Intelligent water quality regulation and control system and method based on multi-parameter real-time monitoring

Abstract

The invention discloses an intelligent water quality regulation and control system and method based on multi-parameter real-time monitoring, and belongs to the technical field of water quality monitoring. According to the intelligent water quality regulation and control system, the water quality data of the water body is obtained in real time through the multi-parameter sensor array, and the monitoring regulation and control server can quickly generate an abnormal report and a water quality regulation and control scheme. The data processing module performs feature extraction on the water quality data to obtain target features; the water quality evaluation module is used for accurately evaluating the water quality by using a pre-trained deep neural network model; the abnormity identification module can timely judge whether the water quality has a pollution risk and generate an abnormity report; and the regulation and control module generates a water quality regulation and control scheme by adopting a multi-objective optimization algorithm. The system realizes real-time performance, accuracy and intelligence of water quality monitoring, can quickly respond to water quality changes, effectively reduces pollution risks, and improves the efficiency and effect of water quality regulation and control.

Description

Technical Field

[0001] The present application relates to the field of water quality monitoring technology, and in particular to a water quality intelligent control system and method based on real-time monitoring of multiple parameters. Background Art

[0002] Water quality monitoring and control have important applications in water environment management, industrial water treatment, and wastewater treatment. Traditional water quality monitoring methods rely primarily on manual sampling and laboratory analysis, which suffer from issues such as insufficient real-time performance and limited parameter detection. In recent years, with the advancement of sensor technology, data processing techniques, and intelligent algorithms, water quality monitoring has gradually become more automated and intelligent.

[0003] Despite progress in water quality monitoring technology, several challenges remain. First, traditional water quality monitoring methods lack real-time performance and cannot meet the demands of refined management of complex aquatic environments. Second, the single parameter tested cannot fully reflect water quality. Furthermore, regulatory responses lag, making real-time water quality control difficult. These issues result in suboptimal water quality monitoring and control, hindering effective responses to water pollution and ecological damage.

[0004] In the process of realizing the present invention, the inventors found that the prior art has the following deficiencies:

[0005] Traditional water quality monitoring methods typically rely on manual sampling and laboratory analysis, resulting in long monitoring cycles and an inability to reflect real-time changes in water quality. For example, manual sampling requires regular on-site water sample collection and laboratory analysis, a process that can take hours or even days. In situations where water quality changes rapidly, such as industrial wastewater discharges and stormwater runoff, traditional monitoring methods are unable to provide timely data support, resulting in delayed water quality management and an inability to effectively respond to water quality changes.

[0006] Traditional water quality monitoring typically focuses on only a few key indicators, failing to fully reflect water quality. For example, when monitoring eutrophication, traditional methods focus solely on the levels of nutrients like nitrogen and phosphorus, while ignoring crucial information such as algae abundance and the toxins they secrete. This single parameter approach fails to accurately assess the overall state of water quality, easily leading to incomplete and inaccurate water quality management decisions.

[0007] Traditional water quality control methods are often based on empirical rules and lack real-time data support, making them incapable of precise and rapid regulation. For example, when addressing eutrophication, traditional methods routinely administer a fixed amount of flocculants or disinfectants based on historical experience and regular monitoring of water quality indicators. However, due to the lack of real-time monitoring and intelligent analysis, this approach is unable to promptly adjust control measures based on real-time changes in water quality, resulting in poor control effectiveness and even secondary pollution. Summary of the Invention

[0008] In view of this, an embodiment of the present invention provides a water quality intelligent control system and method based on real-time monitoring of multiple parameters to solve at least one of the above technical problems.

[0009] To achieve the above objectives, in a first aspect, the present invention provides an intelligent water quality control system based on real-time multi-parameter monitoring, comprising:

[0010] A multi-parameter sensor array and a monitoring and control server, wherein the multi-parameter sensor array is arranged in a water body and is used to obtain raw water quality data of the water body;

[0011] The monitoring and control server is configured to receive the raw water quality data from the multi-parameter sensor array, determine whether there is a pollution risk based on the raw water quality data, and if it is determined that there is a pollution risk, generate an abnormality report and a water quality evaluation result based on the raw water quality data, and generate a water quality control plan based on the abnormality report and the water quality evaluation result;

[0012] The monitoring and control server includes:

[0013] A data processing module is used to obtain raw water quality data, preprocess the raw water quality data to obtain standard water quality data, and perform feature extraction on the standard water quality data to obtain multiple target features;

[0014] A water quality evaluation module, configured to input a plurality of target features into a pre-trained deep neural network model to obtain a water quality evaluation result;

[0015] An anomaly recognition module is used to determine whether the water quality evaluation result has a pollution risk based on the standard water quality data and a preset safety interval. When it is determined that there is a pollution risk, an anomaly recognition algorithm is used to process the standard water quality data to obtain an anomaly report;

[0016] The control module is used to generate a water quality control plan using a multi-objective optimization algorithm according to the water quality evaluation result and the abnormality report.

[0017] In a second aspect, the present invention provides a method for intelligent water quality control based on real-time monitoring of multiple parameters, comprising the following steps:

[0018] Acquiring original water quality data, preprocessing the original water quality data to obtain standard water quality data, and performing feature extraction on the standard water quality data to obtain multiple target features;

[0019] Inputting the plurality of target features into a pre-trained deep neural network model to obtain a water quality evaluation result;

[0020] Determining whether the water quality evaluation result has a pollution risk based on the standard water quality data and a preset safety interval, and when it is determined that there is a pollution risk, processing the standard water quality data using an anomaly recognition algorithm to obtain an anomaly report;

[0021] A water quality control plan is generated using a multi-objective optimization algorithm based on the water quality evaluation results and the abnormality report.

[0022] According to a third aspect, an electronic device is provided, comprising:

[0023] one or more processors;

[0024] a storage device for storing one or more programs,

[0025] When the one or more programs are executed by the one or more processors, the one or more processors implement a water quality intelligent control method based on multi-parameter real-time monitoring as described in the second aspect.

[0026] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, an intelligent water quality control method based on real-time monitoring of multiple parameters as described in the second aspect is implemented.

[0027] The above technical solution has the following beneficial technical effects:

[0028] The intelligent water quality control system of the present invention uses a multi-parameter sensor array to acquire real-time water quality data. The monitoring and control server can quickly generate anomaly reports and water quality control plans. The data processing module extracts features from the water quality data to obtain target features. The water quality assessment module uses a pre-trained deep neural network model to accurately evaluate water quality. The anomaly recognition module can promptly determine whether water quality poses a pollution risk and generate an anomaly report. The control module uses a multi-objective optimization algorithm to generate a water quality control plan. This system achieves real-time, accurate, and intelligent water quality monitoring, can quickly respond to water quality changes, effectively reduce pollution risks, and improve the efficiency and effectiveness of water quality control. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The accompanying drawings are provided for a better understanding of the present invention and are not intended to limit the present invention.

[0030] Figure 1 This is a structural block diagram of a water quality intelligent control system based on multi-parameter real-time monitoring of the present invention;

[0031] Figure 2 This is a structural block diagram of a data processing module in a water quality intelligent control system based on multi-parameter real-time monitoring of the present invention;

[0032] Figure 3is a schematic diagram of a deep neural network in the present invention;

[0033] Figure 4 This is a structural block diagram of an abnormality recognition module in a water quality intelligent control system based on multi-parameter real-time monitoring of the present invention;

[0034] Figure 5 This is a structural block diagram of a control module in a water quality intelligent control system based on multi-parameter real-time monitoring of the present invention;

[0035] Figure 6 This is a structural block diagram of an execution module in a water quality intelligent control system based on multi-parameter real-time monitoring of the present invention;

[0036] Figure 7 This is a flow chart of a water quality intelligent control method based on multi-parameter real-time monitoring of the present invention;

[0037] Figure 8 Schematic diagram of the structure of a computer system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The following description of exemplary embodiments of the present invention is made in conjunction with the accompanying drawings, in which various details of the embodiments of the present invention are included to facilitate understanding. These details should be considered as merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0039] Example 1

[0040] like Figure 1 As shown, this embodiment provides a water quality intelligent control system based on real-time monitoring of multiple parameters, including:

[0041] A multi-parameter sensor array and a monitoring and control server, wherein the multi-parameter sensor array is arranged in a water body and is used to obtain raw water quality data of the water body;

[0042] The monitoring and control server is configured to receive the raw water quality data from the multi-parameter sensor array, determine whether there is a pollution risk based on the raw water quality data, and if it is determined that there is a pollution risk, generate an abnormality report and a water quality evaluation result based on the raw water quality data, and generate a water quality control plan based on the abnormality report and the water quality evaluation result;

[0043] The monitoring and control server includes:

[0044] A data processing module is used to obtain raw water quality data, preprocess the raw water quality data to obtain standard water quality data, and perform feature extraction on the standard water quality data to obtain multiple target features;

[0045] A water quality evaluation module, configured to input a plurality of target features into a pre-trained deep neural network model to obtain a water quality evaluation result;

[0046] An anomaly recognition module is used to determine whether the water quality evaluation result has a pollution risk based on the standard water quality data and a preset safety interval. When it is determined that there is a pollution risk, an anomaly recognition algorithm is used to process the standard water quality data to obtain an anomaly report;

[0047] The control module is used to generate a water quality control plan using a multi-objective optimization algorithm according to the water quality evaluation result and the abnormality report.

[0048] In this embodiment, the multi-parameter sensor array includes any of a surface plasmon resonance sensor, an ion-sensitive field-effect transistor sensor, a chlorophyll sensor, and a heavy metal ion concentration sensor. The surface plasmon resonance sensor is located downstream of the industrial wastewater outlet to monitor the concentration of organic pollutants. The ion-sensitive field-effect transistor sensor is located in the ammonia nitrogen pollution area to monitor the ammonia nitrogen concentration in the water. The chlorophyll sensor is located in the photic zone of the water to monitor the chlorophyll concentration associated with algal biomass. The multi-parameter sensor array uploads raw water quality data of the water to the monitoring and control server via a communication module. Data upload can be independent transmission by each sensor, or unified transmission via relay or aggregation. The communication module can be physically integrated into the sensor or located in an external gateway.

[0049] Specifically, the multi-parameter sensor array includes, but is not limited to, any combination of pH sensors, dissolved oxygen sensors, conductivity sensors, turbidity sensors, surface plasmon resonance sensors, ion-sensitive field-effect transistor sensors, chlorophyll sensors, or heavy metal ion concentration sensors. The pH sensors, dissolved oxygen sensors, and conductivity sensors are positioned at different depths within the water body. The surface plasmon resonance sensor is used to detect the concentration of organic pollutants, the ion-sensitive field-effect transistor sensor is used to monitor ammonia nitrogen, and the chlorophyll sensor is used to monitor algae content. Each sensor is positioned according to the characteristics of the water body. The pH sensor, dissolved oxygen sensor, and conductivity sensor are installed at the water surface and at different depths to monitor changes in water pH, oxygen content, and electrical conductivity, reflecting the water's self-purification capacity and pollution status. Turbidity sensors are primarily deployed in areas with high water flow or at water intakes to detect the concentration of suspended particulate matter in the water and assess sediment deposition and pollutant diffusion. Heavy metal ion concentration sensors are installed near pollution sources or at water sources to monitor potential heavy metal pollution. The surface plasmon resonance sensor can be used for real-time online monitoring of organic pollutants at industrial wastewater outlets. The ion-sensitive field-effect transistor sensor is suitable for ammonia nitrogen monitoring in aquaculture farms and sewage treatment plants. The chlorophyll sensor is used in lakes and reservoirs to assess algae reproduction and predict the risk of algal blooms. The multi-parameter sensor array collects water quality parameters in real time at a high frequency (e.g., 1 to 10 times per second) to ensure the continuity and real-time nature of the data, and uploads the collected data to a monitoring and control server to support subsequent analysis and decision-making. The selection of sensors in the multi-parameter sensor array is based on the water body type (e.g., river, lake, or sewage treatment plant, etc.) and the monitoring target (e.g., pollution monitoring, water quality monitoring, or ecological assessment, etc.).

[0050] The pH sensor is placed in the shallowest water layer (0.5m). Because pH is significantly affected by light, algal activity, and temperature changes, and the pH of surface water fluctuates, installing it in a shallower water layer helps better reflect environmental changes. The dissolved oxygen sensor is installed at a slightly deeper depth (1m). Because dissolved oxygen primarily comes from atmospheric exchange and photosynthesis, dissolved oxygen is higher in surface waters but gradually decreases with depth. Therefore, it is placed between the surface and mid-depth to accurately monitor the distribution of oxygen concentration in the water. If necessary, multiple sensors are deployed at different depths. The conductivity sensor is placed in the deepest water layer (1-5m). Because water conductivity is primarily determined by the concentration of dissolved salts and minerals, and surface waters are easily diluted by rainwater and runoff, resulting in larger data fluctuations, placing it in a deeper water layer helps obtain more stable and representative measurements.

[0051] Specifically, the water quality data is not limited to the monitoring data of the above-mentioned sensors, but also includes any one or more combinations of microplastic concentration, antibiotic concentration, algal toxin concentration, total organic carbon content and water redox potential.

[0052] Specifically, the concentration of microplastics is analyzed by a filtration method combined with a microscope or spectrophotometry (for example, Raman spectroscopy and Fourier transform infrared spectroscopy). The filtration method combined with a microscope or spectrophotometry can identify the morphology and distribution of microplastics, thereby accurately estimating their concentration in the water body. The antibiotic concentration is collected by liquid chromatography or liquid chromatography-mass spectrometry. The algal toxin concentration is collected by enzyme-linked immunosorbent assay, liquid chromatography and mass spectrometry, which can accurately detect harmful algal toxins in water bodies, especially when algal blooms occur, and can quickly identify potential risks. The total organic carbon content is a comprehensive indicator for measuring all organic matter in water, and is determined by ultraviolet-visible spectrophotometry or high-temperature combustion method. The redox potential is measured by an ambient light sensor, such as a silver chloride electrode or a gold electrode.

[0053] Specifically, before water quality monitoring begins, all sensors in the multi-parameter sensor array are inactive. Each sensor undergoes parameter configuration and self-tests. For example, the pH sensor undergoes zero point and slope calibration, the dissolved oxygen sensor's membrane integrity is checked, and the conductivity sensor's temperature compensation is tested. The system automatically detects the sensor's operating status and, if any anomalies are detected, such as drift or signal loss, initiates adjustments or prompts for maintenance.

[0054] Specifically, during the monitoring process, all sensors in the multi-parameter sensor array are initialized simultaneously. Data collection begins when all sensors are confirmed to be properly connected, their communication interfaces are correct, and their timestamps are synchronized, ensuring consistent data collection. This simultaneous data collection ensures that all water quality data is collected at the same time, avoiding errors caused by data lag or differences in collection time, and improving analysis accuracy.

[0055] Specifically, data collected by all sensors in the multi-parameter sensor array is managed according to predefined storage rules. Local caching is used for short-term storage to prevent data loss, while long-term storage utilizes cloud storage or a server database to support historical data backtracking and trend analysis. Abnormal data is prioritized for storage to facilitate subsequent anomaly detection and early warning analysis.

[0056] Specifically, if Figure 2 As shown, the data processing module may include:

[0057] A raw water quality data acquisition unit, configured to acquire raw water quality data from the multi-parameter sensor array;

[0058] a denoising unit, configured to remove noise from the original water quality data using a filtering algorithm to obtain first preprocessed data;

[0059] an outlier processing unit, configured to detect outliers in the first preprocessed data using a machine learning algorithm, and repair the outliers using a variational autoencoder to obtain second preprocessed data;

[0060] a standardization unit, configured to process the second pre-processed data using a normalization algorithm or a standardization algorithm to obtain standard water quality data;

[0061] The feature extraction unit is used to extract features from the standard water quality data using at least one of a time series decomposition method, a wavelet transform method, or a long short-term memory network to obtain multiple target features.

[0062] Specifically, in the original acquisition unit, the original water quality data is directly acquired from a multi-parameter sensor array.

[0063] Specifically, in the denoising unit, since the multi-parameter sensor array is easily affected by factors such as electromagnetic interference, water flow fluctuations or temperature changes in a complex water environment, the original water quality data contains noise. When performing denoising, one of wavelet transform, Kalman filter or median filter is used. The wavelet transform is used to remove high-frequency noise while retaining mutation characteristics and improving data smoothness. The Kalman filter is used for real-time data stream processing, predicting the current value based on historical data, and correcting the noise effect to improve data accuracy. The median filter is used for sudden abnormal values (for example, sudden changes in conductivity) to reduce the impact of pulse noise on data stability.

[0064] Specifically, the outlier is caused by sensor failure, environmental mutation or instantaneous impact of pollutants. In the outlier processing unit, outlier processing is performed by a method based on the interquartile range algorithm, the Z-score algorithm, the machine learning algorithm (such as the isolation forest algorithm or the local anomaly factor algorithm, etc.), the interpolation method or the elimination method. The interquartile range algorithm and the Z-score algorithm are used to identify abnormal data points exceeding a preset range. The machine learning algorithm is used to detect abnormal points in complex data patterns. The interpolation method is used to perform linear interpolation or mean difference correction on mild abnormal data. For serious abnormal data, the elimination method is used to directly delete it to prevent individual abnormal data from affecting the overall analysis results.

[0065] In addition, in the feature extraction unit, the target features include any multiple of water pH, dissolved oxygen concentration, conductivity, turbidity, heavy metal ion concentration, chemical oxygen demand, biological oxygen demand, ammonia nitrogen and chlorophyll concentration; the target features also include multiple time features and environmental features, and the time features include short-term change features, periodic change features and anomaly detection features.

[0066] Specifically, in the feature extraction unit, the target features include any one of water pH, dissolved oxygen concentration, conductivity, turbidity, heavy metal ion concentration (e.g., lead, cadmium, or mercury), chemical oxygen demand, biological oxygen demand, ammonia nitrogen, and chlorophyll concentration. Since changes in water quality data are time-dependent, it is necessary to construct time features to facilitate subsequent trend analysis and prediction. Therefore, the target features also include multiple time features, including short-term change features, periodic change features, and anomaly detection features. The target features may also include environmental features.

[0067] In some embodiments, intelligent outlier repair is performed based on a variational autoencoder (VAE) and adaptive interpolation.

[0068] During water quality monitoring, sensor data can generate outliers due to faults, sudden environmental changes, or pollutant impacts. To ensure data accuracy and integrity, this embodiment uses a variational autoencoder to detect and correct outliers. It also incorporates adaptive interpolation to further optimize the smoothness of the corrected data, ensuring that the data conforms to the natural trends of water quality changes. This method not only identifies and completes outliers but also avoids the distortion associated with traditional interpolation methods, making it suitable for complex environments with large water quality fluctuations. Its working process is as follows:

[0069] The first step is data preprocessing. First, outliers in the water quality data are identified using anomaly detection methods such as the IQR (interquartile range), Z-score, or isolation forest, and their specific locations are recorded. Next, normalization (such as min-max normalization) is used to scale the data to [0, 1] or [-1, 1] to meet the input requirements of the deep learning model.

[0070] In the second step, VAE performs outlier repair. After detecting abnormal data, VAE generates reasonable data completion values through the encoding-decoding mechanism. First, the encoder uses LSTM (Long Short-Term Memory Network) or MLP (Multi-Layer Perceptron) to extract the key features of water quality data and map them to the latent space to learn the pattern and distribution characteristics of water quality data. Then, in the latent variable sampling process, VAE calculates the mean (μ) and variance (σ) of the data. 2 ) and generates the latent variable z by sampling from a standard normal distribution, ensuring that the generated data points conform to the historical data distribution. Finally, the decoder uses a reverse LSTM or MLP structure to generate the most reasonable water quality data points from the latent variable z and replace outliers, ensuring that the corrected data trend conforms to the natural variation of water quality.

[0071] The third step is to perform adaptive interpolation optimization. Although the corrected data generated by the VAE is reasonable, it may have small fluctuations. Therefore, adaptive interpolation is further optimized to make the corrected data smoother and more stable. For abnormal data within a short time range, the sliding window mean interpolation method is used to perform a weighted average of the VAE-generated values at the previous and next n time points to make the data transition smoother and reduce sudden changes. The formula is as follows:

[0072]

[0073] Among them, w i is the weight, and the data points closer to the current moment have greater weight.

[0074] For abnormal data with a longer time span, local trend interpolation (Trend-based Interpolation), such as local linear regression or spline interpolation, is used to ensure that the supplemented data conforms to the historical trend of water quality without over-smoothing and losing important information. For example, when the water flow is temporarily affected by a pollution source, resulting in abnormal pH or conductivity, local trend interpolation can ensure that the corrected data still conforms to the changing pattern of the water body itself.

[0075] This solution avoids the data distortion problem caused by traditional interpolation methods. VAE makes the corrected data more consistent with the real environment by learning the underlying patterns of water data, while the adaptive interpolation method further optimizes the smoothness of the completed data. Compared with the adversarial training of the Generative Adversarial Network (GAN), which requires high computational costs, VAE has lower computational complexity and is more suitable for real-time water quality monitoring systems. In addition, this method can adapt to different water environments, including still water environments (such as lakes and reservoirs) and flowing water bodies (such as rivers and sewage treatment plants). It can effectively repair abnormal water quality data in a variety of application scenarios, thereby improving the stability and data quality of the monitoring system.

[0076] Specifically, in the standardization unit, the original water quality data is scaled to the interval [0,1] by adopting a normalization algorithm, which is suitable for non-normally distributed data; the Z-score normalization algorithm is used for normally distributed data to convert the original water quality data into a standard distribution with a mean of 0 and a variance of 1, thereby improving robustness.

[0077] Specifically, in the feature extraction unit, the target features include any one of water pH, dissolved oxygen concentration, electrical conductivity, turbidity, heavy metal ion concentration (for example, lead, cadmium or mercury, etc.), chemical oxygen demand, biological oxygen demand, ammonia nitrogen and chlorophyll concentration. Since the changes in water quality data are time-related, it is necessary to construct time features to facilitate subsequent trend analysis and prediction. Therefore, the target features also include multiple time features, and the time features include short-term change features, periodic change features and anomaly detection features. The target features may also include environmental features. The time length corresponding to the short-term change feature is shorter than the time length corresponding to the periodic feature.

[0078] Specifically, the pH value of the water body is used to measure the acid-base characteristics of the water body. The dissolved oxygen concentration is used to reflect the oxygen-rich state of the water body. The electrical conductivity is used to measure the ion content in the water, reflecting the mineralization or pollution level of the water body. The turbidity is used to evaluate the transparency of the water body and indirectly reflect the concentration of particulate matter and the level of pollutants. The heavy metal ion concentration is used to identify industrial pollution or water quality problems in mining areas. The chemical oxygen demand is used to measure the content of organic pollutants in the water and is an important indicator of the degree of water pollution. The biological oxygen demand is used to reflect the consumption of oxygen in the water by the degradation of organic matter. The ammonia nitrogen is used to evaluate the nitrogen pollution level in the water body. The chlorophyll concentration is used to judge the reproduction of algae and monitor the degree of eutrophication.

[0079] Specifically, the short-term variation characteristics include, for example, the fluctuation trend of pH value and dissolved oxygen concentration within 24 hours. The periodic characteristics include, for example, seasonal water quality variation characteristics, which are combined with historical data for periodic modeling. The anomaly detection feature refers to the determination of whether the current data is abnormal based on statistical characteristics such as mean and variance over a period of time.

[0080] Specifically, the feature extraction unit is used to extract features from the standard water quality data using one or more methods including the time series decomposition method (Seasonal-Trend Decomposition, STL), the wavelet transform method (Wavelet Transform, WT) or the long short-term memory network (Long Short-Term Memory, LSTM) to obtain multiple target features. STL decomposition can break down the data into long-term trends (Trend), periodic components (Seasonal) and residuals (Residual), thereby facilitating the analysis of short-term and long-term change patterns. WT (Wavelet Transform) is used to analyze the changes in water quality data at different time scales, capturing mutation points and non-stationary patterns. It can detect mutation points and abnormal water quality events (such as instantaneous impacts of pollutants). LSTM is a variant of recurrent neural network (RNN) in deep learning, which is specifically used to process time series data and can learn long-term dependencies. Since water quality data is a time-dependent nonlinear dynamic system, LSTM can capture complex water quality change patterns better than traditional statistical methods.

[0081] If the water quality data has relatively simple characteristics (for example, only long-term trend changes), STL or LSTM can be used alone. However, in complex water quality monitoring scenarios (such as those with both long-term trends and sudden changes), the combined use of STL, WT, and LSTM can provide more comprehensive and accurate feature extraction capabilities.

[0082] In some embodiments, time series decomposition (STL), wavelet transform (WT), and long short-term memory (LSTM) networks are combined to extract long-term trends, cyclical changes, and short-term mutation information from water quality data, making water quality monitoring and prediction more accurate and intelligent. This method is suitable for scenarios such as water quality prediction, pollution monitoring, and water environment anomaly detection, and is particularly suitable for water environments where both long-term trends and short-term mutations need to be considered. It includes the following steps:

[0083] The first step is to perform time series decomposition using STL. First, the STL (Seasonal-Trend Decomposition) method is used to decompose the water quality time series data, breaking down the original water quality data into long-term trends (Trend), periodic components (Seasonal), and residuals (Residual). The long-term trend component represents the overall direction of change in water quality, such as the increase or decrease in pH, dissolved oxygen concentration, or conductivity over a longer time scale; the periodic component is used to capture the periodic fluctuations in water quality over time, such as diurnal temperature changes, seasonal precipitation, or the impact of human activities on water bodies; the residual component contains short-term fluctuation information that cannot be explained by trend or periodic components. Through STL decomposition, noise in water quality data can be effectively removed and a clearer pattern of water quality changes can be provided, making subsequent analysis more stable and accurate.

[0084] In the second step, wavelet transform (WT) is used to process the residual part decomposed by STL. Among the three components decomposed by STL, the residual part usually contains mutation information, random fluctuations and abnormal water quality events (such as instantaneous impact of pollutants). In order to further analyze and extract these short-term high-frequency changes, wavelet transform is used to decompose the residual part. Wavelet transform has good time-frequency analysis capabilities and can identify the change patterns of water quality data at different time scales. By decomposing the residual into different frequency components, it can effectively extract abnormal water quality characteristics in a short period of time, such as pollutant discharge events, sudden algae outbreaks or short-term abnormal water quality fluctuations. Ultimately, the high-frequency components after wavelet transform retain the key mutation information in the water quality data, while the low-frequency part is used to further smooth short-term noise and improve the interpretability of the data.

[0085] In the third step, the trend component and mutation information are fed into the LSTM for feature learning. After STL decomposition and wavelet transform processing, the long-term trend component (representing the long-term evolution of water quality) and the high-frequency mutation information (indicating short-term abnormal changes) extracted by the wavelet transform are fed into a long-short-term memory (LSTM) network for feature learning. As a deep learning model specifically designed for time series data analysis, the LSTM can effectively learn long-term dependencies in water quality data and identify patterns of change across different time scales. By combining long-term trend and mutation information, the LSTM can simultaneously capture both the overall trend of water quality changes and short-term abnormal fluctuations, improving the accuracy of water quality predictions. Furthermore, because the LSTM has the ability to memorize long-term dependencies, this method can also predict future water quality trends.

[0086] Specifically, the data processing module can be deployed in a multi-parameter sensor array through edge computing technology, completing part of the data cleaning and feature extraction at the sensor end, reducing the pressure on data transmission and cloud computing.

[0087] Specifically, if Figure 3 As shown, the deep neural network model may specifically include:

[0088] An input layer, configured to receive a plurality of target features;

[0089] A convolutional layer, used to obtain the correlation between the target features;

[0090] A time series analysis layer is used to predict the water quality change trend based on the target characteristics to obtain the water quality change trend; the time series analysis layer adopts a hybrid model consisting of a long short-term memory network and a transformer model;

[0091] A fully connected layer, comprising multiple hidden layers, each of which is provided with an activation function for adjusting the weight of each target feature;

[0092] The output layer is used to output water quality evaluation results, which include water quality grade, pollution degree and water quality change trend.

[0093] Specifically, in the input layer, the target features include pH, dissolved oxygen, conductivity, turbidity, heavy metal ion concentration, chemical oxygen demand, biological oxygen demand, ammonia nitrogen, or chlorophyll concentration. These target features are organized in a time series format and then divided into training, test, and validation sets, with the division ratio set according to actual conditions. Due to the different data ranges of the target features, they are uniformly normalized or standardized before entering the input layer, and the normalized data is mapped to [0, 1] to facilitate subsequent processing.

[0094] Specifically, the convolutional layer is set according to actual conditions to capture local correlations between target features, such as the mutual influence between pollutants.

[0095] Specifically, the time series analysis layer may adopt a long short-term memory network and / or a transformer model to learn the time series pattern of water quality data and improve the prediction accuracy of water quality change trends.

[0096] Specifically, the fully connected layer includes multiple hidden layers, each hidden layer is provided with a ReLU (Rectified Linear Unit) activation function, and each layer of neurons in the fully connected layer is connected to all neurons in the previous layer. This fully connected structure enables the layer to integrate all feature information in the input data. The input data can contain multiple target features (for example, pH value, dissolved oxygen or ammonia nitrogen, etc.) and time series data of water quality changes. The fully connected layer can comprehensively integrate these feature information of different dimensions and different time points, thereby capturing the overall state and changing trend of water quality. The fully connected layer automatically evaluates the importance of different input features by learning the weight relationship between features. For complex water quality problems, some features are more influential than other features. For example, when evaluating the degree of water pollution, the fully connected layer can learn to assign higher weights to target features such as heavy metal content, thereby forming a comprehensive target feature representation that reflects the relative importance of different features.

[0097] Multiple hidden layers can include primary hidden layers, intermediate hidden layers, and advanced hidden layers. Primary hidden layers (those close to the input layer) are primarily responsible for extracting low-level features from the data, such as basic patterns, edges, colors, and textures. In the context of water quality monitoring, the primary hidden layers capture some basic patterns from water quality sensor data. Intermediate hidden layers are used to learn more complex patterns based on the feature outputs of the previous layer, such as the relationships between different water quality parameters. The output of each layer further processes the features of the previous layer, gradually forming a higher-level abstract representation. Advanced hidden layers (those close to the output layer) learn deeper features of the data, such as a global understanding and pattern recognition of different water quality data. These features are multidimensional and comprehensive, helping the model determine water quality level, pollution level, and other factors.

[0098] Specifically, the output layer outputs include classification results, regression outputs, and trend prediction outputs. The classification output is implemented by setting a softmax function to output water quality grades (e.g., excellent, good, medium, poor, and extremely poor). The regression output is implemented by a Sigmoid activation function or a hyperbolic tangent activation function (Tanh) to output the degree of pollution (e.g., pollutant contribution score, major pollutants, etc.). The trend prediction output is implemented based on a time series analysis layer and a fully connected layer to output water quality change trends.

[0099] Specifically, in some embodiments, the time series analysis layer may adopt a hybrid model consisting of a long short-term memory network and a transformer model. This solution combines the Transformer and LSTM, giving full play to the advantages of both in time series modeling to improve the prediction accuracy and efficiency of water quality data. LSTM is suitable for short-term dependency learning and can effectively capture the changing trends of water quality data within a short time scale, such as the fluctuations of pH, dissolved oxygen, and conductivity over the past few days. However, when processing longer time series, LSTM may encounter problems such as gradient vanishing or difficulty learning long-term dependencies. In contrast, the Transformer can more comprehensively understand the long-term change patterns of water quality data through the self-attention mechanism, and is particularly suitable for analyzing long-term trends, such as seasonal changes and long-term pollution diffusion trends. Therefore, this embodiment adopts a hybrid architecture of LSTM and Transformer, with LSTM first processing short-term dependencies and then Transformer performing long-term trend modeling. Finally, the features of the two are merged through the fusion layer to obtain more stable and accurate prediction results.

[0100] The architecture of this hybrid model consists of three main stages. In the first stage (LSTM processing short-term dependencies), an LSTM network is used to model short-term fluctuations in water quality data and extract local time-dependent features, such as recent pH patterns and short-term pollution shocks. In the second stage (Transformer processing long-term dependencies), the short-term features generated by the LSTM are fed into the Transformer for global time series modeling, capturing complex trends over longer time scales, such as seasonal patterns in water quality and changes in watershed pollution. In the third stage (fusion layer), the outputs of the LSTM and Transformer are fused using a fully connected layer (FC) or a gating mechanism (such as a GRU structure) to enhance feature representation and improve prediction stability.

[0101] The main advantage of this hybrid model is that compared with using LSTM alone, this hybrid model can process longer time series and improve the accuracy of water quality prediction; compared with using Transformer alone, this method is more efficient and avoids the problem of excessive computational complexity of Transformer when processing very long time series, because LSTM first extracts short-term features, reducing the amount of data that Transformer needs to process; it is suitable for short-term water quality prediction (for example, 1 to 7 days) and long-term water quality prediction (for example, 30 days), meeting the needs of predicting water quality changes at different time scales.

[0102] Specifically, during training, the deep neural network model uses a cross-entropy loss function for classification optimization, and a mean squared error loss function for pollution level and trend prediction. The optimizer uses Adaptive Moment Estimation (Adam), and the learning rate uses a dynamic adjustment strategy (e.g., Reduce LR On Plateau). The deep neural network model uses an early stopping strategy to prevent overfitting, and K-fold cross-validation to improve stability.

[0103] Specifically, the test set is used to evaluate the performance of the deep neural network model, using metrics such as classification accuracy, mean squared error, and goodness of fit. When deployed, the deep neural network model is integrated into an edge computing system, processing water quality data in real time and providing intelligent evaluation results on water quality status, pollution levels, and trend forecasts.

[0104] Specifically, the water quality evaluation results are stored or uploaded to the database as historical data after being generated. The water quality evaluation results display the water quality status through a visualization platform, including historical water quality change trends, pollution factor contribution rates and water quality grades.

[0105] Specifically, if Figure 4 As shown, the abnormality identification module may specifically include:

[0106] a judgment unit, configured to judge whether there is a pollution risk based on the standard water quality data and a preset safety interval, and judge that there is a pollution risk when the standard water quality data is not within the safety interval;

[0107] an identification unit, configured to process the standard water quality data using an anomaly identification algorithm to obtain an anomaly detection result when there is a risk of contamination;

[0108] A risk assessment unit, configured to determine a risk level based on the anomaly detection result and a preset risk rating standard;

[0109] A report output unit is used to output an anomaly report based on the risk level and the anomaly detection result.

[0110] Specifically, in the judgment unit, the safety interval is set based on historical data. For example, the safety interval for water pH is 6.5 to 8.5; the safety interval for dissolved oxygen concentration is greater than 3 mg / L, the dissolved oxygen concentration is normally greater than or equal to 5 mg / L, and 3 mg / L is the limit value; the safety interval for ammonia nitrogen is less than 1 mg / L, the ammonia nitrogen is normally less than or equal to 0.5 mg / L, and 1 mg / L is the limit value. The safety interval for conductivity is set based on the water type (e.g., fresh water, seawater, or industrial discharge water).

[0111] Specifically, in the identification unit, the anomaly identification algorithm includes a statistical method (for example, Z-score detection or an analysis method based on the interquartile range), a machine learning algorithm (for example, an isolation forest algorithm or a local anomaly factor algorithm) and a deep learning algorithm. The Z-score detection is based on the normal distribution and identifies anomalies that exceed three times the standard deviation. The analysis method based on the interquartile range is used to detect extreme values that exceed the normal distribution range. The isolation forest algorithm is based on a tree structure to identify rare or extreme abnormal data. The local anomaly factor algorithm detects abnormal patterns by analyzing the density of data points. The deep learning algorithm is suitable for complex water quality trend analysis. By setting a long short-term memory network autoencoder, the normal pattern of the target feature is learned and abnormal data that deviates from the normal pattern is detected. The long short-term memory network autoencoder combines deep learning technology and is suitable for complex time series data, especially for capturing long-term dependencies and trends of water quality parameters.

[0112] Specifically, in the identification unit, the abnormality detection results include instantaneous abnormalities, gradual abnormalities and periodic abnormalities. The instantaneous abnormalities refer to extreme values that appear in a short period of time, such as sudden changes in water pH, sudden drops in dissolved oxygen concentration, etc., which are caused by short-term pollutant emissions or equipment failures. The gradual abnormality refers to a certain target feature that slowly deviates from the normal range over time, such as a continuous increase in chemical oxygen demand, indicating a continuous impact from the pollution source. The periodic abnormality refers to a certain target feature that changes in a specific period, such as the chlorophyll concentration that periodically increases in summer, but when it exceeds the historical normal range, it is necessary to be vigilant about the risk of algal blooms.

[0113] Specifically, in the risk assessment unit, when the abnormality detection result is not within the safety interval, an abnormality value is calculated based on the abnormality detection result and the safety interval. The abnormality value calculation formula is:

[0114] c = min(|x1-y|, |x2-y|);

[0115] Where c is the outlier, x1 is the minimum value of the safety interval, x2 is the maximum value of the safety interval, and y is the outlier detection result. The preset risk rating standard is set according to the outlier. The risk levels include low risk, medium risk, and high risk. Low risk refers to a slight deviation from the safety interval, which is a natural fluctuation and does not require intervention. Medium risk refers to a large deviation from the safety interval, affecting water quality and requiring continuous monitoring. High risk refers to a serious deviation from the safety interval, which may cause a water pollution incident and requires immediate response measures.

[0116] Specifically, in the report output unit, the abnormality report records abnormal water quality parameters, occurrence time, impact range, risk level, and provides visual trend analysis. In addition, historical data can be combined to analyze the source of the abnormality, such as industrial emissions, agricultural runoff, or climate change.

[0117] Specifically, the abnormality identification module further includes an early warning unit, which is used to issue an alarm according to the risk level and send the abnormality report to the terminal device of the staff.

[0118] Specifically, if Figure 5 As shown, the control module may include:

[0119] A historical data acquisition unit, used for acquiring historical water quality data;

[0120] A multi-objective function setting unit, configured to generate a multi-objective function based on the water quality evaluation result and the abnormality report;

[0121] The optimization unit is used to optimize the multi-objective function, the historical water quality data and the water quality data using a multi-objective optimization algorithm to obtain an optimization result; the multi-objective optimization algorithm includes a reinforcement learning algorithm.

[0122] The control unit is used to select a water quality control scheme from a preset water quality control scheme library according to the optimization result.

[0123] Specifically, in the multi-objective function setting unit, the multi-objective function objectives include reducing pollutant concentrations (for example, ammonia nitrogen, heavy metals and chemical oxygen demand), improving the self-purification capacity of water bodies (for example, increasing dissolved oxygen and improving ecological stability) and preventing algal blooms or pollutant accumulation (for example, regulating water flow or controlling nutrients).

[0124] Specifically, the multi-objective function satisfies constraints, and the constraints include cost constraints and environmental constraints. The cost constraints are set by obtaining historical cost data, and the environmental constraints (for example, chemical inventory, aeration equipment power consumption, and water conservancy facility operation restrictions) are set according to the historical environmental data. The environmental constraints are used to avoid the impact of chemical reagents on the ecological balance, and the cost constraints are used to optimize operating costs, giving priority to low-cost and high-effect regulatory measures. The constraints also include legal constraints, which refer to compliance with legal provisions to avoid excessive emissions or improper treatment (for example, the use of banned agents, etc.). The constraints also include energy consumption constraints, which are mainly aimed at energy-consuming equipment such as aeration, circulating water pumps, ultraviolet disinfection systems, and chemical agent delivery equipment involved in the water quality control process. The optimal energy consumption range is set through energy consumption monitoring data to reduce energy consumption and improve the energy utilization efficiency of the control system.

[0125] Specifically, in the optimization unit, optimization is performed by adopting a multi-objective optimization algorithm (for example, a particle swarm algorithm, a genetic algorithm, or a reinforcement learning algorithm). The optimization calculation process simulates the impact of different control strategies on water quality by combining historical water quality data and standard water quality data, thereby evaluating the effectiveness, cost, and feasibility of each water quality control scheme. In this process, the system selects the optimal solution based on the multi-objective function, and considers the effective utilization of resources and the feasibility of actual operation based on the constraints. The system will dynamically adjust the weights in the control strategy, optimize the control strategy based on real-time monitoring results and water quality changes, and ensure that the control measures can always achieve the best results under different environmental conditions.

[0126] Specifically, the specific working process of the optimization unit includes the following steps:

[0127] Step 1: The first step in the optimization unit is to prepare the input data. The system extracts relevant water quality parameters from historical and standard water quality data and generates preliminary water quality assessment results based on this data. Water quality parameters of interest include, but are not limited to, dissolved oxygen, pH, ammonia nitrogen concentration, and turbidity. This data is used as the basis for the multi-objective optimization function.

[0128] Step 2: Based on historical water quality data, standard water quality data, and water quality evaluation results, the optimization unit will generate a multi-objective function. This function takes into account multiple aspects, such as the degree of water quality improvement, control costs, and the feasibility of water quality control plans. The multi-objective function can be expressed as:

[0129]

[0130] Among them, x is the parameter of the control strategy, f i (x) is the water quality optimization goal (e.g., improving water quality, reducing costs, optimizing energy consumption, reducing risk of violations, etc.), w i is the weight of each objective function. The multi-objective function provides an overall evaluation criterion by integrating the impact of each objective.

[0131] Step 3: The optimization unit selects a suitable multi-objective optimization algorithm, including particle swarm optimization (PSO), genetic algorithm (GA), or reinforcement learning (RL).

[0132] Taking the particle swarm optimization algorithm as an example, the particle swarm optimization algorithm searches for the optimal solution by simulating the movement of a group of particles. In this step, each position of the particle represents a set of control parameters.

[0133] The update formula of particle swarm optimization is:

[0134] v i (t+1)=w·v i (t)+c1·rand1(pbest i -x i (t))+c2·rand2·(gbest-x i (t));

[0135] x i (t+1)=x i (t)+v i (t+1);

[0136] Among them, v i (t) is the particle velocity, x i (t) is the particle position, pbest i and gbest are the optimal positions of the particle and the world respectively, c1, c2 are acceleration constants, w is the inertia weight, rand1 and rand2 are random numbers.

[0137] Step 4: In the optimization unit, the control strategy is optimized using a multi-objective optimization algorithm. Based on the multi-objective function F(x) and historical and standard water quality data, the algorithm simulates the impact of different control strategies (such as adjusting chemical dosage or adding water treatment equipment) on water quality and evaluates the effectiveness, cost, and feasibility of each option.

[0138] During optimization calculations, the system simulates different water quality control scenarios to obtain optimization objectives (such as water quality improvement and control costs) and the constraints of each scenario. Each iteration generates a new set of water quality control scenarios, and the optimal solution is selected based on the multi-objective function.

[0139] Step 5: During the optimization process, the system considers constraints, such as resource limitations and operational feasibility. For example, a water quality improvement plan might be constrained by equipment capacity, a maximum limit on the amount of chemical to be administered, or a time window. In this step, constraints influence the choice of optimization solution.

[0140] Assuming the constraint condition is g(x)≤0, the optimization problem can be expressed as a constrained multi-objective optimization problem, which is expressed as follows:

[0141] min F(x)subject to g(x)≤0;

[0142] Under this constraint, the optimization algorithm will dynamically adjust the control strategy to ensure the effective utilization of resources and the feasibility of operations.

[0143] Step 6: In multi-objective optimization, the system will dynamically adjust the weight w of each objective function based on real-time monitoring results and water quality changes. i , in order to cope with different environmental conditions. At this time, the task of the optimization unit is to adjust the weight according to environmental changes, so as to ensure that the control strategy can achieve the best effect in different situations. The relevant formula is as follows:

[0144] w i (t+1)=w i (t)·(1+δ);

[0145] Among them, δ is the weight adjustment factor, which is automatically adjusted based on real-time monitoring data.

[0146] Step 7: Based on the algorithm's calculation results, the optimization unit selects the best water quality control solution from the water quality control solution library and outputs the optimal solution. This solution may include specific control strategy parameters, such as the amount of chemical to be added and the operating mode of the control equipment.

[0147] Step 8: Optimization results are fed back to the control unit, which then performs specific water quality control operations based on the optimization results. Furthermore, standard water quality data continues to be collected and fed back into the optimization unit, forming a closed-loop feedback mechanism. The system continuously monitors water quality changes and dynamically adjusts strategies.

[0148] In some embodiments, the optimization unit adopts an optimization process based on reinforcement learning, and the specific working process is as follows:

[0149] During the water quality optimization process, reinforcement learning (RL) is used to dynamically optimize control strategies to ensure that water bodies maintain optimal conditions under varying environmental conditions. This optimization process includes data preparation, reward function construction, reinforcement learning model initialization, training optimization, constraint handling, dynamic weight adjustment, optimal solution output, and closed-loop feedback optimization. Through continuous learning and adjustment, the system can adaptively optimize water quality control measures to achieve multiple goals, including water quality improvement, cost control, energy optimization, and regulatory compliance.

[0150] First, input data is prepared to ensure that the reinforcement learning algorithm has sufficient information for optimization. Key water quality parameters such as dissolved oxygen, pH, ammonia nitrogen, turbidity, chemical oxygen demand, biological oxygen demand, and heavy metal ion concentrations (lead, cadmium, and mercury) are extracted from historical water quality data, standard water quality data, and real-time monitoring data. In addition, the system also obtains environmental factors (such as weather conditions and pollution source emission data) and resource constraint information (drug inventory and equipment load) to ensure that the reinforcement learning model can fully perceive the water quality conditions and make accurate decisions.

[0151] Next, a reward function for reinforcement learning is constructed to guide the optimization strategy to converge in the optimal direction. The reward function takes into account multiple objectives, including water quality improvement, cost control, energy consumption reduction, environmental impact reduction, and regulatory compliance. Specifically, when the control measures effectively reduce the concentration of pollutants (such as ammonia nitrogen, heavy metals) or increase the dissolved oxygen level, the system gives positive rewards; when the amount of chemical agents added is reduced and the energy consumption of operating equipment (such as aerators, water pumps) is reduced, positive rewards will also be given to encourage energy conservation and cost reduction. At the same time, if the control plan leads to excessive emissions or non-compliance with water quality regulations, negative rewards will be given to ensure that the strategy complies with environmental regulations. The goal of reinforcement learning is to maximize the cumulative reward, that is, to learn the optimal control strategy so that the water quality always maintains the optimal state under different conditions. The formula for maximizing the cumulative reward is as follows:

[0152]

[0153] π* represents the optimal strategy. The system continuously learns to maximize the expected cumulative reward of all future time steps under the current state S0. E[…] represents the expected value, which represents the expected calculation of future rewards under strategy π. ∑(t=0→∞)γ t R t represents the cumulative reward for all time steps t, where R t is the reward value at time t, used to measure the effectiveness of the water quality control strategy at that moment, such as pollutant removal rate and energy consumption optimization. γ (discount factor) is a value between 0 and 1 that indicates the influence of future rewards on current decisions. When γ is closer to 1, reinforcement learning focuses more on long-term returns; when γ is smaller, it tends to focus more on short-term gains. The goal of this formula is to enable the reinforcement learning agent (agent) to find an optimal strategy π* that ensures that the water quality control system remains in optimal condition over the long term, dynamically adjusting optimization measures under varying environmental conditions to achieve the optimal balance between improving water quality, reducing costs, saving energy, and meeting regulatory requirements.

[0154] After constructing the reward function, the system initializes the reinforcement learning model and selects an algorithm suitable for water quality control. Depending on the characteristics of water quality management, reinforcement learning methods such as Deep Q-Network (DQN), Proximal Policy Optimization (PPO), or Deep Deterministic Policy Gradient (DDPG) can be selected. DQN is suitable for discrete action spaces, such as the selection of different dosages of different chemicals; PPO is suitable for complex water quality optimization problems and can handle highly uncertain water environment changes; DDPG is suitable for continuous action spaces, such as the dynamic adjustment of aerator power and water pump flow. During the model initialization phase, the system sets the neural network parameters, defines the exploration rate (ε-greedy strategy), learning rate (α), discount factor (γ), and constructs an experience replay pool (Replay Buffer) to store historical interaction data to improve training stability.

[0155] The reinforcement learning agent then interacts with the water quality control environment, continuously optimizing its water quality management strategy. In each training episode, the system first senses the current water quality state (S), including information such as pH, dissolved oxygen, and pollutant concentrations. This information is combined with historical data and environmental factors to form a complete water quality state vector. The system then selects a water quality control measure based on the current reinforcement learning policy π(A|S), such as adjusting chemical dosage, changing aerator power, or optimizing water flow rate. After selecting the control measure, the system executes the corresponding action and records the new water quality state (S') at the next moment. The system then evaluates the effectiveness of the control measure based on a reward function, calculating a reward value (R) to measure the effectiveness of the policy. The system then updates the policy using a reinforcement learning algorithm. For example, in DQN training, the Q-learning formula is used: Q(S,A)←Q(S,A)+α[R+γmaxQ(S′,A′)-Q(S,A)].

[0156] During PPO or DDPG training, the system uses policy gradient optimization to ensure that the agent's strategy converges to the optimal solution after multiple interactions. By repeatedly repeating this process, the reinforcement learning model gradually optimizes the control strategy, making water quality management more precise, stable, and intelligent.

[0157] During the optimization process, reinforcement learning automatically handles various constraints to ensure that the control plan meets resource limitations and regulatory requirements. Traditional optimization methods require manual setting of resource constraints, regulatory compliance, and energy consumption limits, while reinforcement learning automatically learns these constraints through a reward mechanism. For example, when the inventory of reagents is limited, the system will use negative rewards to prevent over-dosing; when the power consumption of the aerator exceeds the standard, the reward value will also be reduced to encourage energy-saving optimization. In addition, if the control measures cause water quality parameters to exceed the standard, the system will issue a large negative reward, prompting the model to automatically avoid illegal discharge plans. In this way, reinforcement learning can adaptively optimize water quality control strategies while satisfying various constraints.

[0158] Another important feature of reinforcement learning is the ability to dynamically adjust the optimization objective weights so that the system can maintain the best decision-making ability under different environmental conditions. In traditional multi-objective optimization, the weight parameter w i It requires manual setting, while reinforcement learning can dynamically adjust the target weight according to real-time monitoring data. The weight adjustment follows the following formula: i (t+1)=w i (t)·(1+δ);

[0159] Here, δ is a weight adjustment factor, automatically optimized based on water quality monitoring results. For example, if pollutant concentrations rise sharply, the system automatically increases the pollution control weight w1, prioritizing efficient pollution removal measures. If water quality stabilizes but energy consumption is high, the system decreases the pollution control weight and increases the energy optimization weight w2, adjusting the strategy to better balance water quality optimization and energy management.

[0160] After reinforcement learning optimizes the optimal strategy π*, the system outputs the optimal water quality control plan and guides its implementation. This plan includes specific operational parameters, such as optimal chemical dosage, aerator operating parameters, and water facility control modes, to ensure optimal water quality. The system also evaluates the effectiveness of different control plans and, when necessary, further optimizes decisions, making control measures more adaptable and flexible.

[0161] Finally, reinforcement learning relies on a closed-loop feedback mechanism to continuously optimize water quality management strategies, adapting them to new water quality conditions. By monitoring water quality changes in real time, the system can detect deviations between control effects and expected values and adjust the reinforcement learning model accordingly to optimize future decisions. Simultaneously, the system regularly updates the reinforcement learning strategy based on the latest water quality monitoring data, adapting it to long-term water quality trends. Furthermore, in the event of sudden pollution incidents or extreme weather events, the reinforcement learning system can rapidly respond and adjust water quality control plans, improving the stability and resilience of water quality management.

[0162] Specifically, in the control unit, the water quality control program library is comprehensively established based on historical control data and water quality control field data. The water quality control program includes the addition of water treatment agents, aeration intensity adjustment or water flow direction adjustment. The water treatment agent addition is achieved by calculating the optimal dosage of flocculants, disinfectants and nutrient regulators to ensure the maximum pollutant removal rate while reducing the waste of agents. For example, when the chemical oxygen demand exceeds the standard, it is recommended to add an appropriate amount of flocculants to promote the sedimentation of organic matter; when the bacteria exceed the standard, an appropriate amount of disinfectant (for example, sodium hypochlorite) is added. The aeration intensity adjustment increases the dissolved oxygen concentration and improves the water ecology by adjusting the operating frequency of the aeration equipment. If the dissolved oxygen concentration is detected to be lower than 3 mg / L, the aeration volume is increased; if eutrophication of the water body is detected, the aeration power is reduced to reduce algae reproduction. The water flow direction adjustment reduces pollutant enrichment and improves water quality balance by optimizing the water flow distribution of reservoirs, lakes or artificial water systems. For example, in areas with high risk of algal blooms, the water flow direction is adjusted to promote water exchange and prevent local water quality deterioration.

[0163] Specifically, the control module also includes a simulation unit, which is used to simulate the water quality control plan to determine the feasibility of the water quality control plan. If it is judged to be effective, the water quality control plan is sent to the execution module. If it is judged to be infeasible, the parameters in the optimization unit are adjusted, and the optimization is performed again to generate a new water quality control plan.

[0164] Specifically, if Figure 6 As shown, the execution module is used to obtain the water quality control plan and perform operations according to the water quality control plan. The execution module includes:

[0165] A control variable extraction unit, configured to extract control variables from the water quality control scheme;

[0166] an execution device, configured to perform work according to the control variable;

[0167] an effect judgment unit, configured to obtain standard water quality data after the execution device has been operated, and determine whether the regulation is completed based on the standard water quality data after the execution device has been operated and a preset target value; if the regulation is completed, the execution device is turned off; if the regulation is not completed, the dynamic adjustment unit is executed;

[0168] A dynamic adjustment unit is used to generate a dynamic control variable based on the standard water quality data after the execution device works and the target value, and control the execution device to work according to the dynamic control variable.

[0169] When the execution module is working, it first identifies the control variables in the water quality control plan (for example, the type and type of purifier, the adjustment parameters of the aeration system and the water pump flow control target, etc.), and then performs self-inspection on the execution equipment in the execution module to ensure that all execution equipment (for example, the dosing system, the aeration system and the water pump, etc.) are operating normally to avoid the control effect being affected by equipment failure. Thirdly, the control execution equipment works according to the control variables. Then, the water quality data after the work is completed is obtained, and the water quality data after the work is completed is compared with the preset target value to judge the control effect. If the water quality data after the work is completed cannot meet the preset target value, dynamic adjustments are made, such as increasing or decreasing the amount of purifier added, and further optimizing the aeration or water pump flow control plan. Each control operation, including execution time, parameter settings, equipment status, water quality response, etc., is stored in the database. A water quality control log is formed. Combined with historical data, the long-term trend of the control strategy is analyzed to optimize future water quality management plans. Finally, by adjusting the control system of the execution module based on feedback data, for example, by analyzing the long-term effects of water quality control through machine learning algorithms (such as reinforcement learning and adaptive optimization), the control strategy can be dynamically adjusted. The optimal parameter configuration can be identified, and the response strategy for similar situations in the future can be optimized, thereby improving the intelligence level of water quality control. If the water quality is not improved as expected, or if the control leads to unexpected risks (such as excessive dosage of chemicals or water flow disturbances), the emergency plan will be immediately activated. For example, the emergency stop of drug administration can prevent secondary pollution caused by excessive purification agents; the adjustment of water flow direction can reduce the risk of pollutant spread; and the increase of manual inspections to check for equipment failures or interference from external pollution sources.

[0170] Specifically, when the execution device is controlled to work according to the control variables, the work content includes adding purifiers, adjusting the aeration system, and optimizing water flow direction and hydraulic regulation.

[0171] The goal of administering purifiers (e.g., flocculants, disinfectants, nutrient regulators) is to remove organic pollutants, control pathogenic microorganisms, and maintain ecological balance. Execution involves the following steps: Controlling the automated dosing system to administer the optimal dosage based on the strategy's calculations. Adjusting the dosing method, such as continuous dosing (for long-term water quality adjustments) or pulse dosing (for short-term, rapid response). Monitoring water quality changes after the agent is administered. If the desired effect is not achieved, dynamically adjusting the dosage or switching the purifier type.

[0172] The goal of adjusting the aeration system is to improve the water's self-purification capacity, inhibit the accumulation of anaerobic pollutants, and reduce the risk of eutrophication. This involves adjusting the frequency and power of the aeration system to increase or decrease the oxygen supply. In areas of high eutrophication risk (e.g., areas with algae blooms), reduce aeration intensity to inhibit algae growth. In areas of low dissolved oxygen (e.g., bottom water), increase aeration to promote mixing and increase oxygen levels.

[0173] Optimizing water flow and hydraulic regulation aims to reduce pollutant accumulation, accelerate water circulation, and optimize hydrodynamic structure. This includes controlling pump flow to increase or decrease water velocity and prevent pollutant deposition. Adjusting gate openings or hydraulic structures redirects water flow, promotes water exchange, and prevents the formation of pollution hotspots in stagnant water areas. Dynamic adjustments to water flow regulation strategies can be made based on meteorological and hydrological data, such as increasing water flow before rainfall to reduce pollutant accumulation.

[0174] Example 2

[0175] like Figure 7 As shown, this embodiment provides a method for intelligent water quality control based on real-time monitoring of multiple parameters, including the following steps:

[0176] S10: acquiring original water quality data, preprocessing the original water quality data to obtain standard water quality data, and performing feature extraction on the standard water quality data to obtain a plurality of target features;

[0177] S20: Inputting the plurality of target features into a pre-trained deep neural network model to obtain a water quality evaluation result;

[0178] S30: Determine whether the water quality evaluation result has a pollution risk based on the standard water quality data and a preset safety interval. If it is determined that there is a pollution risk, process the standard water quality data using an anomaly recognition algorithm to obtain an anomaly report.

[0179] S40: Generate a water quality control plan using a multi-objective optimization algorithm according to the water quality evaluation result and the abnormality report.

[0180] Specifically, the step S10 includes the following steps:

[0181] S11: acquiring raw water quality data from the multi-parameter sensor array;

[0182] S12: Using a filtering algorithm to remove noise from the original water quality data to obtain first preprocessed data;

[0183] S13: Detecting outliers in the first preprocessed data using a machine learning algorithm, and repairing the outliers using a variational autoencoder to obtain second preprocessed data;

[0184] S14: Processing the second pre-processed data using a normalization algorithm or a standardization algorithm to obtain standard water quality data;

[0185] S15: Using at least one of a time series decomposition method, a wavelet transform method, or a long short-term memory network to perform feature extraction on the standard water quality data to obtain a plurality of target features.

[0186] Specifically, in step S12, outlier processing is performed using one of the following methods: the interquartile range algorithm, the Z-score algorithm, a machine learning algorithm (e.g., the isolation forest algorithm or the local outlier factor algorithm), an interpolation method, or a elimination method, selected based on the actual situation. The wavelet transform is used to remove high-frequency noise while retaining mutation characteristics and improving data smoothness. The Kalman filter is used for real-time data stream processing, predicting current values based on historical data, and correcting for noise effects to improve data accuracy. The median filter is used for sudden outliers (e.g., sudden changes in conductivity) to reduce the impact of impulse noise on data stability.

[0187] Specifically, in step S13, one of the interquartile range algorithm, Z-score algorithm, machine learning algorithm (such as isolation forest algorithm or local outlier factor algorithm, etc.), interpolation method or elimination method is used. The interquartile range algorithm and Z-score algorithm are used to identify abnormal data points exceeding a preset range, the machine learning algorithm is used to detect abnormal points in complex data patterns, the interpolation method is used to perform linear interpolation or mean difference correction on mild abnormal data, and the elimination method is used to directly delete severe abnormal data to prevent individual abnormal data from affecting the overall analysis results.

[0188] Specifically, in step S14, the original water quality data is scaled to the interval [0, 1] by adopting a normalization algorithm, which is suitable for non-normally distributed data; the Z-score normalization algorithm is used for normally distributed data to convert the original water quality data into a standard distribution with a mean of 0 and a variance of 1, thereby improving robustness.

[0189] Specifically, in step S15, the target features include any one of water pH, dissolved oxygen concentration, conductivity, turbidity, heavy metal ion concentration (e.g., lead, cadmium, or mercury), chemical oxygen demand, biological oxygen demand, ammonia nitrogen, and chlorophyll concentration. Since changes in water quality data are time-dependent, it is necessary to construct time features to facilitate subsequent trend analysis and prediction. Therefore, the target features also include multiple time features, including short-term change features, periodic change features, and anomaly detection features. The target features may also include environmental features.

[0190] Specifically, step S30 may include the following steps:

[0191] S31: judging whether there is a pollution risk based on the standard water quality data and a preset safety interval; and judging that there is a pollution risk when the standard water quality data is not within the safety interval;

[0192] S32: When there is a pollution risk, an abnormality recognition algorithm is used to process the standard water quality data to obtain an abnormality detection result;

[0193] S33: Obtaining a risk level based on the abnormality detection result and a preset risk rating standard;

[0194] S34: Outputting an abnormality report according to the risk level and the abnormality detection result.

[0195] Specifically, in step S32, the anomaly identification algorithm includes a statistical method (for example, Z-score detection or an analysis method based on the interquartile range), a machine learning algorithm (for example, an isolation forest algorithm or a local anomaly factor algorithm) and a deep learning algorithm. The Z-score detection is based on the normal distribution and identifies anomalies that exceed three times the standard deviation. The analysis method based on the interquartile range is used to detect extreme values that exceed the normal distribution range. The isolation forest algorithm is based on a tree structure and identifies rare or extreme anomaly data. The local anomaly factor algorithm detects abnormal patterns by analyzing the density of data points. The deep learning algorithm is suitable for complex water quality trend analysis. By setting a long short-term memory network autoencoder, the normal pattern of the target feature is learned and abnormal data that deviates from the normal pattern is detected. The long short-term memory network autoencoder combines deep learning technology and is suitable for complex time series data, especially for capturing the long-term dependence and trend of water quality parameters. The anomaly detection results include transient anomalies, gradual anomalies and periodic anomalies. The transient anomaly refers to an extreme value that appears in a short period of time, such as a sudden change in the pH of the water body, a sudden drop in dissolved oxygen concentration, etc., caused by short-term pollutant emissions or equipment failures. The gradual anomaly refers to the slow deviation of quality indicators from the normal range over time, such as the continuous increase in chemical oxygen demand, indicating the continued impact of the pollution source.

[0196] Specifically, in step S33, when the abnormality detection result is not within the safety interval, an abnormality value is calculated based on the abnormality detection result and the safety interval. The abnormality value calculation formula is:

[0197] c = min(|x1-y|, |x2-y|);

[0198] Where c is the outlier, x1 is the minimum value of the safety interval, x2 is the maximum value of the safety interval, and y is the outlier detection result. The preset risk rating standard is set according to the outlier. The risk levels include low risk, medium risk, and high risk. Low risk refers to a slight deviation from the safety interval, which is a natural fluctuation and does not require intervention. Medium risk refers to a large deviation from the safety interval, affecting water quality and requiring continuous monitoring. High risk refers to a serious deviation from the safety interval, which may cause a water pollution incident and requires immediate response measures.

[0199] Specifically, in step S34, the abnormality report records abnormal water quality parameters, occurrence time, impact range, risk level, and provides visual trend analysis. In addition, historical data can be combined to analyze the source of the abnormality, such as industrial emissions, agricultural runoff, or climate change.

[0200] Specifically, the step S40 includes the following steps:

[0201] S41: Obtain historical water quality data;

[0202] S42: generating a multi-objective function according to the water quality evaluation result and the abnormality report;

[0203] S43: performing optimization using a multi-objective optimization algorithm based on the multi-objective function, the historical water quality data, and the water quality data to obtain an optimization result; the multi-objective optimization algorithm includes a reinforcement learning algorithm;

[0204] S44: Selecting a water quality control scheme from a preset water quality control scheme library according to the optimization result.

[0205] Specifically, in step S42, the objectives of the multi-objective function include reducing the concentration of pollutants (for example, ammonia nitrogen, heavy metals and chemical oxygen demand), improving the self-purification capacity of water bodies (for example, increasing dissolved oxygen and improving ecological stability) and preventing algal blooms or accumulation of pollutants (for example, regulating water flow or controlling nutrients). The multi-objective function satisfies constraints, and the constraints include cost constraints and environmental constraints. The cost constraints are set by obtaining historical cost data, and the environmental constraints (for example, chemical agent inventory, aeration equipment power consumption, water conservancy facility operation restrictions) are set according to the historical environmental data. The environmental constraints are used to avoid the impact of chemical reagents on ecological balance, and the cost constraints are used to optimize operating costs, giving priority to low-cost and high-effect regulatory measures. The constraints also include legal constraints and energy consumption constraints. The legal constraints refer to complying with legal provisions to avoid excessive emissions or improper treatment (for example, the use of banned agents, etc.). The energy consumption constraints are mainly aimed at energy-consuming equipment such as aeration, circulating water pumps, ultraviolet disinfection systems, and chemical agent delivery equipment involved in the water quality control process. The optimal energy consumption range is set through energy consumption monitoring data to reduce energy consumption and improve the energy utilization efficiency of the control system.

[0206] Specifically, in step S43, optimization is performed by adopting a multi-objective optimization algorithm (e.g., a particle swarm optimization algorithm, a genetic algorithm, or a reinforcement learning algorithm). The optimization calculation process simulates the impact of different control strategies on water quality by combining historical water quality data and standard water quality data, thereby evaluating the effect, cost, and feasibility of each water quality control scheme. In this process, the system screens out the optimal solution based on the multi-objective function, and considers the effective utilization of resources and the feasibility of actual operation based on the constraints. The system dynamically adjusts the weights in the control strategy, optimizes the control strategy based on real-time monitoring results and water quality changes, and ensures that the control measures can always achieve the best results under different environmental conditions.

[0207] Specifically, in step S44, the water quality control program library is comprehensively established based on historical control data and water quality control field data, and the water quality control program includes the addition of water treatment agents, aeration intensity adjustment or water flow direction adjustment. The water treatment agent addition is performed by calculating the optimal dosage of flocculants, disinfectants and nutrient regulators to ensure the maximum pollutant removal rate while reducing the waste of agents. The aeration intensity adjustment increases the dissolved oxygen concentration and improves the water ecology by adjusting the operating frequency of the aeration equipment. The water flow direction adjustment reduces pollutant enrichment and improves the balance of water quality by optimizing the water flow distribution of reservoirs, lakes or artificial water systems.

[0208] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-mentioned intelligent water quality control methods based on multi-parameter real-time monitoring.

[0209] The present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing related hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of the above-mentioned various method embodiments. The computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0210] The present invention also provides an electronic device. The electronic device in an embodiment of the present invention includes: one or more processors; and a storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the intelligent water quality control method based on real-time multi-parameter monitoring provided by the present invention.

[0211] Reference below Figure 8 , which shows a schematic structural diagram of a computer system 800 of an electronic device suitable for implementing an embodiment of the present invention. Figure 8 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention.

[0212] like Figure 8 As shown, the computer system 800 includes a central processing unit (CPU) 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage unit 808 into a random access memory (RAM) 803. Various programs and data required for the operation of the computer system 800 are also stored in the RAM 803. The CPU 801, ROM 802, and RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0213] The following components are connected to the I / O interface 805: an input section 806 including a keyboard, a mouse, and the like; an output section 807 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 808 including a hard disk; and a communication section 809 including a network interface card such as a LAN card or a modem. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as needed. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 810 as needed, so that computer programs read therefrom can be installed in the storage section 808 as needed.

[0214] In particular, according to embodiments disclosed herein, the processes described in the main step diagrams above can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for executing the methods shown in the main step diagrams. In the above embodiments, the computer program can be downloaded and installed from a network via the communication section 809 and / or installed from removable media 811. When the computer program is executed by the central processing unit 801, the above-described functions defined in the system of the present invention are performed.

[0215] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. An intelligent water quality control system based on real-time monitoring of multiple parameters, characterized in that: include: A multi-parameter sensor array and a monitoring and control server, wherein the multi-parameter sensor array is arranged in a water body and is used to obtain raw water quality data of the water body; The monitoring and control server is configured to receive the raw water quality data from the multi-parameter sensor array, determine whether there is a pollution risk based on the raw water quality data, and if it is determined that there is a pollution risk, generate an abnormality report and a water quality evaluation result based on the raw water quality data, and generate a water quality control plan based on the abnormality report and the water quality evaluation result; The monitoring and control server includes: A data processing module is used to obtain raw water quality data, preprocess the raw water quality data to obtain standard water quality data, and perform feature extraction on the standard water quality data to obtain multiple target features; A water quality evaluation module, configured to input a plurality of target features into a pre-trained deep neural network model to obtain a water quality evaluation result; An anomaly recognition module is used to determine whether the water quality evaluation result has a pollution risk based on the standard water quality data and a preset safety interval. When it is determined that there is a pollution risk, an anomaly recognition algorithm is used to process the standard water quality data to obtain an anomaly report; The control module is used to generate a water quality control plan using a multi-objective optimization algorithm according to the water quality evaluation result and the abnormality report.

2. The intelligent water quality control system based on multi-parameter real-time monitoring according to claim 1 is characterized in that: The multi-parameter sensor array includes any multiple of a surface plasmon resonance sensor, an ion-sensitive field-effect transistor sensor, a chlorophyll sensor, and a heavy metal ion concentration sensor; The surface plasmon resonance sensor is arranged in the downstream area of the industrial wastewater discharge outlet to monitor the concentration of organic pollutants; The ion-sensitive field-effect transistor sensor is arranged in an ammonia-nitrogen pollution area to monitor the ammonia-nitrogen concentration in water bodies; The chlorophyll sensor is disposed in the photic zone of the water body and is used to monitor the chlorophyll concentration associated with algal biomass; The multi-parameter sensor array uploads the original water quality data of the water body to the monitoring and control server through the communication module.

3. The intelligent water quality control system based on multi-parameter real-time monitoring according to claim 1 is characterized in that: The data processing module specifically includes: A raw water quality data acquisition unit, configured to acquire raw water quality data from the multi-parameter sensor array; a denoising unit, configured to remove noise from the original water quality data using a filtering algorithm to obtain first preprocessed data; an outlier processing unit, configured to detect outliers in the first preprocessed data using a machine learning algorithm, and repair the outliers using a variational autoencoder to obtain second preprocessed data; a standardization unit, configured to process the second pre-processed data using a normalization algorithm or a standardization algorithm to obtain standard water quality data; The feature extraction unit is used to extract features from the standard water quality data using at least one of a time series decomposition method, a wavelet transform method, or a long short-term memory network to obtain multiple target features.

4. The water quality intelligent control system based on multi-parameter real-time monitoring according to claim 1 is characterized in that: The deep neural network model specifically includes: An input layer, configured to receive a plurality of target features; A convolutional layer, used to obtain the correlation between the target features; A time series analysis layer is used to predict the water quality change trend based on the target characteristics to obtain the water quality change trend; the time series analysis layer adopts a hybrid model consisting of a long short-term memory network and a transformer model; A fully connected layer, comprising multiple hidden layers, each of which is provided with an activation function for adjusting the weight of each target feature; The output layer is used to output water quality evaluation results, which include water quality grade, pollution degree and water quality change trend.

5. The water quality intelligent control system based on multi-parameter real-time monitoring according to claim 1 is characterized in that: The abnormality identification module specifically includes: a judgment unit, configured to judge whether there is a pollution risk based on the standard water quality data and a preset safety interval, and judge that there is a pollution risk when the standard water quality data is not within the safety interval; an identification unit, configured to process the standard water quality data using an anomaly identification algorithm to obtain an anomaly detection result when there is a risk of contamination; A risk assessment unit, configured to determine a risk level based on the anomaly detection result and a preset risk rating standard; A report output unit is used to output an anomaly report based on the risk level and the anomaly detection result.

6. The water quality intelligent control system based on multi-parameter real-time monitoring according to claim 1 is characterized in that: The control module specifically includes: A historical data acquisition unit, used for acquiring historical water quality data; A multi-objective function setting unit, configured to generate a multi-objective function based on the water quality evaluation result and the abnormality report; an optimization unit, configured to perform optimization using a multi-objective optimization algorithm based on the multi-objective function, the historical water quality data, and the standard water quality data to obtain an optimization result; the multi-objective optimization algorithm includes a reinforcement learning algorithm; The control unit is used to select a water quality control scheme from a preset water quality control scheme library according to the optimization result.

7. The intelligent water quality control system based on multi-parameter real-time monitoring according to claim 3 is characterized in that: The standard water quality data include any one or more of microplastic concentration, chlorophyll concentration, antibiotic concentration, algal toxin concentration, total organic carbon content and water redox potential; In the feature extraction unit, the target features include any one of water pH, dissolved oxygen concentration, conductivity, turbidity, heavy metal ion concentration, chemical oxygen demand, biological oxygen demand, ammonia nitrogen and chlorophyll concentration.

8. The intelligent water quality control system based on multi-parameter real-time monitoring according to claim 1 is characterized in that: The system further includes an execution module, which is used to obtain the water quality control plan and perform operations according to the water quality control plan. The execution module includes: A control variable extraction unit, configured to extract control variables from the water quality control scheme; an execution device, configured to perform work according to the control variable; an effect judgment unit, configured to obtain standard water quality data after the execution device has been operated, and determine whether the regulation is completed based on the standard water quality data after the execution device has been operated and a preset target value; if the regulation is completed, the execution device is turned off; if the regulation is not completed, the dynamic adjustment unit is executed; A dynamic adjustment unit is used to generate a dynamic control variable based on the standard water quality data after the execution device works and the target value, and control the execution device to work according to the dynamic control variable.

9. The intelligent water quality control system based on multi-parameter real-time monitoring according to claim 6 is characterized in that: The multi-objective function satisfies constraint conditions, which include cost constraint conditions and environmental constraint conditions. The cost constraint conditions are set by obtaining historical cost data, and the environmental constraint conditions are set according to the historical environmental data.

10. A method for intelligent water quality control based on real-time multi-parameter monitoring, characterized in that: The following steps are involved: Acquiring original water quality data, preprocessing the original water quality data to obtain standard water quality data, and performing feature extraction on the standard water quality data to obtain multiple target features; Inputting the plurality of target features into a pre-trained deep neural network model to obtain a water quality evaluation result; Determining whether the water quality evaluation result has a pollution risk based on the standard water quality data and a preset safety interval, and when it is determined that there is a pollution risk, processing the standard water quality data using an anomaly recognition algorithm to obtain an anomaly report; A water quality control plan is generated using a multi-objective optimization algorithm based on the water quality evaluation results and the abnormality report.

Images