Intelligent regulation and control system for sewage treatment process

Abstract

The invention discloses an intelligent regulation and control system for a sewage treatment process. The intelligent regulation and control system comprises an Internet of Things data acquisition instrument, a server and an Internet of Things controller, wherein the server deploys a database, a data preprocessing model, a data model, a mechanism model, an evaluation model, a decision model, a data quality optimization model and application platform software; wherein the database is provided with an original data set, a data model regulation and control parameter set, a mechanism model regulation and control parameter set, an evaluation parameter set, a decision regulation and control parameter set, an optimization data set, a data model parameter set and a mechanism model parameter set; the application platform software has all functions required by man-machine interaction. The mechanism model, the data model and the evaluation model which can operate independently are connected in parallel, the advantages are complementary, and the reliability and the application range of the system are enhanced; a decision model is set to coordinate the relationship among the models, errors in the regulation and control direction and force are avoided, smooth regulation and control are realized, and the usability and interpretability of the intelligent regulation and control system are improved at the same time.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, and more specifically, to an intelligent control system for wastewater treatment processes. Background Technology

[0002] The scale of sewage treatment plants in rural towns in my country is generally 500 to 40,000 m³. 3 The water quality and quantity of wastewater in township wastewater treatment plants vary significantly with the seasons and time, with daily variation coefficients typically reaching 3-5. Their energy consumption per unit is significantly higher than that of medium and large-sized plants. Insufficient professional operational capacity is a prominent problem in the operation of township wastewater treatment plants. The inability to adjust operations in a timely and accurate manner in response to changes leads to insufficient operational stability and increased energy costs. In recent years, as wastewater treatment plant operation and control have moved towards a green and low-carbon trend, professional technicians have developed various intelligent control systems for wastewater treatment systems. Some of these systems employ mechanistic models, while others use data-driven models, and there is a growing trend towards coupling these two models. These models often encounter the following problems in practical applications: (1) The phenomenon of gradually deteriorating data quality in sewage treatment systems is common. Existing technologies cannot resolve the contradiction between low data quality and high model accuracy requirements, making it difficult to meet the high-quality computation of the model, and the control accuracy and precision gradually decline; (2) Existing technologies using mechanistic models require a large number of difficult-to-obtain model parameters, the models are complex, calibration is difficult, a large number of water quality instruments and high-quality data are required, simplified mechanistic models have mismatch problems, it is difficult to cope with the complexity and drastic fluctuations of actual processes, it does not have the machine learning function to automatically adjust model parameters, and the computational load of the coupled hydraulic model increases significantly, making it difficult to adapt to the actual process. (3) Data-driven models have machine learning capabilities, but the generalization of existing technologies heavily relies on large spatial data samples, requiring higher data quality and poor interpretability. They cannot quickly achieve intelligent regulation after the system is put into operation. When regulation is frequent or accuracy requirements are high, the computing power requirement increases sharply, making it difficult to quickly respond to changes in impact loads and adapt to township sewage treatment systems. (4) Existing mechanism models and data-driven models are coupled, resulting in complex architectures and a lack of model quality control methods. If the result of one model is missing or distorted, the overall result will be missing or distorted, making it impossible to function independently and difficult to coordinate the differences in the calculation results of two or more models. Therefore, it is necessary to overcome the above drawbacks, improve data quality, combine multiple models, improve the accuracy of the regulation system, strengthen robustness, and develop an intelligent regulation system suitable for township sewage treatment systems.

[0003] The information disclosed in the background section of this invention is intended only to enhance the understanding of the general background of this invention, and should not be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art. Summary of the Invention

[0004] This invention proposes an intelligent control system for wastewater treatment processes, comprising an IoT data acquisition unit, a server, and an IoT controller. Server deployment includes databases, data preprocessing models, data models, mechanistic models, evaluation models, decision-making models, data quality optimization models, and application platform software; The database includes the original dataset, data model control parameter set, mechanism model control parameter set, evaluation parameter set, decision control parameter set, optimization dataset, data model parameter set, and mechanism model parameter set; the application platform software has all the functions required for human-computer interaction. The IoT data acquisition instrument collects variable data from the process system, filters out invalid data, and uploads the collected dataset to the server database at a set frequency. The collected dataset has the function of manually entering data through the application platform, and its structure is a non-fixed time sequence dataset. The data preprocessing model processes the collected dataset data in batches at a frequency of once per hour. Based on the distribution characteristics of variable values, it filters out outlier data and replaces it with empty data. After integerization, it takes the difference, mean, and mode as needed according to the variables. The processing results are stored in and update the original dataset, which has a standard format dataset with a time interval of 1 hour, and accumulates and updates over time. The model operation control is run in a set step. Each model operation control includes the following three steps: (1) The model operation control is triggered whenever the original dataset is updated a certain number of times at a set frequency. The data model, mechanism model and evaluation model are all operated independently. The model input data are all taken from the original dataset. The mechanism model is a mathematical model for dynamic correction of biochemical dynamics and hydrodynamic coupling parameters. It is divided into different process modules. All recommended parameters are numerical range variables. The mechanism model is stored and updated in the mechanism model control parameter set. The data model is a comprehensive data-driven model for dynamic correction of biochemical dynamics and hydrodynamic coupling parameters. All recommended parameters are in the same order and quantity as the mechanism model. They are all numerical variables. The data model control parameter set is stored and updated. The evaluation model is a mathematical model. The processing effect and energy efficiency evaluation parameters are output according to the settings. They are all numerical variables. The evaluation parameter set is stored and updated. If the above three models exceed the operation time limit, their operation will be stopped and the output parameters that cannot be obtained will be stopped. All values ​​are empty; the output parameter values ​​of the data model that has not completed the first training are all empty; (2) Whenever the data model regulation parameter set, mechanism model regulation parameter set, and evaluation parameter set are all updated, the decision model operation is triggered. The decision model is a decision tree fuzzy logic closed-loop control comprehensive model. Regardless of whether the output parameters of the three models in step (1) are all empty, the decision model can complete the operation by calling the original dataset. The data structure and variable types of the output decision parameters are exactly the same as those of the data model. The decision regulation parameter set is stored and updated. If the operation time of the decision model exceeds the limit, the operation will stop. The output parameter values ​​that cannot obtain the result are all empty; (3) Whenever the decision regulation parameter set is updated, regulation is triggered. According to the automatic or manual setting, the decision parameters are downloaded to the IoT controller, parsed into equipment control parameters, and the current regulation operation is completed. The equipment regulation parameter execution data is transmitted to the collection dataset to form a control iteration closed-loop structure. No parsing and regulation operation is performed when the decision regulation parameters are empty; The data quality optimization model uses the step batch processing of the original dataset data with a set value greater than the model operation control to provide high-quality data for model training and optimization by imputation and repair methods. It includes the following three steps: (1) Take out a certain range of data before the target time from the original dataset and imput data with a continuous missing percentage of less than 5%; (2) Repair data with a continuous missing percentage of 5% to 20% by using data within a certain range before the target time; when the continuous missing data percentage reaches 20% to 50%, perform cross-repair by using data within a certain range before and after the target time; when the continuous missing data percentage reaches more than 50%, no repair is performed; (3) Data quality inspection, reduce the amount of imputed data and increase the amount of repaired data, repeat steps (1) to (2) until the set random sampling inspection repair accuracy standard is met, store and update the optimized dataset, whose data structure is the same as the original dataset, and mark the imputed and repaired data; when the amount of imputed data is reduced to 0 or the operation time exceeds the limit, stop the model operation and do not update the optimized dataset. Optimizing dataset updates triggers the training of data models and mechanistic models; if there is a conflict with model computation, training will be started after the computation is completed, the output model parameter results will be stored in the corresponding model parameter set and the model will be updated; if the model training time exceeds the limit, training will be stopped and the model parameters will not be updated.

[0005] Preferably, the IoT data acquisition instrument has the functions of data acquisition, storage, processing, and uploading to the database according to set requirements; the data acquisition range includes online instrument feedback values, namely various instrument statuses, flow rates, water quality, temperature, sludge concentration, dissolved oxygen, and oxidation-reduction potential. Instantaneous values ​​are acquired using a combination of changing thresholds and set frequencies, while cumulative values ​​are acquired using a set frequency acquisition method. All acquired data is time-stamped; data processing is performed through edge computing, combined with instrument status monitoring, instrument range settings, and pump and fan operating characteristic detection, to filter out abnormal data such as faults, exceeding ranges, exceeding operating conditions, and illegal values, and mark them as empty; data is uploaded via the network and can resume writing after power failure and automatically reconnect to the network for transmission.

[0006] Preferably, the data model is a composite data-driven model that includes at least two of the following: fuzzy neural network, support vector machine, genetic algorithm, and particle swarm optimization algorithm. The initial training of the data model begins when the hourly compliance rate of each water effluent indicator exceeds 85% for three consecutive days after system debugging. After accumulating 90 days of data in the optimized dataset, the model parameters are initialized and debugged, and training is repeated until the data model quality meets the set standards, thus completing the initial training; or A one-year standard optimization dataset created using data from similar process systems was used as the virtual operating data from the previous year for training. The model parameters were adjusted and the training was repeated until the data model quality inspection met the set standards, thus completing the first training. After the data model completes its initial training, it enters normal training. Training is started according to the set training time interval. When the set time is reached, the data of the current 60 days and the same month of the previous year and the adjacent 30 days before and after it are taken from the self-optimized dataset, for a total of 150 days of data. Training is carried out until the data model quality check meets the set standard. The quality verification method of the data model is to compare the predicted values ​​of each set effluent water quality variable on a set date in the optimized dataset with the actual values. If the prediction accuracy of each effluent water quality variable is greater than the set standard, the training is complete. The data model's operation method is as follows: after the model operation is triggered, the water quality index compliance value is set with the safety margin as a constraint, and the minimum predicted power consumption and reagent consumption is the goal. The optimization operation is carried out to obtain the recommended control parameters, and the operation is completed.

[0007] Preferably, the mechanistic model is a simplified model of dynamically corrected parameters coupled with biochemical dynamics and hydraulics. The model simplification is achieved through variable clustering, time scale separation, and parameter selection. Initial model parameters are determined by initializing the constants of the wastewater treatment process system facilities and equipment using empirical functions, completing the initial mechanistic model calculation, and dynamically correcting the model parameters through mechanistic model training. The training method starts when the hourly compliance rate of each effluent indicator exceeds 85% for the first three consecutive days after system debugging, and enters normal training after the optimized dataset has accumulated 30 days of data. Training is started according to the set training time interval, and continues until the mechanistic model quality inspection meets the set standard. The mechanistic model parameters are then updated. If the model calculation time exceeds the limit, the calculation ends, and the original model parameters are used again. The quality inspection method of the mechanistic model is to compare the model's prediction of the values ​​of each set effluent water quality variable on a set date in the optimized dataset with the actual values. If the prediction accuracy of each effluent water quality variable is greater than the set standard, the training is complete.

[0008] Preferably, the evaluation model is a simplified mathematical model of ASM with dynamically corrected model parameters. The input is the original dataset, used for evaluating the recent control effect. The evaluation model steps are set, and the appropriateness of the previous control direction and intensity is evaluated by using the evaluation parameters of the compliance status of each effluent indicator, electricity consumption per ton of water, and chemical consumption per ton of water in the previous step. The parameter correction of the evaluation model is synchronized with the training of the mechanism model. Starting from the first three consecutive days after the system debugging, when the hourly compliance rate of each effluent indicator exceeds 85%, the optimized dataset accumulates data for 30 days and then enters normal training. Training is started according to the set training time interval until the quality test of the evaluation model reaches the set standard, and the model parameters are updated, thus completing the model parameter correction. The quality test method of the evaluation model is to predict the values ​​of each set effluent water quality variable on a set date in the optimized dataset by the model and compare them with the actual values. The prediction accuracy of each evaluation variable is greater than the set standard.

[0009] Preferably, the decision model is a comprehensive model combining decision trees, fuzzy logic control, and closed-loop iterative models, storing the knowledge, experience, and rules of domain experts, and using the computational results of the above models to solve operational problems; before the data model is initially trained, the computational results of the mechanistic model and the evaluation model are used to complete the decision model computation; after the data model is initially trained, for control parameters with a control step of less than 1 day, the weight of the mechanistic model computational results is increased, while for control parameters with a control step of not less than 1 day, the computational results of the mechanistic model and the data-driven model are balanced; the decision model provides parameter adjustment suggestions based on the built-in logical judgment and the evaluation model results.

[0010] Preferably, the IoT controller has the functions of collecting power equipment status, operating current and frequency, parsing adjustment parameters into control parameters, setting control modes and control parameters, operating the equipment according to logic programs or manual operation instructions, and data storage and network transmission; the electromechanical equipment status is collected using a combination of threshold changes and timing, and all collected data is time-stamped; the health status of the equipment is correlated with the variables of voltage, current and frequency to realize equipment fault early warning; the control parameters are adjusted in stages according to the set single adjustment upper limit.

[0011] Preferably, the change threshold is 5%, and the timing interval is less than 1 hour.

[0012] Preferably, the data quality optimization model operation process includes the following two steps: Outlier data were removed using Gaussian, Poisson, and binomial distributions. For data with consecutive missing values ​​within 5%, a combination of linear, autoregressive integral moving average, exponential smoothing, multiple linear regression, and vector autoregression methods were used to impute the missing data, and the imputed data were labeled. Using one or more combinations of models such as LSTM, RNN, and gradient boosting regression tree, and utilizing data from the same period of the previous year and recent neighboring data, we impute data with consecutive missing values ​​of 5%-50% and label the imputed data.

[0013] Its beneficial effects are as follows: First, the system framework adds a front-end data quality optimization model to the existing data model and mechanism model in series, parallel and embedded manner. The data model, mechanism model and evaluation model are connected in parallel and operate independently, and there is a crossover. The functions of the models complement each other, effectively coordinating operation to achieve safety and energy saving. Even if the three models have no calculation results, it does not affect the decision model to complete the regulation using the original data, thus enhancing the overall reliability and robustness of the system.

[0014] Secondly, the IoT data acquisition device leverages edge computing capabilities to block illegal data from being uploaded to the database through logical judgment, thus controlling data from contamination at the source.

[0015] Third, the data quality optimization model can use different methods to filter out outliers and remove variables that exhibit different distributions and fluctuations, and impute locally missing data to ensure data continuity. The method of cross-patching continuous missing data using current data and historical data from the same period can provide high-quality data for stable model operation.

[0016] Fourth, the mechanistic model simplification method significantly improves model efficiency, reduces the calculation cycle to the minute level, and ensures controllable accuracy, making it suitable for wastewater treatment systems where complete model parameters cannot be obtained; the method of dynamically correcting model parameters enhances the model's adaptability to environmental changes, provides machine learning capabilities, and significantly improves generalization performance, thus compensating for the shortcomings of data models in the early stages of system operation when there is insufficient data sample space.

[0017] Fifth, the data model operation adopts an optimization method that ensures the achievement of the target effect and reduces energy consumption. The training and model quality verification methods can control the model quality and accuracy, and effectively control the impact of the decline in data quality on the model quality.

[0018] Sixth, the decision-making model can coordinate the relationships between the results of the mechanistic model, the data model, the evaluation model, and the raw data, avoiding logical errors in the direction and intensity of regulation; it adapts to the daily, weekly, and seasonal changes in water quality and quantity, avoiding excessive single-time regulation that could cause abrupt changes, achieving smooth regulation, and ensuring a stable transition of the biological treatment system. Simultaneously, it improves the usability and interpretability of the intelligent regulation system.

[0019] Seventh, the mechanistic model and data model can record and save the parameters after each correction as a data model parameter set and a mechanistic model parameter set, which is convenient for operators to understand and improves the interpretability of the model calculation results.

[0020] The system of the present invention has other features and advantages that will be apparent from or will be set forth in detail in the accompanying drawings and following detailed description, which together serve to explain the particular principles of the invention. Attached Figure Description

[0021] The above and other objects, features and advantages of the present invention will become more apparent from the more detailed description of exemplary embodiments of the invention in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same parts.

[0022] Figure 1 A block diagram of an intelligent control system for a wastewater treatment process according to an embodiment of the present invention is shown. Detailed Implementation

[0023] Preferred embodiments of the invention will now be described in more detail. While preferred embodiments of the invention are described below, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0024] To facilitate understanding of the solutions and effects of the embodiments of the present invention, a specific application example is given below. Those skilled in the art should understand that this example is merely for the purpose of understanding the present invention, and any specific details therein are not intended to limit the present invention in any way.

[0025] Example 1

[0026] A smart control system for wastewater treatment processes includes an IoT data acquisition unit, a server, and an IoT controller. Server deployment includes databases, data preprocessing models, data models, mechanistic models, evaluation models, decision-making models, data quality optimization models, and application platform software; The database includes the original dataset, data model control parameter set, mechanism model control parameter set, evaluation parameter set, decision control parameter set, optimization dataset, data model parameter set, and mechanism model parameter set; the application platform software has all the functions required for human-computer interaction. The IoT data acquisition instrument collects variable data from the process system, filters out invalid data, and uploads the collected dataset to the server database at a set frequency. The collected dataset has the function of manually entering data through the application platform, and its structure is a non-fixed time sequence dataset. The data preprocessing model processes the collected dataset data in batches at a frequency of once per hour. Based on the distribution characteristics of variable values, it filters out outlier data and replaces it with empty data. After integerization, it takes the difference, mean, and mode as needed according to the variables. The processing results are stored in and update the original dataset, which has a standard format dataset with a time interval of 1 hour, and accumulates and updates over time. The model operation control is run in a set step. Each model operation control includes the following three steps: (1) The model operation control is triggered whenever the original dataset is updated a certain number of times at a set frequency. The data model, mechanism model and evaluation model are all operated independently. The model input data are all taken from the original dataset. The mechanism model is a mathematical model for dynamic correction of biochemical dynamics and hydrodynamic coupling parameters. It is divided into different process modules. All recommended parameters are numerical range variables. The mechanism model is stored and updated in the mechanism model control parameter set. The data model is a comprehensive data-driven model for dynamic correction of biochemical dynamics and hydrodynamic coupling parameters. All recommended parameters are in the same order and quantity as the mechanism model. They are all numerical variables. The data model control parameter set is stored and updated. The evaluation model is a mathematical model. The processing effect and energy efficiency evaluation parameters are output according to the settings. They are all numerical variables. The evaluation parameter set is stored and updated. If the above three models exceed the operation time limit, their operation will be stopped and the output parameters that cannot be obtained will be stopped. All values ​​are empty; the output parameter values ​​of the data model that has not completed the first training are all empty; (2) Whenever the data model regulation parameter set, mechanism model regulation parameter set, and evaluation parameter set are all updated, the decision model operation is triggered. The decision model is a decision tree fuzzy logic closed-loop control comprehensive model. Regardless of whether the output parameters of the three models in step (1) are all empty, the decision model can complete the operation by calling the original dataset. The data structure and variable types of the output decision parameters are exactly the same as those of the data model. The decision regulation parameter set is stored and updated. If the operation time of the decision model exceeds the limit, the operation will stop. The output parameter values ​​that cannot obtain the result are all empty; (3) Whenever the decision regulation parameter set is updated, regulation is triggered. According to the automatic or manual setting, the decision parameters are downloaded to the IoT controller, parsed into equipment control parameters, and the current regulation operation is completed. The equipment regulation parameter execution data is transmitted to the collection dataset to form a control iteration closed-loop structure. No parsing and regulation operation is performed when the decision regulation parameters are empty; The data quality optimization model uses the step batch processing of the original dataset data with a set value greater than the model operation control to provide high-quality data for model training and optimization by imputation and repair methods. It includes the following three steps: (1) Take out a certain range of data before the target time from the original dataset and imput data with a continuous missing percentage of less than 5%; (2) Repair data with a continuous missing percentage of 5% to 20% by using data within a certain range before the target time; when the continuous missing data percentage reaches 20% to 50%, perform cross-repair by using data within a certain range before and after the target time; when the continuous missing data percentage reaches more than 50%, no repair is performed; (3) Data quality inspection, reduce the amount of imputed data and increase the amount of repaired data, repeat steps (1) to (2) until the set random sampling inspection repair accuracy standard is met, store and update the optimized dataset, whose data structure is the same as the original dataset, and mark the imputed and repaired data; when the amount of imputed data is reduced to 0 or the operation time exceeds the limit, stop the model operation and do not update the optimized dataset. Optimizing dataset updates triggers the training of data models and mechanistic models; if there is a conflict with model computation, training will be started after the computation is completed, the output model parameter results will be stored in the corresponding model parameter set and the model will be updated; if the model training time exceeds the limit, training will be stopped and the model parameters will not be updated.

[0027] In one example, the IoT data acquisition instrument has the functions of collecting, storing, processing, and uploading data to the database according to the set requirements. The data acquisition range includes online instrument feedback values, namely various instrument statuses, flow rates, water quality, temperature, sludge concentrations, dissolved oxygen, and oxidation-reduction potentials. Instantaneous values ​​are acquired using a combination of changing thresholds and set frequencies, while cumulative values ​​are acquired using a set frequency method. All acquired data is time-stamped. Data processing is performed through edge computing, combined with instrument status monitoring, instrument range settings, and pump and fan operating characteristic detection. Abnormal data such as faults, exceeding ranges, exceeding operating conditions, and illegal values ​​are filtered out and marked as empty. Data is uploaded via the network and can resume writing after power failure and automatically reconnect to the network for transmission.

[0028] In one example, the data model is a composite data-driven model that includes at least two of the following: fuzzy neural network, support vector machine, genetic algorithm, and particle swarm optimization. The initial training of the data model begins when the hourly compliance rate of each water effluent indicator exceeds 85% for three consecutive days after system debugging. After accumulating 90 days of data in the optimized dataset, the model parameters are initialized and debugged, and training is repeated until the data model quality meets the set standards, thus completing the initial training; or A one-year standard optimization dataset created using data from similar process systems was used as the virtual operating data from the previous year for training. The model parameters were adjusted and the training was repeated until the data model quality inspection met the set standards, thus completing the first training. After the data model completes its initial training, it enters normal training. Training is started according to the set training time interval. When the set time is reached, the data of the current 60 days and the same month of the previous year and the adjacent 30 days before and after it are taken from the self-optimized dataset, for a total of 150 days of data. Training is carried out until the data model quality check meets the set standard. The quality verification method of the data model is to compare the predicted values ​​of each set effluent water quality variable on a set date in the optimized dataset with the actual values. If the prediction accuracy of each effluent water quality variable is greater than the set standard, the training is complete. The data model's operation method is as follows: after the model operation is triggered, the water quality index compliance value is set with the safety margin as a constraint, and the minimum predicted power consumption and reagent consumption is the goal. The optimization operation is carried out to obtain the recommended control parameters, and the operation is completed.

[0029] In one example, the mechanistic model is a simplified model of dynamically corrected parameters coupled with biochemical dynamics and hydraulics. The model simplification is achieved through variable clustering, time scale separation, and parameter selection. Initial model parameters are determined by setting constants for the wastewater treatment process system facilities and equipment using empirical functions, completing the initial mechanistic model calculation. Dynamic correction of model parameters is achieved through mechanistic model training. The training method starts when the hourly compliance rate of each effluent indicator exceeds 85% for the first three consecutive days after system debugging. Normal training begins after the optimized dataset has accumulated 30 days of data. Training is started according to the set time interval and continues until the mechanistic model quality check meets the set standard. The mechanistic model parameters are then updated. If the model calculation time exceeds the limit, the calculation ends, and the original model parameters are used again. The quality check method of the mechanistic model is to predict the values ​​of each set effluent water quality variable on a set date in the optimized dataset and compare them with the actual values. If the prediction accuracy of each effluent water quality variable is greater than the set standard, the training is complete.

[0030] In one example, the evaluation model is a simplified ASM mathematical model with dynamically adjusted model parameters. The input is the original dataset, used for evaluating the recent control effect. The evaluation model's steps are set, and the appropriateness of the previous control direction and intensity is evaluated based on the compliance status of each effluent indicator, electricity consumption per ton of water, and chemical consumption per ton of water in the previous step. The evaluation model parameter adjustment is synchronized with the mechanistic model training. Starting from the first three consecutive days after system debugging, when the hourly compliance rate of each effluent indicator exceeds 85%, normal training begins after the optimized dataset has accumulated 30 days of data. Training is started according to the set training time interval, and continues until the evaluation model quality test meets the set standard. The model parameters are then updated, thus completing the model parameter adjustment. The evaluation model quality test method is to compare the predicted values ​​of each set effluent water quality variable in the optimized dataset on a set date with the actual values. The prediction accuracy of each evaluation variable is greater than the set standard.

[0031] In one example, the decision model is a comprehensive model combining decision trees, fuzzy logic control, and closed-loop iterative models, incorporating the knowledge, experience, and rules of domain experts. The results of these model calculations are used to solve operational problems. Before the data model's initial training is complete, the results of the mechanistic model and evaluation model are used to complete the decision model's calculations. After the data model's initial training is complete, for control parameters with a control step of less than one day, the weight of the mechanistic model's calculation results is increased; for control parameters with a control step of not less than one day, the results of the mechanistic model and the data-driven model are balanced. The decision model provides parameter adjustment suggestions based on its built-in logical judgments and the evaluation model's results.

[0032] In one example, the IoT controller has the functions of collecting power equipment status, operating current and frequency, parsing adjustment parameters into control parameters, setting control methods and control parameters, operating the equipment according to logical programs or manual operation instructions, and data storage and network transmission; the electromechanical equipment status is collected using a combination of changing thresholds and timing, and all collected data is time-stamped; the health status of the equipment is correlated with the variables of voltage, current and frequency to realize equipment fault early warning; the control parameters are adjusted according to the set single adjustment limit, and adjustments are made in stages when the limit is exceeded.

[0033] In one example, the change threshold is 5%, and the timing interval is less than 1 hour.

[0034] In one example, the data quality optimization model operation process includes the following two steps: Outlier data were removed using Gaussian, Poisson, and binomial distributions. For data with consecutive missing values ​​within 5%, a combination of linear, autoregressive integral moving average, exponential smoothing, multiple linear regression, and vector autoregression methods were used to impute the missing data, and the imputed data were labeled. Using one or more combinations of models such as LSTM, RNN, and gradient boosting regression tree, and utilizing data from the same period of the previous year and recent neighboring data, we impute data with consecutive missing values ​​of 5%-50% and label the imputed data.

[0035] Figure 1 A block diagram of an intelligent control system for a wastewater treatment process according to an embodiment of the present invention is shown.

[0036] Specifically, such as Figure 1 As shown, the control system structure includes an IoT data acquisition instrument, a server, and an IoT controller; the server deploys a database, a data preprocessing model, a data model, a mechanistic model, an evaluation model, a decision-making model, a data quality optimization model, and application platform software; the database sets the original dataset, the data model control parameter set, the mechanistic model control parameter set, the evaluation parameter set, the decision control parameter set, the optimized dataset, the data model parameter set, and the mechanistic model parameter set; the application platform software has all the functions required for human-computer interaction.

[0037] The IoT data acquisition instrument collects process system variable data, including DO in the aerobic tank, ORP in the anoxic tank, temperature in the biological treatment tank, external return flow rate, internal return flow rate, sludge concentration, carbon source dosing pump operating time, phosphorus removal dosing pump operating time, influent COD, influent ammonia nitrogen, effluent COD, effluent ammonia nitrogen, effluent total nitrogen, influent flow rate, and electricity consumption. Illegal data is filtered out, and the collected dataset is uploaded to the server database every 5 seconds. The collected dataset has the function of manually entering supplementary data through the application platform, and its structure is a non-fixed time-series dataset. The data preprocessing model processes the collected dataset data in batches at a frequency of 1 time per hour. Based on the distribution characteristics of variable values, outlier data is filtered out and replaced with empty data. After rounding, averaging, on-demand difference, and time labeling, the processing results are stored and updated in the original dataset, which is a standard format dataset with a time-series interval of 1 hour, accumulating and updating over time.

[0038] The model operation control is carried out in steps of 1 day. Each model operation control is divided into 3 steps: (1) The model operation control is triggered every time the original dataset is updated 24 times. The data model, mechanism model and evaluation model are all operated independently. The model input data are all taken from the original dataset. The mechanism model is a mathematical model of dynamic correction of biochemical dynamics and hydraulic coupling parameters. It is divided into different process modules and can be selected according to the actual situation. All recommended parameters are numerical range variables. The mechanism model control parameter set is stored and updated. The data model is a comprehensive data-driven model with dynamic parameter correction. The order and number of all recommended parameters are the same as those of the mechanism model. They are all numerical variables. The data model control parameter set is stored and updated. The evaluation model is a mathematical model. The output of treatment effect and energy efficiency evaluation parameters are set, such as the compliance rate of each water quality index, the overall compliance rate of effluent, the electricity consumption per ton of water, the carbon source consumption per ton of water, and the phosphorus removal consumption per ton of water. They are all numerical variables. The evaluation parameter set is stored and updated. If the above three models have an operation time of more than 0.5 hours, their operation will be stopped. The output parameter values ​​that cannot be obtained are all empty; the output parameter values ​​of the data model that has not completed the first training are all empty; (2) Whenever the data model regulation parameter set, mechanism model regulation parameter set, and evaluation parameter set are all updated, the decision model operation is triggered. The decision model is a decision tree fuzzy logic closed-loop control comprehensive model. Regardless of whether the output parameters of the three models in the previous step are all empty, the decision model can complete the operation by calling the original dataset. The data structure and variable types of the output decision parameters are exactly the same as those of the data model. The decision regulation parameter set is stored and updated. If the operation time of the decision model exceeds 0.5 hours, the operation will stop. The output parameter values ​​that cannot be obtained are all empty; (3) Whenever the decision regulation parameter set is updated, the regulation is triggered. According to the automatic or manual settings, the decision parameters are downloaded to the IoT controller, parsed into equipment control parameters, and the current regulation operation is completed. The equipment regulation parameter execution data is transmitted to the collection dataset to form a control iteration closed-loop structure. No parsing or regulation operation is performed if the decision regulation parameters are empty.

[0039] The data quality optimization model uses imputation and repair methods to provide high-quality data for model training and optimization based on the set step batch processing of the original dataset data that is greater than the model operation control. It consists of three steps: (1) Take out a certain range of data before the target time from the original dataset and imput data with a continuous missing rate of less than 5%; (2) Repair data with a continuous missing rate of 5% to 20% by using data within a certain range before the target time; when the continuous missing data rate reaches 20% to 50%, perform cross-repair by using data within a certain range before and after the target time; when the continuous missing data rate reaches more than 50%, no repair is performed; (3) Data quality inspection, reduce the amount of imputed data and increase the amount of repaired data, repeat steps (1) to (2) until the set random sampling inspection repair accuracy standard is met, store and update the optimized dataset, whose data structure is the same as the original dataset, and mark the imputed and repaired data; when the amount of imputed data is reduced to 0 or the operation time exceeds the limit, the model operation is stopped and the optimized dataset is not updated.

[0040] The data quality optimization model operation is set to a 7-day step, that is, the original dataset data is processed once every 7 days. Imputation and repair methods are used to provide high-quality data for model training and optimization. The process is divided into 3 steps: (1) Take adjacent data from the original dataset and imput data with a continuous missing percentage of less than 5%; (2) Repair data with a continuous missing percentage of 5% to 20% by using data within 2 months before the target time; when the continuous missing data percentage reaches 20% to 50%, cross-repair is carried out by using data within 2 months before and after the target time; when the continuous missing data percentage reaches more than 50%, no repair is performed; (3) Data quality inspection, reduce the amount of imputed data and increase the amount of repaired data, repeat steps (1) to (2) until the set random sampling detection repair accuracy rate of 95% is met, store and update the optimized dataset, whose data structure is the same as the original dataset, and mark the imputed and repaired data; when the amount of imputed data is reduced to 0 or the operation time exceeds 0.5 hours, the model operation is stopped and the optimized dataset is not updated.

[0041] Optimizing dataset updates triggers the training of data models and mechanistic models. If there is a conflict with model computation, training will be started after the computation is completed. The output model parameter results will be stored in the corresponding model parameter set and the model will be updated to improve the accuracy and precision of the model computation results and enhance the model's self-learning function. Training will stop if the model training time exceeds 1 hour and the model parameters will not be updated.

[0042] The IoT data acquisition instrument has functions such as data acquisition, storage, processing, and database upload according to set requirements. The data acquisition range includes online instrument feedback values, namely various instrument status, flow rate, water quality, temperature, sludge concentration, dissolved oxygen, oxidation-reduction potential, etc. It collects instantaneous values ​​with changes exceeding 5% at a frequency of 1 time / 2 seconds, and cumulative values ​​at a frequency of 1 time / 5 seconds. All collected data is time-stamped. Data processing is performed through edge computing, combined with instrument status monitoring, instrument range setting, pump and fan operating characteristic detection, etc., to filter out abnormal data such as faults, over-range, out-of-range, and illegal values, and mark them as empty. Data is uploaded via network and can resume writing after power failure and automatically reconnect to the network for transmission.

[0043] The data model is a composite data-driven model using two or more of the following: fuzzy neural network, support vector machine, genetic algorithm, and particle swarm optimization. There are two initial training methods for the data model: First, starting with three consecutive days after system debugging where the hourly compliance rate of each water output indicator exceeds 85%, the optimized dataset is used to accumulate 90 days of data. Model parameters are then initialized, adjusted, and repeatedly trained until the data model quality meets the set standards, thus completing the training. Second, a one-year standard optimized dataset created using data from a similar process system is used as the virtual operating data from the previous year for training. Model parameters are adjusted and repeatedly trained until the data model quality meets the set standards, thus completing the training. The initial training is completed; after the initial training is completed, normal training begins. The training interval is 15 days. Training starts at the set time, taking the current 60 days of data from the self-optimized dataset and the same month of the previous year plus 30 days before and after it, for a total of 150 days of data, for training until the data model quality verification meets the set standard. The data model quality verification method is to set the output control parameters as daily sludge discharge, internal recirculation ratio, dissolved oxygen in the aerobic zone, carbon source dosage concentration, and phosphorus removal dosage concentration, and compare the COD and ammonia nitrogen values ​​of the last 5 days in the optimization dataset with the actual values. If the prediction accuracy of each effluent water quality variable is greater than 95%, the training is considered complete. After the data model operation is triggered, with the standard limits for effluent COD and ammonia nitrogen set at 30 mg / L and 1 mg / L respectively, or 25 mg / L and 0.7 mg / L respectively if the effluent is required to meet the standard in real time, and with the goal of minimizing the predicted power consumption and reagent consumption, optimization calculation is carried out to obtain the recommended control parameter values, thus completing the calculation.

[0044] Model simplification is achieved through variable clustering, time scale separation, and parameter selection. Initial model parameters are determined by setting constants for wastewater treatment process system facilities and equipment, using empirical functions, and initial mechanistic model calculations are completed. Dynamic correction of model parameters is achieved through mechanistic model training. The training method starts when the hourly compliance rate of each effluent indicator exceeds 85% for the first three consecutive days after system debugging, and enters normal training after the optimized dataset has accumulated 30 days of data. Training is started according to the set training time interval and continues until the mechanistic model quality inspection meets the set standard. The mechanistic model parameters are then updated. If the model calculation time exceeds the limit, the calculation ends, and the original model parameters are used again. The mechanistic model quality inspection method is the same as the data model quality inspection method.

[0045] The mechanistic model is a simplified model with dynamically corrected parameters based on the coupling of biochemical dynamics and hydraulics (ASM-CFD). Model simplification is achieved through variable clustering, time-scale separation, and parameter selection. Initial model parameters are determined by setting constants for the wastewater treatment process system facilities and equipment, using empirical functions, and initial mechanistic model calculations are completed. Dynamic parameter correction is achieved through mechanistic model training. The training method begins when the hourly compliance rate of each effluent indicator exceeds 85% for three consecutive days after system debugging, and normal training begins after accumulating 30 days of optimized dataset data. Training is initiated at 7-day intervals, continuing until the mechanistic model quality inspection meets the set standards (same as the data-driven model). At this point, the mechanistic model parameters are updated, and if the model computation time exceeds the limit, the computation ends, and the original model parameters are reused. The mechanistic model quality inspection method is the same as the data model quality inspection method.

[0046] The evaluation model is a simplified ASM mathematical model with dynamically adjusted model parameters. The input is the original dataset, used for evaluating the control effect. The evaluation model's step size is set to 7 days. Based on the previous step's evaluation parameters, such as the compliance status of various effluent indicators, electricity consumption per ton of water, and chemical consumption per ton of water, the suitability of the control direction and intensity is evaluated. The parameter adjustment of the evaluation model is synchronized with the training of the mechanistic model, using the same method, until the evaluation model's quality verification meets the set standards, at which point the model parameters are updated, completing the process. The quality verification method for the evaluation model is the same as that for the data model. Compared with actual values, the prediction accuracy of each evaluation variable, such as electricity consumption per ton of water and chemical consumption per ton of water, is greater than 95%.

[0047] The decision model is a comprehensive model combining decision trees, fuzzy logic control, and closed-loop iterative models. It stores the knowledge, experience, and rules of domain experts, coordinates the results of the previous model's calculations, and solves operational problems. Regardless of whether the results of the previous model's calculations are all empty, the decision model can complete the calculations using the original dataset and store them in the decision control parameter set. The decision model provides appropriate adjustment parameters based on its built-in logical judgments and evaluation model results.

[0048] The IoT controller has functions such as collecting power equipment status, operating current and frequency, parsing adjustment parameters into control parameters, setting control modes and control parameters, operating equipment according to logic programs or manual operation commands, data storage, and network transmission. The electromechanical equipment status is collected using a combination of status change thresholds and a 1-time / 5-second data acquisition method, and all collected data is time-stamped. By associating equipment operating health status with variables such as voltage, current, and frequency, it can realize equipment fault early warning. Control parameters are adjusted according to the set single adjustment limit, and adjustments are made in stages to reach the limit if it is exceeded.

[0049] The data quality optimization model operates in two steps: First, Gaussian, Poisson, and binomial distributions are used to remove outliers from different variables. For data with consecutive missing values ​​of less than 5%, exponential smoothing and multiple linear regression are used to impute the missing data, and the imputed data are labeled. Second, a combination of LSTM and RNN is used to repair data with consecutive missing values ​​of 5%-50% using the nearest neighbor data, and the repaired data are labeled.

[0050] This invention was applied in a sewage treatment plant in a township in Beijing. Under the premise of achieving stable effluent compliance, the process operation volume was reduced by 25%, power consumption was reduced by 7%, and phosphorus removal and carbon source agents were reduced by 5%, thus achieving the goals of improving process control and energy saving.

[0051] Those skilled in the art should understand that the above description of the embodiments of the present invention is only intended to illustrate the beneficial effects of the embodiments of the present invention, and is not intended to limit the embodiments of the present invention to any of the examples given.

[0052] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. An intelligent control system for wastewater treatment processes, characterized in that, Includes IoT data acquisition devices, servers, and IoT controllers: Server deployment includes databases, data preprocessing models, data models, mechanistic models, evaluation models, decision-making models, data quality optimization models, and application platform software; The database includes the original dataset, data model control parameter set, mechanism model control parameter set, evaluation parameter set, decision control parameter set, optimization dataset, data model parameter set, and mechanism model parameter set; the application platform software has all the functions required for human-computer interaction. The IoT data acquisition instrument collects variable data from the process system, filters out invalid data, and uploads the collected dataset to the server database at a set frequency. The collected dataset has the function of manually entering data through the application platform, and its structure is a non-fixed time sequence dataset. The data preprocessing model processes the collected dataset data in batches at a frequency of once per hour. Based on the distribution characteristics of variable values, it filters out outlier data and replaces it with empty data. After integerization, it takes the difference, mean, and mode as needed according to the variables. The processing results are stored in and update the original dataset, which has a standard format dataset with a time interval of 1 hour, and accumulates and updates over time. The model operation control is run in a set step. Each model operation control includes the following three steps: (1) The model operation control is triggered whenever the original dataset is updated a certain number of times at a set frequency. The data model, mechanism model and evaluation model are all operated independently. The model input data are all taken from the original dataset. The mechanism model is a mathematical model for dynamic correction of biochemical dynamics and hydrodynamic coupling parameters. It is divided into different process modules. All recommended parameters are numerical range variables. The mechanism model is stored and updated in the mechanism model control parameter set. The data model is a comprehensive data-driven model for dynamic correction of biochemical dynamics and hydrodynamic coupling parameters. All recommended parameters are in the same order and quantity as the mechanism model. They are all numerical variables. The data model control parameter set is stored and updated. The evaluation model is a mathematical model. The processing effect and energy efficiency evaluation parameters are output according to the settings. They are all numerical variables. The evaluation parameter set is stored and updated. If the above three models exceed the operation time limit, their operation will be stopped and the output parameters that cannot be obtained will be stopped. All values ​​are empty; the output parameter values ​​of the data model that has not completed the first training are all empty; (2) Whenever the data model regulation parameter set, mechanism model regulation parameter set, and evaluation parameter set are all updated, the decision model operation is triggered. The decision model is a decision tree fuzzy logic closed-loop control comprehensive model. Regardless of whether the output parameters of the three models in step (1) are all empty, the decision model can complete the operation by calling the original dataset. The data structure and variable types of the output decision parameters are exactly the same as those of the data model. The decision regulation parameter set is stored and updated. If the operation time of the decision model exceeds the limit, the operation will stop. The output parameter values ​​that cannot obtain the result are all empty; (3) Whenever the decision regulation parameter set is updated, regulation is triggered. According to the automatic or manual setting, the decision parameters are downloaded to the IoT controller, parsed into equipment control parameters, and the current regulation operation is completed. The equipment regulation parameter execution data is transmitted to the collection dataset to form a control iteration closed-loop structure. No parsing and regulation operation is performed when the decision regulation parameters are empty; The data quality optimization model uses the step batch processing of the original dataset data with a set value greater than the model operation control to provide high-quality data for model training and optimization by imputation and repair methods. It includes the following three steps: (1) Take out a certain range of data before the target time from the original dataset and imput data with a continuous missing percentage of less than 5%; (2) Repair data with a continuous missing percentage of 5% to 20% by using data within a certain range before the target time; when the continuous missing data percentage reaches 20% to 50%, perform cross-repair by using data within a certain range before and after the target time; when the continuous missing data percentage reaches more than 50%, no repair is performed; (3) Data quality inspection, reduce the amount of imputed data and increase the amount of repaired data, repeat steps (1) to (2) until the set random sampling inspection repair accuracy standard is met, store and update the optimized dataset, whose data structure is the same as the original dataset, and mark the imputed and repaired data; when the amount of imputed data is reduced to 0 or the operation time exceeds the limit, stop the model operation and do not update the optimized dataset. Optimizing dataset updates triggers the training of data models and mechanistic models; if there is a conflict with model computation, training will be started after the computation is completed, the output model parameter results will be stored in the corresponding model parameter set and the model will be updated; if the model training time exceeds the limit, training will be stopped and the model parameters will not be updated.

2. The intelligent control system for wastewater treatment processes according to claim 1, wherein, The IoT data acquisition instrument has the functions of collecting, storing, processing and uploading data to the database according to the set requirements; the data acquisition range includes online instrument feedback values, namely various instrument status, flow rate, water quality, temperature, sludge concentration, dissolved oxygen, oxidation-reduction potential, instantaneous values ​​are acquired using a combination of changing threshold and set frequency, cumulative values ​​are acquired using a set frequency acquisition method, and all acquired data are time-stamped. Data processing is performed through edge computing, combined with instrument status monitoring, instrument range setting, and pump and fan operating condition characteristic detection. Abnormal data such as faults, over-range, out-of-range, and illegal values ​​are filtered out and marked as empty. Data is uploaded through the network and can resume writing after power failure and automatically resume transmission online.

3. The intelligent control system for wastewater treatment processes according to claim 1, wherein, The data model is a composite data-driven model that includes at least two of the following: fuzzy neural network, support vector machine, genetic algorithm, and particle swarm optimization algorithm. The first training of the data model begins when the hourly compliance rate of each water output indicator exceeds 85% for three consecutive days after the system is debugged. After the dataset has accumulated 90 days of data, the model parameters are initialized and debugged and trained repeatedly until the data model quality inspection meets the set standard, thus completing the first training. or A one-year standard optimization dataset created using data from similar process systems was used as the virtual operating data from the previous year for training. The model parameters were adjusted and the training was repeated until the data model quality inspection met the set standards, thus completing the first training. After the data model completes its initial training, it enters normal training. Training is started according to the set training time interval. When the set time is reached, the data of the current 60 days and the same month of the previous year and the adjacent 30 days before and after it are taken from the self-optimized dataset, for a total of 150 days of data. Training is carried out until the data model quality check meets the set standard. The quality verification method of the data model is to compare the predicted values ​​of each set effluent water quality variable on a set date in the optimized dataset with the actual values. If the prediction accuracy of each effluent water quality variable is greater than the set standard, the training is complete. The data model's operation method is as follows: after the model operation is triggered, the water quality index compliance value is set with the safety margin as a constraint, and the minimum predicted power consumption and reagent consumption is the goal. The optimization operation is carried out to obtain the recommended control parameters, and the operation is completed.

4. The intelligent control system for wastewater treatment process according to claim 1, wherein, The mechanistic model is a simplified model with dynamically corrected parameters that couples biochemical dynamics and hydraulics. The model simplification is achieved through variable clustering, time scale separation, and parameter screening. The initial model parameters are determined by initializing the constants of the wastewater treatment process system facilities and equipment using empirical functions, and the initial mechanistic model calculation is completed. The model parameters are then dynamically corrected through mechanistic model training. The training method begins when the hourly compliance rate of each effluent indicator exceeds 85% for the first three consecutive days after system debugging. After the optimized dataset has accumulated 30 days of data, normal training begins. Training is started according to the set training time interval and continues until the quality of the mechanistic model meets the set standard. The mechanistic model parameters are then updated. If the model operation time exceeds the limit, the operation ends and the original model parameters are used again. The quality verification method of the mechanistic model is to predict the values ​​of each set effluent water quality variable in the optimized dataset on a set date and compare them with the actual values. If the prediction accuracy of each effluent water quality variable is greater than the set standard, the training is complete.

5. The intelligent control system for wastewater treatment process according to claim 1, wherein, The evaluation model is a simplified mathematical model of ASM with dynamically corrected model parameters. The input is the original dataset, used for evaluating the recent control effect. The evaluation model steps are set, and the appropriateness of the previous control direction and intensity is evaluated by using the evaluation parameters of the compliance status of each effluent indicator, electricity consumption per ton of water, and chemical consumption per ton of water in the previous step. The parameter correction of the evaluation model is synchronized with the training of the mechanism model. Starting from the first three consecutive days after the system is debugged, when the hourly compliance rate of each effluent indicator exceeds 85%, the optimized dataset accumulates data for 30 days and then enters normal training. Training is started according to the set training time interval, and continues until the quality test of the evaluation model reaches the set standard. The model parameters are then updated, thus completing the model parameter correction. The quality test method of the evaluation model is to predict the values ​​of each set effluent water quality variable in the optimized dataset on a set date and compare them with the actual values. The prediction accuracy of each evaluation variable is greater than the set standard.

6. The intelligent control system for wastewater treatment process according to claim 1, wherein, The decision-making model is a comprehensive model combining decision trees, fuzzy logic control, and closed-loop iterative models, incorporating the knowledge, experience, and rules of experts in the storage domain. The results of these models are used to solve operational problems. Before the data model's initial training is complete, the decision-making model is calculated using the results of the mechanistic model and the evaluation model. After the data model's initial training is complete, for control parameters with a step size of less than one day, the weight of the mechanistic model's calculation results is increased; for control parameters with a step size of not less than one day, the results of the mechanistic model and the data-driven model are balanced. The decision-making model provides parameter adjustment suggestions based on its built-in logical judgments and the results of the evaluation model.

7. The intelligent control system for wastewater treatment process according to claim 1, wherein, The IoT controller has the functions of collecting power equipment status, operating current and frequency, parsing adjustment parameters into control parameters, setting control modes and control parameters, operating equipment according to logic programs or manual operation instructions, and data storage and network transmission. The electromechanical equipment status is collected using a combination of threshold changes and timing, and all collected data is time-stamped. By associating the equipment's operating health status with variables such as voltage, current and frequency, equipment fault early warning can be achieved. Control parameters are adjusted in stages according to the set single adjustment limit.

8. The intelligent control system for wastewater treatment process according to claim 7, wherein, The change threshold is 5%, and the timing interval is less than 1 hour.

9. The intelligent control system for wastewater treatment process according to claim 1, wherein, The data quality optimization model operation process includes the following two steps: Outlier data were removed using Gaussian, Poisson, and binomial distributions. For data with consecutive missing values ​​within 5%, a combination of linear, autoregressive integral moving average, exponential smoothing, multiple linear regression, and vector autoregression methods were used to impute the missing data, and the imputed data were labeled. Using one or more combinations of models such as LSTM, RNN, and gradient boosting regression tree, and utilizing data from the same period of the previous year and recent neighboring data, we impute data with consecutive missing values ​​of 5%-50% and label the imputed data.

Images