Intelligent parameter adjustment-based ultra-trace monitoring system for harmful pollutants in sewage

Abstract

This application relates to the field of wastewater treatment technology and discloses a wastewater hazardous pollutant trace monitoring system based on intelligent parameter adjustment. The system includes a data acquisition module for acquiring and preprocessing multi-dimensional data from the entire process; a hazardous pollutant impact event identification module that accurately identifies events based on a dynamic baseline statistical algorithm to avoid misjudgments; a simulation model and control strategy generation module that uses a model predictive control framework to predict water quality dynamics online and generate an optimal equipment linkage control strategy that balances water quality compliance with operating costs; and a post-treatment effect evaluation and model self-evolution module that quantitatively scores the entire treatment process and automatically adjusts and optimizes key parameters of the simulation model based on the scoring results. This invention constitutes an adaptive closed-loop control system that can achieve model self-evolution by learning from historical treatment experience, continuously improving the robustness and accuracy in dealing with unknown hazardous pollutant impacts, ensuring stable compliance of effluent water quality, and optimizing operational efficiency.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, specifically to a wastewater hazardous pollutant trace monitoring system based on intelligent parameter adjustment. Background Technology

[0002] Wastewater treatment plants are critical infrastructure for ensuring water environment safety. Their core relies on microbial systems such as activated sludge to biochemically degrade pollutants. However, the ecological fragility of this microbial system makes it highly susceptible to sudden influx of harmful pollutants into the influent. Illegal discharge of industrial wastewater or cross-connection of pipe networks often leads to the instantaneous entry of trace amounts of wastewater containing heavy metals, cyanide, or other harmful organic substances into the treatment system. This has a strong inhibitory or even inactivating effect on the microbial population, resulting in a sharp reduction in the efficiency of the biochemical treatment unit or even system collapse. Ultimately, this causes serious exceedances of effluent quality and triggers environmental pollution incidents.

[0003] Existing technologies have significant limitations in responding to such shocks of hazardous pollutants. In the identification phase of such shock events, traditional wastewater treatment plant operation and management rely on offline testing of routine water quality indicators or low-frequency online monitoring. This approach is severely delayed in responding to hazardous pollutant shocks, often only detecting problems after microbial activity has been damaged and effluent quality has deteriorated, missing the optimal intervention window. Although online hazardous substance monitoring instruments exist, they mostly use fixed threshold alarms, making it difficult to adapt to the complex fluctuations in water quality under normal operating conditions. This results in high false alarm or false negative rates, failing to provide accurate and timely early warnings.

[0004] In emergency response, current control strategies are generally crude and reactive. Wastewater treatment plants typically rely on the experience of operators for manual intervention, such as increasing aeration, adding emergency chemicals, or activating emergency tanks. These operations lack accurate prediction of dynamic changes in the system, making it difficult to achieve optimized coordination among multiple devices. This can lead to waste of energy and chemicals, and the treatment effect is highly uncertain. Even with the introduction of mechanism-based simulation models (such as the ASM series of activated sludge models) for process simulation, these models suffer from the inherent problem of "model mismatch." The model parameters are usually calibrated offline under specific operating conditions or set based on experience, failing to reflect in real time changes in the dynamic characteristics of the system caused by variations in influent composition, hydraulic load, and microbial population succession. When such static models are applied to complex hazardous pollutant impact scenarios, their predictive accuracy drops significantly, and the control strategies generated based on these models naturally fail to achieve optimal results. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a wastewater hazardous pollutant trace monitoring system based on intelligent parameter adjustment, which solves the problems of delayed response and rigid control strategies in wastewater treatment plants when dealing with sudden hazardous pollutant shocks.

[0006] To achieve the above objectives, the first aspect of the present invention provides a wastewater hazardous pollutant trace monitoring system based on intelligent parameter tuning, comprising:

[0007] The data acquisition module is used to collect and preprocess multidimensional time series data from the wastewater treatment system.

[0008] A hazardous pollutant impact event identification module, connected to the data acquisition module, is used to identify the occurrence of hazardous pollutant impact events based on the multidimensional time series data;

[0009] The simulation model and control strategy generation module is connected to the harmful pollutant impact event identification module. It contains a simulation model to predict the future water quality status when the event occurs and to generate equipment linkage control strategies.

[0010] The post-treatment effect evaluation and model self-evolution module is connected to the simulation model and control strategy generation module, and is used to update the model parameters of the simulation model after the event treatment is completed.

[0011] The hazardous pollutant impact event identification module is configured as follows:

[0012] Using multidimensional time series data from historical normal operating conditions, an unsupervised anomaly detection model is trained to learn normal data patterns.

[0013] Real-time data is input into the model to obtain the real-time reconstruction error that quantifies the degree of deviation;

[0014] When the real-time reconstruction error exceeds a preset dynamic threshold, an event trigger signal is generated.

[0015] The unsupervised anomaly detection model in the hazardous pollutant impact event identification module is an autoencoder neural network model.

[0016] The simulation model and control strategy generation module is configured to use a model predictive control framework to perform rolling optimization and generate the equipment linkage control strategy.

[0017] The post-treatment effect evaluation and model self-evolution module is configured as follows:

[0018] Obtain the actual influent disturbance sequence, actual control execution sequence, and actual water quality response sequence covering the entire process of the event;

[0019] An optimization problem is constructed, with the model parameters of the simulation model as optimization variables, and the objective is to minimize the cumulative prediction error between the predicted water quality response sequence output by the simulation model and the actual water quality response sequence.

[0020] Solve the optimization problem to obtain a set of updated model parameters, and use them to update the simulation model and the simulation model in the control strategy generation module.

[0021] The post-treatment effect evaluation and model self-evolution module is configured to use gradient descent or a variant thereof to solve the optimization problem to obtain updated model parameters.

[0022] The updating of the model parameters of the simulation model includes:

[0023] After the incident is resolved, obtain the actual inflow disturbance sequence, the actual control execution sequence, and the actual water quality response sequence covering the entire incident process;

[0024] An optimization problem is constructed with the model parameters of the simulation model as optimization variables, aiming to minimize the cumulative prediction error between the predicted water quality response sequence output by the simulation model after inputting the actual influent disturbance sequence and the actual control execution sequence and the actual water quality response sequence.

[0025] Solve the optimization problem to obtain a set of updated model parameters, and use them to replace the original model parameters.

[0026] The method of using simulation models to predict future water quality conditions and generate equipment linkage control strategies includes:

[0027] Using the current system state as the initial condition, a model predictive control framework is adopted to perform rolling optimization within a preset prediction time domain. The optimization problem with the objective of minimizing the comprehensive cost function of water quality deviation cost and operation cost is solved to obtain the optimal control sequence, and the first control command of the sequence is executed.

[0028] The unsupervised anomaly detection model is an autoencoder neural network model.

[0029] A second aspect of this invention provides a method for processing data from continuous monitoring of trace amounts of hazardous pollutants in wastewater based on intelligent parameter tuning, comprising the following steps:

[0030] Step S1: Continuously collect hazardous pollutant indicators, conventional water quality indicators, and operating process parameters in the wastewater treatment process;

[0031] Collect and preprocess multidimensional time-series data from wastewater treatment systems;

[0032] Based on the aforementioned multidimensional time series data, the occurrence of hazardous pollutant impact events is identified;

[0033] When a hazardous pollutant impact event is detected, a simulation model is used to predict the future water quality status, and an equipment linkage control strategy is generated based on the prediction results.

[0034] After the event is resolved, the model parameters of the simulation model are updated.

[0035] The identification of hazardous pollutant impact events includes:

[0036] Using multidimensional time series data from historical normal operating conditions, an unsupervised anomaly detection model is trained to learn normal data patterns.

[0037] The real-time collected multidimensional time series data is input into the unsupervised anomaly detection model to obtain the real-time reconstruction error that quantifies the degree of deviation between the current data and the normal pattern;

[0038] When the real-time reconstruction error exceeds a preset dynamic threshold, a hazardous pollutant impact event is determined to have occurred.

[0039] This invention provides a wastewater hazardous pollutant trace monitoring system based on intelligent parameter adjustment. It has the following beneficial effects:

[0040] 1. This invention establishes a feedback and update closed loop within the system, from the final effect of event handling to the upstream control model parameters, by setting up a post-treatment effect evaluation and model self-evolution module, and using the output of this module, namely the evaluation score of the treatment effect, to update the simulation model and control strategy generation module. This structure allows the performance of the control model to be directly affected and corrected by the actual control effect it produces.

[0041] 2. In this invention, the post-treatment effect evaluation and model self-evolution module is based on a quantitative evaluation score and adjusts the core parameters of the simulation model autonomously through a preset parameter update rule. This mechanism enables the system's control logic foundation, i.e. its internal model parameters, to break free from dependence on human experience and achieve automated iteration based on objective operating data, thereby enabling the system to have self-optimization capabilities.

[0042] 3. The system of the present invention can iteratively update the model parameters based on the handling effect of each hazardous pollutant impact event. Its ability to generate control strategies can adaptively improve with the passage of time and the accumulation of handling experience. This continuous self-correction and optimization can enhance the system's response performance and adaptability in the face of future unknown or changing hazardous pollutant impact events. Attached Figure Description

[0043] Figure 1 This is a system architecture diagram of the present invention;

[0044] Figure 2 This is a flowchart of the method of the present invention.

[0045] Among them, 10 is the data acquisition module; 20 is the hazardous pollutant impact event identification module; 30 is the simulation model and control strategy generation module; and 40 is the post-treatment effect evaluation and model self-evolution module. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Example:

[0048] Please see the appendix Figure 1 - Appendix Figure 2 This invention provides a wastewater hazardous pollutant trace monitoring system based on intelligent parameter tuning, comprising:

[0049] The data acquisition module 10 is the perception foundation of the entire system. Its core technical task is to provide high-quality, highly synchronous, and standardized multidimensional data streams for downstream event recognition and decision-making modules. The performance of the data acquisition module 10 directly determines the accuracy of subsequent hazardous pollutant impact event recognition and the timeliness of control strategy generation.

[0050] In this embodiment, the technical implementation details of the data acquisition module 10 are as follows:

[0051] Firstly, in terms of the breadth and depth of data acquisition, the acquisition module 10 connects to the SCADA system of the wastewater treatment plant in real time or directly to the underlying controller via industrial Ethernet or fieldbus protocols. Its data sources are not limited to conventional pollutant indicators, but constitute a multi-dimensional data matrix capable of comprehensively characterizing the system's fingerprint. These data can be specifically divided into:

[0052] Rapid-response water quality indicators: These indicators are extremely sensitive to small or sudden changes in water quality and are key to the early identification of harmful pollutant shocks. Examples include abnormal decreases in dissolved oxygen (DO) concentration, sharp jumps in oxidation-reduction potential (ORP), sudden deviations in pH value, and abnormal fluctuations in conductivity.

[0053] Process operating status parameters: These parameters reflect the actual working status and load of the treatment unit, such as the influent and effluent flow rates of each treatment tank, the blower frequency or valve opening of the aeration system, the rotation speed of the return sludge pump, and the instantaneous flow rate of the chemical dosing pump.

[0054] Conventional pollutant monitoring indicators: These indicators serve as background and validation data, including chemical oxygen demand (COD), ammonia nitrogen (NH3-N), total phosphorus (TP), and total nitrogen (TN) measured by online analytical instruments.

[0055] Secondly, in the data preprocessing stage, in order to transform complex raw data into high-quality input that can be directly used by machine learning models, the data acquisition module 10 integrates a sophisticated processing workflow, rather than simply formatting. This workflow specifically includes:

[0056] Signal noise reduction and feature preservation: Considering that sensor signals are generally subject to high-frequency noise interference, this module does not use a simple moving average filter, as it may cause the signal's abrupt change characteristics to be over-smoothed and lost. In this embodiment, the Savitzky-Golay filter is preferred. This filter can effectively filter out noise while preserving the original form of the signal to the greatest extent through local polynomial fitting. It is especially important for capturing the sharp rise or fall trend of the signal in the early stage of the impact of harmful pollutants.

[0057] Outlier Removal: Isolated outliers that do not conform to physical laws often occur in the raw data stream due to momentary sensor failures or communication interruptions. To remove these outliers, this module adopts a detection method based on interquartile range (IQR). Compared with the traditional 3-sigma rule, the IQR detection method does not rely on the assumption of normality of data distribution and is not sensitive to outliers themselves. Therefore, it exhibits stronger robustness when processing real-world working condition data that often exhibits a skewed distribution, and can effectively avoid misjudging real impact signal spikes as outliers that need to be removed.

[0058] Standardization of multi-scale data: Due to the different physical meanings of the collected data in each dimension, their dimensions and numerical ranges vary greatly. To eliminate the impact of these scale differences on the weight learning of the downstream neural network model, this data acquisition module 10 performs Z-score standardization on all dimensions of the data. This method transforms all data into a standard normal distribution with a mean of 0 and a standard deviation of 1, ensuring that each feature contributes equally to the model training.

[0059] High-precision time alignment: Different sensors and actuators have different sampling frequencies and data upload delays, resulting in misalignment of the original data in terms of timestamps. In order to obtain a complete snapshot vector representing the instantaneous state of the system at any time, this module adopts a time resampling technology based on spline interpolation. By setting a unified, high-frequency time reference, this technology can generate a smooth interpolation function based on several points before and after each data stream, thereby accurately estimating the synchronization value of all variables at each standard timestamp, and finally forming a strictly aligned multidimensional time series data stream, which is provided as output to the hazardous pollutant impact event identification module 20.

[0060] The hazardous pollutant impact event identification module 20 is connected to the data acquisition module 10 and is used to identify the occurrence of hazardous pollutant impact events based on multidimensional time series data.

[0061] In this embodiment, the model is implemented as a carefully designed autoencoder neural network. The specific technical implementation of the hazardous pollutant impact event identification module 20 is divided into two closely linked stages: offline modeling and online monitoring.

[0062] Offline modeling stage:

[0063] The goal of this stage is to enable the model to deeply learn and internalize the inherent patterns and high-dimensional nonlinear coupling relationships contained in the multidimensional time series data provided by the data acquisition module 10 under the healthy operation state of the sewage treatment system for several months or even longer.

[0064] Model Construction: The autoencoder consists of an encoder and a decoder. The encoder network is responsible for compressing the input high-dimensional data into a latent space with a dimension much lower than the original dimension through a series of nonlinear transformations. This compression process forces the model to remove the coarse parts and retain the core, learning the most representative core features in the data. The decoder network is responsible for receiving this compressed feature vector and attempting to accurately reconstruct it back into the original high-dimensional data through inverse nonlinear transformations.

[0065] Training process: The autoencoder is trained using a large amount of clean, normal operating state data. The objective function of the training is to minimize the reconstruction error between the input data vector and the reconstructed data vector output by the decoder. The mean squared error is usually used as the loss function. The network weights are continuously adjusted through the backpropagation algorithm. The model will eventually learn an approximate identity transformation: for any data input that belongs to the normal mode, the model can reconstruct it with a very low error. At this point, it can be said that the model has a deep understanding of what is normal.

[0066] Online monitoring phase:

[0067] Once the model is trained, it is deployed in the online monitoring process to continuously analyze the real-time data stream.

[0068] Real-time reconstruction and error calculation:

[0069] Each frame of standardized, time-aligned multidimensional data vector output by the data acquisition module 10 is fed into the pre-trained autoencoder in real time. The model immediately performs encoding and decoding operations on it and calculates the repackaging error at the current moment.

[0070] Anomaly Criterion Generation: The ingenuity of this method lies in the fact that when the system encounters a sudden surge of harmful pollutants, the chemical composition and harmful substances in the influent water will cause drastic changes in the biochemical reaction kinetics. This change will be quickly reflected in the combination of multiple dimensions such as DO, ORP, and flow rate, forming a data fingerprint that is completely different from the historical normal pattern. When this abnormal fingerprint is input into the autoencoder, since the model has never seen such a pattern during the training phase, its inherent decoding ability based on normal knowledge will be unable to accurately reconstruct this unfamiliar input. As a result, the reconstruction error will instantly and significantly deviate from its low-level stable range under normal operating conditions, producing a sharp upward spike.

[0071] This real-time reconstruction error itself constitutes an objective, data-driven anomaly score. It is no longer a rule set by humans, but a direct quantification of the degree of cognitive dissonance when the model faces unknown patterns.

[0072] Event Triggering: To convert continuous abnormal scores into discrete trigger signals, this module sets a dynamic and statistically significant trigger threshold based on statistical analysis of the reconstruction error distribution under historical normal operating conditions. When the real-time reconstruction error stably exceeds this threshold for several consecutive sampling points, the module determines that a hazardous pollutant impact event has officially occurred and immediately locks the current time as the event start time. Simultaneously, an event trigger signal is generated and transmitted to the simulation model and control strategy generation module 30 to initiate an emergency response.

[0073] The simulation model and control strategy generation module 30 is connected to the hazardous pollutant impact event identification module 20. It contains a simulation model to predict the future water quality status when the event occurs and to generate equipment linkage control strategies.

[0074] In this embodiment, the core technology of the simulation model and control strategy generation module 30 is the model predictive control framework. The reason why the model predictive control framework can effectively cope with the complex dynamic process of harmful pollutant impact is that it inherently possesses the ability to predict, optimize, and handle constraints. The detailed working mechanism of the simulation model and control strategy generation module 30 can be decomposed into the following key technical aspects:

[0075] Simulation model:

[0076] The simulation model and control strategy generation module 30 contains a pre-built dynamic simulation model that can accurately describe the biochemical reaction process of wastewater treatment. This model is the cornerstone of the model predictive control framework. This model can be:

[0077] Mechanistic models: For example, mechanistic models built based on the activated sludge model family published by the International Water Association. These models describe the biochemical processes such as microbial growth, death, substrate consumption, and pollutant transformation through a series of complex differential equations. The key kinetic and stoichiometric parameters in the model constitute the parameter vector. These parameters are precisely the optimization targets of the subsequent self-evolution module 40.

[0078] Data-driven models: In some cases, advanced data-driven models can also be used, such as long short-term memory networks or recurrent neural networks such as gated recurrent units. These models can directly establish an end-to-end dynamic mapping relationship from input to output by learning from massive amounts of historical operating data, without having to know the complex biochemical mechanisms in advance.

[0079] Regardless of the model used, the core functionality remains the same:

[0080] Receive the current system status. Then, it can be based on a given future series of control input sequences. Rapidly extrapolate forward to predict the system's future performance in the prediction time domain. Internal state evolution trajectory .

[0081] Optimization goal:

[0082] The intelligence of the model predictive control framework lies in its ability to find the optimal control strategy at each decision-making moment by solving an optimization problem, the objective function of which is... Carefully designed to balance water quality compliance (effectiveness) and resource consumption (cost):

[0083] ;

[0084] The first item (water quality deviation cost): the predicted effluent water quality for this penalty. Compared with the preset target value The deviation between them, weight matrix The settings reflect the importance attached to different water quality indicators.

[0085] The second item (controlling incremental costs): drastic changes in the penalty control action. It reflects considerations for operational stability and energy / pharmaceutical consumption, avoiding frequent start-ups or significant adjustments to control equipment, thereby reducing losses and operating costs. (Weight matrix) Used to balance economic factors.

[0086] Constraint handling:

[0087] This is another major advantage of model predictive control frameworks compared to traditional control: when solving optimization problems, the physical and operational constraints of the system can be explicitly included.

[0088] Input constraints: For example, the maximum frequency of the aeration blower, the maximum flow rate of the return pump, and the minimum or maximum dosing rate of the dosing pump.

[0089] State constraints: For example, the dissolved oxygen concentration in an aerobic tank must be maintained at a certain minimum level to ensure the survival of microorganisms.

[0090] Scrolling optimization:

[0091] The simulation model and control strategy generation module 30 do not calculate the entire event handling plan at once, but instead adopt a rolling time domain or backward time domain strategy to cyclically execute the following steps within each control cycle:

[0092] Get Status: Obtain the current time from data acquisition module 10. Latest system status .

[0093] Optimization solution: with Given the initial conditions, and assuming all constraints are satisfied, solve the above objective function. The minimization problem yields coverage of the future control time domain. (generally ) optimal control sequence .

[0094] First step: Extract only the first control command from this calculated optimal sequence. It then immediately sends it to the underlying PLC or DCS system for execution.

[0095] Roll forward: Discard the rest of the sequence in the next control cycle (time). When the arrival of the new measurement status, the system will obtain the new measurement status. Then repeat steps one and three.

[0096] Through this rolling optimization mechanism, the simulation model and control strategy generation module 30 can continuously use the latest field feedback information to correct its future decisions, thereby achieving dynamic and precise online closed-loop control of the harmful pollutant impact process.

[0097] The post-treatment effect evaluation and model self-evolution module 40 is connected to the simulation model and control strategy generation module 30, and is used to update the model parameters of the simulation model after the event treatment is completed.

[0098] In this embodiment, the technical implementation details of the post-treatment effect evaluation and model self-evolution module 40 are as follows:

[0099] Step 1: Establishing a digital archive of the entire event process

[0100] When a hazardous pollutant impact event occurs from the initial moment By the end time After completion, this module will first extract and integrate data from the system's historical database to form a complete digital archive of the event, containing three sets of core time-series data. These three sets of data constitute an objective and comprehensive description of the event:

[0101] Actual inflow disturbance sequence This set of data is the trigger for the event. It accurately records the actual changes of all relevant external disturbance variables during the event, such as fluctuations in influent flow, concentration curves of key pollutants, and changes in water temperature.

[0102] Actual control execution sequence This set of data represents the countermeasures taken by the system. It records the precise timing and magnitude of each control operation actually performed by the underlying equipment under the guidance of the control strategy generation module 30, such as the frequency change curve of the blower in the aeration system, the speed record of the emergency carbon source addition pump, and the adjustment history of the internal and external reflux ratio.

[0103] Actual water quality response sequence This set of data represents the final outcome of the event. It is measured and recorded by online sensors and reflects the actual evolution trajectory of key state variables within the wastewater treatment system and in the effluent under the combined effects of the aforementioned disturbances and controls. Examples include the change curves of effluent COD and ammonia nitrogen concentrations, and the actual fluctuations in dissolved oxygen concentration in the aerobic tank.

[0104] Step 2: Constructing an optimization problem based on model entity mismatch.

[0105] After obtaining the complete digital archive, the core task of the post-treatment effect evaluation and model self-evolution module 40 is to quantify and reduce the gap between the simulation model and the real physical world, i.e., model entity mismatch. To this end, the post-treatment effect evaluation and model self-evolution module 40 constructs a key parameter vector of the simulation model. This is a nonlinear optimization problem involving optimizing variables.

[0106] The objective function of this optimization problem is Defined as the cumulative prediction error, which is the sum of the squares of the Euclidean distances between the model-predicted trajectory and the actual response trajectory over the duration of the event:

[0107] ;

[0108] To understand the deeper meaning of this formula: It is an objective fact that has already occurred and cannot be changed, and This is the result of virtual playback, where we will use the actual historical sequence of water inflow disturbances. and actual control execution sequence As input, the current parameter is The simulation model is used to predict how the system should have responded in that historical period. The objective function directly and unbiasedly measures the response under the current parameters. Given the degree of deviation between the model's predictions and reality, our goal is to find a new set of parameters. This deviation To reach the minimum.

[0109] Step 3: Achieve adaptive parameter updates through iterative optimization.

[0110] To solve the above optimization problem, we need to find the solution that makes... Minimize the optimal parameters The post-treatment effect evaluation and model self-evolution module 40 adopts a gradient-based optimization algorithm.

[0111] In this embodiment, the Adam optimization algorithm is used, which is an advanced improvement on the traditional gradient descent method. Its parameter update process is as follows:

[0112] In each iteration, the algorithm first calculates the objective function. Relative to the current parameter gradient This gradient indicates the direction of parameter adjustment that can most quickly reduce prediction error. Unlike simple gradient descent, the Adam algorithm maintains estimates of the first moment (momentum) and the second moment (rate of change) of the gradient. This allows it to adaptively calculate different, more suitable learning rates for each individual model parameter. This not only accelerates convergence but also helps the algorithm escape local optima and find better solutions.

[0113] After multiple iterative calculations, when the objective function After the values ​​converge to a stable state or reach the preset number of iterations, the algorithm outputs a set of optimized model parameters. .

[0114] Finally, the post-treatment effect evaluation and model self-evolution module 40 will use this set of new parameters derived from real events. The parameters are automatically transmitted and overwritten into the simulation model of the simulation model and control strategy generation module 30, replacing the original old parameters. .

[0115] Through the complete process of establishing archives, constructing optimizations, and iteratively solving, the post-treatment effect evaluation and model self-evolution module realizes the closed-loop self-evolution of the model. Each hazardous pollutant impact event is no longer just a crisis that needs to be passively dealt with, but is transformed into a valuable learning opportunity to drive the system's cognitive upgrade, thereby continuously improving the system's prediction accuracy and the optimality of control strategies when dealing with unknown impacts in the future.

[0116] See attached document Figure 2 A data processing method for continuous monitoring of trace amounts of hazardous pollutants in wastewater based on intelligent parameter tuning, including:

[0117] S1. Collect and preprocess multidimensional time series data of the wastewater treatment system;

[0118] S2. Identify the occurrence of hazardous pollutant impact events based on multidimensional time series data;

[0119] S3. When a harmful pollutant impact event is detected, the simulation model is used to predict the future water quality status, and an equipment linkage control strategy is generated based on the prediction results.

[0120] S4. After the event is resolved, update the model parameters of the simulation model.

[0121] S5. Among them, identifying the occurrence of hazardous pollutant impact events includes:

[0122] Using multidimensional time series data from historical normal operating conditions, an unsupervised anomaly detection model is trained to learn normal data patterns.

[0123] The real-time collected multidimensional time series data is input into the unsupervised anomaly detection model to obtain the real-time reconstruction error that quantifies the degree of deviation between the current data and the normal pattern;

[0124] When the real-time reconstruction error exceeds the preset dynamic threshold, a hazardous pollutant impact event is determined to have occurred.

Claims

1. A wastewater hazardous pollutant trace monitoring system based on intelligent parameter adjustment, characterized in that, include: The data acquisition module is used to collect and preprocess multidimensional time series data from the wastewater treatment system. A hazardous pollutant impact event identification module, connected to the data acquisition module, is used to identify the occurrence of hazardous pollutant impact events based on the multidimensional time series data; The simulation model and control strategy generation module is connected to the harmful pollutant impact event identification module. It contains a simulation model to predict the future water quality status when the event occurs and to generate equipment linkage control strategies. The post-treatment effect evaluation and model self-evolution module is connected to the simulation model and control strategy generation module, and is used to update the model parameters of the simulation model after the event treatment is completed; The post-treatment effect evaluation and model self-evolution module is configured as follows: Obtain the actual influent disturbance sequence, actual control execution sequence, and actual water quality response sequence covering the entire process of the event; An optimization problem is constructed, with the model parameters of the simulation model as optimization variables, and the objective is to minimize the cumulative prediction error between the predicted water quality response sequence output by the simulation model and the actual water quality response sequence. Solve the optimization problem to obtain a set of updated model parameters, and use them to update the simulation model and the simulation model in the control strategy generation module; The hazardous pollutant impact event identification module is configured as follows: Using multidimensional time series data from historical normal operating conditions, an unsupervised anomaly detection model is trained to learn normal data patterns. Real-time data is input into the model to obtain the real-time reconstruction error that quantifies the degree of deviation; When the real-time reconstruction error exceeds a preset dynamic threshold, an event trigger signal is generated; The simulation model is used to predict future water quality conditions and generate equipment linkage control strategies, including: Using the current system state as the initial condition, a model predictive control framework is adopted to perform rolling optimization within a preset prediction time domain. The optimization problem with the objective of minimizing the comprehensive cost function of water quality deviation cost and operation cost is solved to obtain the optimal control sequence, and the first control command of the sequence is executed.

2. The wastewater hazardous pollutant trace monitoring system based on intelligent parameter adjustment according to claim 1, characterized in that, The unsupervised anomaly detection model in the hazardous pollutant impact event identification module is an autoencoder neural network model.

3. The wastewater hazardous pollutant trace monitoring system based on intelligent parameter adjustment according to claim 1, characterized in that, The simulation model and control strategy generation module is configured to use a model predictive control framework to perform rolling optimization and generate the equipment linkage control strategy.

4. The wastewater hazardous pollutant trace monitoring system based on intelligent parameter adjustment according to claim 1, characterized in that, The post-treatment effect evaluation and model self-evolution module is configured to use gradient descent or a variant thereof to solve the optimization problem to obtain updated model parameters.

5. The wastewater hazardous pollutant trace monitoring system based on intelligent parameter adjustment according to claim 1, characterized in that, Updating the model parameters of the simulation model includes: After the incident is resolved, obtain the actual inflow disturbance sequence, the actual control execution sequence, and the actual water quality response sequence covering the entire incident process; An optimization problem is constructed with the model parameters of the simulation model as optimization variables, aiming to minimize the cumulative prediction error between the predicted water quality response sequence output by the simulation model after inputting the actual influent disturbance sequence and the actual control execution sequence and the actual water quality response sequence. Solve the optimization problem to obtain a set of updated model parameters, and use them to replace the original model parameters.

6. The wastewater hazardous pollutant trace monitoring system based on intelligent parameter adjustment according to claim 1, characterized in that, The unsupervised anomaly detection model is an autoencoder neural network model.

7. A method for processing data from continuous monitoring of trace amounts of hazardous pollutants in wastewater based on intelligent parameter tuning, according to any one of claims 1-6, characterized in that, Includes the following steps: Collect and preprocess multidimensional time-series data from wastewater treatment systems; Based on the aforementioned multidimensional time series data, the occurrence of hazardous pollutant impact events is identified; When a hazardous pollutant impact event is detected, a simulation model is used to predict the future water quality status, and an equipment linkage control strategy is generated based on the prediction results. After the event is resolved, the model parameters of the simulation model are updated. The identification of hazardous pollutant impact events includes: Using multidimensional time series data from historical normal operating conditions, an unsupervised anomaly detection model is trained to learn normal data patterns. The real-time collected multidimensional time series data is input into the unsupervised anomaly detection model to obtain the real-time reconstruction error that quantifies the degree of deviation between the current data and the normal pattern; When the real-time reconstruction error exceeds a preset dynamic threshold, a hazardous pollutant impact event is determined to have occurred.

Images