Reclaimed water reuse regulation and control method and system based on artificial intelligence and digital twinning

Abstract

The invention relates to a reclaimed water reuse regulation and control method and system based on artificial intelligence and digital twinning, and the method comprises the steps: collecting initial data in a water treatment system, and carrying out the data preprocessing of the initial data to generate preprocessed data; constructing a digital twinning virtual model, and updating model parameters of the digital twinning virtual model based on the preprocessed data through a Bayesian reasoning algorithm to obtain a target digital twinning virtual model; outputting a water quality prediction result in a future preset time period through a pre-constructed water quality prediction model in the decision-making layer, and performing a rehearsal control strategy in the target digital twin virtual model based on the water quality prediction result to generate an optimization control instruction; performing data anomaly detection and fault diagnosis processing based on the preprocessed data through an anomaly detection model and an isolated forest algorithm to generate a fault diagnosis report; and issuing the optimization control instruction to an execution mechanism. The system response speed is improved, the control strategy perspectiveness is enhanced, and the equipment operation and maintenance intelligent level is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control of water treatment, and particularly relates to a reclaimed water reuse regulation method and system based on artificial intelligence and digital twinning. BACKGROUND

[0002] With the acceleration of global urbanization and the increasingly serious problem of water resource shortage, reclaimed water reuse as an important means to alleviate water resource pressure still has significant room for improvement.

[0003] The current traditional reclaimed water recycling treatment system has three major technical bottlenecks: first, the system has low operating efficiency, mainly relies on manual experience to adjust process parameters, resulting in serious lag in response to water quality fluctuations, and is difficult to cope with sudden changes in water quality; second, the intelligent level of equipment operation and maintenance is insufficient, and there is a lack of effective fault prediction mechanism, and equipment abnormalities are often discovered at a serious stage, causing unplanned downtime and rising maintenance costs; third, the control strategy lacks foresight and cannot be adjusted based on water quality trends, resulting in unstable water quality. These problems seriously restrict the reliability and economy of the reclaimed water reuse system. Although artificial intelligence and digital twinning technology have shown application potential in industrial control, there is still a technical gap in key areas such as real-time data fusion, model dynamic updating, and predictive control in the context of reclaimed water reuse. The existing technology needs to be improved in view of the above problems. SUMMARY

[0004] The purpose of the embodiments of the present application is to provide a reclaimed water reuse regulation method and system based on artificial intelligence and digital twinning, which has the advantages of improving system response speed, enhancing control strategy foresight, and improving equipment operation and maintenance intelligence level.

[0005] To solve the above technical problems, the embodiments of the present application provide a reclaimed water reuse regulation method based on artificial intelligence and digital twinning, comprising:

[0006] The initial data in the water treatment system is collected in real time through the perception layer, and the initial data is preprocessed in the edge computing layer to generate preprocessed data, wherein the initial data includes water quality parameters, equipment state data and process data;

[0007] A digital twinning virtual model is constructed in the digital twinning modeling layer, and the model parameters of the digital twinning virtual model are updated based on the preprocessed data through a Bayesian inference algorithm to obtain a target digital twinning virtual model;

[0008] The water quality prediction model pre-constructed in the decision layer outputs water quality prediction results of a future preset time period, and based on the water quality prediction results, a control strategy is pre-performed in the target digital twin virtual model to generate an optimized control instruction;

[0009] The abnormality detection model and the isolation forest algorithm pre-constructed perform data abnormality detection and fault diagnosis processing based on the preprocessed data to generate a fault diagnosis report.

[0010] The optimized control instruction is delivered to an actuator through the edge computing layer.

[0011] To solve the above technical problems, the application embodiment provides a reclaimed water reuse regulation and control system based on artificial intelligence and digital twinning, comprising:

[0012] A data acquisition module is configured to acquire initial data in a water treatment system in real time through a perception layer and perform data preprocessing on the initial data in an edge computing layer to generate preprocessed data, wherein the initial data comprises water quality parameters, equipment state data and process data.

[0013] A digital twin model construction module is configured to construct a digital twin virtual model in a digital twin modeling layer, update model parameters of the digital twin virtual model based on the preprocessed data through a Bayesian inference algorithm, and obtain a target digital twin virtual model.

[0014] An optimized control instruction generation module is configured to output water quality prediction results of a future preset time period through a water quality prediction model pre-constructed in a decision layer, perform a control strategy in the target digital twin virtual model based on the water quality prediction results, and generate an optimized control instruction.

[0015] A fault diagnosis module is configured to perform data abnormality detection and fault diagnosis processing based on the preprocessed data through an abnormality detection model and an isolation forest algorithm pre-constructed to generate a fault diagnosis report.

[0016] An optimized control instruction delivery module is configured to deliver the optimized control instruction to an actuator through the edge computing layer.

[0017] The embodiment of the application provides a reclaimed water reuse regulation method and system based on artificial intelligence and digital twinning. The method comprises the following steps: collecting initial data in a water treatment system in real time through a perception layer, and performing data preprocessing on the initial data in an edge computing layer to generate preprocessed data, wherein the initial data comprises water quality parameters, equipment state data and process data; constructing a digital twinning virtual model in a digital twinning modeling layer, updating model parameters of the digital twinning virtual model based on the preprocessed data through a Bayesian inference algorithm to obtain a target digital twinning virtual model; outputting a water quality prediction result of a future preset time period through a preconstructed water quality prediction model in a decision layer, performing a rehearsal control strategy in the target digital twinning virtual model based on the water quality prediction result to generate an optimized control instruction; performing data anomaly detection and fault diagnosis processing based on the preprocessed data through a preconstructed anomaly detection model and an isolation forest algorithm to generate a fault diagnosis report; and issuing the optimized control instruction to an executing mechanism through the edge computing layer. The embodiment of the application realizes accurate regulation of a reclaimed water reuse system through real-time data collection and preprocessing, dynamic updating of a digital twinning model, predictive control strategy generation and intelligent fault diagnosis, and has the advantages of improving system response speed, enhancing the forward-looking nature of a control strategy and improving the intelligent level of equipment operation and maintenance. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the scheme in the present application, the drawings needed in the description of the embodiments of the present application will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0019] Figure 1 is an implementation flowchart of the method for regulating reclaimed water reuse based on artificial intelligence and digital twinning provided by the embodiment of the present application;

[0020] Figure 2 is an implementation flowchart of the first sub-process in the method for regulating reclaimed water reuse based on artificial intelligence and digital twinning provided by the embodiment of the present application;

[0021] Figure 3 is an implementation flowchart of the second sub-process in the method for regulating reclaimed water reuse based on artificial intelligence and digital twinning provided by the embodiment of the present application;

[0022] Figure 4 is an implementation flowchart of the third sub-process in the method for regulating reclaimed water reuse based on artificial intelligence and digital twinning provided by the embodiment of the present application;

[0023] Figure 5is a fourth sub-process implementation flowchart of the reclaimed water reuse regulation method based on artificial intelligence and digital twinning provided by the embodiment of the application;

[0024] Figure 6 is a fifth sub-process implementation flowchart of the reclaimed water reuse regulation method based on artificial intelligence and digital twinning provided by the embodiment of the application;

[0025] Figure 7 is a sixth sub-process implementation flowchart of the reclaimed water reuse regulation method based on artificial intelligence and digital twinning provided by the embodiment of the application;

[0026] Figure 8 is a schematic diagram of the reclaimed water reuse regulation system based on artificial intelligence and digital twinning provided by the embodiment of the application. DETAILED DESCRIPTION

[0027] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used in the description herein is for describing particular embodiments only and is not intended to be limiting of the application; the description and the drawings are to be regarded as illustrative in nature and are not intended to limit the application; the terminology used in the description and the claims of the present application and the above description of the drawings includes the terms specifically mentioned above as well as their derivatives.

[0028] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase that the phrase in the specification appear in various places in the specification does not necessarily all refer to the same embodiment, or necessarily refer to different or alternative embodiments to each other. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with each other.

[0029] In order to enable persons skilled in the art to better understand the scheme of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the drawings.

[0030] The application will be described in detail below in conjunction with the drawings and embodiments.

[0031] The application also provides a water reuse regulation system, the overall architecture of the system includes a perception layer, an edge computing layer, a digital twin modeling layer, a decision layer and an application interaction layer. The perception layer is deployed in the multi-source sensor network on site, and real-time acquisition of water quality parameters (pH, COD, turbidity, conductivity), liquid level, flow, equipment running state and other key data is provided as the original input for the upper layer. The edge computing layer: the local edge node undertakes the preliminary data cleaning, feature extraction and lightweight AI inference task, reduces the load of the center platform and improves the response speed; the edge computing layer supports local decision and emergency control when the network is disconnected. The digital twin modeling layer constructs a virtual model that evolves synchronously with the physical system, combines CFD fluid dynamics simulation and water quality migration mechanism model, realizes real-time state mapping and process prediction; the digital twin modeling layer dynamically updates the model parameters through the online learning mechanism, to ensure high consistency with the actual system. The decision layer introduces deep learning and reinforcement learning algorithms, combined with the digital twin model to realize multi-objective intelligent optimization: (1) use LSTM, Transformer and other models for water quality trend prediction; (2) based on DDPG, PPO and other reinforcement learning methods, automatically adjust the dosing strategy, aeration intensity, pump station start-stop and other key process parameters; (3) use digital twin as a virtual training ground to perform strategy rehearsal and risk assessment, improve the generalization ability and stability of the AI model. The application interaction layer provides visual monitoring screens, mobile App and remote control terminals, supports multi-role permission management, abnormal alarm push, remote operation instruction issuance and other functions, and realizes efficient human-machine collaboration.

[0032] In the prior art, the reclaimed water reuse system has long relied on manual experience to adjust operating parameters, resulting in a slow response and difficulty in responding to water quality fluctuations. The traditional method uses a fixed model for water quality prediction, which cannot adapt to dynamically changing water quality conditions, and equipment maintenance relies on regular inspection, lacking fault warning capability. For example, when the influent concentration of a certain sewage treatment plant suddenly changes, it takes several hours for the operator to adjust the parameters, during which the effluent indicators exceed the standard, causing resource waste.

[0033] In order to solve the above problems, the system response speed is limited due to the reliance on manual experience, real-time data acquisition and edge computing preprocessing are tried to be introduced, but it is found that the static model still cannot reflect the dynamic working condition. Further research shows that the Bayesian inference algorithm can dynamically update the model parameters combined with real-time data, and the water quality prediction model combined with digital twin rehearsal can generate control strategies in advance. At the same time, it is found that the isolation forest algorithm has an advantage in processing high-dimensional data in equipment anomaly detection, and finally an intelligent regulation and control scheme with multi-level cooperation is formed.

[0034] Therefore, the present application proposes a technical scheme of collecting initial data in a water treatment system in real time through a perception layer, and generating pre-processed data through data preprocessing in an edge computing layer; constructing a virtual model in a digital twin modeling layer and updating model parameters through a Bayesian inference algorithm; outputting prediction results through a water quality prediction model in a decision layer and generating optimized control instructions through strategy rehearsal; performing fault diagnosis through an anomaly detection model and an isolation forest algorithm; and finally issuing instructions to an actuator. Specifically, multi-source sensors continuously collect data such as influent COD, ammonia nitrogen concentration, and aeration machine speed, and generate a standardized data set after removing outliers by an edge node. The digital twin model updates the hydraulic retention time parameter once an hour through the Bayesian inference algorithm, for example, automatically adjusts the pool capacity parameter in the model when the influent flow suddenly increases. The water quality prediction model predicts the effluent total phosphorus concentration in the next 8 hours based on the past 72 hours of data, simulates the effect of different dosing amounts on the effluent in the virtual model, and selects the optimal control scheme. The autoencoder model calculates the reconstruction error of the sensor data, and when the error exceeds the threshold, it triggers an early warning, while the isolation forest algorithm detects abnormal waveforms in the pump vibration spectrum, and the decision tree model judges the fault type in combination with the historical fault library. Finally, the optimized instructions are transmitted to the actuator to adjust the dosing pump frequency through the OPC protocol.

[0035] Referring to Figure 1 , Figure 1 An embodiment of a reclaimed water reuse regulation method based on artificial intelligence and digital twinning is shown.

[0036] It should be noted that the method of the present application is not limited to the order of the flow shown, and includes the following steps: Figure 1

[0037] S1: Collecting initial data in a water treatment system in real time through a perception layer, and performing data preprocessing on the initial data in an edge computing layer to generate pre-processed data. The initial data includes water quality parameters, equipment state data, and process data.

[0038] Referring to Figure 2 , Figure 2 An embodiment of step S1 is shown, described as follows:

[0039] S11: Collecting the initial data in the water treatment system in real time through a multi-source sensor network and an equipment state monitor of the perception layer. S12: Transmitting the initial data to an edge node in the edge computing layer through an industrial bus. S13: Performing data cleaning and feature extraction on the initial data in the edge node to generate the pre-processed data.

[0040] ​Specifically, when the water treatment system is running, the pH sensor and the dissolved oxygen sensor installed in the reaction tank, the sedimentation tank, and the filtration unit continuously collect water quality parameters, and the equipment state monitor at the pump and valve synchronously acquires vibration spectrum and current fluctuation data. These initial data are encapsulated into data frames through an industrial bus protocol and transmitted to the edge node deployed in the control cabinet with millisecond-level delay. The edge node first performs sliding window filtering on the received raw data, for example, calculates the moving average with a 5-second time window to eliminate transient noise; then identifies data points outside the normal fluctuation range based on the 3σ criterion and replaces them by interpolation. The cleaned data enters the feature extraction stage, for example, calculates the hourly mean, standard deviation, and linear regression slope of the turbidity data in the past 30 minutes to generate preprocessed data containing 12 feature vectors. The preprocessed data is stored locally and synchronized to the cloud, providing input for subsequent model updating and decision-making.

[0041] Among them, the multi-source sensor network refers to a cluster of multiple types of sensors deployed in different process sections of the water treatment system, which can be implemented by combining pH sensors, dissolved oxygen sensors, turbidity sensors, and flow meters to cover comprehensive collection of water quality parameters and solve the problem of insufficient data dimension of traditional single sensor. The equipment state monitor refers to an embedded device integrating vibration sensors, temperature sensors, and current detection modules, which can be implemented by an intelligent monitoring terminal with multi-parameter fusion to capture the running state of devices such as pumps and valves in real time and solve the problem of low efficiency of manual inspection. The industrial bus refers to the physical layer and protocol layer specifications that support industrial field communication, which can be implemented by PROFINET, ModbusRTU, or EtherCAT protocols to ensure the real-time performance and anti-interference capability of data transmission and solve the problem of processing lag caused by wireless network delay and packet loss. The edge node refers to a computing unit deployed near the device, which can be implemented by an embedded industrial computer or an edge server to perform preprocessing operations at the data source and solve the bandwidth pressure and response delay caused by centralized processing in the cloud. Data cleaning refers to filtering noise and outliers in raw data, which can be implemented by a sliding window mean filtering algorithm combined with the 3σ criterion to eliminate the influence of sensor drift and transient interference. Feature extraction refers to extracting key statistics from time series data, which can be implemented by calculating the mean, variance, and trend slope within a sliding time window to reduce data dimension and retain effective information.

[0042] S2: Construct a digital twin virtual model in the digital twin modeling layer, update the model parameters of the digital twin virtual model based on the preprocessed data through Bayesian inference algorithm, and obtain a target digital twin virtual model.

[0043] The digital twin virtual model refers to a virtual mirror based on CFD fluid dynamics, which can be implemented by using ANSYS Fluent software, and is used to simulate the real water treatment process. The Bayesian inference algorithm refers to a parameter optimization method based on probability statistics, which can be implemented by using Markov chain Monte Carlo sampling, and is used to dynamically correct model bias.

[0044] Please refer to Figure 3 , Figure 3 An embodiment of step S2 is shown as follows:

[0045] S21: constructing the digital twin virtual model based on a CFD fluid dynamics simulation tool in the digital twin modeling layer. The model parameters of the digital twin virtual model include the reaction tank volume, the hydraulic retention time, the influent concentration fluctuation range, and the sludge return ratio. S22: constructing a Gaussian distribution prior probability based on the model parameters, and updating the model parameters of the digital twin virtual model based on the Gaussian distribution prior probability and the preprocessed data by using the Bayesian inference algorithm to obtain the target digital twin virtual model.

[0046] Specifically, when constructing the digital twin virtual model, a physical model containing the reaction tank structure, fluid motion, and mass transfer is first established by CFD fluid dynamics simulation. The reaction tank volume in the model parameters can be set according to the actual structure size, for example, by inputting the length, width, and height data using a parameterized modeling tool; the hydraulic retention time can be obtained by calculating the flow path and velocity distribution of the fluid in the reaction tank. Subsequently, a Gaussian distribution prior probability is established for each parameter based on historical operation data, for example, the prior mean of the sludge return ratio is set to the typical value under the conventional working condition. In the model updating stage, the residual error between the model prediction value and the actual value is calculated by periodically collecting sensor data, and the Bayesian inference algorithm is used to correct the prior distribution by using the residual error as the observation data, for example, the Markov chain Monte Carlo method is used to iteratively update the parameter posterior distribution, so that the reaction tank volume, the hydraulic retention time, and other parameters of the virtual model dynamically track the changes of the actual system.

[0047] The reaction pool volume refers to the space capacity for accommodating sewage for physical and chemical reactions, which can be realized by using the three-dimensional geometric modeling function in the CFD simulation tool to accurately reflect the treatment capacity of the actual reaction pool. The hydraulic retention time refers to the average residence time of sewage in the reaction pool, which can be simulated by the fluid motion trajectory in the CFD simulation to evaluate the time efficiency of the treatment process. The influent concentration fluctuation range refers to the variation interval of the pollutant concentration entering the reaction pool, which can be obtained by statistical analysis of real-time data collected by sensors to represent the uncertainty of the system input. The sludge reflux ratio refers to the ratio of the sludge reflux amount to the influent amount in the treatment process, which can be realized by correlating the process control parameters with the mass transfer equation in the CFD model to optimize the biological treatment efficiency. The Gaussian distribution prior probability refers to the statistical hypothesis of the initial state of the model parameters, which can be realized by fitting the parameter distribution range based on historical operation data to constrain the initial conditions of Bayesian inference.

[0048] Please refer to Figure 4 , Figure 4 An embodiment of step S22 is shown as follows:

[0049] S221: Constructing the Gaussian distribution prior probability for the model parameters. S222: Collecting actual sensor data from the pre-processed data periodically from the edge node, and calculating the residual based on the actual sensor data and the predicted value output by the digital twin virtual model. S223: Updating the model parameters of the digital twin virtual model based on the Gaussian distribution prior probability and the residual by using the Bayesian inference algorithm to obtain the target digital twin virtual model.

[0050] Specifically, a data collection period is set in the edge computing node, for example, pre-processed sensor measurement data is obtained every 5 minutes. The measurement data is input into the digital twin virtual model to predict the water quality under the current parameters, and the residual sequence of the predicted value and the measured value is obtained by subtraction operation. Based on the pre-constructed Gaussian prior distribution, the residual data is used as the likelihood function by using the Bayesian theorem, and the parameter posterior distribution is derived by probability multiplication operation. With the passage of time, this posterior distribution will serve as the prior distribution for the next round of update, forming a closed-loop iterative optimization mechanism. When the influent concentration suddenly changes, the residual increases to trigger parameter update, so that the model parameters are automatically adjusted to the optimal value under the new working condition. The embodiment of the present application realizes online adaptive optimization of the digital twin model parameters, effectively eliminating the model errors caused by sensor data drift and process fluctuations. This scheme can automatically identify the deviation between the actual running data and the model prediction, gradually correct the parameter distribution through probability iteration, and ensure the dynamic consistency of the virtual model and the physical system. In the influent load fluctuation scenario, the model prediction error can be reduced to within the allowable range of process control, reducing the frequency of manual parameter calibration, and improving the control accuracy of the reclaimed water reuse system.

[0051] wherein the Gaussian distribution prior probability refers to an initial probability distribution assumption of the model parameters, and specifically, a normal distribution function can be used to probabilistically model the reaction pool volume and hydraulic retention time parameters, for representing the uncertainty of the initial state of the parameters. The residual refers to the difference between the actual sensor data and the model prediction value, and specifically, a mean square error algorithm can be used to calculate the difference between the two, for quantifying the deviation of the model output. The Bayesian inference algorithm refers to a parameter updating method based on a probability framework, and specifically, a Markov chain Monte Carlo method can be used to realize iterative calculation of the posterior probability distribution, for fusing real-time data and prior knowledge. S3: outputting water quality prediction results of a future preset time period through a pre-constructed water quality prediction model in the decision layer, and performing a pre-play control strategy in the target digital twin virtual model based on the water quality prediction results to generate an optimized control instruction.

[0052] wherein the water quality prediction model refers to a deep learning model that fuses time series features, and specifically, an LSTM-Transformer hybrid architecture can be used to realize the model, for capturing water quality change rules.

[0053] Please refer to Figure 5 , Figure 5 An embodiment of step S3 is shown, which is described in detail as follows:

[0054] S31: obtaining historical water quality data of a preset time period to obtain target historical water quality data. S32: capturing short-term fluctuation information based on the target historical water quality data through an LSTM layer of the water quality prediction model in the decision layer to output a feature vector. S33: performing dependency relationship analysis based on the feature vector through a Transformer layer of the water quality prediction model to generate the water quality prediction results of a future preset time period. S34: performing a pre-play control strategy in the target digital twin virtual model based on the water quality prediction results to generate the optimized control instruction.

[0055] Specifically, after the target historical water quality data is processed by the LSTM layer, the gating mechanism can filter noise and retain key time series features, such as sudden changes in inflow or fluctuations in dissolved oxygen concentration. After the feature vector is input into the Transformer layer, the self-attention mechanism can automatically identify the correlation weights between different time step water quality parameters, such as the nonlinear relationship between aeration quantity and COD removal rate. The multi-dimensional water quality prediction results output by the fusion of the two models are input into the digital twin virtual model, and by simulating the influence of different control parameter combinations on the treatment process, such as adjusting the dosing amount or sludge return ratio, the optimized control instruction that meets the water quality standard and has the lowest energy consumption is generated. This process forms a closed-loop system of prediction-simulation-optimization, enabling the control strategy to adapt to the water quality change trend in advance.

[0056] The LSTM layer refers to a long short-term memory neural network layer, which can be specifically implemented by using a recurrent neural network structure with a forgetting gate, an input gate, and an output gate, is used to extract short-term fluctuation features of time series from historical water quality data, and solves the problem of response lag of traditional methods to water quality mutations. The Transformer layer refers to a deep learning model layer based on a self-attention mechanism, which can be specifically implemented by using a multi-head attention mechanism and a position encoding structure, is used to capture long-range dependencies between water quality parameters, and overcomes the defect of insufficient modeling of periodic changes by a single model. The rehearsal control strategy refers to simulating the influence of different control instructions on water quality in the digital twin virtual model, which can be specifically implemented by using a reinforcement learning algorithm combined with a simulation environment to perform policy iteration, and realizes dynamic optimization of the control scheme.

[0057] Please refer to Figure 6 , Figure 6 An embodiment of step S34 is shown as follows:

[0058] S341: An optimization objective function is constructed, and a virtual controller is constructed by using a PPO reinforcement learning algorithm. S342: The virtual controller generates the optimization control instruction by performing a rehearsal control strategy in the target digital twin virtual model based on the optimization objective function and the water quality prediction result.

[0059] Specifically, the optimization objective function defines the weight relationship between control objectives such as water quality compliance and energy consumption minimization, and converts the multi-dimensional optimization problem into a solvable numerical form through mathematical modeling. The virtual controller performs a rehearsal operation in the digital twin virtual model based on the PPO algorithm: first, the current water quality prediction result is loaded as the initial state, and then different control parameter combinations (such as aeration amount and reagent dosing rate) are iteratively generated in the simulation environment, and the optimal control instruction is selected by calculating the score of the optimization objective function. In this process, the PPO algorithm limits the step length of policy update to avoid deviation of the simulation result from the actual working condition due to policy mutation, and uses historical training data to accelerate policy convergence. After the rehearsal process is completed, the optimization control instruction is locked and output to the actuator, realizing dynamic generation and verification of the control strategy. The present application realizes dynamic optimization generation of the control strategy, solves the problem of response lag of manual adjustment, and ensures the effectiveness and safety of the control instruction under real working conditions through rehearsal verification in the digital twin environment. The multi-dimensional constraint design of the optimization objective function enables the system to automatically balance the water quality compliance rate and the operating cost, and the application of the PPO reinforcement learning algorithm improves the policy exploration efficiency of the virtual controller in complex scenarios, significantly enhancing the adaptive adjustment capability of the system to water quality fluctuations.

[0060] The optimization objective function refers to a mathematical expression for quantifying the effect of the control strategy, which can be implemented using a multi-objective optimization function. For example, the water quality compliance rate, energy consumption index, and operating cost can be used as constraint conditions, and a comprehensive score function can be constructed by weighted summation to provide quantifiable evaluation criteria for the generation of the control strategy. The PPO reinforcement learning algorithm refers to a reinforcement learning algorithm based on policy gradient, which can be implemented using a combination of offline training and online fine-tuning. For example, different water quality fluctuation scenarios can be simulated in the digital twin virtual model, the reward signal can be obtained by interacting with the environment through the policy network, and the policy parameters can be updated using the importance sampling mechanism to ensure that the virtual controller can adaptively adjust the control strategy in a dynamic environment.

[0061] S4: Perform data anomaly detection and fault diagnosis processing based on the preprocessed data through the pre-constructed anomaly detection model and the isolation forest algorithm, and generate a fault diagnosis report.

[0062] The anomaly detection model refers to a feature reconstruction model based on an autoencoder, which can be implemented using a stacked convolutional autoencoder to identify data anomaly patterns. The isolation forest algorithm refers to a tree structure-based anomaly detection method, which can be implemented by randomly dividing the feature space to detect device state anomalies.

[0063] Please refer to Figure 7 , Figure 7 A specific implementation of step S4 is shown as follows:

[0064] S41: Perform feature encoding and reconstruction error calculation on the preprocessed data through the anomaly detection model to obtain a target reconstruction error. S42: If the target reconstruction error exceeds a preset threshold, trigger a warning and mark it as a potential anomaly. S43: Perform anomaly behavior detection on the device state data in the preprocessed data using the isolation forest algorithm to obtain a target anomaly behavior. S44: Perform fault classification and fault diagnosis on the potential anomaly and the target anomaly behavior based on the expert knowledge base and historical fault records through the decision tree model to generate the fault diagnosis report.

[0065] Specifically, the pre-processed multi-dimensional operation data is first input into the anomaly detection model for feature space conversion, and the reconstruction error value is calculated by comparing the difference between the input data and the reconstructed output. When the error value exceeds the preset threshold, the system automatically triggers the early warning mechanism and marks the abnormal time point in the data stream. For the device state data stream, the isolated forest algorithm is used to construct multiple random partition trees, and the path length of each data point in the tree structure is used to identify abnormal operation behavior. Finally, the detected abnormal events are input into the decision tree model, combined with the fault feature matrix in the knowledge base and the statistical rules of the historical case library, the fault type matching and root cause reasoning are performed, and the structured report containing fault positioning information and disposal suggestions is output. The present application realizes the early discovery and accurate diagnosis of the operation anomaly of the water treatment system, improves the timeliness of the equipment fault warning and the reliability of the diagnosis conclusion, reduces the misjudgment caused by insufficient manual experience, and provides reliable decision support basis for equipment maintenance.

[0066] Among them, the anomaly detection model refers to a deep learning model based on an autoencoder structure, which can specifically realize feature encoding and reconstruction error calculation by using a stacked denoising autoencoder, and learns the normal mode of data through compression-decompression process. The isolated forest algorithm refers to an anomaly detection method based on a random partition strategy, which can specifically calculate the path length of data points to identify abnormal behavior by using a binary tree structure. The decision tree model refers to a classification model constructed based on an information entropy criterion, which can specifically realize fault type discrimination by using a C4.5 algorithm. The expert knowledge base refers to a structured database for storing equipment fault features and processing strategies, which can specifically store historical fault cases and diagnosis rules by using a relational database.

[0067] S5: The optimization control instruction is issued to the execution mechanism through the edge computing layer.

[0068] Specifically, the execution mechanism refers to a programmable logic controller, which can be a PLC execution mechanism, for example, which can be implemented by using Siemens S7-1200 series, and is used for executing the optimization control instruction.

[0069] In the embodiment of the present application, the initial data in the water treatment system is collected in real time through the perception layer, and the initial data is preprocessed in the edge computing layer to generate preprocessed data, wherein the initial data includes water quality parameters, equipment state data and process data; a digital twin virtual model is constructed in the digital twin modeling layer, the model parameters of the digital twin virtual model are updated based on the preprocessed data through the Bayesian inference algorithm to obtain a target digital twin virtual model; the water quality prediction result of a future preset time period is output through a pre-constructed water quality prediction model in the decision layer, and a control strategy is preformed in the target digital twin virtual model based on the water quality prediction result to generate an optimized control instruction; data anomaly detection and fault diagnosis processing are performed based on the preprocessed data through a pre-constructed anomaly detection model and an isolation forest algorithm to generate a fault diagnosis report; and the optimized control instruction is issued to an actuator through the edge computing layer. The embodiment of the present application realizes accurate regulation and control of the reclaimed water reuse system through real-time data collection and preprocessing, dynamic updating of a digital twin model, predictive control strategy generation and intelligent fault diagnosis, and has the advantages of improving system response speed, enhancing the forward-looking nature of the control strategy and improving the intelligent level of equipment operation and maintenance.

[0070] The embodiment of the present application realizes dynamic updating of model parameters through Bayesian inference, overcoming the defect that traditional static models cannot adapt to water quality fluctuations. The LSTM-Transformer hybrid prediction model is used, which can capture long-term dependencies better than a single ARIMA model. In terms of anomaly detection, the self-encoder and isolation forest algorithm are combined, which can identify more complex abnormal patterns than a single threshold alarm method. The digital twin pre-performance mechanism is introduced in the control instruction generation link, which has the advantage of forward-looking compared with traditional PID direct control. The embodiment of the present application realizes real-time dynamic optimization of water treatment parameters, shortening the system response time from hours to minutes. Predictive control strategy reduces the fluctuation range of effluent water quality and reduces the excessive addition of chemical reagents. The dual anomaly detection mechanism improves the accuracy of equipment fault identification and avoids unplanned downtime. The closed-loop control system reduces the frequency of manual intervention and realizes autonomous operation of the reclaimed water reuse system.

[0071] Please refer to Figure 8 , as an implementation of the method shown in Figure 1 , the present application provides an embodiment of a reclaimed water reuse regulation and control system based on artificial intelligence and digital twinning. The device embodiment corresponds to the method embodiment shown in Figure 1 , and the device can be applied to various electronic devices.

[0072] As Figure 8As shown, the reclaimed water reuse regulation system based on artificial intelligence and digital twinning of the embodiment comprises a data acquisition module 61, a digital twinning model construction module 62, an optimization control instruction generation module 63, a fault diagnosis module 64, and an optimization control instruction issuing module 65, wherein:

[0073] The data acquisition module 61 is configured to collect initial data in the water treatment system in real time through the perception layer, and perform data preprocessing on the initial data in the edge computing layer to generate preprocessed data, wherein the initial data comprises water quality parameters, equipment state data, and process data.

[0074] The digital twinning model construction module 62 is configured to construct a digital twinning virtual model in a digital twinning modeling layer, update model parameters of the digital twinning virtual model based on the preprocessed data through a Bayesian inference algorithm, and obtain a target digital twinning virtual model.

[0075] The optimization control instruction generation module 63 is configured to output a water quality prediction result of a future preset time period through a pre-constructed water quality prediction model in the decision layer, perform a pre-play control strategy in the target digital twinning virtual model based on the water quality prediction result, and generate an optimization control instruction.

[0076] The fault diagnosis module 64 is configured to perform data anomaly detection and fault diagnosis processing based on the preprocessed data through a pre-constructed anomaly detection model and an isolation forest algorithm, and generate a fault diagnosis report.

[0077] The optimization control instruction issuing module 65 is configured to issue the optimization control instruction to an execution mechanism through the edge computing layer.

[0078] Further, the digital twinning model construction module 62 comprises:

[0079] The model construction unit is configured to construct the digital twinning virtual model based on a CFD fluid dynamics simulation tool in the digital twinning modeling layer, wherein the model parameters of the digital twinning virtual model comprise a reaction pool volume, a hydraulic retention time, an influent concentration fluctuation range, and a sludge return ratio.

[0080] The model parameter updating unit is configured to construct a Gaussian distribution prior probability based on the model parameters, and update the model parameters of the digital twinning virtual model based on the Gaussian distribution prior probability and the preprocessed data through the Bayesian inference algorithm to obtain the target digital twinning virtual model.

[0081] Further, the model parameter updating unit comprises:

[0082] The Gaussian distribution prior probability construction unit is configured to construct the Gaussian distribution prior probability for the model parameters.

[0083] a residual calculation unit configured to collect actual sensor data from the edge node periodically from the preprocessed data, and calculate a residual according to the actual sensor data and a predicted value output by the digital twin virtual model;

[0084] a target digital twin virtual model generation unit configured to update model parameters of the digital twin virtual model based on the Gaussian distribution prior probability and the residual by the Bayesian inference algorithm, to obtain the target digital twin virtual model.

[0085] Further, the optimization control instruction generation module 63 comprises:

[0086] a historical water quality data construction unit configured to obtain historical water quality data of a preset time period, to obtain target historical water quality data;

[0087] a feature vector output unit configured to capture short-term fluctuation information based on the target historical water quality data by an LSTM layer of the water quality prediction model in the decision layer, to output a feature vector;

[0088] a water quality prediction result generation unit configured to perform dependency relationship analysis based on the feature vector by a Transformer layer of the water quality prediction model, to generate the water quality prediction result of a future preset time period;

[0089] a rehearsal control strategy unit configured to perform a rehearsal control strategy in the target digital twin virtual model based on the water quality prediction result, to generate the optimization control instruction.

[0090] Further, the rehearsal control strategy unit comprises:

[0091] an optimization objective function construction unit configured to construct an optimization objective function, and construct a virtual controller by using a PPO reinforcement learning algorithm;

[0092] an instruction generation unit configured to perform a rehearsal control strategy in the target digital twin virtual model based on the optimization objective function and the water quality prediction result by the virtual controller, to generate the optimization control instruction.

[0093] Further, the fault diagnosis module 64 comprises:

[0094] a reconstruction error calculation unit configured to perform feature encoding and reconstruction error calculation on the preprocessed data by the anomaly detection model, to obtain target reconstruction error;

[0095] a warning triggering unit configured to trigger a warning and mark as a potential anomaly if the target reconstruction error exceeds a preset threshold;

[0096] an abnormal behavior detection unit, configured to perform abnormal behavior detection on the device state data in the preprocessed data by using the Isolation Forest algorithm to obtain target abnormal behavior;

[0097] a fault diagnosis report generation unit, configured to perform fault classification and fault diagnosis on the potential abnormality and the target abnormal behavior based on an expert knowledge base and historical fault records by using a decision tree model to generate a fault diagnosis report.

[0098] Further, the data acquisition module 61 comprises:

[0099] an initial data acquisition unit, configured to collect the initial data in the water treatment system in real time by using a multi-source sensor network and a device state monitor of the perception layer;

[0100] a data transmission unit, configured to transmit the initial data to an edge node in the edge computing layer by using an industrial bus;

[0101] a data preprocessing unit, configured to perform data cleaning and feature extraction on the initial data in the edge node to generate preprocessed data.

[0102] In the embodiments of the present application, the data acquisition module is configured to collect initial data in a water treatment system in real time by using a perception layer and to perform data preprocessing on the initial data in an edge computing layer to generate preprocessed data, wherein the initial data comprises water quality parameters, device state data and process data; the digital twin model construction module is configured to construct a digital twin virtual model in a digital twin modeling layer, to update model parameters of the digital twin virtual model based on the preprocessed data by using a Bayesian inference algorithm, and to obtain a target digital twin virtual model; the optimized control instruction generation module is configured to output a water quality prediction result for a future preset time period by using a pre-constructed water quality prediction model in a decision layer, to perform a pre-implementation control strategy in the target digital twin virtual model based on the water quality prediction result, and to generate an optimized control instruction; the fault diagnosis module is configured to perform data anomaly detection and fault diagnosis processing based on the preprocessed data by using a pre-constructed anomaly detection model and an Isolation Forest algorithm, and to generate a fault diagnosis report; and the optimized control instruction issuing module is configured to issue the optimized control instruction to an actuator via the edge computing layer. The embodiments of the present application realize precise regulation and control of a reclaimed water reuse system by real-time data acquisition and preprocessing, dynamic updating of a digital twin model, predictive control strategy generation and intelligent fault diagnosis, and have the advantages of improving system response speed, enhancing the forward-looking nature of a control strategy and improving the intelligent level of equipment operation and maintenance.

[0103] Obviously, the above-described embodiments are only some embodiments but not all the embodiments of the present application, the preferred embodiments of the present application are shown in the drawings, but do not limit the scope of the present application. The present application can be implemented in many different forms, and conversely, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent replacements to some technical features therein. Any equivalent structure made by using the content of the specification and drawings, directly or indirectly applied to other related technical fields, is also within the protection scope of the present application.

Claims

1. A method for regulating and controlling wastewater reuse based on artificial intelligence and digital twins, characterized in that, include: The initial data of the water treatment system is collected in real time by the sensing layer, and the initial data is preprocessed in the edge computing layer to generate preprocessed data. The initial data includes water quality parameters, equipment status data and process data. A digital twin virtual model is constructed in the digital twin modeling layer. The model parameters of the digital twin virtual model are updated based on the preprocessed data using a Bayesian inference algorithm to obtain the target digital twin virtual model. In the decision-making level, a pre-built water quality prediction model outputs water quality prediction results for a future preset time period. Based on the water quality prediction results, a pre-simulation control strategy is performed in the target digital twin virtual model to generate optimized control instructions. Based on the pre-processed data, a data anomaly detection and fault diagnosis process is performed using a pre-built anomaly detection model and the isolated forest algorithm to generate a fault diagnosis report. The optimized control commands are sent to the actuators via the edge computing layer.

2. The method for regulating and controlling wastewater reuse based on artificial intelligence and digital twins according to claim 1, characterized in that, The process of constructing a digital twin virtual model in the digital twin modeling layer, and updating the model parameters of the digital twin virtual model based on the preprocessed data using a Bayesian inference algorithm to obtain the target digital twin virtual model, includes: In the digital twin modeling layer, the digital twin virtual model is constructed based on CFD fluid dynamics simulation tools. The model parameters of the digital twin virtual model include the reaction tank volume, hydraulic retention time, influent concentration fluctuation range, and sludge return ratio. Based on the model parameters, a Gaussian distribution prior probability is constructed, and the model parameters of the digital twin virtual model are updated using the Bayesian inference algorithm based on the Gaussian distribution prior probability and the preprocessed data to obtain the target digital twin virtual model.

3. The method for regulating and controlling wastewater reuse based on artificial intelligence and digital twins according to claim 2, characterized in that, The process of constructing a Gaussian distribution prior probability based on the model parameters, and updating the model parameters of the digital twin virtual model using the Bayesian inference algorithm based on the Gaussian distribution prior probability and the preprocessed data to obtain the target digital twin virtual model includes: Construct the Gaussian distribution prior probabilities for the model parameters; The actual sensor data is collected periodically from the preprocessed data at the edge nodes, and the residual is calculated based on the actual sensor data and the predicted value output by the digital twin virtual model. The target digital twin virtual model is obtained by updating the model parameters of the digital twin virtual model based on the Gaussian distribution prior probability and the residual using the Bayesian inference algorithm.

4. The method for regulating and controlling wastewater reuse based on artificial intelligence and digital twins according to claim 1, characterized in that, The process involves outputting water quality prediction results for a predetermined time period through a pre-built water quality prediction model at the decision-making level, and performing pre-simulation of control strategies in the target digital twin virtual model based on the water quality prediction results to generate optimized control instructions, including: Obtain historical water quality data for a preset time period to obtain the target historical water quality data; In the decision-making layer, the LSTM layer of the water quality prediction model captures short-term fluctuation information based on the target historical water quality data to output a feature vector; The water quality prediction model uses the Transformer layer to perform dependency analysis based on the feature vector to generate the water quality prediction results for a future preset time period. Based on the water quality prediction results, the control strategy is pre-simulated in the target digital twin virtual model to generate the optimized control command.

5. The method for regulating and controlling wastewater reuse based on artificial intelligence and digital twins according to claim 4, characterized in that, The step of performing a pre-simulation of the control strategy based on the water quality prediction results in the target digital twin virtual model and generating the optimized control instructions includes: An optimization objective function is constructed, and a virtual controller is built using the PPO reinforcement learning algorithm; The virtual controller performs a pre-simulation of the control strategy in the target digital twin virtual model based on the optimization objective function and the water quality prediction results, and generates the optimized control instructions.

6. The method for regulating and controlling wastewater reuse based on artificial intelligence and digital twins according to claim 1, characterized in that, The process involves using a pre-built anomaly detection model and the isolated forest algorithm to perform data anomaly detection and fault diagnosis based on the preprocessed data, generating a fault diagnosis report, including: The target reconstruction error is obtained by performing feature encoding and reconstruction error calculation on the preprocessed data using the anomaly detection model. If the target reconstruction error exceeds a preset threshold, an early warning is triggered and the target is marked as a potential anomaly. The isolated forest algorithm is used to detect abnormal behavior in the device status data of the preprocessed data to obtain the target abnormal behavior; The potential anomalies and the target abnormal behaviors are classified and diagnosed using a decision tree model based on an expert knowledge base and historical fault records, and a fault diagnosis report is generated.

7. The method for regulating and controlling wastewater reuse based on artificial intelligence and digital twins according to any one of claims 1 to 6, characterized in that, The process involves real-time acquisition of initial data from the water treatment system through the sensing layer, followed by data preprocessing at the edge computing layer to generate preprocessed data, including: The initial data of the water treatment system are collected in real time through the multi-source sensor network and equipment status monitor of the sensing layer; The initial data is transmitted to the edge nodes in the edge computing layer via an industrial bus. The initial data is cleaned and features are extracted at the edge nodes to generate the preprocessed data.

8. A wastewater reuse control system based on artificial intelligence and digital twins, characterized in that, include: The data acquisition module is used to collect initial data from the water treatment system in real time through the sensing layer, and to preprocess the initial data in the edge computing layer to generate preprocessed data. The initial data includes water quality parameters, equipment status data and process data. The digital twin model construction module is used to construct a digital twin virtual model in the digital twin modeling layer, and update the model parameters of the digital twin virtual model based on the preprocessed data using a Bayesian inference algorithm to obtain the target digital twin virtual model. The optimized control command generation module is used to output water quality prediction results for a future preset time period through a pre-built water quality prediction model in the decision-making layer, and to perform a pre-drill control strategy in the target digital twin virtual model based on the water quality prediction results to generate optimized control commands. The fault diagnosis module is used to perform data anomaly detection and fault diagnosis processing based on the preprocessed data using a pre-built anomaly detection model and the isolated forest algorithm, and to generate a fault diagnosis report. The optimized control command distribution module is used to distribute the optimized control commands to the actuators via the edge computing layer.

9. The wastewater reuse control system based on artificial intelligence and digital twins according to claim 8, characterized in that, The digital twin model construction module includes: The model building unit is used to build the digital twin virtual model based on CFD fluid dynamics simulation tools in the digital twin modeling layer. The model parameters of the digital twin virtual model include the reaction tank volume, hydraulic retention time, influent concentration fluctuation range, and sludge return ratio. The model parameter update unit is used to construct a Gaussian distribution prior probability based on the model parameters, and update the model parameters of the digital twin virtual model based on the Gaussian distribution prior probability and the preprocessed data through the Bayesian inference algorithm to obtain the target digital twin virtual model.

10. The wastewater reuse control system based on artificial intelligence and digital twins according to claim 9, characterized in that, The model parameter update unit includes: A Gaussian distribution prior probability construction unit is used to construct the Gaussian distribution prior probability for the model parameters; The residual calculation unit is used to collect actual sensor data from the preprocessed data periodically from the edge nodes, and calculate the residual based on the actual sensor data and the predicted value output by the digital twin virtual model. The target digital twin virtual model generation unit is used to update the model parameters of the digital twin virtual model based on the Gaussian distribution prior probability and the residual using the Bayesian inference algorithm to obtain the target digital twin virtual model.

Images