Sewage treatment decision support system based on intelligent control

Abstract

The present invention discloses a sewage treatment decision support system based on intelligent control, which specifically relates to the field of intelligent control technology, including an information collection and preprocessing module, a model building and training module, an integrated review module, and a decision support module. The information collection and preprocessing module is used to monitor information parameters in the sewage treatment process in real time through sensors, and to process the information parameters. The model building and training module is used to build a physical model and a data-driven model, integrate the physical model with the data-driven model to build a hybrid model, and verify the applicability of the hybrid model. The decision support module is used to output decision support based on the prediction results of the hybrid model, which effectively improves the accuracy and reliability of the decision output, and avoids the problem that a simple physical model is difficult to fully describe the actual production status in actual applications, and the quantity and quality of the collected data are difficult to meet the requirements of accurate prediction.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and more specifically, to a sewage treatment decision support system based on intelligent control. Background Art

[0002] The sewage treatment decision support system based on intelligent control is a process that combines control technology with decision support systems to optimize sewage treatment. It uses artificial intelligence, machine learning, sensor technology and data analysis to monitor, analyze and control the operation of sewage treatment facilities in real time. The sewage treatment process has nonlinear and time-varying characteristics. The system behavior during the treatment process is complex and changeable, and it is difficult to describe and predict it through simple linear models or fixed parameter models. The relationship between the output and input of the system is not a simple linear relationship. Changes in input variables will not cause proportional changes in output variables. On the contrary, changes in input may trigger complex reactions in the system output, and may even lead to sudden and unexpected results under certain conditions. At the same time, the parameters and behaviors of the system will change over time. Even under the same input conditions, the system may exhibit different output behaviors at different times.

[0003] The challenges in describing the system essentially come from the mismatch between the physical model and the data-driven model, that is, the misalignment between reality and simulation.

[0004] In order to solve the above defects, a technical solution is now proposed. Summary of the invention

[0005] The purpose of the present invention is to provide a sewage treatment decision support system based on intelligent control to solve the deficiencies in the background technology.

[0006] In order to achieve the above object, the present invention provides the following technical solutions: A sewage treatment decision support system based on intelligent control, comprising an information collection and preprocessing module, a model building and training module, an integrated review module, and a decision support module;

[0007] The information collection and preprocessing module is used to monitor the information parameters in the sewage treatment process in real time through sensors, store the collected information parameters, process the information parameters through data cleaning and preprocessing methods, and mark and extract features from the processed information parameters;

[0008] The model building and training module is used to build a physical model based on sewage treatment and a data-driven model based on historical data. After training the data-driven model, the physical model and the data-driven model are integrated to build a hybrid model.

[0009] The applicability of the hybrid model is verified by establishing a verification model, and after passing the verification, the output of the hybrid model is transmitted to the decision support module;

[0010] The decision support module is used to output decision support based on the prediction results of the hybrid model.

[0011] Preferably, the information parameters include physical parameters, chemical parameters, and biological parameters. Physical parameters include flow rate, temperature, turbidity, and suspended solids concentration. Chemical parameters include pH value, dissolved oxygen, redox potential, chemical oxygen demand, biological oxygen demand, ammonia nitrogen, and phosphate. Biological parameters include microbial activity and biomass concentration.

[0012] Preferably, the method of data preprocessing for information parameters includes abnormal data detection and processing, data denoising, missing data filling, data labeling and feature extraction, and data normalization;

[0013] Detect outliers in data through statistical methods or machine learning algorithms, and delete or fill in outliers;

[0014] Remove high-frequency noise from the data through low-pass filters, sliding average filters or wavelet transform methods, and deduplicate repeated data records;

[0015] For missing data, linear interpolation or polynomial interpolation is used to fill in the missing data for information parameters with small missing amounts and stable changes, and machine learning model prediction is used to fill in the missing data for information parameters with large missing amounts and complex changes;

[0016] In data annotation, the time points of key events in the sewage treatment process are manually marked based on historical records and experimental data, and the historical data are analyzed using automated tools or algorithms to mark abnormal nodes;

[0017] Extract features from the labeled data. The extracted features include time domain, frequency domain, and nonlinear aspects. Time domain features include mean, standard deviation, peak, skewness, and kurtosis, which are used to describe the statistical characteristics of the data. The time series data is converted to the frequency domain through the Fourier transform method. The extracted frequency domain features include spectrum energy and main frequency. The nonlinear features are entropy, fractal dimension, and nonlinear index features, which are used to capture the nonlinear behavior of complex systems.

[0018] Data normalization is used to scale data of different dimensions.

[0019] Preferably, the method for normalizing the data is , where To normalize the data, is the normalized data, is the median value of all data, is a very small constant and Is a positive number.

[0020] Preferably, the method for performing physical modeling is:

[0021] Analyze the sewage treatment process and define the sub-processes to be modeled. The sub-processes include physical process, chemical process and biological process. Physical process includes sedimentation, filtration and mixing. Chemical process includes neutralization, redox reaction and precipitation reaction. Biological process includes microbial metabolism, degradation of organic matter, nitrification and denitrification. Decompose each sub-process into operations and analyze the variables in the operation. The variables include flow rate, concentration and reaction rate.

[0022] Modeling is performed based on the characteristics of the sub-process, including the mass conservation model , Energy Conservation Model , dynamic model , Fluid mechanics model , mass transfer model , using nonlinear regression and least squares method, the parameters in the physical model are calculated through historical data or experimental measurement data, including reaction rate constant, mass transfer coefficient, and heat transfer coefficient;

[0023] The model is calibrated using experimental data or actual operation data, and the physical model output is matched with the actual measured values ​​by adjusting the parameters. After matching, the physical model is output.

[0024] Preferably, the method for establishing a data-driven model is:

[0025] Collect historical operation data, including input variables and output variables, and perform data cleaning and standardization;

[0026] Establish the network architecture, including the number of nodes in the input layer, hidden layer, and output layer, and use nonlinear activation functions in each layer to capture nonlinear relationships;

[0027] Train the neural network model and use the back-propagation algorithm to optimize the weights and biases of the neural network model, and use the gradient descent method to minimize the loss function;

[0028] The performance of the neural network model is verified, and the network architecture and training parameters are adjusted to prevent overfitting or underfitting.

[0029] Preferably, the method of integrating the physical model with the data-driven model to construct a hybrid model is:

[0030] The output of the physical model and the data-driven model are weighted and mixed using the Bayesian update method to define the output of the physical model. and data-driven models The output of is expressed as a probability distribution: and ,in, and are the predicted means of the physical model and the data-driven model, respectively. and are the prediction variances of the physical model and the data-driven model, respectively;

[0031] Set the initial weights of the physical model and the data-driven model based on the performance of historical data and , based on actual data Calculate the likelihood function of the physical model and the data-driven model. The output probability of the model under certain actual data is and ;

[0032] Use Bayes’ theorem to update the weights of the physical model and data-driven model according to the likelihood function. The weights are updated as and , where and are the updated weights of the physical model and the data-driven model, is a normalization constant used to ensure that the updated weight sums to 1;

[0033] Apply the updated weights to the model output to get the mixed output, which is calculated as .

[0034] Preferably, the method for verifying the applicability of the hybrid model by establishing a verification model is:

[0035] The number of adjustable parameters in the calibration physical model is By using the orthogonal test method to analyze the sensitivity of the physical model, the sensitivity transition threshold range of the adjustable parameters is limited to , where is the value of the adjustable parameter, For floating values, when the floating range of the adjustable parameter value in the sensitivity analysis exceeds If the range is too large, the adjustable parameters are marked as sensitive parameters and the number of sensitive parameters is calculated. , the number of hidden layer nodes in the labeled data-driven model is , establish a verification model, the expression is , where is the consistency deviation coefficient, is the time step of the physical model, is the time step of the data-driven model;

[0036] The preset consistency deviation threshold is , when the calculated consistency deviation coefficient Greater than or equal to the consistency deviation threshold is , then the hybrid model operation is marked as having synchronization misalignment;

[0037] When the calculated consistency deviation coefficient Less than the consistency deviation threshold , then the label mixed model output is normal.

[0038] Preferably, the logic of outputting decision support based on the prediction results of the hybrid model is:

[0039] When the hybrid model is marked as out of sync, the physical model and the data-driven model are re-weighted. If the initial weight of the physical model Greater than or equal to the initial weight of the data-driven model , it indicates that the physical model parameters are abnormal, and the output verifies the sensor information decision;

[0040] If the initial weight of the physical model Smaller than the initial weight of the data-driven model , it indicates that the data-driven model is unbalanced and the output optimizes the decision of the neural network model;

[0041] When the hybrid model is marked as having normal output, a threshold is set according to the output parameter of the hybrid model. When the hybrid model prediction parameter exceeds the threshold, an early warning is triggered to notify the operator to intervene.

[0042] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0043] This application establishes a physical model and a data-driven model separately, and uses a parallel hybrid method to establish a hybrid model, while obtaining the interpretability and applicability of the physical model, and improving the ability to capture nonlinear relationships through a data-driven model. This avoids the problem that a simple physical model is difficult to fully describe the actual production status in practical applications, and the quantity and quality of the collected data are difficult to meet the requirements of accurate prediction. This effectively improves the efficiency of utilizing actual production data, provides decision support based on the output results of the hybrid model, and reduces the requirements for human resources in intelligent control of sewage treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0045] Figure 1 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] Example 1: Please refer to Figure 1 As shown, the present invention is a sewage treatment decision support system based on intelligent control, including an information collection and preprocessing module, a model building and training module, an integrated review module, and a decision support module;

[0048] The information collection and preprocessing module is used to monitor the information parameters in the sewage treatment process in real time through sensors, store the collected information parameters, process the information parameters through data cleaning and preprocessing methods, and mark and extract features from the processed information parameters;

[0049] The model building and training module is used to build a physical model based on sewage treatment and a data-driven model based on historical data. After training the data-driven model, the physical model and the data-driven model are integrated to build a hybrid model.

[0050] The applicability of the hybrid model is verified by establishing a verification model, and after passing the verification, the output of the hybrid model is transmitted to the decision support module;

[0051] The decision support module is used to output decision support based on the prediction results of the hybrid model.

[0052] The nonlinear and time-varying characteristics of the sewage treatment process refer to the complex and changeable system behavior during the treatment process, which is not easy to be described and predicted by simple linear models or models with fixed parameters. The nonlinear characteristics mean that the relationship between the output and input of the system is not a simple linear relationship. Changes in input variables will not cause proportional changes in output variables. Changes in input may trigger complex reactions in the system output, and may even lead to sudden and unexpected results under certain conditions;

[0053] For example, in a biological treatment unit, the relationship between microbial growth and pollutant concentration is nonlinear. Low concentrations of pollutants may lead to insufficient microbial activity, while excessively high concentrations of pollutants may inhibit microbial growth.

[0054] The relationship between chemical reaction rate and factors such as temperature and pH value is nonlinear, and different factors will interact with each other during the sewage treatment process, resulting in complex dynamic behavior;

[0055] Time-varying characteristics mean that the parameters and behaviors of a system change over time. Even under the same input conditions, the system may exhibit different output behaviors at different times;

[0056] For example, the water quality characteristics of sewage may change at different times of the day, and the changes will cause the dynamic characteristics of the system to change continuously. The performance of treatment equipment will change with the increase of usage time. For example, the efficiency of aeration equipment may decrease due to aging or improper maintenance of the equipment.

[0057] Information collection and pretreatment are the foundation of the entire system, directly affecting the accuracy of the model and the effectiveness of the control strategy. The information collection and pretreatment module monitors the information parameters in the sewage treatment process in real time through sensors. The information parameters include physical parameters, chemical parameters, and biological parameters. Physical parameters include flow, temperature, turbidity, and suspended solids concentration. Chemical parameters include pH value, dissolved oxygen, redox potential, chemical oxygen demand, biological oxygen demand, ammonia nitrogen, and phosphate. Biological parameters include microbial activity and biomass concentration.

[0058] According to the design of the sewage treatment process, sensors are installed at key process nodes, including the water inlet, sedimentation tank, biological reaction tank, and water outlet to ensure that the information parameters fully reflect the dynamic changes of the entire treatment process;

[0059] Real-time collection of information parameters is achieved through the sensor network, and the information parameters are transmitted to the central control system. Wired or wireless communication technology is used for information transmission, and an information storage system is established in the cloud to store historical data.

[0060] Wired transmission methods such as industrial Ethernet, wireless communication technologies such as LoRa, NB-IoT, etc.

[0061] The stored information parameters are used for real-time control and provide information support for subsequent model training and analysis;

[0062] It should be noted that the choice of transmission method depends on the scale, layout and environmental conditions of the sewage treatment plant, and is set by professional and technical personnel in this field.

[0063] Methods for data preprocessing of information parameters include abnormal data detection and processing, data denoising, missing data filling, data labeling and feature extraction, and data normalization;

[0064] Detect outliers in data through statistical methods or machine learning algorithms, and delete or fill in outliers;

[0065] Remove high-frequency noise from the data through low-pass filters, sliding average filters or wavelet transform methods, and deduplicate repeated data records;

[0066] For missing data, linear interpolation or polynomial interpolation is used to fill in the missing data for information parameters with small missing amounts and stable changes, and machine learning model prediction is used to fill in the missing data for information parameters with large missing amounts and complex changes;

[0067] In data annotation, the time points of key events in the sewage treatment process are manually marked based on historical records and experimental data, and the historical data are analyzed using automated tools or algorithms to mark abnormal nodes;

[0068] Extract features from the labeled data. The extracted features include time domain, frequency domain, and nonlinear aspects. Time domain features include mean, standard deviation, peak, skewness, and kurtosis, which are used to describe the statistical characteristics of the data. The time series data is converted to the frequency domain through the Fourier transform method. The extracted frequency domain features include spectrum energy and main frequency. The nonlinear features are entropy, fractal dimension, and nonlinear index features, which are used to capture the nonlinear behavior of complex systems.

[0069] Data normalization is used to scale data of different dimensions.

[0070] Commonly used statistical methods include Z-score and box plot analysis, etc., machine learning algorithms include isolation forest algorithm, LOF algorithm, etc., commonly used filling methods include interpolation method filling, replacement with the nearest normal value, etc., commonly used machine learning models include KNN model, regression model, etc., key events in the sewage treatment process include abnormal discharge, process failure, shutdown and maintenance, etc.

[0071] The method to normalize the data is , where To normalize the data, is the normalized data, is the median value of all data, is a very small constant and Is a positive number.

[0072] Historical data is preserved for a long time through a data archiving mechanism, and metadata management methods are used to record and manage the data's source, processing methods, and version information.

[0073] Nonlinearity makes it difficult for simple linear models to capture the complex relationships in the system. The presence of nonlinearity requires models to have higher flexibility and complexity, that is, complex mathematical tools and deep computing resources;

[0074] The time-varying characteristics make it impossible for model parameters to be fixed and must be adjusted over time. This requires the model to be able to adapt to system changes in real time, or to introduce time-varying parameters into the model, which increases the complexity of modeling;

[0075] In order to capture nonlinear and time-varying characteristics, a large amount of high-quality data is required to train and validate the model. If the data is insufficient or the data quality is not high, the accuracy of the model will be affected;

[0076] There is a high degree of coupling between the various sub-processes in the sewage treatment process. This coupling relationship is also nonlinear and changes over time. That is, the model not only considers the characteristics of a single process, but also the interaction between sub-processes;

[0077] The various sub-processes in sewage treatment include physical treatment, chemical treatment, and biological treatment.

[0078] The model building and training module is used to build physical models based on sewage treatment and data-driven models based on historical data;

[0079] A physical model is a model built based on the physical, chemical and biological principles of a system. It uses known scientific laws, equations and theories to describe the various processes in the system. For example, biodegradation in sewage treatment can be described by microbial kinetic equations, and chemical reactions can be described by reaction kinetic models. Physical models are based on known scientific principles and can explain the reasons behind system behavior. Because physical models are based on generally applicable physical and chemical principles, they can be extrapolated and expanded under different conditions.

[0080] Data-driven models are models built based on historical data and machine learning and deep learning algorithms. By analyzing and learning a large amount of historical data, patterns and relationships between inputs and outputs are discovered, thereby predicting the behavior of the system. Data-driven models can capture complex nonlinear relationships and adapt to dynamically changing system characteristics. Data-driven models rely on data rather than deep physical or chemical knowledge, and are suitable for situations where system behavior is difficult to accurately describe through physical models.

[0081] The method for physical modeling is:

[0082] Analyze the sewage treatment process and define the sub-processes to be modeled. The sub-processes include physical process, chemical process and biological process. Physical process includes sedimentation, filtration and mixing. Chemical process includes neutralization, redox reaction and precipitation reaction. Biological process includes microbial metabolism, degradation of organic matter, nitrification and denitrification. Decompose each sub-process into operations and analyze the variables in the operation. The variables include flow rate, concentration and reaction rate.

[0083] Modeling is performed based on the characteristics of the sub-process, including the mass conservation model , Energy Conservation Model , dynamic model , Fluid mechanics model , mass transfer model , using nonlinear regression and least squares method, the parameters in the physical model are calculated through historical data or experimental measurement data, including reaction rate constant, mass transfer coefficient, and heat transfer coefficient;

[0084] The model is calibrated using experimental data or actual operation data, and the physical model output is matched with the actual measured values ​​by adjusting the parameters. After matching, the physical model is output.

[0085] The method of establishing a data-driven model is: collecting historical operation data, including input variables and output variables, and performing data cleaning and standardization;

[0086] Establish the network architecture, including the number of nodes in the input layer, hidden layer, and output layer, and use nonlinear activation functions in each layer to capture nonlinear relationships;

[0087] Train the neural network model and use the back-propagation algorithm to optimize the weights and biases of the neural network model, and use the gradient descent method to minimize the loss function;

[0088] The performance of the neural network model is verified, and the network architecture and training parameters are adjusted to prevent overfitting or underfitting.

[0089] Input variables include influent quality, flow rate, pH, temperature, etc. Output variables include effluent quality, pollutant removal rate, etc. The number of nodes in the hidden layer is adjusted by professionals in this field according to the complexity of the data. Commonly used nonlinear activation functions include ReLU, sigmoid and other functions. Training parameters include learning rate and regularization parameter.

[0090] In the complex system describing the sewage treatment process, it is difficult to meet the accuracy and adaptability requirements by using only physical models or data-driven models. Hybrid modeling can effectively improve the description and prediction capabilities of the system by combining the advantages of physical models and data-driven models:

[0091] The physical model provides the basic framework of the system, ensuring that the model is robust and credible under known conditions. The data-driven model can supplement the deficiencies of the physical model. In the nonlinear behavior or dynamic changes that are difficult to accurately describe with the physical model, the hybrid model effectively improves accuracy and reliability.

[0092] Through physical models, the reasons behind system behavior can be explained, thereby enhancing the credibility of the model. Data-driven models can make supplementary predictions for certain complex or unknown system behaviors, making the overall performance of the model more comprehensive.

[0093] Physical models are usually based on the general laws of the system and are applicable to a wide range of operating conditions. Data-driven models continuously learn and adjust through real-time data, and can dynamically adapt to new operating conditions, making up for the limitations of physical models under certain specific conditions;

[0094] Pure data-driven models usually require a large amount of high-quality data to train, while physical models can reduce the data requirements. Hybrid modeling can still provide effective predictions when data is insufficient.

[0095] After training and validating the data-driven model, the data-driven model is output, and the physical model is integrated with the data-driven model to construct a hybrid model;

[0096] The method of integrating physical models with data-driven models to build hybrid models is:

[0097] The output of the physical model and the data-driven model are weighted and mixed using the Bayesian update method to define the output of the physical model. and data-driven models The output of is expressed as a probability distribution: and ,in, and are the predicted means of the physical model and the data-driven model, respectively. and are the prediction variances of the physical model and the data-driven model, respectively;

[0098] Set the initial weights of the physical model and the data-driven model based on the performance of historical data and , based on actual data Calculate the likelihood function of the physical model and the data-driven model. The output probability of the model under certain actual data is and ;

[0099] Use Bayes’ theorem to update the weights of the physical model and data-driven model according to the likelihood function. The weights are updated as and , where and are the updated weights of the physical model and the data-driven model, is a normalization constant used to ensure that the updated weight sums to 1;

[0100] Apply the updated weights to the model output to get the mixed output, which is calculated as .

[0101] The applicability of the hybrid model is verified by establishing a verification model, and after passing the verification, the output of the hybrid model is transmitted to the decision support module;

[0102] The setting of initial weights has a great influence on the final results. The setting of initial weights is determined according to historical data, the maturity of the physical model and the training of the data-driven model. The estimation of model prediction uncertainty is the key basis for the change of model accuracy in the hybrid process. The uncertainty is obtained through historical error analysis, cross-validation, or model-based prediction variance estimation. The Bayesian method is adaptive and dynamically adjusts weights under different operating conditions to ensure that when the model prediction performance is unstable, the system can automatically rely on a more reliable model, and avoid using fixed weight design to affect the output results of the hybrid model when the model deviates.

[0103] The number of adjustable parameters in the calibration physical model is By using the orthogonal test method to analyze the sensitivity of the physical model, the sensitivity transition threshold range of the adjustable parameters is limited to , where is the value of the adjustable parameter, For floating values, when the floating range of the adjustable parameter value in the sensitivity analysis exceeds If the range is too large, the adjustable parameters are marked as sensitive parameters and the number of sensitive parameters is calculated. , the number of hidden layer nodes in the labeled data-driven model is , establish a verification model, the expression is , where is the consistency deviation coefficient, is the time step of the physical model, is the time step of the data-driven model;

[0104] The preset consistency deviation threshold is , when the calculated consistency deviation coefficient Greater than or equal to the consistency deviation threshold is , then the hybrid model operation is marked as having synchronization misalignment;

[0105] When the calculated consistency deviation coefficient Less than the consistency deviation threshold , then the label mixed model output is normal.

[0106] The decision support module is used to output decision support based on the prediction results of the hybrid model.

[0107] When the hybrid model is marked as out of sync, the physical model and the data-driven model are re-weighted. If the initial weight of the physical model Greater than or equal to the initial weight of the data-driven model , it indicates that the physical model parameters are abnormal, and the output verifies the sensor information decision;

[0108] If the initial weight of the physical model Smaller than the initial weight of the data-driven model , it indicates that the data-driven model is unbalanced and the output optimizes the decision of the neural network model;

[0109] When the hybrid model is marked as having normal output, a threshold is set according to the output parameter of the hybrid model. When the hybrid model prediction parameter exceeds the threshold, an early warning is triggered to notify the operator to intervene.

[0110] This application establishes a physical model and a data-driven model separately, and uses a parallel hybrid method to establish a hybrid model, while obtaining the interpretability and applicability of the physical model, and improving the ability to capture nonlinear relationships through a data-driven model. This avoids the problem that a simple physical model is difficult to fully describe the actual production status in practical applications, and the quantity and quality of the collected data are difficult to meet the requirements of accurate prediction. This effectively improves the efficiency of utilizing actual production data, provides decision support based on the output results of the hybrid model, and reduces the requirements for human resources in intelligent control of sewage treatment.

[0111] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0112] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of computer program goods. The computer program goods include one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0113] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0114] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0115] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0116] If the functions are implemented in the form of software functional units and sold or used as independent goods, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can essentially or in other words, the part that contributes to the prior art or the part of the technical solution can be embodied in the form of software goods, and the computer software goods are stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0117] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A sewage treatment decision support system based on intelligent control, characterized in that: It includes information collection and preprocessing module, model building and training module, integrated review module, and decision support module; The information collection and preprocessing module is used to monitor the information parameters in the sewage treatment process in real time through sensors, store the collected information parameters, process the information parameters through data cleaning and preprocessing methods, and mark and extract features from the processed information parameters; The model building and training module is used to build a physical model based on sewage treatment and a data-driven model based on historical data. After training the data-driven model, the physical model and the data-driven model are integrated to build a hybrid model. The applicability of the hybrid model is verified by establishing a verification model, and after passing the verification, the output of the hybrid model is transmitted to the decision support module; The decision support module is used to output decision support based on the prediction results of the hybrid model; The method to build a data-driven model is: Collect historical operation data, including input variables and output variables, and perform data cleaning and standardization; Establish the network architecture, including the number of nodes in the input layer, hidden layer, and output layer, and use nonlinear activation functions in each layer to capture nonlinear relationships; Train the neural network model and use the back-propagation algorithm to optimize the weights and biases of the neural network model, and use the gradient descent method to minimize the loss function; Verify the performance of the neural network model and adjust the network architecture and training parameters to prevent overfitting or underfitting; The method of integrating physical models with data-driven models to build hybrid models is: The Bayesian update method is used to perform a weighted mix of the outputs of the physical model and the data-driven model, and the output y of the physical model is defined p and data-driven model y d The output of is expressed as a probability distribution: and Among them, μ p and μ d are the predicted means of the physical model and the data-driven model, respectively. and are the prediction variances of the physical model and the data-driven model, respectively; Set the initial weights of the physical model and the data-driven model based on the performance of historical data p and ω d , based on the actual data y obs Calculate the likelihood function of the physical model and the data-driven model. The output probability of the model under certain actual data is and Use Bayes’ theorem to update the weights of the physical model and data-driven model according to the likelihood function. The weights are updated as and In the formula, ω up and ω ud are the updated weights of the physical model and the data-driven model, respectively, P(y obs ) is a normalization constant used to ensure that the updated weight sum is 1; Apply the updated weights to the model output to get the mixed output, calculated as y mix =ω up ·y p +ω ud ·y d ; The method for verifying the applicability of the hybrid model by establishing a verification model is as follows: The number of adjustable parameters in the calibration physical model is A p By using the orthogonal test method to conduct sensitivity analysis on the physical model, the sensitivity transition threshold range of the adjustable parameters is limited to [V+V fl ,VV fl ], where V is the value of the adjustable parameter, V fl For floating values, when the floating range of the adjustable parameter value in the sensitivity analysis exceeds [V+V fl ,VV fl ], the adjustable parameters are marked as sensitive parameters, and the number of sensitive parameters S is calculated p , the number of hidden layer nodes in the labeled data-driven model is HL n , establish a verification model, the expression is Where Va l is the consistency deviation coefficient, TS p is the time step of the physical model, TS d is the time step of the data-driven model; The preset consistency deviation threshold is T va , when the calculated consistency deviation coefficient Va l Greater than or equal to the consistency deviation threshold is T va , then the hybrid model operation is marked as having synchronization misalignment; When the calculated consistency deviation coefficient Va l Less than the consistency deviation threshold T va , then the label mixed model output is normal.

2. The sewage treatment decision support system based on intelligent control according to claim 1 is characterized in that: Information parameters include physical parameters, chemical parameters, and biological parameters. Physical parameters include flow, temperature, turbidity, and suspended solids concentration. Chemical parameters include pH value, dissolved oxygen, redox potential, chemical oxygen demand, biological oxygen demand, ammonia nitrogen, and phosphate. Biological parameters include microbial activity and biomass concentration.

3. The sewage treatment decision support system based on intelligent control according to claim 2 is characterized in that: Methods for data preprocessing of information parameters include abnormal data detection and processing, data denoising, missing data filling, data labeling and feature extraction, and data normalization; Detect outliers in data through statistical methods or machine learning algorithms, and delete or fill in outliers; Remove high-frequency noise from the data through low-pass filters, sliding average filters or wavelet transform methods, and deduplicate repeated data records; For missing data, linear interpolation or polynomial interpolation is used to fill in the missing data for information parameters with small missing amounts and stable changes, and machine learning model prediction is used to fill in the missing data for information parameters with large missing amounts and complex changes; In data annotation, the time points of key events in the sewage treatment process are manually marked based on historical records and experimental data, and the historical data are analyzed using automated tools or algorithms to mark abnormal nodes; Extract features from the labeled data. The extracted features include time domain, frequency domain, and nonlinear aspects. Time domain features include mean, standard deviation, peak, skewness, and kurtosis, which are used to describe the statistical characteristics of the data. The time series data is converted to the frequency domain through the Fourier transform method. The extracted frequency domain features include spectrum energy and main frequency. The nonlinear features are entropy, fractal dimension, and nonlinear index features, which are used to capture the nonlinear behavior of complex systems. Data normalization is used to scale data of different dimensions.

4. The sewage treatment decision support system based on intelligent control according to claim 1 is characterized in that: The method for physical modeling is: Analyze the sewage treatment process and define the sub-processes to be modeled. The sub-processes include physical process, chemical process and biological process. Physical process includes sedimentation, filtration and mixing. Chemical process includes neutralization, redox reaction and precipitation reaction. Biological process includes microbial metabolism, degradation of organic matter, nitrification and denitrification. Decompose each sub-process into operations and analyze the variables in the operation. The variables include flow rate, concentration and reaction rate. Modeling is performed based on the characteristics of the sub-process, including the mass conservation model C om 、Energy conservation model C oe , dynamic model D y 、Fluid mechanics model F m , mass transfer model M tm , using nonlinear regression and least squares method, the parameters in the physical model are calculated through historical data or experimental measurement data, including reaction rate constant, mass transfer coefficient, and heat transfer coefficient; The model is calibrated using experimental data or actual operation data, and the physical model output is matched with the actual measured values ​​by adjusting the parameters. After matching, the physical model is output.

5. The sewage treatment decision support system based on intelligent control according to claim 1 is characterized in that: The logic of outputting decision support based on the prediction results of the hybrid model is: When the hybrid model is marked as synchronously misaligned, the physical model and the data-driven model are re-weighted. If the initial weight ω of the physical model p Greater than or equal to the initial weight ω of the data-driven model d , it indicates that the physical model parameters are abnormal, and the output verifies the sensor information decision; If the initial weight ω of the physical model p Less than the initial weight ω of the data-driven model d , it indicates that the data-driven model is unbalanced and the output optimizes the decision of the neural network model; When the hybrid model is marked as having normal output, a threshold is set according to the output parameter of the hybrid model. When the hybrid model prediction parameter exceeds the threshold, an early warning is triggered to notify the operator to intervene.

Images