A wastewater treatment fault early warning method and system

Abstract

This invention relates to the field of fault early warning technology, and discloses a method and system for early warning of sewage treatment faults. The method includes acquiring sewage monitoring data, extracting water quality fluctuation signals and detecting anomalies to obtain preliminary anomaly results; grouping the results, identifying potential fault characteristics if data fluctuations exceed the range, and forming a list of key influencing factors; extracting real-time indicator data from the list, classifying and determining dynamic operating strategy parameters, and generating a purification process adjustment plan. The plan is input into the equipment, and the status values ​​are monitored. If a threshold is exceeded, the anomaly extraction process and the list of optimized factors are updated to obtain real-time changing indicators; water quality trends are predicted, risk control points are determined, and an early warning sequence is generated; high-priority items are extracted, data is regrouped, and if a continuous deviation occurs, a purification process adjustment instruction is generated; the system configuration is updated according to the instruction, the equipment status values ​​are verified, and the accuracy of anomaly extraction and overall environmental risk control are evaluated. This method can solve the problem of insufficient fault early warning capabilities.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, and in particular to a wastewater treatment fault early warning method and system. Background Technology

[0002] Currently, in a certain existing sewage treatment system, by collecting data from the system equipment sensors, extreme values ​​that are significantly beyond the sensor's range are first eliminated. Then, the difference between the single collected data and the preset standard value is calculated. Data with values ​​greater than the threshold or less than the safety lower limit are retained as suspected abnormal signals. Three fixed early warning thresholds are set. After the early warning is triggered, the system only pushes two types of information: "exceeding the standard index and the current value". The operation strategy relies on manual experience for adjustment.

[0003] However, in existing technologies, because the sensor data collection and processing only performs threshold comparisons, it cannot identify subtle anomalies hidden in normal value fluctuations. The water quality data and equipment data collected by the system are not feature-refined, requiring a long time to troubleshoot. The warning threshold is a fixed value, which cannot adapt to the dynamic changes in water quality fluctuations, resulting in delayed warnings. In summary, existing technologies have insufficient fault warning capabilities. Summary of the Invention

[0004] This invention provides a wastewater treatment fault early warning method and system to solve the problem of insufficient fault early warning capability in the prior art.

[0005] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a wastewater treatment fault early warning method, comprising:

[0006] Wastewater monitoring data is acquired, and water quality fluctuation signals are extracted. Anomaly detection is performed based on the water quality fluctuation signals to obtain anomaly signal extraction results.

[0007] The data of the abnormal signal extraction results are grouped. If the fluctuation range of the data in the group deviates from the preset range, it is judged as a potential fault feature identification signal, and a list of key influencing factors is obtained.

[0008] Real-time indicator change data is extracted from the list of key influencing factors, the real-time indicator change data is classified and processed, dynamic operation strategy adjustment parameters are determined, and purification process adjustment plan is obtained.

[0009] The purification process adjustment plan is input into the wastewater treatment process control equipment, and the status monitoring value of the equipment is extracted. If the status monitoring value exceeds the preset status threshold, the abnormal signal extraction process is updated, the list of key influencing factors is further optimized, and the indicator change data is extracted to obtain the real-time change indicators.

[0010] Based on the real-time change indicators, predict future water quality fluctuation trends, identify potential environmental risk control points, and obtain early warning signal sequences.

[0011] High-priority items are extracted from the warning signal sequence, and the wastewater monitoring data are regrouped according to the high-priority items. If the grouping shows a continuous deviation, it is determined that a dynamic operation strategy needs to be executed immediately, and a purification process adjustment instruction is obtained.

[0012] The wastewater treatment system configuration is updated according to the purification process adjustment instructions. The status monitoring values ​​of the adjusted equipment are verified, the accuracy of abnormal signal extraction is determined, and reinforcement learning algorithm is used for iterative optimization to obtain an overall environmental risk control assessment.

[0013] Secondly, the present invention provides a wastewater treatment fault early warning system, comprising:

[0014] The data acquisition module is used to acquire wastewater monitoring data, extract water quality fluctuation signals, perform anomaly detection based on the water quality fluctuation signals, and obtain anomaly signal extraction results.

[0015] The data judgment module is used to group the data of the abnormal signal extraction results. If the fluctuation range of the data in the group deviates from the preset range, it is judged as a potential fault feature identification signal and a list of key influencing factors is obtained.

[0016] The data classification module is used to extract real-time indicator change data from the list of key influencing factors, classify the real-time indicator change data, determine the dynamic operation strategy adjustment parameters, and obtain the purification process adjustment plan.

[0017] The data input module is used to input the purification process adjustment plan into the wastewater treatment process control equipment, extract the status monitoring value of the equipment, and if the status monitoring value exceeds the preset status threshold, update the abnormal signal extraction process, further optimize the list of key influencing factors, and extract the indicator change data to obtain real-time change indicators.

[0018] The data prediction module is used to predict future water quality fluctuation trends based on the real-time changing indicators, determine potential environmental risk control points, and obtain early warning signal sequences.

[0019] The instruction generation module is used to extract high-priority items from the warning signal sequence, regroup the wastewater monitoring data according to the high-priority items, and if the grouping shows a continuous deviation, it is determined that a dynamic operation strategy needs to be executed immediately to obtain a purification process adjustment instruction.

[0020] The data update module is used to update the configuration of the wastewater treatment system according to the purification process adjustment instructions, verify the status monitoring values ​​of the adjusted equipment, and use reinforcement learning algorithms to iteratively optimize the system, determine the accuracy of abnormal signal extraction, and obtain an overall environmental risk control assessment.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] (1) This method separates the trend component and fluctuation component of sewage monitoring data, extracts local features such as the duration and trend of abnormal signals, and can also distinguish between abnormal water quality fluctuations and abnormal equipment status. It breaks through the limitations of single threshold comparison, effectively captures hidden signals with normal single indicators but abnormal combined features, improves the accuracy of hidden anomaly identification, increases the advance warning of anomaly alarms, and reserves sufficient control time compared with existing technologies.

[0023] (2) This method groups the initial abnormal signals by multiple dimensions, mines the intrinsic relationship between data through time series correlation analysis, sorts the key influencing factors, extracts real-time indicator change data from the list of key influencing factors, and determines the dynamic operation strategy adjustment parameters after classification and processing, and generates a targeted purification process adjustment plan, rather than relying on fixed parameters, thereby improving the efficiency of key factor identification, eliminating the need for manual cross-analysis of water quality and equipment data, quickly locating collaborative anomalies, and improving the accuracy of fault tracing. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the wastewater treatment fault early warning method provided in the first embodiment of the present invention;

[0025] Figure 2 This is a schematic diagram of the wastewater treatment fault early warning system provided in the second embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Reference Figure 1 The first embodiment of the present invention provides a wastewater treatment fault early warning method, including the following steps:

[0028] S11, acquire wastewater monitoring data and extract water quality fluctuation signals, perform anomaly detection based on the water quality fluctuation signals, and obtain anomaly signal extraction results;

[0029] S12, group the data of the abnormal signal extraction result. If the fluctuation range of the data in the group deviates from the preset range, it is judged as a potential fault feature identification signal, and a list of key influencing factors is obtained.

[0030] S13, extract real-time indicator change data from the list of key influencing factors, classify the real-time indicator change data, determine the dynamic operation strategy adjustment parameters, and obtain the purification process adjustment plan.

[0031] S14, input the purification process adjustment plan into the wastewater treatment process control equipment, extract the status monitoring value of the equipment, if the status monitoring value exceeds the preset status threshold, update the abnormal signal extraction process, further optimize the list of key influencing factors, and extract the indicator change data to obtain real-time change indicators.

[0032] S15, based on the real-time change indicators, predict the future water quality fluctuation trend, determine potential environmental risk control points, and obtain an early warning signal sequence;

[0033] S16, extract high-priority items from the warning signal sequence, regroup the wastewater monitoring data according to the high-priority items, and if the grouping shows a continuous deviation, it is determined that a dynamic operation strategy needs to be executed immediately to obtain a purification process adjustment instruction;

[0034] S17, update the wastewater treatment system configuration according to the purification process adjustment instruction, verify the status monitoring values ​​of the adjusted equipment, determine the accuracy of abnormal signal extraction, and use reinforcement learning algorithm to iteratively optimize and obtain an overall environmental risk control assessment.

[0035] In step S11, acquiring wastewater monitoring data, extracting water quality fluctuation signals, and performing anomaly detection based on the water quality fluctuation signals to obtain anomaly signal extraction results includes:

[0036] The trend component and fluctuation component are separated from the wastewater monitoring data to obtain the water quality fluctuation signal;

[0037] If the water quality fluctuation signal exceeds the preset water quality fluctuation range, it is judged as a potential abnormal signal, and the specific time point and amplitude data are extracted from the potential abnormal signal.

[0038] Local feature extraction is performed on the specific time points and the amplitude data to determine the duration and trend of the potential abnormal signal, thereby obtaining a detailed feature description;

[0039] The abnormal signals are classified according to the detailed feature description to determine whether they belong to abnormal water quality fluctuations or abnormal equipment status. The classified abnormality types and corresponding time series segments are extracted to obtain the abnormal signal extraction results.

[0040] It should be noted that wastewater monitoring data is collected in real time through a sensor network deployed at key nodes of the wastewater treatment system, such as the inlet, aeration tank, sedimentation tank, and outlet. The core collected indicators include water quality parameters and equipment operation-related parameters. Water quality parameters include turbidity, pH value, dissolved oxygen, COD, and ammonia nitrogen. Equipment operation-related parameters include influent flow rate, aeration tank blower current, and dosing pump speed. The collection frequency is uniformly 10 seconds per data point, and the data is stored in a structured format of "indicator name-collection timestamp-value-sensor ID".

[0041] In this embodiment, the STL time series decomposition algorithm is used to separate trend components and fluctuation components. The core parameters are calibrated based on 500 sets of historical fault data. Referring to the daily cycle fluctuation characteristics of the sewage treatment system, the initial value of the time window width is set to 12 hours, with a step size of 6 hours, and iterative testing is conducted within the range of 6 hours to 36 hours (covering 6 / 12 / 18 / 24 / 30 / 36 hours, a total of 6 candidate values). The trend component explanatory power (the proportion of the original data fluctuation that the trend component can explain, which can be calculated by dividing the trend component variance by the original data variance and multiplying by 100%) and the fault signal extraction rate (the proportion of fault signals contained in the decomposed fluctuation component, which can be calculated by dividing the length of the data containing the fault period in the fluctuation component by the total fault duration and multiplying by 100%) are used for evaluation. Under the 24-hour window width, the trend component can explain 85% of the long-term fluctuation of the original data (adapting to the daily cycle characteristics), while the fault signal extraction rate reaches 95% (only 5% of weak fault signals are not captured), which is determined as the calibration value of the time window width. For example, iterative testing was conducted within the range of 0.4 to 0.8 using a step size of 0.05 (9 candidate values ​​in total: 0.4 / 0.45 / 0.5 / 0.55 / 0.6 / 0.65 / 0.7 / 0.75 / 0.8). A dual-index evaluation was used: weak anomaly capture rate (the number of weak anomaly signals identified in the fluctuation component divided by the actual number of weak anomaly signals multiplied by 100%) and noise filtering rate (the amount of noise data removed from the fluctuation component divided by the total amount of noise data multiplied by 100%). When α=0.75, the weak anomaly capture rate reached 90% (missing only 10% of risk-free, extremely weak noise anomalies), and the noise filtering rate reached 85% (effectively removing invalid noise such as sensor drift), which was determined as the calibration value for the trend fit parameter. Specifically, the decomposition process involved smoothing the raw monitoring data (using a 3-point moving average method to remove instantaneous sensor noise). The long-term trend component (reflecting slow changes in water quality / equipment parameters, such as the natural upward trend of daily COD) was fitted using Loess local weighted regression. Then, the trend component was subtracted from the original data to obtain the fluctuation component, which only contains short-term sudden fluctuations, i.e., the water quality fluctuation signal.

[0042] For example, the fluctuation range adopts a two-dimensional system of basic threshold and scenario-based dynamic adjustment to ensure adaptability to complex operating environments. The basic threshold is based on 1000 sets of fault-free data from nearly 90 days of fault-free operation, calculating the standard deviation of the fluctuation components of each indicator, and setting the basic fluctuation range to between negative two standard deviations and positive two standard deviations. The scenario-based adjustment rules are as follows: if the influent flow exceeds the rated flow by ±30%, the fluctuation range of turbidity and COD is increased by 30% (to adapt to normal fluctuations caused by sudden changes in flow); if it is during the peak period of industrial wastewater discharge (determined by timestamps), the fluctuation range of ammonia nitrogen and total phosphorus is increased by 20% (to adapt to the fluctuation characteristics of industrial pollutant concentrations); if the equipment is in the start-up and shutdown phase, the fluctuation range of the corresponding related parameters (such as dissolved oxygen) is increased by 50% (to avoid misjudging normal fluctuations during equipment start-up and shutdown).

[0043] In one implementation, water quality fluctuation signals are iterated. If the fluctuation value at a certain time point exceeds the fluctuation range of the corresponding scenario, and the fluctuation value does not return to the range for three consecutive acquisition cycles (30 seconds), it is marked as a potential abnormal signal. The specific time point (accurate to the acquisition timestamp) and amplitude data (including fluctuation peak value, deviation from the threshold value, and peak occurrence time) of the potential abnormal signal are recorded.

[0044] For example, a sliding window method is used to extract local features. The window parameters are adapted to a sampling frequency of 1 minute / time, and the window width is set to 15 seconds (containing 3 data points); for example, the sliding step size is set to 1 minute. The feature extraction dimensions include three core indicators: duration, trend, and fluctuation frequency. The duration is calculated as the time difference between the time point when the potential abnormal signal first exceeds the threshold and the time point when the fluctuation value returns to within the threshold. The trend is determined by calculating the linear regression slope of the fluctuation data within the window: a slope > 0.5 mg / L / min indicates a rapid upward trend, -0.5 mg / L / min < slope ≤ 0.5 mg / L / min indicates a stable trend, and a slope ≤ -0.5 mg / L / min indicates a rapid downward trend. The fluctuation frequency is obtained by counting the number of times the fluctuation value crosses the upper / lower limit of the threshold within the window, thus providing a detailed feature description.

[0045] The C4.5 decision tree classification model was used, with detailed feature descriptions as input. Combining the correlation of equipment operating parameters, it distinguished between abnormal water quality fluctuations and abnormal equipment status. Water quality fluctuations were considered abnormal if any of the following conditions were met: the Pearson correlation coefficient between the abnormal indicator and environmental parameters such as influent flow rate and industrial wastewater proportion was ≥0.7 (indicating the abnormality was caused by changes in the external environment); the fluctuation frequency was ≤3 times / 5 minutes and the duration was ≥5 minutes (consistent with the gradual change characteristics of water quality components); the abnormality only occurred at the inlet monitoring node, and the corresponding indicators in subsequent treatment units (aeration tanks, sedimentation tanks) did not fluctuate synchronously. Equipment status was considered abnormal if any of the following conditions were met: the Pearson correlation coefficient between the indicator and equipment operating parameters (such as fan current and pump speed) was ≥0.6 (indicating the abnormality was caused by equipment failure); the fluctuation frequency was ≥5 times / 5 minutes; the abnormality was limited to the monitoring node corresponding to a single piece of equipment (e.g., abnormal dissolved oxygen in one aeration tank while other aeration tanks were normal); and the time of the abnormality coincided with the equipment start-up / shutdown and maintenance timestamps (±10 minutes).

[0046] For example, the anomaly type (abnormal water quality fluctuation or abnormal equipment status) is clearly marked, and quantitative data for classification are added. A complete data segment is extracted from 10 minutes before the start time of the anomaly to 10 minutes after the end time, including the original monitoring data, trend components, fluctuation components and related equipment parameter data, to form an anomaly time series data package and obtain the anomaly signal extraction result.

[0047] The C4.5 decision tree classification model training process involves collecting 1000 valid samples (500 fault samples and 500 normal samples), inputting 12-dimensional feature data (4-dimensional core anomaly features, 3-dimensional environmental association features, 3-dimensional equipment association features, and 2-dimensional scene features), and labeling them with binary classification tags: "1" represents abnormal water quality fluctuations, and "0" represents abnormal equipment status. For 8-dimensional numerical features such as fluctuation peak and duration, an equal-frequency discretization method is used to divide them into 3-5 intervals (the number of intervals is determined based on the principle of maximizing information entropy). The information gain filtering method is used to screen features, calculate the information gain of each feature to the anomaly type label, and remove redundant features with information gain < 0.05. Finally, 9 valid features are retained for training. The minimum number of samples (leaf nodes) is set to 5 (if the number of samples in a node is < 5, splitting is stopped to avoid overfitting); the information gain ratio threshold is 0.1 (a feature is used to split nodes only when its information gain ratio is ≥ 0.1); and the maximum tree depth is 8 (to avoid the tree structure being too deep, which would reduce the generalization ability, and was determined through 5-fold cross-validation).

[0048] Using all samples in the training set as the root node, calculate the information gain ratio (IVR) of each effective feature. Select the feature with the largest IVR as the splitting attribute of the current node, and divide it into child nodes according to the discretization interval of the feature. Repeat the calculation of IVR and the division of child nodes for each child node until the stopping condition is met (number of samples in the node < 5, IVR < 0.1, or all samples belong to the same category). Calculate the pessimistic error rate of each branch (assuming the error rate of the leaf node follows a binomial distribution, add a penalty term of 0.5). If the pessimistic error rate of a branch is ≥ the pessimistic error rate of its parent node, prune the branch (replace the branch with a leaf node, labeled with the anomaly type with the most samples in the branch). After pruning, verify the model accuracy using the validation set. If the accuracy improves by ≥ 3%, retain the pruning results; otherwise, revert to the previous state. Stop training when any of the following conditions are met: the improvement in validation set accuracy over 5 consecutive iterations (adjusting 1 parameter per iteration) is ≤ 0.5%; or the number of training iterations reaches 50 (to avoid infinite iteration).

[0049] In step S12, the data of the abnormal signal extraction results are grouped. If the fluctuation amplitude of the data in the group deviates from the preset amplitude range, it is judged as a potential fault feature identification signal, and a list of key influencing factors is obtained, including:

[0050] The data from the abnormal signal extraction results are grouped, and the feature boundaries of each group are determined to obtain a grouped data set.

[0051] If the fluctuation range of the grouped data set exceeds the preset range, it is judged as a potential fault feature, and the corresponding abnormal data segment is extracted.

[0052] Time series correlation analysis was performed on the abnormal data fragments to identify key influencing factors and obtain a list of factor descriptions;

[0053] The factors are categorized based on the list of factors described, their degree of influence is determined, and they are sorted to obtain a list of key influencing factors.

[0054] It should be noted that, based on the preliminary abnormal signal extraction results, the data was grouped according to three dimensions: abnormality type, monitoring node, and time period. The abnormality type dimension was divided into two main categories: water quality fluctuation abnormalities and equipment status abnormalities. The monitoring node dimension was further subdivided according to the key nodes of the wastewater treatment process, including the inlet, aeration tank, sedimentation tank, and effluent outlet. The time period dimension was divided into 30-minute segments, with each segment containing 60 data points. The K-means clustering algorithm was used to perform secondary optimization on the preliminary grouped data after the three-dimensional division. The core parameters were calibrated based on 500 sets of historical abnormal data. The clustering effect of different K values ​​was calculated using the silhouette coefficient method, and the K value with the largest silhouette coefficient (K=3-5) was selected. The peak value, duration, and slope of the abnormal signal fluctuation were used as clustering features. For each cluster group, ±2 standard deviations of its cluster center were calculated as the feature boundary to determine the feature boundary of each data group, thus obtaining the grouped data set.

[0055] For each grouped dataset, the fluctuation range of the indicators within the group is calculated (the difference between the maximum and minimum values ​​of the indicators within the group, divided by the mean of the indicators within the group, multiplied by 100%). A basic threshold and scenario-based dynamic adjustment system is adopted. The amplitude range is set based on historical data and industry standards. The basic threshold is selected from 1000 groups of similar grouped data during the past 90 days of fault-free operation, and the standard deviation of their fluctuation range is calculated. The basic amplitude range is set between 0 and twice the standard deviation. The scenario adjustment rules are as follows: if the group corresponds to the peak period of industrial wastewater discharge, the basic range is increased by 20%; if the group corresponds to the rainy season water inflow period, the basic range is increased by 30%; if the group is a type of equipment status abnormality, the basic range is decreased by 15%. If the fluctuation range of the grouped data is greater than the preset amplitude range of the corresponding scenario, and the fluctuation does not return for two time periods (60 minutes), it is judged as a potential fault feature. The complete data segment from the first time the fluctuation range of the group exceeds the threshold to the return threshold is extracted, including "segment ID, group ID, indicator name, time range (accurate to the second), number of data points, and peak fluctuation range".

[0056] Pearson correlation coefficient analysis and Granger causality test were used to explore the association between anomalous data fragments and potential influencing factors. Potential influencing factors related to anomalous indicators were selected, and their Pearson correlation coefficients were calculated. Factors with an absolute correlation coefficient ≥ 0.6 (strong correlation) were screened. Granger causality tests were performed on strongly correlated factors (lag order = 3, determined based on the AIC criterion). If the p-value < 0.05, the factor was determined to be an anomalous causal influencing factor. Influencing factors were extracted from three dimensions: environment, equipment, and scenario. Environmental factors included influent flow rate, industrial wastewater proportion, rainwater mixing amount, and influent pH value; equipment factors included chemical dosing pump dosage, aeration fan power, bar screen interception efficiency, and sedimentation tank sludge return ratio; scenario factors included time period type (peak / off-peak), season type (rainy season / dry season), and equipment maintenance status (operating / shutdown). The list includes "factor name, associated anomalous indicator, correlation coefficient, Granger test p-value, direction of influence (positive / negative), and description of the influencing scenario," resulting in a factor description list.

[0057] The list of factors is divided into three categories: direct influencing factors, indirect influencing factors, and auxiliary influencing factors. Direct influencing factors are those with a direct causal relationship to the abnormal indicators; indirect influencing factors affect the abnormal indicators through intermediate indicators; and auxiliary influencing factors only have an impact in specific scenarios. Data weights are calculated based on the absolute values ​​of correlation coefficients. Each data weight equals the corresponding correlation coefficient divided by the sum of the correlation coefficients. The data are then sorted in descending order of weight to obtain the list of key influencing factors.

[0058] In step S13, the step of extracting real-time indicator change data from the list of key influencing factors, classifying and processing the real-time indicator change data, determining dynamic operation strategy adjustment parameters, and obtaining a purification process adjustment plan includes:

[0059] Real-time indicator change data is extracted from the list of key influencing factors. If the fluctuation of the real-time indicator change data exceeds the preset environmental fluctuation range, it is judged as an abnormal change point, and a set of abnormal change points is obtained.

[0060] The set of abnormal change points is grouped and analyzed to identify key points related to the dynamic operation strategy, resulting in a parameter adjustment priority list.

[0061] Based on the parameter adjustment priority list, the corresponding operating status indicators are extracted to determine the direction of the purification process adjustment and obtain a preliminary adjustment plan.

[0062] The preliminary adjustment plan is verified. If the verification result meets the preset verification standard, the preliminary adjustment plan is determined to be the final adjustment plan, and the purification process adjustment plan is obtained.

[0063] It should be noted that, from the list of key influencing factors, real-time monitoring data corresponding to environmental influencing factors were selected, focusing on high-priority factors with a data weight ≥0.18 (based on statistical analysis and process correlation verification of 500 sets of historical fault data, factors with a weight ≥0.18 have a probability of causing a fault ≥15%; factors with a weight <0.18 have a probability of causing a fault <5%), while auxiliary influencing factors were eliminated. A dual-dimensional system of basic threshold and dynamic adjustment of process scenarios was adopted. Based on historical environmental data of nearly 180 days of fault-free operation and industry process standards, the basic threshold was calculated as the "median ± 2 standard deviations" of historical data for each type of environmental factor as the basic fluctuation range. The basic range was adapted according to the real-time process scenario. During the rainy season (daily rainfall ≥50mm), the influent flow rate range was increased by 30%, and the rainwater mixing range was increased by 50%. During the peak period of industrial wastewater discharge, the industrial wastewater proportion range was increased by 20%. In the case of influent impact scenario (flow rate increase ≥50% within 10 minutes), the basic range was temporarily relaxed by 50% (and restored after 1 hour). If the real-time value of a certain environmental factor exceeds the preset fluctuation range of the corresponding scenario for three consecutive collection cycles (30 seconds), it is marked as an abnormal change point, and a set of abnormal change points is obtained.

[0064] Wastewater treatment processes are grouped into two dimensions: process stages and influencing factors. The process stage dimension is divided into influent regulation, chemical dosing and reaction, aeration and sedimentation, and effluent protection, retaining only the stages involved in abnormal changes. The factor type dimension is subdivided into "flow rate (influent flow rate), composition (industrial wastewater proportion, pH value), and external source (rainwater mixing)". For each group of abnormal changes, the corresponding "process control nodes" and "adjustable parameters" are analyzed to establish a mapping relationship of "abnormal factor - control node - adjustment parameter". The priority score is calculated using a graded assignment and summation method, with the degree of impact assigned (peak exceedance ≥20%=3 points, 10%-20%=2 points, <10%=1 point) and the degree of urgency assigned (duration >10 minutes=3 points, 5-10 minutes=2 points, <5 minutes=1 point). A score ≥5 is high priority, 3-4 is medium priority, and <3 is low priority. The parameters are arranged in descending order of priority score to form a parameter adjustment priority list.

[0065] Based on the parameter adjustment priority list, extract the current operating status indicators and target control indicators corresponding to each type of adjustment parameter. The current operating status indicators include the current value of the adjustment parameter, the operating load of associated equipment, and the real-time value of the effluent indicators. Based on the effluent standards of the wastewater treatment plant, set the adjusted target indicators. Determine the adjustment direction based on the deviation between the current indicator and the target indicator. The deviation is equal to the difference between the current value and the target value divided by the target value multiplied by 100%. If the deviation > 0 (the current value exceeds the target value), the adjustment direction is to decrease the parameter value; if the deviation < 0, the adjustment direction is to increase the parameter value. The adjustment range is determined by 0.8 times the deviation range (to avoid over-adjustment leading to new anomalies), resulting in a preliminary adjustment plan.

[0066] The verification process employs a dual approach: simulated operation and historical data comparison. The preliminary adjustment plan is input into wastewater treatment process simulation software (such as GPS-X), with influent water quality and equipment parameters set to match the current settings. The simulation is run for one hour, outputting simulated effluent indicators and equipment load. Historical adjustment records from the past three months, similar to the current scenario, are queried, and the consistency between the adjusted parameters and the predicted effects is compared. If the deviation between the historical adjusted effect and the predicted effect is ≤15% (a cross-statistic is performed on the deviation rate and the effluent compliance rate after adjustment for 300 cases; when the deviation rate is ≤15%, the compliance rate is 100%; when the deviation rate is >15%, the compliance rate is only 20.8%), then the verification is considered successful. If the verification fails, the parameter priorities are readjusted, a new preliminary plan is generated, and the verification is repeated until the verification is successful, at which point the preliminary plan is determined as the final plan.

[0067] In step S14, the purification process adjustment plan is input into the wastewater treatment process control equipment, and the status monitoring values ​​of the equipment are extracted. If the status monitoring values ​​exceed a preset status threshold, the abnormal signal extraction process is updated, the list of key influencing factors is further optimized, and the indicator change data is extracted to obtain real-time change indicators, including:

[0068] The purification process adjustment plan is input into the wastewater treatment process control equipment, and the status monitoring value of the equipment is extracted. If the status monitoring value exceeds the preset status threshold, abnormal signal data is generated to obtain a preliminary set of abnormal signal data.

[0069] The preliminary dataset is grouped, and optimized feature points are extracted to obtain the classified signal feature combination.

[0070] The signal feature combination is matched with the list of key influencing factors. If the matching result meets the preset matching criteria, the priority order in the list of key influencing factors is updated to obtain the set of influencing factors to be optimized.

[0071] Data on changes in indicators are extracted from the set of influencing factors to obtain real-time changing indicators.

[0072] It should be noted that the purification process adjustment scheme is input into the wastewater treatment process control equipment. The monitoring equipment extracts the status monitoring values ​​of the equipment, specifically including turbidity, influent flow rate and equipment vibration frequency. The threshold is set by multiplying the rated parameter by a safety factor. The safety factor is adjusted according to the importance of the indicators (turbidity and influent flow rate are 0.8, vibration frequency is 0.9). If the monitoring value exceeds the threshold for three consecutive collection cycles (30 seconds) or a single value exceeds the threshold by 15%, an abnormal signal is generated. The set includes "abnormal signal ID, sensor ID, monitoring indicator, abnormal timestamp, measured value, preset threshold, and exceedance range", thus obtaining the preliminary data set of the abnormal signal.

[0073] The abnormal signals are grouped according to two dimensions: monitoring dimensions and indicator types. The monitoring dimensions are divided into process indicators (turbidity, flow rate) and equipment operation indicators (vibration frequency). The indicator types are divided into turbidity, flow rate, and vibration frequency. Five core feature points are extracted from each group of abnormal signals to quantify the key attributes of the abnormality, including the time of the abnormality start (time stamp of the first time the threshold is exceeded), the duration of the abnormality, the peak exceedance (the deviation rate between the measured maximum value and the threshold), the fluctuation frequency (the number of times the threshold is exceeded within 1 minute), and the associated indicator values ​​(other core indicator data at the same time stamp). The feature combinations are presented in a structured table to obtain the classified signal feature combinations.

[0074] The system employs a dual-dimensional approach, considering both feature point overlap and process correlation. Feature point overlap is determined by the correlation index between the signal feature combination and key influencing factors, as well as the overlap rate of the influencing scenarios. The system compares the signal feature combination with a specific factor from the list of key influencing factors, comparing each factor individually in terms of correlation index, influencing scenario, peak exceedance range, abnormal process, trend direction, and related equipment. A match is considered successful if the following conditions are met: complete consistency in correlation index, influencing scenario, abnormal process, and related equipment; signal amplitude falling within the peak exceedance range; and signal trend consistent with factor trend direction. The number of successfully matched feature items is counted, and the overlap is calculated as the number of feature items divided by 6 multiplied by 100%.

[0075] Process correlation is based on the wastewater treatment process to determine the causal relationship between abnormal indicators and influencing factors. Using "abnormal indicator - influencing factor" as the analysis object, and combining process link length and data correlation, the causal link length of "influencing factor → abnormal indicator" is determined according to the wastewater treatment process flow diagram. A causal link length of 1 indicates direct causation with a correlation of 100%; a causal link length of 2 indicates indirect causation with a correlation of 60%; a causal link length ≥ 3 indicates distant indirect causation with a correlation of 30%; no link indicates no causal relationship with a correlation of 0%. If the overlap is ≥ 70% (based on statistical verification of 500 historical abnormal signal-key factor matching cases, an overlap rate ≥ 70% indicates a matching effectiveness of 95%, while an overlap rate < 70% reduces the matching effectiveness to 62%) and the correlation is ≥ 60% (cross-analysis of the correlation strength and fault handling success rate of 500 abnormal cases shows a correlation ≥ 60% with a fault handling success rate ≥ 90%; a correlation < 60% indicates a fault handling success rate of only 28%), the match is considered successful. Based on the newly extracted feature points, the priority scores are recalculated, and the factors are reordered according to their scores to obtain the set of influencing factors that need to be optimized.

[0076] Key indicator data for turbidity, flow rate, and vibration frequency are extracted from the set of influencing factors. The extraction range is from 10 minutes before the anomaly to the current moment. The rate of change of the indicator is calculated by dividing the difference between the current value and the value 10 minutes ago by the value 10 minutes ago and multiplying by 100%. The trend direction is determined by the slope of the linear fit: >0 indicates an increase, <0 indicates a decrease, and approximately equal to 0 indicates stability. The indicator is output as a real-time change indicator containing "factor name, monitoring indicator, three-dimensional values ​​(current value, rate of change, trend direction) and generation timestamp".

[0077] In step S15, the step of predicting future water quality fluctuation trends based on the real-time change indicators, determining potential environmental risk control points, and obtaining an early warning signal sequence includes:

[0078] Periodic and trend features are extracted from the real-time changing indicators to obtain a preliminary set of fluctuation anomaly signals;

[0079] The abnormal fluctuation signals are grouped and processed to extract fluctuation trend prediction feature combinations, resulting in classified signal feature groups.

[0080] The signal feature group is compared with a pre-established risk control point database. If the matching result meets the comparison criteria, the point priority is updated and potential environmental risk control points are determined.

[0081] Based on the potential environmental risk control points, relevant early warning feature values ​​are extracted to generate corresponding early warning signal sequences.

[0082] It should be noted that the input is based on real-time changing indicators, focusing on core indicators that are strongly correlated with water quality fluctuations, including turbidity, influent flow rate, COD concentration, ammonia nitrogen concentration, influent pH value, proportion of industrial wastewater, and amount of rainwater mixed in. Fourier transform and periodogram analysis were used to identify periodic fluctuations in water quality indicators. A Fast Fourier Transform (FFT) was performed on time-series data (60 data points) over one hour to convert the time-domain signal to a frequency-domain signal. The periodogram (power spectral density) was calculated to identify the period corresponding to the peak frequency (period = 3600 seconds / frequency). Significant periods (power spectral density > twice the average density) were selected. A sliding window linear regression was used to calculate the trend slope and capture the long-term trend of water quality indicators: the window size was 30 minutes (30 data points), and the sliding step was 5 minutes. Linear fitting was performed on the data within each window, and the fitting slope was calculated. A trend slope > 0.05 indicated a strong upward trend, 0.01 ≤ k ≤ 0.05 indicated a weak upward trend, -0.01 ≤ k ≤ 0.01 indicated a stable trend, -0.05 ≤ k < -0.01 indicated a weak downward trend, and k < -0.05 indicated a strong downward trend.

[0083] When the fluctuation amplitude within a significant period is greater than 30% of the historical average for the same period and the trend slope is strong upward / strong downward and lasts for 2 windows (10 minutes), a fluctuation anomaly signal is generated; the set contains "anomaly signal ID, indicator name, period characteristics, trend characteristics, and start timestamp", thus obtaining a preliminary fluctuation anomaly signal set.

[0084] The water quality was grouped into two dimensions: the treatment process stage and the correlation between indicators. The process stages were divided into the influent equalization tank, the aeration biological tank, the sedimentation tank, and the effluent outlet. The correlation between indicators was based on the Pearson correlation coefficient (|r|≥0.7) for clustering. Five predictive features were extracted from each group of signals to quantify the future trend of water quality fluctuations, including the peak value of the periodic fluctuation (the maximum value of the indicator within the period minus the minimum value), the duration of the trend (the duration from the first determination of a strong upward / downward trend to the present), the synergy of the correlated indicators (the number of indicators with the same trend divided by the total number of indicators in the group multiplied by 100%), the probability of historical recurrence (the number of times the same combination occurred divided by the total number of anomalies multiplied by 100%), and the change in the prediction window (the predicted fluctuation value for the next 30 minutes, based on the ARIMA model (parameters p=2, d=1, q=2)).

[0085] The ARIMA model training process was based on historical monitoring data of the wastewater treatment system over the past two years, with a data collection frequency of 10 minutes per data point, generating a total of 52,560 valid data sets. These data were divided into training and validation sets in a 7:3 ratio. First-order differencing was performed on the original data, followed by an ADF test. The ADF statistic for all indicators was <-3.43 (critical value at the 1% significance level), and the p-value was <0.01, thus rejecting the null hypothesis and determining that the data was stationary after first-order differencing. Based on the stationary data after first-order differencing, autocorrelation function (ACF) and partial autocorrelation function (PACF) plots were drawn to observe the truncation characteristics. The ACF value of the data after first-order differencing fell into the 95% confidence interval at lag 2, and showed no significant fluctuations at subsequent lags, exhibiting a 2nd-order truncation characteristic, indicating q=2 (number of moving average terms = truncation order). The PACF value also fell into the confidence interval at lag 2, and showed no significant fluctuations at subsequent lags, exhibiting a 2nd-order truncation characteristic, indicating p=2 (number of autoregressive terms = truncation order). Initialize model parameters (autoregressive coefficients are 0.3 and 0.2; moving average coefficients are 0.4 and 0.3); input training data, minimize the sum of squared residuals (residuals equal to actual values ​​minus predicted values) using the L-BFGS algorithm, iteratively updating the autoregressive and moving average coefficients; calculate the sum of squared residuals after each iteration, and stop iteration if the sum decreases ≤0.001 for 5 consecutive iterations; if convergence is not achieved after 100 iterations, adjust the initial values ​​(e.g., φ1=0.2, φ2=0.3) and retrain. Output the model's predicted values ​​after convergence.

[0086] The risk control point database is constructed based on the process layout and historical risk records of wastewater treatment plants. It includes basic point information and risk characteristic thresholds. Cosine similarity of feature vectors is used to calculate the matching degree between signal feature groups and database points. The five feature values ​​of the signal feature groups and the risk characteristic thresholds of the database points are converted into standardized vectors (0-1 interval). Cosine similarity is calculated; a cosine similarity ≥ 0.8 (high similarity) indicates a successful match. For successfully matched points, a priority score is calculated by multiplying the risk level by the matching similarity. A score ≥ 4 (requiring immediate monitoring) is high priority; 2.5-4 (for monitoring within 1 hour) is medium priority; and < 2.5 (for monitoring within 24 hours) is low priority. High and medium priority points are selected as potential environmental risk control points.

[0087] Two core early warning features are extracted from potential risk control points to quantify the urgency of the risk: the predicted time of exceedance (the time when the indicator first exceeds the standard based on the ARIMA model) and the magnitude of exceedance (the deviation rate between the predicted value and the emission standard). Three levels of early warning are established based on the magnitude of exceedance and the predicted time of exceedance, corresponding to different response strategies: a red warning is for exceedance magnitude ≥ 50% and predicted exceedance time < 30 minutes, with a response time limit of 5 minutes; a yellow warning is for exceedance magnitude ≤ 30% and predicted exceedance time 30-60 minutes, with a response time limit of 15 minutes; and a blue warning is for exceedance magnitude < 30% and predicted exceedance time > 60 minutes, with a response time limit of 30 minutes. The corresponding early warning signal sequence is obtained by sorting the warnings in chronological order.

[0088] In step S16, extracting high-priority items from the warning signal sequence and regrouping the wastewater monitoring data according to the high-priority items, if the grouping shows a continuous deviation, it is determined that a dynamic operation strategy needs to be executed immediately, and a purification process adjustment instruction is obtained, including:

[0089] The signals in the warning signal sequence are divided into three priority levels: high, medium, and low. The high priority items are extracted to obtain the classified priority signal set.

[0090] The wastewater monitoring data are regrouped according to the priority signal set. If the grouping shows a continuous deviation, the deviation features of the corresponding data are extracted to obtain the deviation feature combination.

[0091] If the combination of deviation features is matched with a pre-established threshold database, the corresponding dynamic operation strategy scheme is extracted to obtain the basis for strategy execution;

[0092] Based on the strategy execution criteria, an adjustment instruction for the purification process is generated, resulting in the purification process adjustment instruction.

[0093] It should be noted that the warning signal sequence is divided into three levels: high, medium, and low. The warning level is red, which is the high priority level; the warning level is yellow, which is the medium priority level; and the warning level is blue, which is the low priority level. The warning signal sequence is traversed, and all high priority level signals are filtered out. Invalid signals that have been processed or are repeated (repeatedly pushed to the same control point within 10 minutes) are removed. The signals are stored in a structured table to obtain the set of priority signals after classification.

[0094] Using high-priority signal sets as the core, wastewater monitoring data are regrouped based on two dimensions: the process stage to which the signal belongs and the associated monitoring indicators. The process stage is defined by the control point process stage corresponding to the high-priority signal, retaining only the monitoring data of that stage and its upstream and downstream related stages. The associated monitoring indicators are defined by the indicators associated with the early warning signals. Persistent deviation is determined by the deviation amplitude, which is equal to the difference between the real-time value of the grouped data and the normal baseline value, divided by the normal baseline value multiplied by 100%. The normal baseline value is the median of the indicators under the same operating conditions over the past 90 days. If the deviation amplitude of the grouped data is > ±15% and the duration is greater than 3 acquisition cycles (30 seconds, adapted to an acquisition frequency of 10 seconds / acquisition), it is determined to be a persistent deviation. Four core deviation features are extracted from the grouped data of persistent deviation to form a deviation feature combination, including the average deviation amplitude (the average deviation amplitude during the persistent deviation period), the duration of the deviation (the cumulative duration from the first deviation to the present), the deviation trend direction (the changing trend of the deviation amplitude), and the synergy of the associated indicators. The deviation feature combination is output in a structured format.

[0095] The threshold database is built based on wastewater treatment process manuals and historical control cases (300 valid cases). Core fields include "deviation characteristic threshold, corresponding dynamic operation strategy, applicable scenario, and associated equipment." Feature overlap matching is used; calculating feature overlap requires counting the number of matches between deviation feature combinations and deviation characteristic thresholds in the database entry, dividing by the total number of features, and multiplying by 100%. A successful match is considered when the feature overlap is ≥80%, and the "dynamic operation strategy," "associated equipment," and "execution parameters" of the database entry are extracted to form the basis for strategy execution. If no matching entry is found (feature overlap <80%), the default strategy is triggered, reducing the load on the corresponding stage by 20%. Based on the strategy execution basis, adjustment instructions that can be directly executed by the equipment are generated, including adjustment parameters, target values ​​after parameter adjustment, and execution time, resulting in purification process adjustment instructions.

[0096] In step S17, the wastewater treatment system configuration is updated according to the purification process adjustment instruction, the status monitoring values ​​of the adjusted equipment are verified, the accuracy of abnormal signal extraction is determined, and reinforcement learning algorithm is used for iterative optimization to obtain an overall environmental risk control assessment.

[0097] It should be noted that, according to the purification process adjustment instructions, the system configuration update needs to synchronize the operating parameters of the equipment. The adjusted monitoring values ​​are compared with the baseline values ​​under normal operating conditions (the average values ​​of indicators under the same conditions over the past 90 days), and the deviation rate is calculated. If the deviation rate is ≤ ±5% and consistently meets the standard for 5 consecutive collection cycles, it is considered to be qualified. Based on the adjusted equipment status monitoring values ​​and actual abnormal situations, the accuracy of abnormal signal extraction is evaluated through three-dimensional indicators: accuracy, recall, and F1 score. Accuracy is equal to the number of correctly extracted abnormal signals divided by the total number of extracted abnormal signals multiplied by 100%. Recall is equal to the number of correctly extracted abnormal signals divided by the number of actual abnormal signals multiplied by 100%. F1 score is equal to the product of accuracy and recall divided by twice the sum of accuracy and recall (a comprehensive evaluation indicator, the higher the better). If the F1 score is ≥90% (based on 500 complete fault cases of the wastewater treatment system, statistical analysis of core performance indicators corresponding to different F1 score ranges shows that when the F1 score is ≥90%, the false negative rate is ≤3%, and all are low-risk anomalies; the false positive rate is ≤5%, the fault handling success rate is ≥95%, and the effluent meets 100% standards), the accuracy of anomaly signal extraction is considered excellent, and the current identification algorithm parameters are maintained; if 70% ≤ F1 score < 90%, the assessment is satisfactory but requires optimization, and the threshold for anomaly signal identification is adjusted; if the F1 score < 70% (when the F1 score is ≥70%, all missed faults are of medium to low risk), the assessment is considered satisfactory. The success rate of fault handling is ≥80%, and the effluent compliance rate is ≥95%. If the F1 score is <70%, the high-risk fault omission rate will rise sharply to 20%. If the fault is deemed unqualified, the algorithm iteration driven by reinforcement learning (based on Q-learning algorithm) will be automatically triggered and the reasons for the deviation will be recorded. Through the trial and error and feedback mechanism of reinforcement learning, the key parameters of the abnormal signal extraction algorithm (such as the water quality fluctuation range threshold and the information gain ratio threshold of C4.5 decision tree) will be adjusted so that the F1 score after iteration is ≥70% (the pass line) and the high-risk fault omission rate is ≤5%, thus obtaining the overall environmental risk control assessment.

[0098] The iterative process based on the Q-learning algorithm involves defining the current environment state of the algorithm iteration, focusing on the performance bottleneck of the anomaly signal extraction algorithm and the current system operation scenario. This includes four dimensions of quantitative parameters: algorithm performance state (current F1 score range), anomaly judgment threshold state (±σ multiple of the current water quality fluctuation range), system operation scenario (current scenario type and deviation rate of core monitoring indicators), and fault risk state (current high-risk fault false positive rate). The adjustable action space of the algorithm parameters is also defined, focusing on the key adjustable parameters of the anomaly signal extraction algorithm. These are all discrete and controllable actions, specifically including three core actions: anomaly judgment threshold adjustment (adjusting the ±σ multiple of the water quality fluctuation range), feature extraction window adjustment (adjusting the sliding window width), and decision tree parameter adjustment. (Adjust the information gain ratio threshold of the C4.5 decision tree); The reward function takes "improving the F1 score and reducing the risk of failure" as its core objective, and comprehensively evaluates the changes in algorithm performance after the action is executed. If the F1 score is improved after the action is executed compared with before the adjustment, the reward for improving the F1 score is 10 times the difference between the F1 score before and after the adjustment; if the F1 score decreases, the reward for improving the F1 score is -20 times the difference between the F1 score before and after the adjustment; if the high-risk false negative rate is lower than before the adjustment, the reward for lowering the high-risk false negative rate is 15 times the difference between the false negative rate before and after the adjustment; if it is higher, the reward for lowering the high-risk false negative rate is -30 times the difference between the false negative rate before and after the adjustment; if the false positive rate is lower than before the adjustment, the reward for lowering the false positive rate is 5 times the difference between the false positive rate before and after the adjustment.

[0099] Construct a Q(S,A) table (state-action value table), initially set to 0. Rows correspond to all combinations in the state space, and columns correspond to the three types of actions in the action space. The iterative process involves collecting the current state; selecting actions based on the ε-greedy strategy; executing actions; adjusting the threshold parameters of the abnormal signal extraction algorithm; collecting the F1 score, false negative rate, and false positive rate of 100 sets of validation data after adjustment; calculating the new state; calculating the reward value according to the reward function; finally updating the Q table; repeating the iterative process until the F1 score is ≥70% and the high-risk false negative rate is ≤5% for 5 consecutive iterations, then stopping the iteration and fixing the current algorithm parameters.

[0100] In summary, this method separates the trend and fluctuation components of wastewater monitoring data, extracts local features such as the duration and trend of abnormal signals, and can distinguish between abnormal water quality fluctuations and abnormal equipment status. It overcomes the limitations of single threshold comparison, effectively capturing latent signals where a single indicator is normal but the combined characteristics are abnormal, improving the accuracy of latent anomaly identification, increasing the lead time for anomaly alarms, and reserving sufficient control time compared to existing technologies. This method groups preliminary abnormal signals by multiple dimensions, mines the inherent relationships between data through time series correlation analysis, ranks key influencing factors, extracts real-time indicator change data from the list of key influencing factors, and determines dynamic operation strategy adjustment parameters after classification and processing, generating targeted purification process adjustment plans, rather than relying on fixed parameters. This improves the efficiency of key factor identification, eliminates the need for manual cross-analysis of water quality and equipment data, quickly locates collaborative anomalies, and improves the accuracy of fault tracing.

[0101] Reference Figure 2 The second embodiment of the present invention provides a wastewater treatment fault early warning system, comprising:

[0102] The data acquisition module is used to acquire wastewater monitoring data, extract water quality fluctuation signals, perform anomaly detection based on the water quality fluctuation signals, and obtain anomaly signal extraction results.

[0103] The data judgment module is used to group the data of the abnormal signal extraction results. If the fluctuation range of the data in the group deviates from the preset range, it is judged as a potential fault feature identification signal and a list of key influencing factors is obtained.

[0104] The data classification module is used to extract real-time indicator change data from the list of key influencing factors, classify the real-time indicator change data, determine the dynamic operation strategy adjustment parameters, and obtain the purification process adjustment plan.

[0105] The data input module is used to input the purification process adjustment plan into the wastewater treatment process control equipment, extract the status monitoring value of the equipment, and if the status monitoring value exceeds the preset status threshold, update the abnormal signal extraction process, further optimize the list of key influencing factors, and extract the indicator change data to obtain real-time change indicators.

[0106] The data prediction module is used to predict future water quality fluctuation trends based on the real-time changing indicators, determine potential environmental risk control points, and obtain early warning signal sequences.

[0107] The instruction generation module is used to extract high-priority items from the warning signal sequence, regroup the wastewater monitoring data according to the high-priority items, and if the grouping shows a continuous deviation, it is determined that a dynamic operation strategy needs to be executed immediately to obtain a purification process adjustment instruction.

[0108] The data update module is used to update the configuration of the wastewater treatment system according to the purification process adjustment instructions, verify the status monitoring values ​​of the adjusted equipment, determine the accuracy of abnormal signal extraction, and use reinforcement learning algorithm to iteratively optimize and obtain an overall environmental risk control assessment.

[0109] It should be noted that the wastewater treatment fault early warning system provided in this embodiment of the invention is used to execute all the process steps of the wastewater treatment fault early warning method in the above embodiment. The working principle and beneficial effect of the two are one-to-one, so they will not be described again.

[0110] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0111] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for early warning of sewage treatment faults, characterized in that, include: Wastewater monitoring data is acquired, and water quality fluctuation signals are extracted. Anomaly detection is performed based on the water quality fluctuation signals to obtain anomaly signal extraction results. The data of the abnormal signal extraction results are grouped. If the fluctuation range of the data in the group deviates from the preset range, it is judged as a potential fault feature identification signal, and a list of key influencing factors is obtained. Real-time indicator change data is extracted from the list of key influencing factors, the real-time indicator change data is classified and processed, dynamic operation strategy adjustment parameters are determined, and purification process adjustment plan is obtained. The purification process adjustment plan is input into the wastewater treatment process control equipment, and the status monitoring value of the equipment is extracted. If the status monitoring value exceeds the preset status threshold, the abnormal signal extraction process is updated, the list of key influencing factors is further optimized, and the indicator change data is extracted to obtain the real-time change indicators. Based on the real-time change indicators, predict future water quality fluctuation trends, identify potential environmental risk control points, and obtain early warning signal sequences. High-priority items are extracted from the warning signal sequence, and the wastewater monitoring data are regrouped according to the high-priority items. If the grouping shows a continuous deviation, it is determined that a dynamic operation strategy needs to be executed immediately, and a purification process adjustment instruction is obtained. The wastewater treatment system configuration is updated according to the purification process adjustment instructions. The status monitoring values ​​of the adjusted equipment are verified, the accuracy of abnormal signal extraction is determined, and reinforcement learning algorithm is used for iterative optimization to obtain an overall environmental risk control assessment. Specifically, the purification process adjustment plan is input into the wastewater treatment process control equipment, and the equipment status monitoring values ​​are extracted. If the status monitoring values ​​exceed a preset status threshold, the abnormal signal extraction process is updated, the list of key influencing factors is further optimized, and indicator change data is extracted to obtain real-time change indicators, including: The purification process adjustment plan is input into the wastewater treatment process control equipment, and the status monitoring value of the equipment is extracted. If the status monitoring value exceeds the preset status threshold, abnormal signal data is generated to obtain a preliminary set of abnormal signal data. The preliminary dataset is grouped, and optimized feature points are extracted to obtain the classified signal feature combination. The signal feature combination is matched with the list of key influencing factors. If the matching result meets the preset matching criteria, the priority order in the list of key influencing factors is updated to obtain the set of influencing factors to be optimized. Data on changes in indicators are extracted from the set of influencing factors to obtain real-time changing indicators.

2. The wastewater treatment fault early warning method according to claim 1, characterized in that, The process of acquiring wastewater monitoring data, extracting water quality fluctuation signals, performing anomaly detection based on the water quality fluctuation signals, and obtaining anomaly signal extraction results includes: The trend component and fluctuation component are separated from the wastewater monitoring data to obtain the water quality fluctuation signal; If the water quality fluctuation signal exceeds the preset water quality fluctuation range, it is judged as a potential abnormal signal, and the specific time point and amplitude data are extracted from the potential abnormal signal. Local feature extraction is performed on the specific time points and the amplitude data to determine the duration and trend of the potential abnormal signal, thereby obtaining a detailed feature description; The abnormal signals are classified according to the detailed feature description to determine whether they belong to abnormal water quality fluctuations or abnormal equipment status. The classified abnormality type and corresponding time series segment are extracted to obtain the abnormal signal extraction result.

3. The wastewater treatment fault early warning method according to claim 1, characterized in that, The data from the extracted abnormal signals are grouped. If the fluctuation range of the data in a group deviates from a preset range, it is determined to be a potential fault feature identification signal, resulting in a list of key influencing factors, including: The data from the abnormal signal extraction results are grouped, and the feature boundaries of each group are determined to obtain a grouped data set. If the fluctuation range of the grouped data set exceeds the preset range, it is judged as a potential fault feature, and the corresponding abnormal data segment is extracted. Time series correlation analysis was performed on the abnormal data segments to identify key influencing factors and obtain a list of factor descriptions; The factors are categorized based on the list of factors described, their degree of influence is determined, and they are sorted to obtain a list of key influencing factors.

4. The wastewater treatment fault early warning method according to claim 1, characterized in that, The process of extracting real-time indicator change data from the list of key influencing factors, classifying and processing the real-time indicator change data, determining dynamic operation strategy adjustment parameters, and obtaining a purification process adjustment plan includes: Real-time indicator change data is extracted from the list of key influencing factors. If the fluctuation of the real-time indicator change data exceeds the preset environmental fluctuation range, it is judged as an abnormal change point, and a set of abnormal change points is obtained. The set of abnormal change points is grouped and analyzed to identify key points related to the dynamic operation strategy, resulting in a parameter adjustment priority list. Based on the parameter adjustment priority list, the corresponding operating status indicators are extracted to determine the direction of the purification process adjustment and obtain a preliminary adjustment plan. The preliminary adjustment plan is verified. If the verification result meets the preset verification standard, the preliminary adjustment plan is determined to be the final adjustment plan, and the purification process adjustment plan is obtained.

5. The wastewater treatment fault early warning method according to claim 1, characterized in that, The step of predicting future water quality fluctuation trends based on the real-time changing indicators, determining potential environmental risk control points, and obtaining an early warning signal sequence includes: Periodic and trend features are extracted from the real-time changing indicators to obtain a preliminary set of fluctuation anomaly signals; The abnormal fluctuation signals are grouped and processed to extract the fluctuation trend prediction feature combination, resulting in the classified signal feature group. The signal feature group is compared with a pre-established risk control point database. If the matching result meets the comparison criteria, the point priority is updated and potential environmental risk control points are determined. Based on the potential environmental risk control points, relevant early warning feature values ​​are extracted to generate corresponding early warning signal sequences.

6. The wastewater treatment fault early warning method according to claim 1, characterized in that, The step of extracting high-priority items from the warning signal sequence, regrouping the wastewater monitoring data according to the high-priority items, and determining that a dynamic operation strategy needs to be implemented immediately if the grouping shows a continuous deviation, and obtaining a purification process adjustment instruction, includes: The signals in the warning signal sequence are divided into three priority levels: high, medium, and low. The high priority items are extracted to obtain the classified priority signal set. The wastewater monitoring data are regrouped according to the priority signal set. If the grouping shows a continuous deviation, the deviation features of the corresponding data are extracted to obtain the deviation feature combination. If the combination of deviation features is matched with a pre-established threshold database, the corresponding dynamic operation strategy scheme is extracted to obtain the basis for strategy execution; Based on the strategy execution criteria, an adjustment instruction for the purification process is generated, resulting in the purification process adjustment instruction.

7. A wastewater treatment fault early warning system, used to implement the wastewater treatment fault early warning method as described in any one of claims 1-6, characterized in that, include: The data acquisition module is used to acquire wastewater monitoring data, extract water quality fluctuation signals, perform anomaly detection based on the water quality fluctuation signals, and obtain anomaly signal extraction results. The data judgment module is used to group the data of the abnormal signal extraction results. If the fluctuation range of the data in the group deviates from the preset range, it is judged as a potential fault feature identification signal and a list of key influencing factors is obtained. The data classification module is used to extract real-time indicator change data from the list of key influencing factors, classify the real-time indicator change data, determine the dynamic operation strategy adjustment parameters, and obtain the purification process adjustment plan. The data input module is used to input the purification process adjustment plan into the wastewater treatment process control equipment, extract the status monitoring value of the equipment, and if the status monitoring value exceeds the preset status threshold, update the abnormal signal extraction process, further optimize the list of key influencing factors, and extract the indicator change data to obtain real-time change indicators. The data prediction module is used to predict future water quality fluctuation trends based on the real-time changing indicators, determine potential environmental risk control points, and obtain early warning signal sequences. The instruction generation module is used to extract high-priority items from the warning signal sequence, regroup the wastewater monitoring data according to the high-priority items, and if the grouping shows a continuous deviation, it is determined that a dynamic operation strategy needs to be executed immediately to obtain a purification process adjustment instruction. The data update module is used to update the configuration of the wastewater treatment system according to the purification process adjustment instructions, verify the status monitoring values ​​of the adjusted equipment, determine the accuracy of abnormal signal extraction, and use reinforcement learning algorithm to iteratively optimize and obtain an overall environmental risk control assessment.

Images