Nitrous nitrogen concentration monitoring method and system based on sensor coupling soft measurement model

Abstract

The invention discloses a nitrite concentration monitoring method based on a sensor coupling soft measurement model. The method comprises the following steps: S1, acquiring target influence factor data; determining a plurality of target influence factors based on the water quality parameters and the nitrification process of the to-be-monitored sewage; s2, establishing a soft measurement model; the input of the soft measurement model is the target influence factor, and the output of the soft measurement model is the concentration of nitrite nitrogen; s3, determining the concentration of nitrite nitrogen; according to the method, the target influence factor is firstly acquired by utilizing the sensor, and the nitrite concentration is calculated or predicted in real time by utilizing the target influence factor, so that the problems that monitoring equipment is expensive, the monitoring range is small, the monitoring is inaccurate and the monitoring equipment cannot directly monitor the nitrite in the seriously polluted wastewater in the nitrite monitoring in the prior art can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of water treatment, and specifically to a method for monitoring nitrite nitrogen concentration based on a sensor-coupled soft measurement model. Background Art

[0002] With the continuous growth of the population, the rapid development of the economy, and the accelerated progress of urbanization, the amount of municipal solid waste generated has shown an increasing trend year by year. Among current waste treatment technologies, landfill is still the main waste disposal method due to its technical maturity, low cost, and simplicity of operation. However, leachate is generated during the treatment process, which has become an environmental problem that urgently needs to be solved.

[0003] The treatment technologies for landfill leachate mainly include biological treatment, physicochemical treatment, and combined treatment. Among these methods, biological treatment is widely used due to its low cost and simple process, including technologies such as activated sludge process, membrane bioreactor (MBR), and sequencing batch reactor (SBR). During the biochemical treatment process, the nitrogen removal effect is affected by various factors such as temperature and influent conditions, and free ammonia (FA) and free nitrous acid (FNA) are the key factors affecting the denitrification effect. Medium and low concentrations of FA and FNA promote the denitrification process, but high concentrations will inhibit the microbial activity and have a negative impact on the nitrification system. Therefore, real-time monitoring of key water quality indicators and dynamically adjusting system parameters to keep FA and FNA within an appropriate concentration range are of crucial significance for ensuring the stable operation of the biological denitrification system.

[0004] Among them, nitrite nitrogen is a key water quality indicator affecting the biochemical system of landfill leachate. Its concentration change is of great significance for indicating the health degree of the system and is an important reference for subsequent regulation. The change of nitrite nitrogen concentration is of great significance for indicating the health degree of the system and is also of important reference significance for subsequent regulation processes such as selecting the carbon source dosage, alkalinity adjustment, sludge supplementation, and anaerobic aeration. Currently, the existing real-time monitoring products for nitrite nitrogen are extremely expensive (hundreds of thousands of yuan per unit) and have a narrow measurement range (about 0 - 10 ppm), and are not applicable to the measurement environment of landfill leachate biochemical tanks or other severely polluted (high-concentration nitrite nitrogen) environments. Summary of the Invention

[0005] To solve the above problems, the present invention provides a method for monitoring nitrite nitrogen concentration based on a sensor-coupled soft measurement model.

[0006] A method for monitoring nitrite nitrogen concentration based on a sensor-coupled soft measurement model includes the following steps:

[0007] S1. Determine multiple target influencing factors based on the water quality parameters and nitrification process of the sewage to be monitored; and obtain the data of the multiple target influencing factors and the concentration data of nitrite nitrogen at multiple historical moments.

[0008] S2. Based on the data of the multiple target influencing factors at the multiple historical moments and the concentration data of nitrite nitrogen, establish a soft-sensing model. The input of the soft-sensing model is the data of the target influencing factors, and the output of the soft-sensing model is the concentration of nitrite nitrogen.

[0009] S3. Obtain the data of the target influencing factors in the sewage at the target moment, and then determine the concentration data of nitrite nitrogen at the target moment based on the soft-sensing model; the target moment includes the current moment or a future moment.

[0010] Note: Through the above method, the concentration of nitrite nitrogen in severely polluted wastewater can be obtained in real time. By using the soft-sensing model, the damage to equipment caused by highly polluted wastewater can be avoided, and the monitoring cost is greatly reduced. By obtaining the data of the target influencing factors through sensors, the cost of data acquisition can be further reduced, and the target influencing factors can be used as inputs to calculate the concentration of nitrite nitrogen more accurately through the model.

[0011] Further, in S1, the method for determining multiple target influencing factors based on the water quality parameters and nitrification process of the sewage to be monitored includes:

[0012] Obtain the data of multiple water quality parameters that affect nitrous acid production. The multiple water quality parameters include NO3--N, NH4+-N, COD, temperature, pH, DO, ORP, total nitrogen, suspended solids, and TOC.

[0013] Screen out multiple water quality parameters that affect the nitrification process of the sewage.

[0014] Perform a multicollinearity analysis on the multiple water quality parameters that affect each nitrification process, and select the water quality parameters with a variance inflation factor not exceeding the standard in the multicollinearity analysis as the target influencing factors.

[0015] Further, the target influencing factors include temperature, pH, DO, ORP, NO3 - -N and NH4 + -N.

[0016] Note: Through the above method for determining multiple target influencing factors, water quality data that has a greater impact on nitrous acid and does not have a multicollinearity problem can be screened out as target influencing factors.

[0017] Further, the method for establishing the soft-sensing model in S2 includes:

[0018] Obtain the data of the multiple target influencing factors and the concentration data of nitrite nitrogen at multiple historical moments to form a data set, and divide the data set into a training set and a test set;

[0019] Based on the data of the training set, using RMSE as the loss function, use a variety of machine learning algorithms to build models, and obtain a variety of algorithm models, including RF model, GBDT model, XGBOOST model, MLP model, and Resnet model;

[0020] Based on a variety of algorithm models, test the data input to the test set, and select an algorithm model with the smallest error in the test results as the soft sensor model.

[0021] Note: Through the above method of establishing a soft sensor model, a model that can determine a more appropriate corresponding relationship between the target influencing factor and nitrite nitrogen can be determined.

[0022] Furthermore, before the test, use the Bayesian optimization algorithm to optimize the key hyperparameters in each algorithm model. The specific operation process of the optimization process includes:

[0023] 1). Preliminary pre-experiment; Manually adjust the range of key hyperparameters and observe the change of model performance to initially determine some hyperparameters and the hyperparameter optimization space;

[0024] 2). Bayesian optimization; Use the Bayesian optimization algorithm to set 30 groups of hyperparameter combinations for iterative testing within the specified hyperparameter range to obtain each algorithm model after optimization processing.

[0025] Note: Through the above method, the accuracy of model operation can be improved.

[0026] Furthermore, in S3, the method for determining the concentration data of nitrite nitrogen at the target moment based on the soft sensor model is:

[0027] When the target moment is the current moment, it is determined by using the data of the target influencing factor at the current moment or the data of the target influencing factor at the historical moment;

[0028] Among them, when determining by using the data of the target influencing factor at the current moment, the data of the target influencing factor at the current moment is used as the input of the soft sensor model for model calculation to obtain the concentration data of nitrite nitrogen at the current moment;

[0029] When determining using the data of the target influencing factors at historical moments, the variation laws of multiple groups of the target influencing factors in a historical time period are used to predict the data of the target influencing factors at the current moment, and then the predicted data of the target influencing factors at the current moment is used as the input of the soft measurement model for model calculation to obtain the concentration data of nitrite nitrogen at the current moment.

[0030] Explanation: Through the above method, it is possible to not only obtain the concentration data of nitrite nitrogen through real-time monitoring, but also infer the concentration data of nitrite nitrogen using historical data. When there are sudden situations in sensors or processes, prediction can be carried out so that researchers can master the operating status of the sewage treatment process.

[0031] Furthermore, when the target moment is a future moment, according to the variation laws of multiple groups of the target influencing factors at the current moment or historical moments, the data of the target influencing factors at the future moment is predicted, and then the predicted data of the target influencing factors at the future moment is used as the input of the soft measurement model for model calculation to obtain the concentration data of nitrite nitrogen at the future moment.

[0032] Explanation: The above method can realize the prediction of the concentration data of nitrite nitrogen at future moments, predict the water quality change trend, make the sewage treatment process more intelligent and automated, significantly reduce the system maintenance cost and operation complexity, and make the daily operation of the system more simple and economical.

[0033] Furthermore, the method for predicting the data of the target influencing factors at a future moment or the current moment according to the variation laws of multiple groups of the target influencing factors at the current moment or historical moments includes:

[0034] First, according to the variation laws of multiple groups of the target influencing factors at the current moment or historical moments, multiple working conditions in the biochemical treatment of high-ammonia-nitrogen wastewater are defined; among them, the multiple working conditions include the short-cut nitrification-denitrification stage working condition, the nitrification reaction inhibited by FA stage working condition, and the nitrification reaction co-inhibited stage working condition;

[0035] Based on the multiple working conditions in the biochemical treatment of high-ammonia-nitrogen wastewater, determine the operating range of the target influencing factor values in each working condition and the fluctuation law of the target influencing factor in each working condition;

[0036] According to the operating range and the fluctuation law, determine the data of the target influencing factors at the future moment or the current moment.

[0037] Explanation: The above method can provide a specific method for prediction, making the prediction result more accurate.

[0038] Furthermore, before forming the data set, data correction is performed on the acquired data of the target impact factor at multiple moments; the data correction method includes:

[0039] Fitting the data of each of the multiple target influencing factors respectively to obtain a fitting equation for each target influencing factor;

[0040] The fitting equation is used to define the numerical range of the target influencing factor, and the data of the target influencing factor that exceeds the numerical range is eliminated.

[0041] Note: The above correction method can eliminate abnormal values ​​generated by sensor measurements, thereby improving data accuracy.

[0042] The present invention also provides a nitrite nitrogen concentration monitoring system based on a sensor coupling soft measurement model, wherein the nitrite nitrogen concentration monitoring system comprises a preprocessing module, a collection module and a calculation module;

[0043] The pretreatment module is used to pre-treat the sewage to be treated;

[0044] The collection module is used to collect the target influencing factors in the pre-treated sewage using sensors;

[0045] The calculation module is used to calculate the target influencing factor obtained by the collection module to obtain the nitrite nitrogen concentration;

[0046] The pretreatment module includes a filtration unit and a dilution unit; the filtration unit uses a ceramic flat membrane with a pore size of 0.5 μm for filtration, and the operation method of the filtration unit is to perform air-water backwashing once every 5 minutes of filtration.

[0047] Note: Through the setting of the above monitoring system, the wastewater to be treated can be systematically treated and pre-treated to prevent the suspended matter and other impurities in the water from damaging the sensor. In addition, the experiment found that the use of the above filtration unit can have little effect on the key influencing parameter factors in the water body, and almost no effect on various water quality parameters. The use of this system can effectively improve the measurement accuracy of the sensor and reduce the frequency of failure of the measuring instrument. It can adapt to different types of sewage treatment environments, and can adapt to harsh sewage treatment environments, and maintain high-precision monitoring capabilities.

[0048] Furthermore, the sewage to be treated is landfill leachate.

[0049] The beneficial effects of the present invention are:

[0050] The present invention solves the problems that are difficult for real-time monitoring, such as high cost of nitrite nitrogen monitoring equipment and narrow measurement range. By combining advanced machine learning algorithms with traditional sensor monitoring technologies, the present invention has developed an innovative online monitoring system. The present invention covers the sensor design of the system, precise error correction methods, and soft measurement technologies based on the Resnet model. The monitoring system of the present invention can not only monitor key influencing parameter factors in real time, but also predict the water quality change trend through a machine learning model. This prediction ability is not available in traditional monitoring technologies, which makes the sewage treatment process more intelligent and automated, significantly reduces the maintenance cost and operation complexity of the system, and makes the daily operation of the system more simple and economical. In the present invention, a correction method is adopted to correct the measurement results of the sensor through a specific algorithm, ensuring the accuracy of the data. Description of the Drawings

[0051] Figure 1 is the flow chart of the monitoring method of the embodiment of the present invention;

[0052] Figure 2 is the data graph of the membrane flux recovery of different cleaning methods for the ceramic flat membrane in the filtration unit in the embodiment of the present invention;

[0053] Figure 3 is the flux attenuation of the ceramic flat membrane in the filtration unit after different cleaning methods in the embodiment of the present invention;

[0054] Figure 4 is the flow chart of the monitoring system in the embodiment of the present invention;

[0055] Figure 5 is the soft measurement model construction process of NO 2- -N in the embodiment of the present invention;

[0056] Figure 6 is the comparison of prediction errors based on different algorithm models in the embodiment of the present invention;

[0057] Figure 7 is the flow chart of the landfill leachate treatment process in the embodiment of the present invention;

[0058] Figure 8 is the sensor coupling online monitoring device of the monitoring system in the embodiment of the present invention;

[0059] Figure 9 is the Pearson correlation between different water quality parameters in the embodiment of the present invention;

[0060] Figure 10 is the accuracy test data graph of the COD sensor in actual water samples in the embodiment of the present invention, where (a) represents the distribution of the comparison test data of the actual water samples, and (b) represents the relative error distribution of the actual water sample test data;

[0061] Figure 11 It is the error correction data diagram of the COD sensor in the embodiment of the present invention, where (a) represents the correction curve and (b) represents the residual distribution of the ammonia nitrogen test data after correction;

[0062] Figure 12 It is the accuracy test data diagram of the nitrate nitrogen sensor in the actual water sample in the embodiment of the present invention, where (a) represents the distribution of the comparison test data of the actual water sample and (b) represents the relative error distribution of the test data of the actual water sample;

[0063] Figure 13 It is the error correction data diagram of the ammonia nitrogen sensor in the embodiment of the present invention, where (a) represents the correction curve and (b) represents the residual distribution of the ammonia nitrogen test data after correction;

[0064] Figure 14 It is the accuracy test data diagram of the ammonia nitrogen sensor in the actual water sample in the embodiment of the present invention, where (a) represents the distribution of the comparison test data of the actual water sample and (b) represents the relative error distribution of the test data of the actual water sample;

[0065] Figure 15 It is the online monitoring data change diagram of the ammonia nitrogen sensor in the embodiment of the present invention;

[0066] Figure 16 It is the online monitoring data change diagram of the ammonia nitrogen sensor directly put into the SBR-O tank in the embodiment of the present invention;

[0067] Figure 17 It is the online monitoring data change diagram of the nitrate nitrogen sensor in the embodiment of the present invention;

[0068] Figure 18 It is the online monitoring data change diagram of the COD sensor in the embodiment of the present invention;

[0069] Figure 19 It is the operation cycle of the SBR reactor in the embodiment of the present invention;

[0070] Figure 20 It is the SBR experimental device in the experimental example of the present invention; Figure 21 It is the online monitoring data change diagram of the nitrate nitrogen sensor directly put into the SBR-O tank in the embodiment of the present invention. Specific implementation manners

[0071] To further elaborate the methods adopted and the effects achieved by the present invention, the technical solutions of the present invention will be clearly and completely described below in combination with experiments.

[0072] In the biochemical treatment process of high-concentration ammonia nitrogen wastewater such as landfill leachate and highly polluted wastewater, denitrifying microorganisms often face the dual inhibition of high free ammonia (FA) and high free nitrous acid (FNA), which poses a significant threat to their normal growth and metabolic activities. To ensure the stable operation of the denitrification system, it is particularly crucial and necessary for the staff to monitor the key water quality parameters in the biochemical system in real time and take corresponding control measures in a timely manner based on these parameters.

[0073] However, in the current biochemical treatment process of sewage treatment plants, the instruments used for on-line monitoring have certain limitations in measurement accuracy and long-term stability. In addition, due to technical limitations, these instruments cannot directly and accurately measure nitrite nitrogen in water bodies. Specifically, the monitoring instruments are mainly divided into two categories: one is large-scale on-line analyzers, and the other is sensors based on ion-selective electrode method and spectroscopy. Large-scale analyzers are known for their high accuracy, but their high cost limits their application scope and makes them more suitable for specific monitoring tasks; relatively speaking, sensors have lower costs and are more suitable for full-process monitoring. However, in a complex environment, sensors may be interfered, so regular maintenance work is required. Although the on-line water quality monitoring technology is quite mature in some fields, in the biochemical pool of sewage treatment plants, this technology is still in the stage of research and exploration.

[0074] In addition, in the current field of landfill leachate treatment, the monitoring and control of the biochemical treatment process are the key links to ensure the treatment effect and efficiency. Traditional monitoring methods rely on manual sampling and laboratory analysis. Although this method can provide relatively accurate data, it is not only time-consuming and laborious, but also unable to achieve real-time monitoring and continuous on-line monitoring of the treatment process, resulting in the inability to detect and adjust abnormal situations in the treatment process in a timely manner.

[0075] With the development of sensor technology, some studies have attempted to apply on-line monitoring devices such as dissolved oxygen sensors and pH sensors to the monitoring of the leachate treatment process. However, these attempts still have many limitations. For example, due to the particularity of high ammonia-nitrogen wastewater, conventional sensors often cannot adapt to its harsh environmental conditions, are easily contaminated and damaged, and it is difficult to ensure the stability and accuracy of sensors in complex environments. Moreover, there is a lack of effective data processing and analysis methods to improve the reliability of monitoring results. In existing technical solutions, some studies have attempted to improve the monitoring accuracy by improving sensor technology. For example, using electrochemical sensors or spectroscopic analysis technology still cannot effectively solve the monitoring problems in high FA and FNA environments, and the cost is relatively high, making it difficult to be widely applied in actual projects. There are also studies that have attempted to improve the monitoring accuracy by improving sensor technology. For example, some studies have extended the service life of sensors by adding a protective layer to the sensors, or adopted specific cleaning and calibration procedures to improve their stability. However, these technologies often only target a single monitoring index, cannot comprehensively reflect the state of the entire treatment process, and cannot solve the problems of long-term stability and accuracy of sensors in high ammonia-nitrogen environments.

[0076] In summary, there is still a lack of a technical solution in the current market that can effectively monitor the key water quality parameters (nitrite nitrogen) during the treatment of high ammonia-nitrogen wastewater. Therefore, in view of this situation, this study aims to explore a new method, that is, by combining sensor technology with machine learning algorithms, to develop a soft-sensing model to achieve on-line monitoring of key water quality indicators in the biochemical treatment process of landfill leachate. The present invention provides a method for monitoring nitrite nitrogen concentration based on a sensor-coupled soft-sensing model to improve the stability and efficiency of the entire denitrification system. This method not only has the potential to improve the monitoring accuracy, but also can provide more stable and reliable monitoring data for sewage treatment plants, thus ensuring the efficient and stable operation of the entire denitrification system. Through this method, an innovative solution can be provided for the treatment of high-concentration ammonia-nitrogen wastewater to address the challenges faced by current technologies.

[0077] Example 1: Combining the above content, the following is the specific solution in the embodiment of the present invention: As Figure 1 shown, a method for monitoring nitrite nitrogen concentration based on a sensor-coupled soft-sensing model includes the following steps:

[0078] S1. Based on the water quality parameters of the sewage to be monitored and the nitrification process, determine multiple target influencing factors; and obtain data of the multiple target influencing factors and nitrite nitrogen concentration data at multiple historical moments;

[0079] Specifically, the method for obtaining the target impact factor data includes collecting existing data or collecting data through sensors; the method for obtaining nitrite nitrogen includes obtaining it from sewage in a laboratory (for example, manually measuring it using the N-(1-Naphthyl)ethylenediamine dihydrochloride spectrophotometric method).

[0080] In S1, the method for determining multiple target impact factors based on the water quality parameters and nitrification process of the sewage to be monitored includes the following S1-1 to S1-3:

[0081] S1-1. Obtain the data of multiple water quality parameters that affect nitrite production. The multiple water quality parameters include NO 3- -N, NH4 + -N, COD, temperature, pH, DO, ORP, total nitrogen, total phosphorus, suspended solids, and TOC;

[0082] S1-2. Screen out multiple water quality parameters that affect the nitrification process of the sewage;

[0083] Specifically, as a typical biochemical process, the nitrification process is mainly carried out by AOB and NOB bacteria. Temperature is one of the main factors affecting the activity of nitrifying bacteria. Therefore, there must be a great correlation between the accumulation amount of nitrite nitrogen and temperature during the nitrification process, and it is selected as a water quality parameter for the soft measurement model of nitrite nitrogen. Secondly, pH, DO, and ORP are important indicators reflecting the progress of the nitrification process. During the nitrification reaction process, nitrifying bacteria will continuously consume alkalinity, resulting in a decrease in pH, and when the nitrification process ends, the pH will rise again, thus forming an "ammonia valley". And DO is even more one of the important factors affecting the nitrification process. Nitrifying bacteria are aerobic bacteria and require a large amount of oxygen during the nitrification process. ORP will also gradually increase during the nitrification process and will decrease during the denitrification process. The decrease process will also show a characteristic point of "nitrite knee point" indicating the end of the denitrification process. Therefore, pH, DO, and ORP are selected as water quality parameters for the soft measurement model of nitrite nitrogen. Finally, the concentration changes of NH4 + -N and NO 3- -N are directly indicators reflecting the nitrification degree, and there is a direct relationship between their concentration changes and the concentration changes of NO 2- -N. Therefore, NH4 + -N and NO 3- -N are selected as water quality parameters for the soft measurement model of nitrite nitrogen. Since the biodegradability of COD in landfill leachate is poor and the nitrification process is an autotrophic process, COD is included in the initially selected water quality parameters.

[0084] S1-3. Conduct a multicollinearity analysis among the multiple water quality parameters that affect each nitrification process, and select the water quality parameters with a variance inflation factor not exceeding the standard in the multicollinearity analysis as the target impact factors.

[0085] It is understandable that the problem of multicollinearity refers to the high linear correlation between two or more target influencing factors due to improper variable selection. When multicollinearity occurs, it will affect the stability and interpretability of the model, making the estimation of model parameters unstable and inaccurate, resulting in significant changes in the estimation results even for small changes or errors.

[0086] Multicollinearity analysis method:

[0087] To explore the problem of multicollinearity among target influencing factors, this study also used the variance inflation factor (VIF) to measure the strength of multicollinearity between independent variables. The VIF analysis results among water quality parameters in this study are shown in Table 1. ORP, T, NH4 + The VIF values of the three water quality parameters are the highest, which are 4.43, 3.37, and 3.19 respectively, which is consistent with the results of Pearson correlation coefficient analysis. It can be seen that although the Pearson correlation coefficients among these three indicators are strong, according to the VIF analysis results, the multicollinearity relationship among these indicators does not reach the level that needs to be excluded (i.e., not exceeding the standard, VIF > 10). Therefore, it can be considered to input all the above six water quality parameters as target influencing factors into the model.

[0088] Table 1 VIF results among different input indicators

[0089] Index T pH DO ORP NH4+ NO3- VIF 3.37 3.13 1.67 4.43 3.19 3.08

[0090] Pearson correlation analysis verification:

[0091] The collected data is sorted into the original dataset in the order of collection time. The summary statistics of the maximum value, minimum value, average value, quartiles, and standard deviation of each target influencing factor are shown in Table 2.

[0092] Table 2 Summary statistics table of water quality parameter data

[0093] Index Unit Maximum value Minimum value Average value Median T ℃ 26.69 16.26 24.92 25.91 pH - 9.31 6.09 8.29 8.53 DO mg / L 14.48 0.00 5.55 5.88 ORP mv 278 -618 -83.20 -36.5 <![CDATA[NH4 + > mg / L 268.21 6.63 97.88 86.54 <![CDATA[NO3 - > mg / L 97.4 4.30 35.31 28.95

[0094] The Pearson correlation coefficient method is used to analyze the correlation between variables in the dataset to avoid the occurrence of collinearity problems. The calculation results are as Figure 9 shown.

[0095] Among the input variables, NO3 -There is a strong negative correlation with pH, and the Pearson correlation coefficient reaches -0.71. In addition, the Pearson coefficients of ORP with T and pH also reach 0.68 and -0.66 respectively, showing a strong correlation. The Pearson correlation coefficients of NH4+ with T and ORP are -0.62 and -0.69 respectively, showing a strong negative correlation. Generally speaking, although there is a certain degree of linear correlation among the variables, it is not very obvious.

[0096] The target influencing factors include temperature, pH, DO, ORP, NO3 - -N and NH4 + -N.

[0097] To sum up, this study selects the real-time concentrations of temperature T, pH value, redox potential ORP, dissolved oxygen DO, NH4 + , NO3 - as the target influencing factors, a total of 6 water quality parameters.

[0098] S2. Establish a soft sensor model;

[0099] Based on the data of multiple target influencing factors at multiple historical moments and the concentration data of nitrite nitrogen, establish a soft sensor model. The input of the soft sensor model is the data of the target influencing factors, and the output of the soft sensor model is the concentration of nitrite nitrogen;

[0100] The method for establishing the soft sensor model in S2 includes the following S2-1 to S2-3: as Figure 5 shown,

[0101] S2-1. Obtain the data of the multiple target influencing factors and the concentration data of nitrite nitrogen at multiple historical moments to form a data set, and divide the data set into a training set and a test set;

[0102] Exemplarily, 330 pieces of collected original data are divided according to the ratio of 8:2, where 80% is used as model training data and 20% is used as the test set to test the generalization ability of the model on unknown data sets;

[0103] S2-2. Use the data in the training set, take RMSE as the loss function, and use a variety of machine learning algorithms for modeling to obtain a variety of algorithm models. The variety of algorithm models include RF model, GBDT model, XGBOOST model, MLP model, and Resnet model;

[0104] Use the Bayesian optimization algorithm to optimize the key hyperparameters in each algorithm model; the specific operation process of the optimization process includes:

[0105] 1), Preliminary pre-experiment: By manually adjusting the range of key hyperparameters and observing the changes in model performance, some hyperparameters and the hyperparameter optimization space are initially determined;

[0106] 2), Bayesian optimization: Using the Bayesian optimization algorithm, 30 sets of iterative tests of hyperparameter combinations are set within the specified hyperparameter range to obtain each algorithm model after optimization processing.

[0107] S2-3, Based on multiple algorithm models, the data of the test set is input for testing, and the model corresponding to the machine learning algorithm with the smallest error in the test results is selected as the soft sensor model.

[0108] Exemplarily, the error of the soft sensor model on the test set is evaluated, and three model evaluation metrics are used: root mean square error (RMSE), coefficient of determination (R 2 ), and mean absolute error (MAE), so as to determine the optimal modeling algorithm. The results are as Figure 6 shown. From R 2 viewpoint, Resnet and XGBOOST are the best; Continuing to observe the two evaluation metrics of MAE and RMSE, it can be found that the Resnet model is superior to the other four models. The MAE of the Resnet model is 12.93, while the MAEs of the four models of RF, GBDT, XGBOOST, and MLP are 15.0, 19.26, 14.12, and 13.91 respectively. The MAE of Resnet is reduced by about 10-30% compared with the other four models. The RMSE of the Resnet model is 17.13, and the RMSEs of the other four models are 22.67, 26.64, 18.21, and 22.07 respectively, with a reduction of 5%-40% compared with the other four models.

[0109] This shows that under the same tuning conditions, the learning ability of the Resnet model is significantly higher than that of RF, GBDT, XGBOOST, and MLP. And the MAE is 12.93. From the relationship between MAE and absolute error, it can be known that the average absolute error of the model is less than 12.93 mg / L, and the range of nitrite nitrogen concentration on the test set is within 0-250 mg / L. This shows that the full-scale error of the model on the test set is less than 5%. Therefore, choosing the Resnet model as the soft sensor for nitrite nitrogen meets the accuracy requirements for regulation.

[0110] Furthermore, before forming the data set, data correction is performed on the data of the target influencing factors at multiple moments obtained by the sensor; the data correction method includes:

[0111] Fitting the data of each of the multiple target influencing factors respectively to obtain the fitting equation of each target influencing factor;

[0112] Using the fitting equation (the fitting equation curve is obtained by linear fitting with the least squares method), the numerical range of the target influencing factor is delimited, and the data of the target influencing factor exceeding the numerical range are excluded.

[0113] Exemplarily, taking ammonia nitrogen, nitrate nitrogen, and COD as the target influencing factors, the data correction is as follows:

[0114] (1) Determination of the actual usage range of the ammonia nitrogen sensor, correction of the measurement error, and evaluation of the monitoring results

[0115] Take the water in the SBR A tank and the SBR O tank, and obtain a series of water samples within different ammonia nitrogen concentration ranges by means of diluting the ammonia nitrogen concentration with tap water and increasing the ammonia nitrogen concentration by adding ammonium chloride. After the water samples are filtered through a 0.5 μm ceramic flat membrane, they are then measured by the sensor. A total of 100 sets of comparative test data of the ammonia nitrogen sensor in actual water samples are measured, and the measurement results are as Figure 14 shown. It can be clearly seen from Figure 14 (a) that although there is a large deviation between the measured value of the ammonia nitrogen sensor and the manual measurement value using the standard analysis method, there is a great positive correlation between the two, which indicates that technical means such as error correction can be used to correct the measured value, thereby improving the measurement accuracy. In addition, it can also be found from Figure 14 (b) that as the ammonia nitrogen concentration in the mixed liquid increases, the measurement deviation of the sensor also increases. In the range where the measured value of the sensor is greater than 200, the relative error of a large amount of data exceeds 30%, while in the range where the indication value of the sensor is [0, 200], the relative error is basically distributed within ±30%. This relative error is still very large and far from meeting the requirement of 15% specified in HJ355-2019. In order to obtain more accurate original data, [0, 200] is selected as the measurement range of the ammonia nitrogen sensor. In subsequent experiments, all measurements of the ammonia nitrogen concentration using this sensor are carried out within this range. Then, on this basis, the measured value of the sensor is further corrected for error. In order to obtain more accurate original data, [0, 200] mg / L is selected as the measurement range of the ammonia nitrogen sensor. In subsequent experiments, all measurements of the ammonia nitrogen concentration using this sensor are carried out within this range. Then, on this basis, the measured value of the sensor is further corrected for error.

[0116] Select the test data with sensor readings within 200 from 100 groups of original datasets. There are a total of 70 groups. Divide 54 groups of data according to the ratio of 8:2 for mathematical modeling and 16 groups of data as the validation set. The data selected as the validation set is extracted from the original dataset by stratified sampling to ensure that it can cover the entire measurement range. Since it can be clearly found that the actual concentration in the water sample and the test concentration of the sensor show a unary linear regular distribution, Excel is used to perform regression analysis on these 54 groups of data. The regression analysis results are as Figure 13 shown. From Figure 13 (a), it can be seen that the regression analysis results are very good, and the R 2 reaches 0.96. To further explore the generalization ability of the obtained correction curve, it is also necessary to look at its performance on the unknown dataset. It can be clearly seen from Figure 13 (b) that after the curve correction, the relative errors of the ammonia nitrogen concentrations measured by the sensor are basically distributed within the range of ±15%, which meets the requirement of a relative error less than 15% specified in HJ 355-2019.

[0117] In summary, the measurement range of the ammonia nitrogen sensor is determined to be within [0, 200] mg / L, and the correction formula is corrected using the equation obtained from Figure 13 (a), and the obtained corrected value is used as the final test value of the ammonia nitrogen sensor. All subsequent data measured using the ammonia nitrogen sensor will be measured within this range, and the above-obtained correction formula will be used for correction.

[0118] Evaluation of the monitoring results of the ammonia nitrogen sensor

[0119] Select the actual water samples during the daily decantation of the SBR-O tank from June 17 to July 19, 2024, and analyze the comparison test results. The results are as Figure 15 shown. First, observing the direct measurement results of the sensor, the sensor test correction results, and the laboratory analysis measurement results, it can be found that the change trends of the three are basically the same, indicating that the measurement performance of the sensor is good. However, it is also found that the sensor test correction results are closer to the laboratory analysis results, indicating that the ammonia nitrogen sensor error correction equation explored in the third chapter is effective and can improve the measurement accuracy of the sensor to a certain extent.

[0120] Continue to observe Figure 15In the scatter plot of the relative error distribution, it is found that the relative error between the direct measurement result of the sensor and the laboratory analysis measurement result is roughly in the range of 20% - 40%, while the relative error of the corrected test result of the sensor is distributed within the range of ±20%. This further shows that the correction equation can effectively improve the measurement accuracy of the sensor. However, it should be noted that even after correction, there are still a large number of samples with relative errors distributed in the range of 15% - 20%, showing a gap of about 5% compared with the result in Chapter 3 where the relative error can be made within 15% after being corrected by the correction equation. It is speculated that the reasons for this gap are as follows: (1) There are differences in the physicochemical properties of the water samples used to construct the correction equation and the water samples in the current production process; (2) After long-term use, the measurement characteristics of the sensor have shifted; (3) The measurement errors of different testers are different. Since the object of this study is process monitoring, a relative error of 20% is still within an acceptable range. Figure 16 For the change of the online monitoring data of the ammonia nitrogen sensor directly put into the SBR-O pool for one week, the red line indicates that the sensor was cleaned and calibrated on the 5th day after operation. In the first 5 days, the relative error between the direct test data of the sensor, the corrected test data and the laboratory analysis value showed a continuous upward trend, and the relative error on the 4th and 5th days had exceeded 40%. After cleaning and calibration, the relative error of the sensor measurement dropped to around 20%, and the measurement performance returned to normal. Looking back at Figure 15 the change of the overall relative error in, it is found that there is no such significant continuous upward trend of the relative error. This indicates that the pollutants attached to the ammonia nitrogen sensor may affect the ion exchange membrane, thus interfering with the measurement performance of the sensor. By comparing the ammonia nitrogen sensor placed in the online monitoring system after 30 days of use and directly placed in the SBR-O pool after 5 days of use, it can be very clearly found that there are great differences between the two. There are almost no traces of fouling on the surface of the ammonia nitrogen sensor put into the online monitoring system, while the sensor probe directly put into the SBR-O pool is covered with sludge all over. This is because the online monitoring system has filtration and dilution units, so that the sensor is in a measurement environment with better water quality conditions. Even after long-term operation, it is difficult for the surface of the sensor to be fouled; while the water quality conditions in the biochemical pool are poor, and due to the high temperature in summer, the growth rate of microorganisms is extremely fast, and the surface of the sensor will be covered by the growth of attached sludge after short-term operation.

[0121] In summary, using the error correction equation to correct the direct measurement results of the ammonia nitrogen sensor is a simple and effective method to improve the measurement accuracy of the sensor. In addition, compared with directly putting the ammonia nitrogen sensor into the aeration tank, using it under the online monitoring system designed in this study can very effectively avoid the contamination of the sensor probe, which may lead to a sharp decline in the measurement accuracy, and greatly reduce the maintenance frequency of the sensor.

[0122] (2) Determination of the actual measurement range of the nitrate nitrogen sensor and evaluation of its monitoring results;

[0123] Samples from the SBR-A tank and the O tank were taken, and a series of water samples with different nitrate nitrogen concentration ranges were obtained by diluting the nitrate nitrogen concentration with tap water and increasing the nitrate nitrogen concentration by adding sodium nitrate. After passing through a 0.5-μm ceramic flat membrane filter, the water samples were measured by the sensor... Since the measurement accuracy of the nitrate nitrogen sensor was found to be good during the test, a total of 60 groups of actual water sample comparison test data were measured, and the results are as Figure 11 shown. As can be seen from Figure 12 (a), the measurement performance of the nitrate nitrogen sensor is somewhat similar to that of the ammonia nitrogen sensor. That is, the measurement accuracy is better when the concentration of this index in the water body is low, and then as the concentration of this water quality index increases, the measurement accuracy gradually deteriorates. However, it is also very different from the ammonia nitrogen sensor. As can be seen from Figure 12 (b), the test errors of the nitrate nitrogen sensor are basically distributed within plus or minus 30%, and when the measured value of the sensor is in the range of [0, 250], the vast majority of the test errors are distributed within plus or minus 15%, basically meeting the requirement of the relative error less than or equal to 15% specified in HJ355-2019. Therefore, [0, 250] mg / L is selected as the measurement range of the nitrate nitrogen sensor. In subsequent experiments, all measurements of the nitrate nitrogen concentration using this sensor are carried out within this range.

[0124] Evaluation of the monitoring results of the nitrate nitrogen sensor

[0125] Similarly, the comparison test results of the actual water samples during the daily decantation of the SBR-O tank from June 17 to July 19, 2024 were selected for analysis, and the results are as Figure 17As shown in the figure. It can be found that the changing trend of the direct measurement value of the sensor is basically consistent with the results of laboratory analysis and measurement, indicating that the basic measurement performance of the nitrate nitrogen sensor is good during the entire monitoring process. Continuing to observe the change of the relative error between the sensor measurement value and the laboratory analysis and detection value, it is found that the relative error basically fluctuates around 20% during the entire monitoring period, slightly larger than the direct measurement error of about 15% of the nitrate nitrogen sensor determined in Chapter 3. It is speculated that the reason for this increase in relative error is the same as that of the ammonia nitrogen sensor described above. Since the ammonia nitrogen sensor electrode and the nitrate nitrogen sensor electrode are on the same sensor probe, the comparison and change situation of the nitrate nitrogen sensor placed in the on-line monitoring system after 30 days and directly placed in the SBR-O tank for 5 days is the same as that of the ammonia nitrogen sensor, and the nitrate nitrogen sensor directly placed in the SBR-O tank has suffered greater fouling. Figure 21 The figure shows the change of on-line monitoring data of the nitrate nitrogen sensor directly put into the SBR-O tank for one week. The red line indicates that the sensor is cleaned and calibrated on the 5th day of operation. It can be found that within the first 5 days, the relative error gradually increases, and the relative error approaches 30% on the 5th day. After re-cleaning and calibration on the 6th day, the measurement error returns to normal. This shows that when the nitrate nitrogen sensor is directly placed in the SBR-O tank, due to the complex and harsh environment and the blocking effect of pollutants on the electrode membrane, the measurement performance of the sensor will show a faster deviation. Continuing to observe Figure 21 the change of the relative error in the figure, it is found that there is no such drastic fluctuation, indicating that it is helpful for the nitrate nitrogen sensor to maintain its measurement performance when monitoring in a relatively mild measurement environment.

[0126] In summary, when the nitrate nitrogen sensor is used under the on-line monitoring system designed in this study, it can very effectively avoid the contamination of the sensor probe, effectively prevent the sharp decline of the sensor measurement accuracy, and achieve relatively accurate measurement for a long time.

[0127] (3) Determination of the actual use range of the COD sensor, correction of measurement error and evaluation of its monitoring effect;

[0128] Take the water samples from the SBR-A tank and the SBR O tank, and obtain a series of water samples with different COD concentration ranges by gradient dilution with tap water. After the water samples are filtered through a 0.5μm ceramic flat membrane, they are then measured by the sensor. A total of 100 groups of actual water sample comparison experiment data are measured, and the measurement results are as Figure 10 shown. From Figure 10It can be seen from (a) that the measurement deviation of the COD sensor is very large. However, in terms of the absolute error, its distribution pattern is also consistent with that of the ammonia nitrogen sensor and the nitrate nitrogen sensor. When the COD concentration in the water body is relatively high, the absolute deviation of the measured value of the COD sensor is relatively small. However, as the COD concentration in the water body increases, the measurement deviation of the COD sensor also becomes larger and larger. From Figure 10 It can be seen from (b) that the relative errors of the test data are basically all greater than 30%, which is far from the 15% specified in HJ355-2019. Moreover, as the COD concentration in the water body increases, the relative error also becomes larger. Although the relative errors of the measured data are very large, from Figure 10 It can be seen from (a) that the data distribution shows a strong linear pattern. Therefore, it is considered to perform regression modeling on it. The data is split into 77 groups for modeling according to a ratio of approximately 8:2; 23 groups are used as the test set to verify the effectiveness of the model. The data split out as the validation set is sampled from the original dataset in a stratified sampling manner to ensure that it can cover the entire measurement range. Excel is used for regression analysis to obtain the regression curve as shown in Figure 11 . It can be seen from Figure 11 (a) that the curve fitting effect is very good, and R2 reaches 0.98. To further explore the generalization ability of the obtained correction curve, it is also necessary to look at its performance on the unknown dataset. It can be clearly seen from Figure 11 (b) that after being corrected by the regression curve, the relative error is significantly reduced compared with that before correction. Moreover, the errors after correction are basically distributed within ±15%, meeting the requirements specified in HJ355-2019. The maximum range of the COD sensor is 1000 mg / L, and it is feasible to measure the sensor readings within the range of [0, 1000]. Therefore, the actual operating range can be the same as the theoretical range of the COD sensor, and the sensor test values need to be corrected using the correction formula obtained from Figure 11 (a).

[0129] Evaluation of the Monitoring Results of the COD Sensor

[0130] Similarly, the actual water samples taken during the daily decantation of the SBR-O tank from June 17 to July 19, 2024, are selected for comparative test result analysis. The results are as shown in Figure 18 . The entire test result can be divided into three stages. Stage I is from June 17 to July 4, during which the measurement performance of the COD sensor is good; Stage II is from July 5 to July 12, during which the measurement performance of the COD sensor gradually declines; Stage III is from July 13 to July 19, during which the COD sensor is cleaned and its measurement performance is restored.

[0131] First of all, from Figure 18It can be clearly found that the direct measurement value of the COD sensor is far from the laboratory analysis value, and the relative error is basically above 40%; while the relative error between the COD value corrected by the correction equation and the laboratory analysis value is basically distributed in the range of ±25%, and the relative error is reduced by about 15%, which greatly improves the measurement accuracy of the COD sensor. This shows that the error correction equation explored in Chapter 3 is effective in improving measurement accuracy. Through the changes in the COD sensor 25 days after it was put into use (at the end of Phase II), it can be found that the inside of the sensor has been covered with yellow pollutants. This change can well explain why Figure 18 In the process of measuring the COD sensor, there will be a phase II in which the sensor's measurement performance gradually deteriorates. This is because the pollutants attached to and grown on the sensor affect the measurement light path, causing the photometric absorption value to be too large, thereby interfering with the measurement of the COD sensor. Therefore, after cleaning the COD sensor in phase III, the sensor's measurement accuracy quickly returned to the measurement accuracy level of phase I. Since the COD sensor used in this study is greatly affected by SS concentration and cannot be used directly in the aeration tank, no experiment was conducted to test the sensor in the aeration tank.

[0132] In summary, using the error correction equation to correct the direct measurement results of the COD sensor can greatly improve the measurement accuracy of the sensor. In addition, the online monitoring system designed in this study expands the scope of use of this COD sensor and greatly reduces the operating and maintenance costs.

[0133] S3. determining the concentration of nitrite nitrogen;

[0134] Obtaining data of target influencing factors in sewage at a target time, and then determining the concentration data of nitrite nitrogen at a target time based on the soft sensor model; the target time includes the current time or a future time;

[0135] The method for determining the concentration data of nitrite nitrogen at a specific target time based on the soft sensor model is:

[0136] When the target time is the current time, it is determined by using the data of the target influencing factor at the current time or the data of the target influencing factor at a historical time; among them, when determining by using the data of the target influencing factor at the current time, the data of the target influencing factor at the current time is used as the input of the soft-sensing model for model calculation to obtain the concentration data of nitrite nitrogen at the current time; when determining by using the data of the target influencing factor at a historical time, the variation law of multiple groups of the target influencing factors in a historical time period is used to predict the data of the target influencing factor at the current time, and then the predicted data of the target influencing factor at the current time is used as the input of the soft-sensing model for model calculation to obtain the concentration data of nitrite nitrogen at the current time

[0137] When the target time is a future time, according to the variation law of multiple groups of the target influencing factors at the current time or a historical time, the data of the target influencing factor at the future time is predicted, and then the predicted data of the target influencing factor at the future time is used as the input of the soft-sensing model for model calculation to obtain the concentration data of nitrite nitrogen at the future time.

[0138] The method for predicting the data of the target influencing factor at a future time or the current time according to the variation law of multiple groups of the data of the target influencing factors at the current time or a historical time includes:

[0139] S3-1. First, according to the variation law of multiple groups of the data of the target influencing factors at the current time or a historical time, multiple working conditions in the biochemical treatment of landfill leachate are defined; among them, the multiple working conditions include the short-cut nitrification-denitrification stage working condition, the nitrification reaction inhibited by FA stage working condition, and the nitrification reaction co-inhibited stage working condition;

[0140] (1) Short-cut nitrification-denitrification stage working condition: In this stage, nitrifying bacteria are inhibited by FA and FNA and cannot completely oxidize ammonia nitrogen to nitrate nitrogen, and a stable short-cut nitrification-denitrification reaction is formed in the whole system. During the life cycle of the whole working condition, the influent ammonia nitrogen gradually increases the ammonia nitrogen concentration from 60 mg / L until the operation of this working condition collapses.

[0141] (2) Nitrification reaction inhibited by FA stage working condition: In this stage, due to the extremely high FA in the reaction system, the activities of AOB and NOB are greatly inhibited, and the nitrification rate in the reaction system is extremely slow. Similarly, the influent ammonia nitrogen is increased gradually until the reactor collapses.

[0142] (3) Nitrification reaction co-inhibited stage working condition: It is the situation where the nitrification reaction is co-inhibited by FA and FNA. In this stage, the concentrations of FA and FNA in the reaction system gradually increase until the reactor operation collapses.

[0143] S3-2. Based on multiple operating conditions in the biochemical treatment of high-ammonia-nitrogen wastewater (such as landfill leachate), determine the operating range of the target impact factor values in each operating condition, as well as the fluctuation law of the target impact factor in each operating condition;

[0144] Define different operating conditions through the variation laws of water quality indicators. Because under the above different operating conditions, the variation laws of each water quality indicator are inconsistent; but within the same operating condition, the variation laws of each water quality indicator remain consistent. Therefore, the operating range and fluctuation law (the fluctuation law is, for example, wavy) of the target impact factor values can be obtained;

[0145] S3-3. According to the operating range and fluctuation law, determine the value of the target impact factor at the future moment or the current moment.

[0146] Since there is a close relationship among the concentrations of nitrite nitrogen, nitrate nitrogen, and ammonia nitrogen, it can reflect the nitrification reaction process and biochemical processes such as the hydrolysis and ammonification of organic matter; and pH, DO, and ORP can supplement relevant information on the nitrification process from the perspective of physical indicators. Then, according to the common variation law of the target impact factor, the value at the future moment can be inferred.

[0147] In summary, the soft measurement technology of the present invention is based on the machine learning Resnet model. By analyzing historical data and real-time monitoring results through the model, it not only improves the monitoring accuracy, but the Resnet model optimized by Bayesian can also predict the change trend of water quality parameters, providing an intelligent solution for sewage treatment, the online monitoring system of the present invention; the present invention also particularly emphasizes the innovation of data processing algorithms. By introducing advanced data fusion technology, the present invention can integrate data from different sensors, improving the accuracy and reliability of monitoring results. At the same time, the present invention has also developed an intelligent prediction function, which can predict the change trend of water quality parameters based on historical data and real-time monitoring results, providing decision support for the optimization of the sewage treatment process.

[0148] Correspondingly, a nitrite nitrogen concentration monitoring system based on a sensor-coupled soft measurement model is applied to the nitrite nitrogen concentration monitoring method of the above sensor-coupled soft measurement model. The nitrite nitrogen concentration monitoring system includes a pretreatment module, a collection module, and an operation module;

[0149] The pretreatment module is used to pretreat the sewage to be treated;

[0150] The collection module is used to collect the target impact factor in the pretreated sewage by using sensors;

[0151] The operation module is used to operate on the target impact factor obtained by the collection module to obtain the nitrite nitrogen concentration;

[0152] Among them, as Figure 4 shown, the pretreatment module includes a filtration unit and a dilution unit; the filtration unit uses a ceramic flat membrane with a pore size of 0.5 μm for filtration, and the operation method of the filtration unit is to perform air-water backwashing every 5 minutes of filtration

[0153] Exemplarily, taking ammonia nitrogen, nitrate nitrogen, and COD as the target influencing factors in the pretreatment module, the research is carried out as follows: After pretreating the sewage using the measurement method of this monitoring system and then measuring, the relative error of the sensor measurement value can meet 15% of the technical requirements specified in HJ355-2019.

[0154] 1.1 Test of basic performance indicators of sensors in water

[0155] Under the landfill leachate environment investigated in the experiment, the stable response time of the ammonia nitrogen sensor is determined to be 8 minutes, the stable response time of the nitrate nitrogen sensor is 10 minutes, and the stable response time of the COD sensor is 1 minute. The ammonia nitrogen, nitrate nitrogen, and COD sensors can all achieve accurate measurement under the laboratory prepared water conditions, but there are deviations of varying degrees between the measured values in actual wastewater and the manually measured values, and the relative errors do not meet 15% of the technical requirements specified in HJ355-2019.

[0156] 1.2 Core design parameters of the filtration unit

[0157] In this study, in order to minimize the impact of filtration on water quality indicators such as ammonia nitrogen, nitrate nitrogen, and COD in water as much as possible, a ceramic flat membrane with a pore size of 0.5 μm is selected. Relevant research shows that although the initial membrane flux of this pore size membrane is very large, it is extremely vulnerable to pollution. Therefore, the designed operation mode is to perform air-water backwashing every 5 minutes of filtration. When using a 1 m2 flat membrane, about 1.25 L of water can be produced within 5 minutes, fully meeting the measurement use of the sensor. The membrane cleaning method (the membrane cleaning method is to perform air-water backwashing every time after water sampling and measurement; when the membrane flux drops to about 40% of the original value, the original value is 41.09 L / (m2·h), and then chemical cleaning is carried out to restore the membrane flux) selects alkali washing, acid washing, and sodium hypochlorite washing. By comparing the membrane flux of the wastewater after membrane cleaning and the size of the wastewater flux recovery rate of the ceramic microfiltration membrane, the effect of cleaning the ceramic microfiltration membrane is measured, and the cleaning method combination is shown in Table 3.

[0158] Table 3 Membrane cleaning methods

[0159] Number Cleaning method A 0.1mol / L NaOH solution B <![CDATA[0.1mol / L HNO3 solution]]> C 0.1mol / L NaClO solution

[0160] As Figure 2As shown, the wastewater membrane flux of the new ceramic microfiltration membrane is 41.09 L / (m2·h), and the wastewater membrane fluxes after cleaning with alkali, acid, and sodium hypochlorite are 30, 17.45, and 32.91 L / (m2·h), respectively ( Figure 2 ). It can be seen that the cleaning effect of sodium hypochlorite solution is the best, and it can be restored to 80.09% of the wastewater flux of the new membrane; the effect of alkali cleaning is the second best, and it can be restored to 73.01% of the wastewater flux of the new membrane; the cleaning effect of nitric acid solution is the worst, and it can only be restored to 42.46% of the wastewater flux of the new membrane.

[0161] To further explore the optimal cleaning method, this paper also studied the change of membrane flux of ceramic microfiltration membrane after chemical cleaning, and the results are as Figure 3 shown.

[0162] From Figure 3 it can be seen that although the recovery effect of NaClO on membrane flux is the best, compared with the change of membrane flux after cleaning with NaOH solution, the membrane flux decay rate of the membrane cleaned with NaClO solution is faster. Therefore, in the subsequent experiments of this paper, 0.1 mol / L NaOH solution is used for chemical cleaning of the ceramic flat membrane.

[0163] 1.4 Design of Pretreatment Device for On-line Monitoring of High Ammonia-nitrogen Wastewater Quality

[0164] The pretreatment unit proposed in the embodiment of the present invention solves two core problems encountered by water quality sensors in practical engineering applications. First, aiming at the problem that there are many impurities in the sewage mixed liquor, blocking the ion exchange membrane or the laser channel, thus interfering with the measurement results, the filtration unit can remove the vast majority of SS, which can enable the sensor to measure in an environment with better water quality, avoiding the blocking problem from the source; reducing the instrument failure rate is beneficial to the long-term accurate monitoring of the sensor. Second, aiming at the problem of whether the error between the on-line monitoring value and the laboratory manual measurement value meets the specified range, a dilution unit is set up, so that the sensor can measure within the best measurement range explored above, and these measurement values are then corrected by relevant correction formulas, and the accuracy can meet the technical requirements specified in HJ355-2019.

[0165] Experimental Example: Productive Experiment of On-line Water Quality Monitoring - Soft Measurement Model of Nitrite

[0166] (1) Experimental Device and Operation Cycle, the experimental device is as Figure 20 shown, and the operation cycle is as Figure 19 shown;

[0167] Figure 20Among them, 1 is the influent water tank, 2 is the peristaltic pump, 3 is the effluent water tank, 4 is the aeration pump, 5 is the pH, DO, ORP sensor, 6 is the submersible mixer, 7 is the ceramic flat membrane, 8 is the sensor monitoring water tank, 9 is the ammonia nitrogen and nitrate nitrogen sensor, and 10 is the multi-parameter water quality monitor;

[0168] This experimental device is composed of an influent water tank, an SBR reactor, and an effluent water tank connected in sequence. The effective volume of the SBR reactor is 20L. Aeration is carried out using an aeration head, and stirring is carried out using a submersible mixer. At the same time, pH, DO, and ORP sensors are also installed to monitor the operating state of the reactor. The raw water adjusted by adding COD is pumped from the influent water tank into the SBR reactor by a peristaltic pump. Subsequently, the mixed liquid completes denitrification and nitrification reactions in the SBR reactor successively. During the nitrification reaction stage, the changes in relevant water quality indicators are monitored every 50 minutes; the pH, DO, and ORP indicators are directly read by a multi-parameter water quality detector; for the monitoring of ammonia nitrogen and nitrate nitrogen, the mixed liquid is first filtered by a peristaltic pump through a ceramic flat membrane into the sensor monitoring water tank, and then obtained by sensor monitoring. At the same time, a syringe is used to extract the mixed liquid in the reactor and measure the concentration of nitrite nitrogen in it.

[0169] The operation cycle of the reactor is as Figure 19 shown, adopting the anoxic / aerobic (A / O) mode, including four stages: influent water, stirring, aeration, sedimentation and drainage. Each cycle runs for 12 hours, and two cycles are run every day. The submersible mixer is turned on in the anoxic section to fully stir the mixed liquid to ensure the smooth progress of denitrification. In the aerobic section, the aeration pump is started for aeration, and at the same time, the submersible mixer is kept turned on to ensure the smooth progress of the nitrification reaction. Finally, the aeration pump and the submersible mixer stop, and the sedimentation and drainage stage is entered. The initial inoculation sludge concentration of the mixed liquid is 12000mg / L, and sludge is not manually discharged.

[0170] Combined with the research results of the previous two parts, an online monitoring system integrating sensor monitoring and machine learning soft sensing was successfully constructed and applied in an actual sewage treatment plant to verify its actual application effect.

[0171] The case study site selected in this chapter is a landfill leachate treatment plant in Luohu District, Shenzhen City, which is also the water intake site for the experimental sewage in the previous two chapters.

[0172] In the entire treatment process, the SBR A and O pools are the most core treatment units, mainly responsible for degrading COD and nitrogen removal. Therefore, the online monitoring system is deployed near the SBR O pool to mainly monitor the changes in various water quality indicators in this pool. The treatment process flow of the sewage treatment plant is as Figure 7 shown. The specific process is as follows:

[0173] As Figure 8As shown, Unit 1 is a water intake pump, whose main function is to pump the muddy water mixture from the aeration tank into the filtration functional unit. Unit 2 is a filtration unit, inside which there are 0.5μm ceramic flat membrane modules. The mixed liquid is filtered in this unit, and the filtered solution enters Unit 3. Unit 3 is a dilution and measurement unit, where the filtered solution is diluted and the water quality sensor completes the measurement work. Unit 4 is the fixed box for the transmitter of the water quality sensor, which is used to fix the transmitter of the sensor.

[0174] The research results show that the designed on-line monitoring system significantly improves the measurement accuracy of the sensor and reduces the maintenance frequency. Adjusting the measurement results of ammonia nitrogen and COD sensors through an error correction model is a simple and efficient strategy. This system effectively avoids the pollution of the sensor probe, reduces the risk of the decline of measurement accuracy, and thus reduces the maintenance cost.

[0175] In addition, the soft measurement model of nitrite nitrogen in the SBR-O pool constructed based on the Resnet model, after Bayesian optimization, the R 2 reaches 0.73 and the RMSE is 8.16. The model can better fit the change trend of nitrite nitrogen. Through the sensitivity analysis of the input variables of the constructed Resnet soft measurement model of nitrite nitrogen, it is found that the four water quality indicators of ammonia nitrogen concentration, nitrate nitrogen concentration, temperature, and dissolved oxygen have a very significant impact on the prediction of nitrite nitrogen. The influence contribution degrees of pH and ORP are relatively smaller.

[0176] Therefore, the present invention proposes that the pretreatment module pretreats the original water sample before measurement, avoiding the pollution of the sensor probe, effectively improving the measurement accuracy of the sensor, and reducing the occurrence frequency of measurement instrument failures. The system of the present invention has been proved to be able to effectively cope with various challenges in the treatment process of high ammonia nitrogen wastewater such as landfill leachate, and can accurately monitor key water quality parameters, making the landfill leachate treatment process more efficient and environmentally friendly, and further promoting the development of the environmental protection cause.

[0177] In summary, the present invention combines sensor monitoring data with model soft measurement. Sensor monitoring provides a data basis for model prediction, and the machine learning model conducts learning and pre-judgment. Moreover, through the exploration of water quality parameters affecting sensor monitoring, the determination of the sensor measurement range, and the pretreatment link, the purpose is to obtain more accurate and stable monitoring data, so as to better predict in the soft measurement stage.

Claims

1. A method for monitoring nitrite nitrogen concentration based on a sensor-coupled soft measurement model, characterized in that, It includes the following steps: S1. Based on the water quality parameters and nitrification process of the sewage to be monitored, determine multiple target influencing factors; And obtain the data of the multiple target influencing factors and the concentration data of nitrite nitrogen at multiple historical moments; S2. Based on the data of the multiple target influencing factors at the multiple historical moments and the concentration data of nitrite nitrogen, establish a soft sensor model. The input of the soft sensor model is the data of the target influencing factors, and the output of the soft sensor model is the concentration of nitrite nitrogen; S3. Obtain the data of the target influencing factors in the sewage at the target moment, and then based on the soft sensor model, determine the concentration data of nitrite nitrogen at the target moment; the target moment includes the current moment or a future moment.

2. The nitrite nitrogen concentration monitoring method based on a sensor-coupled soft measurement model according to claim 1, wherein In S1, the method for determining multiple target influencing factors based on the water quality parameters and nitrification process of the sewage to be monitored includes: Obtain data of multiple water quality parameters that affect nitrite, and the multiple water quality parameters include NO 3- -N, NH4 + -N, COD, temperature, pH, DO, ORP, total nitrogen, total phosphorus, suspended solids, and TOC; Screen out multiple water quality parameters that have an impact on the nitrification process of the sewage; Perform a multicollinearity analysis on the multiple water quality parameters that have an impact on each nitrification process, and select the water quality parameters with a variance inflation factor not exceeding the standard in the multicollinearity analysis as the target influencing factors.

3. A method for monitoring nitrite nitrogen concentration based on a sensor-coupled soft measurement model according to claim 2, characterized in that, The target influencing factors include temperature, pH, DO, ORP, NO3 - -N, and NH4 + -N.

4. The nitrite nitrogen concentration monitoring method based on a sensor-coupled soft measurement model according to claim 3, wherein The method for establishing a soft sensor model in S2 includes: Obtain the data of the multiple target influencing factors and the concentration data of nitrite nitrogen at multiple historical moments to form a data set, and divide the data set into a training set and a test set; Based on the data of the training set, using RMSE as the loss function, use multiple machine learning algorithms for modeling to obtain multiple algorithm models. The multiple algorithm models include RF model, GBDT model, XGBOOST model, MLP model, and Resnet model; Based on the multiple algorithm models, test the data input into the test set, and select the algorithm model with the smallest error in the test results as the soft sensor model.

5. The nitrite nitrogen concentration monitoring method based on a sensor-coupled soft measurement model according to claim 5, characterized in that, In S3, the method for determining the concentration data of nitrite nitrogen at the target moment based on the soft sensor model is: When the target moment is the current moment, it is determined using the data of the target influencing factors at the current moment or the data of the target influencing factors at historical moments; Among them, when determining using the data of the target influencing factors at the current moment, the data of the target influencing factors at the current moment is used as the input of the soft sensor model for model calculation to obtain the concentration data of nitrite nitrogen at the current moment; When determining using the data of the target influencing factors at historical moments, use the change law of multiple groups of the target influencing factors in the historical time period to predict the data of the target influencing factors at the current moment, and then use the predicted data of the target influencing factors at the current moment as the input of the soft sensor model for model calculation to obtain the concentration data of nitrite nitrogen at the current moment.

6. The nitrite nitrogen concentration monitoring method based on a sensor-coupled soft measurement model according to claim 5, characterized in that, When the target moment is a future moment, predict the data of the target influencing factors at the future moment according to the change law of multiple groups of the target influencing factors at the current moment or historical moments, and then use the predicted data of the target influencing factors at the future moment as the input of the soft sensor model for model calculation to obtain the concentration data of nitrite nitrogen at the future moment.

7. The nitrite nitrogen concentration monitoring method based on a sensor-coupled soft measurement model according to claim 6, characterized in that, The method for predicting the data of the target influencing factor at a future moment or the current moment according to the variation rules of multiple groups of the target influencing factors at the current moment or historical moments includes: Firstly, according to the variation rules of multiple groups of the target influencing factors at the current moment or historical moments, define multiple working conditions in the biochemical treatment of high-ammonia-nitrogen wastewater; wherein, the multiple working conditions include the short-cut nitrification-denitrification stage working condition, the nitrification reaction inhibited by FA stage working condition, and the nitrification reaction co-inhibited stage working condition; Based on the multiple working conditions in the biochemical treatment of high-ammonia-nitrogen wastewater, determine the operating range of the numerical values of the target influencing factor in each working condition, and the fluctuation rule of the target influencing factor in each working condition; According to the operating range and the fluctuation rule, determine the data of the target influencing factor at the future moment or the current moment.

8. The nitrite nitrogen concentration monitoring method based on a sensor-coupled soft measurement model according to claim 5, characterized in that Before forming the data set, perform data correction on the data of the target influencing factor at multiple moments obtained; the data correction method includes: Respectively fit the data of multiple target influencing factors of each type to obtain the fitting equation of each target influencing factor; Use the fitting equation to delimit the numerical range of the target influencing factor, and eliminate the data of the target influencing factor that exceeds the numerical range.

9. A nitrite nitrogen concentration monitoring system based on a sensor-coupled soft measurement model, based on the nitrite nitrogen concentration monitoring method based on the sensor-coupled soft measurement model according to any one of claims 1-8, characterized in that, The nitrite nitrogen concentration monitoring system includes a pretreatment module, a collection module, and an operation module; The pretreatment module is used to pretreat the sewage to be treated; The collection module is used to collect the target influencing factor in the pretreated sewage by using a sensor; The operation module is used to perform operations on the target influencing factor obtained by the collection module to obtain the nitrite nitrogen concentration; Wherein, the pretreatment module includes a filtration unit and a dilution unit; the filtration unit uses a ceramic flat membrane with a pore size of 0.5 μm for filtration, and the operation method of the filtration unit is to perform air-water backwashing once every 5 minutes of filtration.

Images