A sewage treatment performance intelligent prediction method and device

Abstract

The present application belongs to the field of sewage treatment, and particularly relates to a sewage treatment performance intelligent prediction method and device. The specific technical scheme is as follows: a sewage treatment performance prediction method, which takes the microbial respiration rate as a characteristic index for prediction. The present application uses the microbial activity index to construct an intelligent prediction model of the pollutant removal performance such as the pollutant removal rate in sewage, strengthens the description of the influence of the microbial action in the artificial intelligent model on the sewage treatment performance, and solves the problem that the model prediction result is inaccurate due to the lack of online data of water quality in the current numerous sewage treatment systems.

Description

Technical Field

[0001] This invention belongs to the field of wastewater treatment, specifically relating to a method and device for intelligent prediction of wastewater treatment performance. Background Technology

[0002] Wastewater treatment processes are characterized by nonlinearity, large inertia, large time delay, and strong coupling. Therefore, current operation and management methods based primarily on human experience often suffer from lag and uncertainty, making it difficult to meet the requirements for stable, economical, and efficient operation of wastewater treatment systems.

[0003] To improve wastewater treatment efficiency and reduce the risk of over-discharge during operation, an increasing number of theoretical studies and engineering practices are constructing predictive models using various methods to support the operation and management of wastewater treatment systems. Support Vector Machines (SVM), Autoregressive Integral Moving Averages (ARIMA), and Nonlinear Autoregressive Networks (NAR) algorithms have been used to build time series prediction models. However, many wastewater treatment plants (such as rural wastewater treatment facilities, township wastewater treatment plants, and newly built wastewater treatment plants) lack online water quality data, resulting in poor model prediction accuracy and difficulty in effectively guiding the optimization and control of wastewater treatment systems.

[0004] To improve the prediction accuracy of pollutant removal efficiency in wastewater, some researchers have considered incorporating noise into the original data to enhance the generalization ability of learning methods. However, traditional machine learning relies on the consistency of sample distribution. Therefore, these methods of generating virtual samples do not accurately reflect the distribution of the original data, leading to unsatisfactory prediction results. Other studies have considered combining transfer learning algorithms to transfer data with similar features from other wastewater treatment systems to the target system when data is insufficient, thereby improving the prediction accuracy of pollutant removal efficiency. However, due to differences in influent water quality, operating conditions, and even wastewater treatment processes among different wastewater treatment systems, the construction and application of prediction models based on transfer learning algorithms still face significant limitations.

[0005] Therefore, given the lack of online water quality data, taking measures to effectively improve the accuracy of wastewater treatment performance prediction is of great significance for real-time monitoring of the operation of wastewater treatment systems and guiding process optimization and stable operation. Summary of the Invention

[0006] The purpose of this invention is to provide a method and apparatus for intelligent prediction of wastewater treatment performance.

[0007] To achieve the above-mentioned objectives, the technical solution adopted by the present invention is: a method for predicting the removal performance of pollutants in wastewater, wherein the method uses microbial respiration rate as a characteristic indicator for prediction.

[0008] Preferably, the dissolved oxygen consumed per unit weight of activated sludge per unit time, SOUR, is used to characterize the microbial respiration rate.

[0009] Preferably, the SOUR is calculated using ATP as the biomass.

[0010] Preferably, the method includes the following steps:

[0011] (1) Collect raw data, including: respiratory rate (SOUR) in aerobic and anoxic pools. AOB and SOUR NOB SOUR in aerobic tank, SOUR in anoxic tank, and SOUR of denitrifying bacteria. 反 ;

[0012] (2) Obtain the feature vector x for predicting ammonia nitrogen removal rate:

[0013] x = [aerobic tank SOUR, aerobic tank SOUR] AOB aerobic tank SOUR NOB aerobic tank SOUR AOB / Aerobic tank SOUR NOB ];

[0014] and / or;

[0015] Obtain the eigenvector x for the predicted total nitrogen removal rate:

[0016]

[0017] and / or;

[0018] Obtain the feature vector x for predicting COD removal rate:

[0019] x = [aerobic tank SOUR, aerobic tank SOUR] AOB aerobic tank SOUR NOB aerobic tank SOUR AOB / Aerobic tank SOUR NOB Anoxic pool SOUR, anoxic pool SOUR 反 SOUR (anoxic pool) 反 / Anoxic pool SOUR];

[0020] (3) The feature vector x is used as the input of the random forest, and the pollutant removal performance in wastewater is calculated and predicted by the random forest model.

[0021] Preferably, each SOUR is calculated using ATP as the biomass, and the calculation formula is as follows:

[0022] SOUR = (DO1 - DO2) / (ATP × T)

[0023] SOURNOB =SOUR-SOUR1

[0024] SOUR AOB =SOUR1-SOUR2

[0025] SOUR 反 =SOUR-SOUR AOB -SOUR NOB

[0026] SOUR1 is the SOUR value measured after adding NaClO3, and SOUR2 is the SOUR value measured after adding allyl thiourea.

[0027] Preferably, when the dissolved oxygen value in the breathing chamber decreases by 1 / 3, allyl thiourea is added to inhibit the activity of ammonia-oxidizing bacteria (AOB); and when the dissolved oxygen value decreases by 2 / 3, NaClO3 is added to inhibit the activity of nitrite-oxidizing bacteria (NOB).

[0028] Preferably, the method includes the following steps:

[0029] (1) Collect raw data, including: respiratory rate (SOUR) in aerobic and anoxic pools. AOB and SOUR NOB SOUR in aerobic tank, SOUR in anoxic tank, and SOUR of denitrifying bacteria. 反 ;

[0030] (2) Randomly divide the data into training and test sets;

[0031] In the preprocessing stage, all data are normalized, and then individual base learners are trained using the training set data. The prediction results of the base learner training data are combined into a new training set, and the new training set is used to train the meta learner.

[0032] (3) Using K-fold cross-validation, the training set is divided into K parts, of which K-1 parts are used to train the base learners, and the remaining part is predicted by each trained base learner. The prediction values ​​of each base learner for the K parts form a new feature matrix.

[0033] (4) For the test data, use K trained learners with different base learners to make predictions on it, and use the mean of the predicted test set results as the new feature matrix of the test data.

[0034] (5) Use the new feature matrix for training and prediction of the meta-learner respectively to obtain the prediction results.

[0035] Preferably, the basic learner includes: Support Vector Machine (SVR), K-nearest Neighbor Regression (KNR), Linear Regression (LR), Decision Tree Model (DT), and Random Forest (RF).

[0036] Accordingly, a terminal device includes at least one processor and at least one memory; at least one program is stored in the memory and configured to be executed by the processor.

[0037] Accordingly, a computer-readable storage medium stores program code that executes the method when the program code is run by a processor.

[0038] The present invention has the following beneficial effects: The present invention provides a new method for predicting the performance of wastewater treatment, which can predict the treatment performance of the entire wastewater treatment system, especially the removal rates of ammonia nitrogen and total nitrogen, by using microbial activity indicators under the condition that influent water quality indicators are lacking or missing.

[0039] In biological wastewater treatment systems, microorganisms are the main functional components, and the composition and activity of the microbial community are intrinsic factors determining the system's performance. However, existing detection methods struggle to quickly obtain information about the microbial community in wastewater treatment systems. Microbial activity indicators, on the other hand, reflect the physiological state and metabolic activity of microorganisms, and their detection methods are relatively mature and simple. Therefore, this invention innovatively chooses to describe and predict the operating status of wastewater treatment systems based on these indicators. SOUR (Soluble Oxygen Ratio) refers to the amount of dissolved oxygen consumed per unit weight of activated sludge per unit time. It can reflect the physiological state of activated sludge and the metabolic status of the substrate from the perspective of microbial respiration rate, thus potentially directly reflecting the operating status of the wastewater biochemical treatment system. ATP is the main carrier of usable energy in cell metabolism and can characterize the number of microorganisms and the rate of microbial metabolism during biodegradation.

[0040] This invention leverages the unique advantages of microbial activity indicators in characterizing the physiological status and metabolic activity of microorganisms. It directly utilizes these indicators to construct an intelligent prediction model for pollutant removal performance in wastewater, such as the pollutant removal rate. This strengthens the description of the impact of microbial activity on wastewater treatment performance in artificial intelligence models and solves the problem of inaccurate model prediction results in many current wastewater treatment systems due to a lack of online water quality data. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the fusion model algorithm provided by the present invention. Detailed Implementation

[0042] This invention provides a method for intelligently predicting the pollutant removal performance in wastewater treatment. Taking the prediction of pollutant removal rate as an example, the method specifically includes the following steps:

[0043] 1. Collect raw data, including: respiration rate (SOUR) in aerobic and anoxic pools. AOB and SOURNOB SOUR in aerobic tank, SOUR in anoxic tank, and SOUR of denitrifying bacteria. 反 .

[0044] The preferred approach is that each SOUR is calculated based on ATP as biomass. Wherein,

[0045] SOUR ATP = (DO1-DO2) / (ATP×T)

[0046] SOUR ATP_NOB =SOUR ATP -SOUR ATP_1

[0047] SOUR ATP_AOB =SOUR ATP_1 -SOUR ATP_2

[0048] SOUR ATP_反 =SOUR ATP -SOUR ATP_AOB -SOUR ATP_NOB

[0049] SOUR ATP_1 SOUR is the solubility of oxygen (SOUR) measured after adding NaClO3 with ATP as the biomass. ATP_2 The SOUR value is the result of adding allyl thiourea with ATP as the biomass.

[0050] 2. Obtain the eigenvector x for predicting ammonia nitrogen removal rate:

[0051] x = [aerobic tank SOUR, aerobic tank SOUR] AOB OUR aerobic pool NOB aerobic tank SOUR AOB / Aerobic tank SOUR NOB ]

[0052] Obtain the eigenvector x for the predicted total nitrogen removal rate:

[0053]

[0054] Obtain the feature vector x for predicting COD removal rate:

[0055] x = [aerobic tank SOUR, aerobic tank SOUR] AOB aerobic tank SOUR NOB aerobic tank SOUR AOB / Aerobic tank SOUR NOB Anoxic pool SOUR, anoxic pool SOUR 反 SOUR (anoxic pool)反 / Anoxic pool SOUR]

[0056] 3. Using the feature vector x as input to the random forest, the removal rates of COD, ammonia nitrogen, and total nitrogen in wastewater are calculated and predicted through the random forest model.

[0057] 4. A more preferred approach is as follows: After collecting the raw data in step 1, the raw data is randomly divided into a training set and a test set. During the preprocessing stage, all data is normalized using the StandScaler() function from the sklearn library in Python. Then, individual base learners are trained using the training set data. The base learners include: Support Vector Machine (SVR), K-Nearest Neighbor Regression (KNR), Linear Regression (LR), Decision Tree Model (DT), and Random Forest (RF). The prediction results from the training data of these base learners are combined to form a new training set, which is then used to train the meta-learner.

[0058] Using K-fold cross-validation, the training set is divided into K parts. K-1 parts are used to train the base learners, and the remaining part is used for prediction by each trained base learner. The predictions from each base learner for the K parts form a new feature matrix. For the test data, predictions are made using K learned learners trained with different base learners. The mean of the predictions for the test set is used as the new feature matrix for the test data. This new feature matrix is ​​then used for both training and prediction of the meta-learner to obtain the prediction results.

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Unless otherwise specified, the technical means used in the embodiments are conventional means well known to those skilled in the art. All obtained data are average values ​​obtained after at least three repetitions, and each repetition yields valid data.

[0060] Example 1: Raw data collection and correlation analysis between microbial respiratory activity and pollutant removal rate

[0061] 1. Data on influent and effluent water quality, SOUR data, and respiration rates (SOUR) of nitrifying bacteria (ammonia oxidizing bacteria AOB and nitrite oxidizing bacteria NOB) in continuously treated wastewater at a wastewater treatment plant in Chengdu, Sichuan Province. AOB_MLSS and SOUR NOB_MLSS ), ATP and denitrifying bacteria respiration rate (SOUR) in anoxic and aerobic tanks 反_MLSSThe dissolved oxygen concentration (MLSS) of activated sludge in the aerobic tank and the dissolved oxygen concentration (MLSS) of activated sludge in the anoxic tank were monitored online. SOUR data was obtained by continuously monitoring the dissolved oxygen (DO1) in the aerobic / anoxic tanks of the wastewater treatment system and the dissolved oxygen (DO2) in the breathing chamber of the online SOUR monitoring device. ATP data was obtained using an online ATP monitoring device. Influent and effluent water quality data, and MLSS in the anoxic / aerobic tanks, were obtained using online monitoring equipment at the wastewater treatment plant. AOB SOUR NOB and SOUR 反 Data were obtained through online DO probe detection in the breathing chamber. When the DO value in the breathing chamber decreased by 1 / 3, allyl thiourea (ATU) was added to inhibit AOB activity; and when the DO value decreased by 2 / 3, NaClO3 was added to inhibit NOB activity. After aligning the influent and effluent water quality data according to the actual hydraulic retention time (HRT), the raw data are shown in Table 1.

[0062] Table 1. Raw Data Collection

[0063]

[0064]

[0065] 2. Calculate the removal rates of COD, ammonia nitrogen, and total nitrogen based on Table 1. The calculation formulas are as follows:

[0066] COD removal rate = (Influent COD concentration - Effluent COD concentration) / HRT

[0067] Ammonia nitrogen removal rate = (influent ammonia nitrogen concentration - effluent ammonia nitrogen concentration) / HRT

[0068] Total nitrogen removal rate = (Influent total nitrogen concentration - Effluent total nitrogen concentration) / HRT

[0069] (1) Calculate SOUR(SOUR) based on MLSS MLSS Pearson correlation analysis was performed on the data in Table 1 to analyze the microbial respiration rate (SOUR). MLSS The correlation between the pollutant removal rate and the pollutant removal rate is shown in Table 2.

[0070] SOUR MLSS =(DO1-DO2) / (MLSS×T)

[0071] SOUR NOB_MLSS =SOUR MLSS -SOUR MLSS_1

[0072] SOUR AOB_MLSS =SOUR MLSS_1 -SOURMLSS_2

[0073] SOUR 反_MLSS =SOUR MLSS -SOUR MLSS_AOB -SOUR MLSS_NOB

[0074] SOUR MLSS_1 The SOUR value is the result of adding NaClO3 with MLSS as the biomass. MLSS_2 The SOUR value is the result of adding allyl thiourea to MLSS as the biomass.

[0075] Table 2 Microbial respiration rate (SOUR) MLSS Comparison table of the relationship between pollutant removal rate and pollutant removal rate

[0076] Indicator 1 Indicator 2 Correlation <![CDATA[Aerobic tank SOUR NOB_MLSS > COD removal rate r＝0.3,p＝0.007 <![CDATA[Aerobic pond SOUR AOB_MLSS / Aerobic pond SOUR NOB_MLSS > COD removal rate r＝-0.29,p＝0.01 <![CDATA[Anoxic Tank SOUR 反_MLSS / Anoxic Tank SOUR MLSS > COD removal rate r＝0.30,p＝0.008 <![CDATA[Aerobic tank SOUR MLSS > Total nitrogen removal rate 0.22，p＝0.05 <![CDATA[Anoxic Pond SOUR AOB_MLSS / Anoxic Pond SOUR MLSS > Total nitrogen removal rate -0.19,p＝0.09 <![CDATA[Anoxic Tank SOUR NOB_MLSS / Anoxic Tank SOUR MLSS > Total nitrogen removal rate -0.21,p＝0.067 <![CDATA[(Aerobic tank SOUR AOB_MLSS + Aerobic tank SOUR NOB_MLSS ) / Aerobic SOUR MLSS > ammonia nitrogen removal rate 0.23,p＝0.041

[0077] The results showed that SOUR was closely and significantly correlated with COD removal and denitrification in wastewater treatment, proving that microbial activity indicators are expected to indicate and predict wastewater treatment performance.

[0078] (2) Based on ATP calculation of SOUR, the correlation between microbial respiration rate (SOUR) and pollutant removal rate was analyzed. ATP The calculation formulas are as follows (SOUR for aerobic and anoxic tanks) ATP The calculation formulas are all as follows, where T is the time required for the mud-water mixture to fill the breathing chamber at a certain flow rate q.

[0079] SOUR ATP = (DO1-DO2) / (ATP×T)

[0080] SOUR NOB_ATP =SOUR ATP -SOUR ATP_1

[0081] SOUR AOB_ATP =SOUR ATP_1 -SOUR ATP_2

[0082] SOUR 反_ATP =SOUR ATP -SOUR AOB_ATP -SOUR NOB_ATP

[0083] SOUR ATP_1 SOUR is the solubility of oxygen (SOUR) measured after adding NaClO3 with ATP as the biomass. MLSS_2 The SOUR value is the result of adding allyl thiourea with ATP as the biomass.

[0084] Pearson correlation analysis was performed on the above calculation results to analyze the microbial respiration rate (SOUR). ATP The correlation between the pollutant removal rate and the pollutant removal rate is shown in Table 3.

[0085] Table 3 Microbial respiration rate (SOUR) ATP Comparison table of the relationship between pollutant removal rate and pollutant removal rate

[0086] Indicator 1 Indicator 2 Correlation <![CDATA[Aerobic tank SOUR ATP > COD removal rate r = 0.52, p = 1.14e-6 <![CDATA[Aerobic tank SOUR NOB_ATP > COD removal rate r = 0.43, p = 8.67e-5 <![CDATA[Anaerobic pond SOUR 反_ATP > COD removal rate r = 0.47, p = 1.90e-5 <![CDATA[Anoxic tank SOUR ATP > Total nitrogen removal rate 0.27,p＝0.017 <![CDATA[Anoxic pond SOUR 反_ATP > Total nitrogen removal rate 0.23,p＝0.041 <![CDATA[Aerobic tank SOUR ATP > Total nitrogen removal rate 0.24，p＝0.038 <![CDATA[Aerobic tank SOUR AOB_ATP > Total nitrogen removal rate 0.3，p＝0.008 <![CDATA[Anoxic Tank SOUR NOB_ATP / Anoxic Tank SOUR ATP > Total nitrogen removal rate -0.21,p＝0.067 <![CDATA[Anoxic Pond SOUR AOB_ATP / Anoxic Pond SOUR ATP > Total nitrogen removal rate -0.19，p＝0.091 <![CDATA[Aerobic tank SOUR ATP > ammonia nitrogen removal rate 0.57, p = 5.33e-8 <![CDATA[Aerobic tank SOUR AOB_ATP > ammonia nitrogen removal rate 0.41,p＝0.00025 <![CDATA[Aerobic tank SOUR NOB_ATP > ammonia nitrogen removal rate 0.524, p = 1.02e-6

[0087] The results in Table 3 further demonstrate that SOUR is significantly correlated with COD removal and denitrification in wastewater treatment and can be used to predict wastewater treatment results. At the same time, compared with SOUR calculated using MLSS, SOUR calculated using ATP is more strongly correlated with wastewater denitrification, especially ammonia nitrogen removal rate.

[0088] Example 2: Using microbial activity indicators (SOUR) MLSS Predicting wastewater treatment performance

[0089] Random forests are ensemble algorithms that average the results of multiple decision trees. Integrating the results of multiple decision trees effectively reduces the risk of overfitting. Each decision tree is trained on an independent dataset sampled from the original training set. For each tree, input variables are used to generate nodes. These variables are partially randomly selected as subsets at each split, and then at a certain split point, the variable that contributes the least to the sum of the impurities of the two child nodes is selected as the split variable. This process is repeated until the tree no longer splits. The impurity of a node is calculated as follows, where R... m It is the set of sample features of the m-th node, N m It is R m Medium sample size, y i It is x i The response It is R m middle y i The mean of I m For R m The impurity.

[0090]

[0091] Using the data from Example 1, the feature vector x for predicting the ammonia nitrogen removal rate was obtained as follows:

[0092]

[0093] The eigenvector x for the predicted total nitrogen removal rate is obtained as follows:

[0094]

[0095] The feature vector x for predicting COD removal rate is obtained as follows:

[0096] x = [aerobic tank SOUR, aerobic tank SOUR] AOB aerobic tank SOUR NOB aerobic tank SOUR AOB / Aerobic tank SOUR NOB Anoxic pool (SOUR), anoxic pool (SOUR) 反 SOUR (anoxic pool) 反 / Anoxic pool SOUR]

[0097] Using the feature vector x as input to the random forest, the removal rates of ammonia nitrogen and total nitrogen in wastewater are calculated and predicted through the random forest model.

[0098] Simultaneously, the root mean square error (RMSE) and coefficient of determination (R²) are used as indicators. 2 The evaluation criterion is y, and the prediction effect is assessed using the formula shown below. i Indicates the truth value. Indicates the predicted value. The mean of the true values.

[0099]

[0100]

[0101] The results are shown in Table 4.

[0102] Table 4 SOUR MLSS Predicted Wastewater Treatment Performance Comparison Table

[0103]

[0104] Example 3: Using microbial activity indicators (SOUR) ATP Predicting wastewater treatment performance

[0105] Using the data from Example 1, and referring to the method of Example 2, the removal rates of COD, ammonia nitrogen, and total nitrogen were predicted, except that the SOUR... MLSS Replace with the corresponding SOUR ATP Each SOUR ATP The calculation formula is the same as in Example 1. The feature vector x for predicting the ammonia nitrogen removal rate is:

[0106] x = [aerobic tank SOUR, aerobic tank SOUR] AOB aerobic tank SOUR NOB aerobic tank SOUR AOB / Aerobic tank SOUR NOB ]

[0107] The eigenvector x for predicting the total nitrogen removal rate is:

[0108]

[0109] The feature vector x for predicting COD removal rate is:

[0110] x = [aerobic tank SOUR, aerobic tank SOUR] AOB aerobic tank SOUR NOB aerobic tank SOUR AOB / Aerobic tank SOUR NOB Anoxic pool (SOUR), anoxic pool (SOUR) 反 SOUR (anoxic pool) 反 / Anoxic pool SOUR]

[0111] The results are shown in Table 5.

[0112] Table 5 SOUR ATP Predicted Wastewater Treatment Performance Comparison Table

[0113]

[0114] Compared with SOUR MLSS Compared to predictions based on features, predictions based on SOUR... ATP The prediction results are better, especially the accuracy of predictions for ammonia nitrogen and total nitrogen is significantly improved.

[0115] Example 4: Based on SOUR ATP Fusion model predicts wastewater treatment performance

[0116] Using the data from Example 1 and the feature vectors from Example 3, a fusion model algorithm is used to predict wastewater treatment performance. Specifically, the steps include:

[0117] Raw data was collected as described in Example 1 for six consecutive months. The collected raw data was then randomly divided into an 80% training set and a 20% test set. During preprocessing, the `StandScaler()` function from the `sklearn` library in Python was used to normalize all data. Then, individual base learners were trained using the training set data. These base learners included: Support Vector Machine (SVR), K-Nearest Neighbor Regression (KNR), Linear Regression (LR), Decision Tree Model (DT), and Random Forest (RF). These base learners were chosen because their results on this batch of data were relatively similar, making them suitable for fusion. The prediction results from the training data of these base learners were combined to form a new training set, which was then used to train a meta-learner (in this example, linear regression was chosen as the meta-learner).

[0118] The algorithm flow of the fusion algorithm is as follows: Figure 1 As shown in Table 6, K-fold cross-validation is used to divide the training set into K parts. K-1 parts are used to train the base learners, and the remaining part is predicted by each trained base learner. The predictions from each base learner for the K parts form a new feature matrix. Similarly, for the test data, K learneders trained with different base learners are used to predict the test data. The mean of the predicted test data is used as the new feature matrix for the test data. Then, the new feature matrix is ​​used for training and prediction of the meta-learners to obtain the final prediction results for the test data. In this embodiment, K is manually tuned to 15. The prediction results are shown in Table 6.

[0119] Table 6 Comparison of Wastewater Treatment Performance Predicted by Fusion Algorithms

[0120]

[0121]

[0122] The results show that when predicting wastewater treatment performance based solely on microbial activity data, the prediction accuracy for pollutant removal rates, especially ammonia nitrogen removal rates, is relatively high. Model fusion can improve the prediction accuracy to some extent.

[0123] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Any modifications, alterations, substitutions, or variations made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention shall fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for predicting the performance of a contaminant removal from wastewater, characterized by: The method comprises the following steps: (1) Collecting raw data, including: the respiration rate SOUR in the aerobic tank and the anoxic tank AOB and SOUR NOB , SOUR in the aerobic tank, SOUR in the anoxic tank, the respiration rate SOUR of denitrifying bacteria 反 ; (2) obtaining a feature vector x of a predicted ammonia nitrogen removal rate: ； and / or; obtaining a feature vector x of a predicted total nitrogen removal rate: ； and / or; obtaining a feature vector x of a predicted COD removal rate: x = [aerobic tank SOUR, aerobic tank SOUR AOB , aerobic tank SOUR NOB , aerobic tank SOUR AOB / aerobic tank SOUR NOB , anoxic tank SOUR, anoxic tank SOUR 反 , anoxic tank SOUR 反 / anoxic tank SOUR] (3) taking the feature vector x as an input of a random forest, and calculating and predicting a pollutant removal performance in sewage through a random forest model; Each SOUR is calculated based on ATP as biomass, and the SOUR value calculation formula corresponding to the aerobic tank and the anoxic tank is as follows: SOUR= ; SOUR NOB = SOUR-SOUR1 SOUR AOB = SOUR1-SOUR2 SOUR 反 = SOUR-SOUR AOB -SOUR NOB SOUR1 is the SOUR value measured after NaClO3 is added, SOUR2 is the SOUR value measured after propylene thiourea is added, T is the time required for the sludge-water mixture to fill the respiration chamber at a certain flow rate q, DO1 is the dissolved oxygen value in the aerobic tank or the anoxic tank in the sewage treatment system, and DO2 is the dissolved oxygen value in the respiration chamber in the SOUR online detection device.

2. The method of claim 1, wherein: When the dissolved oxygen value in the respiration chamber decreases by 1 / 3, propylene thiourea is added to inhibit the activity of ammonia-oxidizing bacteria AOB, and when the dissolved oxygen value decreases by 2 / 3, NaClO3 is added to inhibit the activity of nitrite-oxidizing bacteria NOB.

3. The method of claim 1, wherein: The method comprises the following steps: (1) Collecting raw data, including: Respiratory rate SOUR in the aerobic tank and the anoxic tank AOB and SOUR NOB , SOUR in the aerobic tank, SOUR in the anoxic tank, Respiratory rate SOUR of denitrifying bacteria 反 ; (2) randomly dividing the data into a training set and a test set; In the preprocessing stage, all data are normalized, and then the training set data are used to train individual basic learners; the prediction results of the basic learner training data are used to form a new training set, and the new training set is used to train the meta-learner; (3) using K-fold cross-validation to divide the training set into K parts, wherein K-1 parts are used to train the basic learners, and the remaining part is predicted by the trained basic learners, and the prediction values of the K parts by each basic learner form a new feature matrix; (4) for test data, different basic learners are used to predict the test data, and the mean value of the prediction results of the test data is used as a new feature matrix of the test data; (5) the new feature matrix is used for training and prediction of the meta-learner respectively, and a prediction result is obtained.

4. The method of claim 3, wherein: The basic learners include support vector machine SVR, K-order nearest neighbor regression KNR, linear regression LR, decision tree model DT and random forest RF.

5. A terminal device, characterized by: The computer readable storage medium stores a program code, and when the program code is run by a processor, the method of any one of claims 1-4 is executed.

6. A computer-readable storage medium, characterized in that: The computer readable storage medium stores a program code, and when the program code is run by a processor, the method of any one of claims 1-4 is executed.

Images