time correlation of pollutant concentration changes in the multidimensional dataset to obtain a time correlation distribution; a feature fusion module, which performs deep information mining on the time correlation distribution to obtain time correlation features, and performs multi-level fusion of the time correlation features with the key emission characteristic patterns to obtain comprehensive prediction features; and a water quality prediction module, which uses a pre-trained water quality prediction model to intelligently predict the effluent quality of the wastewater treatment process based on the comprehensive prediction features, and obtains water quality prediction results.

[0013] This invention acquires a multidimensional dataset related to emission characteristics during wastewater treatment, performs preliminary emission characteristic analysis on the dataset, and obtains preliminary analysis results, providing necessary basic data for subsequent prediction. Key patterns are extracted from the preliminary analysis results to obtain key emission characteristic patterns, improving the data's interpretability. The temporal correlation of pollutant concentration changes in the multidimensional dataset is calculated to obtain the temporal correlation distribution. Deep information mining is performed on the temporal correlation distribution to obtain temporal correlation features. These features are then fused with the key emission characteristic patterns at multiple levels to obtain comprehensive prediction features, enhancing feature representation capabilities. A pre-trained water quality prediction model is used to intelligently predict the effluent quality of wastewater treatment based on the comprehensive prediction features, yielding water quality prediction results. Therefore, the intelligent water quality prediction method and system for wastewater treatment based on emission characteristics proposed in this invention can solve the problems of insufficient accuracy and poor adaptability of existing water quality prediction methods in complex emission scenarios. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating a wastewater treatment water quality intelligent prediction method based on emission characteristics, provided in an embodiment of the present invention.

[0015] Figure 2 A schematic diagram of the process for extracting key emission characteristic patterns and mining time-related features according to an embodiment of the present invention;

[0016] Figure 3 This is a functional block diagram of a wastewater treatment water quality intelligent prediction system based on emission characteristics, provided in an embodiment of the present invention. Detailed Implementation

[0017] This invention provides a method and system for intelligent prediction of wastewater treatment water quality based on emission characteristics. Its core lies in achieving high-precision prediction of wastewater treatment effluent quality through multi-dimensional data analysis, key pattern extraction, and multi-level feature fusion, combined with a pre-trained water quality prediction model. The following is in conjunction with the appendix... Figure 1 To be continued Figure 3 The specific embodiments of the present invention will be described in detail below.

[0018] First of all, Figure 1 In the illustrated process, the implementation of this invention begins with a data acquisition module, which is responsible for acquiring multidimensional datasets related to the emission characteristics during wastewater treatment. These datasets include, but are not limited to, pollutant types, concentration changes, and time-series information. In practical applications, it is assumed that a wastewater treatment plant receives wastewater daily from industrial and residential areas, and its emission characteristics exhibit significant fluctuations. The data acquisition module collects pollutant concentrations in the wastewater in real time through a sensor network and stores this data as a multidimensional dataset. To ensure data integrity, the data acquisition module also detects and corrects outliers, for example, by using a sliding window algorithm to remove abnormal data points caused by equipment malfunctions.

[0019] Next, the data acquisition module performs preliminary emission characteristic analysis on the multidimensional dataset, obtaining preliminary emission characteristic analysis results. This process consists of three steps: pollutant type identification, concentration fluctuation detection, and pattern integration. In the pollutant type identification stage, a pollutant type classification rule base is established, containing information such as the chemical properties, sources, and detection thresholds of various pollutants. For example, for heavy metal ions such as lead and cadmium in industrial wastewater, the rule base defines their concentration ranges and detection methods. The pollutant types in the multidimensional dataset are labeled using the rule base, ultimately obtaining the pollutant type distribution. In the concentration fluctuation detection stage, Fourier transform is used to perform spectral analysis on the time series of pollutant concentrations, identifying the main frequency components of concentration fluctuations. For example, the periodic fluctuations in ammonia nitrogen concentration over a certain period may be related to intermittent industrial emissions. Finally, by integrating the pollutant type distribution and concentration fluctuation distribution, preliminary emission characteristic analysis results are formed. This result provides a foundation for subsequent key pattern extraction.

[0020] In the pattern extraction module, the preliminary emission characteristic analysis results are further processed to extract key patterns. Specifically, the preliminary emission characteristic analysis results are divided into multiple sub-intervals according to the time series, and the length of each sub-interval can be adjusted according to the actual application scenario. For example, in a typical application scenario, a day's data can be divided into 24 sub-intervals, each corresponding to one hour. Then, the pollutant concentration change trend of each sub-interval is statistically analyzed to obtain trend statistics. The trend statistics can be calculated using a linear regression model; for example, for a certain sub-interval, if the pollutant concentration shows an upward trend over time, its slope parameter is positive. A heatmap is constructed based on the trend statistics, and the color intensity of the heatmap indicates the severity of the pollutant concentration change. Through image processing techniques such as edge detection and region segmentation, significant patterns in the heatmap are extracted, ultimately yielding key emission characteristic patterns. These key patterns reflect the core laws of pollutant emissions, such as the concentrated emission behavior of certain pollutants within a specific time period. Meanwhile, the time analysis module is responsible for calculating the time correlation of the concentration changes of each pollutant in the multidimensional dataset, obtaining the time correlation distribution. The formula for calculating time correlation is:

[0021] in Let represent the correlation between pollutant i and pollutant j at time t, Cov represent the covariance, and σi and σj represent the standard deviations of pollutant i and pollutant j, respectively. This represents the concentration monitoring value of the i-th pollutant at time t. This represents the concentration monitoring value of pollutant j at time t. This formula can quantify the correlation between changes in the concentrations of different pollutants. For example, if the correlation between ammonia nitrogen concentration and total phosphorus concentration is high, it indicates that these two pollutants may originate from the same emission source. The temporal correlation distribution provides a basis for further in-depth information mining.

[0022] In the feature fusion module, the temporal correlation distribution undergoes hierarchical analysis, pattern clustering, and feature reconstruction to obtain temporal correlation features. The hierarchical analysis process involves decomposing the temporal correlation distribution layer by layer and extracting high-contribution segments. For example, using Principal Component Analysis (PCA), the temporal correlation distribution is projected into a low-dimensional space, retaining principal components with a cumulative contribution rate of over 95% as high-contribution segments. Subsequently, pattern clustering is performed on the high-contribution segments, using the K-means clustering algorithm to group similar segments into one class, resulting in temporal correlation pattern clusters. Finally, feature reconstruction is performed on the temporal correlation pattern clusters, transforming them into temporal correlation features using nonlinear mapping methods such as kernel functions. These features, along with key emission characteristic patterns, are mapped to the same feature space to form a fused feature matrix. The fused feature matrix undergoes structured decomposition to obtain low-dimensional sub-features, which are then weighted and superimposed to generate comprehensive prediction features. The formula for weighted superposition is... Where F is the comprehensive prediction feature, For the i-th low-dimensional sub-feature, The weighting coefficients are determined using cross-validation.

[0023] Before using water quality prediction models for intelligent prediction, the models need to be trained and optimized. Specifically, a large amount of historical datasets with water quality prediction labels are collected. These datasets include historical pollutant concentrations, emission characteristics, and corresponding effluent water quality indicators. The historical datasets undergo the same time correlation analysis and feature fusion processing as described above to obtain historical comprehensive prediction features. These historical comprehensive prediction features are divided into training and validation sets. Machine learning models such as support vector machines or random forests are used to predict water quality on the training set to obtain training prediction results. The error between the training prediction results and the actual water quality labels is evaluated. The error is calculated using the root mean square error (RMSE) formula, i.e.:

[0024] in This is the actual value. Here, n represents the predicted value, and n is the number of samples. If the error value is greater than or equal to a preset threshold, the model parameters are adjusted and the model is retrained. If the error value is less than the preset threshold, the model's performance is evaluated using a validation set. If the model fails the performance evaluation, the model structure is optimized and the evaluation is repeated. If the model passes the performance evaluation, the model training is confirmed to be complete, and a water quality prediction model is obtained.

[0025] In the actual prediction phase, the water quality prediction module processes the comprehensive prediction features using the trained water quality prediction model. First, feature enhancement is performed on the comprehensive prediction features by extracting higher-level abstract features through an autoencoder network, resulting in enhanced features. Then, a recurrent neural network (RNN) is used to dynamically model the enhanced features, capturing the long-term dependencies of pollutant concentration changes, thus obtaining dynamic prediction results. Finally, a decision layer outputs the dynamic prediction results, which can use a softmax function or other classifiers to determine the water quality level across multiple categories. For example, if the prediction result shows that ammonia nitrogen concentration exceeds the standard, the output water quality level is "unqualified".

[0026] Figure 2 This document demonstrates a detailed workflow for extracting key emission characteristic patterns and mining time-related features. In this workflow, the preliminary analysis results of emission characteristics are processed through time series segmentation, trend statistics, and heatmap construction to ultimately extract key patterns. The time-related distribution is then transformed into time-related features through hierarchical analysis, pattern clustering, and feature reconstruction. These features and key patterns are then fused at multiple levels in the feature fusion module to generate comprehensive predictive features.

[0027] Figure 3 This is a functional block diagram of a wastewater treatment water quality intelligent prediction system based on emission characteristics. The system includes a data acquisition module, a pattern extraction module, a time analysis module, a feature fusion module, and a water quality prediction module. The data acquisition module is responsible for acquiring multidimensional datasets and performing preliminary analysis; the pattern extraction module extracts key emission characteristic patterns; the time analysis module calculates the time correlation distribution; the feature fusion module generates comprehensive prediction features; and the water quality prediction module uses the trained model to perform water quality prediction. The modules collaborate with each other through data flow and control flow to jointly complete the water quality prediction task.

[0028] In summary, this invention acquires a multidimensional dataset related to emission characteristics during wastewater treatment, performs preliminary emission characteristic analysis, extracts key patterns, analyzes temporal correlations, and fuses multi-level features. Finally, it utilizes a pre-trained water quality prediction model to achieve intelligent prediction of wastewater effluent quality. This method not only solves the problems of insufficient accuracy and poor adaptability of existing water quality prediction methods in complex emission scenarios but also provides technical support for the intelligent management of wastewater treatment plants.

Claims

1. A method for intelligent prediction of wastewater treatment water quality based on emission characteristics, characterized in that, The method includes: A multidimensional dataset related to emission characteristics during wastewater treatment is obtained. A preliminary emission characteristic analysis is performed on the multidimensional dataset. The preliminary emission characteristic analysis is carried out through three steps: pollutant type identification, concentration fluctuation detection, and pattern integration, and the preliminary emission characteristic analysis results are obtained. Key patterns are extracted from the preliminary analysis results of the emission characteristics to obtain key emission characteristic patterns. The preliminary analysis results of the emission characteristics are divided into sub-intervals according to the time series. The pollutant concentration change trends in each sub-interval are statistically analyzed and a heat map is constructed. Significant patterns are extracted using image processing technology. The temporal correlation of the concentration changes of each pollutant in the multidimensional dataset is calculated to obtain the temporal correlation distribution. Deep information mining is performed on the time correlation distribution to obtain time correlation features. The time correlation distribution is then subjected to hierarchical analysis, pattern clustering, and feature reconstruction. The K-means clustering algorithm is used to obtain time correlation pattern clusters. Feature reconstruction is performed through kernel functions. The time correlation features are then fused with the key emission characteristic patterns at multiple levels to obtain comprehensive prediction features. The pre-trained water quality prediction model is used to intelligently predict the effluent quality of wastewater treatment based on the comprehensive prediction features, and the water quality prediction results are obtained. Before using the pre-trained water quality prediction model to intelligently predict the effluent quality of wastewater treatment based on the comprehensive prediction features, the method further includes: A large amount of historical datasets with water quality prediction labels were collected, and the temporal correlation of the concentration changes of each pollutant in the historical datasets was calculated to obtain the historical temporal correlation distribution. Deep information mining is performed on the historical time correlation distribution to obtain historical time correlation features and extract key emission characteristics of the historical dataset. The historical time correlation features are fused with the key emission characteristic patterns at multiple levels to obtain historical comprehensive prediction features. The historical comprehensive prediction features are divided into a training set and a validation set. A machine learning model is used to predict water quality on the training set to obtain the training prediction results. Evaluate the error between the training prediction results and the actual water quality labels; If the error value is greater than or equal to the preset threshold, the above steps of using the machine learning model to predict water quality on the training set are re-executed after adjusting the model parameters. If the error value is less than a preset threshold, the model performance is evaluated using the validation set. If the model fails the performance evaluation, the above steps of evaluating the model using the validation set will be repeated after optimizing the model structure. If the model passes the performance evaluation, the model training is confirmed to be complete, and a water quality prediction model is obtained.

2. The intelligent prediction method for wastewater treatment water quality based on emission characteristics as described in claim 1, characterized in that, The preliminary emission characteristic analysis of the multidimensional dataset yields preliminary emission characteristic analysis results, including: Pollutant type identification is performed on the multidimensional dataset to obtain the pollutant type distribution; Concentration fluctuation detection was performed on the multidimensional dataset to obtain the concentration fluctuation distribution; By integrating the distribution of pollutant types and the distribution of concentration fluctuations, preliminary analysis results of emission characteristics are obtained.

3. The intelligent prediction method for wastewater treatment water quality based on emission characteristics as described in claim 2, characterized in that, The step of identifying pollutant types from the multidimensional dataset to obtain the pollutant type distribution includes: Establish a database of rules for classifying pollutant types; The pollutant types in the multidimensional dataset are labeled according to the classification rule base to obtain the distribution of pollutant types.

4. The intelligent prediction method for wastewater treatment water quality based on emission characteristics as described in claim 1, characterized in that, The preliminary analysis results of the emission characteristics are used to extract key patterns, resulting in key emission characteristic patterns, including: The preliminary analysis results of the emission characteristics were divided into multiple sub-intervals according to the time series. The trend of pollutant concentration changes in each sub-interval is statistically analyzed to obtain the trend statistics results; A heat map is constructed based on the trend statistics, and significant patterns are extracted from the heat map to obtain key emission characteristic patterns.

5. The intelligent prediction method for wastewater treatment water quality based on emission characteristics as described in claim 1, characterized in that, The process of performing deep information mining on the time correlation distribution to obtain time correlation features includes: The time correlation distribution is analyzed hierarchically to extract high-contribution segments from the time correlation distribution. Pattern clustering is performed on the high-contribution segments to obtain time-related pattern clusters; The time-related pattern clusters are reconstructed to obtain time-related features.

6. The intelligent prediction method for wastewater treatment water quality based on emission characteristics as described in claim 1, characterized in that, The process of fusing the time-related features with the key emission characteristic patterns at multiple levels to obtain comprehensive prediction features includes: The time-related features and the key emission characteristic patterns are mapped to the same feature space to obtain a fused feature matrix; The fused feature matrix is ​​structurally decomposed to obtain low-dimensional sub-features; The low-dimensional sub-features are weighted and superimposed to obtain the comprehensive prediction features.

7. The intelligent prediction method for wastewater treatment water quality based on emission characteristics as described in claim 1, characterized in that, The evaluation of the error between the training prediction results and the actual water quality labels includes: The error value is quantified based on the degree of deviation between the training prediction results and the actual water quality labels, wherein the error value reflects the prediction accuracy.

8. The intelligent prediction method for wastewater treatment water quality based on emission characteristics as described in claim 1, characterized in that, The method of using a pre-trained water quality prediction model to intelligently predict the effluent quality of wastewater treatment based on the comprehensive prediction features, and obtaining water quality prediction results, includes: The comprehensive prediction features are enhanced using the water quality prediction model to obtain enhanced features; The enhanced features are dynamically modeled using a recursive network through the water quality prediction model to obtain dynamic prediction results. The water quality prediction model is used to output the dynamic prediction results through a decision-making layer, thus obtaining the water quality prediction results.

9. A wastewater treatment water quality intelligent prediction system based on emission characteristics, characterized in that, The system includes: The data acquisition module acquires a multidimensional dataset related to emission characteristics during the wastewater treatment process, performs preliminary emission characteristic analysis on the multidimensional dataset, and conducts preliminary emission characteristic analysis through three steps: pollutant type identification, concentration fluctuation detection, and pattern integration, to obtain preliminary emission characteristic analysis results. The pattern extraction module extracts key patterns from the preliminary analysis results of emission characteristics to obtain key emission characteristic patterns. It divides the preliminary analysis results of emission characteristics into sub-intervals according to time series, statistically analyzes the pollutant concentration change trends in each sub-interval and constructs a heat map, and extracts significant patterns through image processing technology. The time analysis module calculates the time correlation of the concentration changes of each pollutant in the multidimensional dataset and obtains the time correlation distribution. The feature fusion module performs deep information mining on the time correlation distribution to obtain time correlation features, performs hierarchical analysis, pattern clustering and feature reconstruction on the time correlation distribution, uses the K-means clustering algorithm to obtain time correlation pattern clusters, performs feature reconstruction through kernel functions, and fuses the time correlation features with the key emission characteristic patterns at multiple levels to obtain comprehensive prediction features. The water quality prediction module uses a pre-trained water quality prediction model to intelligently predict the quality of wastewater effluent based on the comprehensive prediction features, and obtains the water quality prediction results. Before using the pre-trained water quality prediction model to intelligently predict the effluent quality of wastewater treatment based on the comprehensive prediction features, the method further includes: collecting a large amount of historical datasets with water quality prediction labels, calculating the temporal correlation of the concentration changes of each pollutant in the historical datasets, and obtaining the historical temporal correlation distribution. Deep information mining is performed on the historical time correlation distribution to obtain historical time correlation features and extract key emission characteristics of the historical dataset. The historical time correlation features are fused with the key emission characteristic patterns at multiple levels to obtain historical comprehensive prediction features. The historical comprehensive prediction features are divided into a training set and a validation set. A machine learning model is used to predict water quality on the training set to obtain the training prediction results. Evaluate the error between the training prediction results and the actual water quality labels; If the error value is greater than or equal to the preset threshold, the above steps of using the machine learning model to predict water quality on the training set are re-executed after adjusting the model parameters. If the error value is less than a preset threshold, the model performance is evaluated using the validation set. If the model fails the performance evaluation, the above steps of evaluating the model using the validation set will be repeated after optimizing the model structure. If the model passes the performance evaluation, the model training is confirmed to be complete, and a water quality prediction model is obtained.

Images