Quality detection and evaluation method for terminal effluent carbon source of sewage treatment plant

Abstract

The invention provides a sewage treatment plant terminal effluent carbon source quality detection and evaluation method, which realizes full-flow dynamic evaluation and regulation of carbon source quality through on-line monitoring and intelligent algorithm fusion. According to the method, an online water quality full-spectrum detector is used for collecting original spectrum data flow, and after preprocessing such as variational mode decomposition denoising and mutual information feature selection, organic matter content quantification, variation trend analysis and anomaly detection are completed in combination with algorithms such as a support vector machine and an autoregressive moving average model. An entropy weight method is introduced to dynamically adjust the weight of the evaluation model, model parameters are optimized based on a gradient descent algorithm, a process adjustment instruction is generated through reinforcement learning and fuzzy logic, and an automatic system is linked to execute regulation and control. According to the method, the problems of hysteresis and singleness of traditional offline analysis are solved, multi-dimensional real-time evaluation, abnormal quick response and process dynamic optimization of the quality of the carbon source are realized, the sewage treatment efficiency and the effluent quality stability are improved, and a technical support is provided for continuous standard reaching of the quality of the carbon source.

Description

Technical Field

[0001] The present invention relates to the field of effluent carbon source quality inspection and evaluation, and in particular to a method for inspecting and evaluating the quality of effluent carbon sources at a sewage treatment plant terminal. Background Art

[0002] The quality of carbon sources in the terminal effluent of sewage treatment plants directly affects the effectiveness of subsequent treatment processes and the compliance of water quality standards. Traditional evaluation methods mostly rely on offline sampling and analysis, which have problems such as data lag and single coverage dimension, making it difficult to dynamically reflect changes in carbon source quality in real time. With the development of online monitoring technology and intelligent algorithms, there is an urgent need for an automated evaluation method that integrates real-time data collection, multi-dimensional analysis and dynamic regulation to solve the defects of existing technologies such as insufficient data timeliness, delayed abnormal response, and poor flexibility of evaluation models. Among the existing technologies, spectral analysis, time series models, machine learning algorithms, etc. have been initially applied in the field of water quality monitoring, but the full-process automated algorithm system for carbon source quality evaluation has not yet matured, especially in the collaborative mechanism of multi-source data fusion, anomaly detection and dynamic model updating. There is a technical gap; Therefore, it is necessary to build a complete technical solution covering data collection, feature extraction, anomaly determination, model optimization and process control to improve the accuracy and real-time performance of carbon source quality evaluation of terminal effluent from sewage treatment plants. Summary of the Invention

[0003] The main purpose of the present invention is to provide a method for inspecting and evaluating the carbon source quality of the terminal effluent of a sewage treatment plant, so as to solve the problem that the full-process automated algorithm system for carbon source quality evaluation is not yet mature, especially in the collaborative mechanism of multi-source data fusion, anomaly detection and dynamic model updating.

[0004] To solve the above technical problems, the technical solution adopted by the present invention is: a method for inspecting and evaluating the quality of carbon sources in the terminal effluent of a sewage treatment plant, the method comprising: Obtain the original spectral data stream of the organic matter content in the terminal outlet water through the online water quality full spectrum detector equipment; Obtaining preliminary quantitative results of organic matter content based on the raw spectral data stream; Based on the preliminary quantitative results, obtaining change trend data of organic matter content; If the change trend data exceeds the preset threshold range, the anomaly detection mechanism is triggered and the anomaly determination result is output; According to the abnormality determination result, an adapted evaluation parameter combination is obtained to update the weight distribution of the evaluation model, and the carbon source quality of the current terminal effluent is scored in multiple dimensions using the updated evaluation model weight distribution to obtain a comprehensive quality score value; Based on the comprehensive quality score, obtain process adjustment parameter suggestions corresponding to the score and output specific process control instructions; According to the process control instructions, the automation control system of the sewage treatment plant is linked to obtain feedback data after execution; The feedback data after the execution is compared and analyzed with the initial comprehensive quality score value. If the feedback data shows that the effluent quality does not meet the standard, the evaluation model parameters are readjusted and new process control instructions are output.

[0005] In a preferred embodiment, a water quality full spectrum detector device is used to perform a preliminary scan on the initial signal data to obtain the original spectrum data stream related to the carbon source quality index; Performing a preprocessing operation on the original spectral data stream to remove noise interference and obtain a cleaned spectral data set; extracting key characteristic values ​​related to quality indicators based on the cleaned spectral data set to determine the quality characteristics of the carbon source in the effluent; If the extracted characteristic value does not conform to a preset threshold range, performing a secondary calibration process on the spectral data set to obtain a calibrated characteristic data set; Using the calibrated feature data set, a support vector machine algorithm is used to classify and judge the water outlet feature parameters to obtain classification result data; Determine the final effluent characteristic parameter description based on the classification result data and in combination with the pre-established characteristic parameter mapping relationship; Among them, in the step of removing noise interference, the variational mode decomposition algorithm is used to transform the original spectral data stream Decomposed into K modal components , and its optimization model is: ; The constraints are ,By solving the model, the effective signal components are obtained,and a cleaned spectral data set is constructed.

[0006] In a preferred embodiment, obtaining a preliminary quantitative result of the organic matter content based on the original spectral data stream includes: The raw spectral data stream is preliminarily cleaned by a data stream processing module to remove irrelevant interference signals and obtain a processed spectral data set; Based on the processed spectral data set, a feature extraction method is used to screen key signal points to obtain a feature signal set related to the organic matter content; Among them, when using the feature extraction method to screen key signal points, the feature selection formula based on mutual information is introduced to calculate the feature and organic matter content target value Mutual information : ; Sort by mutual information value, and select the value with mutual information higher than the set threshold The features of constitute the feature signal set; Using the characteristic signal set, applying a pre-built evaluation model library in combination with a support vector regression algorithm, a regression calculation is performed on the signal set to obtain an initial quantization value; If the initial quantization value does not conform to the preset threshold range, performing secondary correction processing on the characteristic signal set to obtain a corrected quantization data set; Based on the corrected quantitative data set, a multi-dimensional comparison is performed through a data analysis layer to determine the distribution characteristics of the organic matter content; Based on the distribution characteristics, a preset classification rule is used to perform stratification processing to determine the final content assessment result; The final content assessment result is combined with a pre-established mapping relationship table to obtain an analysis data description that matches the business goal.

[0007] In a preferred embodiment, obtaining the variation trend data of the organic matter content based on the preliminary quantitative results includes: By preprocessing the original monitoring data, the initial time series data of organic matter content is obtained, and a standardized basic data set is obtained; Based on the normalized basic data set, a time series analysis method is used to construct a variation trend model of organic matter content, and an autoregressive moving average model is used to fit the data to determine trend characteristic parameters; For the trend characteristic parameters, extract the periodic component in the change trend, separate the long-term trend and the short-term fluctuation part, and obtain the distribution law of the periodic characteristics; Among them, when extracting the periodic component, the improved algorithm of empirical mode decomposition is used to convert the time series Decompose into Intrinsic Mode Function and residual components : ; By calculating each The frequency characteristics of the periodic characteristic components are screened out to obtain the distribution law of the periodic characteristics; If there is a significant fluctuation period in the distribution pattern of the periodic characteristics, the fluctuation period is decomposed by a frequency domain analysis method to obtain a specific period frequency value; Based on the periodic frequency values ​​and combined with long-term trend data, a water quality fluctuation prediction framework is constructed to determine whether the fluctuation pattern meets the preset periodic threshold range; Based on the results of the fluctuation pattern, a correlation map between water quality fluctuation and organic matter content change is generated to determine the key time nodes of the fluctuation impact; If the fluctuation impact of the key time node exceeds a preset threshold range, the data of the relevant time period is weighted to obtain adjusted change trend data.

[0008] In the preferred solution, if the change trend data exceeds a preset threshold range, the anomaly detection mechanism is triggered and the anomaly determination result is output, including: By collecting real-time data transmitted by environmental sensors, obtaining change trend information, storing it in the data processing module, and obtaining a preliminary change trend record; According to the change trend record, a preset threshold is used for comparison. If the change trend exceeds the preset threshold range, an anomaly detection process is triggered to determine a preliminary abnormal signal; For the preliminary abnormal signal, historical data is obtained from the database, and multi-dimensional comparison is performed in combination with the current data. The support vector machine algorithm is used to classify the water quality fluctuation characteristics to determine whether there is a significant deviation; Among them, when judging whether there is a significant deviation, the local anomaly factor algorithm is introduced to calculate the sample points The local reachable density : ; Then get the sample points Local anomaly factor ,when Greater than the set threshold When , it is judged that there is a significant deviation; If the support vector machine algorithm determines that there is a significant deviation in the water quality fluctuation, the fluctuation data is matched with the abnormal state standard to obtain abnormal state confirmation information; Generate determination result data based on the abnormal state confirmation information, store it in the result output module, and obtain a structured abnormality determination record; Through the abnormality determination record, the system adjustment module is linked to automatically generate adjustment parameter suggestions and output them to the relevant control unit; Obtain the execution feedback data of the adjustment parameter suggestions, compare and analyze it with historical data, determine whether the water quality fluctuation after the system adjustment has returned to the normal range, and output the final verification result.

[0009] In a preferred embodiment, according to the abnormality determination result, an adapted evaluation parameter combination is obtained to update the weight distribution of the evaluation model, including: By obtaining relevant data for abnormality determination from the system database, preliminary screening is performed on records in abnormal states to obtain the initial abnormality data set; Extracting key features of water quality fluctuations using a feature analysis method based on the initial abnormal data set and determining a distribution pattern of the fluctuation features; If the distribution pattern of the fluctuation characteristics exceeds a preset threshold range, the corresponding rules are called to match applicable evaluation parameters from the system database to obtain the corresponding parameter combination; By standardizing the obtained parameter combination, analyzing its adaptability to the evaluation model, and judging the effectiveness of the parameter combination; Among them, when analyzing the adaptability of parameter combination and evaluation model, the adaptability calculation formula based on Mahalanobis distance is adopted, and the parameter combination vector is set as , the mean vector of the evaluation model parameter distribution is , the covariance matrix is , then the fitness for: ; when Less than the set threshold When , the parameter combination is determined to be valid; If the parameter combination is judged to be valid, it is applied to the evaluation model, the weight distribution is updated, and a new model configuration is generated; Based on the new model configuration, real-time monitoring of water quality fluctuations under abnormal conditions is performed to obtain updated evaluation results; The sustained effect of the model update is determined by storing and comparing the updated evaluation results.

[0010] In the preferred solution, the updated evaluation model weight distribution is used to perform a multi-dimensional evaluation on the carbon source quality of the current terminal effluent to obtain a comprehensive quality score, including: By collecting the terminal effluent data, we can obtain the relevant indicators of carbon source quality, and combine them with the organic matter content and water quality fluctuation information to obtain a preliminary data set; Based on the preliminary data set, a pre-established evaluation model is used to perform a multi-dimensional scoring of the carbon source quality, and a scoring result for each dimension is determined; Based on the multi-dimensional scoring results, a weighted calculation is performed in combination with the weight distribution to obtain the comprehensive quality score value; Among them, when performing weighted calculation of comprehensive quality score, a dynamic weight adjustment formula based on entropy weight method is introduced, and the first The dimension rating is , its weight The calculation formula is: ; in , is the sample size, is the number of dimensions, based on which the comprehensive quality score is calculated ; If the comprehensive quality score is lower than the preset threshold, the water quality fluctuation data is analyzed to determine whether the fluctuation range exceeds the normal range; If the fluctuation amplitude exceeds the normal range, the abnormal time period characteristics are extracted from the fluctuation data, the process link information corresponding to the abnormal fluctuation is obtained, and the adjustment direction is determined; By combining the process link information corresponding to the abnormal fluctuation with the comprehensive quality score, a support vector machine algorithm is used to optimize and simulate the process parameters to obtain specific parameter values ​​of the adjustment plan; Based on the specific parameter values ​​of the adjustment scheme, guidance data for process adjustment is generated to determine the final process optimization direction.

[0011] In a preferred embodiment, for the comprehensive quality score, obtaining process adjustment parameter suggestions corresponding to the score and outputting specific process control instructions include: By extracting key indicators from the quality score data and using preset mapping rules to compare them with the process optimization data in the parameter library, preliminary matching results are obtained; Based on the preliminary matching result, the accuracy of the score matching is verified. If the verification result is lower than a preset threshold, a closer adjustment parameter is obtained from the parameter library to determine the final parameter recommendation; Among them, when checking the accuracy of score matching, the cosine similarity formula is used to calculate the similarity between the preliminary matching result and the actual score. : ; in Optimize the data index value for the initial matching process, is the indicator value corresponding to the actual score, is the number of indicators, when Below the set threshold When , re-obtain the adjustment parameters; After obtaining the final parameter suggestions, the parameter suggestions are converted into specific control instructions in combination with process control requirements to generate an executable process adjustment plan; The real-time optimization module is used to obtain the current status data of the sewage treatment process and compare it with the generated control instructions to determine whether it meets the requirements of the process adjustment; If the process adjustment requirements are met, the control instructions are sent to the sewage treatment equipment to update the process parameters in real time; Obtaining the actual effect of the process adjustment based on the feedback data from the equipment after the execution, and determining whether further parameter adjustment is required based on the deviation between the effect and the expectation; If the deviation exceeds a preset range, the process optimization data is re-extracted from the parameter library and combined with the real-time optimization feedback to generate new control instructions.

[0012] In a preferred embodiment, the feedback data after the execution is compared with the initial comprehensive quality score value. If the feedback data indicates that the effluent quality does not meet the standard, the evaluation model parameters are readjusted and new process control instructions are output, including: By obtaining feedback data from the monitoring system, the water quality is preliminarily compared with the preset threshold to determine whether it meets the standards; If the feedback data indicates that the effluent quality does not meet the standards, a data comparison method is used to conduct a detailed analysis of the feedback data and the initial quality score to determine the deviation range; Adjusting parameters of the evaluation model according to the deviation range obtained from the analysis, and optimizing the parameters using a pre-established support vector machine model to obtain adjusted model parameters; Generate new process control instructions through the adjusted model parameters, conduct real-time regulation on key links in the sewage treatment process, and determine the specific execution content of the instructions; Obtain real-time feedback data after regulation and compare it with the adjusted quality score to determine whether the water quality meets the expected standards; If the re-comparison still shows that the standard is not met, the process control instructions will be adjusted through historical data analysis combined with the quality analysis results to obtain an optimized control plan; According to the optimized control scheme, feedback data from the sewage treatment process is continuously monitored, and dynamic adjustments are made based on changing trends.

[0013] In the preferred solution, when adjusting the evaluation model parameters, an adaptive learning rate adjustment formula based on gradient descent is adopted, and the model parameters are set to , the loss function is , learning rate The dynamic adjustment formula is: ; in is the attenuation coefficient, For the The gradient of the loss function with respect to the parameters at the iteration is used to update the model parameters to optimize the control instructions.

[0014] The present invention provides a method for inspecting and evaluating the quality of carbon sources in terminal effluent from sewage treatment plants. The method uses an online full-spectrum water quality detector to achieve high-frequency real-time acquisition of raw spectral data on the organic matter content in the terminal effluent, and combines algorithms such as variational mode decomposition to remove noise interference, ensuring data accuracy and timeliness. Mutual information feature selection, support vector machine classification, and regression algorithms are used to achieve multi-dimensional and precise analysis from raw data to feature extraction, quantification results, and anomaly determination, comprehensively reflecting the quality characteristics of the carbon source. A dynamic weight adjustment mechanism and an adaptive model optimization algorithm are introduced to dynamically update the weight distribution of the evaluation model based on real-time feedback data, thereby improving the model's adaptability to water quality fluctuations. By constructing a time series analysis and frequency domain decomposition model, the long-term trend and periodic fluctuation of organic matter content changes are effectively separated, providing multi-dimensional data support for anomaly detection. The method combines a local anomaly factor algorithm with Mahalanobis distance fitness calculation to achieve rapid detection of abnormal states and verification of the effectiveness of parameter combinations, thereby shortening the abnormal response time. Reinforcement learning and fuzzy logic algorithms are used to generate process adjustment parameter recommendations, which are linked to the automatic control system for dynamic regulation. Control instructions are optimized through a closed-loop feedback data system, forming a "detection-evaluation-regulation-verification" process. The complete intelligent closed loop significantly improves sewage treatment efficiency and effluent quality stability, ensuring that the carbon source quality continues to meet standards. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The present invention will be further described below with reference to the accompanying drawings and examples: Figure 1 It is a flow chart of the present invention; DETAILED DESCRIPTION Example 1 like Figure 1 As shown, a method for inspecting and evaluating the quality of carbon sources in terminal effluent of a sewage treatment plant comprises: Obtain the original spectral data stream of the organic matter content in the terminal outlet water through the online water quality full spectrum detector equipment; Obtaining preliminary quantitative results of organic matter content based on the raw spectral data stream; Based on the preliminary quantitative results, obtaining change trend data of organic matter content; If the change trend data exceeds the preset threshold range, the anomaly detection mechanism is triggered and the anomaly determination result is output; According to the abnormality determination result, an adapted evaluation parameter combination is obtained to update the weight distribution of the evaluation model, and the carbon source quality of the current terminal effluent is scored in multiple dimensions using the updated evaluation model weight distribution to obtain a comprehensive quality score value; Based on the comprehensive quality score, obtain process adjustment parameter suggestions corresponding to the score and output specific process control instructions; According to the process control instructions, the automation control system of the sewage treatment plant is linked to obtain feedback data after execution; The feedback data after the execution is compared and analyzed with the initial comprehensive quality score value. If the feedback data shows that the effluent quality does not meet the standard, the evaluation model parameters are readjusted and new process control instructions are output.

[0016] First, the online full-spectrum water quality detector is used to collect the original spectral data stream of the organic matter content in the terminal effluent, and then the original spectral data stream is processed to obtain the preliminary quantitative results of the organic matter content. Based on the preliminary quantitative results, its change trend data is obtained. If the trend data exceeds the preset threshold range, the anomaly detection mechanism is triggered to output the anomaly judgment result. According to the anomaly judgment result, the adapted evaluation parameter combination is obtained to update the weight distribution of the evaluation model, and then the terminal effluent carbon source quality is scored in multiple dimensions through the updated model to obtain a comprehensive quality score value. For this score value, the corresponding process adjustment parameter suggestions are obtained and specific process control instructions are output. The instructions are executed and feedback data is obtained through the linkage automation control system. The feedback data is compared and analyzed with the initial comprehensive quality score value. If the effluent quality does not meet the standard, the evaluation model parameters are readjusted and new process control instructions are output.

[0017] Real-time data collection is achieved through online water quality full-spectrum detector equipment to ensure data timeliness and accuracy; the processing of raw spectral data streams and multi-dimensional scoring can comprehensively and accurately evaluate the quality of carbon sources; the anomaly detection mechanism can promptly detect water quality anomalies; based on the abnormal results, the evaluation model weight distribution is updated and process control instructions are output to achieve dynamic adjustment of the sewage treatment process; through comparative analysis of feedback data and initial scores, the evaluation model and process control instructions are continuously optimized to effectively improve the quality of carbon sources in the terminal effluent of sewage treatment plants, enhance the efficiency and effectiveness of sewage treatment, and ensure that the effluent water quality is stable and meets the standards.

[0018] The equipment signal of online water quality full spectrum detector is: L800D from French HEMERA: can measure multiple components in water or fluids, and can measure up to 7 components simultaneously.

[0019] The HD-S1000 Hyperspectral Intelligent Water Quality Detector: This instrument eliminates the need for chemical reagents and displays results in real time. Its portability, speed, all-weather capability, and intelligence make it ideal for rapid, real-time online monitoring of surface water, sewage, and other applications.

[0020] Hikvision's hyperspectral water quality multi-parameter monitor: Installed on the water surface like a probe, it can identify the spectral characteristics of different water qualities, collect 11 water quality data such as chlorophyll, permanganate index, and transparency in seconds, and effectively observe the distribution of chlorophyll and suspended matter in the water body, thereby identifying common water pollution problems such as eutrophication and algae blooms. It has been used in many lakes and drinking water sources in Jiangsu, Zhejiang, Shanghai, South China, Northwest China and other places.

[0021] Example 2 Using a water quality full spectrum detector device to preliminarily scan the initial signal data to obtain the original spectrum data stream related to the carbon source quality index; Performing a preprocessing operation on the original spectral data stream to remove noise interference and obtain a cleaned spectral data set; extracting key characteristic values ​​related to quality indicators based on the cleaned spectral data set to determine the quality characteristics of the carbon source in the effluent; If the extracted characteristic value does not conform to a preset threshold range, performing a secondary calibration process on the spectral data set to obtain a calibrated characteristic data set; Using the calibrated feature data set, a support vector machine algorithm is used to classify and judge the water outlet feature parameters to obtain classification result data; Determine the final effluent characteristic parameter description based on the classification result data and in combination with the pre-established characteristic parameter mapping relationship; Among them, in the step of removing noise interference, the variational mode decomposition algorithm is used to transform the original spectral data stream Decomposed into K modal components , and its optimization model is: ; The constraints are ,By solving the model, the effective signal components are obtained,and a cleaned spectral data set is constructed.

[0022] Use online sensors to collect high-frequency data on the terminal outlet water to obtain initial signal data related to the organic matter content: set the sampling frequency of the online sensor , to ensure that in unit time Completed Each time a sample is taken, the sensor converts the detected light signal into an electrical signal and records it as ,in , forming the initial signal data sequence. To ensure data accuracy, the sensor needs to be calibrated before sampling. The spectral data of a known standard solution is compared with the sensor measurement value to correct the measurement deviation.

[0023] Assume that the response function of the sensor is , the actual light intensity is , the electrical signal output by the sensor The relationship with light intensity can be expressed as: ; in, and is the wavelength range detected by the sensor, This formula describes the sensor's conversion process from optical signals to electrical signals. It can be used to analyze the relationship between sensor measurement results and actual light intensity, facilitating subsequent data calibration and error analysis.

[0024] The initial signal data is preliminarily scanned using spectral analysis technology to obtain the original spectral data stream related to the carbon source quality index: the initial signal data is input into the spectral analysis instrument, and the instrument uses a dispersion element to separate the light signal into wavelengths. Expand and detect light intensity at different wavelength positions to obtain the original spectral data stream In order to improve the spectral resolution, the method of multiple scan averaging can be used to perform spectral averaging on the same water sample. Scans are performed and the results of each scan are weighted averaged: .

[0025] In spectral analysis, the relationship between light intensity and wavelength can be converted from time domain signals to frequency domain spectrum through Fourier transform. Assume that the time domain signal is , its Fourier transform is: ; in, is the angular frequency, In the spectrum analysis scenario, the wavelength can be and frequency Establish a corresponding relationship (such as , The time-domain electrical signal collected by the sensor is converted into frequency-domain spectral data through the Fourier transform formula, thereby obtaining the original spectral data stream for subsequent analysis of the relationship between light intensity and carbon source quality indicators at different wavelengths.

[0026] The original spectral data stream is preprocessed to remove noise interference and obtain a cleaned spectral data set: the variational mode decomposition (VMD) algorithm is used to first set the number of decomposed modes. and penalty factor (Used to balance decomposition accuracy and computational complexity). Perform VMD decomposition and iteratively solve the optimization model to obtain modal components and the corresponding center frequency The proportion of retained energy exceeds a certain threshold The modal components of ,in is the set of modal indices that meet the energy threshold condition.

[0027] The optimization model of VMD is: ; The constraints are ; in, Indicates time Seek the derivative, is the Dirac function, $*$ represents the convolution operation, for This formula decomposes the original spectral data stream into modal components with different frequency characteristics by minimizing the sum of the bandwidths of each modal component, thereby separating the noise from the effective signal, removing noise interference, and retaining the effective spectral signal related to the carbon source quality.

[0028] Based on the cleaned spectral data set, key eigenvalues ​​related to quality indicators are extracted to determine the quality characteristics of the carbon source in the water: a feature extraction method based on Local Fisher Discriminant Analysis (LFDA) is used. First, the cleaned spectral data are divided into different categories according to the known carbon source quality category labels (such as high, medium, and low quality). The local neighborhood of each data point is calculated. For each data point , determine its Neighbor point set . Calculate the intra-class scatter matrix and the between-class scatter matrix , by solving the generalized eigenvalue problem , and obtain the feature vector , select the eigenvector corresponding to the larger eigenvalue to construct the projection matrix , project the spectral data into a low-dimensional space and obtain the key eigenvalue vector .

[0029] Intra-class scatter matrix The calculation formula is: ; Between-class scatter matrix The calculation formula is: ; in, is the number of categories, For the Class data collection, For the The mean of the class data, is the mean of all data, For the The eigenvectors obtained by solving the generalized eigenvalue problem are used to construct a projection matrix, mapping the high-dimensional spectral data into a low-dimensional space, and extracting key eigenvalues ​​with high discrimination between carbon source quality categories, thereby determining the carbon source quality characteristics of the effluent.

[0030] If the extracted eigenvalues ​​do not match the preset threshold range, the spectral data set is subjected to secondary calibration to obtain a calibrated feature data set: a nonlinear mapping model between eigenvalues ​​and spectral data is established, and kernel-based support vector regression (KSVR) is used. The deviation between the extracted eigenvalues ​​and the preset threshold is used as the training label, the original spectral data is used as the training sample, and a suitable kernel function (such as radial basis kernel function) is selected. ) Train the KSVR model. Use the trained model to predict the original spectral data to obtain calibrated spectral data. Then re-extract the eigenvalues ​​until the eigenvalues ​​fall within the preset threshold range to form a calibrated feature dataset.

[0031] The optimization objective function of KSVR is: ; The constraints are: ; ; ; in, is the weight vector, is the bias term, is the slack variable, is the penalty parameter, To tolerate errors, is the actual label, , This is a nonlinear mapping that maps input data to a high-dimensional space using a kernel function. By minimizing structural and empirical risks, this formula establishes a nonlinear relationship model between eigenvalue deviation and spectral data. This is used to calibrate spectral data, ensuring that the extracted eigenvalues ​​meet preset requirements and improving the accuracy of carbon source quality assessment.

[0032] The calibrated feature data set is used to classify and judge the water characteristic parameters using the support vector machine algorithm to obtain classification result data: the calibrated feature data set is divided into a training set and a test set, and an improved multi-class support vector machine (MSVM) algorithm is used, such as one-vs-one (OVO) or one-vs-rest (OVR) strategy. For the OVO strategy, In the categories, two-category training In the test phase, each sample votes for all binary classification models, and the category with the most votes is taken as the final classification result, thereby obtaining the classification result data of the water outlet characteristic parameters.

[0033] The decision function of the binary SVM is: ; in, is a symbolic function. In the multi-classification scenario, for the OVO strategy, we have Category, and The decision function of the SVM model trained for each category is ,sample The final classification result is:

[0034] in, is the indicator function, when The formula is 1 when the water quality is good, otherwise it is 0. This formula realizes the classification judgment of multiple categories of effluent characteristic parameters through the combination of multiple binary classification SVM models, and provides classification results for carbon source quality evaluation.

[0035] Based on the classification results and pre-established characteristic parameter mapping relationships, the final effluent characteristic parameter description is determined. A characteristic parameter mapping table is established, which records the carbon source quality characteristic parameter descriptions corresponding to different classification results. Based on the classification results, the corresponding record in the mapping table is searched to obtain the final effluent characteristic parameter description information, such as "high concentration easily degradable carbon source" and "low concentration difficult to degrade carbon source."

[0036] Let the classification result be , the characteristic parameter mapping relationship is , then the final water characteristic parameters describe It can be expressed as:

[0037] This formula simply and intuitively describes the mapping process from classification results to the final characteristic parameter description. Through the pre-established mapping relationship, the abstract classification results are converted into practical and understandable carbon source quality characteristic descriptions, providing a clear basis for subsequent sewage treatment process adjustments.

[0038] Example 3 Based on the raw spectral data stream, preliminary quantitative results of organic matter content are obtained, including: The raw spectral data stream is preliminarily cleaned by a data stream processing module to remove irrelevant interference signals and obtain a processed spectral data set; Based on the processed spectral data set, a feature extraction method is used to screen key signal points to obtain a feature signal set related to the organic matter content; Among them, when using the feature extraction method to screen key signal points, the feature selection formula based on mutual information is introduced to calculate the feature and organic matter content target value Mutual information : ; Sort by mutual information value, and select the value with mutual information higher than the set threshold The features of constitute the feature signal set; Using the characteristic signal set, applying a pre-built evaluation model library in combination with a support vector regression algorithm, a regression calculation is performed on the signal set to obtain an initial quantization value; If the initial quantization value does not conform to the preset threshold range, performing secondary correction processing on the characteristic signal set to obtain a corrected quantization data set; Based on the corrected quantitative data set, a multi-dimensional comparison is performed through a data analysis layer to determine the distribution characteristics of the organic matter content; Based on the distribution characteristics, a preset classification rule is used to perform stratification processing to determine the final content assessment result; The final content assessment result is combined with a pre-established mapping relationship table to obtain an analysis data description that matches the business goal.

[0039] The original spectral data stream is preliminarily cleaned by the data stream processing module to remove irrelevant interference signals and obtain the processed spectral data group: The method of combining mathematical morphology with adaptive median filtering is adopted. First, define the structural element , for the original spectral data stream Perform opening operation , remove the peak noise in the data; then perform a closed operation , filling the valley noise in the data. Then, the median filter window size is dynamically adjusted according to the local variance of the data, and the median filter window size is calculated for each data point. Neighborhood Variance ,like (high variance threshold), then increase the window; if (low variance threshold), then reduce the window and perform adaptive median filtering on the data to obtain the processed spectral data set .

[0040] Opening formula: ,in is the corrosion operation, is the dilation operation. The erosion operation is defined as , the dilation operation is defined as . Closed operation formula: .

[0041] Adaptive median filter window adjustment formula: ; in, For data points The window size at is the minimum window, is the maximum window, Adjust the step size for the window.

[0042] These formulas remove noise peaks and valleys through mathematical morphological operations, and adaptive median filtering dynamically adjusts the window according to the local characteristics of the data, effectively removing irrelevant interference signals and retaining the true characteristics of the spectral data.

[0043] According to the processed spectral data set, the key signal points are screened by feature extraction method to obtain the characteristic signal set related to the organic matter content: based on the combination of mutual information and improved genetic algorithm (GA). First, using the mutual information formula Calculate each wavelength characteristic and organic matter content target value The mutual information value of Then, the mutual information value is used as the fitness function of the genetic algorithm individual, and the individual is encoded (such as binary encoding, 1 means retaining the feature, 0 means removing the feature). Through selection, crossover, and mutation operations, multiple generations of evolution are carried out. In each generation, individuals with high fitness (large mutual information value) are retained and individuals with low fitness are eliminated. Finally, the optimal feature combination is obtained to form the feature signal set. .

[0044] The selection operation adopts the roulette wheel selection method. Probability of being selected for: ; The crossover operation uses two-point crossover, and there are two parent individuals and , randomly select two intersection points and ( ), exchange two parent individuals in and The gene fragments between the two are used to obtain the offspring individuals. The mutation operation is based on the mutation probability Flip individual genes.

[0045] Through these formulas, using mutual information as fitness to guide the genetic algorithm search, key features with a high correlation with organic matter content can be efficiently screened out from numerous wavelength features to form a feature signal set.

[0046] Through the feature signal set, the pre-built evaluation model library is applied, combined with the support vector regression algorithm, to perform regression calculation on the signal set to obtain the initial quantization value: using the support vector regression based on quantum kernel function (QKSVR). First, select the appropriate basic model from the evaluation model library and transform the feature signal set into As input data , the known true value of organic matter content is used as output data . Define the quantum kernel function ,in is the quantum state mapping function. Construct the optimization objective function of QKSVR: ; The constraints are: ; ; ; in By solving the optimization problem, we can get the regression model parameters and , and then perform regression calculation on the feature signal set to obtain the initial quantization value .

[0047] A specific form of quantum kernel function (taking the quantum rotating gate kernel function as an example):

[0048] in, is the feature dimension, and Data and No. dimensional features, is the angle parameter of the quantum rotating gate.

[0049] If the initial quantization value does not match the preset threshold range, the feature signal set is subjected to secondary correction processing to obtain a corrected quantization data set: Local weighted regression based on Bayesian optimization (BO-LWR) is introduced. Calculate the initial quantization value With the preset threshold range Deviation With deviation Minimization is the goal, and the Bayesian optimization algorithm is used to adjust the parameters of the local weighted regression (such as the bandwidth parameter of the weight function ). For each data point , according to its relationship with the current point Distance calculation weight , build a local weighted regression model ,in is the kernel function mapping. Parameters are continuously updated through Bayesian optimization , perform regression correction on the characteristic signal set to obtain the corrected quantitative data set .

[0050] In Bayesian optimization, the Gaussian process model is used to predict the objective function (deviation ) at the new parameter point. Assuming that Sample points , , construct the covariance function of the Gaussian process (such as the squared exponential covariance function ,in is the signal variance, is the length scale parameter), predict the new parameter point The objective function value at The mean and variance : ; ; in, , , is the noise variance, is the identity matrix.

[0051] Based on the corrected quantitative data set, a multi-dimensional comparison is performed through the data analysis layer to determine the distribution characteristics of the organic matter content: a method based on density peak clustering (DPC) and fractal dimension analysis is used. First, each data point in the corrected quantitative data set is calculated. The local density ,in is a unit step function, is the distance between data points, is the cutoff distance. Then calculate the distance of the data point . With local density is the horizontal axis, distance Draw a decision diagram for the vertical axis and identify the cluster centers. Then, use the box dimension method to calculate the fractal dimension of each cluster. : ,in The side length is The number of clustered data points covered by the box is . Based on the clustering results and fractal dimension, the distribution characteristics of organic matter content in different areas are determined, such as uniform distribution, clustered distribution, etc.

[0052] Approximate formula for box dimension calculation (actual calculation Take finite value): ; According to the distribution characteristics, the preset classification rules are used for hierarchical processing to determine the final content assessment results: a classification model based on fuzzy rules and hierarchical analysis method (AHP) is constructed. The distribution characteristic indicators such as fractal dimension and number of cluster centers are used as fuzzy input variables, and multiple fuzzy sets are defined (such as "low fractal dimension", "high fractal dimension", "few cluster centers", "many cluster centers", etc.). The degree to which each indicator belongs to different fuzzy sets is determined according to the membership function. A fuzzy rule base is established, for example, "if the fractal dimension is low and there are few cluster centers, the distribution of organic matter content is simple". The hierarchical analysis method is used to determine the weight of each fuzzy rule, and a judgment matrix is ​​constructed. The weight vector is obtained by calculating the eigenvector and the maximum eigenvalue of the matrix. For each sample, multiple classification results are obtained based on fuzzy rule reasoning, and weighted summation is performed in combination with weights to determine the final content assessment result, such as "uniform low content" and "aggregated high content".

[0053] Fuzzy rule reasoning uses the Mamdani reasoning method, which is based on the principle of “if yes and yes ,but yes " Type rule as an example, the output fuzzy set C' is: ; In the AHP, the judgment matrix The maximum eigenvalue of and eigenvectors satisfy , by solving the equation, we get the weight vector, which is used to weight the fuzzy reasoning results.

[0054] These formulas are combined with fuzzy logic to handle the ambiguity of data and use the analytic hierarchy process to determine the rule weights to achieve accurate classification and evaluation of the distribution characteristics of organic matter content.

[0055] Through the final content assessment results, combined with the pre-established mapping relationship table, the analytical data description that matches the business objectives is obtained: a dynamically updated mapping relationship table is established, in which the business description information corresponding to different content assessment results is recorded. At the same time, a reinforcement learning mechanism is introduced to update the mapping relationship table based on the actual effect feedback after the sewage treatment process is adjusted. Define the state space is the content evaluation result, action space Describe options for the business, reward function The setting is based on the improvement degree of effluent quality after process adjustment. Then, select Action Get the business description and get the new status after performing process adjustment operations and rewards , update the Q value table through the Q-learning algorithm:

[0056] in, is the learning rate, Based on the updated Q value table, the mapping relationship is optimized to obtain the analysis data description that best matches the business goal.

[0057] Example 4 Based on the preliminary quantitative results, obtain the trend data of organic matter content, including: By preprocessing the original monitoring data, the initial time series data of organic matter content is obtained, and a standardized basic data set is obtained; Based on the normalized basic data set, a time series analysis method is used to construct a variation trend model of organic matter content, and an autoregressive moving average model is used to fit the data to determine trend characteristic parameters; For the trend characteristic parameters, extract the periodic component in the change trend, separate the long-term trend and the short-term fluctuation part, and obtain the distribution law of the periodic characteristics; Among them, when extracting the periodic component, the improved algorithm of empirical mode decomposition is used to convert the time series Decompose into Intrinsic Mode Function and residual components : ; By calculating each The frequency characteristics of the periodic characteristic components are screened out to obtain the distribution law of the periodic characteristics; If there is a significant fluctuation period in the distribution pattern of the periodic characteristics, the fluctuation period is decomposed by a frequency domain analysis method to obtain a specific period frequency value; Based on the periodic frequency values ​​and combined with long-term trend data, a water quality fluctuation prediction framework is constructed to determine whether the fluctuation pattern meets the preset periodic threshold range; Based on the results of the fluctuation pattern, a correlation map between water quality fluctuation and organic matter content change is generated to determine the key time nodes of the fluctuation impact; If the fluctuation impact of the key time node exceeds a preset threshold range, the data of the relevant time period is weighted to obtain adjusted change trend data.

[0058] By preprocessing the original monitoring data, the initial time series data of organic matter content are obtained, and the normalized basic data set is obtained: a preprocessing method based on singular spectrum analysis (SSA) combined with quantile normalization is used. First, the original monitoring data are arranged in chronological order to form the original time series data. , perform singular spectrum analysis on it and decompose the time series into multiple subsequences , each subsequence corresponds to a different time scale feature. Calculate the mean of each subsequence and standard deviation , handle outliers (if the data point satisfy , it is considered as an outlier and repaired by adjacent point interpolation method). Then, the quantile normalization method is used to map the processed subsequences to the same quantile space, making different time series comparable, and finally obtaining a normalized basic data set. .

[0059] Based on the standardized basic data set, a time series analysis method was used to construct a trend model for organic matter content. The data were fitted with an autoregressive moving average model to determine the trend characteristic parameters: an improved adaptive autoregressive moving average (ARMA) model was used. First, the order of the ARMA model was preliminarily determined using the minimum information criterion (AIC) and the Bayesian information criterion (BIC). and , get the initial model Then, the dynamic weight factor is introduced , dynamically adjust the model parameters according to the time correlation and volatility of the data. Calculate the local autocorrelation coefficient of the data and partial autocorrelation coefficient , through the function Determine the weight factor. When updating the model parameters in each iteration, the weight factor is incorporated into the parameter estimation, so that the model can better adapt to the changes in data and ultimately determine the trend characteristic parameters. .

[0060] The improved algorithm based on empirical mode decomposition (EMD) is adopted, namely, ensemble empirical mode decomposition (EEMD) combined with variational mode decomposition (VMD). First, the normalized time series data is decomposed by EEMD. Then, white noise is added multiple times and EMD decomposition is performed. Then, the decomposition results are averaged to obtain multiple intrinsic mode functions (IMFs). and residual components , effectively suppressing the modal aliasing phenomenon. Then, the obtained IMF components are decomposed by VMD, and each IMF component is further decomposed into multiple modal components with different frequency characteristics. By calculating the center frequency of each modal component and bandwidth , filter out the modal components with periodic characteristics (if Within the preset periodic frequency range and If the modal component is less than a certain threshold, it is considered that the modal component is periodic. Finally, the filtered periodic modal components are statistically analyzed to obtain the distribution law of periodic characteristics, and the remaining components are combined to obtain the long-term trend and short-term fluctuation parts.

[0061] If there is a significant fluctuation period in the distribution law of the periodic characteristics, the fluctuation period is decomposed by the frequency domain analysis method to obtain the specific period frequency value: a frequency domain analysis method based on the fusion of Hilbert-Huang transform (HHT) and wavelet packet transform (WPT) is used. First, the time series with significant fluctuation periods is subjected to Hilbert-Huang transform, and multiple IMF components are obtained through empirical mode decomposition. Then, each IMF component is subjected to Hilbert transform to obtain the Hilbert spectrum. , preliminarily analyze the frequency range of the fluctuation period from the Hilbert spectrum. Then, perform wavelet packet transform on the original time series, decompose the signal into different frequency bands, and obtain the wavelet packet coefficients According to the frequency range determined by the Hilbert spectrum, the corresponding wavelet packet coefficients are selected for reconstruction to obtain sub-signals with different frequency components. Finally, the power spectral density of the sub-signal is calculated. , determine the specific periodic frequency value (the frequency corresponding to the peak of the power spectrum density is the periodic frequency value).

[0062] Based on the cyclical frequency values ​​and long-term trend data, a water quality fluctuation prediction framework is constructed to determine whether the fluctuation pattern meets the preset periodic threshold range. A hybrid prediction framework based on a gated recurrent unit (GRU) and seasonal decomposition is constructed. First, the cyclical frequency values ​​and long-term trend data are encoded and converted into a format suitable for model input. The data is divided into training and test sets. During the training phase, seasonal decomposition methods (such as the X-13 method) are used to separate seasonal, trend, and random components from the training data. The seasonal and trend components are used as input features for the GRU model. The model is trained by adjusting GRU parameters (such as the number of hidden layer neurons and the learning rate) to learn the data's changing patterns and periodic patterns. During the prediction phase, the test data is input into the trained model to obtain predicted water quality fluctuation data. The cyclical frequency and fluctuation amplitude of the predicted data are then calculated and compared with the preset periodic threshold range to determine whether the fluctuation pattern meets the requirements.

[0063] By judging the fluctuation patterns, we generate a correlation map between water quality fluctuations and organic matter content changes, and identify the key time nodes affected by the fluctuations: we use a correlation map method based on dynamic time warping (DTW) and cross-correlation analysis. First, we normalize the water quality fluctuation data and the organic matter content change data to make them comparable. Then, we use the dynamic time warping algorithm to calculate the similarity distance between the two. , find the best matching path between the two time series and determine their corresponding relationship in time. Then, calculate the cross-correlation function between the two ,in is the water quality fluctuation data, is the data of changes in organic matter content, and According to the value of the cross-correlation function, the correlation degree and time delay between water quality fluctuations and organic matter content changes are determined. Finally, combined with the results of the fluctuation law, the time points with significant fluctuation impact are found in the correlation map and determined as key time nodes.

[0064] If the fluctuation impact of the key time node exceeds the preset threshold range, the data of the relevant time period is weighted to obtain the adjusted change trend data: a weighted processing method based on fuzzy logic and adaptive weight allocation is adopted. First, the fuzzy input variables are defined as the fluctuation amplitude and duration at the key time node, which are divided into multiple fuzzy sets (such as "low amplitude and short duration", "high amplitude and long duration", etc.), and the degree to which they belong to different fuzzy sets is determined according to the membership function. A fuzzy rule base is established, for example, "If the fluctuation amplitude is high and the duration is long, the data weight should be greatly increased". Then, reasoning is performed according to the fuzzy rules to obtain the fuzzy weight of each data point. At the same time, an adaptive weight adjustment mechanism is introduced to dynamically adjust the fuzzy weight according to the local variance of the data and the degree of change of the overall trend. Assume that the data point The local variance of The overall trend change rate is , through the function Adjust the weights. Finally, multiply the adjusted weights by the original data to obtain the weighted adjusted change trend data.

[0065] Fuzzy rule reasoning uses the Mamdani reasoning method, and the output fuzzy set C' is ; in and is the membership function of the input variable.

[0066] The adaptive weight adjustment formula is: , the final weighted data ,in is the weight obtained by fuzzy reasoning.

[0067] These formulas use fuzzy logic to process data uncertainty, combined with adaptive weight allocation, to reasonably adjust data weights according to fluctuations at key time nodes to obtain more accurate change trend data.

[0068] Example 5 If the change trend data exceeds the preset threshold range, the anomaly detection mechanism is triggered and the anomaly determination result is output, including: By collecting real-time data transmitted by environmental sensors, obtaining change trend information, storing it in the data processing module, and obtaining a preliminary change trend record; According to the change trend record, a preset threshold is used for comparison. If the change trend exceeds the preset threshold range, an anomaly detection process is triggered to determine a preliminary abnormal signal; For the preliminary abnormal signal, historical data is obtained from the database, and multi-dimensional comparison is performed in combination with the current data. The support vector machine algorithm is used to classify the water quality fluctuation characteristics to determine whether there is a significant deviation; Among them, when judging whether there is a significant deviation, the local anomaly factor algorithm is introduced to calculate the sample points The local reachable density : ; Then get the sample points Local anomaly factor ,when Greater than the set threshold When , it is judged that there is a significant deviation; If the support vector machine algorithm determines that there is a significant deviation in the water quality fluctuation, the fluctuation data is matched with the abnormal state standard to obtain abnormal state confirmation information; Generate determination result data based on the abnormal state confirmation information, store it in the result output module, and obtain a structured abnormality determination record; Through the abnormality determination record, the system adjustment module is linked to automatically generate adjustment parameter suggestions and output them to the relevant control unit; Obtain the execution feedback data of the adjustment parameter suggestions, compare and analyze it with historical data, determine whether the water quality fluctuation after the system adjustment has returned to the normal range, and output the final verification result.

[0069] By collecting real-time data transmitted by environmental sensors, obtaining change trend information, storing it in the data processing module, and obtaining preliminary change trend records: Distributed multi-sensor fusion and data compression algorithm is adopted. First, multiple environmental sensors (such as water quality sensors, flow sensors, etc.) collect data in real time. The data of each sensor is represented as ,in Indicates the sensor number, Represents time. The data of each sensor is preprocessed, including noise removal (using wavelet threshold denoising method, assuming the wavelet function is , the threshold is , the denoised data is ,in is the wavelet coefficient) and normalization processing (mapping the data to the $[0,1]$ interval, Then, a distributed multi-sensor fusion algorithm is used, such as a fusion method based on D-S evidence theory, to fuse the data of multiple sensors. Assume that each sensor's judgment on the change trend is an evidence, and the basic probability distribution function of the evidence is ,in The basic probability distribution function after fusion is calculated by D-S synthesis rule. : ; in is the normalization constant. Finally, the fused data is subjected to lossless data compression (using a dictionary-based compression algorithm, such as the LZ77 algorithm, to build a dictionary , the data sequence is represented as in the form of is the position of the matching string in the dictionary, The length of the matching string, is the next character), stored in the data processing module, and a preliminary record of the change trend is obtained. .

[0070] According to the change trend record, the preset threshold is used for comparison. If the change trend exceeds the preset threshold range, the abnormality detection process is triggered to determine the preliminary abnormal signal: a comparison method based on dynamic threshold and fuzzy logic is used. First, the initial preset threshold range is determined based on historical data and expert experience. Then, through the analysis of historical data, a dynamic threshold adjustment model is established. The time series analysis method (such as ARIMA model) is used to model the historical change trend data and predict the threshold range in the future. Assume that the predicted threshold range is In the actual comparison process, the current change trend data Compare with the dynamic threshold range. At the same time, fuzzy logic is introduced to fuzzify the change trend data and the threshold range. Let the fuzzy set of the change trend data be , the fuzzy set of the threshold range is , define the membership function and Use fuzzy reasoning (such as Mamdani reasoning method) to determine whether the change trend exceeds the threshold range. , then the anomaly detection process is triggered to determine the preliminary anomaly signal.

[0071] ARIMA model prediction threshold formula: Assume that the historical trend data is , modeled by ARIMA (p, d, q) model, the predicted value for ,in is the autoregressive coefficient, is the moving average coefficient, is a white noise sequence. Adjust the threshold range according to the predicted value .

[0072] Fuzzy membership function example (taking Gaussian membership function as an example): ,in is the mean, is the standard deviation, which is used to fuzzy the change trend data and threshold range for fuzzy reasoning.

[0073] For the preliminary abnormal signal, historical data is obtained from the database, and multi-dimensional comparison is performed with the current data. The support vector machine algorithm is used to classify the water quality fluctuation characteristics to determine whether there is a significant deviation: a method combining multi-kernel support vector machine (MK-SVM) and principal component analysis (PCA) is used. First, historical data with a time span close to that of the current data is obtained from the database. . The current data With historical data Splice and get the data set . For the dataset Perform principal component analysis to reduce the data dimension and extract the main features. Suppose the data set after principal component analysis is Then, the multi-core support vector machine algorithm is used to Multi-core support vector machine combines multiple kernel functions (such as linear kernel function , Gaussian kernel function etc.), to improve the accuracy of classification. Suppose the kernel function combination is ,in is the weight coefficient. By solving the optimization problem: ; The constraints are , ,in is the weight vector, is the bias term, is the slack variable, is the penalty parameter, is a function mapped to a high-dimensional space through a kernel function. A classification model is obtained to determine whether the current water quality fluctuation characteristics deviate significantly from historical data.

[0074] If the support vector machine algorithm determines that there is a significant deviation in the water quality fluctuation, the fluctuation data is matched with the abnormal state standard to obtain abnormal state confirmation information: an abnormal state matching algorithm based on deep learning is used. First, a deep neural network (such as a convolutional neural network CNN) is constructed, and the fluctuation data is used as input and the abnormal state standard is used as a label for training. Assume that the structure of the deep neural network is Layer, The output of the layer is , and its calculation formula is ,in is the weight matrix, is the bias term, For activation functions (such as ReLU function ). During the training process, the cross entropy loss function is used ,in is the true label, To predict the label, the network parameters are updated using a backpropagation algorithm. After training is complete, the current fluctuation data is fed into the trained network to obtain the predicted abnormal state label. The predicted label is then matched against the abnormal state criteria to obtain abnormal state confirmation information.

[0075] According to the abnormal state confirmation information, the judgment result data is generated and stored in the result output module to obtain a structured abnormal judgment record: a distributed storage and encryption algorithm based on blockchain is used. First, the abnormal state confirmation information is encoded and converted into a digital signature form. Assume that the abnormal state confirmation information is , through the hash function Generate a digital signature Then, the digital signature is stored in a distributed ledger using blockchain technology. The blockchain consists of multiple blocks, each of which contains the hash value, timestamp, data (i.e. digital signature) of the previous block. ) and other information. The hash value of the block is , and its calculation formula is ,in is the hash value of the previous block, is the timestamp, The data in the current block is stored in the result output module, which generates a structured anomaly determination record.

[0076] Through the abnormal judgment record, the system adjustment module is linked to automatically generate adjustment parameter suggestions and output them to the relevant control unit: an intelligent adjustment algorithm based on reinforcement learning is used. First, the state space of the system adjustment module is defined , including current water quality parameters (such as organic matter content, pH value, dissolved oxygen, etc.), equipment operating status (such as pump flow, aeration time, etc.) and other information. Define action space , that is, the adjustment measures that can be taken (such as increasing the pump flow, extending the aeration time, etc.). Assume that the current state is , the actions taken , the reward function obtained The optimal adjustment strategy is learned through reinforcement learning algorithms (such as deep Q-network DQN). The deep Q-network consists of a neural network (such as a multi-layer perceptron MLP) as the Q-value function estimator, and its input is the state , the output is the Q value of each action During the training process, by constantly trying different actions, the parameters of the Q value function are updated according to the reward function. Let the parameters of the Q value function be , the stochastic gradient descent algorithm is used to update the parameters: ; in is the learning rate, is the discount factor. After training is completed, according to the current state Choose the action with the largest Q value Output to relevant control units as adjustment parameter suggestions.

[0077] Obtain the feedback data after the execution of the parameter adjustment suggestion, compare and analyze it with the historical data, determine whether the water quality fluctuation after the system adjustment has returned to the normal range, and output the final verification result: adopt the verification method based on grey correlation analysis and time series prediction. First, obtain the feedback data after the execution of the parameter adjustment suggestion , including water quality parameters, equipment operating status and other information. With historical data Perform grey correlation analysis and calculate the correlation degree. Assume that the sequence of feedback data is , the sequence of historical data is , grey relational degree The calculation formula is: ; in is the resolution coefficient, usually set to $0.5$. Then, a time series prediction method (such as LSTM model) is used to predict water quality fluctuations in the future. Assume that the input of the LSTM model is historical data and feedback data , the output is the predicted water quality fluctuation data . The predicted data Compare the predicted data with the normal range threshold to determine whether the water quality fluctuations after the system adjustment have returned to the normal range. If the predicted data is within the normal range, the final verification result is output as a successful adjustment; otherwise, the adjustment is output as a failure and the optimization adjustment strategy continues.

[0078] Example 6 According to the abnormality determination result, an adapted evaluation parameter combination is obtained to update the weight distribution of the evaluation model, including: By obtaining relevant data for abnormality determination from the system database, preliminary screening is performed on records in abnormal states to obtain the initial abnormality data set; all records related to abnormality determination are queried from the system database, which contain various water quality parameters, timestamps, abnormality determination results and other information.

[0079] For records in abnormal states, preliminary screening is performed based on pre-set screening conditions. For example, screening can be performed based on the type of abnormality (such as abnormal organic matter content, abnormal pH value, etc.), the time range of the abnormality, and other conditions.

[0080] The screened records were collated into an initial anomaly dataset, which served as the basis for subsequent analyses.

[0081] Extracting key features of water quality fluctuations using a feature analysis method based on the initial abnormal data set and determining a distribution pattern of the fluctuation features; The water quality fluctuation data in the initial abnormal data set were preprocessed, including denoising, normalization and other operations, to improve data quality and comparability.

[0082] The principal component analysis (PCA) method is used to extract features from the preprocessed data. The covariance matrix of the data is calculated, and then the principal components are obtained through eigenvalue decomposition.

[0083] The key features are determined based on the contribution rate of the principal components. Usually, the first few principal components with higher contribution rates are selected as key features.

[0084] Analyze key features to determine the distribution pattern of fluctuation characteristics. For example, you can observe the distribution of features by drawing histograms, box plots, etc.

[0085] If the distribution pattern of the fluctuation characteristics exceeds a preset threshold range, the corresponding rules are called to match applicable evaluation parameters from the system database to obtain the corresponding parameter combination; The determined fluctuation characteristic distribution pattern is compared with a preset threshold range. The threshold range can be set based on historical data, experience, or industry standards.

[0086] If the distribution pattern of the fluctuation characteristics exceeds the preset threshold range, the corresponding rules are triggered. These rules can be based on expert knowledge, machine learning models, or other empirical rules.

[0087] According to the triggered rules, matching evaluation parameters are searched from the system database, which stores various evaluation parameters and their applicable conditions.

[0088] The matched evaluation parameters are combined into a parameter combination, which will be used for subsequent model updates.

[0089] By standardizing the obtained parameter combination, analyzing its adaptability to the evaluation model, and judging the effectiveness of the parameter combination; Among them, when analyzing the adaptability of parameter combination and evaluation model, the adaptability calculation formula based on Mahalanobis distance is adopted, and the parameter combination vector is set as , the mean vector of the evaluation model parameter distribution is , the covariance matrix is , then the fitness for: ; when Less than the set threshold When , the parameter combination is determined to be valid; If the parameter combination is judged to be valid, it is applied to the evaluation model, the weight distribution is updated, and a new model configuration is generated; If the parameter combination is determined to be valid, it is applied to the evaluation model, which can be various machine learning models or statistical models.

[0090] Update the weight distribution of the evaluation model based on the parameter combination. The specific updating method depends on the type and structure of the evaluation model.

[0091] Generate a new model configuration, including updated weight distribution, model parameters and other information.

[0092] Based on the new model configuration, real-time monitoring of water quality fluctuations under abnormal conditions is performed to obtain updated evaluation results; The new model configuration is applied to the real-time monitoring system to monitor water quality fluctuations under abnormal conditions in real time.

[0093] Water quality data is collected in real time and input into the updated evaluation model for calculation.

[0094] Obtain updated evaluation results, which may include comprehensive evaluation scores of water quality, degree of abnormality, and other information.

[0095] The sustained effect of the model update is determined by storing and comparing the updated evaluation results.

[0096] The updated evaluation results are stored in the database for subsequent analysis and comparison.

[0097] Compare the updated evaluation results with the historical evaluation results regularly or irregularly.

[0098] Through comparative analysis, the sustainability of model updates can be determined. For example, changes in indicators such as stability and accuracy of evaluation results can be observed.

[0099] Example 7 The updated evaluation model weight distribution is used to perform a multi-dimensional evaluation of the carbon source quality of the current terminal effluent, and a comprehensive quality score is obtained, including: By collecting the terminal effluent data, we can obtain the relevant indicators of carbon source quality, and combine them with the organic matter content and water quality fluctuation information to obtain a preliminary data set; Based on the preliminary data set, a pre-established evaluation model is used to perform a multi-dimensional scoring of the carbon source quality, and a scoring result for each dimension is determined; Based on the multi-dimensional scoring results, a weighted calculation is performed in combination with the weight distribution to obtain the comprehensive quality score value; Among them, when performing weighted calculation of comprehensive quality score, a dynamic weight adjustment formula based on entropy weight method is introduced, and the first The dimension rating is , its weight The calculation formula is: ; in , is the sample size, is the number of dimensions, based on which the comprehensive quality score is calculated ; If the comprehensive quality score is lower than the preset threshold, the water quality fluctuation data is analyzed to determine whether the fluctuation range exceeds the normal range; If the fluctuation amplitude exceeds the normal range, the abnormal time period characteristics are extracted from the fluctuation data, the process link information corresponding to the abnormal fluctuation is obtained, and the adjustment direction is determined; By combining the process link information corresponding to the abnormal fluctuation with the comprehensive quality score, a support vector machine algorithm is used to optimize and simulate the process parameters to obtain specific parameter values ​​of the adjustment plan; Based on the specific parameter values ​​of the adjustment scheme, guidance data for process adjustment is generated to determine the final process optimization direction.

[0100] By collecting the terminal effluent data, we can obtain the relevant indicators of carbon source quality, and combine the organic matter content and water quality fluctuation information to obtain a preliminary data set. Collect terminal effluent data, including but not limited to organic matter content, pH value, dissolved oxygen, turbidity and other indicators.

[0101] The collected data is preprocessed, such as removing noise and filling missing values, to improve data quality.

[0102] The information on organic matter content and water quality fluctuations was integrated with other relevant indicators to form a preliminary data set.

[0103] Based on the preliminary data set, a pre-established evaluation model is used to perform a multi-dimensional scoring of carbon source quality, and a scoring result for each dimension is determined: the preliminary data set is input into the pre-established evaluation model. The evaluation model can be constructed based on a machine learning algorithm (e.g., a neural network, a decision tree, etc.) or a statistical method (e.g., multiple linear regression, etc.).

[0104] The evaluation model scores the carbon source quality in multiple dimensions based on the input data, and each dimension corresponds to a specific evaluation indicator (such as the biodegradability and stability of the carbon source).

[0105] Output the scoring results for each dimension, which will serve as the basis for subsequent comprehensive quality scoring.

[0106] For the multi-dimensional scoring results, combined with the weight distribution, weighted calculation is performed to obtain the comprehensive quality score value: a dynamic weight adjustment formula based on the entropy weight method is introduced. First, the entropy value of each dimension score is calculated. , the formula is ,in is the sample size, It is Dimension The score of the samples.

[0107] Calculate the weight of each dimension based on the entropy value , the formula is ,in is the number of dimensions.

[0108] Perform weighted calculation on the multi-dimensional scoring results to obtain a comprehensive quality score .

[0109] If the comprehensive quality score is below a preset threshold, the water quality fluctuation data is analyzed to determine whether the fluctuation is outside the normal range: the comprehensive quality score is compared with a preset threshold. The preset threshold can be determined based on historical data, industry standards, or experience.

[0110] If the comprehensive quality score value is lower than the preset threshold, the water quality fluctuation data will be further analyzed.

[0111] Calculate the amplitude of water quality fluctuations. For example, the amplitude of fluctuations can be measured by calculating statistics such as standard deviation and range.

[0112] Compare the calculated fluctuation range with the normal range to determine whether the fluctuation range exceeds the normal range.

[0113] If the fluctuation amplitude exceeds the normal range, the abnormal time period characteristics are extracted from the fluctuation data, the process link information corresponding to the abnormal fluctuation is obtained, and the adjustment direction is determined: when the fluctuation amplitude exceeds the normal range, the fluctuation data is subjected to time series analysis. Methods such as sliding window method or wavelet transform can be used to extract abnormal time period characteristics. Combined with the flowchart and relevant knowledge of the sewage treatment process, the relationship between abnormal fluctuations and various process links is analyzed. Determine the process link information that may cause abnormal fluctuations, such as the failure of a certain equipment, the abnormality of a certain operating parameter, etc. Based on the determined process link information, formulate adjustment directions, such as adjusting equipment operating parameters, repairing equipment failures, etc.

[0114] By combining the process link information corresponding to the abnormal fluctuation with the comprehensive quality score value, the support vector machine algorithm is used to optimize and simulate the process parameters to obtain the specific parameter values ​​of the adjustment plan: the process link information corresponding to the abnormal fluctuation and the comprehensive quality score value are used as input data to construct a support vector machine (SVM) model.

[0115] The SVM model is trained using historical data or simulated data as training samples, and the parameters of the model are adjusted so that it can accurately predict the optimized values ​​of the process parameters.

[0116] Use the trained SVM model to optimize and simulate the current process parameters, input the current process link information and comprehensive quality score value, and obtain the specific parameter values ​​of the adjustment plan.

[0117] Based on the specific parameter values ​​of the adjustment plan, guidance data for process adjustment is generated to determine the final process optimization direction: Based on the specific parameter values ​​of the obtained adjustment plan, guidance data for process adjustment is generated. This data may include specific operating instructions, equipment parameter settings, etc.

[0118] Communicate the guidance data for process adjustment to relevant operators or automated control systems to implement process adjustments.

[0119] Monitor and evaluate the effects of process adjustments, further optimize the adjustment plan based on actual conditions, and determine the final process optimization direction.

[0120] Example 8 Based on the comprehensive quality score, obtain process adjustment parameter suggestions corresponding to the score and output specific process control instructions, including: By extracting key indicators from the quality score data and using preset mapping rules to compare them with the process optimization data in the parameter library, preliminary matching results are obtained; Based on the preliminary matching result, the accuracy of the score matching is verified. If the verification result is lower than a preset threshold, a closer adjustment parameter is obtained from the parameter library to determine the final parameter recommendation; Among them, when checking the accuracy of score matching, the cosine similarity formula is used to calculate the similarity between the preliminary matching result and the actual score. : ; in Optimize the data index value for the initial matching process, is the indicator value corresponding to the actual score, is the number of indicators, when Below the set threshold When , re-obtain the adjustment parameters; After obtaining the final parameter suggestions, the parameter suggestions are converted into specific control instructions in combination with process control requirements to generate an executable process adjustment plan; The real-time optimization module is used to obtain the current status data of the sewage treatment process and compare it with the generated control instructions to determine whether it meets the requirements of the process adjustment; If the process adjustment requirements are met, the control instructions are sent to the sewage treatment equipment to update the process parameters in real time; Obtaining the actual effect of the process adjustment based on the feedback data from the equipment after the execution, and determining whether further parameter adjustment is required based on the deviation between the effect and the expectation; If the deviation exceeds a preset range, the process optimization data is re-extracted from the parameter library and combined with the real-time optimization feedback to generate new control instructions.

[0121] By extracting key indicators from the quality score data, using preset mapping rules, and comparing them with the process optimization data in the parameter library, preliminary matching results are obtained: key indicators are determined from the quality score data. These indicators may include organic matter content, water quality fluctuation range, carbon source stability, etc.

[0122] For each key indicator, it is mapped to the corresponding process optimization data in the parameter library according to the preset mapping rules. The mapping rules can be based on experience, historical data or models.

[0123] Search the parameter library for the process optimization data closest to the mapped key indicators to obtain preliminary matching results.

[0124] Based on the preliminary matching results, the accuracy of the score matching is verified. If the verification result is lower than the preset threshold, a closer adjustment parameter is obtained from the parameter library to determine the final parameter recommendation: the cosine similarity formula is used to calculate the similarity between the preliminary matching result and the actual score. The formula is ,in Optimize the data index value for the initial matching process, is the indicator value corresponding to the actual score, is the number of indicators.

[0125] The calculated similarity With the set threshold Make a comparison.

[0126] if Below the set threshold , it is considered that the accuracy of the score matching is low, and it is necessary to obtain closer adjustment parameters from the parameter library again.

[0127] Repeat the above steps until the adjustment parameters that meet the accuracy requirements are found and determine the final parameter recommendations.

[0128] After obtaining the final parameter recommendations, the parameter recommendations are converted into specific control instructions in combination with the process control requirements to generate an executable process adjustment plan: clarify the specific requirements of process control, such as adjusting the operating parameters of sewage treatment equipment, changing the process flow, etc.

[0129] Based on the final parameter recommendations, they are converted into specific control instructions, which may involve converting the parameter values ​​into a signal or instruction format acceptable to the device.

[0130] Combined with the process and logic of process control, an executable process adjustment plan is generated. The plan should include specific operation steps, time schedule, responsible persons and other information.

[0131] Through the real-time optimization module, the current status data of the sewage treatment process is obtained and compared with the generated control instructions to determine whether it meets the requirements of the process adjustment: the real-time optimization module continuously monitors the current status data of the sewage treatment process, including equipment operating parameters, water quality indicators, etc.

[0132] Compare the current state data with the generated control instructions to check whether the actual operation meets the adjustment requirements. Various comparison methods and indicators can be used to evaluate consistency. If the process adjustment requirements are met, the control instructions will be sent to the sewage treatment equipment to update the process parameters in real time: if the comparison result between the current status data and the control instructions meets the process adjustment requirements, the control instructions will be sent to the sewage treatment equipment.

[0133] After receiving the control instructions, the sewage treatment equipment updates the process parameters in real time to adjust the sewage treatment process.

[0134] Based on the equipment feedback data after the execution, the actual effect of the process adjustment is obtained, and based on the deviation between the effect and the expectation, it is determined whether further parameter adjustment is required. Suggestions: Collect feedback data after the sewage treatment equipment executes the control instructions, including water quality improvement, equipment operating status, etc.

[0135] Based on the feedback data, evaluate the actual effects of process adjustments, such as calculating changes in water quality indicators, reductions in equipment energy consumption, etc. Compare the actual effects with the expected effects and calculate the deviation.

[0136] Determine whether the deviation exceeds the preset range. If so, it is considered necessary to further adjust the parameter recommendations.

[0137] If the deviation exceeds the preset range, the process optimization data is re-extracted from the parameter library and new control instructions are generated in combination with the feedback of real-time optimization. When the deviation exceeds the preset range, the process optimization data is re-extracted from the parameter library. According to the direction and size of the deviation, relevant process optimization data can be selected in a targeted manner.

[0138] Combined with the feedback information provided by the real-time optimization module, the re-extracted process optimization data is adjusted and optimized.

[0139] Based on the optimized process optimization data, new control instructions are generated and the above process is repeated until a satisfactory sewage treatment effect is achieved.

[0140] Example 9 The feedback data after the execution is compared and analyzed with the initial comprehensive quality score. If the feedback data indicates that the effluent quality does not meet the standard, the evaluation model parameters are readjusted and new process control instructions are output, including: By obtaining feedback data from the monitoring system, the water quality is preliminarily compared with the preset threshold to determine whether it meets the standards; If the feedback data indicates that the effluent quality does not meet the standards, a data comparison method is used to conduct a detailed analysis of the feedback data and the initial quality score to determine the deviation range; Adjusting parameters of the evaluation model according to the deviation range obtained from the analysis, and optimizing the parameters using a pre-established support vector machine model to obtain adjusted model parameters; Generate new process control instructions through the adjusted model parameters, conduct real-time regulation on key links in the sewage treatment process, and determine the specific execution content of the instructions; Obtain real-time feedback data after regulation and compare it with the adjusted quality score to determine whether the water quality meets the expected standards; If the re-comparison still shows that the standard is not met, the process control instructions will be adjusted through historical data analysis combined with the quality analysis results to obtain an optimized control plan; According to the optimized control scheme, feedback data from the sewage treatment process is continuously monitored, and dynamic adjustments are made based on changing trends.

[0141] In the preferred embodiment, when adjusting the evaluation model parameters, an adaptive learning rate adjustment formula based on gradient descent is adopted, and the model parameters are set to , the loss function is , learning rate The dynamic adjustment formula is: ; in is the attenuation coefficient, For the The gradient of the loss function with respect to the parameters at the iteration is used to update the model parameters to optimize the control instructions.

[0142] The above embodiments are merely preferred technical solutions of the present invention and should not be construed as limiting the present invention. The scope of protection of the present invention shall be the technical solutions set forth in the claims, including equivalent alternatives to the technical features of the technical solutions set forth in the claims. In other words, equivalent alternatives and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A method for inspecting and evaluating the quality of carbon sources in the terminal effluent of a sewage treatment plant, characterized by: The method includes: Obtain the original spectral data stream of the organic matter content in the terminal outlet water through the online water quality full spectrum detector equipment; Obtaining preliminary quantitative results of organic matter content based on the raw spectral data stream; Based on the preliminary quantitative results, obtaining change trend data of organic matter content; If the change trend data exceeds the preset threshold range, the anomaly detection mechanism is triggered and the anomaly determination result is output; According to the abnormality determination result, an adapted evaluation parameter combination is obtained to update the weight distribution of the evaluation model, and the carbon source quality of the current terminal effluent is scored in multiple dimensions using the updated evaluation model weight distribution to obtain a comprehensive quality score value; Based on the comprehensive quality score, obtain process adjustment parameter suggestions corresponding to the score and output specific process control instructions; According to the process control instructions, the automation control system of the sewage treatment plant is linked to obtain feedback data after execution; The feedback data after the execution is compared and analyzed with the initial comprehensive quality score value. If the feedback data shows that the effluent quality does not meet the standard, the evaluation model parameters are readjusted and new process control instructions are output.

2. The method for evaluating the quality of carbon sources in the effluent from a sewage treatment plant according to claim 1, wherein: Using a water quality full spectrum detector device to preliminarily scan the initial signal data to obtain the original spectrum data stream related to the carbon source quality index; Performing a preprocessing operation on the original spectral data stream to remove noise interference and obtain a cleaned spectral data set; extracting key characteristic values ​​related to quality indicators based on the cleaned spectral data set to determine the quality characteristics of the carbon source in the effluent; If the extracted characteristic value does not conform to a preset threshold range, performing a secondary calibration process on the spectral data set to obtain a calibrated characteristic data set; Using the calibrated feature data set, a support vector machine algorithm is used to classify and judge the water outlet feature parameters to obtain classification result data; Determine the final effluent characteristic parameter description based on the classification result data and in combination with the pre-established characteristic parameter mapping relationship; Among them, in the step of removing noise interference, the variational mode decomposition algorithm is used to transform the original spectral data stream Decomposed into K modal components , and its optimization model is: ； The constraints are ,By solving the model, the effective signal components are obtained,and a cleaned spectral data set is constructed.

3. The method for evaluating the quality of carbon sources in the terminal effluent of a sewage treatment plant according to claim 1, wherein: Based on the raw spectral data stream, preliminary quantitative results of organic matter content are obtained, including: The raw spectral data stream is preliminarily cleaned by a data stream processing module to remove irrelevant interference signals and obtain a processed spectral data set; Based on the processed spectral data set, a feature extraction method is used to screen key signal points to obtain a feature signal set related to the organic matter content; Among them, when using the feature extraction method to screen key signal points, the feature selection formula based on mutual information is introduced to calculate the feature and organic matter content target value Mutual information : ； Sort by mutual information value, and select the value with mutual information higher than the set threshold The features constitute the feature signal set; Using the characteristic signal set, applying a pre-built evaluation model library in combination with a support vector regression algorithm, a regression calculation is performed on the signal set to obtain an initial quantization value; If the initial quantization value does not conform to the preset threshold range, performing secondary correction processing on the characteristic signal set to obtain a corrected quantization data set; Based on the corrected quantitative data set, a multi-dimensional comparison is performed through a data analysis layer to determine the distribution characteristics of the organic matter content; Based on the distribution characteristics, a preset classification rule is used to perform stratification processing to determine the final content assessment result; The final content assessment result is combined with a pre-established mapping relationship table to obtain an analysis data description that matches the business goal.

4. A method for inspecting and evaluating the quality of carbon sources in terminal effluent from a sewage treatment plant according to claim 3, characterized in that: Based on the preliminary quantitative results, obtain the trend data of organic matter content, including: By preprocessing the original monitoring data, the initial time series data of organic matter content is obtained, and a standardized basic data set is obtained; Based on the normalized basic data set, a time series analysis method is used to construct a variation trend model of organic matter content, and an autoregressive moving average model is used to fit the data to determine trend characteristic parameters; For the trend characteristic parameters, extract the periodic component in the change trend, separate the long-term trend and the short-term fluctuation part, and obtain the distribution law of the periodic characteristics; Among them, when extracting the periodic component, the improved algorithm of empirical mode decomposition is used to convert the time series Decompose into Intrinsic Mode Function and residual components : ； By calculating each The frequency characteristics of the periodic characteristic components are screened out to obtain the distribution law of the periodic characteristics; If there is a significant fluctuation period in the distribution pattern of the periodic characteristics, the fluctuation period is decomposed by a frequency domain analysis method to obtain a specific period frequency value; Based on the periodic frequency values ​​and combined with long-term trend data, a water quality fluctuation prediction framework is constructed to determine whether the fluctuation pattern meets the preset periodic threshold range; Based on the results of the fluctuation pattern, a correlation map between water quality fluctuation and organic matter content change is generated to determine the key time nodes of the fluctuation impact; If the fluctuation impact of the key time node exceeds a preset threshold range, the data of the relevant time period is weighted to obtain adjusted change trend data.

5. A method for inspecting and evaluating the quality of carbon sources in terminal effluent from a sewage treatment plant according to claim 4, characterized in that: If the change trend data exceeds the preset threshold range, the anomaly detection mechanism is triggered and the anomaly determination result is output, including: By collecting real-time data transmitted by environmental sensors, obtaining change trend information, storing it in the data processing module, and obtaining a preliminary change trend record; According to the change trend record, a preset threshold is used for comparison. If the change trend exceeds the preset threshold range, an anomaly detection process is triggered to determine a preliminary abnormal signal; For the preliminary abnormal signal, historical data is obtained from the database, and multi-dimensional comparison is performed in combination with the current data. The support vector machine algorithm is used to classify the water quality fluctuation characteristics to determine whether there is a significant deviation; Among them, when judging whether there is a significant deviation, the local anomaly factor algorithm is introduced to calculate the sample points The local reachable density : ； Then get the sample points Local anomaly factor ,when Greater than the set threshold When , it is judged that there is a significant deviation; If the support vector machine algorithm determines that there is a significant deviation in the water quality fluctuation, the fluctuation data is matched with the abnormal state standard to obtain abnormal state confirmation information; Generate determination result data based on the abnormal state confirmation information, store it in the result output module, and obtain a structured abnormality determination record; Through the abnormality determination record, the system adjustment module is linked to automatically generate adjustment parameter suggestions and output them to the relevant control unit; Obtain the execution feedback data of the adjustment parameter suggestions, compare and analyze it with historical data, determine whether the water quality fluctuation after the system adjustment has returned to the normal range, and output the final verification result.

6. A method for inspecting and evaluating the quality of carbon sources in terminal effluent from a sewage treatment plant according to claim 5, characterized in that: According to the abnormality determination result, an adapted evaluation parameter combination is obtained to update the weight distribution of the evaluation model, including: By obtaining relevant data for abnormality determination from the system database, preliminary screening is performed on records in abnormal states to obtain the initial abnormality data set; Extracting key features of water quality fluctuations using a feature analysis method based on the initial abnormal data set and determining a distribution pattern of the fluctuation features; If the distribution pattern of the fluctuation characteristics exceeds a preset threshold range, the corresponding rules are called to match applicable evaluation parameters from the system database to obtain the corresponding parameter combination; By standardizing the obtained parameter combination, analyzing its adaptability to the evaluation model, and judging the effectiveness of the parameter combination; Among them, when analyzing the adaptability of parameter combination and evaluation model, the adaptability calculation formula based on Mahalanobis distance is adopted, and the parameter combination vector is set as , the mean vector of the evaluation model parameter distribution is , the covariance matrix is , then the fitness for: ； when Less than the set threshold When , the parameter combination is determined to be valid; If the parameter combination is judged to be valid, it is applied to the evaluation model, the weight distribution is updated, and a new model configuration is generated; Based on the new model configuration, real-time monitoring of water quality fluctuations under abnormal conditions is performed to obtain updated evaluation results; The sustained effect of the model update is determined by storing and comparing the updated evaluation results.

7. A method for inspecting and evaluating the quality of carbon sources in terminal effluent from a sewage treatment plant according to claim 6, characterized in that: The updated evaluation model weight distribution is used to perform a multi-dimensional evaluation of the carbon source quality of the current terminal effluent, and a comprehensive quality score is obtained, including: By collecting the terminal effluent data, we can obtain the relevant indicators of carbon source quality, and combine them with the organic matter content and water quality fluctuation information to obtain a preliminary data set; Based on the preliminary data set, a pre-established evaluation model is used to perform a multi-dimensional scoring of the carbon source quality, and a scoring result for each dimension is determined; Based on the multi-dimensional scoring results, a weighted calculation is performed in combination with the weight distribution to obtain the comprehensive quality score value; Among them, when performing weighted calculation of comprehensive quality score, a dynamic weight adjustment formula based on entropy weight method is introduced, and the first The dimension rating is , its weight The calculation formula is: ； in , is the sample size, is the number of dimensions, based on which the comprehensive quality score is calculated ; If the comprehensive quality score is lower than the preset threshold, the water quality fluctuation data is analyzed to determine whether the fluctuation range exceeds the normal range; If the fluctuation amplitude exceeds the normal range, the abnormal time period characteristics are extracted from the fluctuation data, the process link information corresponding to the abnormal fluctuation is obtained, and the adjustment direction is determined; By combining the process link information corresponding to the abnormal fluctuation with the comprehensive quality score, a support vector machine algorithm is used to optimize and simulate the process parameters to obtain specific parameter values ​​of the adjustment plan; Based on the specific parameter values ​​of the adjustment scheme, guidance data for process adjustment is generated to determine the final process optimization direction.

8. A method for inspecting and evaluating the quality of carbon sources in terminal effluent from a sewage treatment plant according to claim 7, characterized in that: Based on the comprehensive quality score, obtain process adjustment parameter suggestions corresponding to the score and output specific process control instructions, including: By extracting key indicators from the quality score data and using preset mapping rules to compare them with the process optimization data in the parameter library, preliminary matching results are obtained; Based on the preliminary matching result, the accuracy of the score matching is verified. If the verification result is lower than a preset threshold, a closer adjustment parameter is obtained from the parameter library to determine the final parameter recommendation; Among them, when checking the accuracy of score matching, the cosine similarity formula is used to calculate the similarity between the preliminary matching result and the actual score. : ； in Optimize the data index value for the initial matching process, is the indicator value corresponding to the actual score, is the number of indicators, when Below the set threshold When , re-obtain the adjustment parameters; After obtaining the final parameter suggestions, the parameter suggestions are converted into specific control instructions in combination with process control requirements to generate an executable process adjustment plan; The real-time optimization module is used to obtain the current status data of the sewage treatment process and compare it with the generated control instructions to determine whether it meets the requirements of the process adjustment; If the process adjustment requirements are met, the control instructions are sent to the sewage treatment equipment to update the process parameters in real time; Obtaining the actual effect of the process adjustment based on the feedback data from the equipment after the execution, and determining whether further parameter adjustment is required based on the deviation between the effect and the expectation; If the deviation exceeds a preset range, the process optimization data is re-extracted from the parameter library and combined with the real-time optimization feedback to generate new control instructions.

9. A method for inspecting and evaluating the quality of carbon sources in terminal effluent of a sewage treatment plant according to claim 8, characterized in that: The feedback data after the execution is compared and analyzed with the initial comprehensive quality score. If the feedback data indicates that the effluent quality does not meet the standard, the evaluation model parameters are readjusted and new process control instructions are output, including: By obtaining feedback data from the monitoring system, the water quality is preliminarily compared with the preset threshold to determine whether it meets the standards; If the feedback data indicates that the effluent quality does not meet the standards, a data comparison method is used to conduct a detailed analysis of the feedback data and the initial quality score to determine the deviation range; Adjusting parameters of the evaluation model according to the deviation range obtained from the analysis, and optimizing the parameters using a pre-established support vector machine model to obtain adjusted model parameters; Generate new process control instructions through the adjusted model parameters, conduct real-time regulation on key links in the sewage treatment process, and determine the specific execution content of the instructions; Obtain real-time feedback data after regulation and compare it with the adjusted quality score to determine whether the water quality meets the expected standards; If the re-comparison still shows that the standard is not met, the process control instructions will be adjusted through historical data analysis combined with the quality analysis results to obtain an optimized control plan; According to the optimized control scheme, feedback data from the sewage treatment process is continuously monitored, and dynamic adjustments are made based on changing trends.

10. A method for inspecting and evaluating the quality of carbon sources in terminal effluent of a sewage treatment plant according to claim 9, characterized in that: in, When adjusting the evaluation model parameters, the adaptive learning rate adjustment formula based on gradient descent is adopted, and the model parameters are set to , the loss function is , learning rate The dynamic adjustment formula is: ； in is the attenuation coefficient, For the The gradient of the loss function with respect to the parameters at the iteration is used to update the model parameters to optimize the control instructions.

Images