An ai monitoring and evaluation method and system for nitrogen reduction in sewage treatment

Abstract

This invention discloses an AI monitoring and evaluation method and system for nitrogen reduction in wastewater treatment, relating to the field of wastewater treatment technology. The method includes: collecting multi-dimensional parameters related to nitrogen reduction in wastewater treatment; screening and calibrating parameters using a nitrogen feature independent component screening algorithm; determining the parameter importance ranking using the SHAP nitrogen feature contribution analysis algorithm; constructing a dissolved oxygen coupled nitrogen conversion prediction model based on the calibrated parameters and their contributions to predict nitrogen conversion rate and efficiency; inputting the prediction results and real-time operating parameters into the SOURATP nitrogen process intelligent early warning and judgment platform to identify abnormal operating conditions; and finally generating an AI monitoring and evaluation report. The system comprises six units. This method and system eliminate the need for manual sampling and offline detection, enabling real-time capture of parameter changes, improving monitoring real-time performance. The integration of multiple algorithms for collaborative analysis enhances the accuracy and comprehensiveness of the evaluation. It can also provide early warnings of anomalies and pinpoint influencing factors, improving anomaly response efficiency and facilitating efficient management and control of nitrogen reduction in wastewater treatment.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, and in particular to an AI monitoring and evaluation method and system for nitrogen reduction in wastewater treatment. Background Technology

[0002] In the current wastewater treatment industry, nitrogen reduction efficiency directly impacts water quality compliance and ecological environmental safety. With increasingly stringent national requirements for water environment governance, the need for precise monitoring and efficient assessment of nitrogen reduction processes in wastewater treatment systems is becoming increasingly urgent. Traditional wastewater treatment nitrogen reduction monitoring relies heavily on manual sampling and offline detection. This not only makes it difficult to capture the dynamic changes of multi-dimensional parameters in real time during wastewater treatment but also fails to quickly identify key influencing factors in the nitrogen conversion process. This results in a lag in the assessment of nitrogen reduction efficiency, failing to meet the practical needs of stable operation and optimized regulation of wastewater treatment systems. Against this backdrop, there is an urgent need to leverage artificial intelligence technology to integrate multi-dimensional monitoring parameters and construct methods and systems capable of real-time monitoring, accurate prediction, and intelligent early warning of nitrogen reduction processes, thereby improving the scientific rigor and efficiency of nitrogen reduction management in wastewater treatment.

[0003] Existing technologies for monitoring and assessing nitrogen reduction in wastewater treatment have two significant drawbacks: First, existing technologies often employ a single algorithm or model to analyze the nitrogen reduction process, failing to effectively integrate multiple technologies such as feature screening, contribution analysis, predictive modeling, and intelligent early warning. This results in a relatively singular analytical dimension of the nitrogen reduction process, failing to comprehensively reflect the combined impact of multi-dimensional parameters on nitrogen reduction effectiveness, thus affecting the accuracy and comprehensiveness of monitoring and assessment results. Second, existing technologies lack the ability to accurately identify and assess abnormal operating conditions during the nitrogen reduction process. They are unable to provide early warnings of potential anomalies through system parameter correlation analysis, nor can they quickly locate key influencing factors after anomalies occur. This makes it difficult for wastewater treatment systems to take timely and targeted control measures when faced with nitrogen reduction anomalies, easily leading to a decrease in nitrogen reduction efficiency or even failure to meet effluent quality standards. Summary of the Invention

[0004] In order to overcome the shortcomings and deficiencies of existing technologies, this invention provides an AI monitoring and evaluation method and system for nitrogen reduction in wastewater treatment.

[0005] The technical solution adopted in this invention is an AI monitoring and evaluation method for nitrogen reduction in wastewater treatment, comprising: Step S1, collecting multi-dimensional parameters related to nitrogen reduction during wastewater treatment, including influent total nitrogen concentration, effluent total nitrogen concentration, dissolved oxygen concentration in the nitrification tank, dissolved oxygen concentration in the denitrification tank, sludge return ratio, mixed liquor return ratio, nitrification tank temperature, denitrification tank temperature, hydraulic retention time, and sludge age; Step S2, inputting the collected multi-dimensional parameters into a nitrogen feature independent component screening algorithm, which performs feature separation and screening on the multi-dimensional parameters to obtain a set of calibration feature parameters that significantly affect nitrogen reduction; Step S3, inputting the calibration feature parameter set into a SHAP nitrogen feature contribution analysis algorithm, which calculates the contribution of different calibration feature parameters to the nitrogen reduction process. Contribution values ​​are used to determine the importance ranking of different calibration feature parameters; Step S4: Based on the calibration feature parameter set and the contribution values ​​of different parameters, a dissolved oxygen coupled nitrogen conversion prediction model is constructed. This model is used to predict the nitrogen conversion rate and nitrogen reduction efficiency in the wastewater treatment process, and the prediction results are output; Step S5: The prediction results of the dissolved oxygen coupled nitrogen conversion prediction model and the real-time operating parameters of the wastewater treatment system are input into the SOURATP nitrogen process intelligent early warning and judgment platform. The platform is used to identify and judge abnormal operating conditions in the nitrogen reduction process; Step S6: Based on the judgment results of the SOURATP nitrogen process intelligent early warning and judgment platform, an AI monitoring and evaluation report on the nitrogen reduction process in wastewater treatment is generated. The report includes the contribution value of calibration feature parameters, nitrogen conversion prediction results, and abnormal operating condition identification results.

[0006] Furthermore, the expression for the SHAP nitrogen feature contribution parsing algorithm is as follows: ,in, The SHAP contribution value of the i-th nitrogen feature parameter. For a feature subset that does not contain the i-th feature parameter, To calibrate the feature parameter set, To determine the total number of parameters in the characteristic parameter set, The nitrogen reduction impact assessment function for the feature subset. The weighting coefficients for the dissolved oxygen concentration parameter are: This is the adjustment coefficient for the total nitrogen concentration parameter.

[0007] Furthermore, the expression for the dissolved oxygen coupled nitrogen transformation prediction model is as follows: ,in, For nitrogen conversion rate, This refers to the dissolved oxygen concentration in the nitrification tank. Temperature of the nitrification tank This refers to the dissolved oxygen concentration in the denitrification tank. The hydraulic residence time, The sludge return ratio is... The total nitrogen concentration in the influent. These are the nitrification reaction coefficient, temperature influence coefficient, denitrification reaction coefficient, hydraulic retention time coefficient, and sludge return influence coefficient of the model, respectively.

[0008] Furthermore, the expression for the nitrogen feature independent component screening algorithm is as follows: ,in, For the j-th independent component, Let be the separation coefficient between the j-th independent component and the k-th original parameter. For the k-th original nitrogen characteristic parameter, This represents the total number of original parameters. Let j be the error correction term for the j-th independent component. The influencing factors of sludge age parameter, This is the weighting factor for the concentration parameter of suspended solids in the mixture.

[0009] Furthermore, the abnormal operating condition identification expression of the SOURATP nitrogen process intelligent early warning and judgment platform is as follows: ,in, The alert result is indicated by 1 for abnormality and 0 for normality. This represents the predicted nitrogen conversion rate output by the dissolved oxygen coupled nitrogen conversion prediction model. This is the standard value for nitrogen conversion rate. This represents the change in dissolved oxygen concentration. This represents the change in total nitrogen concentration. These are the weights for predicted values, standard values, dissolved oxygen changes, and total nitrogen changes, respectively. This is a correction factor for the specific oxygen consumption rate parameter.

[0010] Furthermore, the comprehensive index expression for AI monitoring and evaluation of nitrogen reduction in the wastewater treatment is as follows: ,in, A comprehensive index for monitoring and evaluating AI. To determine the number of feature parameters, Let be the real-time monitoring value of the i-th calibration feature parameter. For the maximum nitrogen conversion rate, This is the influence coefficient for abnormal operating conditions. This is the baseline coefficient for normal operating conditions. , The evaluation weights are for dissolved oxygen, total nitrogen, and early warning, respectively.

[0011] Further, step S3 includes the following sub-steps: S31, obtaining the calibration feature parameter set output in step S2, arranging each parameter in the calibration feature parameter set according to the time sequence of the wastewater treatment process to form a time-series feature sequence; S32, inputting the time-series feature sequence into the feature processing module of the SHAP nitrogen feature contribution analysis algorithm, performing sliding window segmentation on the time-series feature sequence to obtain multiple feature subsequences, each feature subsequence including 5-8 consecutive time-series data points; S33, calling the calculation module of the SHAP nitrogen feature contribution analysis algorithm, calculating the local contribution value of different parameters in each feature subsequence to the nitrogen reduction process based on each feature subsequence, and obtaining the global contribution value of different parameters through weighted averaging; S34, sorting the global contribution values ​​of different parameters in descending order to generate a calibration feature parameter importance ranking table, as the output result of step S3.

[0012] Further, step S4 includes the following sub-steps: S41, extracting the top 8-10 most important parameters from the importance ranking table of the calibration feature parameters output in step S3, and using them as the input parameter set for the dissolved oxygen coupled nitrogen conversion prediction model; S42, constructing the network structure of the dissolved oxygen coupled nitrogen conversion prediction model, which includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is consistent with the number of parameters in the input parameter set. The hidden layer includes 2-3 sub-layers, with the number of nodes in each sub-layer being 1.5-2 times the number of nodes in the input layer. The output layer includes 2 nodes, corresponding to the nitrogen conversion rate and nitrogen reduction efficiency, respectively; S43, training the constructed dissolved oxygen coupled nitrogen conversion prediction model using historical nitrogen reduction data from wastewater treatment, and adjusting the coefficients of different terms of the model using the gradient descent algorithm until the prediction error of the model converges to below a preset threshold; S44, inputting the calibration feature parameter set output in step S2 into the trained dissolved oxygen coupled nitrogen conversion prediction model, running the model to obtain the predicted values ​​of nitrogen conversion rate and nitrogen reduction efficiency, which are used as the prediction results of step S4.

[0013] Further, step S5 includes the following sub-steps: S51, collecting real-time operating parameters of the wastewater treatment system, including real-time dissolved oxygen concentration in the nitrification tank, real-time dissolved oxygen concentration in the denitrification tank, real-time sludge return ratio, real-time mixed liquor return ratio, and real-time total nitrogen concentration in the effluent; S52, aligning the prediction results output in step S4 with the real-time operating parameters to ensure consistency in the time dimension, forming an aligned comprehensive dataset; S53, inputting the comprehensive dataset into the data analysis module of the SOURATP nitrogen process intelligent early warning and judgment platform, which calculates the deviation rate and deviation trend by comparing the deviation between the prediction results and the real-time operating parameters; S54, calling the early warning and judgment module of the SOURATP nitrogen process intelligent early warning and judgment platform, identifying the abnormal operating conditions in the nitrogen reduction process based on the deviation rate, deviation trend, and preset early warning thresholds, determining the severity of the abnormal operating conditions, and generating early warning and judgment results.

[0014] An AI monitoring and evaluation system for nitrogen reduction in wastewater treatment is disclosed. This system, applied to an AI monitoring and evaluation method for nitrogen reduction in wastewater treatment, includes the following units: a multi-dimensional nitrogen parameter acquisition and transmission unit, which collects multi-dimensional parameters during wastewater treatment, including influent total nitrogen concentration, effluent total nitrogen concentration, dissolved oxygen concentration in the nitrification tank, dissolved oxygen concentration in the denitrification tank, sludge return ratio, mixed liquor return ratio, nitrification tank temperature, denitrification tank temperature, hydraulic retention time, and sludge age, and transmits the collected parameters to the multi-dimensional nitrogen parameter acquisition and transmission unit via industrial Ethernet; a nitrogen feature independent component screening and processing unit, connected to the multi-dimensional nitrogen parameter acquisition and transmission unit, receiving the transmitted multi-dimensional parameters, performing feature separation and screening on the multi-dimensional parameters using a nitrogen feature independent component screening algorithm, and outputting a calibration feature parameter set to the SHAP nitrogen feature contribution analysis unit; and a SHAP nitrogen feature contribution analysis unit, connected to the nitrogen feature independent component screening and processing unit, receiving the output calibration feature parameter set, and calculating different calibrations using the SHAP nitrogen feature contribution analysis algorithm. The contribution values ​​of characteristic parameters are used to generate a parameter importance ranking table, which is then transmitted to the dissolved oxygen coupled nitrogen conversion prediction unit. This unit, connected to the SHAP nitrogen characteristic contribution analysis unit, receives the transmitted parameter importance ranking table, constructs a dissolved oxygen coupled nitrogen conversion prediction model based on the table, predicts the nitrogen conversion rate and nitrogen reduction efficiency, and outputs the prediction results to the SOURATP nitrogen process intelligent early warning and judgment unit. The SOURATP nitrogen process intelligent early warning and judgment unit, connected to both the dissolved oxygen coupled nitrogen conversion prediction unit and the multi-dimensional nitrogen parameter acquisition and transmission unit, receives the prediction results and real-time operating parameters, identifies abnormal operating conditions through the SOURATP nitrogen process intelligent early warning and judgment platform, generates judgment results, and transmits them to the AI ​​monitoring and evaluation report generation unit. The AI ​​monitoring and evaluation report generation unit, connected to the SOURATP nitrogen process intelligent early warning and judgment unit, receives the transmitted judgment results, integrates and calibrates the contribution values ​​of characteristic parameters, nitrogen conversion prediction results, and abnormal operating condition identification results, and generates and outputs an AI monitoring and evaluation report on the nitrogen reduction process in wastewater treatment.

[0015] Beneficial Effects: This invention proposes an AI monitoring and evaluation method and system for nitrogen reduction in wastewater treatment. It collects multi-dimensional nitrogen reduction-related parameters, filters and calibrates parameters using a nitrogen feature independent component screening algorithm, clarifies the parameter importance ranking using the SHAP nitrogen feature contribution analysis algorithm, and combines a dissolved oxygen coupled nitrogen conversion prediction model to predict nitrogen conversion rate and efficiency. Finally, the SOURATP nitrogen process intelligent early warning and judgment platform identifies abnormal operating conditions and generates evaluation reports. The entire process requires no manual sampling or offline detection, and can capture dynamic parameter changes in real time and quickly locate key influencing factors. This improves the real-time performance of nitrogen reduction monitoring and solves the problem of lagging monitoring and evaluation in existing systems. Furthermore, this method and system integrate features... The technology, encompassing screening, contribution analysis, predictive modeling, and intelligent early warning, utilizes multi-algorithm collaboration to achieve multi-dimensional analysis of the nitrogen reduction process. This comprehensively reflects the combined impact of multiple parameters, significantly improving the accuracy and comprehensiveness of monitoring and evaluation results, and overcoming the shortcomings of existing technologies that rely on single-algorithm analysis dimensions. Furthermore, its SOURATP nitrogen process intelligent early warning and judgment platform can provide early warnings of potential anomalies through parameter correlation analysis. After an anomaly occurs, it can quickly locate influencing factors by calibrating parameter contribution rates, helping the wastewater treatment system to take timely and targeted control measures. This not only improves the timeliness and accuracy of responding to abnormal operating conditions but also avoids a decline in nitrogen reduction efficiency and substandard effluent quality, addressing the shortcomings of existing technologies in anomaly identification and judgment. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the overall steps of the method of the present invention;

[0017] Figure 2 This is a flowchart of method step S3 of the present invention;

[0018] Figure 3 This is a flowchart of method step S4 of the present invention;

[0019] Figure 4 This is a flowchart of step S5 of the method of the present invention;

[0020] Figure 5 This is a diagram showing the system unit composition of the present invention. Detailed Implementation

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] like Figure 1 As shown, an AI monitoring and assessment method for nitrogen reduction in wastewater treatment includes:

[0023] Step S1: Collect multi-dimensional parameters related to nitrogen reduction during the wastewater treatment process. These parameters include total nitrogen concentration in the influent, total nitrogen concentration in the effluent, dissolved oxygen concentration in the nitrification tank, dissolved oxygen concentration in the denitrification tank, sludge return ratio, mixed liquor return ratio, temperature in the nitrification tank, temperature in the denitrification tank, hydraulic retention time, and sludge age.

[0024] Specifically, step S1 involves collecting multi-dimensional parameters related to nitrogen reduction during the wastewater treatment process, providing basic data support for subsequent analysis. The quality of its implementation directly affects the accuracy of subsequent feature selection and model prediction. During implementation, the specific types of parameters to be collected were first determined, including 10 calibration parameters: influent total nitrogen concentration, effluent total nitrogen concentration, dissolved oxygen concentration in the nitrification tank, dissolved oxygen concentration in the denitrification tank, sludge return ratio, mixed liquor return ratio, nitrification tank temperature, denitrification tank temperature, hydraulic retention time, and sludge age. Secondly, the collection locations and equipment for each parameter were specified. Influent and effluent total nitrogen concentrations were collected using online total nitrogen analyzers installed on the influent and effluent pipes, respectively. The selected equipment was an online ultraviolet spectrophotometer with a range of 0-100 mg / L and a detection accuracy of ±0.5 mg / L. Dissolved oxygen concentrations in the nitrification and denitrification tanks were collected using fluorescence dissolved oxygen sensors submerged 0.5-1.0 m below the liquid level in the tanks. The sensor response time was ≤30 seconds, and the measurement range was 0-20 mg / L. The reflux ratio and mixed liquor reflux ratio are calculated using an electromagnetic flowmeter installed on the reflux pipe in conjunction with the frequency of the variable frequency pump. The flowmeter accuracy is ±0.2%. The temperatures of the nitrification and denitrification tanks are collected using platinum resistance temperature sensors inserted into the tank walls, with a measurement range of 0-50℃ and an accuracy of ±0.1℃. The hydraulic retention time is obtained by calculating the effective volume of the reaction tank and the influent flow rate. The sludge age is calculated using the sludge discharge and the total amount of sludge in the reaction tank. Finally, the acquisition frequency is set, and all parameters are continuously acquired every 5 minutes. The acquired data is stored in real time in the database of the industrial control computer to ensure the continuity and timeliness of the data. This step, through comprehensive, accurate, and real-time parameter acquisition, avoids subsequent analysis deviations due to data loss or errors, providing a reliable data foundation for AI monitoring and evaluation of nitrogen reduction.

[0025] Step S2: Input the collected multi-dimensional parameters into the nitrogen feature independent component screening algorithm. The algorithm performs feature separation and screening on the multi-dimensional parameters to obtain a set of calibration feature parameters that have a significant impact on nitrogen reduction.

[0026] Specifically, step S2 involves feature separation and screening of the multi-dimensional parameters collected in S1. The key feature parameter set that significantly affects nitrogen reduction is extracted by using the nitrogen feature independent component screening algorithm, thereby reducing the interference of redundant parameters on subsequent analysis and improving analysis efficiency and accuracy. During implementation, the multi-dimensional parameters of S1 stored in the database are first integrated according to time series to form an original dataset indexed by time and including 10 parameters. During the integration process, abnormal data caused by equipment malfunctions (such as dissolved oxygen concentrations suddenly exceeding 20 mg / L or falling below 0 mg / L, or temperatures exceeding the 0-50℃ range) are removed to ensure the integrity and validity of the original dataset. Secondly, the processed original dataset is input into the computation module of the nitrogen feature independent component analysis algorithm. The algorithm first standardizes each parameter (not normalized, but only through data shifting and scaling to ensure all parameters are of the same order of magnitude, avoiding the influence of dimensional differences on the screening results). Then, it uses the independent component analysis principle to perform feature separation on the standardized parameters and calculates the correlation coefficient between each parameter and the nitrogen reduction process. The correlation coefficient calculation is based on the synergistic analysis of parameter change trends and nitrogen reduction efficiency change trends; the higher the synergy, the better. The larger the correlation coefficient, the better. Next, a correlation coefficient threshold is set, which is determined to be 0.6 through multiple experiments. Parameters with a correlation coefficient greater than 0.6 are identified as key feature parameters that significantly affect nitrogen reduction. Typically, the correlation coefficients of 5-6 parameters—total nitrogen concentration in influent, dissolved oxygen concentration in nitrification tank, dissolved oxygen concentration in denitrification tank, sludge return ratio, and sludge age—will exceed the threshold, forming a set of key feature parameters. Finally, the effectiveness of the selected set of key feature parameters is verified by comparing the prediction errors of this parameter set with those of the original dataset in subsequent models. If the verification results show that the increase in prediction error is less than 5%, then the set of key feature parameters is determined to be a valid output. This step, through precise parameter selection and calibration, reduces the computational load of subsequent algorithms (reducing the number of parameters from 10 to 5-6) while ensuring that parameters with a core impact on nitrogen reduction are retained, laying the foundation for the feature contribution analysis of S3.

[0027] Step S3: Input the calibration feature parameter set into the SHAP nitrogen feature contribution analysis algorithm. The algorithm calculates the contribution value of different calibration feature parameters to the nitrogen reduction process and determines the importance ranking of different calibration feature parameters.

[0028] Specifically, step S3 uses the SHAP nitrogen feature contribution analysis algorithm to calculate the contribution value of each parameter in the key feature parameter set selected in S2 to the nitrogen reduction process, and determines the importance ranking of each parameter, clarifying the core factors affecting nitrogen reduction, and providing a basis for parameter priority for subsequent model construction. During implementation, the key feature parameter set output by S2 is first retrieved from the database. Parameter data for 72 consecutive hours (864 data sets, calculated at a collection frequency of 5 minutes) is extracted according to the time series and used as input data for the SHAP nitrogen feature contribution analysis algorithm. Data extraction ensures stable operation of the wastewater treatment system within the time interval (no large-scale equipment maintenance, no drastic fluctuations in influent water quality, etc.) to avoid special operating conditions affecting the contribution calculation results. Next, the input data is imported into the algorithm's calculation module. The algorithm first constructs a mapping relationship model between parameters and nitrogen reduction efficiency, based on a data-driven nonlinear fitting method. Then, using the SHAP value calculation principle, the marginal contribution of each key feature parameter to nitrogen reduction efficiency under different values ​​is analyzed. The marginal contribution calculation considers the interaction between parameters (such as the synergistic effect of dissolved oxygen concentration in the nitrification tank and sludge age), ultimately yielding the average contribution value of each parameter over the entire time interval. Then... The average contribution values ​​of each parameter are sorted, and a ranking table of key feature parameters is generated in descending order of contribution value. Typically, the dissolved oxygen concentration in the nitrification tank has the highest contribution value (average contribution value approximately 0.35-0.45), followed by the influent total nitrogen concentration (average contribution value approximately 0.25-0.35), the dissolved oxygen concentration in the denitrification tank (average contribution value approximately 0.15-0.25), the sludge return ratio (average contribution value approximately 0.08-0.15), and the sludge age (average contribution value approximately 0.05-0.10). Finally, the importance ranking table and contribution values ​​are stored in a database, and parameters with a contribution value greater than 0.2 are marked as core influencing parameters. This step quantifies the contribution of each parameter, clearly defining the weight of different parameters in nitrogen reduction, avoiding excessive focus on secondary parameters in subsequent model construction, improving the targeting and efficiency of model construction, and providing a basis for factor localization under abnormal operating conditions.

[0029] Step S4: Based on the calibration feature parameter set and the contribution values ​​of different parameters, construct a dissolved oxygen coupled nitrogen conversion prediction model. Use this model to predict the nitrogen conversion rate and nitrogen reduction efficiency in the wastewater treatment process and output the prediction results.

[0030] Specifically, step S4, based on the key feature parameter set and contribution values ​​of each parameter output from S3, constructs a dissolved oxygen coupled nitrogen conversion prediction model to predict the nitrogen conversion rate and nitrogen reduction efficiency during wastewater treatment, providing a predictive basis for the dynamic management of the nitrogen reduction process. During implementation, the input and output variables of the model are first determined. The input variables are the core influencing parameters marked in S3 (usually dissolved oxygen concentration in the nitrification tank, total nitrogen concentration in the influent, and dissolved oxygen concentration in the denitrification tank). The output variables are the nitrogen conversion rate and nitrogen reduction efficiency. The nitrogen conversion rate is calculated by the change in total nitrogen concentration in the reaction tank per unit time, and the nitrogen reduction efficiency is calculated by (total nitrogen concentration in the influent - total nitrogen concentration in the effluent) / total nitrogen concentration in the influent × 100%. Secondly, the model structure is constructed. The model adopts a multi-layer feedforward neural network architecture, with the number of input layer nodes and the number of input variables... The system is consistent (3 nodes), with 2 hidden layers. The first hidden layer has 1.8 times the number of nodes as the input layer (5 nodes), and the second hidden layer has 1.2 times the number of nodes as the first layer (6 nodes). The output layer has 2 nodes (corresponding to nitrogen conversion rate and nitrogen reduction efficiency, respectively). The ReLU function is used as the activation function for the hidden layers, and the Sigmoid function is used as the activation function for the output layers (ensuring that the output values ​​are within a reasonable range, with the nitrogen conversion rate output ranging from 0-5 mg / (L•h) and the nitrogen reduction efficiency output ranging from 0-100%). Then proceed... Model training used key feature parameters and corresponding nitrogen conversion rate and nitrogen reduction efficiency data from the past 6 months (approximately 52,560 data sets, collected every 5 minutes). The gradient descent optimization algorithm was used, with a learning rate of 0.001 and 1000 iterations. After each iteration, the mean squared error between the model's predicted and actual values ​​was calculated. Training was stopped when the mean squared error was less than 0.005 and showed no significant decrease after 10 consecutive iterations, at which point the model's weights and bias parameters were determined. Finally, model validation was performed using the training data... One month's worth of data (approximately 8640 data sets) is used as a validation set to verify the model's prediction accuracy. If the prediction error of nitrogen conversion rate is less than 5% and the prediction error of nitrogen reduction efficiency is less than 3%, the model is validated. Subsequently, the set of key feature parameters output from S2 is input into the validated model, and the model is run to obtain real-time predicted values ​​of nitrogen conversion rate and nitrogen reduction efficiency. This step, by constructing an accurate prediction model, can predict changes in key indicators of the nitrogen reduction process in advance, avoid the decline in nitrogen reduction efficiency caused by delayed regulation, and provide a prediction benchmark for the identification of abnormal operating conditions in S5.

[0031] Step S5: Input the prediction results of the dissolved oxygen coupled nitrogen conversion prediction model and the real-time operating parameters of the wastewater treatment system into the SOURATP nitrogen process intelligent early warning and judgment platform, and use the platform to identify and judge abnormal operating conditions in the nitrogen reduction process.

[0032] Specifically, step S5 inputs the prediction results of the dissolved oxygen coupled nitrogen conversion prediction model output from S4, along with the real-time operating parameters of the wastewater treatment system, into the SOURATP nitrogen process intelligent early warning and judgment platform. This enables the identification and judgment of abnormal operating conditions during nitrogen reduction, providing a basis for timely control measures. During implementation, firstly, the data type and source of the input to the platform are clarified. The prediction results include predicted values ​​of nitrogen conversion rate and nitrogen reduction efficiency (from the model output of S4). The real-time operating parameters include real-time data of all 10 parameters collected in S1 (accessed in real-time via an industrial control computer, with a data delay of no more than 10 seconds). Simultaneously, it is ensured that the prediction results and real-time operating parameters are precisely aligned in time (using timestamps as a benchmark, ensuring that each set of prediction results corresponds to the real-time operating parameters at the same point in time). Secondly, the aligned data is input into the SOURATP nitrogen process intelligent early warning and judgment platform. The data analysis module first calculates the deviation between the predicted results and the corresponding indicators in the real-time operating parameters (such as the difference between the real-time nitrogen conversion rate and the predicted nitrogen conversion rate, and the difference between the real-time nitrogen reduction efficiency and the predicted nitrogen reduction efficiency). The deviation calculation uses a combination of absolute and relative differences (the absolute difference reflects the magnitude of the deviation, and the relative difference reflects the degree of the deviation). Then, deviation thresholds are set. The absolute difference thresholds are obtained through statistical analysis of historical abnormal data (the absolute difference threshold for nitrogen conversion rate is 0.5 mg / (L•h), and the absolute difference threshold for nitrogen reduction efficiency is 5%). The relative difference threshold is set to 1. When the absolute difference of any indicator exceeds the corresponding threshold or the relative difference exceeds 10%, an anomaly warning is triggered. The platform's warning analysis module then further analyzes the anomaly that triggered the warning. Combining this with the importance ranking table of key characteristic parameters output by S3, it identifies the core parameters causing the anomaly (e.g., if the nitrogen conversion rate deviation exceeds the standard, parameters with high contribution, such as dissolved oxygen concentration in the nitrification tank and total nitrogen concentration in the influent, are prioritized for investigation). Simultaneously, based on the duration of the deviation (a duration threshold of 15 minutes is set; if the deviation lasts longer than 15 minutes, it is judged as a stable anomaly; otherwise, it is an instantaneous fluctuation) and the magnitude of the deviation... Abnormal operating conditions are classified into three levels: mild abnormality (relative deviation 10%-20%), moderate abnormality (relative deviation 20%-30%), and severe abnormality (relative deviation exceeding 30%). Finally, an abnormal operating condition identification report is generated, which includes information such as the time of abnormality occurrence, abnormal indicators, deviation values, core influencing parameters, and abnormality level. This information is stored in the database and pushed to the wastewater treatment system monitoring terminal. This step, through the system's abnormality identification and analysis, solves the problem that traditional manual monitoring is difficult to detect abnormalities in real time, provides accurate direction for rapid regulation, and avoids water quality exceeding standards due to the expansion of abnormalities.

[0033] Step S6: Based on the analysis results of the SOURATP nitrogen process intelligent early warning and analysis platform, generate an AI monitoring and evaluation report on the nitrogen reduction process in wastewater treatment. The report includes the contribution of calibration characteristic parameters, nitrogen conversion prediction results, and abnormal operating condition identification results.

[0034] Specifically, step S6, based on the analysis results of the SOURATP nitrogen process intelligent early warning and judgment platform output from S5, generates an AI monitoring and evaluation report on the nitrogen reduction process in wastewater treatment. This report comprehensively presents key information about the nitrogen reduction process, providing a complete basis for the optimization of wastewater treatment system operation and management decisions. During implementation, the core content modules of the report are first determined, including three core parts: a key characteristic parameter contribution analysis module, a nitrogen conversion prediction result analysis module, and an abnormal operating condition identification result module. The content of each module is closely integrated with the data from previous steps to ensure the completeness and relevance of the report. In the key characteristic parameter contribution analysis module, the importance ranking table and contribution values ​​output from S3 are integrated, presenting the average contribution value and importance ranking of each calibrated parameter in tabular form. Simultaneously, a trend chart is used to display the contribution change trend of core parameters (contribution value greater than 0.2) over the past 24 hours (e.g., the contribution fluctuation of dissolved oxygen concentration in the nitrification tank at different time periods), analyzing the contribution... The correlation between temperature changes and adjustments to the operation of the wastewater treatment system (such as sludge return ratio adjustments) helps managers understand the dynamic changes in parameter impact. In the nitrogen conversion prediction result analysis module, the predicted values ​​of nitrogen conversion rate and nitrogen reduction efficiency output from S4 for the past 24 hours are summarized and compared with the real-time values ​​for the same period. The prediction accuracy is calculated (prediction accuracy = (1 - |predicted value - real-time value| / real-time value) × 100%), and the average prediction accuracy is calculated (required to be no less than 92%). Simultaneously, a line graph is used to display the changing trends of the predicted and real-time values, marking periods with large prediction deviations (periods exceeding 5%), and analyzing the causes of deviations (such as sudden fluctuations in influent water quality leading to model changes). (Predictive lag); The abnormal operating condition identification results module integrates the abnormal operating condition identification report output by S5, listing all abnormal events that occurred in the past 24 hours in chronological order, including the occurrence and end time of the abnormality, abnormal indicators, core influencing parameters, abnormality level, and handling suggestions (handling suggestions are formulated based on the abnormality level and core parameters, such as fine-tuning the dissolved oxygen concentration in the nitrification tank for mild abnormalities, and urgently adjusting the sludge return ratio and checking the equipment operating status for severe abnormalities). It also statistically analyzes the frequency of abnormal occurrences (categorized by abnormality level) and the handling status (such as the handling measures and effects of handled abnormalities). After the report is generated, it is formatted in a fixed format (PDF format, including report details). The report (including basic information such as generation time, data statistics period, and compiling unit) is formatted to ensure clear data charts and concise and accurate text descriptions, avoiding redundant information. The report is then stored in a database and simultaneously pushed to the terminals of the wastewater treatment plant's technical and management personnel via email, supporting online viewing and download. This step, by generating a comprehensive assessment report, integrates the previously scattered analytical data into a systematic decision-making basis, solving the problem that traditional monitoring can only provide single data points and lacks comprehensive analysis. This helps management personnel fully grasp the status of the nitrogen reduction process, formulate scientific operation optimization plans, and improve the precision of nitrogen reduction management in the wastewater treatment system.

[0035] Preferably, the expression for the SHAP nitrogen feature contribution parsing algorithm is: ,in, The SHAP contribution value of the i-th nitrogen feature parameter. For a feature subset that does not contain the i-th feature parameter, To calibrate the feature parameter set, To determine the total number of parameters in the characteristic parameter set, The nitrogen reduction impact assessment function for the feature subset. The weighting coefficients for the dissolved oxygen concentration parameter are: This is the adjustment coefficient for the total nitrogen concentration parameter.

[0036] Specifically, the SHAP nitrogen feature contribution analysis algorithm quantifies the contribution of each nitrogen feature parameter to the nitrogen reduction process by setting a calculation method, providing a precise calculation basis for clarifying the importance of calibration parameters. During implementation, the basic data required for the algorithm calculation is first determined, including the key feature parameter set selected in step S2 and the nitrogen reduction impact assessment results corresponding to each feature subset. The key feature parameter set typically includes 5-6 parameters, and the feature subsets cover all combinations without a single parameter, ensuring the comprehensiveness of the calculation. Secondly, the technical parameters and numerical settings involved in the calculation process are clarified. The number of feature subsets is determined based on the total number of calibration parameters. If the total number of calibration parameters is 5, then the number of feature subsets is 31. The total number of parameters is consistent with the total number of calibration parameters. The weighting coefficient of the dissolved oxygen concentration parameter is determined through historical data regression analysis, with a value range of 0.3-0.5. The adjustment coefficient of the total nitrogen concentration parameter is set according to the fluctuation range of the influent total nitrogen concentration; when the fluctuation of the influent total nitrogen concentration is less than 20 mg / L, the value is 1. The value is 0.0, 1.2 when the fluctuation is between 20-50 mg / L, and 1.5 when the fluctuation is greater than 50 mg / L. The calculation process proceeds step by step according to a specific logic. First, the nitrogen reduction impact assessment value corresponding to each feature subset without a single parameter is calculated one by one. Then, the assessment value of the feature subset after adding the parameter is calculated. The difference between the two is the local contribution of the parameter in that subset. Then, the weight coefficient is calculated by combining the number of feature subsets and the total number of parameters. The local contributions are weighted and summed to obtain the total contribution value of the parameter. This algorithm accurately distinguishes the degree of influence of different parameters on nitrogen reduction by quantifying the parameter contribution, avoiding the misjudgment of parameter importance caused by subjective judgment. It provides objective data support for determining the parameter importance ranking in step S3, and at the same time ensures that the subsequent model construction can focus on the core parameters, improving the model's computational efficiency and prediction accuracy.

[0037] Preferably, the expression for the dissolved oxygen coupled nitrogen conversion prediction model is: ,in, For nitrogen conversion rate, This refers to the dissolved oxygen concentration in the nitrification tank. Temperature of the nitrification tank This refers to the dissolved oxygen concentration in the denitrification tank. The hydraulic residence time, The sludge return ratio is... The total nitrogen concentration in the influent. These are the nitrification reaction coefficient, temperature influence coefficient, denitrification reaction coefficient, hydraulic retention time coefficient, and sludge return influence coefficient of the model, respectively.

[0038] Specifically, the dissolved oxygen coupled nitrogen conversion prediction model achieves accurate prediction of nitrogen conversion rate by constructing a set model expression, providing a specific calculation method for the prediction step S4. During implementation, the first step was to clarify the range and acquisition method of the model input parameters. The dissolved oxygen concentration in the nitrification tank was obtained from real-time data collected in step S1, with a range of 1.5-4.0 mg / L. The temperature in the nitrification tank was also obtained from real-time data collection, with a range of 15-35℃. The dissolved oxygen concentration in the denitrification tank ranged from 0.2-0.5 mg / L. The hydraulic retention time was calculated using the effective volume of the reaction tank and the influent flow rate, with a range of 8-16 hours. The sludge return ratio ranged from 50%-150%, and the total nitrogen concentration in the influent was obtained from real-time monitoring data, with a range of 20-80 mg / L. Secondly, the numerical settings of various coefficients in the model were determined. The nitrification reaction coefficient was obtained by fitting nitrification experimental data, with a range of 0.02-0.05. The temperature influence coefficient was set based on the relationship between nitrifying bacteria activity and temperature, with a value of 0.06-0.08. The denitrification reaction coefficient was determined through denitrification experiments, with a value of 0.01. -0.03, the hydraulic retention time coefficient is set according to the nitrogen conversion efficiency under different hydraulic retention times, with a value range of 0.8-1.2, and the sludge return influence coefficient is set according to the influence of the sludge return ratio on the sludge concentration, with a value range of 0.1-0.3. The model calculation is performed in a specific order, first calculating the nitrification reaction related terms, the denitrification reaction related terms, and the sludge return influence terms respectively. Among them, the nitrification reaction related terms combine the square of the dissolved oxygen concentration in the nitrification tank with the temperature exponential function, and the denitrification reaction related terms combine the cube of the dissolved oxygen concentration in the denitrification tank with the logarithm of the hydraulic retention time. Finally, the three results are combined according to a specific logic to obtain the nitrogen conversion rate. This model realizes the coupled calculation of dissolved oxygen with other calibration parameters, breaks through the limitations of traditional single parameter prediction, improves the accuracy of nitrogen conversion rate prediction, and the prediction error can be controlled within 5%. It provides core data for the nitrogen reduction efficiency assessment in step S4, and provides an accurate prediction benchmark for subsequent abnormal operating condition identification.

[0039] Preferably, the expression for the nitrogen feature independent component screening algorithm is: ,in, For the j-th independent component, Let be the separation coefficient between the j-th independent component and the k-th original parameter. For the k-th original nitrogen characteristic parameter, This represents the total number of original parameters. Let j be the error correction term for the j-th independent component. The influencing factors of sludge age parameter, This is the weighting factor for the concentration parameter of suspended solids in the mixture.

[0040] Specifically, the nitrogen feature independent component screening algorithm effectively separates and filters the original parameters by constructing an algorithmic expression, providing a specific implementation method for obtaining the key feature parameter set in step S2. During implementation, the requirements and preprocessing methods for the algorithm input data are first clarified. The original nitrogen feature parameters include the 10 parameters collected in step S1. Outliers (such as data exceeding the normal range) are first removed to ensure data validity. The total number of original parameters is fixed at 10. The number of independent components is set according to actual needs, usually 3-5, to ensure coverage of major influencing factors. Secondly, the numerical settings of the calibration parameters in the algorithm are determined. The separation coefficient is calculated iteratively through the independent component analysis algorithm, with a value range of -1.0 to 1.0. The larger the absolute value, the stronger the correlation between the corresponding original parameter and the independent component. The error correction term is set according to the data acquisition accuracy: 0.01 when the acquisition equipment accuracy is ±0.5%, and 0.005 when the accuracy is ±0.2%. The influence factor of the sludge age parameter is determined according to the degree of influence of sludge age on nitrogen reduction. The degree of separation is set, with a value range of 0.2-0.4. The weighting factor for the mixed liquor suspended solids concentration parameter is set according to the relationship between the mixed liquor suspended solids concentration and microbial activity, with a value range of 0.15-0.35. The algorithm implementation process is carried out in stages. First, the original parameters are standardized (non-normalized). Then, the separation coefficient is determined through iterative calculation. Subsequently, the independent components are calculated by combining the error correction term, the sludge age influence factor and the mixed liquor suspended solids concentration weighting factor. The core value of this algorithm lies in extracting independent features that play a dominant role in nitrogen reduction through multi-parameter coupling separation, avoiding multicollinearity interference between original parameters, and reducing the redundancy of the key feature parameter set by more than 40%. This provides technical support for improving the feature screening accuracy in step S2, and lays a high-quality data foundation for subsequent contribution analysis and model construction.

[0041] Preferably, the abnormal operating condition identification expression of the SOURATP nitrogen process intelligent early warning and judgment platform is as follows: ,in, The alert result is indicated by 1 for abnormality and 0 for normality. This represents the predicted nitrogen conversion rate output by the dissolved oxygen coupled nitrogen conversion prediction model. This is the standard value for nitrogen conversion rate. This represents the change in dissolved oxygen concentration. This represents the change in total nitrogen concentration. These are the weights for predicted values, standard values, dissolved oxygen changes, and total nitrogen changes, respectively. This is a correction factor for the specific oxygen consumption rate parameter.

[0042] Specifically, the abnormal operating condition identification function of the SOURATP nitrogen process intelligent early warning and judgment platform can accurately determine abnormal operating conditions by constructing early warning expressions, and provide specific calculation logic for the early warning and judgment in step S5. During implementation, firstly, the source and processing requirements of the input data for the expression are clarified. The predicted nitrogen conversion rate comes from the model output in step S4. The standard value of the nitrogen conversion rate is determined based on the design parameters of the wastewater treatment plant and historical best operating data, with a range of 2.0-4.0 mg / (L•h). The change in dissolved oxygen concentration is calculated by the difference in dissolved oxygen concentration between two consecutive collection periods, with a range of -0.5 to 0.5 mg / L. The change in total nitrogen concentration is also calculated by the difference between consecutive periods, with a range of -10 to 10 mg / L. Secondly, the weights and coefficients of each item in the expression are set. The weight of the predicted value is set according to the reliability of the prediction results, with a range of 0.4-0.6. The weight of the standard value is set with a range of 0.2-0.3. The weight of the dissolved oxygen change is set according to the sensitivity of dissolved oxygen to nitrogen conversion, with a range of 0.15-0.25. The weight of the total nitrogen change is... The value range is 0.1-0.2. The correction coefficient for the specific oxygen consumption rate parameter is set based on microbial activity monitoring data. When the specific oxygen consumption rate is greater than 2.0 mg / (g•h), the value is 1.2; when it is between 1.0 and 2.0 mg / (g•h), the value is 1.0; and when it is less than 1.0 mg / (g•h), the value is 0.8. The calculation process first multiplies each input data with its corresponding weight, then calculates the intermediate result according to a specific logical combination, and then judges the positive or negative sign of the intermediate result through a sign function. Combined with the specific oxygen consumption rate correction coefficient, the warning result identifier is obtained. This expression realizes multi-indicator collaborative warning, breaking through the limitations of traditional single-indicator warning. It can effectively identify instantaneous fluctuations and stable anomalies, and the accuracy of abnormal operating condition identification is improved to over 90%. It provides a reliable technical means for timely detection of nitrogen reduction anomalies in step S5, and at the same time provides accurate anomaly location basis for subsequent formulation of targeted control measures.

[0043] Preferably, the comprehensive index expression for AI monitoring and evaluation of nitrogen reduction in wastewater treatment is: ,in, A comprehensive index for monitoring and evaluating AI. To determine the number of feature parameters, Let be the real-time monitoring value of the i-th calibration feature parameter. For the maximum nitrogen conversion rate, This is the influence coefficient for abnormal operating conditions. This is the baseline coefficient for normal operating conditions. , The evaluation weights are for dissolved oxygen, total nitrogen, and early warning, respectively.

[0044] Specifically, the AI ​​monitoring and evaluation comprehensive index for nitrogen reduction in wastewater treatment achieves a comprehensive assessment of the nitrogen reduction process by constructing a comprehensive index expression, providing core quantitative indicators for generating the assessment report in step S6. During implementation, the acquisition methods and calculation standards for each data item in the expression are first clarified. The number of key characteristic parameters is consistent with the total number of calibration parameters selected in step S2, typically 5-6 items. The SHAP contribution value of each parameter comes from the calculation results in step S3. The real-time monitoring values ​​of parameters are taken from the real-time data collected in step S1. The maximum nitrogen conversion rate is determined based on historical best operating data, with a value range of 4.5-6.0 mg / (L・h). The abnormal operating condition influence coefficient is set according to the abnormality level: 0.2 for mild abnormality, 0.5 for moderate abnormality, and 0.8 for severe abnormality. The normal operating condition baseline coefficient is fixed at 1.0. The dissolved oxygen assessment weight is set based on the core impact of dissolved oxygen on nitrogen reduction, with a value range of 0.4-0.5. The total nitrogen assessment weight ranges from... The weighting ranges from 0.3 to 0.4, while the weighting range for early warning assessment is 0.1 to 0.2. The calculation process is carried out in modules: first, the weighted average of the contribution of the calibration parameters is calculated; then, the relative value of the nitrogen conversion rate is calculated; subsequently, the impact of abnormal operating conditions is calculated; and finally, the comprehensive index is obtained by combining the weights of each assessment item. The comprehensive index ranges from 0 to 100, with scores above 80 being excellent, 60-80 being good, and below 60 being unqualified. The core value of this comprehensive index lies in integrating parameter contribution, prediction results, and early warning information to achieve a multi-dimensional quantitative assessment of the nitrogen reduction process, avoiding the one-sidedness of single-indicator assessments. It provides the core basis for generating a comprehensive assessment report in step S6, while also helping managers quickly grasp the overall status of nitrogen reduction and providing intuitive data support for operational optimization decisions.

[0045] Preferred, such as Figure 2 As shown, step S3 includes the following sub-steps: S31, obtain the calibration feature parameter set output in step S2, and arrange each parameter in the calibration feature parameter set according to the time sequence of the wastewater treatment process to form a time-series feature sequence; S32, input the time-series feature sequence into the feature processing module of the SHAP nitrogen feature contribution analysis algorithm, perform sliding window segmentation on the time-series feature sequence to obtain multiple feature subsequences, each feature subsequence including 5-8 consecutive time-series data points; S33, call the calculation module of the SHAP nitrogen feature contribution analysis algorithm, calculate the local contribution value of different parameters in each feature subsequence to the nitrogen reduction process, and obtain the global contribution value of different parameters by weighted average; S34, sort the global contribution values ​​of different parameters in descending order to generate a calibration feature parameter importance ranking table, which is the output result of step S3.

[0046] Specifically, step S3 clarifies the implementation process of the SHAP nitrogen feature contribution analysis algorithm through four sub-steps, providing an operational basis for accurately calculating the contribution of key feature parameters and determining their importance ranking. During implementation, S31 is executed first, retrieving the set of key feature parameters from the output of step S2. The data for each parameter is arranged according to the time sequence of the wastewater treatment process (with a collection frequency of once every 5 minutes as the time unit), forming a continuous time-series feature sequence. This ensures the sequence covers at least 72 hours of operational data to guarantee the representativeness of the analysis. Next, S32 is performed, inputting the time-series feature sequence into the feature processing module of the SHAP nitrogen feature contribution analysis algorithm. A sliding window segmentation method is used to divide the sequence, with the window size set to include 5-8 consecutive time-series data points. The window sliding step size is consistent with the data collection frequency (5 minutes). This operation decomposes the long time-series sequence into multiple independent... The algorithm first identifies the feature subsequences to facilitate subsequent local contribution calculations. Then, in step S33, the core calculation module is invoked. For each feature subsequence, the local contribution value of each parameter is calculated based on the values ​​of each parameter within the subsequence and the nitrogen reduction efficiency data for the corresponding time period. Weights are then assigned according to the runtime of each subsequence (25-40 minutes per subsequence), and a weighted average is performed to obtain the global contribution value of each parameter. Finally, step S34 sorts the global contribution values ​​of each parameter in descending order, generating a ranking table of key feature parameters' importance, including parameter name, global contribution value, and ranking position. This table serves as the final output of step S3. This step-by-step implementation effectively reduces the interference of a single data segment on the calculation results, keeping the contribution calculation error within 3%, and providing accurate data support for subsequent model construction focusing on core parameters.

[0047] Preferred, such as Figure 3 As shown, step S4 includes the following sub-steps: S41, extract the top 8-10 most important parameters from the importance ranking table of the calibration feature parameters output in step S3, and use them as the input parameter set for the dissolved oxygen coupled nitrogen conversion prediction model; S42, construct the network structure of the dissolved oxygen coupled nitrogen conversion prediction model, which includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is consistent with the number of parameters in the input parameter set. The hidden layer includes 2-3 sub-layers, and the number of nodes in each sub-layer is 1.5-2 times the number of nodes in the input layer. The output layer includes 2 nodes, corresponding to the nitrogen conversion rate and nitrogen reduction efficiency, respectively; S43, train the constructed dissolved oxygen coupled nitrogen conversion prediction model using historical nitrogen reduction data from wastewater treatment, and adjust the coefficients of different terms of the model using the gradient descent algorithm until the prediction error of the model converges to below a preset threshold; S44, input the calibration feature parameter set output in step S2 into the trained dissolved oxygen coupled nitrogen conversion prediction model, run the model to obtain the predicted values ​​of nitrogen conversion rate and nitrogen reduction efficiency, and use them as the prediction results of step S4.

[0048] Specifically, step S4 involves a four-step process for constructing and running a dissolved oxygen coupled nitrogen conversion prediction model to ensure the model accurately outputs predictions for nitrogen conversion rate and nitrogen reduction efficiency. During implementation, first, step S41 is executed. From the key feature parameter importance ranking table output in step S3, the top 8-10 most important parameters (typically including core parameters such as dissolved oxygen concentration in the nitrification tank and total nitrogen concentration in the influent) are extracted and determined as the input parameter set for the dissolved oxygen coupled nitrogen conversion prediction model, ensuring that the input parameters comprehensively cover factors that have a key impact on nitrogen conversion. Next, step S42 is performed to construct the model's network structure. The number of nodes in the input layer is consistent with the number of parameters in the input parameter set (8-10 nodes). The hidden layer has 2-3 sub-layers. The first sub-layer has 1.5-2 times the number of nodes in the input layer (12-20 nodes), and the number of nodes in subsequent sub-layers decreases by 0.8 times the number of nodes in the previous layer (9-16 nodes). The output layer has 2 nodes corresponding to... The nitrogen conversion rate and nitrogen reduction efficiency were determined, with the hidden layer using the ReLU activation function and the output layer using the Sigmoid activation function. Then, step S43 was performed, using historical nitrogen reduction data from the past six months of wastewater treatment (including parameter data corresponding to the input parameter set and actual nitrogen conversion rate and nitrogen reduction efficiency data for the same period, totaling approximately 52,560 data sets) to train the model. Gradient descent optimization was used to adjust the model coefficients, with a learning rate set to 0.001 and 1000 iterations. Training stopped when the model prediction error (mean squared error) converged to below 0.005. Finally, step S44 was executed, inputting the key feature parameter set output from step S2 into the trained model, running the model to obtain real-time predicted values ​​for nitrogen conversion rate and nitrogen reduction efficiency, which were used as the prediction results from step S4. This step-by-step process ensures sufficient model training, improving prediction accuracy to over 92%, and providing a reliable prediction benchmark for subsequent abnormal operating condition identification.

[0049] Preferred, such as Figure 4As shown, step S5 includes the following sub-steps: S51, collecting real-time operating parameters of the wastewater treatment system, including real-time dissolved oxygen concentration in the nitrification tank, real-time dissolved oxygen concentration in the denitrification tank, real-time sludge return ratio, real-time mixed liquor return ratio, and real-time total nitrogen concentration in the effluent; S52, aligning the prediction results output in step S4 with the real-time operating parameters to ensure consistency in the time dimension, forming an aligned comprehensive dataset; S53, inputting the comprehensive dataset into the data analysis module of the SOURATP nitrogen process intelligent early warning and judgment platform, which calculates the deviation rate and deviation trend by comparing the deviation between the prediction results and the real-time operating parameters; S54, calling the early warning and judgment module of the SOURATP nitrogen process intelligent early warning and judgment platform, identifying the abnormal operating conditions in the nitrogen reduction process based on the deviation rate, deviation trend, and preset early warning threshold, determining the severity of the abnormal operating conditions, and generating early warning and judgment results.

[0050] Specifically, step S5 clarifies the abnormal operating condition identification process of the SOURATP nitrogen process intelligent early warning and judgment platform through four sub-steps, so as to achieve accurate judgment of abnormalities in the nitrogen reduction process. During implementation, S51 is executed first to collect real-time operating parameters of the wastewater treatment system. These parameters include real-time dissolved oxygen concentration in the nitrification tank (collected every 5 minutes, with a measurement accuracy of ±0.1 mg / L), real-time dissolved oxygen concentration in the denitrification tank (same as the nitrification tank parameter standard), real-time sludge return ratio (calculated using an electromagnetic flowmeter and variable frequency pump frequency, with an accuracy of ±0.2%), real-time mixed liquor return ratio (same as the sludge return ratio parameter standard), and real-time total nitrogen concentration in the effluent (collected by an online monitoring instrument, with an accuracy of ±0.5 mg / L), ensuring that parameter collection covers key nitrogen reduction stages. Next, S52 is performed to align the predicted results (nitrogen conversion rate prediction and nitrogen reduction efficiency prediction) output from step S4 with the real-time operating parameters collected in S51 in terms of time dimension. Using the timestamp of the predicted results as a benchmark, real-time operating parameters at the same time point are matched to form an aligned comprehensive dataset, avoiding analytical errors caused by time deviations. Finally, S53 is performed to... The aggregated dataset is input into the platform's data analysis module. This module calculates the deviation between the predicted results and the corresponding real-time operating parameters (absolute deviation = |predicted value - real-time value|, relative deviation = absolute deviation / real-time value × 100%), and analyzes the deviation trend over five consecutive acquisition cycles (25 minutes) to determine whether the deviation is continuously increasing, fluctuating steadily, or gradually decreasing. Finally, step S54 is executed, calling the platform's early warning and judgment module. Based on the deviation rate (relative deviation), deviation trend, and preset early warning thresholds calculated in step S53 (a relative deviation exceeding 10% triggers an early warning), it identifies abnormal operating conditions (such as dissolved oxygen anomalies, reflux ratio anomalies, etc.) and classifies the severity of the anomaly based on the duration of the deviation (more than 15 minutes is considered a stable anomaly) and the deviation amplitude (mild: relative deviation 10%-20%, moderate: 20%-30%, severe: exceeding 30%), generating early warning and judgment results including the time of occurrence, type, and severity of the anomaly. This step-by-step process enables the response time for abnormal operating condition identification to be controlled within 10 minutes, and the identification accuracy to be improved to over 90%, providing a precise basis for timely control measures.

[0051] The SHAP nitrogen feature contribution analysis algorithm in this invention is a technical tool based on the SHAP value principle, specifically used to quantify the influence of key feature parameters related to nitrogen reduction in the wastewater treatment process on the nitrogen reduction effect. SHAP stands for SHApley Additive exPlanations. The implementation process is based on the key feature parameter set selected in step S2. First, continuous time-series data of at least 72 hours are extracted at a fixed acquisition frequency of 5 minutes / time. These data are then organized into a complete time-series feature sequence in chronological order to ensure that the data covers the typical cycle of stable operation of the wastewater treatment system. Next, the time-series feature sequence is processed using a sliding window segmentation method. The window size is set to include 5-8 continuous data points, and the sliding step size is consistent with the data acquisition frequency (i.e., 5 minutes), thereby decomposing the long time-series into multiple independent feature subsequences. Subsequently, for each feature subsequence, the local contribution of each parameter in the subsequence is calculated by combining the actual values ​​of each parameter in the subsequence with the nitrogen reduction efficiency data of the corresponding time period. Then, different weights are assigned according to the actual running time (25-40 minutes) of each subsequence, and the global contribution of each parameter is calculated by weighted average. Finally, the global contribution of all calibrated parameters is sorted in descending order to generate a key feature parameter importance ranking table including parameter name, global contribution value, and ranking position. The algorithm's function is to accurately distinguish the influence weight of different calibration parameters on the nitrogen reduction process. For example, it clearly defines the contribution differences of core parameters such as dissolved oxygen concentration in the nitrification tank and total nitrogen concentration in the influent, avoiding subjectivity in the judgment of parameter importance. It strictly controls the contribution calculation error to within 3%, providing objective and reliable data support for the subsequent dissolved oxygen-coupled nitrogen conversion prediction model to select core input parameters. This reduces redundant parameters in the model calculation, lowers the computational load, and ultimately improves the accuracy of the entire AI monitoring and evaluation system in analyzing the nitrogen reduction process.

[0052] The dissolved oxygen coupled nitrogen conversion prediction model in this invention is a nonlinear prediction model that integrates dissolved oxygen parameters with other multi-dimensional key feature parameters to predict nitrogen conversion rate and nitrogen reduction efficiency in wastewater treatment. Its implementation involves four steps: First, extract the top 8-10 most important parameters from the key feature parameter importance ranking table output in step S3 (typically including dissolved oxygen concentration in the nitrification tank, total nitrogen concentration in the influent, dissolved oxygen concentration in the denitrification tank, sludge return ratio, etc.), and determine these parameters as the input parameter set of the model; Second, construct the model's network structure, with the number of nodes in the input layer exactly matching the number of parameters in the input parameter set (8-10 nodes). The hidden layer has 2-3 sub-network layers. The number of nodes in the first sub-network layer is 1.5-2 times the number of nodes in the input layer (i.e., 12-20 nodes), and the number of nodes in subsequent sub-network layers decreases sequentially by 0.8 times the number of nodes in the previous layer (e.g., the second layer has 9-16 nodes). The output layer has 2 nodes, corresponding to the nitrogen conversion rate and nitrogen reduction efficiency predictions, respectively. The model is trained using several steps. First, it measures key performance indicators (KPIs). The hidden layer uses ReLU activation to address the vanishing gradient problem, while the output layer uses Sigmoid activation to ensure the predictions are within a reasonable range (nitrogen conversion rate 0-5 mg / (L•h), nitrogen reduction efficiency 0-100%). Second, it trains the model using historical nitrogen reduction data from the past six months (approximately 52,560 sets, collected every 5 minutes). Gradient descent optimization is used to adjust the model's weights and bias parameters, with a learning rate of 0.001 and 1000 iterations, until the model's mean square error converges below 0.005. Third, the real-time key feature parameter set output from step S2 is input into the trained model. Running the model yields real-time predicted nitrogen conversion rate and nitrogen reduction efficiency. This model aims to predict nitrogen conversion rate and nitrogen reduction efficiency in advance, achieving a stable prediction accuracy exceeding 92%, providing advance warning for nitrogen reduction process control. Breaking through the limitations of traditional prediction models that rely on only a single parameter, this method achieves multi-parameter coupled prediction, providing an accurate prediction benchmark for the SOURATP nitrogen process intelligent early warning and judgment platform. This effectively avoids the decline in nitrogen reduction efficiency caused by control lag and ensures the dynamic controllability of the nitrogen reduction process in the wastewater treatment system.

[0053] The nitrogen feature independent component screening algorithm in this invention is a technical method based on the principle of independent component analysis to separate and screen independent feature parameters that significantly affect the nitrogen reduction process from multi-dimensional raw parameters related to nitrogen reduction in wastewater treatment. Its implementation process is as follows: First, the 10 raw parameters collected in step S1 (influent total nitrogen concentration, effluent total nitrogen concentration, dissolved oxygen concentration in the nitrification tank, etc.) are preprocessed to remove outliers caused by equipment failure or abnormal data transmission (such as dissolved oxygen concentration <0 mg / L or >20 mg / L, temperature <0℃ or >50℃) to ensure the validity and integrity of the raw data; then, the preprocessed valid data is standardized (only data translation and scaling are used to bring the parameters to the same order of magnitude, without normalization, to avoid data loss); then, the separation coefficient between each raw parameter and the potential independent component is determined through iterative calculation. The separation coefficient ranges from -1.0 to 1.0, with a larger absolute value indicating a higher correlation between the corresponding raw parameter and the potential independent component. The stronger the correlation between quantities, the better. Subsequently, error correction terms, sludge age influencing factors, and mixed liquor suspended solids concentration weighting factors are introduced to optimize the calculation results. The error correction term is set according to the accuracy of the data acquisition equipment (0.005 for ±0.2% accuracy, 0.01 for ±0.5% accuracy). The sludge age influencing factor ranges from 0.2 to 0.4, and the mixed liquor suspended solids concentration weighting factor ranges from 0.15 to 0.35. These parameters are adjusted to obtain 3-5 independent components that reflect the core influencing factors of nitrogen reduction. Finally, the correlation coefficient between each independent component and nitrogen reduction efficiency is calculated, and a correlation coefficient threshold of 0.6 is set. The original parameters corresponding to independent components with correlation coefficients greater than 0.6 are selected to form the final set of key feature parameters. The algorithm's function is to remove multicollinearity interference between original parameters, reduce the influence of invalid and redundant parameters, and reduce parameter redundancy by more than 40%. This provides high-quality, low-redundancy calibration parameter data for the subsequent SHAP nitrogen feature contribution analysis algorithm, avoiding interference from invalid parameters on the contribution calculation results. It ensures the analytical accuracy of the entire AI monitoring and evaluation system from the data source, laying a solid data foundation for subsequent model construction and early warning judgment.

[0054] The SOURATP nitrogen process intelligent early warning and judgment platform in this invention is an intelligent platform that integrates prediction results and real-time operation data, and is specifically used to identify abnormal operating conditions and judge the severity of abnormalities in the nitrogen reduction process of wastewater treatment. SOURATP stands for Specific Oxygen Uptake Rate linked to Adenosine Triphosphate. The process involves multi-dimensional data acquisition and analysis: First, real-time operating parameters of the wastewater treatment system are collected, including dissolved oxygen concentration in the nitrification tank (collected every 5 minutes, with a measurement accuracy of ±0.1 mg / L), dissolved oxygen concentration in the denitrification tank (same as the nitrification tank parameter standard), sludge return ratio (calculated using an electromagnetic flowmeter and variable frequency pump installed on the return pipe, with an accuracy of ±0.2%), mixed liquor return ratio (same as the sludge return ratio parameter standard), and total nitrogen concentration in the effluent (collected using an online UV spectrophotometer, with an accuracy of ±0.5 mg / L), ensuring that the parameters cover key aspects of nitrogen reduction. Next, the predicted nitrogen conversion rate and nitrogen reduction efficiency output from step S4 are aligned with the real-time operating parameters in terms of time. Using the timestamp of the prediction results as a benchmark, real-time parameters at the same time point are matched to form a comprehensive dataset including both predicted and real-time data, avoiding analytical errors caused by time deviations. Finally, the comprehensive dataset is... The data analysis module of the input platform calculates the absolute deviation (|predicted value - real-time value|) and relative deviation (absolute deviation / real-time value × 100%) between the predicted value and the corresponding real-time value. It also analyzes the deviation trend over five consecutive data collection cycles (25 minutes in total) to determine whether the deviation is continuously increasing, fluctuating steadily, or gradually decreasing. Finally, the platform's early warning and judgment module is invoked. Based on a preset early warning threshold (a relative deviation exceeding 10% triggers an early warning), combined with the deviation duration (more than 15 minutes indicates a stable anomaly, otherwise an instantaneous fluctuation) and deviation amplitude, the abnormal operating conditions are classified into three levels: mild anomaly (relative deviation 10%-20%), moderate anomaly (20%-30%), and severe anomaly (more than 30%). Simultaneously, the core influencing parameters causing the anomaly are located using the parameter importance ranking table from step S3. The platform generates an early warning and judgment result including the anomaly occurrence time, anomaly type, core influencing parameters, and severity. The platform's function is to identify abnormal operating conditions in the nitrogen reduction process in real time, quickly locate core influencing factors, control the early warning response time within 10 minutes, and achieve an anomaly identification accuracy rate exceeding 90%. Its significance lies in completely solving the problems of delayed anomaly detection and difficulty in locating the cause under traditional manual monitoring methods. It provides accurate basis for sewage treatment system operation and maintenance personnel to take timely and targeted control measures (such as fine-tuning the dissolved oxygen concentration in the nitrification tank and adjusting the sludge return ratio), effectively avoiding the decline in nitrogen reduction efficiency or even the failure of effluent quality to meet standards due to the expansion of abnormal operating conditions, and ensuring the stable and efficient operation of the nitrogen reduction process of the sewage treatment system.

[0055] like Figure 5 As shown, an AI monitoring and evaluation system for nitrogen reduction in wastewater treatment is described. This system, applied to an AI monitoring and evaluation method for nitrogen reduction in wastewater treatment, includes the following units: a multi-dimensional nitrogen parameter acquisition and transmission unit, which collects multi-dimensional parameters during wastewater treatment, including influent total nitrogen concentration, effluent total nitrogen concentration, dissolved oxygen concentration in the nitrification tank, dissolved oxygen concentration in the denitrification tank, sludge return ratio, mixed liquor return ratio, nitrification tank temperature, denitrification tank temperature, hydraulic retention time, and sludge age, and transmits the collected parameters to the multi-dimensional nitrogen parameter acquisition and transmission unit via industrial Ethernet; a nitrogen feature independent component screening and processing unit, connected to the multi-dimensional nitrogen parameter acquisition and transmission unit, receiving the transmitted multi-dimensional parameters, performing feature separation and screening on the multi-dimensional parameters using a nitrogen feature independent component screening algorithm, and outputting a calibration feature parameter set to the SHAP nitrogen feature contribution analysis unit; and a SHAP nitrogen feature contribution analysis unit, connected to the nitrogen feature independent component screening and processing unit, receiving the output calibration feature parameter set, and calculating different calibration features using the SHAP nitrogen feature contribution analysis algorithm. The contribution values ​​of characteristic parameters are determined, a parameter importance ranking table is generated, and transmitted to the dissolved oxygen coupled nitrogen conversion prediction unit. The dissolved oxygen coupled nitrogen conversion prediction unit, connected to the SHAP nitrogen characteristic contribution analysis unit, receives the transmitted parameter importance ranking table, constructs a dissolved oxygen coupled nitrogen conversion prediction model based on the table, predicts the nitrogen conversion rate and nitrogen reduction efficiency, and outputs the prediction results to the SOURATP nitrogen process intelligent early warning and judgment unit. The SOURATP nitrogen process intelligent early warning and judgment unit, connected to both the dissolved oxygen coupled nitrogen conversion prediction unit and the multi-dimensional nitrogen parameter acquisition and transmission unit, receives the prediction results and real-time operating parameters, identifies abnormal operating conditions through the SOURATP nitrogen process intelligent early warning and judgment platform, generates judgment results, and transmits them to the AI ​​monitoring and evaluation report generation unit. The AI ​​monitoring and evaluation report generation unit, connected to the SOURATP nitrogen process intelligent early warning and judgment unit, receives the transmitted judgment results, integrates and calibrates the contribution values ​​of characteristic parameters, nitrogen conversion prediction results, and abnormal operating condition identification results, and generates and outputs an AI monitoring and evaluation report on the nitrogen reduction process in wastewater treatment.

[0056] An AI-based monitoring and evaluation method and system for nitrogen reduction in wastewater treatment is proposed. This method collects multi-dimensional parameters related to nitrogen reduction in wastewater treatment, accurately filters and calibrates parameters using a nitrogen feature independent component screening algorithm, and then clarifies the importance ranking of each parameter for nitrogen reduction using the SHAP nitrogen feature contribution analysis algorithm. The entire process requires no manual sampling or offline detection, and can capture dynamic changes in parameters in real time, significantly improving the real-time performance of nitrogen reduction monitoring and solving the evaluation lag problem caused by reliance on manual methods in existing monitoring. Furthermore, it integrates multiple technologies such as feature screening, contribution analysis, and predictive modeling, using multi-algorithm collaborative analysis of the comprehensive impact of multi-dimensional parameters on nitrogen reduction, rather than the one-sided analysis of a single algorithm. This comprehensively presents key information about the nitrogen reduction process, significantly improving the comprehensiveness and accuracy of monitoring and evaluation results, overcoming the shortcomings of existing technologies that rely on a single analytical dimension and fail to reflect the comprehensive effects of parameters.

[0057] This invention utilizes a dissolved oxygen coupled nitrogen conversion prediction model to accurately predict nitrogen conversion rate and nitrogen reduction efficiency. Combined with the SOURATP nitrogen process intelligent early warning and judgment platform, the prediction results are integrated with real-time operating parameters to achieve early identification and accurate judgment of abnormal operating conditions during nitrogen reduction, improving the timeliness of abnormal condition warnings. Furthermore, after an anomaly occurs, it can quickly locate influencing factors based on the contribution of calibrated parameters, providing a clear direction for the control of the wastewater treatment system. This helps to take timely and targeted measures, improving the accuracy and efficiency of anomaly response, avoiding a decline in nitrogen reduction efficiency or substandard effluent quality, and overcoming the shortcomings of existing technologies such as weak anomaly identification capabilities and difficulty in timely control.

[0058] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0059] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An AI monitoring and evaluation method for nitrogen reduction in wastewater treatment, characterized in that, include: Step S1: Collect multi-dimensional parameters related to nitrogen reduction during wastewater treatment. These parameters include influent total nitrogen concentration, effluent total nitrogen concentration, dissolved oxygen concentration in the nitrification tank, dissolved oxygen concentration in the denitrification tank, sludge return ratio, mixed liquor return ratio, nitrification tank temperature, denitrification tank temperature, hydraulic retention time, and sludge age. Step S2: Input the collected multi-dimensional parameters into a nitrogen feature independent component screening algorithm. This algorithm performs feature separation and screening on the multi-dimensional parameters to obtain a set of calibration feature parameters that significantly affect nitrogen reduction. Step S3: Input the calibration feature parameter set into the SHAP nitrogen feature contribution analysis algorithm. This algorithm calculates the contribution value of different calibration feature parameters to the nitrogen reduction process and determines the importance ranking of different calibration feature parameters. Sequence; Step S4: Based on the calibrated feature parameter set and the contribution values ​​of different parameters, construct a dissolved oxygen coupled nitrogen conversion prediction model, and use this model to predict the nitrogen conversion rate and nitrogen reduction efficiency in the wastewater treatment process, and output the prediction results; Step S5: Input the prediction results of the dissolved oxygen coupled nitrogen conversion prediction model and the real-time operating parameters of the wastewater treatment system into the SOURATP nitrogen process intelligent early warning and judgment platform, and use the platform to identify and judge abnormal operating conditions in the nitrogen reduction process; Step S6: Based on the judgment results of the SOURATP nitrogen process intelligent early warning and judgment platform, generate an AI monitoring and evaluation report on the nitrogen reduction process in wastewater treatment, the report including the contribution of calibrated feature parameters, nitrogen conversion prediction results, and abnormal operating condition identification results; The abnormal operating condition identification expression of the SOURATP nitrogen process intelligent early warning and judgment platform is as follows: ， in, The alert result is indicated by 1 for abnormality and 0 for normality. This represents the predicted nitrogen conversion rate output by the dissolved oxygen coupled nitrogen conversion prediction model. This is the standard value for nitrogen conversion rate. This represents the change in dissolved oxygen concentration. This represents the change in total nitrogen concentration. These are the weights for predicted values, standard values, dissolved oxygen changes, and total nitrogen changes, respectively. This is a correction factor for the specific oxygen consumption rate parameter.

2. The AI ​​monitoring and evaluation method for nitrogen reduction in wastewater treatment according to claim 1, characterized in that, The expression for the SHAP nitrogen feature contribution parsing algorithm is as follows: ， in, Let be the SHAP contribution value of the i-th nitrogen feature parameter, S be the feature subset excluding the i-th feature parameter, F be the calibration feature parameter set, M be the total number of parameters in the calibration feature parameter set, and f(·) be the nitrogen reduction impact assessment function of the feature subset. The weighting coefficients for the dissolved oxygen concentration parameter are: This is the adjustment coefficient for the total nitrogen concentration parameter.

3. The AI ​​monitoring and evaluation method for nitrogen reduction in wastewater treatment according to claim 1, characterized in that, The expression for the dissolved oxygen coupled nitrogen conversion prediction model is as follows: ， Where, r TN For nitrogen conversion rate, This refers to the dissolved oxygen concentration in the nitrification tank. Temperature of the nitrification tank The dissolved oxygen concentration in the denitrification tank is HRT, which is the hydraulic retention time, and R is R. S TN is the sludge return ratio. in The total nitrogen concentration in the influent. These are the nitrification reaction coefficient, temperature influence coefficient, denitrification reaction coefficient, hydraulic retention time coefficient, and sludge return influence coefficient of the model, respectively.

4. The AI ​​monitoring and evaluation method for nitrogen reduction in wastewater treatment according to claim 1, characterized in that, The expression for the nitrogen feature independent component screening algorithm is as follows: ， Among them, s j For the j-th independent component, Let be the separation coefficient between the j-th independent component and the k-th original parameter. Let N be the k-th original nitrogen feature parameter, and N be the total number of original parameters. Let j be the error correction term for the j-th independent component. The influencing factors of sludge age parameter, This is the weighting factor for the concentration parameter of suspended solids in the mixture.

5. The AI ​​monitoring and evaluation method for nitrogen reduction in wastewater treatment according to claim 1, characterized in that, The comprehensive index expression for AI monitoring and evaluation of nitrogen reduction in wastewater treatment is as follows: ， in, AI is a comprehensive index for monitoring and evaluation, and K is the number of calibrated characteristic parameters. Let be the real-time monitoring value of the i-th calibration feature parameter. For the maximum nitrogen conversion rate, This is the influence coefficient for abnormal operating conditions. This is the baseline coefficient for normal operating conditions. The evaluation weights are for dissolved oxygen, total nitrogen, and early warning, respectively.

6. The AI ​​monitoring and evaluation method for nitrogen reduction in wastewater treatment according to claim 1, characterized in that, Step S3 includes the following sub-steps: S31, obtain the calibration feature parameter set output by step S2, and arrange each parameter in the calibration feature parameter set according to the time sequence of the sewage treatment process to form a time sequence feature sequence; S32. Input the time-series feature sequence into the feature processing module of the SHAP nitrogen feature contribution analysis algorithm, perform sliding window segmentation on the time-series feature sequence to obtain multiple feature subsequences, each feature subsequence including 5-8 consecutive time-series data points; S33. Call the calculation module of the SHAP nitrogen feature contribution analysis algorithm, calculate the local contribution value of different parameters in each feature subsequence to the nitrogen reduction process, and obtain the global contribution value of different parameters by weighted average; S34. Sort the global contribution values ​​of different parameters in descending order to generate a ranking table of the importance of calibrated feature parameters, as the output result of step S3.

7. The AI ​​monitoring and evaluation method for nitrogen reduction in wastewater treatment according to claim 1, characterized in that, Step S4 includes the following sub-steps: S41, extract the top 8-10 most important parameters from the importance ranking table of the calibration feature parameters output in step S3, and use them as the input parameter set for the dissolved oxygen coupled nitrogen conversion prediction model; S42, construct the network structure of the dissolved oxygen coupled nitrogen conversion prediction model, which includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is consistent with the number of parameters in the input parameter set. The hidden layer includes 2-3 sub-layers, and the number of nodes in each sub-layer is 1.5-2 times the number of nodes in the input layer. The output layer includes 2 nodes, corresponding to the nitrogen conversion rate and nitrogen reduction efficiency, respectively; S43, train the constructed dissolved oxygen coupled nitrogen conversion prediction model using historical nitrogen reduction data from wastewater treatment, and adjust the coefficients of different terms of the model using the gradient descent algorithm until the prediction error of the model converges to below a preset threshold; S44, input the calibration feature parameter set output in step S2 into the trained dissolved oxygen coupled nitrogen conversion prediction model, run the model to obtain the predicted values ​​of nitrogen conversion rate and nitrogen reduction efficiency, and use them as the prediction results of step S4.

8. The AI ​​monitoring and evaluation method for nitrogen reduction in wastewater treatment according to claim 1, characterized in that, Step S5 includes the following sub-steps: S51, collecting real-time operating parameters of the wastewater treatment system, including real-time dissolved oxygen concentration in the nitrification tank, real-time dissolved oxygen concentration in the denitrification tank, real-time sludge return ratio, real-time mixed liquor return ratio, and real-time total nitrogen concentration in the effluent; S52, aligning the prediction results output in step S4 with the real-time operating parameters to ensure consistency in the time dimension, forming an aligned comprehensive dataset; S53, inputting the comprehensive dataset into the data analysis module of the SOURATP nitrogen process intelligent early warning and judgment platform, which calculates the deviation rate and deviation trend by comparing the deviation between the prediction results and the real-time operating parameters; S54, calling the early warning and judgment module of the SOURATP nitrogen process intelligent early warning and judgment platform, identifying the abnormal operating conditions in the nitrogen reduction process based on the deviation rate, deviation trend, and preset early warning thresholds, determining the severity of the abnormal operating conditions, and generating early warning and judgment results.

9. An AI monitoring and evaluation system for nitrogen reduction in wastewater treatment, characterized in that, The system is applied to the AI ​​monitoring and evaluation method for nitrogen reduction in wastewater treatment as described in claim 1, and includes the following unit: a multi-dimensional nitrogen parameter acquisition and transmission unit, which is used to acquire multi-dimensional parameters such as total nitrogen concentration in influent, total nitrogen concentration in effluent, dissolved oxygen concentration in nitrification tank, dissolved oxygen concentration in denitrification tank, sludge return ratio, mixed liquor return ratio, nitrification tank temperature, denitrification tank temperature, hydraulic retention time, and sludge age during the wastewater treatment process, and transmits the acquired parameters to the multi-dimensional nitrogen parameter acquisition and transmission unit via industrial Ethernet; The nitrogen feature independent component screening and processing unit is connected to the multi-dimensional nitrogen parameter acquisition and transmission unit. It receives the transmitted multi-dimensional parameters, performs feature separation and screening on the multi-dimensional parameters using the nitrogen feature independent component screening algorithm, and outputs a set of calibrated feature parameters to the SHAP nitrogen feature contribution analysis unit. The SHAP nitrogen feature contribution analysis unit is connected to the nitrogen feature independent component screening and processing unit. It receives the output set of calibrated feature parameters, calculates the contribution value of different calibrated feature parameters using the SHAP nitrogen feature contribution analysis algorithm, generates a parameter importance ranking table, and transmits it to the dissolved oxygen coupled nitrogen conversion prediction unit. The dissolved oxygen coupled nitrogen conversion prediction unit is connected to the SHAP nitrogen feature contribution analysis unit. It receives the transmitted parameter importance ranking table, constructs a dissolved oxygen coupled nitrogen conversion prediction model based on the table, predicts the nitrogen conversion rate and nitrogen reduction efficiency, and outputs the prediction results to the SOURATP nitrogen process intelligent early warning and judgment unit. The SOURATP nitrogen process intelligent early warning and judgment unit is connected to both the dissolved oxygen coupled nitrogen conversion prediction unit and the multi-dimensional nitrogen parameter acquisition and transmission unit. It receives prediction results and real-time operating parameters, identifies abnormal operating conditions through the SOURATP nitrogen process intelligent early warning and judgment platform, generates judgment results, and transmits them to the AI ​​monitoring and evaluation report generation unit. The AI ​​monitoring and evaluation report generation unit is connected to the SOURATP nitrogen process intelligent early warning and judgment unit. It receives the transmitted judgment results, integrates the contribution of calibrated characteristic parameters, nitrogen conversion prediction results, and abnormal operating condition identification results, and generates and outputs an AI monitoring and evaluation report on the nitrogen reduction process in wastewater treatment.

Images