A method for predicting the amount of aeration required to maintain dissolved oxygen stability in activated sludge systems

Abstract

The present invention discloses a method for predicting the amount of aeration required to maintain the stability of dissolved oxygen in an activated sludge system, including measuring the influent flow rate, influent organic matter concentration, influent ammonia nitrogen concentration, activated sludge system dissolved oxygen concentration, sludge concentration, and aeration air volume data of a biochemical pool at intervals of t1 time period within a certain time period; in view of the problem of hysteresis in the change of dissolved oxygen concentration in the activated sludge system, the dissolved oxygen concentration data at the time of measurement is replaced with the dissolved oxygen concentration data after the t2 time period at the time; the replaced dissolved oxygen concentration data is screened to form a data set; a random forest model is constructed, and a machine learning matrix is ​​created using the data in the data set to train the random forest model; the prediction performance of the trained random forest model is evaluated; and the trained random forest model is used to predict the air volume value required to achieve the target dissolved oxygen value after the t2 time period, thereby adjusting the output air volume of the aeration fan. The present invention solves the problem of hysteresis in the change of dissolved oxygen concentration in the prior art in regulating the aeration amount in the activated sludge system, thereby maintaining the dissolved oxygen concentration within the target range.

Description

Technical Field

[0001] The invention relates to the technical field of sewage treatment, and in particular to a method for predicting the aeration amount required to maintain dissolved oxygen stability in an activated sludge system. Background Art

[0002] The activated sludge system of a sewage treatment plant mainly removes pollutants from sewage through the metabolism of microorganisms to achieve the purpose of purifying water quality. The microbial metabolism process consumes oxygen, so the dissolved oxygen concentration (DO) is an important indicator for the normal operation of the activated sludge system. In order to maintain a certain dissolved oxygen concentration in the activated sludge environment, the aeration fan needs to continuously aerate the activated sludge system. However, too much or too little air supply will cause the dissolved oxygen concentration to fluctuate, resulting in unstable microbial metabolic activities in the activated sludge system, affecting the sewage treatment efficiency and effluent water quality.

[0003] At present, sewage treatment plants generally regulate the dissolved oxygen in activated sludge systems based on experience. By manually adjusting the amount of aeration, sewage treatment plants usually maintain the dissolved oxygen in activated sludge systems at 2-4 mg / L. Due to the hysteresis of dissolved oxygen regulation, aeration cannot be adjusted in time, resulting in excessive air being injected into the system when the dissolved oxygen has reached or exceeded the expected value, resulting in energy waste and unstable system operation. Therefore, accurately adjusting the aeration volume to maintain the stability of the dissolved oxygen concentration in the activated sludge system is of great significance to the stable operation and energy saving of sewage treatment plants. The current manual regulation method relies too much on experience, requires a lot of manual operation, has poor stability and high energy consumption. Therefore, accurately predicting and calculating the aeration volume required to maintain the stability of dissolved oxygen in the activated sludge system and realizing the automation and intelligent regulation of aeration volume are one of the most concerned issues in the field of sewage treatment in recent years.

[0004] At present, some aeration control prediction models and control strategies for activated sludge systems have been reported. Chinese patent document CN202010398290.5 discloses a method for implementing precise aeration based on a deep learning algorithm model. Combined with the collection of historical operation data, deep learning is used to predict the system's dissolved oxygen demand, ensuring that the lowest aerator energy consumption is obtained while meeting the dissolved oxygen concentration requirements in the wastewater, thereby reducing the cost of wastewater treatment. However, there is still a hysteresis problem in the change of dissolved oxygen concentration during the aeration volume adjustment process, that is, the dissolved oxygen concentration in the biochemical treatment tank cannot immediately reach the target value after adjusting the aeration volume. Summary of the invention

[0005] Purpose of the invention: The purpose of the present invention is to provide a method for predicting the amount of aeration required to maintain the stability of dissolved oxygen in an activated sludge system. Based on the hysteresis problem of changes in dissolved oxygen concentration in an activated sludge system, the aeration amount is adjusted in advance to achieve stability of dissolved oxygen in the sludge system.

[0006] Technical solution: To achieve the above purpose, the method for predicting the aeration amount required to maintain the dissolved oxygen stability of an activated sludge system according to the present invention comprises the following steps:

[0007] Step 1: Within a certain period of time, measure and record the influent flow rate, influent organic matter concentration, influent ammonia nitrogen concentration, activated sludge system dissolved oxygen concentration, sludge concentration, and aeration air volume data of the biochemical pool of the sewage treatment plant at intervals of t1;

[0008] Step 2: In view of the hysteresis problem of dissolved oxygen concentration change in the activated sludge system, the dissolved oxygen concentration data at the time of measurement is replaced with the dissolved oxygen concentration data after the time period t2 at that time based on each record in step 1;

[0009] Step 3: Screen the dissolved oxygen concentration data after the replacement in step 2, delete the data that exceeds the normal range of dissolved oxygen concentration, and form a data set with the screened dissolved oxygen concentration data and the influent flow rate, influent organic matter concentration, influent ammonia nitrogen concentration, activated sludge system dissolved oxygen concentration, sludge concentration, and aeration air volume data measured in step (1);

[0010] Step 4: Build a random forest model, use the data in the dataset to build a machine learning matrix, divide the machine learning matrix into a training set and a test set, and train and evaluate the prediction performance of the random forest model respectively;

[0011] Step 5: The dissolved oxygen concentration that needs to be reached after the t2 time period at the current moment is taken as the target dissolved oxygen value, and the target dissolved oxygen value and the measured influent flow rate, influent organic matter concentration, influent ammonia nitrogen concentration, and sludge concentration data at the current moment are input into the trained random forest model to predict the air volume value required to achieve the target dissolved oxygen value, and adjust the output air volume of the aeration fan according to the air volume value.

[0012] The time period t2 described in step 2 is the delayed feedback time of the change of dissolved oxygen in the activated sludge system, which is an integer multiple of the time period t1.

[0013] Among them, the normal range of dissolved oxygen concentration in step 3 is usually 2.0-3.0 mg / L.

[0014] Among them, the machine learning matrix described in step 4 is:

[0015]

[0016] Among them, FL1~FL m is the aeration air volume data, COD1~COD m Influent chemical oxygen demand data, intFL1~intFL mis the water volume data, N1~N m is the influent ammonia nitrogen concentration data, DO t-1 ~DO t-m is the dissolved oxygen concentration data of the activated sludge system, SS1~SS m is the sludge concentration; m is the total number of records measured at intervals of t1 within a certain period of time.

[0017] Among them, the random forest model training process described in step 4 is: first randomly extract the training set to form multiple sub-data sets, then independently build a decision tree based on each sub-data set to generate multiple decision trees, and in the construction process of each decision tree, select a random feature subset to divide the nodes until the predetermined number of leaf nodes is reached or it cannot be divided.

[0018] The prediction performance evaluation method described in step 4 is: by calculating the mean absolute percentage error MAPE and the determination coefficient R of the random forest model after training in the test set. 2 The error between the output predicted value and the measured value is used to characterize the model prediction performance. 2 >0.8 and MAPE<20%, the model has good predictive power, otherwise the dissolved oxygen concentration data in the data set are screened again.

[0019] Among them, the method for re-screening the dissolved oxygen concentration data in the data set is: adjusting the value of the t2 time period and the normal range of dissolved oxygen concentration, replacing the dissolved oxygen concentration data at the measurement time in the data set with the data of the dissolved oxygen concentration after the adjusted t2 time period at that moment, and using the data set to re-construct the machine learning matrix, so as to re-train the random forest model.

[0020] The adjusted time period t2 is the delayed feedback time of the change of dissolved oxygen in the activated sludge system, which is usually in the range of 5-30 minutes and is an integer multiple of the time period t1.

[0021] Among them, the use of the trained random forest model in step 5 to predict the air volume required to achieve the target dissolved oxygen value is used when the direct monitoring value of the dissolved oxygen concentration in the activated sludge system fluctuates greatly or is about to exceed the normal range of the target value. The required air volume value predicted by the trained random forest model is used to adjust the aeration fan output air volume in advance so that the dissolved oxygen concentration reaches the required target value after the t2 time period.

[0022] Among them, the direct monitoring value of the dissolved oxygen concentration in the activated sludge system in step 5 is within the range of the target value ±0.5, and the dissolved oxygen concentration is considered to be in a stable state. When the dissolved oxygen concentration exceeds this range, the trained random forest model predicts the required air volume value, and the aeration fan output air volume is regulated in advance.

[0023] Beneficial effects: The present invention has the following advantages: 1. This method fully considers the problem of delayed reaction of dissolved oxygen concentration change, replaces and screens the dissolved oxygen concentration data at the time of measurement, ensures the timeliness and accuracy of the data, and at the same time, by predicting in advance the aeration air volume required to maintain the dissolved oxygen stability of the activated sludge system, ensures that the dissolved oxygen concentration can always be maintained within the target value range, so as to achieve the purpose of maintaining the dissolved oxygen concentration stability in the activated sludge system;

[0024] 2. This method constructs a machine learning data matrix to train the random forest model, which not only improves the model's fit to the biochemical pool operation data of the sewage treatment plant, but also enhances the accuracy of the predicted air volume, providing strong support for the optimized operation of the activated sludge system. It has high reliability and practicality in practical applications, and helps to achieve more low-cost and efficient sewage treatment;

[0025] 3. This method measures the operating data of the biochemical pool of the sewage treatment plant, including key parameters such as influent flow rate, influent organic matter concentration, influent ammonia nitrogen concentration, dissolved oxygen concentration of the activated sludge system, sludge concentration, and aeration air volume, and fully considers the impact of various operating parameters on dissolved oxygen changes, providing rich data support for the machine learning data matrix. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Schematic diagram of the process of this method;

[0027] Figure 2 This is the fitting effect diagram of aeration air volume prediction using this method. DETAILED DESCRIPTION

[0028] The technical solution of the present invention is described in detail below in conjunction with the embodiments and drawings.

[0029] like Figure 1 As shown, the method for predicting the aeration amount required to maintain the dissolved oxygen stability of an activated sludge system according to the present invention comprises the following steps:

[0030] Step 1: Within a certain period of time, measure and record the influent flow rate, influent organic matter concentration, influent ammonia nitrogen concentration, activated sludge system dissolved oxygen concentration, sludge concentration, and aeration air volume data of the biochemical pool of the sewage treatment plant at intervals of t1. The value range of t1 is 1-10 minutes.

[0031] Step 2: In order to solve the problem of hysteresis in the change of dissolved oxygen concentration in the activated sludge system, the dissolved oxygen concentration data at the time of measurement is replaced with the dissolved oxygen concentration data after the time period t2 at that time based on each record in step 1.

[0032] Among them, the time period t2 is the delayed feedback time of the change of dissolved oxygen in the activated sludge system, and the value range of t2 is 5-30 minutes, which is an integer multiple of the time period t1.

[0033] Step 3: Filter the dissolved oxygen concentration data after the replacement in step 2, delete the data that exceeds the normal range of dissolved oxygen concentration (2.0-3.0 mg / L), and combine the filtered dissolved oxygen concentration data and the influent flow rate, influent organic matter concentration, influent ammonia nitrogen concentration, activated sludge system dissolved oxygen concentration, sludge concentration, and aeration air volume data measured in step (1) to form a data set.

[0034] Step 4: Build a random forest model, use the data in the dataset to build a machine learning matrix, divide the machine learning matrix into a training set and a test set, and evaluate the training and prediction performance of the random forest model respectively.

[0035] Among them, the machine learning matrix:

[0036]

[0037] Among them, FL1~FL m is the aeration air volume data, COD1~COD m Influent chemical oxygen demand data, intFL1~intFL m is the water volume data, N1~N m is the influent ammonia nitrogen concentration data, DO t-1 ~DO t-m is the dissolved oxygen concentration data of the activated sludge system, SS1~SS m is the sludge concentration; m is the total number of records measured at intervals of t1 within a certain period of time.

[0038] The training process of the random forest model is as follows: first, randomly extract the training set to form multiple sub-datasets, then independently build a decision tree based on each sub-dataset to generate multiple decision trees. During the construction of each decision tree, a random feature subset is selected to divide the nodes until the predetermined number of leaf nodes is reached or it cannot be divided.

[0039] The prediction performance evaluation method is to calculate the error evaluation indexes such as the mean absolute percentage (MAPE) and the determination coefficient R of the random forest model after training in the test set. 2 The error between the predicted value and the measured value is used to characterize the prediction performance of the model. Generally speaking, a MAPE less than 10% is considered a good prediction model, and a MAPE between 10% and 20% is considered to be an acceptable prediction accuracy. 2>0.8 and the error between the predicted value and the measured value is within 20%, then the model has good predictive power, otherwise the dissolved oxygen concentration data in the data set are screened again.

[0040] The method for re-screening the dissolved oxygen concentration data in the data set is: adjusting the value of the t2 time period and the normal range of dissolved oxygen concentration, replacing the dissolved oxygen concentration data at the time of measurement in the data set with the data of the dissolved oxygen concentration after the adjusted t2 time period at that time, and using the data set to re-construct the machine learning matrix, so as to re-train the random forest model. The adjusted t2 time period is the lag feedback time of the dissolved oxygen change in the activated sludge system, which usually ranges from 5 to 30 minutes, which is an integer multiple of the t1 time period.

[0041] In this embodiment, t1 is set to 1 minute, t2 is set to 10 minutes, and the error indicators obtained by running the function in scikit-learn are: MSE = 11528.36, MAE = 59.68, RMSE = 102.25, R 2 =0.9828>0.8. 2 Close to 1 indicates that the training effect of the model is good. The calculated values ​​of other error indicators are related to the amount of data. In this embodiment, they are used to compare with the modeling effect of the unreplaced and screened data set. Based on the prediction data set composed of the model prediction values ​​and the corresponding actual sewage treatment aeration plant air volume data output, the relative error between the air volume prediction value obtained by the model and the measured value in the prediction data set is calculated, and finally the average absolute error of the prediction data set MAPE = 4.53% <10% is obtained. The fitting degree of the prediction model effect can be obtained, as shown in the attached figure. Figure 2 As shown, it is shown that the random forest model constructed according to the historical operating parameters of the sewage treatment plant has a good prediction effect.

[0042] In order to compare the prediction effect of the replaced and filtered data sets, the original data set (the dissolved oxygen concentration data at the time of measurement was not replaced with the dissolved oxygen concentration data after the time period t2 at that time, and the dissolved oxygen concentration data was not filtered) was also input into the above random forest model for training, and the MSE of the model was calculated to be 12122.84, MAE = 54.23, RMSE = 110.10, R 2 =0.9719, MAPE of the prediction data set =5.59%>4.53%. The comprehensive evaluation indicators show that the accuracy of the model in predicting aeration air volume can be improved by screening the training data set.

[0043] Step 5: The dissolved oxygen concentration that needs to be reached after the t2 time period at the current moment is taken as the target dissolved oxygen value, and the target dissolved oxygen value and the measured influent flow rate, influent organic matter concentration, influent ammonia nitrogen concentration, and sludge concentration data at the current moment are input into the trained random forest model to predict the air volume value required to achieve the target dissolved oxygen value, and adjust the output air volume of the aeration fan according to the air volume value.

[0044] Among them, the air volume value required to achieve the target dissolved oxygen value predicted by the trained random forest model in step 5 is used when the direct monitoring value of the dissolved oxygen concentration in the activated sludge system fluctuates greatly or is about to exceed the normal range of the target value. The required air volume value predicted by the trained random forest model is used to adjust the aeration fan output air volume in advance so that the dissolved oxygen concentration reaches the required target value after the t2 time period. The direct monitoring value of the dissolved oxygen concentration in the activated sludge system is within the range of ±0.5 of the target value, and the dissolved oxygen concentration is considered to be in a stable state. When the dissolved oxygen concentration exceeds this range, the required air volume value predicted by the trained random forest model is used to adjust the aeration fan output air volume in advance. In this embodiment, the target dissolved oxygen concentration is 2.5 mg / L. When the monitoring value of the dissolved oxygen concentration in the activated sludge system fluctuates significantly or is about to exceed the ±0.5 mg / L range of the set target value, that is, 2.0-3.0 mg / L, it is considered that the dissolved oxygen in the system cannot maintain stability. At this time, it is necessary to predict the aeration air volume through the model, and perform aeration control according to the predicted air volume value obtained.

Claims

1. A method for predicting the amount of aeration required to maintain dissolved oxygen stability in an activated sludge system, characterized in that: The following steps are involved: Step 1: Within a certain period of time, measure and record the influent flow rate, influent organic matter concentration, influent ammonia nitrogen concentration, activated sludge system dissolved oxygen concentration, sludge concentration, and aeration air volume data of the biochemical pool of the sewage treatment plant at intervals of t1; Step 2: In view of the hysteresis problem of dissolved oxygen concentration change in the activated sludge system, the dissolved oxygen concentration data at the time of measurement is replaced with the dissolved oxygen concentration data after the time period t2 at that time based on each record in step 1; Step 3: Screen the dissolved oxygen concentration data after the replacement in step 2, delete the data that exceeds the normal range of dissolved oxygen concentration, and combine the screened dissolved oxygen concentration data and the influent flow rate, influent organic matter concentration, influent ammonia nitrogen concentration, activated sludge system dissolved oxygen concentration, sludge concentration, and aeration air volume data measured in step (1) to form a data set; Step 4: Build a random forest model, use the data in the dataset to build a machine learning matrix, divide the machine learning matrix into a training set and a test set, and train and evaluate the prediction performance of the random forest model respectively; Step 5: The dissolved oxygen concentration that needs to be reached after the t2 time period at the current moment is taken as the target dissolved oxygen value, and the target dissolved oxygen value and the measured influent flow rate, influent organic matter concentration, influent ammonia nitrogen concentration, and sludge concentration data at the current moment are input into the trained random forest model to predict the air volume value required to achieve the target dissolved oxygen value, and adjust the output air volume of the aeration fan according to the air volume value; The machine learning matrix described in step 4: ， Among them, FL1~FL m is the aeration air volume data, COD1~COD m Influent chemical oxygen demand data, intFL1~intFL m is the water volume data, N1~N m is the influent ammonia nitrogen concentration data, DO t-1 ~DO t-m is the dissolved oxygen concentration data of the activated sludge system, SS1~SS m is the sludge concentration; m is the total number of records measured at intervals of t1 in a certain period of time; The prediction performance evaluation method described in step 4 is: by calculating the mean absolute percentage error MAPE and the determination coefficient R of the trained random forest model on the test set 2 The error between the output predicted value and the measured value is used to characterize the model prediction performance. 2 >0.8 and MAPE<20%, the model has good predictive power, otherwise the dissolved oxygen concentration data in the data set are replaced again; The method for replacing the dissolved oxygen concentration data in the data set again is: adjusting the value of the t2 time period and the normal range of the dissolved oxygen concentration, replacing the dissolved oxygen concentration data at the measurement time in the data set with the data of the dissolved oxygen concentration after the adjusted t2 time period at that time, and using the data set to re-construct the machine learning matrix, so as to re-train the random forest model; The adjusted time period t2 is the delayed feedback time of the change of dissolved oxygen in the activated sludge system, and the value range is 5-30 minutes, which is an integer multiple of the time period t1; The method of using the trained random forest model to predict the air volume required to achieve the target dissolved oxygen value in step 5 is used when the direct monitoring value of the dissolved oxygen concentration in the activated sludge system fluctuates greatly or is about to exceed the normal range of the target value. The trained random forest model is used to predict the required air volume, and the aeration fan output air volume is adjusted in advance so that the dissolved oxygen concentration reaches the required target value after the t2 time period.

2. The method for predicting the amount of aeration required to maintain dissolved oxygen stability in an activated sludge system according to claim 1, characterized in that: The time period t2 described in step 2 is the lag feedback time of the change of dissolved oxygen in the activated sludge system, which is an integer multiple of the time period t1.

3. The method for predicting the amount of aeration required to maintain dissolved oxygen stability in an activated sludge system according to claim 1, characterized in that: The normal range of dissolved oxygen concentration in step 3 is usually 2.0-3.0 mg / L.

4. The method for predicting the amount of aeration required to maintain dissolved oxygen stability in an activated sludge system according to claim 1, characterized in that: In step 5, the direct monitoring value of the dissolved oxygen concentration in the activated sludge system is within the range of the target value ±0.5 mg / L, and the dissolved oxygen concentration is considered to be in a stable state. When the dissolved oxygen concentration exceeds this range, the trained random forest model predicts the required air volume value, and the aeration fan output air volume is regulated in advance.

Images